ABSTRACT 
 
 
Title of Dissertation: MODELING THE RELATIONSHIP 
BETWEEN THE HOUSING FIRST 
APPROACH AND HOMELESSNESS 
  
 David Boston, Doctor of Philosophy, 2020 
  
Dissertation directed by: Associate Professor Willow Lung-Amam, 
Department of Urban Studies and Planning 
 
A growing body of evidence from individual-level studies demonstrating that 
the Housing First approach is effective at keeping those experiencing homelessness in 
stable housing has led to the approach being championed by many leading experts, 
especially as a way to address chronic homelessness (O'Flaherty, 2019). This helps us 
understand the relationship between Housing First and an individual?s homelessness, 
but we know very little about the relationship between implementation of a Housing 
First approach and overall homelessness rates in a community. 
In a 2019 survey of homelessness research published by the Journal of 
Housing Economics, Brendan O?Flaherty wrote: 
?What has been missing in studies of Housing First are estimates of 
aggregate impact: does operating a Housing First program actually reduce the total 
amount of homelessness in a community?? 
 
 
Through this study, I sought to understand if Continuums of Care (CoC) that 
have adopted a Housing First approach by dedicating a higher proportion of their 
resources towards permanent housing units are associated with a lower proportion of 
people experiencing homelessness between the years 2009 and 2017 than CoCs 
dedicating a higher proportion of their resources towards emergency shelter and other 
short-term solutions. Additionally, I sought to understand how that relationship 
between the implementation of a Housing First approach and homelessness rates 
change as the values of median rent, unemployment, and other covariates typically 
associated with homelessness rates change. I hypothesized that CoCs adopting a 
Housing First approach, as defined in the context of this study, would experience 
lower homelessness rates. 
The hypothesis that homelessness rates would decrease as the Housing First 
index increases was supported by the results, but the relationship is more complex 
than hypothesized. The relationship between Housing First and homelessness rates 
was quadratic in nature and influenced by an interaction effect with housing tenure. 
Jurisdictions that adopted a Housing First approach generally experienced lower 
homelessness rates, except where a vast majority of households are owner-occupied. 
  
 
 
 
 
 
 
 
 
MODELING THE RELATIONSHIP BETWEEN THE HOUSING FIRST 
APPROACH AND HOMELESSNESS 
 
by 
David Boston 
 
 
Dissertation submitted to the Faculty of the Graduate School of the 
University of Maryland, College Park, in partial fulfillment 
of the requirements for the degree of 
Doctor of Philosophy 
2020 
 
 
 
 
Advisory Committee: 
Dr. Willow Lung-Amam, Chair 
Dr. Casey Dawkins 
Dr. Ariel Bierbaum 
Dr. Gerrit Knaap 
Dr. Christopher Foreman 
 
 
 
 
 
 
 
 
I, David Boston, confirm that the work presented in this dissertation is my own. 
Where information has been derived from other sources, I confirm that this has been 
indicated in the dissertation. 
 
 
 
 
 
 
? Copyright by 
David Boston 
2020 
 
 
 
 
 
Preface 
To paraphrase the ethical principles of the American Planning Association 
(APA), we in the planning community must continuously pursue and faithfully serve 
the public interest by striving to expand choice and opportunity for all persons, 
recognizing a special responsibility to plan for the needs of disadvantaged groups and 
persons (1992). To strive to better understand why people experience homelessness is 
striving to uphold our profession?s ethical principles, because this allows us to better 
plan for some of the most disadvantaged people in our communities. It is my hope 
that the contributions of this study are beyond academic in nature, and that the results 
of this work may benefit practitioners who work in the realm of planning, housing, 
and development, and those who work with people experiencing homelessness.  
 
  
ii 
 
Dedication 
I dedicate this research to my grandmother, Patricia Given, who made me the 
person I am today. You taught me to speak up, lend my voice to the voiceless, strive 
to be clever, and not give up a good fight. 
I also dedicate this research to anyone struggling with homelessness. You are 
seen and this struggle will not define you.  
 
iii 
 
Acknowledgements 
I?d like to thank Willow Lung-Amam for serving as my mentor and the chair 
for my dissertation committee. Your guidance through this long and rigorous process 
made me a better scholar. 
I?d like to thank my dissertation committee for their time and expertise. Your 
reviews and advice throughout the process made this dissertation more valuable. 
I?d also like to thank my employers that have allowed me to work on flexible 
schedules and take leave to support my academic endeavors throughout my Ph.D. 
journey. To all of my professional colleagues---you taught me far more than I would 
have learned in school alone. 
Lastly, I?d like to thank my family and friends for their emotional support. I?d 
like to especially thank my wife, Caitlin Kelly, for being my rock. Caitlin, your 
shoulder to lean on, ear to bend, words of wisdom, and constant encouragement made 
this research possible. Thank you. 
 
iv 
 
Table of Contents 
List of Tables ............................................................................................................. viii 
List of Figures .......................................................................................................... xxiii 
List of Equations ..................................................................................................... xxvii 
List of Abbreviations ............................................................................................. xxviii 
Chapter 1: Introduction ................................................................................................. 1 
1.1: Research Question ......................................................................................................... 2 
1.2: Debate ............................................................................................................................ 2 
1.3: Contribution of this Study ............................................................................................. 4 
Chapter 2: Progress Understanding and Ending Homelessness ................................. 11 
2.1: Housing First ............................................................................................................... 12 
2.1.1: Development of the Housing First approach ......................................................................12 
2.1.2: Debate regarding the efficacy of Housing First .................................................................14 
2.2: Measures of Homelessness .......................................................................................... 17 
2.2.1: 1984 HUD estimate ............................................................................................................17 
2.2.2: 1987 Burt sheltered population survey ...............................................................................18 
2.2.3: Census 1990 S-night enumeration ......................................................................................20 
2.2.4: HUD Continuums of Care ..................................................................................................22 
2.3: Determinants of Homelessness.................................................................................... 25 
2.3.1: Housing affordability ..........................................................................................................30 
2.3.2: Income and poverty ............................................................................................................32 
2.3.3: Unemployment and underemployment ................................................................................34 
2.3.4: Vacancy ..............................................................................................................................35 
2.3.5: Tenure .................................................................................................................................36 
2.3.6: Race and ethnicity ..............................................................................................................37 
2.3.7: Climate ...............................................................................................................................41 
2.3.8: Inclusionary zoning ............................................................................................................42 
2.3.9: Eviction filing rates ............................................................................................................43 
2.3.10: Housing First strategies ...................................................................................................43 
2.4: Prior Models ................................................................................................................ 44 
Chapter 3: Methodology, Data, and Constructing the Model ..................................... 47 
3.1: Study Variables ........................................................................................................... 48 
v 
 
3.2: Creation of a Housing First Index ............................................................................... 51 
3.3: Conducting the Panel Analyses ................................................................................... 53 
3.3.1: Linear mixed models procedure .........................................................................................53 
3.3.2: Methodology for model selection ........................................................................................55 
Chapter 4: Descriptive Statistics ................................................................................. 58 
4.1: Housing First Index Scores ......................................................................................... 58 
4.2: Means Comparison ...................................................................................................... 59 
4.2.1: Means comparison split by homelessness rates ..................................................................60 
4.2.2: Means comparison split by Housing First index scores .....................................................66 
4.3: Distributions ................................................................................................................ 68 
4.4: Spatial Characteristics ................................................................................................. 75 
4.4.1: Homelessness rates across the United States .....................................................................76 
4.4.2: Housing First index scores across the United States ..........................................................78 
Chapter 5: Panel Analysis ........................................................................................... 82 
5.1: Initial Model ................................................................................................................ 82 
5.2: Final Model Results..................................................................................................... 84 
5.2.1: Results of the Housing First index ......................................................................................87 
5.2.2: Other main effects results and interpretation .....................................................................91 
5.2.3: Other interaction effects results and interpretation ............................................................97 
5.2.4: Implications of the final model ...........................................................................................99 
Chapter 6: Conclusion............................................................................................... 100 
6.1: Housing First and Homelessness ............................................................................... 100 
6.2: Policy Recommendations .......................................................................................... 103 
6.3: Limitations of this Study ........................................................................................... 106 
6.4: Scholarly Implications ............................................................................................... 108 
Appendix 1: Data Collection and Model Assembly Process .................................... 113 
Data Collection ................................................................................................................. 113 
Estimates of homelessness rates and CoC data ..........................................................................113 
Census estimates of county or county equivalent data ................................................................115 
Climate data ................................................................................................................................117 
Eviction filing data ......................................................................................................................118 
Inclusionary zoning data .............................................................................................................119 
Data Assembly and Constructing the Model .................................................................... 121 
Assembling county-level data......................................................................................................122 
vi 
 
Converting county-level data to CoC-level data .........................................................................123 
Adding additional CoC-level and point data ..............................................................................124 
Importing study data into SPSS ..................................................................................................126 
Appendix 2: Model Selection Process ...................................................................... 128 
Individual variable regressions ......................................................................................... 128 
Model testing .................................................................................................................... 131 
Appendix 3: Additional Descriptive Statistics .......................................................... 140 
Appendix 4: Individual Variable Regressions .......................................................... 161 
Appendix 5: Model Testing Results ......................................................................... 211 
Appendix 6: Final Model Results for Subsets of Homelessness .............................. 313 
Glossary .................................................................................................................... 323 
References ................................................................................................................. 324 
 
vii 
 
List of Tables 
Table 1: Prior research of homelessness rates and associated variables ..................... 46 
Table 2: Variables used in this model ......................................................................... 50 
Table 3: Housing First index scores............................................................................ 59 
Table 4: Means of variables stratified by observations with above- and below-median 
homelessness rates ...................................................................................................... 60 
Table 5: Means of variables stratified by observations with above- and below-median 
Housing First index scores .......................................................................................... 67 
Table 6: Distribution descriptives for homelessness and the Housing First index ..... 69 
Table 7: Homelessness rate percentiles ...................................................................... 73 
Table 8: Type III tests of fixed effects in the first round of main effects testing ....... 83 
Table 9: Information criteria for final model .............................................................. 85 
Table 10: Type III tests of fixed effects for final model ............................................. 85 
Table 11: Fixed effects estimates for final model ....................................................... 86 
Table 12: Estimated homelessness rates for values of the Housing First index and 
percentage of renter-occupied households .................................................................. 89 
Table 13: Estimated homelessness rates for values of the vacancy rate and median 
gross rent ..................................................................................................................... 98 
Table 14: CoC Mergers, 2009-2017 ......................................................................... 114 
Table 15: Census data source tables ......................................................................... 116 
Table 16: Summary of individual variable regression results .................................. 129 
Table 17: Information criteria for first round of main effects testing ....................... 133 
Table 18: Type III tests of fixed effects in the first round of main effects testing ... 134 
viii 
 
Table 19: Summary of results from the first phase of model testing ........................ 136 
Table 20: Summary of results from the second phase of model testing ................... 138 
Table 21: Descriptive summary of variables ............................................................ 140 
Table 22: Outliers ..................................................................................................... 156 
Table 23: Distribution descriptives for homelessness using coerced values ............ 160 
Table 24: Model fit results regressing homelessness rates by total population ........ 161 
Table 25: ANOVA table regressing homelessness rates by total population ........... 162 
Table 26: Coefficient table regressing homelessness rates by total population ........ 163 
Table 27: Model fit results regressing homelessness rates by percentage of population 
with a bachelor?s degree or higher ............................................................................ 164 
Table 28: ANOVA table regressing homelessness rates by percentage of population 
with a bachelor?s degree or higher ............................................................................ 165 
Table 29: Coefficient table regressing homelessness rates by percentage of population 
with a bachelor?s degree or higher ............................................................................ 166 
Table 30: Model fit results regressing homelessness rates by median household 
income ....................................................................................................................... 167 
Table 31: ANOVA table regressing homelessness rates by median household income
................................................................................................................................... 168 
Table 32: Coefficient table regressing homelessness rates by median household 
income ....................................................................................................................... 168 
Table 33: Model fit results regressing homelessness rates by median gross rent ..... 169 
Table 34: ANOVA table regressing homelessness rates by median gross rent ........ 170 
Table 35: Coefficient table regressing homelessness rates by median gross rent .... 171 
ix 
 
Table 36: Model fit results regressing homelessness rates by median home value .. 172 
Table 37: ANOVA table regressing homelessness rates by median home value ..... 173 
Table 38: Coefficient table regressing homelessness rates by median home value . 174 
Table 39: Model fit results regressing homelessness rates by percentage of renters in a 
family with children .................................................................................................. 175 
Table 40: ANOVA table regressing homelessness rates by percentage of renters in a 
family with children .................................................................................................. 176 
Table 41: Coefficient table regressing homelessness rates by percentage of renters in 
a family with children ............................................................................................... 176 
Table 42: Model fit results regressing homelessness rates by percentage of renter-
occupied housing units .............................................................................................. 177 
Table 43: ANOVA table regressing homelessness rates by percentage of renter-
occupied housing units .............................................................................................. 178 
Table 44: Coefficient table regressing homelessness rates by percentage of renter-
occupied housing units .............................................................................................. 178 
Table 45: Model fit results regressing homelessness rates by percentage of renters 
identifying as white, non-Hispanic ........................................................................... 179 
Table 46: ANOVA table regressing homelessness rates by percentage of renters 
identifying as white, non-Hispanic ........................................................................... 180 
Table 47: Coefficient table regressing homelessness rates by percentage of renters 
identifying as white, non-Hispanic ........................................................................... 181 
Table 48: Model fit results regressing homelessness rates by percentage of renters 
without any college education .................................................................................. 182 
x 
 
Table 49: ANOVA table regressing homelessness rates by percentage of renters 
without any college education .................................................................................. 183 
Table 50: Coefficient table regressing homelessness rates by percentage of renters 
without any college education .................................................................................. 184 
Table 51: Model fit results regressing homelessness rates by unemployment rate .. 185 
Table 52: ANOVA table regressing homelessness rates by unemployment rate ..... 186 
Table 53: Coefficient table regressing homelessness rates by unemployment rate .. 186 
Table 54: Model fit results regressing homelessness rates by poverty rate .............. 187 
Table 55: ANOVA table regressing homelessness rates by poverty rate ................. 188 
Table 56: Coefficient table regressing homelessness rates by poverty rate .............. 189 
Table 57: Model fit results regressing homelessness rates by eviction filing rate .... 190 
Table 58: ANOVA table regressing homelessness rates by eviction filing rate ....... 191 
Table 59: Coefficient table regressing homelessness rates by eviction filing rate ... 192 
Table 60: Model fit results regressing homelessness rates by percentage of rent-
burdened households ................................................................................................. 193 
Table 61: ANOVA table regressing homelessness rates by percentage of rent-
burdened households ................................................................................................. 194 
Table 62: Coefficients table regressing homelessness rates by percentage of rent-
burdened households ................................................................................................. 194 
Table 63: Model fit results regressing homelessness rates by Gini index ................ 195 
Table 64: ANOVA table regressing homelessness rates by Gini index ................... 196 
Table 65: Coefficient table regressing homelessness rates by Gini index ................ 196 
Table 66: Model fit results regressing homelessness rates by vacancy rate ............. 197 
xi 
 
Table 67: ANOVA table regressing homelessness rates by vacancy rate ................ 198 
Table 68: Coefficients table regressing homelessness rates by vacancy rate ........... 199 
Table 69: Model fit summary regressing homelessness rates by mean temperature 200 
Table 70: ANOVA table regressing homelessness rates by mean temperature ........ 201 
Table 71: Coefficients table regressing homelessness rates by mean temperature .. 202 
Table 72: Model fit results regressing homelessness rates by total precipitation ..... 203 
Table 73: ANOVA table regressing homelessness rates by total precipitation ........ 204 
Table 74: Coefficients table regressing homelessness rates by total precipitation ... 205 
Table 75: Model fit results regressing homelessness rates by Housing First index . 206 
Table 76: ANOVA table regressing homelessness rates by Housing First index ..... 207 
Table 77: Coefficients table regressing homelessness rates by Housing First index 207 
Table 78: Model fit results regressing homelessness rates by HUD CoC funding ... 208 
Table 79: ANOVA table regressing homelessness rates by HUD CoC funding ...... 209 
Table 80: Coefficients table regressing homelessness rates by HUD CoC funding . 210 
Table 81: Autoregressive residual covariance matrix for yearcoded repeating variable
................................................................................................................................... 211 
Table 82: Information criteria for Model 1 measuring main effects of all variables 211 
Table 83: Type III tests of fixed effects for Model 1 measuring main effects of all 
variables .................................................................................................................... 212 
Table 84: Fixed effects estimates for Model 1 measuring main effects of all variables
................................................................................................................................... 213 
Table 85: Information criteria for Model 2 removing gini ....................................... 214 
Table 86: Type III tests of fixed effects for Model 2 removing gini ........................ 214 
xii 
 
Table 87: Fixed effects estimates for Model 2 removing gini .................................. 215 
Table 88: Information criteria for Model 3 adding inczon * gini interaction ........... 217 
Table 89: Type III tests of fixed effects for Model 3 adding inczon * gini interaction
................................................................................................................................... 217 
Table 90: Fixed effects estimates for Model 3 adding inczon * gini interaction ...... 218 
Table 91: Information criteria for Model 4 adding coccat * gini interaction ........... 220 
Table 92: Type III tests of fixed effects for Model 4 adding coccat * gini interaction
................................................................................................................................... 220 
Table 93: Fixed effects estimates for Model 4 adding coccat * gini interaction ...... 221 
Table 94: Information criteria for Model 21 adding gini * hf interaction ................ 223 
Table 95: Type III tests of fixed effects for Model 21 adding gini * hf interaction . 223 
Table 96: Fixed effects estimates for Model 21 adding gini * hf interaction ........... 224 
Table 97: Information criteria for Model 24 removing evic ..................................... 226 
Table 98: Type III tests of fixed effects for Model 24 removing evic ...................... 226 
Table 99: Fixed effects estimates for Model 24 removing evic ................................ 227 
Table 100: Information criteria for Model 45 removing renocc ............................... 228 
Table 101: Type III tests of fixed effects for Model 45 removing renocc ................ 228 
Table 102: Fixed effects estimates for Model 45 removing renocc .......................... 229 
Table 103: Information criteria for Model 50 adding renocc * medinc interaction . 230 
Table 104: Type III tests of fixed effects for Model 50 adding renocc * medinc 
interaction ................................................................................................................. 230 
Table 105: Fixed effects estimates for Model 50 adding renocc & medinc interaction
................................................................................................................................... 231 
xiii 
 
Table 106: Information criteria for Model 62 adding renocc * hf interaction .......... 232 
Table 107: Type III tests of fixed effects for Model 62 adding renocc * hf interaction
................................................................................................................................... 232 
Table 108: Fixed effects estimates for Model 62 adding renocc * hf interaction..... 233 
Table 109: Information criteria for Model 65 removing renfam .............................. 235 
Table 110: Type III tests of fixed effects for Model 65 removing renfam ............... 235 
Table 111: Fixed effects estimates for Model 65 removing renfam ......................... 236 
Table 112: Information criteria for Model 84 removing precip ............................... 237 
Table 113: Type III tests of fixed effects for Model 84 removing precip ................ 237 
Table 114: Fixed effects estimates for Model 84 removing precip .......................... 238 
Table 115: Information criteria for Model 102 removing unemp............................. 239 
Table 116: Type III tests of fixed effects for Model 102 removing unemp ............. 239 
Table 117: Fixed effects estimates for Model 102 removing unemp ....................... 240 
Table 118: Information criteria for Model 119 removing medinc ............................ 241 
Table 119: Type III tests of fixed effects for Model 119 removing medinc ............. 241 
Table 120: Fixed effects estimates for Model 119 removing medinc ...................... 242 
Table 121: Information criteria for Model 122 adding pop * medinc interaction .... 243 
Table 122: Type III tests of fixed effects for Model 122 adding pop * medinc 
interaction ................................................................................................................. 243 
Table 123: Fixed effects estimates for Model 122 adding pop * medinc interaction 244 
Table 124: Information criteria for Model 124 adding medinc * medren interaction
................................................................................................................................... 245 
xiv 
 
Table 125: Type III tests of fixed effects for Model 124 adding medinc * medren 
interaction ................................................................................................................. 245 
Table 126: Fixed effects estimates for Model 124 adding medinc * medren interaction
................................................................................................................................... 246 
Table 127: Information criteria for Model 125 adding medinc * medval interaction
................................................................................................................................... 247 
Table 128: Type III tests of fixed effects for Model 125 adding medinc * medval 
interaction ................................................................................................................. 247 
Table 129: Fixed effects estimates for Model 125 adding medinc * medval interaction
................................................................................................................................... 248 
Table 130: Information criteria for Model 126 adding medinc * renwhi interaction 249 
Table 131: Type III tests of fixed effects for Model 126 adding medinc * renwhi 
interaction ................................................................................................................. 249 
Table 132: Fixed effects estimates for Model 126 adding medinc * renwhi interaction
................................................................................................................................... 250 
Table 133: Information criteria for Model 127 adding medinc * renedu interaction 251 
Table 134: Type III tests of fixed effects for Model 127 adding medinc * renedu 
interaction ................................................................................................................. 251 
Table 135: Fixed effects estimates for Model 127 adding medinc * renedu interaction
................................................................................................................................... 252 
Table 136: Information criteria for Model 128 adding medinc * pov interaction .... 253 
Table 137: Type III tests of fixed effects for Model 128 adding medinc * pov 
interaction ................................................................................................................. 253 
xv 
 
Table 138: Fixed effects estimates for Model 128 adding medinc * pov interaction 254 
Table 139: Information criteria for Model 129 adding medinc * burd interaction ... 255 
Table 140: Type III tests of fixed effects for Model 129 adding medinc * burd 
interaction ................................................................................................................. 255 
Table 141: Fixed effects estimates for Model 129 adding medinc * burd interaction
................................................................................................................................... 256 
Table 142: Information criteria for Model 130 adding medinc * vac interaction .... 257 
Table 143: Type III tests of fixed effects for Model 130 adding medinc * vac 
interaction ................................................................................................................. 257 
Table 144: Fixed effects estimates for Model 130 adding medinc * vac interaction 258 
Table 145: Information criteria for Model 132 adding medinc * hf interaction ....... 259 
Table 146: Type III tests of fixed effects for Model 132 adding medinc * hf 
interaction ................................................................................................................. 259 
Table 147: Fixed effects estimates for Model 132 adding medinc * hf interaction . 260 
Table 148: Information criteria for Model 133 adding medinc * fund interaction ... 261 
Table 149: Type III tests of fixed effects for Model 133 adding medinc * fund 
interaction ................................................................................................................. 261 
Table 150: Fixed effects estimates for Model 133 adding medinc * fund interaction
................................................................................................................................... 262 
Table 151: Information criteria for Model 134 adding medinc * yearcoded interaction
................................................................................................................................... 263 
Table 152: Type III tests of fixed effects for Model 134 adding medinc * yearcoded 
interaction ................................................................................................................. 263 
xvi 
 
Table 153: Fixed effects estimates for Model 134 adding medinc * yearcoded 
interaction ................................................................................................................. 264 
Table 154: Information criteria for Model 135 removing burd ................................ 265 
Table 155: Type III tests of fixed effects for Model 135 removing burd ................. 265 
Table 156: Fixed effects estimates for Model 135 removing burd ........................... 266 
Table 157: Information criteria for Model 141 adding medval * burd interaction ... 267 
Table 158: Type III tests of fixed effects adding medval * burd interaction ............ 267 
Table 159: Fixed effects estimates for Model 141 adding medval * burd interaction
................................................................................................................................... 268 
Table 160: Information criteria for Model 150 removing medren ........................... 269 
Table 161: Type III tests of fixed effects for Model 150 removing medren ............ 269 
Table 162: Fixed effects estimates for Model 150 removing medren ...................... 270 
Table 163: Information criteria for Model 155 adding medren * medval interaction
................................................................................................................................... 271 
Table 164: Type III tests of fixed effects for Model 155 adding medren * medval 
interaction ................................................................................................................. 271 
Table 165: Fixed effects estimates for Model 155 adding medren * medval interaction
................................................................................................................................... 272 
Table 166: Information criteria for Model 156 adding medren * renwhi interaction273 
Table 167: Type III tests of fixed effects for Model 156 adding medren * renwhi 
interaction ................................................................................................................. 273 
Table 168: Fixed effects estimates for Model 156 adding medren * renwhi interaction
................................................................................................................................... 274 
xvii 
 
Table 169: Information criteria for Model 158 adding medren * pov interaction .... 275 
Table 170: Type III tests of fixed effects for Model 158 adding medren * pov 
interaction ................................................................................................................. 275 
Table 171: Fixed effects estimates for Model 158 adding medren * pov interaction276 
Table 172: Information criteria for Model 159 adding medren * vac interaction .... 277 
Table 173: Type III tests of fixed effects for Model 159 adding medren * vac 
interaction ................................................................................................................. 277 
Table 174: Fixed effects estimates for Model 159 adding medren * vac interaction 278 
Table 175: Information criteria for Model 164 removing renwhi ............................ 279 
Table 176: Type III tests of fixed effects for Model 164 removing renwhi ............. 279 
Table 177: Fixed effects estimates for Model 164 removing renwhi ....................... 280 
Table 178: Information criteria for Model 167 adding pop * renwhi interaction ..... 281 
Table 179: Type III tests of fixed effects for Model 167 adding pop * renwhi 
interaction ................................................................................................................. 281 
Table 180: Fixed effects estimates for Model 167 adding pop * renwhi interaction 282 
Table 181: Information criteria for Model 168 adding bach * renwhi interaction ... 283 
Table 182: Type III tests of fixed effects for Model 168 adding bach * renwhi 
interaction ................................................................................................................. 283 
Table 183: Fixed effects estimates for Model 168 adding bach * renwhi interaction
................................................................................................................................... 284 
Table 184: Information criteria for Model 169 adding medval * renwhi interaction 285 
Table 185: Type III tests of fixed effects for Model 169 adding medval * renwhi 
interaction ................................................................................................................. 285 
xviii 
 
Table 186: Fixed effects estimates for Model 169 adding medval * renwhi interaction
................................................................................................................................... 286 
Table 187: Information criteria for Model 173 adding renwhi * temp interaction ... 287 
Table 188: Type III tests of fixed effects for Model 173 adding renwhi * temp 
interaction ................................................................................................................. 287 
Table 189: Fixed effects estimates for Model 173 adding renwhi * temp interaction
................................................................................................................................... 288 
Table 190: Information criteria for Model 174 adding renwhi * hf interaction ....... 289 
Table 191: Type III tests of fixed effects for Model 174 adding renwhi * hf 
interaction ................................................................................................................. 289 
Table 192: Fixed effects estimates for Model 174 adding renwhi * hf interaction .. 290 
Table 193: Information criteria for Model 177 reintroducing gini ........................... 291 
Table 194: Type III tests of fixed effects for Model 177 reintroducing gini ............ 291 
Table 195: Fixed effects estimates for Model 177 reintroducing gini ...................... 292 
Table 196: Information criteria for Model 180 reintroducing inczon * gini interaction
................................................................................................................................... 293 
Table 197: Type III tests of fixed effects for Model 180 reintroducing inczon * gini 
interaction ................................................................................................................. 293 
Table 198: Fixed effects estimates for Model 180 reintroducing inczon * gini 
interaction ................................................................................................................. 294 
Table 199: Information criteria for Model 181 reintroducing coccat * gini interaction
................................................................................................................................... 295 
xix 
 
Table 200: Type III tests of fixed effects for Model 181 reintroducing coccat * gini 
interaction ................................................................................................................. 295 
Table 201: Fixed effects estimates for Model 181 reintroducing coccat * gini 
interaction ................................................................................................................. 296 
Table 202: Information criteria for Model 182 reintroducing hf * gini interaction.. 297 
Table 203: Type III tests of fixed effects for Model 182 reintroducing hf * gini 
interaction ................................................................................................................. 297 
Table 204: Fixed effects estimates for Model 182 reintroducing hf * gini interaction
................................................................................................................................... 298 
Table 205: Information criteria for Model 201 adding medval * pov interaction .... 299 
Table 206: Type III tests of fixed effects for Model 201 adding medval * pov 
interaction ................................................................................................................. 299 
Table 207: Fixed effects estimates for Model 201 adding medval * pov interaction 300 
Table 208: Information criteria for Model 215 adding pov * hf interaction ............ 301 
Table 209: Type III tests of fixed effects for Model 215 adding pov * hf interaction
................................................................................................................................... 301 
Table 210: Fixed effects estimates for Model 215 adding pov * hf interaction ....... 302 
Table 211: Information criteria for Model 221 adding vac * yearcoded interaction 304 
Table 212: Type III tests of fixed effects for Model 221 adding vac * yearcoded 
interaction ................................................................................................................. 304 
Table 213: Fixed effects estimates for Model 221 adding vac * yearcoded interaction
................................................................................................................................... 305 
xx 
 
Table 214: Information criteria for Model 224 adding temp * yearcoded interaction
................................................................................................................................... 307 
Table 215: Type III tests of fixed effects for Model 224 adding temp * yearcoded 
interaction ................................................................................................................. 307 
Table 216: Fixed effects estimates for Model 224 adding temp * yearcoded 
interaction ................................................................................................................. 308 
Table 217: Information criteria for Model 225 adding hf * fund interaction ........... 310 
Table 218: Type III tests of fixed effects for Model 225 adding hf * fund interaction
................................................................................................................................... 310 
Table 219: Fixed effects estimates for Model 225 adding hf * fund interaction ...... 311 
Table 220: Information criterion for sheltered homelessness model ........................ 313 
Table 221: Type III tests of fixed effects for sheltered homelessness model ........... 313 
Table 222: Fixed effects estimates for sheltered homelessness model ..................... 314 
Table 223: Information criterion for unsheltered homelessness model .................... 315 
Table 224: Type III tests of fixed effects for unsheltered homelessness model ....... 315 
Table 225: Fixed effects estimates for unsheltered homelessness model ................. 316 
Table 226: Information criteria for family homelessness model .............................. 317 
Table 227: Type III tests of fixed effects for family homelessness model ............... 317 
Table 228: Fixed effects estimates for family homelessness model ......................... 318 
Table 229: Information criteria for chronic homelessness model ............................ 319 
Table 230: Type III tests of fixed effects for chronic homelessness model ............. 319 
Table 231: Fixed effects estimates for chronic homelessness model ....................... 320 
Table 232: Information criteria for veteran homelessness model ............................. 321 
xxi 
 
Table 233: Type III tests of fixed effects for veteran homelessness ......................... 321 
Table 234: Fixed effects estimates for veteran homelessness model ....................... 322 
xxii 
 
List of Figures 
Figure 1: Bohanon's economic theory of homelessness ............................................. 28 
Figure 2: Total precipitation by mean temperature linear regression ......................... 65 
Figure 3: Distribution of homelessness rate observations below the median 
homelessness rate ........................................................................................................ 71 
Figure 4: Distribution of homelessness rate observations above the median 
homelessness rate ........................................................................................................ 71 
Figure 5: Box plot of homelessness rate cases............................................................ 72 
Figure 6: Box plot of homelessness rate cases using coerced values ......................... 74 
Figure 7: Distribution of homelessness rate observations above the median 
homelessness rate using coerced values ..................................................................... 75 
Figure 8: Map of average people per 1,000 experiencing homelessness by CoC: 2009 
to 2017 ........................................................................................................................ 76 
Figure 9: Map of change in people per 1,000 experiencing homelessness by CoC from 
2009 to 2017 ............................................................................................................... 77 
Figure 10: Map of average Housing First index scores by CoC: 2009 to 2017.......... 79 
Figure 11: Map of change in Housing First index scores by CoC from 2009 to 2017 80 
Figure 12: Quadratic relationship between homelessness rates and the Housing First 
index .......................................................................................................................... 130 
Figure 13: Distribution histogram for variable: bach ............................................... 142 
Figure 14: Distribution histogram for variable: medinc ........................................... 142 
Figure 15: Distribution histogram for variable: medren ........................................... 143 
Figure 16: Distribution histogram for variable: medval ........................................... 143 
xxiii 
 
Figure 17: Distribution histogram for variable: aff ................................................... 144 
Figure 18: Distribution histogram for variable: renocc ............................................ 144 
Figure 19: Distribution histogram for variable: renfam ............................................ 145 
Figure 20: Distribution histogram for variable: renwhi ............................................ 145 
Figure 21: Distribution histogram for variable: renedu ............................................ 146 
Figure 22: Distribution histogram for variable: unemp ............................................ 146 
Figure 23: Distribution histogram for variable: pov ................................................. 147 
Figure 24: Distribution histogram for variable: evic ................................................ 147 
Figure 25: Distribution histogram for variable: burd ................................................ 148 
Figure 26: Distribution histogram for variable: gini ................................................. 148 
Figure 27: Distribution histogram for variable: vac ................................................. 149 
Figure 28: Distribution histogram for variable: temp ............................................... 149 
Figure 29: Distribution histogram for variable: precip ............................................. 150 
Figure 30: Distribution histogram for variable: coccat ............................................. 150 
Figure 31: Distribution histogram for variable: inczon ............................................ 151 
Figure 32: Distribution histogram for variable: hf .................................................... 151 
Figure 33: Distribution histogram for variable: fund ................................................ 152 
Figure 34: Distribution histogram for variable: hl .................................................... 152 
Figure 35: Distribution histogram for variable: hls .................................................. 153 
Figure 36: Distribution histogram for variable: hlu .................................................. 153 
Figure 37: Distribution histogram for variable: hlf................................................... 154 
Figure 38: Distribution histogram for variable: hlc .................................................. 154 
Figure 39: Distribution histogram for variable: hlv .................................................. 155 
xxiv 
 
Figure 40: Distribution histogram for variable: hlco ................................................ 155 
Figure 41: Homelessness rates by total population quadratic regression ................. 161 
Figure 42: Homelessness rates by percentage of population with a bachelor?s degree 
or higher cubic and linear regressions ...................................................................... 164 
Figure 43: Homelessness rates by median household income quadratic regression 167 
Figure 44: Homelessness rates by median gross rent linear regression .................... 169 
Figure 45: Homelessness rates by median home value linear regression ................. 172 
Figure 46: Homelessness rates by percentage of renters in a family with children 
quadratic regression .................................................................................................. 175 
Figure 47: Homelessness rates by percentage of renter-occupied housing units linear 
regression .................................................................................................................. 177 
Figure 48: Homelessness rates by percentage of renters identifying as white, non-
Hispanic linear regression ......................................................................................... 179 
Figure 49: Homelessness rates by percentage of renters without any college education 
cubic and linear regressions ...................................................................................... 182 
Figure 50: Homelessness rates by unemployment rate linear regression ................. 185 
Figure 51: Homelessness rates by poverty rate linear regression ............................. 187 
Figure 52: Homelessness rates by eviction filing rate quadratic regression ............. 190 
Figure 53: Homelessness rates by percentage of rent-burdened households linear 
regression .................................................................................................................. 193 
Figure 54: Homelessness rates by Gini Index linear regression ............................... 195 
Figure 55: Homelessness rates by vacancy rate linear regression ............................ 197 
Figure 56: Homelessness rates by mean temperature in January linear regression .. 200 
xxv 
 
Figure 57: Homelessness rates by total precipitation in January linear regression .. 203 
Figure 58: Homelessness rates by Housing First index quadratic regression ........... 206 
Figure 59: Homelessness rates by CoC funding in the previous year linear regression
................................................................................................................................... 208 
  
xxvi 
 
List of Equations 
Equation 1: Housing First index ................................................................................... 9 
Equation 2: Housing First index ................................................................................. 51 
Equation 3: Inner and outer fences ............................................................................. 73 
Equation 4: Estimating homelessness rates by the Housing First index and its 
interaction with the percentage of renter-occupied households .................................. 88 
Equation 5: Estimating homelessness rates by the vacancy rate and its interaction with 
the median gross rent of renter-occupied households ................................................. 98 
  
xxvii 
 
List of Abbreviations 
ACS: American Community Survey 
AFDC: Aid to Families with Dependent Children 
AIC: Akaike?s Information Criterion 
ANOVA: analysis of variance 
APA: American Planning Association 
CoC: Continuum of Care 
ESRI: Environmental Systems Research Institute 
FIPS: Federal Information Processing Standard 
GIS: geographic information system 
GLM: general linear model 
HAP: Homeless Assistance Program 
HEARTH: Homeless Emergency and Rapid Transition to Housing Program 
HF: Housing First 
HIC: Housing Inventory Count 
HMIS: Homelessness Management Information System 
HPRP: Homelessness Prevention and Rapid Re-housing Program 
HUD: United States Department of Housing and Urban Development 
IQR: interquartile range 
IRS: Internal Revenue Service 
LMM: linear mixed models 
nClimDiv: National Climate Divisional Database 
NCSHA: National Council of State Housing Agencies 
xxviii 
 
NGO: nongovernmental organization 
NOAA: National Oceanic and Atmospheric Administration 
NOFA: Super Notice of Fund Availability 
NYHS: New York Housing Study 
PIT: Point-in-Time 
RCT: randomized controlled trial 
REML: restricted maximum likelihood 
RM ANOVA: repeated measures analysis of variance 
RRHD: Rapid Re-housing for Homeless Families Demonstration 
S+C: Shelter Plus Care Program 
SAMHSA: Substance Abuse and Mental Health Services Administration 
SHP: Supportive Housing Program 
SPSS: Statistical Package for the Social Sciences 
SRO: single-room occupancy 
SSI: Social Security Income 
TF: treatment first 
TIGER: Topologically Integrated Geographic Encoding and Referencing 
USICH: United States Interagency Council on Homelessness 
VARCOMP: variance components 
xxix 
 
Chapter 1: Introduction 
The Housing First (HF) model of addressing homelessness was developed by 
Sam Tsemberis, a psychologist in New York, who started a program to move people 
experiencing homelessness with severe disabling conditions directly into permanent 
housing with wrap-around services instead of a shelter or hospital ward (Padgett, 
Henwood, & Tsemberis, 2016). Tsemberis and other researchers have studied the 
results of the HF approach versus traditional Treatment First (TF) approaches, in 
which participants must graduate from shelter to transitional housing and then 
eventually to housing of their own, in a series of randomized controlled trials (RCT) 
studying individuals? ability to remaining housed in each program. A review of 12 
such RCTs found that in 11 of the 12 RCTs, Housing First produced greater housing 
retention than the TF approach (Kertesz & Johnson, 2017).  
This growing body of evidence from individual-level studies demonstrating 
that the Housing First approach is effective at keeping those experiencing 
homelessness in stable housing has led to the approach being championed by many 
leading experts, especially as a way to address chronic homelessness (O'Flaherty, 
2019). Beginning in 2013, the United States Department of Housing and Urban 
Development began encouraging local agencies applying for federal funds to 
demonstrate that they plan to address homelessness in their communities using a 
Housing First approach (HUD, 2013). However, not all scholars or service providers 
are convinced that the Housing First model is the best way to reduce homelessness. 
This study seeks to determine the efficacy of the Housing First approach in reducing 
1 
 
homelessness rates by answering the following research question to help move us 
closer to resolving this debate. 
 
1.1: Research Question 
Are Continuums of Care (CoC) that have adopted a Housing First approach by 
dedicating a higher proportion of their resources towards permanent housing and 
support services associated with a lower proportion of people experiencing 
homelessness between the years 2009 and 2017 than CoCs that dedicate a higher 
proportion of their resources towards emergency shelter and other short-term 
solutions? Additionally, how does that relationship between the implementation of a 
Housing First approach and homelessness rates change as the values of median rent, 
unemployment, and other covariates typically associated with homelessness rates 
change? 
 
1.2: Debate 
In an examination of Housing First research conducted in the United States 
and Australia published by the Australian Economic Review, Kertesz and Johnson 
(2017) found that despite credible housing outcomes in longitudinal studies, some 
claims made on behalf of the Housing First approach remained controversial. Some 
of those controversies include how effective supportive services provided as part of a 
Housing First approach are for treating poor physical and mental health, or how cost-
2 
 
effective the Housing First approach is in comparison to the traditional Treatment 
First approach (Kertesz & Johnson, 2017).  
However, the question of whether Housing First is an effective approach to 
ending homelessness at an individual level is largely settled. Many researchers 
conducting longitudinal studies comparing housing retention rates in Housing First 
programs versus Treatment First programs have found housing retention rates to be 
higher among participants of a Housing First program. Mares and Rosenheck (2007), 
Pearson, Montgomery, and Locke (2009), and Stergiopoulos et al. (2015) conducted 
longitudinal studies including multiple cities and found housing retention rates to be 
higher among Housing First participants in each city. Tsemberis, Kent, and Respress 
(2012), Stefancic et al. (2013), and Collins, Malone, and Clifasefi (2013) all found 
housing retention rates to be higher among participants in Housing First programs in 
single jurisdictions as well. 
The debate on Housing First as a solution to homelessness has been light, and 
that?s because this research is addressing a gap in the existing literature that has not 
seen much attention as opposed to answering a question that is central to a 
longstanding debate. Most longitudinal studies of the effectiveness of the Housing 
First approach have consistently shown that people experiencing homelessness who 
participate in a Housing First program are much more likely to retain their housing 
than a person experiencing homelessness participating in a program that follows a 
traditional treatment-first approach, so there isn?t much room for debate there. A 
small number of opponents look at the issue from a different perspective. Benston 
(2015) points out that problems with the existing body of research, such as varying 
3 
 
measurements of attrition, lack of detail on housing conditions and supportive 
services, selection bias among both participants and administrators, and lack of 
standardized program models and definitions limit internal validity, the ability to 
generalize findings, and efforts to replicate research conditions. Essentially, Benston 
(2015) and Kertesz and Johnson (2017) argue that the results of existing studies are 
unique to their environments. 
 
1.3: Contribution of this Study 
However, just because this research isn?t answering a longstanding question in 
the field subject to fierce debate does not mean that it is unimportant. HUD is pushing 
implementation of the Housing First approach across the country by giving some 
level of funding preference to Continuums of Care, or CoCs, that demonstrate a 
commitment to Housing First. And to some degree, it?s working. I found in this 
research that CoCs are slowly moving towards a Housing First model to a greater 
degree each year. 
I am a member of the CoC where I live. The CoC?s resources for combatting 
homelessness are scarce and some service agency representatives are concerned that 
spending money on permanent supportive housing and rapid rehousing to implement 
a Housing First approach will help a very small fraction of people compared to the 
number of people who could be helped if the money was spread across less expensive 
options. A longitudinal study does not take money into account. It does not take 
overall homelessness levels into account. Some people worry that a Housing First 
approach means that a few people are helped very effectively at the expense of 
4 
 
homelessness rising in the CoC as a result of the lack of other services the CoC could 
not afford. The research community in this field have a moral imperative to develop 
an evidence-based approach to ending homelessness, and researchers must look at the 
relationship between Housing First and homelessness rates to develop that approach. 
In a 2019 survey of homelessness research published by the Journal of 
Housing Economics focusing primarily, but not exclusively, on research conducted in 
the United States, Brendan O?Flaherty wrote:  
?What has been missing in studies of Housing First are estimates of 
aggregate impact: does operating a Housing First program actually reduce the total 
amount of homelessness in a community?? 
This study will address this knowledge gap in the field of homelessness 
research by investigating the relationship between implementation of a Housing First 
approach at the Continuum of Care (CoC) level and homelessness rates in 
communities across the United States using a linear mixed models panel analysis. 
Past research on the efficacy of the Housing First approach has consisted of 
longitudinal studies of specific groups of participants experiencing chronic 
homelessness in a particular place or handful of places to compare housing retention 
rates of those in a Housing First program to those enrolled in a Treatment First 
program1. These studies of Housing First have not looked at the relationship between 
Housing First and homelessness rates or studied homelessness nationwide using CoC 
boundaries or data. On the other hand, models studying the relationship between 
 
1 Longitudinal studies of this nature have been conducted by Tsemberis and Eisenberg (2000); 
Culhane, Metraux, and Hadley (2002); Gulcur, Stefancic, Shinn, Tsemberis, and Fischer (2003); 
Tsemberis, Gulcur, and Nakae (2004); Mares and Rosenheck (2007); Stefancic and Tsemberis (2007); 
Culhane and Metraux (2008); Pearson et al. (2009); Sadowski, Kee, VanderWeele, and Buchanan 
(2009); Tsemberis (2010); Stefancic et al. (2013); Stergiopoulos et al. (2015); and Aubry et al. (2016). 
5 
 
homelessness and the variables associated with homelessness using CoC data and 
boundaries have not included a measurement of Housing First or conducted a panel 
analysis to account for interaction effects and within-subject correlations over time. 
This study is an attempt to bridge the gap between longitudinal research on Housing 
First and modeling research on the determinants of homelessness, and to move each 
body of research a couple steps forward by incorporating new variables and methods. 
By studying the relationship between Housing First and homelessness, this 
research will address some, though admittedly not all, of the criticisms levied against 
past longitudinal research. Since homelessness and the implementation of Housing 
First is done at the CoC level instead of the individual household or even program 
level, the level of detail of a longitudinal study is lost. Validity remains similarly 
limited by the quality of available data, because each local CoC is responsible for 
conducting their own point-in-time (PIT) count of people experiencing homelessness, 
and the people conducting these PIT counts are often untrained volunteers. This 
method does obviously eliminate the risk of selection bias, because a randomized 
controlled trial is not being conducted. This study also uses a standard definition of 
Housing First and homelessness across all CoCs, so this type of method gives future 
researchers the ability to replicate research conditions and generalize results very 
easily. 
By studying variables associated with homelessness using a linear mixed 
models panel analysis, the analysis is able to analyze interaction effects and control 
for within-subject correlations over time. Interaction effects look at how the 
relationship between a primary independent variable and a dependent variable change 
6 
 
in association with changes in a control or interaction variable. Past models studying 
variables associated with homelessness rates have been multivariate regression 
analyses only capable of measuring direct effects of variables while controlling for 
others, so incorporating interaction effects helps one to develop a more 
comprehensive and nuanced understanding of how these factors are associated with 
homelessness rates. The inclusion of these interaction effects turned out to be crucial, 
because the relationship between Housing First and homelessness rates was shown to 
be much more nuanced and complex than I had hypothesized. 
The other benefit associated with panel analyses is that they control for 
within-subject correlations over time, whereas multivariate regression models like 
those used in past analyses assume that the error terms are not correlated across 
observations. When one has multiple observations within the same CoC over the 
course of several years, however, it is very likely that the error terms of those 
observations in the same CoC are correlated with one another. 
This study will also include variables that have not been used in past models 
studying variables associated with homelessness, such as a measure of CoC funding 
rates, inclusionary zoning/housing policies, eviction filing rates, income inequality 
using CoC data, and a measurement of the degree to which a CoC has implemented a 
Housing First approach. The CoCs are the primary bodies responsible for 
coordinating the full range of homelessness services in a geographic area (HUD, 
2018a), including the distribution of HUD funds, so it is important that we develop a 
better understanding of the effectiveness of these funds and the CoCs facilitating their 
use.  
7 
 
To measure the degree to which a CoC has implemented a Housing First 
approach, a Housing First index will be created. Investigating the fidelity of each 
permanent housing program in the country to ensure that it is following a Housing 
First model is not possible, so the scale of this research necessitates an adjustment to 
the definition of Housing First for use in the index. Definitions of Housing First used 
by O'Flaherty (2019) or Katz, Zerger, and Hwang (2017) refer exclusively to the use 
of permanent supportive housing, which is long-term housing provided for 
individuals with disabilities experiencing homelessness or families experiencing 
homelessness in which one member of the household has a disability and supportive 
services that are designed to meet the needs of the program participants.  
This study uses HUD?s (2019b) slightly broader definition of Housing First, 
which includes two components: (1) individuals are rapidly placed and stabilized in 
permanent housing without any preconditions regarding income, work effort, sobriety 
or any other factor, and (2) once in housing, individuals never face requirements to 
participate in services as a condition of retaining their housing. While the Housing 
First approach has been particularly more effective than the Treatment First approach 
in addressing the needs of people experiencing chronic homelessness who require 
permanent supportive housing, this definition is broad enough to include the use of 
rapid re-housing and other permanent housing solutions. As O'Flaherty (2019) 
discussed, an estimate regarding the relationship between Housing First and 
homelessness in a community is missing from the existing literature, so an index by 
which to measure the degree to which Housing First is being implemented in a 
community has not been created before. With no previous examples to build from, 
8 
 
this study is making a first attempt at creating a Housing First index with the hope 
that future research will improve upon this design. In this study, the degree to which a 
CoC is implementing a Housing First approach is calculated using data from the 
Housing Inventory Count (HIC), which is submitted by CoCs to HUD each year. 
Equation 1: Housing First index 
???? + ?? + ?? + ??
??? = 2  
??
 
The Housing First index is equal to the sum of the number of transitional 
housing beds (?) divided by two, the number of rapid re-housing beds (?), the number 
of permanent supportive housing beds (?), and the number of other permanent 
housing beds (?) divided by the total number of beds (?) in the CoC, which also 
includes emergency shelter and safe haven beds. This results in an index score 
between 0 and 1 where a higher value indicates more reliance on permanent housing 
strategies and a stronger alignment with the Housing First approach. This index does 
not measure the fidelity of each permanent housing program with the Housing First 
model, so this index does not really tell me if the permanent housing solutions used in 
the CoC provides people experiencing homelessness with a home without barriers to 
entry. Data regarding the fidelity of each program in the United States to the Housing 
First model do not yet exist and collecting such data would require an incredible 
amount of resources. This index instead attempts to serve as a proxy measurement of 
implementation of the Housing First approach by measuring the degree to which a 
CoC prioritizes permanent housing solutions over transitional or temporary shelter 
solutions. The assumption behind this approach being that CoCs will not be able to 
9 
 
prioritize permanent housing solutions that do not follow a Housing First approach 
without pushing many people back into temporary shelter or onto the streets because 
participants were not able to obtain or remain in permanent housing units with high 
barriers to entry or strict requirements to remain in their homes. Therefore, if a large 
portion of a CoC?s beds are in permanent housing units and the CoC is able to fill 
those beds, this study assumes that a significant number of those permanent housing 
units must be provided through a program following the Housing First model. 
This research will use a study period of 2009 to 2017, because the United 
States Census Bureau?s American Community Survey (ACS) five-year estimate data 
are used for many of the variables in this study. The ACS was started in 2005, so 
ACS five-year estimates are available starting in 2009. At the time of data collection 
and model construction, 2017 data were the latest available for most of the variables 
used in this study. 
 
  
10 
 
Chapter 2: Progress Understanding and Ending Homelessness 
The U.S. Department of Housing and Urban Development (HUD, 2018a) 
generally defines a homeless individual or family as one ?who lacks a fixed, regular, 
and adequate nighttime residence.? The many people who have dedicated their lives 
to the pursuit of understanding and ending homelessness know the implications that 
hide behind a rather sterile definition. Homelessness is a state of being that changes 
the way a person sees and experiences the world. The world becomes a more 
dangerous and uncertain place (Huey, 2010; B. A. Lee & Schreck, 2005), and most of 
the people you interact with are unkind and judgmental (Anderson, Snow, & Cress, 
1994; Roschelle & Kaufman, 2004). Holding on to your possessions is difficult (B. A. 
Lee & Schreck, 2005). Finding food to eat is difficult (Bowen & Irish, 2018). Finding 
a safe and warm place to sleep is difficult (Bao, Whitbeck, & Hoyt, 2000; Huey, 
2010). Many people live without a car, go long periods of time without a shower, and 
survive without a network of friends and family (Meanwell, 2012). 
Despite all of this, many people do get back on their feet. Not only that, but 
homelessness had been steadily declining since 2007 until the number plateaued (and 
slightly increased) in the two years after 2016 (HUD, 2018a). This dissertation seeks 
to help determine what strategies to alleviate the crisis of homelessness were driving 
that reduction in homelessness rates so that we can continue to invest in strategies that 
work in the future.  
 
11 
 
2.1: Housing First  
The Housing First (HF) strategy is to give people experiencing homelessness 
immediate access to housing and support services. The traditional or Treatment First 
(TF) approach to homelessness alleviation is to feed and shelter people while treating 
them for their various addictions, mental illnesses, or other personal characteristics 
deemed to be a barrier to them living in permanent housing before moving them 
through the system. Service providers using the TF approach viewed Housing First 
with great skepticism, and advocates of the HF approach were quick to try to support 
their idea with data. 
2.1.1: Development of the Housing First approach 
Psychologist Sam Tsemberis and others developed the Housing First approach 
and founded Pathways to Housing, Inc. in 1992 based on the idea that attempts to 
treat poor mental health are much more effective when a person has a safe and private 
place to call home. Shortly after the founding of Pathways, a study was conducted 
through a collaboration between Pathways to Housing, New York City?s Human 
Resources Administration, and New York State?s Nathan Kline Institute. Researchers 
compiled data for several thousand people experiencing homelessness who were 
participating in traditional continuum of care programs or the Pathways to Housing 
program over a five-year period. They analyzed rates at which people remained 
sheltered or housed, controlling for differences in client characteristics before 
program entry. In a comparison between traditional programs and the Housing First 
approach, the results showed that 88 percent of Housing First participants remained 
12 
 
housed compared to 47 percent of traditional program participants (Tsemberis & 
Eisenberg, 2000). 
In 1996, the federal Substance Abuse and Mental Health Services 
Administration (SAMHSA) issued a request for proposals for grant funding to study 
mental illness and homelessness. SAMHSA awarded six grants and required that all 
recipients of funding use a common set of outcome measures so that results could be 
compared across the various study areas. Pathways to Housing was awarded one of 
the grants, and was the only program testing HF (Padgett et al., 2016). 
The project began in 1997 was called the New York Housing Study (NYHS). 
The longitudinal study followed participants for two years and lasted for about four 
years total. Participants were recruited between 1997 and 1999 and were required to 
have spent 15 of the last 30 days unsheltered, have a history of homelessness over the 
past six months, and have a psychiatric diagnosis of severe mental illness. About 90 
percent of the 225 people enrolled in the study also struggled with substance abuse 
(Padgett et al., 2016). Of the 225 people enrolled, 99 of them were randomly assigned 
to the HF group and 126 of them were randomly assigned to the control group, which 
was a TF program that provided ?treatment as usual.? People in the HF group were 
immediately placed in a small studio or one-bedroom apartment in an affordable area. 
Participants were required to pay 30 percent of their income, which many times was 
Social Security Income (SSI) benefits, toward their rent. They were also required to 
allow the support services team to visit their apartment on a weekly basis. In the 
control group, participants were placed in a group home, shelter, or single-room 
occupancy (SRO) building with shared sleeping, cooking, and bathing facilities. 
13 
 
Participants were expected to remain drug and alcohol free, stick to curfews, and 
follow other rules typical of a TF program in the hopes that they may ultimately be 
rewarded with a home (Padgett et al., 2016). 
The results of the NYHS showed that participants in the HF group spent 
approximately 80 percent of their time in stable housing compared to 30 percent of 
the TF group participants after two years. The study also had a high participant 
retention rate of 94 percent after 12 months and 87 percent by the conclusion of the 
study, giving researchers a relatively large sample size to analyze for a longitudinal 
study of this length and detail (Tsemberis et al., 2004). The NYHS also found that HF 
group participants spent less time hospitalized for psychiatric problems, and that 
housing people struggling with drug or alcohol abuse problems in a private apartment 
may be more effective at reducing rates of substance abuse than an abstinence 
program in a group setting, where disruptive behaviors are more likely to impinge on 
others (Gulcur et al., 2003). 
2.1.2: Debate regarding the efficacy of Housing First 
After the NYHS, other research efforts attempted to gauge the efficacy of the 
Housing First approach by conducting longitudinal studies on people experiencing 
homelessness and comparing the housing retention rate of people in Housing First 
programs versus the housing retention rate of people in traditional or treatment-first 
programs. The results of these studies have generally provided overwhelming support 
for the Housing First approach. In 2004, the United States Interagency Council on 
Homelessness (USICH) provided grant funding for projects intended to address 
chronic homelessness in eleven cities. Seven of those eleven cities used the HF 
14 
 
model. After 12 months, the housing retention rate among the HF project participants 
in those seven cities was 85 percent (Mares & Rosenheck, 2007). A similar three-city, 
12-month study by HUD achieved an 84 percent housing retention rate (Pearson et 
al., 2009). Studies in Washington, DC (Tsemberis et al., 2012); the State of Vermont 
(Stefancic et al., 2013); and Seattle, Washington (Collins et al., 2013) all found 
similar results with housing retention rates of 84 percent, 85 percent, and 77 percent, 
respectively. A study in four Canadian cities (Vancouver, Winnipeg, Toronto, and 
Montreal) supported American results as well. Among 1,198 participants, the study 
found that people housed using a Housing First program were housed 63 to 77 
percent of the time in the two-year study period, while those in the control group were 
housed only 24 to 39 percent of the time (Stergiopoulos et al., 2015). 
However, some scholars have pointed out that there are scientific limitations 
to the results of longitudinal studies regarding the efficacy of Housing First conducted 
thus far. Benston (2015) argues that researchers have used various methods of 
measuring attrition, making comparisons difficult. In a review of 12 longitudinal 
studies, seven reported attrition as the percentage of participants housed at the end of 
the study, two reported attrition as the percentage of participants completing follow-
up interviews, two reported attrition as the proportion of time spent homeless, and 
one reported days spent homeless. Additionally, all of the studies reporting attrition 
included participants who had dropped out of the study in their statistical calculations 
through methods of inputting missing data on the basis of assumed values, weighting 
adjustments for nonrespondents, and analyzing relationships between baseline scores 
and participant characteristics for those who stayed in the study and those who 
15 
 
dropped out (Benston, 2015). High attrition rates cause a decrease of statistical power 
in the model, and all of these methods of dealing with attrition rely on unstable 
assumptions about chance events that if wrong could lead to inaccurate or misleading 
conclusions (Ribisl et al., 1996).  
A lack of detail on housing conditions and supportive services, selection bias 
among both participants and administrators, and lack of standardized program models 
and definitions limit internal validity, the ability to generalize findings, and efforts to 
replicate research conditions. Essentially, Benston (2015) and Kertesz and Johnson 
(2017) argue that the results of existing studies are unique to their environments. 
Some researchers have also found that using a Housing First approach is cost-
effective compared to other homelessness alleviation strategies. Culhane et al. (2002) 
analyzed administrative data for several thousand people experiencing homelessness 
with severe mental illness in New York City who were placed in housing between 
1989 and 1997 and a control group of people with severe mental illness experiencing 
homelessness who were not placed in housing tracked over the same time period. 
Their findings indicated that the average annual cost of shelter use, hospitalization, 
and incarceration for a person experiencing homelessness with severe mental illness 
was $40,451.2 This number was reduced by an average of $16,281 per year when the 
person was placed in a home, and the average cost of a home was $17,277 (Culhane 
et al., 2002). This did not result in a comprehensive cost-savings, but putting people 
 
2 Their findings also indicated that approximately 10 percent of people experiencing homelessness 
were responsible for 50 percent of service costs (in shelters, hospitals, and jails). This subgroup was 
labeled the ?chronically homeless? (Padgett et al., 2016). 
16 
 
in housing did result in a reduction in homelessness for only an overall $996 per unit 
per year in New York City. 
 
2.2: Measures of Homelessness 
About a decade before the Housing First approach was developed, with the 
rise of visible, chronic homelessness and the ?worst housing crisis since the Great 
Depression? (Appelbaum, Dolny, Dreier, & Gilderbloom, 1991) in the 1980s, 
researchers began attempting to measure the size and composition of the homeless 
population. Since then, four major sources of homelessness data have been used to 
study the homeless population of the United States: (1) a 1984 HUD homelessness 
estimate; (2) the 1987 sheltered population survey conducted by Martha Burt and 
others with the Urban Institute for cities with populations of at least 100,000; (3) the 
1990 Census S-Night (Shelter and Street-Night) Enumeration of people experiencing 
both sheltered and unsheltered homelessness in five cities (New York, Los Angeles, 
Chicago, New Orleans, and Phoenix); and (4) HUD point-in-time (PIT) count data 
gathered in the last week of January from 2007 to the present. 
2.2.1: 1984 HUD estimate 
In 1984, HUD published one of the first nationwide assessments of the 
population experiencing homelessness. HUD determined that there were 
approximately 250,000 to 350,000 people experiencing homelessness throughout the 
country on an average night (HUD, 1984). To arrive at that number, researchers used 
four different methods: (1) estimates from local studies; (2) 500 key informant 
interviews in 60 metropolitan areas; (3) surveys of 184 shelter operators in 60 
17 
 
metropolitan areas; and (4) estimates of ratios of sheltered and unsheltered 
populations (Honig & Filer, 1993).3 
The HUD estimate was provided almost immediately after the visible 
population of people experiencing homelessness exploded during the 1981-1982 
recession (Burt, 1992). Some of the earliest studies that found quantitative evidence 
of a relationship between housing costs and homelessness were based on the 
estimates of the population experiencing homelessness from the 1984 HUD study 
(Bohanon, 1991; Elliot & Krivo, 1991; Honig & Filer, 1993). 
2.2.2: 1987 Burt sheltered population survey 
In a study commissioned by HUD, the Urban Institute surveyed local officials 
in all cities with a population of at least 100,000 in 1986. This sample included 147 
central cities and 35 suburban jurisdictions. The principal investigator, Martha Burt, 
used Comprehensive Homeless Assistance Plans that local officials were required to 
submit to HUD to develop a nationwide list of shelter providers in major cities. Burt 
surveyed nongovernmental organizations (NGOs) and homeless service coordinators 
to identify additional shelter providers in each city (Quigley et al., 2001). The survey 
used a probability sample of service-using homeless individuals and estimated a total 
of 500,000-600,000 on any given night in March 1987 (Burt, 1992; Burt & Cohen, 
1989). 
 
3 The results of this count were used in combination with 1980 decennial census data for some of the 
first analyses of the variables associated with homelessness (Appelbaum et al., 1991; Bohanon, 1991; 
Elliot & Krivo, 1991; Honig & Filer, 1993; Quigley & Portney, 1990), summarized and discussed in 
more detail in sections 1.4 and 1.5 below. 
18 
 
Two obvious methodological problems with using shelter bed capacity to 
study variables associated with homelessness are that not all people experiencing 
homelessness stay in emergency shelters and using shelter bed capacity measures a 
response to homelessness rather than homelessness itself. Although a street-to-shelter 
ratio was used in an attempt to estimate a full count, that ratio is not constant across 
metropolitan areas (Quigley et al., 2001). 
O?Flaherty (1996) brought up two different scenarios in which the results of 
the Burt survey could be misleading. One scenario being that if homeless services are 
normal goods, wealthier cities will allocate more funds for homeless shelters, thus 
introducing a spurious positive correlation between this measure of homelessness and 
mean household income. Alternatively, wealthier areas may devote fewer resources to 
homeless shelters or oppose the opening of shelters through local land-use controls so 
as not to attract the users of such services. If rents are higher in wealthier areas, this 
activity would weaken the relationship between homelessness, as measured by shelter 
capacity, and local rents. 
The results of the Burt survey were released close to the release of the Census 
1990 S-night enumeration results, which was thereafter the preferred source of 
homelessness data among researchers. While the Burt survey was referenced in many 
academic works, data from the Burt survey were not used as the dependent variable 
for homelessness in any major quantitative efforts to determine the variables most 
strongly associated with homelessness. 
19 
 
2.2.3: Census 1990 S-night enumeration 
As part of the 1990 Census, the U.S. Census Bureau conducted a one-night 
count of people experiencing homelessness in urban places with populations of at 
least 50,000. The ?S-night? (street and shelter night) enumeration was conducted on 
March 20?21, 1990, and consisted of three components. From 6 p.m. to midnight, 
enumerators counted everyone sleeping or staying at a predesignated list of shelters 
that was intended to be a complete list of all known shelters (Write & Devine, 1992). 
Between 2 a.m. and 4 a.m., enumerators attempted to count everyone experiencing 
homelessness on the streets at locations designated by local officials as known 
congregating areas. Enumerators counted everyone they saw except for people in 
uniform and people engaging in obvious money-making activities other than 
panhandling. Enumerators did not talk to any of the people they counted. (Early & 
Olsen, 2002). Between 4 a.m. and 8 a.m., enumerators attempted to count all 
individuals exiting pre-designated abandoned buildings (Write & Devine, 1992). 
Researchers have criticized the way the S-night counts were conducted and 
the accuracy of the resulting data. One common criticism is that a point-in-time count 
is not a representation of the number of people experiencing homelessness in a 
particular year, but rather a count of the number of people experiencing homelessness 
on a single day in that particular year. Additionally, enumerators of the S-night count 
were instructed to not enter abandoned buildings, but only to wait outside until 8 a.m. 
and count those who left, therefore anyone experiencing homelessness staying in one 
of those abandoned buildings who left after 8 a.m. were not counted (Grimes & 
Chressanthis, 1997). 
20 
 
The S-night enumeration likely underestimated the number of people 
experiencing homelessness (Grimes & Chressanthis, 1997; Hudson, 1993; Quigley et 
al., 2001). To evaluate the S-night count, the Census Bureau sponsored research in 
five cities in which evaluators were deployed undercover to impersonate people 
experiencing homelessness at the listed street locations for enumerations. Evaluators 
observed the behavior of enumerators to monitor whether they showed up and 
followed directions, estimated the number of people experiencing homelessness at 
listed street locations, and reported whether they themselves had been counted. The 
proportion of counted ?decoys? provides a rough estimate of the degree of 
undercounting at listed street locations across these five cities. The actual degree of 
undercounting is likely higher to the degree that people experiencing homelessness 
were in places other than the listed street and shelter locations used in each city. The 
percentages of decoys explicitly indicating that they had not been counted was 10% 
in New Orleans, 10% in Phoenix, 13% in Los Angeles, 20% in New York, and 25% 
in Chicago (Martin, 1992). 
Evaluations of the enumeration efforts found that enumerators in Chicago 
simply dropped off census forms at the largest shelter, and only people staying in the 
shelter who expressed interest in filling out a form were provided a form by shelter 
staff. Otherwise, people in the largest shelter in Chicago were not counted (Edin, 
1992). Evaluations of the count also revealed that many enumerators either failed to 
visit many of the sites and shelters or did not follow the predetermined protocol in 
counting the number of people at the location (Quigley et al., 2001). In Chicago, the 
police department generated locations for enumerators to count unsheltered people 
21 
 
experiencing homelessness without consulting anyone in the city?s social services 
network (Edin, 1992). Reasons cited for low-quality enumeration in the S-night count 
include poor training (Hopper, 1992), poorly-defined geographies and incorrect 
addresses (Edin, 1992), and enumerator concerns for their personal safety (Quigley et 
al., 2001). 
The Census Bureau estimated a 1990 population of people experiencing 
homelessness of 230,000 based on the S-night enumeration, but the consensus among 
researchers was that the population of people experiencing homelessness in 1990 was 
between 550,000 and 600,000 (Quigley et al., 2001). 
2.2.4: HUD Continuums of Care 
Since 1994, HUD has provided support under the Super Notice of Fund 
Availability (NOFA) program to assist people experiencing homelessness achieve 
self-sufficiency and permanent housing. Eligible counties seeking funding were 
required to submit a ?continuum of care? plan to HUD. These plans justified 
community requests for funding under a variety of federal programs, such as the 
Supportive Housing Program (SHP) and the Shelter Plus Care Program (S+C). HUD 
guidelines for completion of these continuum of care plans encouraged consistency 
among estimate methodologies and data schemas (Quigley et al., 2001; HUD, 1994). 
HUD?s official definition of a CoC plan was the following (HUD, 1999): 
A Continuum of Care Plan is a community plan to organize and deliver 
housing and services to meet the specific needs of people who are homeless as they 
move to stable housing and maximum self-sufficiency. It includes action steps to end 
homelessness and prevent a return to homelessness. 
22 
 
Beginning in 2005, HUD mandated that jurisdictions conduct a point-in-time 
(PIT) count at least once every two years in the last week of January to receive 
federal aid for homelessness programs (Schwartz, 2010). Before CoC plans began 
incorporating PIT counts into the requirements for federal aid, people experiencing 
homelessness were not counted in the decennial Census, the American Community 
Survey, the Current Population Survey, the American Housing Survey, or any other 
national quantitative dataset of the population or households (Schwartz, 2010). 
Division B of the Act to Prevent Mortgage Foreclosures and Enhance 
Mortgage Credit Availability, called the Homeless Emergency Assistance and Rapid 
Transition to Housing Act of 2009 (HEARTH Act), amended the McKinney-Vento 
Homeless Assistance Act and established the Continuum of Care (CoC) Program by 
consolidating and amending the SHP, S+C, & Section 8/SRO programs. The purpose 
of consolidating these programs into the CoC Program was to improve efficiency and 
enhance the response coordination of these programs to better meet the needs of 
homeless individuals and families (HUD, 2011). President Obama signed the 
HEARTH Act into law in 2009, and HUD published the CoC Program Interim Rule 
in 2012 to formally implement the CoC Program.  
The CoC Program is designed to promote community-wide commitment to 
the goal of ending homelessness, quickly re-house individuals and families 
experiencing homelessness, provide those individuals and families with access to 
supportive services and programs to keep them in housing, and to optimize self-
sufficiency among program participants (HUD, 2014a). The CoC program was 
designed to prioritize strategies like permanent supportive housing and rapid re-
23 
 
housing that results from the New York Housing Study (NYHS), the Rapid Re-
housing for Homeless Families Demonstration (RRHD) program, and the 
Homelessness Prevention and Rapid Re-housing Program (HPRP) showed to be 
effective at ending homelessness for program participants (Burt et al., 2016; Finkel et 
al., 2016; Padgett et al., 2016). 
For purposes of distributing federal aid and conducting regional counts, most 
of the nation is split into CoC areas. In almost all cases, a CoC area comprises a 
county or group of counties and CoC boundaries are drawn along county lines. HUD 
encourages coordination and cooperation among jurisdictions to create 
comprehensive packages of services and solutions to homelessness, and CoC 
applications that demonstrate good coordination are more competitive (HUD, 2012). 
Participating in a CoC is voluntary, and jurisdictions that are not interested in 
applying for federal funds are not required to participate in a CoC or conduct PIT 
counts. 
One of the primary responsibilities of the CoC Board4 is to understand the 
extent and nature of homelessness in the geographic area that the CoC services, partly 
by conducting annual or biennial point-in-time (PIT) and annual housing inventory 
counts (HIC) through the homelessness management information system (HMIS). 
Each CoC develops a methodology that best fits their geographic area in accordance 
with HUD?s minimum standards for conducting the PIT count. CoCs may either 
 
4 The CoC Board is the entity established by the CoC to act on its behalf. The CoC?s Board must be 
representative of the CoC and must include at least one homeless or previously homeless individual. 
The responsibilities of the Board depend on how much authority is delegated to the Board by the CoC, 
in accordance with the CoC?s governance charter (HUD, 2014a). 
24 
 
conduct a complete census or one or more sampling and extrapolation methods. HUD 
evaluates the nature and basis for estimation and extrapolation of CoCs sheltered 
and/or unsheltered counts in the annual CoC Program Competition (HUD, 2014b). 
A criticism of the S-Night enumeration that also applies to the CoC PIT count 
is that a point-in-time count is not a representation of the number of people 
experiencing homelessness in a particular year, but rather a count of the number of 
people experiencing homelessness on a single day in that particular year (Grimes & 
Chressanthis, 1997). Additionally, enumerators for PIT counts are typically teams of 
CoC agency representatives and community volunteers, so levels of volunteer 
participation and training can significantly affect the quality or completeness of a 
count, and there may be inconsistencies in the way that communities measure 
homelessness over time (Hanratty, 2017). 
Overall, the PIT count data are the best available data for the number of 
people experiencing homelessness, and researchers have found PIT count data to be 
an improvement over past sources of homelessness data. Byrne et al. found that the 
explanatory power of their homelessness model increased from 35 percent to 58 
percent compared to Lee et al.?s model and attributed that increase to the higher 
quality of CoC data (2013). 
 
2.3: Determinants of Homelessness 
Research on structural determinants of homelessness emerged in the 1990s as 
a response to studies focused on the personal characteristics of those experiencing 
homelessness. In the context of homelessness research, individual-level variables 
25 
 
measure characteristics of people experiencing homelessness, such as their age, race, 
whether they abuse drugs or experience a mental illness, etc. Structural variables 
measure characteristics of the communities in which people experience homelessness.  
These early studies were primarily conducted by researchers in the medical 
and social services fields. By their nature, individual-level studies focused on 
characteristics and conditions of individuals and households, and were based on 
theoretical models that conceptualized homelessness as a result of individual-level 
factors as varied as adverse childhood experiences, disability, mental illness, 
substance abuse disorders, lack of social or human capital, a history of institutional 
involvement, and exogenous health and income shocks (Byrne et al., 2013). These 
researchers studied only personal characteristics of those already experiencing 
homelessness and some found that their hypothesized causes of homelessness 
explained only a small proportion of the variance in the length of time a person 
experienced homelessness (Calsyn & Roades, 1994), while others resulted in findings 
that mental illness was the primary determinant of homelessness and that emergency 
shelters were replacing institutions that had previously been dedicated to people with 
mental health conditions (Bassuk, Rubin, & Lauriat, 1984; Jones, 1983). Other 
researchers, like Freeman and Hall (1987), argued that deinstitutionalization cannot 
be cited as a significant direct cause of homelessness because deinstitutionalization 
began in the late 1950s and early 1960s, while homelessness did not begin to spike 
until the 1980s, roughly 20 years later. 
Quigley et al. (2001) explain that the tendency to downplay housing 
availability as an explanation for homelessness appears to be justified by the traits of 
26 
 
the people experiencing homelessness. Research describes a group suffering 
disproportionately from mental illness, drug and alcohol addiction, and extreme social 
isolation. Nearly one-third of people experiencing homelessness suffer from mental 
illness and one-half abuse drugs or alcohol. Three-quarters of the homeless have been 
institutionalized (Burt & Cohen, 1989; Shlay & Rossi, 1992). Given this confluence 
of personal problems and the relatively low incidence of homelessness, several 
authors have dismissed the explanations of homelessness that focus on housing 
market conditions (Jencks, 1994). 
Point-prevalence estimates fail to account for turnover among the homeless 
and thus understate the likelihood of experiencing a homelessness spell. Culhane et 
al. (1994) show that, although on any day 0.1% of the population of New York City is 
homeless, 1% of the population experiences homelessness over the course of a year, 
and larger percentages of people experience homelessness when measured over 
longer periods. Moreover, turnover among the homeless suggests that point-
prevalence samples are disproportionately composed of individuals suffering long 
spells. Phelan and Link (1999) demonstrate that this composition bias overstates the 
prevalence of personal problems and social isolation among people experiencing 
homelessness, those overemphasizing those factors? importance as an explanation for 
homelessness. 
One of the earliest economic theories attempting to explain how homelessness 
was primarily a structural problem was laid out by Cecil Bohanon (1991), depicted in 
panels A-D of Figure 1. 
27 
 
 
Figure 1: Bohanon's economic theory of homelessness 
In all four panels, the horizontal H axis represents housing consumed where a 
quantity of housing less than H* depicts homelessness, the vertical G axis represents 
all other goods consumed, the line BB represents the budget constraint of a household 
living below the poverty threshold, I represents the indifference curve (or the limits 
within which spending could change without a corresponding change in quality of 
life), and point X represents the quantity of housing consumed by a household living 
in poverty that is not considered to be experiencing homelessness. 
Panel A represents a household living below the poverty line before any 
changes that may cause the household to begin experiencing homelessness, which 
28 
 
could include a decrease in income (represented in panel B), an increase in the cost of 
housing (represented in panel C), or a change in household spending priorities 
(represented in panel D). In panels B, C, and D, a new point Y, Z, or W, respectively, 
represents the quantity of housing consumed by a household living in poverty after a 
change resulted in a situation in which a household could logically choose 
homelessness without experiencing a decrease in their quality of life, represented by a 
new indifference curve, I?. Albeit simplified, this theoretical framework provides a 
foundation for understanding structural determinants of homelessness. 
Like Bohanon and many researchers who conducted studies prior to this one, 
this study is written from the perspective that there will always be people in our 
communities with personal problems that affect their ability to be self-sufficient, and 
research on determinants of homelessness has moved towards a general consensus 
that individual and structural explanations are not mutually exclusive (Byrne et al., 
2013; Culhane, Lee, & Wachter, 1996; O'Flaherty, 2004), but structural conditions 
determine whether or not those most vulnerable members of our society fall into 
homelessness. Below are descriptions of the most commonly studied structural 
determinants of homelessness with arguments as to why they could potentially affect 
homelessness rates and past findings.5 These variables are all included in the panel 
analysis conducted for this study. 
 
5 One variable that was frequently studied in the early 1990s and not included in this analysis is rent 
control. William Tucker (1987) wrote an article for The National Review (a conservative political 
magazine), and various spinoffs of the article began circulating through the media. Several scholars 
included rent control as a variable in models analyzing the relationship between homelessness and 
associated variables to test the validity of the claim and the found the claim that rent control causes 
increases in homelessness to be false almost to the point of fraudulent (Appelbaum et al., 1991; 
Bohanon, 1991; Honig & Filer, 1993; Quigley & Portney, 1990). 
29 
 
2.3.1: Housing affordability 
In a response to Randall Filer?s (1990) claim in an article of the Wall Street 
Journal that ?we know almost nothing about the connection between homelessness 
and housing markets. There is no reliable evidence that homelessness is more 
extensive in cities with tight housing markets,? Bohanon (1991) conducted a cross-
sectional multivariate regression analysis using HUD 1984 sheltered population data 
and 1980 Census data and found that median rent, the most common measurement of 
housing affordability, was the leading factor associated with homelessness rates with 
statistical significance at the one percent level. 
In the early 1980s, the rent-to-income ratio rose so sharply that by 1983, 22 
percent of renters paid 50 percent or more of their income towards rent. In addition, 
declining federal support for public housing construction, growing waiting lists for 
public housing, increasing home ownership costs, more frequent displacement and 
abandonment of residential buildings, and widespread demolition of single-room 
occupancy hotels have all operated to decrease the supply of low-cost housing, 
especially among the very lowest priced units (Hartman, 1986; Hopper & Hamberg, 
1986; Rossi & Wright, 1987). Consequently, some individuals are unable to find 
alternative living arrangements short of emergency shelters or the streets (Elliot & 
Krivo, 1991). 
It cost a young family with children 23 percent of their income to take out a 
mortgage on an average-priced house in 1973, and by 1988 the same scenario would 
cost over half of a young family?s income (Children's Defense Fund, 1988). In 1988, 
the average single-parent household with a head under the age of 25 paid 81 percent 
30 
 
of its income on housing alone (Children's Defense Fund, 1988). Between 1970 and 
1983 rents tripled, while renters? income only doubled. As a result, the average rent-
income ratio grew from roughly one-quarter to one-third, and by 1985 close to one 
out of every four renters paid over half of their income for housing costs. Among 
households living below the poverty line, roughly 45 percent paid more than 70 
percent of their income on housing in 1985, 65 percent paid more than half, and 85 
percent spent more than 30 percent of their income on housing (Appelbaum et al., 
1991). 
Without available or affordable housing, some people will live with relatives 
or friends, thereby increasing the level of household doubling-up (Mutchler & Krivo, 
1989). Other people, especially those with weaker or less resourceful social networks 
(Rossi, 1989), will not find a home at all (Elliot & Krivo, 1991). 
Using data as early as the 1984 HUD survey of opinions used for the agency?s 
national estimate of homeless, Elliot and Krivo (1991) found that the supply of low-
rent housing was one of the two strongest predictors of homelessness, along with per 
capita expenditures on mental health care.6 Bohanon (1991) found that median rent 
was the leading variable associated with homelessness rates with statistical 
significance at the one percent level. Honig and Filer (1993) found a strong 
 
6 The data source for per capita expenditures on mental health care used by Elliot and Krivo has since 
been discontinued. The National Association of State Mental Health Program Directors provided data 
on mental health expenditures by state mental health agencies at the state level. I was unable to find a 
suitable alternative for use in this analysis. Grimes and Chressanthis (1997) believe they found that 
rent control has a highly significant, albeit small, positive impact on homelessness by making rent 
control an endogenous variable instead of an exogenous variable, as it was treated in past studies. 
However, they failed to address the likelihood that the supposed significance of predicted rates of rent 
control was merely masking the significant impact of variables used to predict whether or not rent 
control would be implemented, such as the price of an apartment at the city?s 10th percentile of the 
rent distribution and the percent of the total housing stock which are renter-occupied units. 
31 
 
relationship between measures of housing cost and informed opinion about the 
incidence of homelessness. Hanratty (2017) found that once area-fixed effects are 
included, median rent was the only variable that remained positive and significant in 
its relationship to homelessness. A finding that median rent is statistically significant 
and positively associated with homelessness is common among scholars studying 
homelessness (Burt, 1992; Byrne et al., 2013; Early & Olsen, 2002; Fargo, Munley, 
Byrne, Montgomery, & Culhane, 2013; Grimes & Chressanthis, 1997; Barrett A. Lee, 
Price-Spratlen, & Kanan, 2003; Quigley & Portney, 1990; Quigley et al., 2001). 
I hypothesize that median rent will be statistically significant and positively 
associated with homelessness rates. This has been a consistent finding in past 
research, and I expect that a CoC?s ability to implement Housing First strategies is 
dependent on the availability of affordable housing. 
2.3.2: Income and poverty 
When more people live in poverty, the rate of homelessness may also be 
higher because more people are forced to choose between paying for housing and 
meeting other needs such as food, clothing, and medical care. An area?s poverty rate 
has commonly been included in models studying homelessness, but these models 
have rarely found any statistically significant association between poverty rates and 
homelessness rates. Other measures of income have sometimes been included in 
models studying homelessness, and the results regarding the relationship between 
income and homelessness have been inconsistent. 
For example, Burt (1992) found that poverty and income had no statistically 
significant impact on homelessness rates. Fargo et al. (2013) did not test for poverty, 
32 
 
but found that income is statistically significant and negatively associated with 
homelessness rates. Quigley et al. (2001) found that median household income is 
statistically significant and positively associated with homelessness rates, but found 
per capita income and poverty rates to be statistically insignificant. 
Quigley et al. (2001) tested a theory that homelessness increases with the 
degree of income inequality by regressing homelessness on vacancy rates, median 
rents, the proportion of households with incomes under $15,000, and median 
household income. Their model used median household income as a measurement of 
income inequality by holding the proportion of households with incomes under 
$15,000 constant. The results of their model showed that higher levels of income 
inequality were associated with higher levels of homelessness. To this author?s 
knowledge, a measure of income inequality has not been used in any models studying 
homelessness subsequent to Quigley et al.?s 2001 research, and a measure of income 
inequality will be included in the panel analysis used in this study. 
Marta Elliot and Lauren Krivo (1991) studied income indirectly by using a 
proportion of unskilled jobs as a covariate in their model. They found that unskilled 
jobs, which often do not provide enough income to support monthly housing costs, 
may be the only jobs available to individuals experiencing homelessness (Elliot & 
Krivo, 1991). 
An increase in the percentage of people living below the poverty line from 
11.4 percent in 1978 to 14.4 percent in 1984 support this argument (Elliot & Krivo, 
1991). Reductions in federal means-tested benefit programs, stricter eligibility 
requirements for disability benefits, and decreases in the real value of income 
33 
 
maintenance programs in the early 1980s worsened housing instability for people 
living below the poverty line (Elliot & Krivo, 1991; Hopper & Hamberg, 1986; 
Redburn & Buss, 1986; Rossi & Wright, 1987). 
Despite these findings, income and poverty were found to be statistically 
insignificant in their associations with homelessness more often than they were 
significant (Appelbaum et al., 1991; Burt, 1992; Byrne et al., 2013; Early & Olsen, 
2002; Elliot & Krivo, 1991; Grimes & Chressanthis, 1997; Honig & Filer, 1993; 
Barrett A. Lee et al., 2003; Quigley & Portney, 1990; Quigley et al., 2001). I 
hypothesize that income will not be statistically significant in this model, but that 
income inequality will be statistically significant and positively associated with 
homelessness rates. 
2.3.3: Unemployment and underemployment 
A 1984 survey of nearly 1,000 homeless persons in Ohio (Roth & Bean, 1986) 
emphasizes the importance of unemployment as a cause of homelessness wherein 22 
percent of respondents listed unemployment as the primary reason for their 
homelessness. The long-term shift in employment from manufacturing to service 
industries has increased the proportion of unstable, nonadvancing, low-paying jobs 
(Elliot & Krivo, 1991). Burt and Cohen (1989) reported that single women, women 
with children, and single men experiencing homelessness were without a steady job 
for at least three months for an average of 3.4, 3.8, and 4.2 years respectively. 
However, among these same single men and women experiencing homelessness, 
income from working was the most common single source of income, indicating that 
34 
 
a notable portion of people experiencing homelessness work in unstable jobs for short 
periods of time (Elliot & Krivo, 1991). 
Scholars are split in determining the influence that unemployment rates have 
on homelessness. Unemployment was found to be insignificant in several models that 
tested the impact of unemployment on homelessness rates (Byrne et al., 2013; Early 
& Olsen, 2002; Elliot & Krivo, 1991; Hanratty, 2017; Barrett A. Lee et al., 2003; 
Quigley et al., 2001). However, there have been a nearly equivalent number of studies 
that found unemployment to be positively and significantly associated with the 
homelessness rate (Appelbaum et al., 1991; Bohanon, 1991; Burt, 1992; Troutman, 
Jackson, & Ekelund, 1999). In Hanratty?s (2017) model, unemployment became 
significant and positively associated with homelessness when included in a model 
that used small area estimates of unemployment and vacancy rates. I hypothesize that 
unemployment will be statistically significant and positively associated with 
homelessness in this study. 
2.3.4: Vacancy 
A low residential rental vacancy rate is a signal of a tight housing market, and 
many prior models studying homelessness have included vacancy rates. In models 
that include both median rent and vacancy rates as variables possibly associated with 
homelessness rates, multicollinearity is likely an issue because those two variables are 
likely to be highly correlated with one another. In the case of a tight housing market, 
one would expect vacancy rates to be low and median rent to be high, and vice versa 
in the opposite scenario. 
35 
 
This is likely to be why past findings are split on whether vacancy rates are 
statistically significant in their association with homelessness. In some studies, the 
vacancy rate was found to be statistically significant and negatively associated with 
homelessness, meaning that homelessness is higher in tight housing markets with low 
vacancy rates (Appelbaum et al., 1991; Quigley & Portney, 1990; Quigley et al., 
2001; Troutman et al., 1999). In other studies, the vacancy rate has been found to be 
statistically insignificant (Burt, 1992). I am not aware of any studies that have found 
the vacancy rate to be positively associated with homelessness. I hypothesize that the 
vacancy rate will be statistically insignificant in this model. 
2.3.5: Tenure 
As housing prices inflate, the transition to home ownership becomes more 
difficult and competition in the rental market increases, pushing previously affordable 
rental units out of reach for low-income households at risk of experiencing 
homelessness (Barrett A. Lee et al., 2003). Like the vacancy rate, there is a possibility 
of multicollinearity problems between the effect that tenure has on homelessness rates 
and the effect of housing affordability, since higher proportions of renter-occupied 
households tend to occur in tight housing markets in which low-income households 
are less likely to be able to afford the down payment or acquire financing to purchase 
a home. However, the recent housing crisis has shown that this is not always the case, 
and it is possible for low-income households to receive financing for homes even as 
housing values are inflated. 
Many past studies have not included tenure as an independent variable in 
models analyzing the relationship of variables associated with homelessness rates. 
36 
 
However, past research that has included tenure has been split in their results. The 
majority has found that areas with large proportions of renter-occupied households 
tend to have higher homelessness rates (Appelbaum et al., 1991; Byrne et al., 2013; 
Fargo et al., 2013; Hanratty, 2017). At least one study has found that tenure has no 
statistically significant association with homelessness rates (Barrett A. Lee et al., 
2003), although the rate of homeownership was negatively associated with 
homelessness rates?a result in line with past research. I hypothesize that the 
proportion of renter-occupied households will be statistically significant and 
positively associated with homelessness rates in this study.  
2.3.6: Race and ethnicity 
Several studies and reports have found that black and Hispanic people have 
been heavily overrepresented among people experiencing homelessness (Burt & 
Cohen, 1989; Rossi, 1989; Roth, Bean, Lust, & Saveanu, 1985; HUD, 1984; HUD, 
2018a). Black households in particular are highly segregated (Massey & Denton, 
1993) and often face discrimination in the housing market (Berkovec, Canner, 
Hannan, & Gabriel, 1997; Galster, 1987; Massey, Rugh, Steil, & Albright, 2016; 
Munnell, Tootell, Browne, & McEneaney, 1996; Schafer & Ladd, 1981; Wienk, Reid, 
Simonson, & Eggers, 1979). This discrimination worsened in the lead up to the Great 
Recession. In the mid-1990s, an estimated 10 to 35 percent of people issued subprime 
loans were eligible for prime loans (Mahoney & Zorn, 1996). About 10 years later, as 
we approached the Great Recession, this percentage grew until 62 percent of 
subprime borrowers, disproportionately black and Hispanic, actually qualified for 
prime loans in 2006 (Brooks & Simon, 2007). 
37 
 
Massey et al. (2016) analyzed a randomly-selected sample of 220 deposition 
statements and testimonies from cases where discrimination in the real estate market 
during the housing boom prior to the Great Recession was alleged and the case went 
to trial to identify instances of structural or individual discrimination. They found that 
structural racism was evident in 76 percent of the 220 deposition statements and 
testimonies, and that individual racism was only evident in 11 percent of the same 
texts. In the deposition statements analyzed, defendants referred to subprime 
mortgages as ?ghetto loans? and black customers were referred to as ?less 
sophisticated and intelligent,? ?easier to manipulate,? ?people who don?t pay their 
bills,? and even ?mud people.? Some of the strategies used by lending institutions that 
were discussed in these cases included cold-calling black potential lenders multiple 
times per day, deliberate deception and misrepresentation of lending terms, 
falsification of loan documents, recruiting community leaders to unwittingly build 
trust for predatory lenders, targeting elderly households for aggressive high-pressure 
marketing, the organization of sales events for subprime loans labeled ?wealth-
building seminars,? and the use of a particularly deceptive practice of mailing ?live 
draft checks? to targeted households in black communities (Massey et al., 2016): 
Wells Fargo would mail checks in the amount of $1,000 or $1,500 to leads. 
Once these checks were deposited or cashed, they instantly became loans with Wells 
Fargo at very high interest rates. Individuals who cashed these checks became an 
instant ?lead? target for a home equity refinance loan, which of course would end up 
placing the borrower?s home at risk. 
During the Great Recession, the level of black-white segregation powerfully 
predicted the rate of foreclosures (Rugh & Massey, 2010). For example, majority-
black Prince George?s County experienced the largest concentration of foreclosures 
38 
 
following the Great Recession compared to any other county in Maryland (Boston, 
2012). According to an analysis of Home Mortgage Disclosure Act data, this was a 
trend across the country, as minority and particularly black households were more 
likely to receive subprime loans than white households in the lead up to the Great 
Recession (Goldstein & Urevick-Ackelsberg, 2008). 
These practices unfairly diminished the wealth of black borrowers and 
exposed them to elevated risk of foreclosure and repossession, making those 
households unstably housed and increasing pressure on the rental market in black 
communities (Massey et al., 2016). In the rental market, evictions are much more 
common among black households than white households (Desmond, 2016).  
Racial discrimination in housing intersects with gender, and there is evidence 
that black women are affected specifically. In Milwaukee?s poorest black 
neighborhoods, one female renter in 17 was evicted through the court system each 
year, which was twice as often as men from those neighborhoods and nine times as 
often as women from the city?s poorest white areas. Women from black 
neighborhoods made up nine percent of Milwaukee?s population and 30 percent of its 
evicted tenants (Desmond, 2016). Desmond summarized the proliferation of evictions 
being carried out against low-income black women (2016): 
If incarceration had come to define the lives of men from impoverished black 
neighborhoods, eviction was shaping the lives of women. Poor black men were locked 
up. Poor black women were locked out.  
Structural racism has worked its way into social media platforms used for real 
estate advertising as well. The Department of Housing and Urban Development 
(HUD) recently charged Facebook with violating the Fair Housing Act and stated that 
39 
 
Facebook is "encouraging, enabling, and causing" housing discrimination through its 
advertising platform. An article published by CNN cited an investigation reportedly 
conducted by ProPublica in November 2017 that found discriminatory ads were being 
published on Facebook and thereby violating the Fair Housing Act. ProPublica was 
able to purchase dozens of home-rental ads that specifically excluded "African 
Americans, mothers of high school kids, people interested in wheelchair ramps, Jews, 
expats from Argentina and Spanish speakers" (Yurieff, 2019). Regarding the case, 
HUD Secretary Ben Carson said the following in a statement (Yurieff, 2019):  
Facebook is discriminating against people based upon who they are and 
where they live. Using a computer to limit a person's housing choices can be just as 
discriminatory as slamming a door in someone's face. 
Discrimination in the real estate market based on race and ethnicity for both 
homeowners and renters limits peoples? housing choices and theoretically increases 
the risk of homelessness for at-risk households who experience either structural or 
individual discrimination while searching for a home. HUD (2018a) notes that nearly 
half of all people experiencing homelessness (49% or 270,568 people) identified their 
race as white, and nearly 6 in 10 people (59%) experiencing unsheltered 
homelessness were white. While comprising nearly half of the population of people 
experiencing homelessness, people identifying as white were underrepresented 
compared to their share of the U.S. population (72 percent).  In comparison, almost 
40 percent of people experiencing homelessness identified their race as black, while 
the population of the United States is under 13 percent black (Census, 2013-2017; 
HUD, 2018a). 
40 
 
Much of the prior research conducted on variables associated with 
homelessness has not included race or ethnicity as a tested variable, and the research 
that has included race or ethnicity experienced mixed results. Two known studies 
found that the relative size of the black population was statistically significant and 
positively associated with homelessness rates (Elliot & Krivo, 1991; Honig & Filer, 
1993). Honig and Filer (1993) found that the relative size of the black population had 
an especially strong association with the rate at which households doubled-up with 
family members or friends. Other studies have found no statistically significant 
association between race or ethnicity and homelessness rates (Byrne et al., 2013; 
Early & Olsen, 2002; Barrett A. Lee et al., 2003; Troutman et al., 1999). I 
hypothesize that the percentage of renters identifying as white, non-Hispanic will be 
statistically significant and negatively associated with homelessness rates. 
2.3.7: Climate 
Theoretical frameworks for understanding structural determinants of 
homelessness, such as those proposed by Bohanon (1991) and O?Flaherty (1996, 
2010, 2012) look at homelessness as, to some extent, a result of a rational economic 
decision-making process for low-income households between housing and other 
goods. From this theoretical vantage point, the inclusion of climate in a model 
studying variables associated with homelessness makes sense, because a household?s 
tolerance for homelessness in favor of other goods may be higher in areas with a 
temperate climate and low precipitation. It may also be possible that people find 
temporary housing that is not picked up in the PIT count and that is not normally 
available to them when the climate is particularly severe, or that some migration of 
41 
 
people experiencing chronic homelessness to areas with milder temperatures and less 
rainfall takes place, which would mean that climate variables may be statistically 
significant, but not as substantively relevant. 
Past research has sometimes found mean temperature to be statistically 
significant and positively associated with homelessness (Appelbaum et al., 1991; 
Troutman et al., 1999). Other studies found that higher temperatures are only 
statistically significant and positively associated with specifically unsheltered 
homelessness (Corinth & Lucas, 2018; Early & Olsen, 2002). Some studies have 
found that climatic variables are not associated with homelessness rates in a 
statistically significant way (Bohanon, 1991). Byrne et al. (2013) determined that it 
was not feasible to include measures of climate given that CoCs, which formed their 
unit of analysis, can be large enough that there was significant within-CoC climate 
variation. This study also uses CoCs as the unit of analysis and will attempt to include 
temperature and precipitation data using methods described in the next chapter. I 
hypothesize that climatic variables will be statistically significant, and that 
temperature will be positively associated with homelessness rates while precipitation 
is negatively associated with homelessness rates. 
2.3.8: Inclusionary zoning 
Grounded Solutions Network conducted a national census of local 
inclusionary housing programs and a national survey of state-level legislation and 
judicial decisions related to the adoption of inclusionary housing programs to create 
an inclusionary housing database in 2016. The existence of an inclusionary housing 
policy mandating or incentivizing the creation of affordable housing may have an 
42 
 
impact on homelessness rates, so data from the Grounded Solutions database was 
included in this study.  
2.3.9: Eviction filing rates 
A national database of eviction data became available from Princeton 
University?s Eviction Lab in 2018, and past researchers have not yet had the 
opportunity to study the association between eviction filing rates and homelessness 
rates at the national level. However, Collinson and Reed (2018) found that evictions 
cause households in New York City to be more likely to become homeless. 
2.3.10: Housing First strategies 
Research on the effectiveness of Housing First strategies has been limited thus 
far to small-scale longitudinal studies in one or several CoC areas. This study will 
broaden the scope to all CoCs to determine whether those CoCs that are more 
diligently implementing Housing First strategies such as permanent supportive 
housing and rapid re-housing are more effective in reducing homelessness than CoCs 
that rely more on emergency shelters and other high-barrier, short-term, or 
transitional strategies. 
Given past literature?s findings that the availability of affordable housing 
tends to be the most commonly cited statistically significant variable associated with 
homelessness rates, and the Housing First approach is geared towards finding ways to 
place people experiencing homelessness in an affordable housing unit permanently, I 
hypothesize that the implementation of Housing First strategies will have a 
statistically significant and negative association with homelessness rates. 
 
43 
 
2.4: Prior Models 
Past studies have found that structural socio-economic factors such as rent 
levels, race, and unemployment contribute to homelessness. Most studies analyzed a 
cross-section of data from a single year. Quigley et al. (2001) conducted a panel 
analysis using the Aid to Families with Dependent Children Homeless Assistance 
Program (AFDC-HAP) eight-year dataset as part of their study, but this analysis was 
limited to counties in California.  
Hanratty (2017) and Corinth (2017) also used a panel analysis like the one 
used in this study. Hanratty (2017) broadened the scope from California to CoCs 
across the country, and she studied the effect of right-to-shelter policies. Corinth 
(2017) studied the relationship between permanent supportive housing and 
homelessness and found that a significant negative relationship existed, but that the 
relationship was not very substantive and permanent supportive housing was not 
responsible for the most recent reductions in homelessness. 
This study builds from the progress made by the modeling research of 
Hanratty (2017) and Corinth (2017) and introduces an index measuring the 
implementation of Housing First into a modeling study, thereby bridging the gap 
between longitudinal studies of Housing First and modeling studies on the 
relationship between homelessness rates and various independent variables. 
Methodologically, this study also builds from the research of Hanratty (2017) and 
Corinth (2017) by adding an analysis of interaction effects. This allows one to better 
understand how a third variable may interact with an independent variable to change 
the independent variable?s relationship with homelessness rates. The results of this 
44 
 
analysis will provide a deeper and more nuanced understanding of the relationship 
that Housing First and other variables have with homelessness in the United States. 
For a simplified summary of studies that included models analyzing the 
relationship between homelessness rates and variables associated with homelessness 
and that were discussed most frequently in the literature review preceding this 
section, please see Table 1 below. The table includes each study?s authors, data 
source for homelessness rates, and simplified indications of a small selection of key 
variables? statistical significance and the variable?s positive or negative relationship 
with homelessness rates. Key variables in Table 1 include rent, vacancy rates, 
unemployment, poverty levels, housing tenure (positive and negative correlations 
relate to renter levels), race (positive and negative correlations relate to percent 
black), and income levels. 
45 
 
Table 1: Prior research of homelessness rates and associated variables 
Study Y Data Rent Vac. Unem. Pov. Tenure Race (renter) (black) Income 
Quigley (1990) HUD, 1984 +** -** N/A 0 N/A N/A N/A 
Elliot & Krivo HUD, 
(1991) 1984 +** N/A 0 0 N/A +** N/A 
Appelbaum et HUD, 
al. (1991) 1984 0 -* +** 0 +** N/A N/A 
Bohanon HUD, 
(1991) 1984 +** N/A +** N/A N/A N/A N/A 
Burt (1992) Burt, 1989 +** 0 +** 0 N/A N/A 0 
Honig & Filer HUD, 
(1993) 1984 +** 0 N/A 0 N/A +* N/A 
Grimes & 
Chressanthis Census, 1990 +** N/A N/A 0 N/A N/A N/A (1997) 
Troutman et Census, 
al. (1999) 1990 0 -** +** +* N/A 0 N/A 
Quigley et al. Census, 
(2001) 1990 +** -** 0 0 N/A N/A +** 
Early & Olsen Census, 
(2002) 1990 +** 0 0 0 N/A 0 N/A 
Lee et al. Census, 
(2003) 1990 +** 0 0 0 0 0 N/A 
Byrne et al. HUD, 
(2013) 2009 +** 0 0 0 +** 0 N/A 
Fargo et al. HUD, 
(2013) 2009 +** N/A N/A N/A +* N/A -* 
Hanratty HUD, 
(2017) 2007-14 +** 0 0 +** +** N/A N/A 
Corinth (2017) HUD, 2007-14 +** N/A +** N/A N/A N/A N/A 
**significant at a 5% level; *significant at a 10% level; 0 indicates factor was not 
significant; N/A indicates factor was not tested, or was only included within a 
composite variable in which individual significance was not tested 
Note: In several cases, studies ran several different models and/or tested multiple 
methods or subpopulations of people experiencing homelessness as the dependent 
variable; therefore, the correlations and levels of significance reported in this table will 
not reflect the complexity and nuance of past studies? results. 
46 
 
Chapter 3: Methodology, Data, and Constructing the Model 
To answer this study?s research question, a linear mixed models panel analysis 
is conducted using time-series data between 2009 and 2017. The independent 
variables include many of the commonly cited variables associated with homelessness 
discussed in the previous chapter, along with an independent variable measuring the 
degree to which a CoC?s response to homelessness follows the Housing First 
approach. Determining the degree to which this variable is associated with 
homelessness rates is the primary focus of this study. The dependent variable will be 
homelessness rates. 
Both direct effects and interaction effects of the implementation of the 
Housing First approach on homelessness rates while controlling for other variables 
associated with homelessness will be analyzed as a part of this study. This study will 
thus (a) analyze whether Housing First is associated with decreases in homelessness 
rates at the CoC level, and (b) analyze whether Housing First is associated with 
decreases in homelessness rates at the CoC level under particular types of structural 
and community conditions, as indicated by the interaction effects of changes in other 
variables associated with homelessness rates. 
This study uses Continuum of Care (CoC) boundaries, as described in section 
2.2.4: , as the units of analysis and uses 355 of the 384 CoCs across the country, 
including the two CoCs in Puerto Rico, to conduct the analyses. The study analyzes 
data between years 2009 and 2017 using a linear mixed model to determine the 
variables most significantly associated with homelessness rates and to learn the nature 
of those relationships. 
47 
 
3.1: Study Variables 
The dependent variables used in this study are rates of homelessness and rates 
of homelessness experienced by different subpopulations. The independent variables 
are listed in Table 2 below and include a mixture of variables used in past studies and 
new variables made available thanks to the efforts of researchers creating new sources 
of data subsequent to much of the prior research being conducted on variables 
associated with homelessness. 
The selection of several variables matches the variables used in prior studies 
because some prior studies have found those independent variables to be related to 
homelessness rates in a statistically significant way and because including the same 
variables allows for the results of this study to be more easily compared to prior 
studies. Some of these variables include the percentage of renter-occupied 
households, the unemployment rate, the poverty rate, the vacancy rate, and climate 
variables. Other variables used in this study have been used less frequently in prior 
studies and I will briefly explain some of those variable choices.  
The racial composition of a community has been studied infrequently in past 
research, and this study used the percentage of renters who identified as non-Hispanic 
white as a measure of race instead of a breakdown of racial composition to avoid 
collinearity issues in the model. Racial segregation or concentration was not 
considered, because many CoCs are composed of multiple counties and the scale was 
too large for a measurement of segregation to be meaningful. 
Rent burden was used a measure of housing affordability in each CoC, and 
collinearity concerns are also why other measures of housing affordability were not 
48 
 
included. Rent burden was chosen instead of an overall percentage of households 
burdened with housing costs, because renters are more vulnerable to homelessness if 
they can no longer afford their home.  
Two variables were included in this model because of the availability of new 
data measuring inclusionary zoning and eviction filing rates from the Grounded 
Solutions Network and Princeton University?s Eviction Lab. Both of these variables 
have an intuitive link to homelessness rates, but the existence of relationship between 
either variable and homelessness rates has not been studied. 
The Housing First index was included in this study to test the primary 
research question of this study, which is related to the relationship between the 
Housing First approach to ending homelessness and homelessness rates in CoCs 
across the country. The Housing First approach has been studied in longitudinal 
research, but the index will allow the relationship between Housing First and 
homelessness rates to be studied as well. 
 
49 
 
Table 2: Variables used in this model 
Name Type Description Notes 
hl y people per 1,000 experiencing homelessness  
hls y people per 1,000 experiencing sheltered  
homelessness 
hlu y people per 1,000 experiencing unsheltered  
homelessness 
hlf y people per 1,000 in families experiencing  
homelessness 
hlc y people per 1,000 experiencing chronic  
homelessness 
hf x Housing First index  
medren x median gross rent of renter-occupied  
housing units 
medval x median home value of owner-occupied  
housing units 
medinc x median household income  
renocc x percentage of renter-occupied housing units  
renfam x percentage of renters in a family with no 2009 data 
children 
renwhi x percentage of renters identifying as white,  
non-Hispanic 
renedu x percentage of renters without a college  
degree 
unemp x unemployment rate  
pov x poverty rate no 2009 data 
fund x HUD CoC funding in the previous year per  
person 
evic x eviction filing rate no PR or 2017 data; 
17% missing 
burd x percentage of rent-burdened households  
gini x gini index no 2009 data 
vac x vacancy rate  
coccat x CoC category: balance of state, smaller city not longitudinal 
or county, or major city 
temp x average January temperature no DC, HI, or PR 
data 
precip x total January precipitation no DC, HI, or PR 
data 
inczon x inclusionary zoning policy dummy variable not longitudinal 
Sources: U.S. Census Bureau ACS 5-year estimates; Eviction Lab, Princeton 
University; HUD PIT count data; HUD allocations and awards data 
 
 
50 
 
3.2: Creation of a Housing First Index 
The primary independent variable used in this study is a Housing First index. 
As O'Flaherty (2019) discussed, an estimate regarding the relationship between 
Housing First and homelessness in a community is missing from the existing 
literature, so an index by which to measure the degree to which Housing First is being 
implemented in a community has not been created before. With no previous examples 
to build from, this study will create a Housing First index with the hope that future 
research will improve upon this design. In this study, the degree to which a CoC is 
implementing a Housing First approach is calculated using data from the Housing 
Inventory Count (HIC), which is submitted by CoCs to HUD each year. 
 The index is constructed by using housing inventory count data gathered from 
HUD Exchange (2018b) and a simple formula: 
Equation 2: Housing First index 
???2? + ?? + ?? + ????? =  
??
 
The Housing First index is equal to the sum of the number of transitional 
housing beds (?) divided by two, the number of rapid re-housing beds (?), the number 
of permanent supportive housing beds (?), and the number of other permanent 
housing beds (?) divided by the total number of beds (?) in the CoC, which also 
includes emergency shelter and safe haven beds. This results in an index score 
between 0 and 1 where a higher value indicates more reliance on permanent housing 
strategies and a stronger alignment with the Housing First approach. Since the idea of 
the Housing First approach is to move people experiencing homelessness into a 
51 
 
permanent housing unit as quickly as possible to stabilize them and to give them the 
best chance of retaining their housing, a permanent unit of any kind is worth the most 
points, emergency shelter options are worth no points, and transitional housing falls 
somewhere in between, so those beds are worth half points. See a brief definition of 
Housing First and each category of the housing inventory in the glossary of this study 
for more information (HUD, 2017b; 2019a). 
Of the 3,195 possible index scores for this variable, 3,148 or 98.5% of the 
total possible scores were calculated. Unlike the point-in-time counts that are required 
to be conducted once every two years, housing inventory count data come from a 
CoC?s mandatory maintenance of a Homelessness Management Information System 
(HMIS). Data are required to be collected from service providers and submitted to 
HUD each year. Although some entries are missing, the amount of data available 
nonetheless provides a large enough sample size for use in this study, likely thanks to 
the HMIS requirement. 
There were changes to the components of this index score over the duration of 
the study period. At the beginning of the study period in 2009, the only categories of 
housing included in the housing inventory count were emergency shelter, transitional 
housing, safe haven, and permanent supportive housing. In 2013, after HUD had 
published the CoC Program Interim Rule to formally implement the CoC Program 
adopted by the HEARTH Act of 2009, rapid re-housing was added to the housing 
inventory count. In 2014, other permanent housing was added to the housing 
inventory count as well.  
 
52 
 
3.3: Conducting the Panel Analyses 
Panel analysis is a way to study the relationships between dependent variables 
and independent variables over both space and time, i.e. the data are both cross-
sectional and longitudinal in nature. This study is analyzing the relationships between 
various types of homelessness and a number of independent variables across a set of 
349 CoCs over a span of nine years. The study will use the linear mixed models 
(LMM) procedure in SPSS to conduct the panel analyses. 
3.3.1: Linear mixed models procedure 
The linear mixed models (LMM) procedure has several benefits over other 
types of methods for conducting panel analyses. One such benefit is that the LMM 
procedure is better able to handle data in an unbalanced design, i.e. a dataset with 
missing values. In the case of this study, of the 3,141 observations left for analysis 
after the data were cleaned as described in previous sections, only 1,724 of those 
observations would be usable in a panel analysis that required a balanced design that 
excluded cases listwise as opposed to pairwise. In a comparison between a balanced 
and unbalanced dataset, a general linear model (GLM) procedure will produce 
different results in terms of its fixed effects estimates and its estimates of covariance 
parameters, while a variance components (VARCOMP) procedure and LMM 
procedure can produce the same estimates in an unbalanced design. This is because 
the LMM and VARCOMP procedures offer maximum likelihood or restricted 
maximum likelihood methods of estimation, while GLM estimates are based on the 
method-of-moments approach. LMM is generally preferred because it is 
asymptotically efficient (minimum variance), whether or not the data are balanced, 
53 
 
while GLM only achieves its optimum behavior when the data are balanced (SPSS 
Inc., 2005). In the case of this study, the ability of the model to handle an unbalanced 
dataset well is necessary. 
In cases where observations are repeated for each subject over time, as is the 
case in this study, the LMM procedure accounts for the assumption that the error 
terms within a subject may be correlated, but independent across subjects. The GLM 
and VARCOMP procedures ignore possible correlations within the data, which may 
lead to incorrect conclusions regarding the significance of independent variables in 
the model. Although VARCOMP is a subset of LMM and the two procedures 
produce the same variance estimates, VARCOMP only fits relatively simple models 
because no statistics on fixed effects are produced and it can only handle random 
effects that are independent and identically distributed. For these reasons, LMM is the 
preferred alternative to GLM and VARCOMP when data are likely correlated 
because they come from the same subject (SPSS Inc., 2005). 
Compared to a repeated measures analysis of variance (RM ANOVA) 
procedure, the LMM procedure is capable of considering time-dependent or time-
varying continuous covariates, while the RM ANOVA procedure is only capable of 
considering baseline values of continuous covariates. Like GLM, the RM ANOVA 
procedure is also unable to analyze unbalanced datasets well. A single missing 
variable value in a case will cause that case to be dropped from the analysis in a 
process known as listwise, as opposed to pairwise, deletion (West, 2009).  
The LMM procedure can uniquely consider random effects to explain random 
between-subject (CoCs, in the case of this study) variance in trajectories, and to also 
54 
 
then analyze several different alternative covariance structures for random effects and 
compare those covariance structure models to determine the model with best fit to a 
longitudinal dataset (West, 2009; West, Welch, & Galecki, 2015). I chose to measure 
fixed effects for every independent variable except for time and place, because I am 
analyzing the entire population of CoCs and attempting to understand the relationship 
between the independent variables and homelessness rates in this population. This is 
different from testing within a sample to extrapolate information about the 
relationships to apply in a broader context, in which it may be more appropriate to 
test more of the independent variables? relationship with homelessness rates for 
random effects. All in all, the LMM procedure is the best method to use in the case of 
an unbalanced panel dataset in which I expect longitudinal observations in the same 
CoC to be at least partially dependent on one another due to the conditions and 
environment in which that CoC is operating. 
3.3.2: Methodology for model selection 
M?ller, Scealy, and Welsh (2013) reviewed a large body of literature on linear 
mixed models selection and compared four different model selection methods. These 
methods include using information criteria, shrinkage methods based on penalized 
loss functions, the Fence procedure, and Bayesian techniques. According to M?ller et 
al. (2013), using information criteria such as Akaike?s Information Criterion (AIC) is 
the preferred method, paired with ensuring that the model is supported by the 
literature. For this study, the analysis will be run in SPSS using the MIXED 
procedure, which allows for a report of the AIC and other information criteria in each 
model run.  
55 
 
To select a final model, first a simple linear or polynomial regression was run, 
testing each potential covariate?s independent relationship with homelessness rates 
(see Appendix 4: Individual Variable Regressions). Those variables without a 
statistically significant relationship with homelessness rates and without strong 
support in the literature suggesting a relationship with homelessness rates were 
excluded from the linear mixed models. Those variables without a statistically 
significant relationship in an independent linear or polynomial regression that are 
strongly supported by the literature may be important in interactions with other 
variables, which were tested in the linear mixed models. All remaining variables were 
included in the linear mixed models, and those variables that did not show a 
statistically significant relationship in their main effects with homelessness rates were 
removed one at a time in order of their p score values from highest to lowest and 
tested for interaction effects with all remaining variables before being removed if they 
were not significant in their interaction effects. Any interactions that showed 
significance and lowered the AIC were kept in the model even after main effects were 
removed. The model was continually reduced in this manner until all the variables left 
in the model were statistically significant. 
Once all the model?s independent variables were significant, the goal of the 
model refinement process became lowering the AIC. At that point, all main effect 
variables that were removed were reintroduced one at a time, and interaction effects 
between all remaining variables were tested one at a time to see if inclusion of a main 
effect or interaction variable can further decrease the AIC or change the significance 
of existing variables. After the reintroduction of main effect variables and the 
56 
 
inclusion of interaction effects was tested, this resulting model with the lowest AIC 
and statistically significant variables was considered the final model from which the 
bulk of the study?s conclusions are drawn. 
This method will produce a model that will show which variables had the 
most statistically significant relationship with homelessness rates, and will help one to 
understand the relationship or lack of a relationship between implementation of the 
Housing First approach and homelessness rates in CoCs across the United States.  
  
57 
 
Chapter 4: Descriptive Statistics 
This chapter includes descriptive statistics of some of the primary variables 
used in the panel analyses to provide a foundational understanding of the values and 
distributions of each variable before attempting to interpret the results of the panel 
analyses. Additional descriptive statistical results can be found in the appendices.  
 
4.1: Housing First Index Scores 
Despite changes to the way the Housing First index is measured over the 
study period due to changes in available data from HUD, the average Housing First 
index score across all CoCs slowly and steadily increased each year of the study 
period (see Table 3 below). This suggests that this index score is accurately reflecting 
a trend of service providers transitioning to more of a Housing First approach, and 
that the addition of more categories in the housing inventory count did not include 
service providers assisting people experiencing homelessness with permanent housing 
options that were previously excluded from the housing inventory count, but rather 
that service providers? permanent housing programs were simply recategorized into 
more refined categories that more accurately describe the services they provide. 
58 
 
Table 3: Housing First index scores 
Year Housing First index score 
2009 0.502 
2010 0.510 
2011 0.528 
2012 0.531 
2013 0.543 
2014 0.562 
2015 0.582 
2016 0.609 
2017 0.623 
 
This indication that more service providers across the country are moving 
towards a Housing First model makes it even more important that we study the 
relationship between the Housing First approach and homelessness rates. 
 
4.2: Means Comparison 
Descriptive statistics were run on the variables used in this study to better 
understand the dataset. 7 To establish a foundation for understanding the variables 
used in this study and their relationships with homelessness, variable means stratified 
by the median homelessness rate were calculated and provided in Table 4. 
 
7 See Appendix 1 for a complete set of descriptive statistics run for the dataset used in this study, 
including variables that were not used in the final panel analyses. 
59 
 
4.2.1: Means comparison split by homelessness rates 
Table 4: Means of variables stratified by observations with above- and below-
median homelessness rates 
  Below-Median Above-Median 
Variable Full Sample Homelessness Homelessness 
People per 1,000 homeless 1.97 (.036) 0.86 (.008) 3.09 (.059) 
Total population (1000s) 729.81 (18.27) 745.20 (23.20) 713.99 (28.36) 
% Bachelor?s degree or higher 29.13 (.221) 28.69 (.349) 29.59 (.267) 
Median household income  55,411 (360) 56,879 (601) 53,903 (384) 
Median gross rent 894.72 (5.82) 865.41 (9.25) 924.85 (6.90) 
Median home value (1000s) 214.50 (2.36) 190.41 (2.67) 239.26 (3.81) 
% Renters 33.92 (.215) 31.01 (.332) 36.90 (.252) 
% Renters with children 34.26 (.204) 34.97 (.343) 33.50 (.210) 
% Renters, white non-Hispanic 63.79 (.405) 67.20 (.461) 60.28 (.502) 
% Renters, no college 45.21 (.243) 47.11 (.408) 43.27 (.408) 
Unemployment rate 8.41 (.055) 7.97 (.085) 8.86 (.066) 
CoC funding per person 5.16 (.093) 3.72 (.079) 6.62 (.160) 
Poverty rate 15.11 (.133) 14.56 (.225) 15.69 (.132) 
Eviction filing rate 6.30 (.171) 6.50 (.287) 6.10 (.181) 
% Rent-burdened 31.04 (.126) 30.56 (.225) 31.54 (.106) 
Gini index 0.46 (.002) 0.45 (.004) 0.46 (.002) 
Vacancy rate 12.65 (.126) 12.41 (.176) 12.89 (.182) 
Mean temperature in Jan 35.45 (.302) 33.19 (.465) 37.79 (.374) 
Total precipitation in Jan 2.96 (.049) 2.79 (.059) 3.14 (.078) 
CoC category* 1.96 (.008) 2.08 (.010) 1.84 (.012) 
Inclusionary zoning* 0.34 (.008) 0.26 (.011) 0.42 (.012) 
Housing First index 0.54 (.003) 0.54 (.004) 0.54 (.004) 
Standard errors are in parentheses. 
* Non-continuous variable. For CoC categories, 1 indicates a major city, 2 indicates a smaller town or 
county jurisdiction and 3 indicates a balance-of-state CoC. Inclusionary zoning is a dummy variable in 
which 1 indicates the adoption of an inclusionary zoning policy in at least one jurisdiction within the 
CoC and 0 indicates that no inclusionary zoning policies have been adopted within the CoC. 
 
Though the means comparison is not part of the primary analysis and does not 
control for the effects of any of the independent variables, Table 4 shows no 
difference between the mean Housing First index scores of observations below or and 
above the median homelessness rate. Across observations, meaning a single year 
within a single CoC, where homelessness rates were above or below the median (1.40 
60 
 
people per 1,000 experiencing homelessness) across the study period, the mean 
Housing First index score was 0.54. This tells me that though Housing First is slowly 
being adopted by more service providers each year, it is being adopted at the same 
rate by CoCs with both above median and below median homelessness rates. This 
could be an early indication that Housing First will not be associated with a decrease 
in homelessness rates, but this means comparison is simply descriptive and does not 
control for any other variables or describe a relationship between Housing First and 
homelessness rates. 
Another highlight from the means comparison is that there is a considerable 
disparity in the mean number of people per 1,000 experiencing homelessness between 
the two groups. In the below-median homelessness group, the mean homelessness 
rate is only 0.86 people per 1,000 experiencing homelessness. That number is the 
above-median homelessness group is almost three and a half times greater at 3.09 
people per 1,000 experiencing homelessness. This tells me that there is a high amount 
of disparity between areas with low homelessness rates and areas with high 
homelessness rates, and this adds some weight to the differences in means of the 
other variables. While this table only shows me correlations with no attempt to 
determine causality, any discernable difference in the means of other variables should 
be studied very carefully. 
In observations with homelessness rates above the median, the means of the 
median gross rent, median home values, and the percentage of renters in the housing 
market were all considerably higher, but the median household income was lower and 
the poverty rate was higher. This shows that in areas with higher homelessness rates, 
61 
 
there are more renters, more expensive housing, and lower incomes. These 
correlations appear to line up with past findings that homelessness is largely a 
structural economic problem driven at least partially by a lack of affordable housing 
accessible to the most vulnerable households. The mean values for the percentage of 
rent-burdened households line up with past findings of that nature as well, being that 
jurisdictions with above-median homelessness rates have a higher mean percentage of 
rent-burdened households, although only slightly. Vacancy rates, on the other hand, 
do not. A slightly higher mean vacancy rate in observations with homelessness rates 
above the median implies that the housing market is tighter in areas with below-
median homelessness rates, though the difference in means for that variable is small 
(12.41 to 12.89).  
Another interesting means comparison is that the percentage of renters who 
are white, non-Hispanic, is lower in the above-median homelessness group with 
observations above the median homelessness rate. HUD?s tabulations of point-in-time 
(PIT) count data (2018a) show that while nearly half of all people experiencing 
homelessness in 2018 (49 percent or 270,568) identified their race as white, and 59 
percent of people experiencing unsheltered homelessness identified their race as 
white, people identifying as white were still underrepresented in the population of 
people experiencing homelessness compared to their share of the U.S. population, 
which is 72 percent (HUD, 2018a). This could be indicative of areas with large white, 
non-Hispanic, populations having more wealth, more accommodating access to 
services, and stronger safety nets in place, either institutional or social in nature, that 
62 
 
prevent vulnerable populations from experiencing homelessness and merits careful 
study during the primary analyses. 
Observations with homelessness rates above the median also had a higher 
mean temperature in January, when the point-in-time count is conducted. This lines 
up with some past findings (Appelbaum et al., 1991; Early & Olsen, 2002; Troutman 
et al., 1999) and suggests that those people experiencing homelessness who have the 
ability to move from one area of the country to another may move to warmer areas 
during the winter. It could also be an indication that people experiencing 
homelessness in colder areas are more likely to seek temporary housing in the last 
week of January that causes them to be missed during the PIT count, such as doubling 
up with family or friends, or staying in a motel room. 
The mean CoC funding per person and the rate at which CoCs had at least one 
inclusionary zoning policy in place were also both higher in observations where the 
homelessness rate was above the median for all CoCs. This may be indicative of 
responses to homelessness and a lack of affordable housing, respectively, but the 
differences merit further investigation during the primary analyses of this study. 
Other variables that tended to have higher means in the above-median 
homelessness group included unemployment and precipitation. The unemployment 
rate has been shown to be positively and significantly associated with homelessness 
rates in several prior modeling studies (Appelbaum et al., 1991; Bohanon, 1991; Burt, 
1992; Corinth, 2017; Troutman et al., 1999), so it is not surprising that the mean 
unemployment rate is a little higher in CoCs with above-median homelessness rates, 
63 
 
where the mean unemployment rate is 8.86, compared to CoCs with below-median 
homelessness rates where the unemployment rate is 7.97.  
Precipitation being higher in CoCs with above-median homelessness rates is 
not very intuitive, nor does it align with my hypothesis for the relationship between 
precipitation and homelessness rates. It could be that any changes in precipitation 
between CoCs with above-median homelessness rates and below-median 
homelessness rates is purely coincidental. It could also be an indication that 
precipitation is not significantly related to homelessness, but temperature is positively 
and significantly associated with homelessness rates, and precipitation is simply 
higher in CoCs with warmer mean temperatures in January. A simple regression was 
run to test whether this is true for the data used in this study, and the results are 
plotted in Figure 2. Though it is obvious from the scatterplot that the relationship 
between temperature and precipitation is not best described linearly, there is a clear 
positive slope to the linear fit line in which precipitation goes up by 0.05 inches with 
each increase in the mean temperature of one degree Fahrenheit. 
64 
 
 
Figure 2: Total precipitation by mean temperature linear regression 
 
The percentage of renters with no college education and the CoC category had 
lower mean values in CoCs with above-median homelessness rates. This tells me that 
homelessness rates are higher in CoCs with more educated renters and in major cities. 
These two variables could certainly be associated with one another and variables like 
income and median rent, so it will be important to test for multicollinearity when 
refining the model for use in the panel analysis.  
Variables that did not show a noteworthy difference in means between the two 
groups included total population, the total percentage of people with a bachelor?s 
degree or higher (though this percentage was slightly higher in above-median 
homelessness CoCs), the percentage of renters with children, the eviction filing rate, 
and the Gini index. None of these are particularly surprising, because they do not 
have a foundation in the literature for being significantly associated with 
homelessness, except for income inequality measured via the Gini index. Since the 
65 
 
Gini index is not noticeably different in the above-median homelessness CoCs when 
compared to those CoCs with below-median homelessness rates, this may be an early 
indication that my hypothesis that income inequality is positively and significantly 
associated with homelessness rates will not be supported. 
4.2.2: Means comparison split by Housing First index scores 
Another interesting means comparison is the means of variables for 
observations in which the Housing First index score was below median compared to 
observations in which the Housing First index score was above the median. The 
Housing First index is the independent variable of primary interest in this study, so it 
is important to understand the characteristics of CoCs that have adopted more of a 
Housing First model compared to CoCs that are relying more heavily on short-term 
shelter options when making conclusions about the outcomes of this study. Results of 
that means comparison can be found in Table 5 below. 
66 
 
Table 5: Means of variables stratified by observations with above- and below-
median Housing First index scores 
  Below-Median Above-Median 
Variable Full Sample HF Index Score HF Index Score 
People per 1,000 homeless 1.97 (.036) 2.01 (.050) 1.94 (.051) 
Total population (1000s) 729.81 (18.27) 703.76 (25.35) 767.35 (26.96) 
% Bachelor?s degree or higher 29.13 (.221) 26.96 (.241) 30.91 (.250) 
Median household income  55,411 (360) 53,629 (363) 56,356 (372) 
Median gross rent 894.72 (5.82) 852.00 (5.80) 922.27 (6.38) 
Median home value (1000s) 214.50 (2.36) 203.70 (2.84) 222.64 (3.46) 
% Renters 33.92 (.215) 32.36 (.204) 34.71 (.214) 
% Renters with children 34.26 (.204) 34.94 (.172) 32.79 (.168) 
% Renters, white non-Hispanic 63.79 (.405) 65.84 (.463) 60.80 (.489) 
% Renters, no college 45.21 (.243) 46.60 (.227) 42.79 (.242) 
Unemployment rate 8.41 (.055) 8.29 (.065) 8.44 (.062) 
CoC funding per person 5.16 (.093) 3.37 (.077) 6.90 (.155) 
Poverty rate 15.11 (.133) 14.70 (.135) 15.23 (.140) 
Eviction filing rate 6.30 (.171) 5.41 (.156) 6.73 (.198) 
% Rent-burdened 31.04 (.126) 30.32 (.065) 31.16 (.061) 
Gini index 0.46 (.002) 0.45 (.001) 0.46 (.001) 
Vacancy rate 12.65 (.126) 12.98 (.174) 11.94 (.156) 
Mean temperature in Jan 35.45 (.302) 35.72 (.465) 34.11 (.351) 
Total precipitation in Jan 2.96 (.049) 2.77 (.061) 3.07 (.068) 
CoC category* 1.96 (.008) 2.05 (.012) 1.88 (.011) 
Inclusionary zoning* 0.34 (.008) 0.32 (.012) 0.36 (.012) 
Standard errors are in parentheses. 
* Non-continuous variable. For CoC categories, 1 indicates a major city, 2 indicates a smaller town or 
county jurisdiction and 3 indicates a balance-of-state CoC. Inclusionary zoning is a dummy variable in 
which 1 indicates the adoption of an inclusionary zoning policy in at least one jurisdiction within the 
CoC and 0 indicates that no inclusionary zoning policies have been adopted within the CoC. 
Areas with Housing First index scores above the median tend to have larger 
populations with higher incomes, home values, rent payments, rates of educational 
attainment, while also experiencing higher poverty rates, higher unemployment, more 
evictions, and a higher percentage of rent-burdened households. This tells me that 
areas that have adopted more of a Housing First approach tend to be a little wealthier, 
but more of the population may be vulnerable to homelessness. The renter 
populations are also noticeably different between the two groups. In CoCs with 
67 
 
Housing First index scores above the median, a larger percentage of the population 
rents, fewer of those renters have children, fewer are white non-Hispanic, and more 
have gone to college. 
The largest difference in means that Table 6 shows me is that CoCs with 
Housing First index scores above the median receive over twice the number of grant 
dollars per person from HUD that CoCs with Housing First index scores below the 
median receive for assistance with programs and services designed to assist people 
experiencing homelessness. This makes HUD?s preference for CoCs that have 
adopted a Housing First approach clear. 
Despite these CoCs receiving over the twice the funding per person from 
HUD, CoCs with Housing First index scores above the median also experience 
slightly lower homelessness rates than CoCs with Housing First index scores below 
the median. This provides evidence that the increased funding is not linked to need, 
but rather, to a dedication to the strategies that HUD thinks are most likely to be 
effective in yielding positive result in reducing homelessness. 
 
4.3: Distributions 
Calculating the distributions of the values introduced in the means comparison 
provides a clearer understanding of the variables used in this study. To build off of 
the foundational means comparison, distributions were calculated for variable values 
with observations stratified again by the median homelessness rate. Descriptive 
statistics on the distribution of Housing First index scores and the homelessness rate 
68 
 
are provided in Table 6 below. See Appendix 3 for a more complete accounting of 
distributions for all variables. 
Table 6: Distribution descriptives for homelessness and the Housing First index 
 Above and Below Median Homelessness Rates Statistic Std. Error 
People per 1,000 Below Mean .8550 .00777 
Experiencing 95% Confidence Lower Bound .8397  
Homelessness Interval for Mean Upper Bound .8702  
5% Trimmed Mean .8600  
Median .8553  
Variance .095  
Std. Deviation .30769  
Minimum .02  
Maximum 1.40  
Range 1.38  
Interquartile Range .50  
Skewness -.137 .062 
Kurtosis -.872 .123 
Above Mean 3.0906 .05880 
95% Confidence Lower Bound 2.9753  
Interval for Mean Upper Bound 3.2059  
5% Trimmed Mean 2.7505  
Median 2.2144  
Variance 5.431  
Std. Deviation 2.33048  
Minimum 1.40  
Maximum 16.78  
Range 15.38  
Interquartile Range 1.73  
Skewness 2.677 .062 
Kurtosis 8.507 .123 
Housing First Below Mean .5371 .00436 
index 95% Confidence Lower Bound .5285  
Interval for Mean Upper Bound .5457  
5% Trimmed Mean .5424  
Median .5505  
Variance .030  
69 
 
Std. Deviation .17352  
Minimum .00  
Maximum .97  
Range .97  
Interquartile Range .24  
Skewness -.419 .062 
Kurtosis .013 .123 
Above Mean .5377 .00359 
95% Confidence Lower Bound .5306  
Interval for Mean Upper Bound .5447  
5% Trimmed Mean .5411  
Median .5437  
Variance .020  
Std. Deviation .14224  
Minimum .10  
Maximum .87  
Range .78  
Interquartile Range .20  
Skewness -.317 .062 
Kurtosis -.273 .124 
 
There are a few results in Table 6 that really stand out. The five percent 
trimmed mean for people per 1,000 experiencing homelessness is 2.75 compared to a 
mean of 3.09, which is an indication that there may be outliers in that group of 
observations. The high variance, range, interquartile range (IQR), skewness, and 
kurtosis values further validate the presence of outliers and clustering on one side of a 
leptokurtic distribution, which in this case is clustering on the low end of the values 
according to the difference between the median and the mean. To confirm these 
interpretations of the table, the distributions are visualized as histograms in Figure 3 
and Figure 4 below. 
70 
 
 
Figure 3: Distribution of homelessness rate observations below the median 
homelessness rate 
 
 
Figure 4: Distribution of homelessness rate observations above the median 
homelessness rate 
 
71 
 
As suspected, the distribution of homelessness rate observations below the 
median homelessness rate depicted in Figure 3 is relatively well-distributed while the 
distribution of homelessness rate observations above the median depicted in Figure 4 
is heavily skewed and leptokurtic due to the presence of outliers on the high end of 
the spectrum. To address this, the outliers should be identified and coerced. Coercion 
replaces outliers with the nearest value inside an acceptable range, such as within the 
outer fence, which is three times the IQR. To identify the outliers, the categories were 
removed so that the entire dataset can be analyzed, and a box plot was created to 
visualize the distribution of specific cases as shown in Figure 5. 
 
Figure 5: Box plot of homelessness rate cases 
 
Due to the number of outliers, it is difficult to identify cases to coerce based 
on the box plot alone. Additionally, a box plot shows all cases outside of the inner 
fence as potential outliers, which in this case may be too restrictive. To calculate the 
72 
 
outer fence, I need the IQR for the entire dataset. One way to determine the IQR is to 
calculate Tukey?s Hinges from the percentiles, shown in Table 7 below. 
Table 7: Homelessness rate percentiles 
Percentiles 
 5 10 25 50 75 90 95 
Weighted Average People per 1,000 .4381 .5696 .8553 1.3957 2.2147 4.0306 5.9313 
(Definition 1) Experiencing Homelessness 
Tukey's Hinges People per 1,000   .8557 1.3957 2.2144   
Experiencing Homelessness 
 
Box plots run in SPSS are based on a definition of quartiles that use Tukey?s 
Hinges as the upper and lower limits of the box. To calculate the inner and outer 
fence, the following formulas were used. 
Equation 3: Inner and outer fences 
?? = ??3 + ?????? ? 1.5 ? ??1 ? ?????? ? 1.5 
?? = ??3 + ?????? ? 3 ? ??1 ? ?????? ? 3 
 
These values are simply the IQR multiplied by 1.5 in the case of the inner 
fence and the IQR multiplied by three in the case of the outer fence, which is then 
either added to the Tukey?s Hinge value at the 75th percentile (Q3) or subtracted from 
the value at the 25th percentile (Q1). In the case of this study, the outer fence is the 
threshold used to determine whether a value is an outlier. Therefore, 6.2905 is the 
upper limit of the outer fence and the lower limit of the outer fence is a negative 
value, thereby negating the lower limit. A new variable was created by recoding 
homelessness rates into a dummy variable indicating that the case is an outlier when 
73 
 
equal to or greater than 6.2905. This identified 139 observations across 31 CoCs8 as 
outliers. The homelessness rates in these outlier values were then coerced to 6.2905 in 
a new variable called hlco so that the primary analyses may be conducted using both 
coerced and original values. In Appendix 3 of this study, the full list of outliers is 
included in Table 22, distribution descriptives for homelessness rates using coerced 
values are included in Table 23.  
An updated box plot for homelessness rates using coerced values for the 
outliers beyond the outer fence is included in Figure 6 below, and an updated 
histogram showing homelessness rates above the median using coerced values for the 
outliers beyond the outer fence is included in Figure 7. 
 
Figure 6: Box plot of homelessness rate cases using coerced values 
 
 
8 Outliers were found in CA-501, CA-504, CA-506, CA-508, CA-509, CA-522, CA-523, CA-524, CA-
603, CA-613, CA-614, DC-500, FL-501, FL-505, FL-512, FL-517, FL-518, FL-519, FL-604, GA-500, 
HI-500, LA-503, MA-500, MA-504, MD-501, MD-508, MO-602, NC-516, NY-600, NY-607, and 
OR-500. 
74 
 
 
Figure 7: Distribution of homelessness rate observations above the median 
homelessness rate using coerced values 
As shown in Figure 6, coercing values to the upper limit of the outer fence 
noticeably decreased the number of outliers. The primary analysis will still be 
performed using actual values but coercing the values of these outliers down to the 
upper limit of the outer fence will allow me to test if the results change as a result of 
questionably high homelessness rates in these outlier CoCs. Figure 7 shows that the 
number of CoCs with coerced values is significantly high. 
 
4.4: Spatial Characteristics 
The spatial distribution of homelessness rates and Housing First index scores 
in CoCs across the country is important information to consider when interpreting the 
results of the primary analysis. In this section, I have included maps showing average 
values of homelessness rates and Housing First index scores over the study period, as 
75 
 
well as maps showing how homelessness rates and Housing First index scores have 
changed between 2009 and 2017 in CoCs across the country. 
4.4.1: Homelessness rates across the United States 
The map in Figure 8 shows how homelessness rates are distributed in CoCs 
across the United States. Dashed CoCs were among the 29 removed from the analysis 
due to CoC boundary mergers that occurred over the study period. See Table 14 in 
Appendix 1: Data Collection and Model Assembly Process for more information 
about CoCs removed from the analysis. Data from the remaining 355 CoCs are 
included in the maps in this section. 
 
Figure 8: Map of average people per 1,000 experiencing homelessness by CoC: 
2009 to 2017 
76 
 
As shown in Figure 8, homelessness rates are particularly high in California, 
Florida, and Hawaii. This intuitively lends support to the notion that homelessness 
rates are higher in these areas due to their favorable climate. However, other CoCs in 
the south experience very low homelessness rates, and higher homelessness rates 
extend farther north along the west coast where temperatures are considerably colder, 
so this tells us that other factors are likely influencing homelessness rates in a 
significant way. 
 
Figure 9: Map of change in people per 1,000 experiencing homelessness by CoC 
from 2009 to 2017 
Figure 9 shows that homelessness rates have gone up the most over the course 
of the study period in Montana, Wyoming, the Dakotas, Puerto Rico, and parts of 
77 
 
New England. Large decreases have occurred in parts of Florida, California, Utah, 
Arizona, Washington, Colorado, and Georgia.  
Those balance-of-state and other large multicounty CoCs stand out, but 
looking more closely at smaller CoCs, the distribution of CoCs with increasing or 
decreasing rates of homelessness do not appear to be spatially concentrated in any 
meaningful way. This tells me that factors that are specific to certain regions of the 
country, like precipitation and temperature, are not largely responsible for changes in 
homelessness rates. 
4.4.2: Housing First index scores across the United States 
The map in Figure 10 below shows the degree to which Housing First is 
implemented in CoCs across the United States. Again, dashed CoCs were among the 
29 removed from the analysis due to CoC boundary mergers that occurred over the 
study period. 
78 
 
 
Figure 10: Map of average Housing First index scores by CoC: 2009 to 2017 
As shown in Figure 10, a Housing First approach seems to be used by CoCs in 
Minnesota, Georgia, and parts of New York, California, Puerto Rico, and Florida, as 
well as other small CoCs scattered across the country. 
Places that do not rely on a Housing First approach as strongly seem to be 
concentrated somewhat in the south and the mountainous regions of the west. 
However, some smaller, more densely populated CoCs are the exception to this rule. 
This map reveals that there may be some geographic component to the 
distribution of the Housing First approach across the United States, because smaller 
CoCs seem to generally implement Housing First to a greater degree than larger 
balance-of-state CoCs. 
79 
 
 
Figure 11: Map of change in Housing First index scores by CoC from 2009 to 2017 
Figure 11 shows how Housing First index scores have changed in CoCs across 
the United States. It is difficult to discern a spatial pattern to the changes. Housing 
First index scores have increased significantly in some CoCs where the average score 
was low over the course of the study period, such as West Virginia, Mississippi, and 
Montana. In other places, like Alabama and South Dakota, the index score was low 
on average and does not appear to be increasing. There does not seem to be any 
identifiable pattern across smaller CoCs either. 
This suggests that a CoC?s decision to implement a Housing First approach is 
likely to be a decision made independently of their neighbors. The panel analysis used 
in this study helped to shed light on how that decision may have influenced 
80 
 
homelessness rates in these CoCs, and on the relationship between homelessness rates 
and other variables identified by previous literature to be significant. 
 
  
81 
 
Chapter 5: Panel Analysis 
This study uses a linear mixed model procedure to conduct a panel analysis, 
analyzing the relationship between homelessness rates and a Housing First index 
created for this study along with other variables thought to be associated with 
homelessness from the year 2009 to 2017 in CoCs across the United States. The 
benefits of the linear mixed model procedure and panel analyses more generally are 
described in Chapter 3, along with the methodology used in this study. This chapter 
will describe the results of the panel analysis. 
 
5.1: Initial Model 
All the factors and covariates were included in main effects testing in this 
initial model. In the specifications for the model?s random effects, only the intercept 
and a single interaction term for the interaction between the recoded continuous 
variable for year and the CoC name variable were included. The CoC number and 
category variables were included as a subject grouping to test the random effects. The 
autoregressive covariance type was selected again for the random effect, because I 
expect a random effect on a residual to be most closely related to the random effect 
on a residual within the same CoC in the previous year. By including these random 
effects in the model, I can separate out the portion of the error in each observation 
that is due to the year and CoC in which an observation was made. However, the 
fixed effects results are the subject of this study. 
82 
 
Table 8: Type III tests of fixed effects in the first round of main effects testing 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 6859.020 .103 .748 
coccat 2 311.003 4.061 .018 
inczon 1 320.766 1.049 .307 
pop 0 . . . 
bach 1 810.553 3.016 .083 
medinc 1 803.556 .468 .494 
medren 1 922.549 18.395 .000 
medval 1 828.139 10.432 .001 
renfam 1 666.197 .540 .463 
renocc 1 545.749 .164 .686 
renwhi 1 585.687 3.854 .050 
renedu 1 746.652 5.821 .016 
unemp 1 1647.906 1.409 .235 
pov 1 1320.456 .779 .378 
evic 1 1422.967 .114 .736 
burd 1 1729.809 6.754 .009 
gini 1 1002.101 .046 .830 
vac 1 452.685 2.584 .109 
temp 1 1837.697 2.424 .120 
precip 1 1696.294 1.837 .176 
hf 1 1977.624 .299 .584 
fund 1 753.774 24.652 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 633.257 .508 .476 
evic2 1 1880.448 .146 .702 
hf2 1 1966.959 1.150 .284 
yearcoded 1 1339.173 23.222 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
As seen in Table 8, all of the variables discussed in the literature and in 
Chapter 2 of this study were tested in the panel analysis for a significant relationship 
with homelessness rates. Most of the independent variables have been used in past 
modeling studies, some more than others. The median rent is used frequently. 
83 
 
Unemployment, poverty, and the vacancy rate are also used fairly frequently. The 
rest, aside from the three new variables, have been used at least once in a past 
modeling study. For more detail on the model selection process, please see Appendix 
2: Model Selection Process. 
 
5.2: Final Model Results 
The information criteria scores, type III tests of fixed effects results, and fixed 
effects estimates for Model 159 are included in Table 9, Table 10, and Table 11 
below. Model 159 was also run on subsets of homelessness9 as the dependent 
variable, and the results from those models are included in Appendix 6: Final Model 
Results for Subsets of Homelessness. The information criteria measure the quality of 
the model in measuring relationships between variables in an existing dataset, and 
lower scores indicate a better model. The scores are not meaningful by themselves; 
they are only meaningful in relation to other models analyzing the same dataset.  
  
 
9 Subsets of homelessness included in Appendix 6 include sheltered, unsheltered, families with 
children, chronic, and veterans. 
84 
 
Table 9: Information criteria for final model 
Information Criteriaa 
-2 Restricted Log Likelihood 6496.511 
Akaike's Information Criterion (AIC) 6500.511 
Hurvich and Tsai's Criterion (AICC) 6500.516 
Bozdogan's Criterion (CAIC) 6514.262 
Schwarz's Bayesian Criterion (BIC) 6512.262 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 10: Type III tests of fixed effects for final model 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 731.017 10.247 .001 
inczon 1 370.459 3.340 .068 
coccat 2 358.777 6.309 .002 
pop 0 . . . 
bach 1 567.519 38.995 .000 
medval 0 . . . 
renwhi 1 452.256 5.032 .025 
renedu 1 804.002 21.964 .000 
pov 1 978.303 11.100 .001 
vac 1 692.624 2.914 .088 
temp 1 2576.312 4.665 .031 
hf 1 1467.766 6.315 .012 
fund 1 849.387 25.282 .000 
pop2 0 . . . 
hf2 1 1999.958 9.052 .003 
yearcoded 1 1666.435 65.213 .000 
renocc * hf 1 1016.894 13.133 .000 
renocc * hf2 1 1654.524 16.773 .000 
medren * vac 1 707.213 9.298 .002 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
The Type III tests of fixed effects table displays results similar to the full table 
of fixed effects estimates, except that it measures the overall significance of 
85 
 
categorical variables, instead of separately measuring the significance of each 
individual category with homelessness rates. This is helpful in the case of the models 
compared in this study, because the CoC category variable and the inclusionary 
zoning variable are both categorical variables. 
 
Table 11: Fixed effects estimates for final model 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.416004 1.175257 677.533 2.907 .004 1.108421 5.723587 
[inczon=0] -.315321 .172524 370.459 -1.828 .068 -.654570 .023928 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .903861 .340010 358.375 2.658 .008 .235194 1.572527 
[coccat=2] .056515 .302442 358.997 .187 .852 -.538265 .651296 
[coccat=3] 0b 0 . . . . . 
pop -6.285754E-7 1.872585E-7 360.187 -3.357 .001 -9.968326E-7 -2.603181E-7 
bach -.080024 .012815 567.519 -6.245 .000 -.105194 -.054853 
medval 5.724649E-6 7.762778E-7 671.692 7.374 .000 4.200426E-6 7.248872E-6 
renwhi .010954 .004884 452.256 2.243 .025 .001357 .020552 
renedu -.051451 .010978 804.002 -4.687 .000 -.073000 -.029901 
pov .072196 .021670 978.303 3.332 .001 .029672 .114721 
vac -.052941 .031015 692.624 -1.707 .088 -.113835 .007953 
temp .005839 .002703 2576.312 2.160 .031 .000538 .011141 
hf -4.751981 1.890942 1467.766 -2.513 .012 -8.461218 -1.042743 
fund .066042 .013135 849.387 5.028 .000 .040262 .091822 
pop2 7.07668E-14 2.28433E-14 359.300 3.098 .002 2.584347E-14 1.156903E-13 
hf2 6.556069 2.179028 1999.958 3.009 .003 2.282666 10.829472 
yearcoded -.128228 .015879 1666.435 -8.075 .000 -.159372 -.097083 
renocc * hf .192248 .053048 1016.894 3.624 .000 .088151 .296345 
renocc * hf2 -.256817 .062707 1654.524 -4.095 .000 -.379810 -.133823 
medren * vac 9.134685E-5 2.995657E-5 707.213 3.049 .002 3.253240E-5 .000150 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
 
86 
 
The fixed effects estimates table includes the estimate or coefficient 
associated with each variable, the standard error, degrees of freedom of the 
denominator, a t score, the statistical significance of each variable, and the confidence 
interval values. The most important results for answering my research question are 
the coefficients and the significance. The coefficient represents the value change in 
homelessness rates with an increase of one in the measured variable, just like a 
coefficient in a standard regression analysis. The significance value, or p score, is the 
result of a statistical test in which a value below 0.05 commonly justifies rejecting the 
null hypothesis. This was the threshold used in this study as well. 
The results of the final model are discussed in terms of main effects, 
interaction effects, and how they answer the research question. Since the results 
related to the Housing First index are critical to understanding how the model 
answers the research question, I will start with those. 
5.2.1: Results of the Housing First index 
The relationship between the Housing First index (hf) and homelessness rates 
was measured in four different ways in the final model. This is because an interaction 
between the Housing First index and the percentage of renter-occupied households 
was included in the final model, and because the Housing First index was found to 
have a quadratic relationship with homelessness rates in which homelessness rates 
were lowest either when the Housing First index was very low or very high. 
Therefore, in Table 11, hf represents the relationship between the Housing First index 
and homelessness rates for base values of the Housing First index, and hf2 represents 
the relationship between squared values of the Housing First index and homelessness 
87 
 
rates. Refer to Figure 12 for a representation of the quadratic relationship between the 
Housing First index and homelessness rates. 
Both the hf and the hf2 variables have p scores below 0.05 in the final model, 
indicating that the Housing First index is associated with homelessness rates in a 
statistically significant way. Likewise, the interaction effect between the percentage 
of renter-occupied households (renocc) and the Housing First index is also associated 
with homelessness rates in a statistically significant way. The introduction of the 
interaction effect also flipped the coefficient of the Housing First index so that the hf 
coefficient is now negative and the hf2 coefficient is now positive. 
The hypothesis that homelessness rates would decrease as the Housing First 
index increases is somewhat supported by the results, but the relationship is more 
complex than hypothesized. Instead, the final model used in this study tells me that 
the relationship between the Housing First index and homelessness rates can be 
simplified by controlling for other effects in the model in Equation 4 below. 
Equation 4: Estimating homelessness rates by the Housing First index and its 
interaction with the percentage of renter-occupied households 
??? = 3.4160 ? 4.7520(???) + 6.5561(???2) + .1922(??? ? ????????????)
? .2568(???2 ? ????????????) 
 
In other words, ceteris paribus, a jurisdiction with 50 percent renter-occupied 
households and a Housing First index score of 0.50 would be expected to have a 
homelessness rate of 4.2740, or 4.2740 people experiencing homelessness per 1,000 
people living in the CoC boundary. If the Housing First index score is reduced to 
0.25, the estimated homelessness rate becomes 4.2378. If the Housing First index 
88 
 
score is increased to 0.75, the estimated homelessness rate becomes 3.5248. The 
estimated homelessness rate decreases very slightly as the Housing First index score 
decreases from 0.50, and the homelessness rate decreases much more substantively as 
the Housing First index score increases from 0.50. 
Since there is a statistically significant interaction effect between the Housing 
First index and the percentage of households that are renter-occupied in the final 
model as well, the relationship between Housing First and homelessness rates is 
altered by changes in the percentage of renter-occupied households. For example, 
with a Housing First index of 0.50, a drop in the percentage of renter-occupied 
households from 50 percent to 25 percent changes the estimated homelessness rate 
from 4.2740 to 3.4765. Likewise, with an increase in the percentage of renter-
occupied households to 75 percent, the estimated homelessness rate changes to 
5.0715. When the Housing First index is 0.25, a renocc value of 25 decreases the 
estimated homelessness rate from 4.2378 to 3.4378, and a renocc value of 75 
increases the estimated homelessness rate to 5.0378. Finally, when the Housing First 
index is 0.75, a renocc value of 25 slightly increases the estimated homelessness rate 
from 3.5248 to 3.5323, and a renocc value of 75 slightly decreases the estimated 
homelessness rate to 3.5173. 
Table 12: Estimated homelessness rates for values of the Housing First index and 
percentage of renter-occupied households 
 renocc = 25 renocc = 50 renocc = 75 
hf = 0.25 3.4378 4.2378 5.0378 
hf = 0.50 3.4765 4.2740 5.0715 
hf = 0.75 3.5323 3.5248 3.5173 
 
89 
 
My interpretation of these results is that the CoCs that have done very little to 
implement a Housing First approach tend to have slightly lower homelessness rates 
than jurisdictions that have done some implementation, because the slightly lower 
homelessness rates do not encourage the political will to adopt a Housing First 
approach. In other words, in CoCs where the Housing First index is low, the 
homelessness rates are driving the Housing First index, as opposed to the Housing 
First index driving the homelessness rate. In CoCs that are leading the country in 
adoption of a Housing First approach, as indicated by higher Housing First index 
scores, homelessness rates are substantially lower than both jurisdictions that have 
done very little implementation and jurisdictions with average index scores, and in 
those CoCs it seems that the implementation of a Housing First approach is driving 
the lower homelessness rates. 
Interestingly, a higher percentage of renter-occupied households seems to 
result in substantively higher estimated homelessness rates, except when the Housing 
First index is high. The implementation of a Housing First approach seems to nullify 
the association between renter-occupied households and homelessness rates. This 
could be an indication that while a larger renting population generally means that a 
higher proportion of the population is unstable in their housing and more vulnerable 
to episodes of homelessness, that population can be stabilized and protected from 
episodes of homelessness by adopting a Housing First approach. This approach 
makes it possible to help renters in a financial or other type of crisis that may have 
otherwise resulted in homelessness by immediately moving them in to new 
90 
 
permanent supportive housing or a housing unit without associated supportive 
services through a rapid rehousing program. 
5.2.2: Other main effects results and interpretation 
The inclusionary zoning variable (inczon) was pushed beyond the 0.05 p value 
significance threshold but was still fairly significant with a p value of 0.068. Counter 
to my hypothesis, the adoption of an inclusionary zoning policy was positively 
associated with homelessness rates. This could be due to higher homelessness rates 
driving the adoption of inclusionary zoning policies. However, the adoption of an 
inclusionary zoning policy only increases the estimated homelessness rate by 
approximately 0.3 people experiencing homelessness per 1,000 residents and the 
variable is insignificant at the 0.05 p value level. Therefore, the results show that this 
variable does not have a very substantively meaningful relationship with 
homelessness rates. This could be due to incomplete data in the Grounded Solutions 
database at the time that data were exported and a lack of nuance by treating the 
adoption of an inclusionary zoning policy as a dummy variable when there is a fair 
amount of complexity in the differences between inclusionary zoning policies as they 
are adopted across the country that is not considered in the available data. 
The eviction filing rate (evic) was removed from the final model because the 
variable?s relationship with homelessness rates was statistically insignificant. It also 
did not have a statistically significant relationship with homelessness rates in an 
interaction term with another variable. It is possible that this absence of a relationship 
was due to the incompleteness of these data, so I propose future research expanding 
91 
 
this database so that the relationship can be studied more effectively in the future. I 
include more discussion about this in Chapter 6. 
The CoC category (coccat) is significant, but not each category is statistically 
significant. The difference between Category 1, which indicates a major city CoC, 
and Category 3, which indicates a balance-of-state CoC, is statistically significant. 
However, there is no statistically significant difference between homelessness rates in 
Category 2, which indicates a smaller town or county CoC, compared to 
homelessness rates in other CoC categories. Homelessness rates are about 0.9 people 
per thousand higher in major city CoCs than in balance-of-state CoCs. Since housing 
affordability, rent, race, poverty, and other variables that tend to highlight 
socioeconomic differences between rural and urban areas were included in the model, 
and furthermore, no significant interaction effects including CoC categories emerged 
from the model refinement process, I consider it more likely that this difference is 
attributable to the way that the dependent variable is measured. Homelessness is 
measured through the point-in-time (PIT) count, which involves a collection of 
volunteers that the CoC is responsible for recruiting counting people experiencing 
homelessness in the CoC. Both balance-of-state CoCs and major city CoCs are likely 
to weigh their count data based on a sample, but the increased area to cover in a 
balance-of-state CoC would make an undercount much more likely due to the 
increased difficulty in coordinating the PIT count. 
Total population (pop) is statistically significant and negatively associated 
with homelessness, but not substantively relevant. A CoC with 1 million residents has 
experiences a 0.08 reduction in their homelessness rate compared to a CoC with 10 
92 
 
thousand residents. The effect of total population on homelessness rate estimates is 
likely always spurious in nature and is nullified in the model by controlling for many 
of the factors typically associated with changes in total population. 
The percentage of the population with a bachelor?s degree or higher (bach) 
was statistically significant and negatively associated with homelessness. Ceteris 
paribus, a 10 percent increase in the percentage of the population with a bachelor?s 
degree or higher is estimated to decrease the homelessness rate by approximately 0.8 
people per thousand. This could mean that populations more highly educated 
residents are more likely to politically support effective solutions to homelessness. 
Median value of an owner-occupied home (medval) is statistically significant 
and positively associated with homelessness rates. A $10,000 increase in the median 
value of an owner-occupied home results in an increase of 0.057 person experiencing 
homelessness per thousand residents in a CoC. This is not a very substantive increase, 
largely because renter-occupied households are more vulnerable to homelessness and 
the vulnerability of renter-occupied households is more directly measured by other 
variables in the model. 
The percentage of renters identifying as white, non-Hispanic (renwhi) is 
statistically significant and positively associated with homelessness rates, but it is not 
very substantively meaningful. A 10 percent increase in the percentage of renters 
identifying as white, non-Hispanic only increases the estimates homelessness rate by 
0.110 people experiencing homelessness per thousand residents. The positive 
association is interesting nonetheless, since black and Hispanic populations are 
disproportionately experiencing homelessness (HUD, 2018a). CoCs with a larger 
93 
 
proportion of non-white residents tend to do a better job of safeguarding their 
residents from homelessness, ceteris paribus, and yet non-white residents are much 
more likely to experience homelessness. This seems to mean that CoCs with a higher 
proportion of non-Hispanic white residents, even when they are renters who are more 
vulnerable to homelessness than owner-occupied households, are less likely to 
prioritize strategies that prevent or end peoples? homelessness, and non-white 
residents in those jurisdictions are the people to disproportionately experience 
homelessness. 
The percentage of renters without any college education (renedu) is 
statistically significant and negatively associated with homelessness rates. This seems 
to run counter to the variable measuring the percentage of residents with a bachelor?s 
degree or higher (bach) that is also negatively associated with homelessness, but the 
variables are measuring education in different populations. The bach variable is 
measuring educational attainment across all residents while the renedu variable is 
only measuring educational attainment among renters. This model seems to tell me 
that high levels of educational attainment in the overall population, particularly 
among owner-occupied units helps alleviate homelessness, while high levels of 
educational attainment among renters seems to exacerbate the problem of 
homelessness. My interpretation of these results is that the renedu variable is picking 
up a spurious relationship with homelessness rates due to the absence of a variable 
measuring rental affordability like the medren or burd variables did before they were 
removed due to statistical insignificance, and that a higher level of educational 
94 
 
attainment among the renter population is more indicative of an unaffordable housing 
stock than higher levels of educational attainment across the population as a whole. 
The poverty rate (pov) is statistically significant and positively associated with 
homelessness rates. With a 10 percent increase in the poverty rate, the estimated 
homelessness rate would increase by 0.722 people experiencing homelessness per 
thousand residents in the CoC. As people become more financially vulnerable, they 
are more susceptible to homelessness. This result did not line up with my hypothesis 
that poverty and income would not be statistically significant, but that income 
inequality, as measured through the Gini index (gini), would be statistically 
significant and positively associated with homelessness rates. Instead, the Gini index 
was removed from the model for statistical insignificance and the poverty rate was 
statistically significant and positively associated with homelessness rates. 
The vacancy rate (vac) is not statistically significant at the 0.05 p score level, 
but it is still significant below the 0.10 p score level, and the vacancy rate is 
negatively associated with homelessness rates. The results did not support my 
hypothesis that the vacancy rate would be statistically insignificant. The negative 
association with homelessness rates is likely due to the fact that vacancy rates are low 
in tight housing markets, where the most vulnerable households are less likely to find 
housing that they can afford. 
The mean temperature in January (temp) is statistically significant and 
positively associated with homelessness rates. The results did support my hypothesis, 
except that I also hypothesized that precipitation would be another statistically 
significant variable and that it would be negatively associated with homelessness. 
95 
 
Precipitation was removed from the model during the model refinement process due 
to statistical insignificance. While temperature is statistically significant, it is not very 
substantively relevant. With a 10 degree (Fahrenheit) increase in mean January 
temperature, the estimated homelessness rate only increases by about 0.058 people 
experiencing homelessness per thousand residents in the CoC. Therefore, though 
people experiencing homelessness who are able to travel may seek out a place with 
milder winters, the proportion of people experiencing homelessness in CoCs with 
warmer temperatures who traveled there for the temperature is miniscule. 
HUD CoC funding in the previous year per person (fund) is statistically 
significant and positively associated with homelessness rates. For every dollar 
increase in CoC funding per resident, the estimated homelessness rate increases by 
0.066 people experiencing homelessness per thousand residents. This is likely 
because HUD awards more funding to CoCs with larger homelessness problems, so 
homelessness is driving funding as opposed to funding driving homelessness. 
Lastly, the year for which data were collected (yearcoded) was statistically 
significant and negatively associated with homelessness. In the linear mixed model, 
the year was included as the repeated measures variable with an autoregressive 
covariance structure to ensure that the model accounted for the fact that observations 
in the same CoC were not considered to be independent from one another and that 
each subsequent observation was most likely to be correlated with the previous 
observation in the same CoC. By including the year as its own independent variable 
in the model as well, I can attribute a decrease of 0.128 people experiencing 
homelessness per thousand each year over the study period. This is a substantial 
96 
 
decrease, and it is due to forces otherwise outside the scope of the variables studied in 
this model. This tells me that strategies being implemented across the country are 
working. I know that the Housing First approach is gaining traction and the model 
used in this study tells me that increases in the Housing First index does have a 
statistically significant and substantive association with a decrease in homelessness 
rates. The year variable could be picking up on characteristics of the Housing First 
approach that this study leaves out, such as reducing barriers to housing and 
improving access to supportive services. The index used in this study does not know 
if a person experiencing homelessness had to graduate from a religious program 
before the permanent supportive housing unit became available to them. The index 
only knows the proportion of units in a CoC that are permanent supportive housing 
units and uses that information to assume that barriers are breaking down, because 
programs with a lot of barriers in place would typically need a higher proportion of 
emergency shelter and transitional housing units. Therefore, the change in 
homelessness rates attributable to the year may be picking up on progress that the 
index does not.  
5.2.3: Other interaction effects results and interpretation 
The only other significant interaction effect in the final model besides the 
interaction effect involving the Housing First index (hf) and the percentage of renter-
occupied households (renocc) discussed above was an interaction effect between the 
median gross rent of renter-occupied households (medren) and the vacancy rate (vac). 
The main effect of the vacancy rate remained in the final model, but the median gross 
rent of renter-occupied households was removed due to statistical insignificance, and 
97 
 
it remained in the final model only through its statistically significant interaction with 
vacancy rates. 
The equation for this interaction effect is less complex than the interaction 
between hf and renocc, because neither of the variables in this interaction have a 
relationship with homelessness rates best expressed by a polynomial, such as the 
quadratic relationship between the Housing First index and homelessness rates. 
Instead, the interaction between medren and vac is simplified by controlling for other 
effects in the model in Equation 5 below. 
Equation 5: Estimating homelessness rates by the vacancy rate and its interaction 
with the median gross rent of renter-occupied households 
??? = 3.4160 ? .05294(??????) + .00009(?????? ? ????????????) 
 
This shows that changes in median gross rent affects the relationship between 
vacancy rates and homelessness rates. In other words, ceteris paribus, increases in the 
vacancy rate will be associated with homelessness rates differently depending on the 
median gross rent. See Table 13 below for examples of how changes in each value of 
this interaction term result in changes in estimated homelessness rates. 
Table 13: Estimated homelessness rates for values of the vacancy rate and median 
gross rent 
 medren = 500 medren = 800 medren = 900 medren = 1000 
vac = 10 3.3366 3.6066 3.6966 3.7866 
vac = 12 3.3207 3.6447 3.7527 3.8607 
vac = 14 3.3048 3.6828 3.8088 3.9348 
 
The interaction between vacancy rates and median gross rent illuminates a 
very interesting caveat to the negative association between homelessness rates and the 
vacancy rate. Inclusion of this interaction term shows me that homelessness rates only 
98 
 
decrease as the vacancy rate increases when the median gross rent in a CoC is very 
low. Otherwise, the homelessness rate estimated by my final model actually increases 
as the vacancy rate increases. 
This provides evidence that the vacancy rate is actually more of a metric for 
the gap between supply and demand of affordable housing than it is an indication of a 
tight housing market. In other words, as median gross rent increases and the positive 
relationship between the vacancy rate and the homelessness rate becomes more 
substantive, the model shows that the available housing stock is built or renovated for 
a shrinking proportion of potential renters in the market. Therefore, vacancy is 
increasing not because peoples? housing demands are being met, but because peoples? 
housing demands are being ignored by the development community. 
5.2.4: Implications of the final model 
The results of the final model show that it was an important choice to include 
interaction effects in the analysis, because two of these interaction effects turned out 
to be significant in answering the research question. They add nuance to an 
understanding of the relationship between homelessness and Housing First as well as 
the relationship between homelessness and other variables deemed by prior literature 
to be a significant determinant of homelessness rates. 
Without needing to force the inclusion of the Housing First index in the 
model, the index turned out to have a statistically significant, substantively relevant, 
and nuanced relationship with homelessness rates. I will now cover the implications 
of that relationship in the conclusion. 
99 
 
Chapter 6: Conclusion 
This chapter summarizes the key takeaways from this study, including the 
answer to the research question, policy recommendations as a result of the findings, 
limitations of this study, and some suggested avenues for future research. 
 
6.1: Housing First and Homelessness 
HUD is giving preference to CoCs that implement a Housing First approach to 
ending homelessness in their communities, and it is important to know if that 
preference is justified. Some of the most vulnerable members of our society depend 
on professionals responsible for homelessness services using the limited resources 
available to them to most effectively reduce and end homelessness. 
Across the country?s CoCs, this study found that those that implemented a 
Housing First approach to ending homelessness saw a statistically significant and 
relatively substantive decrease in estimated homelessness rates in most, but not all, 
cases. As a reminder, the research question of this study was the following: 
Are Continuums of Care (CoC) that have adopted a Housing First approach 
by dedicating a higher proportion of their resources towards permanent housing and 
support services associated with a lower proportion of people experiencing 
homelessness between the years 2009 and 2017 than CoCs that dedicate a higher 
proportion of their resources towards emergency shelter and other short-term 
solutions? Additionally, how does that relationship between the implementation of a 
Housing First approach and homelessness rates change as the values of median rent, 
unemployment, and other covariates typically associated with homelessness rates 
change? 
I found in this study that the relationship between the Housing First index, 
used to measure the degree to which a CoC had implemented a Housing First 
100 
 
approach, and homelessness rates was more complex than originally hypothesized. In 
areas with low proportions of renter-occupied households, the Housing First approach 
seems to be ineffective in reducing estimated homelessness rates as predicted by this 
study?s final model. Implementation of the Housing First approach typically 
manifests as an intervention in the rental market, so it stands to reason that the 
approach would be less impactful in places with a large majority of owner-occupied 
households. 
Across the observations used in this study, the median percentage of renter-
occupied housing units was about 32 percent, and the percentages ranged from 12.56 
at the lowest to 69.22 at the highest. In a CoC with the median percentage of renter-
occupied households, the Housing First index begins to decrease estimated 
homelessness rates after the index score increases above 0.42. To provide more 
clarity as to what this means, the median Housing First index score was 0.55 and 
ranged from 0 to 0.97. So in a CoC with a median percentage of renters and a median 
Housing First index score, increasing the degree to which that CoC adheres to a 
Housing First approach will decrease the estimated homelessness rate produced by 
the final model of this study. However, in a CoC with the lowest observed percentage 
of renters, at 12.56 percent, increasing the Housing First index never decreased 
estimated homelessness rates. The formula produced by the final model of this study 
estimates that the percentage of renters living in a CoC needs to be at least 27 percent 
before the homelessness rate is estimated to decrease as a result of an increase in the 
Housing First index score from the median of 0.55 to 0.65. 
101 
 
Also, as the Housing First index increases in a CoC, the estimated 
homelessness rate rises slightly before decreasing more substantially. This quadratic 
relationship with homelessness is likely because homelessness needs to rise to a 
certain level of severity before CoCs begin to engage in more sophisticated 
approaches to ending homelessness in their communities than emergency shelters. So, 
homelessness rates drive the Housing First index until the index gets to about 0.42, 
and then the Housing First index begins to drive homelessness rates down more 
substantially in CoCs that have dedicated themselves to taking intentional steps in 
following a Housing First model. 
The relationship between Housing First and homelessness was more complex 
than I hypothesized. The results show that implementation of a Housing First 
approach will be associated with decreases in homelessness in most cases. In about 
21.2 percent of the observations used in this study, the percentage of renter-occupied 
households was under 27 percent and increases in the Housing First index do not 
relate to decreases in the estimated homelessness rates in those cases. In the 
remaining 78.8 percent of observations, increases in the Housing First index does 
relate to estimated decreases in homelessness with the decrease becoming larger as 
the percentage of renters living in the CoC increases. For example, in a CoC with a 
million residents and 50 percent renters, an increase in the Housing First index score 
from 0.5 to 0.8 decreases the estimated number of people experiencing homelessness 
by 993 people. 
The relationship explained by this research helps to shed some light on the 
question posed by O?Flaherty (2019) by providing a foundation from which 
102 
 
researchers can begin to understand how Housing First is related to homelessness. 
The nuance of this relationship should be studied further by researchers interested in 
understanding homelessness and taken into account when developing policy 
recommendations that are flexible enough to account for differences in local 
conditions. 
 
6.2: Policy Recommendations 
Despite the complexity of the relationship between Housing First and 
homelessness rates, the results of this study support continued investment in the 
Housing First approach. Due to the finding that increases in the Housing First index 
are not associated with decreases in estimated homelessness rates in the 21.2 percent 
of observations where the percentage of renters was under 27 percent, HUD should 
send representatives specializing in homelessness alleviation to CoCs where the 
percentage of renters is under 27 percent to better understand the challenges service 
agencies face and to provide technical assistance. HUD should also sponsor 
additional research to better understand the relationship between the Housing First 
approach and homelessness rates in areas with a large percentage of owner-occupied 
units in the housing market. I will discuss this more in the recommendations for 
future research. 
In the 78.8 percent of observations where the percentage of renters was at 
least 27 percent, this study found that increases in the Housing First index score were 
associated with estimated decreases in the homelessness rate. In CoCs where the 
percentage of renters is over 27 percent, HUD should also send representatives 
103 
 
specializing in homelessness alleviation who will provide additional technical 
assistance, particularly to those with low Housing First index scores, to ensure that 
they have agencies that are proficient in providing permanent supportive housing, 
rapid re-entry programs, case management, coordinated entry services, and other 
functions necessary for successful implementation of the Housing First model. Over 
the span of the next several funding cycles, HUD should require CoCs to transition 
from a model that utilizes a high proportion of emergency shelter and transitional 
housing to a Housing First model that utilizes a large proportion of permanent 
housing units, especially in CoCs with low Housing First index scores or high 
percentages of renter-occupied households. Plans laid out in CoC applications for 
funding should be reviewed critically to ensure that all necessary partner agencies are 
committed, and to ensure that the plan is feasible. Each plan should include a 
monitoring and evaluation component to ensure that important transition benchmarks 
are met and to evaluate whether implemented strategies are working as designed. 
This study conducted an analysis with the best data available, but the quality 
of future homelessness research and an understanding of homelessness in the United 
States would be improved if researchers could rely on the availability of high-quality 
homelessness data. HUD should require that CoCs conduct a PIT count every year 
instead of every two years in order to apply for CoC grant funding. Many CoCs 
already conduct a new count every year, and this would improve the ability to 
conduct research related to homelessness and evaluate the efficacy of programs and 
funded projects in each CoC. HUD should also provide more oversight to ensure that 
CoCs are using consistent methodologies to conduct their PIT counts and that the 
104 
 
counts are being conducted as their methodologies state they will be. HUD should 
send a representative to monitor counts every three years and to require corrective 
actions based on any deficiencies in the way the PIT count is conducted. 
This study found that in most cases, following a Housing First approach is 
associated with lower homelessness rates. Implementing the Housing First approach 
requires the availability of affordable housing units that can be used in a rapid 
rehousing program or as permanent supportive housing for people who need wrap-
around services. Across the country and in most CoCs, housing markets suffer from 
an affordable housing shortage. The Housing First approach cannot be effectively 
implemented without increasing the supply of affordable units. The next two policy 
recommendations address this need. 
State housing agencies that administer housing credits from the low-income 
housing tax credit (LIHTC) program to give additional preference to applications that 
include a percentage of units available to people experiencing homelessness, either as 
permanent supportive housing units or units available as part of rapid re-housing 
program. This provision should be included in the qualified allocation plan (QAP) by 
the housing credit agency that most often operates at the state level as part of the 
selection criteria for projects applying for housing credits. State housing agencies 
should work with the National Council of State Housing Agencies (NCSHA), the 
Internal Revenue Service (IRS), HUD, and their Congressional representatives to 
ensure that a coordinated approach to addressing homelessness through the LIHTC 
program is developed. Furthermore, additional housing credits should be made 
available to areas with the greatest shortages of affordable housing. 
105 
 
To further expand the inventory of affordable housing while leveraging the 
existing housing stock, the federal legislature should eliminate the sequestration 
spending caps and increase funding for Housing Choice Vouchers. Additionally, in 
places where a high percentage of vouchers are being returned due to an inability of 
the recipient to find a unit, the process for determining Fair Market Rents, approving 
units, and investigating Fair Housing complaints should be critically evaluated to 
ensure that every voucher recipient is given a fair opportunity to acquire housing. 
Implementing these policy recommendations would put the country in a much 
better position to address the problem of people experiencing homelessness in many 
communities. 
 
6.3: Limitations of this Study 
When considering the results of this study, the following limitations should be 
taken into consideration. 
The most significant limitation of this study is one that has affected all studies 
of homelessness to date, which is the quality of homelessness data available for 
research. Implementation of my policy recommendations related to the PIT count 
would help to alleviate this issue in the future. 
Another limitation of this study is that I was forced to leave important 
elements of the Housing First approach out of the Housing First index used in this 
study due to the scope. Longitudinal studies of the efficacy of the Housing First 
approach conducted in the past have evaluated specific programs to ensure that the 
program has eliminated barriers to housing and is providing appropriate levels of 
106 
 
voluntary wrap-around services before the program can even qualify as following a 
Housing First model. Since this study was conducted by a single researcher 
evaluating data from every CoC in the country, that level of detail was not possible. 
The Housing First index used in this study instead looked at the proportion of units 
available to people experiencing homelessness in each CoC, and assumed that CoCs 
with a higher proportion of permanent housing units was following more of a 
Housing First approach, because a CoC with many programs that enforced strict 
barriers to housing would be forced to utilize a higher proportion of emergency 
shelter and transitional housing units, because people would not make it through the 
programs to the point when they are placed in a permanent housing unit. There are 
likely some CoCs in which this assumption is violated, and in those cases the 
Housing First index is too simple of a tool for measuring the degree to which that 
jurisdiction is truly following a Housing First approach. Case study research could 
help to uncover when the Housing First index is helpful as a tool and when it is not. 
The final model used in this study may unfortunately suffer from endogeneity 
bias to some degree. After interpreting the results of this study, it is possible that in 
CoCs where the Housing First index is low, the homelessness rates are driving the 
Housing First index, as opposed to the Housing First index driving the homelessness 
rate. I did not control for the possibility that the dependent variable may be 
influencing independent variables. In prior research in the field, this possibility was 
dismissed due to the size of the population of people experiencing homelessness 
being too small to influence socio-economic variables in the community in which 
107 
 
they live. However, because the Housing First approach is a direct response to 
homelessness, that concern should not be dismissed. 
These limitations should all be taken into account when considering the 
results of this study, and future research should attempt to mitigate for as many of 
these limitations as possible. 
 
6.4: Scholarly Implications 
The results of this study provided several interesting scholarly implications 
and possible avenues for future research. It is important that funding be made 
available to pursue a greater understanding of homelessness so that we can, as a 
society and a community, do a better job of providing everyone a safe and decent 
home. 
Due to the continued need for improved homelessness data, additional 
research should be conducted on data collected as part of the point-in-time (PIT) 
count and the housing inventory count (HIC). An example would be to research the 
methodologies of counts used in CoCs across the country and to evaluate the quality 
of those methodologies through field research on the days of counts and through 
analysis of their numbers to look for outliers or inexplicable increases or decreases in 
the results. Another example would be to evaluate HIC data to determine the 
feasibility of sorting housing units into more descriptive categories to indicate the 
presence of barriers or restrictions, such as transitional housing units that are only 
available to single men, or permanent housing units that are only available after 
108 
 
graduation from a program that requires Bible study and a long period of abstinence 
from addictive substances. 
The formula provided by the final model used in this study yielded results 
showing that increases in the Housing First index scores was only associated with 
decreases in estimated homelessness rates after the Housing First index score 
increased above 0.42 and only in CoCs where at least 27 percent of housing units 
were renter-occupied. This effectively meant that increases in the Housing First index 
was only associated with decreases in estimated homelessness rates for about 78.8 
percent of cases. This result merits research into why that is the case. Case study 
research could be conducted in one or several CoCs with housing units that are over 
73 percent owner-occupied to gain a more comprehensive understanding of the 
challenges associated with implementing a Housing First model in a place where a 
large majority of residents own their homes. 
Another important avenue of future research is to evaluate how the Housing 
First index could be expanded upon or improved to better determine the degree to 
which a CoC is implementing a Housing First approach. The index used in this study 
provides a starting point for a variable that has not been used in this way in the past. 
If PIT or HIC data were improved to include more specificity regarding whether beds 
or units provided were done so within the context of a Housing First program, that 
would help improve the quality of a future index. The index may also be reconfigured 
to measure how effectively a CoC is serving the needs of a subset of people 
experiencing homelessness. For example, if the researcher?s primary focus was on 
people experiencing chronic homelessness, then it may be appropriate to reconfigure 
109 
 
the index to put a greater focus on permanent supportive housing units than on 
permanent units provided through a rapid re-housing program. The index is also far 
too simple of a tool to use in case study research, because it is designed with the 
limitations of the national database in mind. In case study research, the index should 
be expanded to include considerations for how people are placed into units, what 
types of services are available, how the ownership structure is designed, how long 
people have to wait for a unit, and other potentially relevant factors that may vary 
across different CoCs. This case study research could help future researchers to 
understand when it is appropriate to use a standardized measure like the Housing First 
index and when another method may be more appropriate. 
The inclusionary zoning variable used in this study was statistically 
significant, but the presence of an inclusionary zoning policy was positively 
associated with homelessness rates instead of negatively associated with 
homelessness rates, as hypothesized. The best explanation for this phenomenon that I 
can construct is that homelessness rates drives the adoption of inclusionary zoning 
policies, as opposed to inclusionary zoning policies affecting homelessness rates. 
Future research should investigate this question more thoroughly. Grounded Solutions 
is currently in the process of updating their database. Perhaps an updated, more 
comprehensive database will result in different results. Or perhaps to properly 
research the relationship between inclusionary zoning policies and homelessness, the 
inclusionary zoning variable should be constructed in a way that allows the model to 
consider differences in the policies, instead of treating the presence of an inclusionary 
zoning policy as a dummy variable. 
110 
 
The Eviction Lab database that provided the eviction filing rate used in this 
study is another great step forward in the availability of data for research purposes. 
However, there was still a fair amount of missing data and it is likely that many of the 
evictions against the most vulnerable households that are most likely to experience 
homelessness are not evicted legally. Future research should focus on building upon 
the existing eviction database and augmenting these data on legal evictions with 
estimates of evictions that are done outside of the court system as well. This could be 
done via confidential interviews with landlords or perhaps by comparing a database 
for utility bills or tenant registries with court records, if the database differentiated 
between tenants that left voluntary versus those that did not. 
In prior research, Quigley et al. (2001) tested a theory that homelessness 
increases with the degree of income inequality in a community. This study did not 
research the relationship between income inequality and homelessness rates in as 
detailed of a way as Quigley et al. (2001) did, but it should be noted that this study 
did not find that income inequality as measured by the Gini index was related to 
homelessness rates in a statistically significant way. Future research should 
investigate why these findings were inconsistent. 
This study also merits follow-up research on homelessness rates in balance-of-
state CoCs and other CoCs that rely largely on emergency shelters, because this study 
found that balance-of-state CoCs tended to have lower homelessness rates and lower 
Housing First index scores. It would be interesting to know if balance-of-state CoCs 
experience lower homelessness rates simply because they encompass a larger area 
and PIT counts are more difficult to coordinate, thereby resulting in an undercount, or 
111 
 
if there is some other explanation. Also, the quadratic relationship between the 
Housing First index and homelessness rates revealed that homelessness rates increase 
slightly before decreasing substantially as the Housing First index increases. The 
implication of this result is that there are potentially CoCs where most of the beds 
available are in short-term emergency shelters and increasing the Housing First index 
would have an adverse impact on homelessness. Conversely, it could also mean that 
lower homelessness rates are driving the lack of sophisticated responses to 
homelessness, because there is not as great a need or perceived need. It could also 
imply that there is something more intentional happening in these CoCs that drives 
the lower homelessness rate outside the realm of the Housing First model. 
The panel analysis method used by Corinth (2017) and Hanratty (2017) 
provides benefits over simple multivariate regression analyses, because a panel 
analysis can better control for the passing of time over a multiyear period. The results 
of this study show the importance of studying interaction terms in panel analyses. 
Two interaction terms that were included in the final model of this study provided a 
deeper and more complex understanding of the relationship between the independent 
variables included in this analysis and homelessness rates.  
Future research should continue to build upon this study and the work of other 
scholars to provide a more complete understanding of the problem of homelessness so 
that communities may be better equipped with the knowledge necessary to end this 
critical problem once and for all. 
  
112 
 
Appendix 1: Data Collection and Model Assembly Process 
Data Collection 
Data were gathered from the U.S. Census Bureau 5-year estimates, Princeton 
University?s Eviction Lab database, and the National Oceanic and Atmospheric 
Administration?s Climate Divisional Database at the county level, from the Grounded 
Solutions Network at the point level, and from HUD at the CoC level. Data were then 
aggregated to CoC boundaries, which are the geographies used for the dependent 
variables in this study and are primarily counties or groups of counties. 
Estimates of homelessness rates and CoC data 
Homelessness data were gathered from the United States Department of 
Housing and Urban Development (HUD), downloaded from the HUD Exchange. The 
number of people experiencing homelessness, sheltered homelessness, unsheltered 
homelessness, families experiencing homelessness, chronic homelessness, and who 
were veterans experiencing homelessness were gathered from point-in-time (PIT) 
count data (HUD, 2018b). Data related to the services available in each CoC, which 
were used to gauge the extent that a CoC?s response to homelessness follows a 
Housing First model, were gathered from housing inventory count (HIC) data (HUD, 
2018b). 
Unfortunately for the purposes of this study, 61 CoC mergers occurred during 
the study period sometime between 2009 and 2017, thus changing the geography of 
the area in which homelessness point-in-time counts and housing inventory counts 
were conducted. These 61 CoC mergers affected 29 of the 384 CoCs included in this 
study, listed in Table 14 below. For the primary analysis, those 29 CoCs were 
113 
 
removed from the dataset. Data from the remaining 355 CoCs were analyzed for this 
study. 
Table 14: CoC Mergers, 2009-2017 
CoC CoC Merger CoC CoC Number Number Year Number Number 
Merger 
Pre-Merger Post Merger Pre-Merger Post Merger Year 
AR-502 AR-503 2010 VA-518 VA-513 2012 
AR-506 AR-503 2010 CA-605 CA-611 2013 
AR-509 AR-503 2010 CT-501 CT-505 2013 
AR-510 AR-503 2010 MA-512 MA-516 2013 
AR-511 AR-503 2010 NJ-505 NJ-503 2013 
CT-504 CT-505 2010 NJ-520 NJ-503 2013 
CT-507 CT-505 2010 NY-524 NY-508 2013 
MI-522 MI-500 2010 TX-501 TX-607 2013 
SC-504 SC-503 2010 TX-504 TX-607 2013 
TX-613 TX-607 2010 VA-509 VA-521 2013 
CA-610 CA-601 2011 VA-510 VA-521 2013 
CT-500 CT-505 2011 VA-517 VA-521 2013 
CT-509 CT-505 2011 CT-506 CT-503 2015 
CT-510 CT-505 2011 CT-508 CT-503 2015 
IL-505 IL-511 2011 CT-512 CT-505 2015 
MN-510 MN-503 2011 FL-516 FL-503 2015 
NE-503 NE-500 2011 MA-513 MA-516 2015 
NE-504 NE-500 2011 MA-520 MA-511 2015 
NE-505 NE-500 2011 NJ-518 NJ-503 2015 
NE-506 NE-500 2011 NY-509 NY-505 2015 
NJ-519 NJ-516 2011 NY-517 NY-508 2015 
OR-504 OR-505 2011 PA-507 PA-509 2015 
TX-704 TX-607 2011 PA-602 PA-601 2015 
VA-512 VA-501 2011 TX-703 TX-607 2015 
VA-519 VA-501 2011 CT-502 CT-505 2016 
AR-507 AR-503 2012 NY-502 NY-505 2016 
ME-501 ME-500 2012 IN-500 IN-502 2017 
NY-605 NY-603 2012 LA-504 LA-509 2017 
TX-610 TX-607 2012 MA-518 MA-516 2017 
TX-702 TX-607 2012 ME-502 ME-500 2017 
 
CoCs were only included in the database for this study if they existed in 2017, 
at the end of the study period. Many of the CoC mergers that occurred in the study 
114 
 
period resulted in several smaller CoCs joining a larger CoC, so the smaller CoCs that 
no longer existed in 2017 were not collected in the initial data collection phase. For 
this reason, the overall impact of CoC mergers on the study dataset were limited. 
Census estimates of county or county equivalent data 
The majority of county or county equivalent data were collected from the 
United States Census Bureau?s American FactFinder service (Census, 2013-2017). 
Data were downloaded using the Advanced Search functionality, using ?All Counties 
within the United States and Puerto Rico? as the geographic filter and American 
Community Survey (ACS) 5-year estimates as the dataset. County or county 
equivalent estimates were gathered for total population, educational attainment, 
income, rent, home value, occupied housing units, race, unemployment, poverty, 
tenure, family composition, rent burden, and economic inequality. 
  
115 
 
Table 15: Census data source tables 
Data Source Table 
Total population B01003 
Educational attainment S1501 
Median household income B19013 
Median gross rent B25064 
Median owner-occupied home value B25077 
Demographic characteristics for occupied housing 
units S2502 
Occupancy characteristics S2501 
Employment status S2301 
Poverty status in the past 12 months S1701 
Median gross rent as a percentage of household 
income B25071 
 
The ACS was chosen for this study because there are updates available every 
year, which is essential for a longitudinal panel analysis like the one used in this 
study. The 5-year estimates were chosen for this study because they use the largest 
sample size, are therefore the most reliable estimates, and data are available for 
counties of all populations. The ACS 3-year estimates are only available for areas 
with populations of 20,000 or more and the ACS 1-year estimates are only available 
for areas with populations of 65,000 or more (U.S. Census Bureau, 2019). These 
datasets would filter out many of the more rural counties. Since the ACS 5-year 
estimates are based on 60 months of collected data, they are the least current. This is 
less important than reliability in the case of the panel analysis used in this study, 
116 
 
because the goal of the study is to accurately determine the importance of 
implementation of the Housing First approach in association with homelessness rates, 
not to project present-day trends. The ACS was started in 2005, so ACS 5-year 
estimates are available as early as 2009, which makes 2009 the beginning of the study 
period. 
Climate data 
Average January temperature and precipitation data were collected from the 
National Oceanic and Atmospheric Administration?s (NOAA) Gridded Climate 
Divisional Database (2014b). Temperature is measured in degrees Fahrenheit and 
precipitation is measured in inches to 100ths. Data are available from 1895 through 
the latest month available and are updated monthly. Climate data have been used in 
previous studies of variables associated with homelessness, but the Gridded Climate 
Divisional Database (nClimDiv) was established relatively recently in 2014 and 
improves the quality of climate data by including additional station networks, quality 
assurance reviews, and temperature bias adjustments along with improvements to 
computational method (Vose et al., 2014a).10 
 
10 Methodology statement from NOAA: County values in nClimDiv were derived from area-weighted 
averages of grid-point estimates interpolated from station data. A nominal grid resolution of 5 km was 
used to ensure that all divisions had sufficient spatial sampling (only four small divisions had less than 
100 points) and because the impact of elevation on precipitation is minimal below 5 km. Station data 
were gridded via climatologically aided interpolation to minimize biases from topographic and 
network variability. The Global Historical Climatology Network (GHCN) Daily dataset is the source 
of station data for nClimDiv. GHCN-Daily contains several major observing networks in North 
America, five of which are used here. The primary network is the National Weather Service (NWS) 
Cooperative Observing (COOP) program, which consists of stations operated by volunteers as well as 
by agencies such as the Federal Aviation Administration. To improve coverage in western states and 
along international borders, nClimDiv also includes the National Interagency Fire Center (NIFC) 
Remote Automatic Weather Station (RAWS) network, the USDA Snow Telemetry (SNOTEL) 
network, the Environment Canada (EC) network (south of 52?N), and part of Mexico?s Servicio 
Meteorologico Nacional (SMN) network (north of 24?N). Note that nClimDiv does not incorporate 
precipitation data from RAWS because that networks tipping-bucket gauges are unheated, leading to 
117 
 
Climate data from the nClimDiv are stored as text files without file extensions 
and were opened in Excel by specifying custom delimiters used in the file. State 
codes used in the climate data differed from state codes used in data gathered from 
the Census Bureau, so it was necessary to convert state codes in the climate database 
prior to transferring climate data to a common database with Census data via FIPS 
codes. The two variables used from the nClimDiv database were average temperature 
and total precipitation in January of each year within the study period. January values 
were used to align with when the point-in-time (PIT) counts are conducted, which are 
the source of this study?s dependent variables. 
Eviction filing data 
This study will analyze the relationship between eviction filing rates and 
homelessness rates by utilizing a relatively new dataset (Desmond et al., 2018b) from 
The Eviction Lab at Princeton University.11 The threat of eviction looms over the 
same vulnerable low-income households that are the most susceptible to experiencing 
 
suspect cold-weather data. All GHCN-Daily stations are routinely processed through a suite of logical, 
serial, and spatial quality assurance reviews to identify erroneous observations. For nClimDiv, all such 
data were set to missing before computing monthly values, which in turn were subjected to additional 
serial and spatial checks to eliminate residual outliers. Stations having at least 10 years of valid 
monthly data since 1950 were used in nClimDiv. For temperature, bias adjustments were computed to 
account for historical changes in observation time, station location, temperature instrumentation, and 
siting conditions. Changes in observation time are only problematic for the COOP network whereas 
changes in station location and instrumentation occur in almost all surface networks. As in the U.S. 
Historical Climatology Network version 2.5, the method of Karl et al. (1986) was applied to remove 
the observation time bias from the COOP network, and the pairwise method of Menne and Williams 
(2009) was used to address changes in station location and instrumentation in all networks.  Because 
the pairwise method also largely accounts for local, unrepresentative trends that arise from changes in 
siting conditions, nClimDiv contains no separate adjustment in that regard. 
11 This research uses data from The Eviction Lab at Princeton University, a project directed by 
Matthew Desmond and designed by Ashley Gromis, Lavar Edmonds, James Hendrickson, Katie 
Krywokulski, Lillian Leung, and Adam Porton. The Eviction Lab is funded by the JPB, Gates, and 
Ford Foundations as well as the Chan Zuckerberg Initiative. More information is found at 
evictionlab.org. 
118 
 
homelessness, so this is an important relationship to study now that researchers have 
the means to do so. 
The Eviction Lab database was created by assembling court records for 
eviction cases gathered from court clerks via Freedom of Information Act (FOIA) 
requests or automated record collection, when possible. Eviction Lab researchers 
supplemented these data by purchasing additional eviction data from two companies, 
LexisNexis Risk Solutions and American Information Research Services (Desmond 
et al., 2018a). The Eviction Lab database includes both an eviction rate, representing 
the number of legal evictions12 per 100 renter homes, and an eviction filing rate, 
representing the number of eviction filings per 100 renter homes. This study uses the 
eviction filing rate due to a greater availability of data. 
Eviction filing data were also gathered at the county or county equivalent 
level, so it was simple to combine eviction data from Princeton University?s Eviction 
Lab with data from the U.S. Census Bureau. 
Inclusionary zoning data 
Inclusionary zoning, also known as inclusionary housing, policy data were 
collected from the Grounded Solutions Network Inclusionary Housing Database 
(2018). The purpose of inclusionary zoning policies is to incentivize or mandate the 
construction of affordable housing as a part of residential development projects. An 
effective inclusionary zoning policy should increase the supply of affordable housing 
 
12 It is important to note that the eviction filing rate used in this study is very likely an underestimate of 
the number of evictions that actually take place. Desmond estimates that informal evictions that 
happen without an eviction case being filed account for 48 percent of all forced moves and that formal 
evictions only account for 24 percent. Another 23 percent of forced moves are due to landlord 
foreclosure and the remaining five percent are a result of building condemnations (Desmond, 2016). 
119 
 
in the jurisdiction compared to the amount of affordable housing that would exist in 
the absence of such a policy. 
Data from the Grounded Solutions Network includes x-y coordinates for the 
center of jurisdictions that have adopted inclusionary zoning policies. These x-y 
coordinates were imported into ArcGIS Pro and a Spatial Join was executed to 
identify CoC boundaries that included at least one jurisdiction that has adopted an 
inclusionary zoning policy. Of the 355 CoCs analyzed as a part of this study, 121 of 
them included a jurisdiction that has adopted an inclusionary zoning policy. These 
data are not available in a longitudinal form, so the dummy values 1 (indicating the 
presence of a policy) and 0 (indicating the absence of a policy) based on 2018 data 
were used for every year of the study period. It is possible that some of these policies 
did not exist during the study period, or only existed for part of the study period, so 
the results regarding this variable are to be interpreted with caution. 
Additionally, this dummy variable does not allow for differentiation between 
a CoC with 10 adopted inclusionary zoning policies across its jurisdictions versus a 
CoC with only one adopted policy. Though this differentiation is possible and readily 
available via a simple GIS analysis, the quality of the data available from the 
Grounded Solutions Network did not allow for an accurate measurement of the 
degree to which these policies were being implemented or the scale of their impact. 
There were 1,169 points in the inclusionary zoning database at the time of 
collection. At that time, 238 of the points included a survey completed by the 
jurisdiction with information about the program (e.g. income restrictions, number of 
units created, affordability period, mandatory vs voluntary, incentives, minimum 
120 
 
numbers of units for the policy requirements to kick in, etc.) and 931 of the points 
were for a policy with no survey. For those 931 points, no information is available 
regarding the policy. Therefore, attempting to measure the degree to which 
inclusionary zoning requirements were implemented within a CoC based on the 
number of policies that exist within its boundaries may be very misleading. A CoC 
with 10 municipalities or counties that have all passed voluntary programs with very 
little result is not necessarily indicative of a stronger commitment to providing 
affordable housing via an inclusionary zoning policy than a single jurisdiction that 
adopted a mandatory policy with incentives that resulted in thousands of affordable 
units being developed.  
The data simply do not include enough information to make a reliable 
determination of scale. A dummy variable can still show me which CoCs include 
jurisdictions that have adopted policies, so it may still provide value in the panel 
analysis. 
 
Data Assembly and Constructing the Model 
The model was constructed and analyzed using three different programs. 
ESRI?s ArcGIS Pro was used for geographic aggregation of county-level data to CoC 
boundaries. Microsoft?s Excel was used for cleaning data and constructing pivot 
tables, which summarize spreadsheet data by identifying characteristics, such as CoC 
numbers. IBM?s SPSS was used for statistical analysis of the database in its final 
format. Data used in this study from any of these programs are available upon 
request. 
121 
 
Assembling county-level data 
A spreadsheet containing every county or county equivalent in rows was 
created using county-level population estimates from the U.S. Census Bureau?s 
American Community Survey (ACS) five-year estimates to ensure that every county 
or county equivalent intended for study was included in the spreadsheet. This 
included 3,209 counties or county equivalents from the 50 states, the District of 
Columbia, and Puerto Rico. County-level data downloaded from the U.S. Census 
Bureau and Princeton University?s Eviction Lab were first cleaned in Excel by 
removing all fields that were not to be used in the study and assigning variable names 
to the first row of each column. The sheet used to house aggregated data is in ?wide? 
format with estimates from multiple years for a single variable stored in multiple 
columns and a single row for each county or county equivalent. Each row in the sheet 
includes a five-digit county Federal Information Processing Standard (FIPS) code in 
the first column, the name of each county or county equivalent in the second column, 
and all subsequent columns are used to store variable values.  
Since not all county-level data are observed in every county every year, 
separate sheets stored data for individual variables on each year. Data were moved to 
the assembly sheet using IF statements that included the VLOOKUP function to use 
county FIPS codes to match data to the appropriate rows and the IFERROR function 
to leave cells blank in the case of missing data. After data were moved into the 
assembly sheet, each column was copied and pasted in place as values to avoid 
overburdening the program and the computer?s memory, and also to sever 
122 
 
connections to the original data source that could otherwise be a potential source of 
errors if original data sources were moved or deleted. 
Converting county-level data to CoC-level data 
Boundaries for counties or county equivalents were downloaded from the U.S. 
Census Bureau?s Topologically Integrated Geographic Encoding and Referencing 
(TIGER) database (2017). CoC boundaries were downloaded from the HUD 
Exchange website (2017a). Both the county or county equivalent boundaries and CoC 
boundaries were imported into ArcGIS Pro. 
Using the Spatial Join tool in ArcGIS Pro, county or county equivalent 
boundaries were identified as the target feature, CoC boundaries were identified as 
the join feature, and ?have their center in? was the selected match option.13 This 
operation assigned a CoC to each county or county equivalent with a center that falls 
within the boundaries of a CoC to ensure an accurate assignment. The Table to Excel 
tool was then used to convert the attribute table of the output feature of the Spatial 
Join into an Excel spreadsheet. In the assembly sheet, the VLOOKUP tool was used 
to populate fields with CoC numbers and names. 
Before summarizing county or county equivalent-level data at the CoC level, 
it was important to weigh counties by the number of occupied housing units so that 
data from lesser-populated counties did not disproportionately impact CoC values. To 
create weights for county-level weights, a pivot table was created on a new sheet 
 
13 It was not necessary to specify a join operation or a merge rule because no join features share the 
same spatial relationship with a single target feature in this operation. This is because the ?have their 
center in? match option was chosen. There are a few cases where more than one CoC overlaps with a 
county boundary, but no cases in which multiple CoC boundaries overlap the center of a county. This 
is because CoC boundaries do not overlap with one another. 
123 
 
using the assembly sheet as its data source. CoC numbers were loaded as the rows, 
columns were set to values, and the number of occupied housing units were loaded as 
values. Value fields were summarized by their sum, thus giving the total number of 
occupied housing units in each CoC. The results of this pivot table were then copied 
and pasted into a new sheet as values only. In the assembly sheet, a new column was 
created for weights in each of the study period years. These weights were calculated 
by dividing the number of occupied housing units in each county or county equivalent 
by the total number of occupied housing units in the associated CoC. A second 
column was then created adjacent to variables in the assembly sheet to store weighted 
values for those variables. The values in these new columns were calculated by 
multiplying variable values by the weight of each county or county equivalent. An 
IFERROR function was used to avoid errors in cases of missing values. 
A second pivot table was then created on a new sheet using the assembly sheet 
as its data source. CoC numbers were loaded as the rows, columns were set to values, 
and each of the weighted value columns from the assembly sheet were loaded into the 
pivot table as values. Value fields were summarized by their sum, thus giving a 
weighted average of each county or county equivalent-level variable at the CoC level.  
Adding additional CoC-level and point data 
After county and county-equivalent data were assembled at the CoC level, 
additional data available as points, such as inclusionary zoning policy data from the 
Grounded Solutions Network, and homelessness or service data from HUD available 
at the CoC level were added. 
124 
 
Inclusionary zoning policy data from the Grounded Solutions Network was 
available as a table with x-y coordinates. This table was added to a map file in 
ArcGIS Pro that also included a layer for CoC boundaries downloaded from HUD. 
Using ArcGIS Pro?s Make XY Event Layer geoprocessing tool, the x-y coordinates in 
the table were converted to spatial data and displayed on the map. This layer file was 
then exported as a saved feature class to allow for further manipulation and the 
original table and visualization layer were removed from the project. A Spatial Join 
was then conducted to identify all CoCs that contained at least one point for an 
adopted inclusionary zoning policy within their boundaries. 
The results of the Spatial Join were exported from ArcGIS Pro via the Table 
to Excel tool. The table values were then brought into the primary worksheet via 
Excel?s VLOOKUP tool using CoC numbers as the common identifier. 
CoC data gathered from HUD Exchange included the number of people 
experiencing homelessness, the number of people experiencing sheltered 
homelessness, the number of people experiencing unsheltered homelessness, the 
number of people in families experiencing homelessness, the number of people 
experiencing chronic homelessness, the number of veterans experiencing 
homelessness, the amount of HUD CoC funding distributed to each CoC, and housing 
inventory count data.14 Each of the variables measuring the number of people 
experiencing homelessness were divided by the total population within the CoC and 
 
14 Housing inventory count data show the total number of beds available for people experiencing 
homelessness in a CoC each year. Beds are separated by categories, including emergency shelter, safe 
haven, transitional housing, rapid re-housing, permanent supportive housing, and other permanent 
housing. 
125 
 
then multiplied by 1,000 to convert the estimates into homelessness rates measuring 
the number of people experiencing homelessness per 1,000 people living in the CoC. 
HUD CoC funding in the previous year was also divided by the total population in 
each CoC to calculate a rate measuring the amount of HUD CoC funding in the 
previous year per person. Since these data were gathered at the CoC level, they were 
all added to the primary worksheet in Excel via the VLOOKUP tool without any 
necessary conversions. 
Importing study data into SPSS 
To conduct descriptive statistics and the primary panel analyses used in this 
study, data were migrated from Excel to SPSS after data collection and conversion to 
CoC geographies was complete. To do this, data were imported into SPSS and 
restructured from a wide format database to a long format database using the 
Restructure tool in the SPSS Data menu. The ?long? format is necessary for 
descriptive statistics and panel analyses to be run in SPSS.   
Using the Restructure tool, each of the variables that previously used a 
separate column for each year of the study period were manually placed into groups 
and listed in chronological order for each variable. The CoC numbers were used for 
case group identification and an index variable was created for years and applied to 
variables in the order in which they were listed in the groups. To properly index 
variables based on the years of the study period, some placeholders were included in 
variable groups and then subsequently removed after the data were restructured. For 
example, the poverty rate, Gini index, and the percentage of renters in families with 
children are variables that were not included in the 2009 ACS five-year estimates. To 
126 
 
properly index these variables, 2010 data for those variables were included in the 
group list twice to fill the space for 2009 data. 
After the data were restructured, the index variable was renumbered from 1 to 
9 to 2009 to 2017 to reflect the years of the study and duplicate data were removed 
from variables that were missing either 2009 or 2017 data. Cases that previously 
included only one row for each of the CoCs now included nine rows for each CoC 
with a separate row for each year of the study period. Variable data that were 
previously spread across nine columns (or fewer, in cases where data were not 
available for each year of the study period) for each variable were now consolidated 
into a single column for each variable. Descriptive labels were then applied to each of 
the variables to facilitate effective visualization in graphs and tables. 
  
127 
 
Appendix 2: Model Selection Process 
The model selection process began by regressing homelessness rates by 
individual variables to determine whether each variable that could potentially be 
included in the panel analysis as a covariate had a statistically significant relationship 
with homelessness rates in a linear or polynomial regression. Next, main effects of 
each variable were tested using the linear mixed models procedure and variables with 
main effects that were not related to homelessness rates in a statistically significant 
way were removed from the model and tested for interaction effects one at a time in 
order of the weakness of the relationship as measured by the p value. Finally, main 
effect variables that remained in the model after all independent variables were 
significant were tested for interaction effects to determine if any additional interaction 
effects were significant and could further decrease the AIC or change the significance 
of main effect variables. The model with the lowest AIC and statistically significant 
independent variables was the final preferred model of this study. Because the 
Housing First index is the independent variable of primary interest in this study, it 
was not removed from the model. 
 
Individual variable regressions 
A full description of the results of each regression can be found in Appendix 
4: Individual Variable Regressions. A summary of each variable and their associated 
individual regression results, including significance, the R2 value, and the regression 
type that provided the best fit can be found in Table 16 below. Factor values of each 
variable were created to test linear, quadratic, cubic, and quartic regressions. 
128 
 
 
Table 16: Summary of individual variable regression results 
Variable Sig Direction R2  Best fit 
Total population (1000s) .000 U-shaped .021 Quadratic 
% Bachelor?s degree or higher .000 Positive .005 Linear* 
Median household income  .001 Inverse U-shaped .004 Quadratic 
Median gross rent .000 Positive .053 Linear 
Median home value (1000s) .000 Positive .118 Linear 
% Renters .000 Positive .145 Linear 
% Renters with children .000 U-shaped .019 Quadratic 
% Renters, white non-Hispanic .000 Negative .018 Linear 
% Renters, no college .000 Negative .024 Linear* 
Unemployment rate .000 Positive .032 Linear 
CoC funding per person .000 Positive .152 Linear 
Poverty rate .000 Positive .016 Linear 
Eviction filing rate .000 U-shaped .008 Quadratic 
% Rent-burdened .000 Positive .081 Linear 
Gini index .000 Positive .072 Linear 
Vacancy rate .000 Positive .017 Linear 
Mean temperature in Jan .000 Positive .090 Linear 
Total precipitation in Jan .000 Positive .026 Linear 
Housing First index .000 Inverse U-shaped .009 Quadratic 
* See Figure 42 and Figure 49 in Appendix 4: Individual Variable Regressions for both a cubic and 
linear regression plot for the % bachelor?s degree or higher variable and the % renters, no college 
variable, respectively. Both of these variables shared a very similar pattern in which the linear 
regression showed a significant relationship between homelessness rates and the variable in question, 
the significance was lost in the quadratic regression, and returned in the cubic regression, and the R2 
value came close to doubling from the linear to cubic regression. However, based on the scatterplot, 
this seemed to be a coincidence and this author could think of no substantive explanation for a cubic 
relationship between education and homelessness, so the linear regressions were used for both. 
 
The individual variable regressions revealed that five of the relationships were 
best expressed quadratically. These included the relationships between homelessness 
rates and total population, median household income, the percentage of renters with 
children, the eviction filing rate, and the Housing First index. To account for these 
quadratic relationships in a linear mixed model, factor values of these five covariates 
were included in the model (IBM, 2019). The quadratic relationship between 
129 
 
homelessness rates and the Housing First index appears as an inverse u-shape on the 
scatterplot, and values on the scatterplot appear to be close to randomly distributed. If 
not for the dense clustering of homelessness rate values, the relationship may not 
have appeared statistically significant at all, and the explanatory power of the 
quadratic regression is quite low at an R2 value of 0.009. Unless accounting for 
random within-subject effects reveals hidden significance of the Housing First index, 
this may be an early indication that the variable is not very significant. 
 
Figure 12: Quadratic relationship between homelessness rates and the Housing 
First index 
 
The variable regressions found every variable to be significantly associated 
with homelessness rates, at least individually. This means that every variable will be 
included in a preliminary version of the linear mixed model panel analysis for main 
effects testing. 
130 
 
All but one of the variables? best fit regressions were highly significant with a 
p score of .000 except for median household income, which came very close with a p 
score of .001 and was also highly significant. This is to be expected with many of the 
variables that are strongly supported by the literature, but for new or less-tested 
variables like the Housing First index, the eviction filing rate, the Gini index, CoC 
funding, and rent-burden/affordability, it is more surprising that all of the variables 
had a statistically significant relationship with homelessness rates. However, two of 
the variables, total population and the Housing First index, with best fit regressions 
that were quadratic in nature were not significant in a linear relationship with 
homelessness rates. 
 
Model testing 
A preliminary version of the linear mixed model panel analysis was created to 
test the significance of main effects of each variable on homelessness rates. The CoC 
name and category variables were included as the subjects and a variable for years 
that had been recoded to continuous values 0 through 8 was entered as the repeated 
variable with an autoregressive repeated covariance type. This means that the model 
will expect within-CoC values across observations repeated over the study period 
years to be more correlated with one another than between-CoC values, and any 
observed value is likely to be most correlated with the value within the same CoC 
from the previous observed year. 
Homelessness rates were included as the dependent variable, the inclusionary 
zoning variable and the CoC names were included as factors because they are 
131 
 
nominal variables that are categorical in nature, and the rest of the variables tested for 
their individual relationship with homelessness rates were included as covariates in 
this initial model. Factor values for the five covariates that have quadratic 
relationships with homelessness rates were also included as separate covariates to 
account for their polynomial relationship with the dependent variable. Additionally, 
the recoded continuous variable for year was included as a covariate. 
In the specifications for the model?s fixed effects, only main effects were 
included for now. All the factors and covariates were included in main effects testing 
in this initial model. In the specifications for the model?s random effects, only the 
intercept and a single interaction term for the interaction between the recoded 
continuous variable for year and the CoC name variable were included. The CoC 
number and category variables were included as a subject grouping to test the random 
effects. The autoregressive covariance type was selected again for the random effect, 
because I expect a random effect on a residual to be most closely related to the 
random effect on a residual within the same CoC in the previous year. By including 
these random effects in the model, I can separate out the portion of the error in each 
observation that is due to the year and CoC in which an observation was made. 
However, the fixed effects results are the subject of this study. 
The first round of main effects testing resulted in an AIC score of 5123.713. 
The goal of refining linear mixed models is to bring this number down while 
maintaining statistically significant variables, which indicates that the model has 
increased its explanatory power or eliminated redundant variables. 
132 
 
Table 17: Information criteria for first round of main effects testing 
Information Criteriaa 
-2 Restricted Log Likelihood 5117.713 
Akaike's Information Criterion (AIC) 5123.713 
Hurvich and Tsai's Criterion (AICC) 5123.725 
Bozdogan's Criterion (CAIC) 5143.484 
Schwarz's Bayesian Criterion (BIC) 5140.484 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
The relationship between homelessness rates and thirteen of the variables 
became insignificant in the first round of main effects testing. This is likely because 
the regression analyses did not control for correlations of repeated measures in the 
same CoC or over time, so those correlations are inappropriately associated with the 
independent variable being tested. The main effects testing also simply includes more 
variables, so collinearity issues and controlling for changes in other variables reduces 
the significance of individual variables. The Housing First index was one of the 
variables that became insignificant during the first round of main effects testing. 
Out of the thirteen variables with insignificant relationships with 
homelessness rates, the Gini index had the highest p value at .830 (see Table 18 
below) and therefore had the weakest direct association with homelessness rates and 
was removed from main effects testing to be tested for interaction effects.  
133 
 
Table 18: Type III tests of fixed effects in the first round of main effects testing 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 6859.020 .103 .748 
coccat 2 311.003 4.061 .018 
inczon 1 320.766 1.049 .307 
pop 0 . . . 
bach 1 810.553 3.016 .083 
medinc 1 803.556 .468 .494 
medren 1 922.549 18.395 .000 
medval 1 828.139 10.432 .001 
renfam 1 666.197 .540 .463 
renocc 1 545.749 .164 .686 
renwhi 1 585.687 3.854 .050 
renedu 1 746.652 5.821 .016 
unemp 1 1647.906 1.409 .235 
pov 1 1320.456 .779 .378 
evic 1 1422.967 .114 .736 
burd 1 1729.809 6.754 .009 
gini 1 1002.101 .046 .830 
vac 1 452.685 2.584 .109 
temp 1 1837.697 2.424 .120 
precip 1 1696.294 1.837 .176 
hf 1 1977.624 .299 .584 
fund 1 753.774 24.652 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 633.257 .508 .476 
evic2 1 1880.448 .146 .702 
hf2 1 1966.959 1.150 .284 
yearcoded 1 1339.173 23.222 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
The model was rerun every time a main effect variable was removed, or 
interaction variables were added to test for changes in variable significance and the 
AIC score. While there are too many iterations of the model to include in this study, 
the full set of model testing results are available in SPSS upon request. A summary of 
134 
 
noteworthy models from the first phase of testing are included in Table 19 below, and 
the results from those models are included in Appendix 5: Model Testing Results. 
Table 19 covers the first phase of the model refinement process, in which 
variable significance was prioritized over decreases in the AIC. The first phase ended 
once all remaining variables in the model were significant.  
For some of the models, Table 19 indicates that a part of an interaction effect 
is significant. What this means is that one or both variables used in the interaction 
term are polynomials, and the interaction between some, but not all, of the directions 
of a quadratic relationship with homelessness rates is significant. A practical example 
of this can be found in Model 62, in which an interaction between the percentage of 
renter-occupied households (renocc) and the Housing First index (hf) was included in 
the model, and the interaction between renocc and hf was insignificant, but the 
interaction between renocc and hf2 was significant. This means that for squared 
values of the Housing First index, changes in the percentage of renter-occupied 
households changes the Housing First index?s relationship with homelessness rates in 
a statistically significant way, and changes in squared values of the Housing First 
index affect the relationship between the percentage of renter-occupied households 
and homelessness rates in a statistically significant way as well. Interestingly, when 
accounting for the interaction between hf and renocc, the main effect of the hf 
coefficients switched their signs so that the coefficient for hf became negative and the 
coefficient for hf2 became positive.  
135 
 
Table 19: Summary of results from the first phase of model testing 
 Model Change AIC 
Model 1 First model testing only main effects of all variables 5123.713 
Model 2 Removed gini (increased AIC) 5126.298 
Model 3 Added inczon * gini interaction (insignificant, lowered AIC) 5116.382 
Model 4 Added coccat * gini interaction (insignificant, lowered AIC) 5107.325 
Model 21 Added gini * hf interaction (insignificant, lowered p score of hf, 5108.665 
lowered AIC) 
Model 24 Removed evic (lowered AIC) 5106.639 
Model 45 Removed renocc (substantially increased AIC) 6549.698 
Model 50 Added renocc * medinc interaction (significant, increased AIC) 6619.417 
Model 62 Added renocc * hf interaction (part significant, lowered AIC, flipped 6544.710 
hf quadratic coefficients) 
Model 65 Removed renfam (lowered AIC) 6526.660 
Model 84 Removed precip (lowered AIC) 6518.533 
Model 102 Removed unemp (increased AIC) 6546.904 
Model 119 Removed medinc (lowered AIC, lowered p score of hf) 6508.198 
Model 122 Added pop * medinc interaction (part significant, increased AIC) 6795.405 
Model 124 Added medinc * medren interaction (significant, increased AIC) 6556.442 
Model 125 Added medinc * medval interaction (significant, increased AIC) 6553.851 
Model 126 Added medinc * renwhi interaction (significant, increased AIC) 6575.396 
Model 127 Added medinc * renedu interaction (part significant, increased AIC) 6575.481 
Model 128 Added medinc * pov interaction (significant, increased AIC) 6560.349 
Model 129 Added medinc * burd interaction (part significant, increased AIC) 6559.636 
Model 130 Added medinc * vac interaction (significant, increased AIC) 6563.869 
Model 132 Added medinc * hf interaction (part significant, increased AIC) 6607.937 
Model 133 Added medinc * fund interaction (significant, increased AIC) 6566.545 
Model 134 Added medinc * yearcoded interaction (significant, increased AIC) 6570.855 
Model 135 Removed burd (lowered AIC) 6502.810 
Model 141 Added medval * burd interaction (significant, increased AIC) 6528.730 
Model 150 Removed medren (lowered AIC) 6490.568 
Model 155 Added medren * medval interaction (significant, increased AIC) 6521.769 
Model 156 Added medren * renwhi interaction (significant, increased AIC) 6505.315 
Model 158 Added medren * pov interaction (significant, increased AIC) 6499.428 
Model 159 Added medren * vac interaction (significant, increased AIC) 6500.511 
Model 164 Removed renwhi (lowered AIC) 6484.055 
Model 167 Added pop * renwhi interaction (part significant, increased AIC) 6579.869 
Model 168 Added bach * renwhi interaction (significant, increased AIC) 6495.177 
Model 169 Added medval * renwhi interaction (significant, increased AIC) 6501.534 
Model 173 Added renwhi * temp interaction (significant, increased AIC) 6495.668 
Model 174 Added renwhi * hf interaction (significant, increased AIC) 6491.493 
 
136 
 
With the removal of the percentage of renter-occupied households (renocc) 
variable in Model 45, the AIC jumped considerably from 5106.639 to 6549.698, 
indicating a substantial weakening of the model with the removal of an insignificant 
variable. This likely means that the significance of the percentage of renter-occupied 
households was masked by collinearity with other variables or that it was serving a 
valuable function as a control variable, and renocc, along with any other variables 
removed that resulted in an increase to the AIC, was reintroduced to the model in the 
second phase of model testing. Aside from the major leap upwards in Model 45 and 
small increases in the AIC with the removal of the Gini index and the unemployment 
rate in Models 2 and 102, the AIC generally trended downwards with the elimination 
of statistically insignificant variables, indicating improvement of the model. While all 
models with interaction effects that either lowered the AIC or were significant were 
noted in Table 19, interaction effects were not kept in the model unless they both 
lowered the AIC and were at least partially significant. 
In the second phase of the model refinement process, decreases in the AIC 
were prioritized over the significance of individual variables. In this phase, all main 
effect variables that increased the AIC when they were removed during the first phase 
were reintroduced to the model in the order that they were removed and were kept if 
their reintroduction lowered the AIC. Then, any insignificant interactions that 
lowered the AIC in the first phase will be reintroduced in the order they were tested in 
the first phase. Finally, all main effect variables remaining in the model were tested 
for interaction effects. A summary of the results of the second phase of model testing 
can be found in Table 20 below. 
137 
 
Table 20: Summary of results from the second phase of model testing 
 Model Change AIC 
Model 177 Reintroduced gini (lowered AIC, included) 6478.585 
Model 178 Reintroduced renocc (increased AIC, not included) 6480.641 
Model 179 Reintroduced unemp (increased AIC, not included) 6484.303 
Model 180 Reintroduced inczon * gini interaction (lowered AIC, included) 6471.642 
Model 181 Reintroduced coccat * gini interaction (lowered AIC, included) 6456.868 
Model 182 Reintroduced gini * hf interaction (lowered AIC, included) 6441.414 
Model 201 Added medval * pov interaction (lowered AIC, included) 6438.321 
Model 215 Added pov * hf interaction (lowered AIC, included) 6431.524 
Model 221 Added vac * yearcoded interaction (lowered AIC, included) 6429.769 
Model 224 Added temp * yearcoded interaction (lowered AIC, included) 6425.655 
Model 225 Added fund * hf interaction (lowered AIC, included) 6424.719 
 
A total of 227 models were run across both phases of testing. The second 
phase of testing prioritized lowering the AIC score over variable significance, and the 
AIC was only lowered from 6478.585 to 6424.719, which is still higher than the 
baseline model AIC of 5123.713 that included only main effects for all variables. In 
both the baseline model and Model 225, which provided the lowest AIC score at the 
conclusion of second phase testing, most variables used in the model were 
statistically insignificant. Since the primary purpose of this study is to understand 
relationships between homelessness and other variables as opposed to making 
predictions about homelessness, the reduction in the AIC is not substantial enough to 
justify sacrificing statistical significance of the variables.  
Therefore, the results of the second phase of model testing are discarded in 
favor of Model 159. I chose Model 159 instead of Model 164, because the difference 
in the AIC is only 21.926, and by adding an interaction term for the median rent and 
vacancy rate, the percentage of renters who are white is shown to have a statistically 
significant relationship with homelessness rates, which is supported by the literature 
138 
 
and I hypothesize that this relationship is meaningful. Inclusionary zoning and the 
vacancy rate are no longer significant at the 0.05 p-score level in Model 159, but they 
are still significant at the 0.10 p-score level, so I do not have to sacrifice much to 
continue to consider the relationship between race and homelessness in the model. 
  
139 
 
Appendix 3: Additional Descriptive Statistics 
 
Table 21: Descriptive summary of variables 
Std. 
 N Minimum Maximum Mean Deviation Skewness Kurtosis 
Percentage of 3141 11.37 74.10 28.9309 9.91841 1.011 1.223 
Population with a 
Bachelor's Degree or 
Higher 
Median Household 3141 17208.45 129588.00 54989.43 14631.66 1.382 2.617 
Income 40 806 
Median Gross Rent of 3141 390.50 1973.00 887.0583 243.9512 1.118 1.309 
Renter-Occupied 6 
Housing Units 
Median Home Value 3141 67165 927400 213148.0 125688.8 2.029 5.112 
of Owner-Occupied 8 78 
Housing Units 
Number of Occupied 2792 11644 3281845 272360.4 369437.5 4.627 27.287 
Housing Units 6 18 
Percentage of Renters 2795 13.90 58.04 33.8157 6.45946 .297 1.021 
in a Family with 
Children 
Percentage of Renter- 3141 12.56 69.22 33.5296 8.37018 .913 1.738 
Occupied Housing 
Units 
Percentage of Renters 3141 .73 98.40 63.3276 19.02170 -.461 -.227 
Identifying as White, 
Non-Hispanic 
Percentage of Renters 3141 13.20 68.64 44.7024 9.49701 -.251 -.168 
Without any College 
Education 
Unemployment Rate 3141 2.68 19.68 8.3145 2.50746 .937 1.426 
Poverty Rate 2795 3.20 48.38 14.8705 5.16948 1.256 6.101 
Eviction Filing Rate 2347 .00 39.52 6.0189 6.06029 1.694 3.035 
Percentage of Rent- 3141 22.69 39.50 30.7397 2.53258 .491 .212 
Burdened Households 
Gini Index 2795 .36 .55 .4516 .03019 .285 .560 
Vacancy Rate 3141 3.67 46.47 12.4655 6.57217 1.778 4.426 
140 
 
Mean Temperature in 3039 -12.80 72.50 34.9229 13.06098 .236 -.208 
January in Fahrenheit 
Total Precipitation in 3031 .01 24.56 2.9166 2.50880 2.604 10.806 
January in Inches 
CoC Category 3141 1 3 1.96 .456 -.148 1.761 
Adoption of an 3141 0 1 .34 .474 .668 -1.554 
Inclusionary Zoning 
Policy 
Housing First Index 3136 .00 .97 .5375 .15871 -.388 .035 
HUD CoC Funding in 3091 .01 46.90 5.1527 5.15511 2.902 11.663 
the Previous Year per 
Person 
People per 1,000 3141 .02 16.78 1.9731 2.00319 3.219 13.614 
Experiencing 
Homelessness 
People per 1,000 3141 .01 12.19 1.2273 1.08133 3.979 24.899 
Experiencing 
Sheltered 
Homelessness 
People per 1,000 3141 .00 16.37 .7458 1.54983 4.746 29.416 
Experiencing 
Unsheltered 
Homelessness 
People per 1,000 in 3141 .00 13.55 .6886 .86739 5.720 51.790 
Families Experiencing 
Homelessness 
People per 1,000 3141 .00 6.30 .3691 .56605 4.358 25.824 
Experiencing Chronic 
Homelessness 
Veterans Experiencing 2448 .00 1.85 .1743 .21014 2.837 10.953 
Homelessness per 
1,000 People 
People per 1,000 3141 .02 6.29 1.8405 1.46589 1.693 2.362 
Experiencing 
Homelessness 
(Coerced) 
  
141 
 
 
Figure 13: Distribution histogram for variable: bach 
 
 
 
Figure 14: Distribution histogram for variable: medinc 
  
142 
 
 
Figure 15: Distribution histogram for variable: medren 
 
 
 
Figure 16: Distribution histogram for variable: medval 
  
143 
 
 
Figure 17: Distribution histogram for variable: aff 
 
 
 
Figure 18: Distribution histogram for variable: renocc 
  
144 
 
 
Figure 19: Distribution histogram for variable: renfam 
 
 
 
Figure 20: Distribution histogram for variable: renwhi 
  
145 
 
 
Figure 21: Distribution histogram for variable: renedu 
 
 
 
Figure 22: Distribution histogram for variable: unemp 
  
146 
 
 
Figure 23: Distribution histogram for variable: pov 
 
 
 
Figure 24: Distribution histogram for variable: evic 
  
147 
 
 
Figure 25: Distribution histogram for variable: burd 
 
 
 
Figure 26: Distribution histogram for variable: gini 
  
148 
 
 
Figure 27: Distribution histogram for variable: vac 
 
 
 
Figure 28: Distribution histogram for variable: temp 
  
149 
 
 
Figure 29: Distribution histogram for variable: precip 
 
 
 
Figure 30: Distribution histogram for variable: coccat 
  
150 
 
 
Figure 31: Distribution histogram for variable: inczon 
 
 
 
Figure 32: Distribution histogram for variable: hf 
  
151 
 
 
Figure 33: Distribution histogram for variable: fund 
 
 
 
Figure 34: Distribution histogram for variable: hl 
  
152 
 
 
Figure 35: Distribution histogram for variable: hls 
 
 
 
Figure 36: Distribution histogram for variable: hlu 
  
153 
 
 
Figure 37: Distribution histogram for variable: hlf 
 
 
 
Figure 38: Distribution histogram for variable: hlc 
  
154 
 
 
Figure 39: Distribution histogram for variable: hlv 
 
 
 
Figure 40: Distribution histogram for variable: hlco 
  
155 
 
Table 22: Outliers 
 CoC Number Year People per 1,000 Experiencing Homelessness 
1 FL-517 2009 16.78 
2 FL-517 2010 16.72 
3 CA-509 2011 16.64 
4 CA-509 2012 16.45 
5 CA-509 2014 16.03 
6 CA-509 2013 15.91 
7 FL-517 2013 15.31 
8 CA-614 2009 14.61 
9 CA-614 2010 14.42 
10 FL-604 2010 14.23 
11 CA-509 2016 14.21 
12 CA-509 2017 14.15 
13 FL-604 2009 14.05 
14 CA-509 2009 13.97 
15 CA-509 2010 13.74 
16 CA-508 2013 13.35 
17 CA-508 2014 13.21 
18 DC-500 2016 12.67 
19 FL-604 2012 12.55 
20 FL-604 2011 12.37 
21 DC-500 2014 12.23 
22 LA-503 2010 12.01 
23 FL-505 2011 11.70 
24 DC-500 2012 11.48 
25 LA-503 2009 11.35 
26 DC-500 2015 11.27 
27 DC-500 2010 11.19 
28 DC-500 2017 11.11 
29 DC-500 2013 11.08 
30 CA-523 2013 11.04 
31 DC-500 2011 11.02 
32 CA-509 2015 10.82 
33 CA-508 2011 10.68 
34 DC-500 2009 10.58 
35 CA-522 2009 10.50 
36 FL-505 2009 10.48 
37 CA-508 2012 10.43 
156 
 
38 CA-603 2009 10.25 
39 FL-505 2012 10.21 
40 CA-522 2010 10.18 
41 FL-519 2009 9.97 
42 FL-519 2010 9.92 
43 CA-603 2010 9.91 
44 MO-602 2012 9.86 
45 FL-519 2011 9.75 
46 FL-505 2010 9.61 
47 FL-518 2013 9.55 
48 CA-504 2011 9.49 
49 FL-519 2012 9.35 
50 CA-504 2012 9.27 
51 FL-604 2014 9.01 
52 CA-508 2009 9.01 
53 CA-523 2014 9.01 
54 FL-518 2011 8.94 
55 NY-600 2015 8.94 
56 NY-600 2017 8.94 
57 CA-524 2013 8.87 
58 LA-503 2011 8.87 
59 FL-518 2012 8.86 
60 CA-508 2010 8.82 
61 FL-604 2013 8.79 
62 CA-504 2013 8.78 
63 CA-522 2015 8.74 
64 NY-600 2016 8.69 
65 CA-504 2014 8.67 
66 CA-614 2013 8.66 
67 FL-518 2009 8.64 
68 CA-614 2014 8.63 
69 CA-501 2013 8.57 
70 MA-500 2015 8.55 
71 FL-518 2010 8.51 
72 FL-518 2016 8.48 
73 CA-522 2016 8.39 
74 FL-518 2015 8.28 
75 CA-508 2017 8.23 
76 CA-501 2016 8.23 
77 FL-604 2017 8.22 
157 
 
78 NY-607 2012 8.15 
79 MA-500 2016 8.13 
80 NY-600 2014 8.12 
81 FL-604 2015 8.10 
82 CA-501 2015 8.06 
83 MA-500 2014 8.00 
84 MA-500 2013 7.99 
85 FL-518 2014 7.96 
86 CA-614 2011 7.95 
87 CA-501 2017 7.94 
88 GA-500 2010 7.91 
89 CA-614 2012 7.88 
90 MA-500 2017 7.86 
91 MD-508 2009 7.80 
92 NY-600 2013 7.75 
93 MA-500 2012 7.74 
94 CA-501 2014 7.73 
95 MA-500 2011 7.68 
96 MD-508 2010 7.67 
97 GA-500 2011 7.50 
98 FL-512 2011 7.48 
99 FL-604 2016 7.45 
100 OR-500 2010 7.43 
101 CA-501 2010 7.38 
102 GA-500 2012 7.33 
103 CA-501 2009 7.30 
104 CA-501 2012 7.30 
105 MA-500 2010 7.29 
106 CA-508 2015 7.29 
107 FL-512 2013 7.29 
108 FL-512 2012 7.23 
109 CA-508 2016 7.23 
110 NY-607 2013 7.21 
111 GA-500 2009 7.11 
112 CA-501 2011 7.10 
113 FL-519 2014 7.10 
114 FL-512 2009 7.07 
115 FL-519 2013 7.06 
116 CA-504 2010 7.06 
117 NC-516 2009 7.01 
158 
 
118 CA-504 2009 6.99 
119 HI-500 2016 6.99 
120 MA-500 2009 6.98 
121 FL-512 2010 6.93 
122 NY-600 2012 6.91 
123 CA-522 2014 6.91 
124 FL-512 2014 6.89 
125 CA-506 2017 6.84 
126 FL-505 2013 6.84 
127 NY-607 2011 6.77 
128 MD-501 2011 6.60 
129 NY-600 2010 6.58 
130 OR-500 2009 6.49 
131 CA-522 2011 6.45 
132 HI-500 2015 6.44 
133 CA-522 2013 6.42 
134 MA-504 2015 6.41 
135 CA-613 2017 6.41 
136 FL-501 2009 6.40 
137 FL-505 2014 6.39 
138 LA-503 2012 6.33 
139 CA-522 2012 6.30 
  
159 
 
Table 23: Distribution descriptives for homelessness using coerced values 
 Above and Below Median Homelessness Rates Statistic Std. Error 
People per 1,000 Above Mean 2.8255 .03794 
Experiencing 95% Confidence Lower Bound 2.7510  
Homelessness Interval for Mean Upper Bound 2.8999  
(Coerced) 5% Trimmed Mean 2.7109  
Median 2.2144  
Variance 2.261  
Std. Deviation 1.50366  
Minimum 1.40  
Maximum 6.29  
Range 4.89  
Interquartile Range 1.73  
Skewness 1.256 .062 
Kurtosis .335 .123 
Below Mean .8550 .00777 
95% Confidence Lower Bound .8397  
Interval for Mean Upper Bound .8702  
5% Trimmed Mean .8600  
Median .8553  
Variance .095  
Std. Deviation .30769  
Minimum .02  
Maximum 1.40  
Range 1.38  
Interquartile Range .50  
Skewness -.137 .062 
Kurtosis -.872 .123 
  
160 
 
Appendix 4: Individual Variable Regressions 
 
Table 24: Model fit results regressing homelessness rates by total population 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of the R Square F Sig. F 
Model R Square Square Estimate Change Change df1 df2 Change 
1 .003a .000 .000 2.00350 .000 .026 1 3139 .873 
2 .144b .021 .020 1.98286 .021 66.704 1 3138 .000 
3 .144c .021 .020 1.98317 .000 .016 1 3137 .899 
4 .146d .021 .020 1.98295 .001 1.703 1 3136 .192 
a. Predictors: (Constant), Total Population 
b. Predictors: (Constant), Total Population, pop2 
c. Predictors: (Constant), Total Population, pop2, pop3 
d. Predictors: (Constant), Total Population, pop2, pop3, pop4 
 
 
Figure 41: Homelessness rates by total population quadratic regression 
  
161 
 
Table 25: ANOVA table regressing homelessness rates by total population 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression .103 1 .103 .026 .873b 
Residual 12600.021 3139 4.014   
Total 12600.124 3140    
2 Regression 262.366 2 131.183 33.365 .000c 
Residual 12337.758 3138 3.932   
Total 12600.124 3140    
3 Regression 262.429 3 87.476 22.242 .000d 
Residual 12337.695 3137 3.933   
Total 12600.124 3140    
4 Regression 269.125 4 67.281 17.111 .000e 
Residual 12330.999 3136 3.932   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Total Population 
c. Predictors: (Constant), Total Population, pop2 
d. Predictors: (Constant), Total Population, pop2, pop3 
e. Predictors: (Constant), Total Population, pop2, pop3, pop4 
 
  
162 
 
Table 26: Coefficient table regressing homelessness rates by total population 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.969 .044  44.927 .000 
Total Population 5.516E-9 .000 .003 .160 .873 
2 (Constant) 2.253 .056  40.534 .000 
Total Population -5.758E-7 .000 -.298 -7.295 .000 
pop2 8.896E-14 .000 .334 8.167 .000 
3 (Constant) 2.258 .068  32.977 .000 
Total Population -5.904E-7 .000 -.306 -4.216 .000 
pop2 9.514E-14 .000 .357 1.898 .058 
pop3 -5.228E-22 .000 -.017 -.126 .899 
4 (Constant) 2.194 .084  26.069 .000 
Total Population -3.420E-7 .000 -.177 -1.447 .148 
pop2 -9.104E-14 .000 -.342 -.602 .547 
pop3 3.778E-20 .000 1.218 1.275 .203 
pop4 -2.246E-27 .000 -.665 -1.305 .192 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
163 
 
Table 27: Model fit results regressing homelessness rates by percentage of 
population with a bachelor?s degree or higher 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .070a .005 .005 1.99858 .005 15.508 1 3139 .000 
2 .073b .005 .005 1.99854 .000 1.112 1 3138 .292 
3 .099c .010 .009 1.99431 .005 14.330 1 3137 .000 
a. Predictors: (Constant), Percentage of Population with a Bachelor's Degree or Higher 
b. Predictors: (Constant), Percentage of Population with a Bachelor's Degree or Higher, bach2 
c. Predictors: (Constant), Percentage of Population with a Bachelor's Degree or Higher, bach2, bach3 
 
 
Figure 42: Homelessness rates by percentage of population with a bachelor?s 
degree or higher cubic and linear regressions 
  
164 
 
Table 28: ANOVA table regressing homelessness rates by percentage of population 
with a bachelor?s degree or higher 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 61.944 1 61.944 15.508 .000b 
Residual 12538.180 3139 3.994   
Total 12600.124 3140    
2 Regression 66.387 2 33.194 8.311 .000c 
Residual 12533.737 3138 3.994   
Total 12600.124 3140    
3 Regression 123.380 3 41.127 10.340 .000d 
Residual 12476.744 3137 3.977   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Percentage of Population with a Bachelor's Degree or Higher 
c. Predictors: (Constant), Percentage of Population with a Bachelor's Degree or Higher, bach2 
d. Predictors: (Constant), Percentage of Population with a Bachelor's Degree or Higher, bach2, bach3 
 
  
165 
 
Table 29: Coefficient table regressing homelessness rates by percentage of 
population with a bachelor?s degree or higher 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.563 .110  14.216 .000 
Percentage of Population .014 .004 .070 3.938 .000 
with a Bachelor's Degree or 
Higher 
2 (Constant) 1.298 .275  4.729 .000 
Percentage of Population .032 .017 .157 1.864 .062 
with a Bachelor's Degree or 
Higher 
bach2 .000 .000 -.089 -1.055 .292 
3 (Constant) 3.463 .634  5.461 .000 
Percentage of Population -.180 .058 -.892 -3.081 .002 
with a Bachelor's Degree or 
Higher 
bach2 .006 .002 2.065 3.591 .000 
bach3 -5.621E-5 .000 -1.153 -3.785 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
166 
 
Table 30: Model fit results regressing homelessness rates by median household 
income 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .043a .002 .002 2.00167 .002 5.764 1 3139 .016 
2 .065b .004 .004 1.99965 .002 7.348 1 3138 .007 
3 .066c .004 .003 1.99976 .000 .663 1 3137 .416 
a. Predictors: (Constant), Median Household Income 
b. Predictors: (Constant), Median Household Income, medinc2 
c. Predictors: (Constant), Median Household Income, medinc2, medinc3 
 
 
Figure 43: Homelessness rates by median household income quadratic regression 
  
167 
 
Table 31: ANOVA table regressing homelessness rates by median household 
income 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 23.094 1 23.094 5.764 .016b 
Residual 12577.030 3139 4.007   
Total 12600.124 3140    
2 Regression 52.477 2 26.239 6.562 .001c 
Residual 12547.647 3138 3.999   
Total 12600.124 3140    
3 Regression 55.127 3 18.376 4.595 .003d 
Residual 12544.997 3137 3.999   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Median Household Income 
c. Predictors: (Constant), Median Household Income, medinc2 
d. Predictors: (Constant), Median Household Income, medinc2, medinc3 
 
Table 32: Coefficient table regressing homelessness rates by median household 
income 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 2.295 .139  16.524 .000 
Median Household Income -5.861E-6 .000 -.043 -2.401 .016 
2 (Constant) 1.218 .421  2.892 .004 
Median Household Income 2.992E-5 .000 .219 2.229 .026 
medinc2 -2.749E-10 .000 -.266 -2.711 .007 
3 (Constant) 1.965 1.010  1.946 .052 
Median Household Income -7.170E-6 .000 -.052 -.151 .880 
medinc2 3.019E-10 .000 .292 .422 .673 
medinc3 -2.793E-15 .000 -.296 -.814 .416 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
168 
 
Table 33: Model fit results regressing homelessness rates by median gross rent 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .230a .053 .053 1.94965 .053 175.848 1 3139 .000 
2 .249b .062 .062 1.94057 .009 30.423 1 3138 .000 
3 .251c .063 .062 1.94020 .001 2.202 1 3137 .138 
a. Predictors: (Constant), Median Gross Rent of Renter-Occupied Housing Units 
b. Predictors: (Constant), Median Gross Rent of Renter-Occupied Housing Units, medren2 
c. Predictors: (Constant), Median Gross Rent of Renter-Occupied Housing Units, medren2, medren3 
 
 
Figure 44: Homelessness rates by median gross rent linear regression 
  
169 
 
Table 34: ANOVA table regressing homelessness rates by median gross rent 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 668.420 1 668.420 175.848 .000b 
Residual 11931.704 3139 3.801   
Total 12600.124 3140    
2 Regression 782.986 2 391.493 103.960 .000c 
Residual 11817.138 3138 3.766   
Total 12600.124 3140    
3 Regression 791.273 3 263.758 70.067 .000d 
Residual 11808.851 3137 3.764   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Median Gross Rent of Renter-Occupied Housing Units 
c. Predictors: (Constant), Median Gross Rent of Renter-Occupied Housing Units, medren2 
d. Predictors: (Constant), Median Gross Rent of Renter-Occupied Housing Units, medren2, medren3 
 
  
170 
 
Table 35: Coefficient table regressing homelessness rates by median gross rent 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) .295 .131  2.252 .024 
Median Gross Rent of .002 .000 .230 13.261 .000 
Renter-Occupied Housing 
Units 
2 (Constant) -1.874 .414  -4.522 .000 
Median Gross Rent of .006 .001 .788 7.681 .000 
Renter-Occupied Housing 
Units 
medren2 -2.238E-6 .000 -.566 -5.516 .000 
3 (Constant) -.181 1.214  -.150 .881 
Median Gross Rent of .001 .004 .146 .328 .743 
Renter-Occupied Housing 
Units 
medren2 2.922E-6 .000 .739 .834 .404 
medren3 -1.581E-9 .000 -.681 -1.484 .138 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
171 
 
Table 36: Model fit results regressing homelessness rates by median home value 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .344a .118 .118 1.88138 .118 420.768 1 3139 .000 
2 .344b .118 .118 1.88168 .000 .003 1 3138 .959 
3 .346c .120 .119 1.88007 .002 6.368 1 3137 .012 
4 .350d .123 .121 1.87766 .003 9.080 1 3136 .003 
a. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units 
b. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units, medval2 
c. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units, medval2, medval3 
d. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units, medval2, medval3, medval4 
 
 
Figure 45: Homelessness rates by median home value linear regression 
  
172 
 
Table 37: ANOVA table regressing homelessness rates by median home value 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 1489.348 1 1489.348 420.768 .000b 
Residual 11110.776 3139 3.540   
Total 12600.124 3140    
2 Regression 1489.357 2 744.678 210.319 .000c 
Residual 11110.767 3138 3.541   
Total 12600.124 3140    
3 Regression 1511.864 3 503.955 142.575 .000d 
Residual 11088.260 3137 3.535   
Total 12600.124 3140    
4 Regression 1543.878 4 385.969 109.477 .000e 
Residual 11056.246 3136 3.526   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units 
c. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units, medval2 
d. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units, medval2, medval3 
e. Predictors: (Constant), Median Home Value of Owner-Occupied Housing Units, medval2, medval3, medval4 
 
  
173 
 
Table 38: Coefficient table regressing homelessness rates by median home value 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) .805 .066  12.182 .000 
Median Home Value of 5.479E-6 .000 .344 20.513 .000 
Owner-Occupied Housing 
Units 
2 (Constant) .800 .123  6.527 .000 
Median Home Value of 5.522E-6 .000 .346 6.283 .000 
Owner-Occupied Housing 
Units 
medval2 -6.293E-14 .000 -.003 -.051 .959 
3 (Constant) 1.280 .226  5.657 .000 
Median Home Value of 2.478E-10 .000 .000 .000 1.000 
Owner-Occupied Housing 
Units 
medval2 1.666E-11 .000 .747 2.472 .013 
medval3 -1.373E-17 .000 -.435 -2.523 .012 
4 (Constant) .217 .419  .517 .605 
Median Home Value of 1.619E-5 .000 1.016 2.760 .006 
Owner-Occupied Housing 
Units 
medval2 -6.044E-11 .000 -2.711 -2.285 .022 
medval3 1.240E-16 .000 3.930 2.694 .007 
medval4 -8.014E-23 .000 -1.926 -3.013 .003 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
174 
 
Table 39: Model fit results regressing homelessness rates by percentage of renters 
in a family with children 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .082a .007 .006 1.95246 .007 18.832 1 2793 .000 
2 .138b .019 .018 1.94052 .012 35.474 1 2792 .000 
3 .146c .021 .020 1.93874 .002 6.122 1 2791 .013 
a. Predictors: (Constant), Percentage of Renters in a Family with Children 
b. Predictors: (Constant), Percentage of Renters in a Family with Children, renfam2 
c. Predictors: (Constant), Percentage of Renters in a Family with Children, renfam2, renfam3 
 
 
Figure 46: Homelessness rates by percentage of renters in a family with children 
quadratic regression 
  
175 
 
Table 40: ANOVA table regressing homelessness rates by percentage of renters in a 
family with children 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 71.791 1 71.791 18.832 .000b 
Residual 10647.162 2793 3.812   
Total 10718.953 2794    
2 Regression 205.371 2 102.685 27.269 .000c 
Residual 10513.582 2792 3.766   
Total 10718.953 2794    
3 Regression 228.380 3 76.127 20.253 .000d 
Residual 10490.573 2791 3.759   
Total 10718.953 2794    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Percentage of Renters in a Family with Children 
c. Predictors: (Constant), Percentage of Renters in a Family with Children, renfam2 
d. Predictors: (Constant), Percentage of Renters in a Family with Children, renfam2, renfam3 
 
Table 41: Coefficient table regressing homelessness rates by percentage of renters 
in a family with children 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 2.778 .197  14.111 .000 
Percentage of Renters in a -.025 .006 -.082 -4.340 .000 
Family with Children 
2 (Constant) 6.350 .631  10.065 .000 
Percentage of Renters in a -.238 .036 -.784 -6.568 .000 
Family with Children 
renfam2 .003 .001 .711 5.956 .000 
3 (Constant) 10.489 1.787  5.868 .000 
Percentage of Renters in a -.625 .161 -2.061 -3.891 .000 
Family with Children 
renfam2 .015 .005 3.384 3.114 .002 
renfam3 .000 .000 -1.426 -2.474 .013 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
176 
 
Table 42: Model fit results regressing homelessness rates by percentage of renter-
occupied housing units 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .380a .145 .144 1.85290 .145 531.037 1 3139 .000 
2 .397b .158 .157 1.83885 .013 49.139 1 3138 .000 
3 .402c .162 .161 1.83477 .004 14.994 1 3137 .000 
a. Predictors: (Constant), Percentage of Renter-Occupied Housing Units 
b. Predictors: (Constant), Percentage of Renter-Occupied Housing Units, renocc2 
c. Predictors: (Constant), Percentage of Renter-Occupied Housing Units, renocc2, renocc3 
 
 
Figure 47: Homelessness rates by percentage of renter-occupied housing units 
linear regression 
  
177 
 
Table 43: ANOVA table regressing homelessness rates by percentage of renter-
occupied housing units 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 1823.177 1 1823.177 531.037 .000b 
Residual 10776.947 3139 3.433   
Total 12600.124 3140    
2 Regression 1989.334 2 994.667 294.160 .000c 
Residual 10610.790 3138 3.381   
Total 12600.124 3140    
3 Regression 2039.810 3 679.937 201.979 .000d 
Residual 10560.314 3137 3.366   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Percentage of Renter-Occupied Housing Units 
c. Predictors: (Constant), Percentage of Renter-Occupied Housing Units, renocc2 
d. Predictors: (Constant), Percentage of Renter-Occupied Housing Units, renocc2, renocc3 
 
Table 44: Coefficient table regressing homelessness rates by percentage of renter-
occupied housing units 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) -1.079 .137  -7.905 .000 
Percentage of Renter- .091 .004 .380 23.044 .000 
Occupied Housing Units 
2 (Constant) 1.448 .385  3.759 .000 
Percentage of Renter- -.053 .021 -.222 -2.535 .011 
Occupied Housing Units 
renocc2 .002 .000 .613 7.010 .000 
3 (Constant) 5.272 1.060  4.975 .000 
Percentage of Renter- -.379 .087 -1.583 -4.370 .000 
Occupied Housing Units 
renocc2 .011 .002 3.371 4.697 .000 
renocc3 -7.212E-5 .000 -1.445 -3.872 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
178 
 
Table 45: Model fit results regressing homelessness rates by percentage of renters 
identifying as white, non-Hispanic 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .135a .018 .018 1.98518 .018 58.248 1 3139 .000 
2 .158b .025 .024 1.97852 .007 22.167 1 3138 .000 
3 .158c .025 .024 1.97883 .000 .000 1 3137 1.000 
a. Predictors: (Constant), Percentage of Renters Identifying as White, Non-Hispanic 
b. Predictors: (Constant), Percentage of Renters Identifying as White, Non-Hispanic, renwhi2 
c. Predictors: (Constant), Percentage of Renters Identifying as White, Non-Hispanic, renwhi2, renwhi3 
 
 
Figure 48: Homelessness rates by percentage of renters identifying as white, non-
Hispanic linear regression 
  
179 
 
Table 46: ANOVA table regressing homelessness rates by percentage of renters 
identifying as white, non-Hispanic 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 229.551 1 229.551 58.248 .000b 
Residual 12370.573 3139 3.941   
Total 12600.124 3140    
2 Regression 316.326 2 158.163 40.404 .000c 
Residual 12283.798 3138 3.915   
Total 12600.124 3140    
3 Regression 316.326 3 105.442 26.927 .000d 
Residual 12283.798 3137 3.916   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Percentage of Renters Identifying as White, Non-Hispanic 
c. Predictors: (Constant), Percentage of Renters Identifying as White, Non-Hispanic, renwhi2 
d. Predictors: (Constant), Percentage of Renters Identifying as White, Non-Hispanic, renwhi2, renwhi3 
 
  
180 
 
Table 47: Coefficient table regressing homelessness rates by percentage of renters 
identifying as white, non-Hispanic 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 2.873 .123  23.332 .000 
Percentage of Renters -.014 .002 -.135 -7.632 .000 
Identifying as White, Non-
Hispanic 
2 (Constant) 1.735 .271  6.396 .000 
Percentage of Renters .029 .009 .277 3.104 .002 
Identifying as White, Non-
Hispanic 
renwhi2 .000 .000 -.420 -4.708 .000 
3 (Constant) 1.735 .391  4.436 .000 
Percentage of Renters .029 .024 .277 1.227 .220 
Identifying as White, Non-
Hispanic 
renwhi2 .000 .000 -.421 -.788 .431 
renwhi3 1.379E-9 .000 .000 .000 1.000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
181 
 
Table 48: Model fit results regressing homelessness rates by percentage of renters 
without any college education 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .154a .024 .023 1.97974 .024 75.842 1 3139 .000 
2 .154b .024 .023 1.98005 .000 .008 1 3138 .927 
3 .216c .047 .046 1.95673 .023 76.246 1 3137 .000 
a. Predictors: (Constant), Percentage of Renters Without any College Education 
b. Predictors: (Constant), Percentage of Renters Without any College Education, renedu2 
c. Predictors: (Constant), Percentage of Renters Without any College Education, renedu2, renedu3 
 
 
Figure 49: Homelessness rates by percentage of renters without any college 
education cubic and linear regressions 
  
182 
 
Table 49: ANOVA table regressing homelessness rates by percentage of renters 
without any college education 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 297.251 1 297.251 75.842 .000b 
Residual 12302.873 3139 3.919   
Total 12600.124 3140    
2 Regression 297.284 2 148.642 37.913 .000c 
Residual 12302.840 3138 3.921   
Total 12600.124 3140    
3 Regression 589.214 3 196.405 51.297 .000d 
Residual 12010.910 3137 3.829   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Percentage of Renters Without any College Education 
c. Predictors: (Constant), Percentage of Renters Without any College Education, renedu2 
d. Predictors: (Constant), Percentage of Renters Without any College Education, renedu2, renedu3 
 
  
183 
 
Table 50: Coefficient table regressing homelessness rates by percentage of renters 
without any college education 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 3.421 .170  20.125 .000 
Percentage of Renters -.032 .004 -.154 -8.709 .000 
Without any College 
Education 
2 (Constant) 3.373 .558  6.048 .000 
Percentage of Renters -.030 .026 -.142 -1.159 .247 
Without any College 
Education 
renedu2 -2.712E-5 .000 -.011 -.092 .927 
3 (Constant) -8.765 1.495  -5.862 .000 
Percentage of Renters .922 .112 4.372 8.232 .000 
Without any College 
Education 
renedu2 -.024 .003 -9.809 -8.691 .000 
renedu3 .000 .000 5.371 8.732 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
184 
 
Table 51: Model fit results regressing homelessness rates by unemployment rate 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .178a .032 .031 1.97166 .032 102.236 1 3139 .000 
2 .184b .034 .033 1.96950 .002 7.880 1 3138 .005 
3 .203c .041 .040 1.96241 .007 23.725 1 3137 .000 
a. Predictors: (Constant), Unemployment Rate 
b. Predictors: (Constant), Unemployment Rate, unemp2 
c. Predictors: (Constant), Unemployment Rate, unemp2, unemp3 
 
 
Figure 50: Homelessness rates by unemployment rate linear regression 
  
185 
 
Table 52: ANOVA table regressing homelessness rates by unemployment rate 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 397.435 1 397.435 102.236 .000b 
Residual 12202.689 3139 3.887   
Total 12600.124 3140    
2 Regression 428.002 2 214.001 55.170 .000c 
Residual 12172.122 3138 3.879   
Total 12600.124 3140    
3 Regression 519.370 3 173.123 44.955 .000d 
Residual 12080.754 3137 3.851   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Unemployment Rate 
c. Predictors: (Constant), Unemployment Rate, unemp2 
d. Predictors: (Constant), Unemployment Rate, unemp2, unemp3 
 
Table 53: Coefficient table regressing homelessness rates by unemployment rate 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) .793 .122  6.511 .000 
Unemployment Rate .142 .014 .178 10.111 .000 
2 (Constant) -.017 .313  -.054 .957 
Unemployment Rate .329 .068 .411 4.834 .000 
unemp2 -.010 .004 -.239 -2.807 .005 
3 (Constant) 3.300 .749  4.405 .000 
Unemployment Rate -.821 .245 -1.027 -3.343 .001 
unemp2 .112 .025 2.723 4.435 .000 
unemp3 -.004 .001 -1.584 -4.871 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
186 
 
Table 54: Model fit results regressing homelessness rates by poverty rate 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .127a .016 .016 1.94318 .016 45.742 1 2793 .000 
2 .164b .027 .026 1.93286 .011 30.894 1 2792 .000 
3 .164c .027 .026 1.93310 .000 .325 1 2791 .569 
4 .164d .027 .026 1.93339 .000 .156 1 2790 .692 
a. Predictors: (Constant), Poverty Rate 
b. Predictors: (Constant), Poverty Rate, pov2 
c. Predictors: (Constant), Poverty Rate, pov2, pov3 
d. Predictors: (Constant), Poverty Rate, pov2, pov3, pov4 
 
 
Figure 51: Homelessness rates by poverty rate linear regression 
  
187 
 
Table 55: ANOVA table regressing homelessness rates by poverty rate 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 172.718 1 172.718 45.742 .000b 
Residual 10546.235 2793 3.776   
Total 10718.953 2794    
2 Regression 288.139 2 144.069 38.563 .000c 
Residual 10430.814 2792 3.736   
Total 10718.953 2794    
3 Regression 289.353 3 96.451 25.811 .000d 
Residual 10429.600 2791 3.737   
Total 10718.953 2794    
4 Regression 289.938 4 72.485 19.391 .000e 
Residual 10429.015 2790 3.738   
Total 10718.953 2794    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Poverty Rate 
c. Predictors: (Constant), Poverty Rate, pov2 
d. Predictors: (Constant), Poverty Rate, pov2, pov3 
e. Predictors: (Constant), Poverty Rate, pov2, pov3, pov4 
 
  
188 
 
Table 56: Coefficient table regressing homelessness rates by poverty rate 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.224 .112  10.930 .000 
Poverty Rate .048 .007 .127 6.763 .000 
2 (Constant) .357 .192  1.860 .063 
Poverty Rate .156 .021 .412 7.549 .000 
pov2 -.003 .001 -.303 -5.558 .000 
3 (Constant) .539 .372  1.446 .148 
Poverty Rate .122 .063 .322 1.926 .054 
pov2 -.001 .003 -.120 -.368 .713 
pov3 -2.607E-5 .000 -.104 -.570 .569 
4 (Constant) .333 .639  .521 .603 
Poverty Rate .176 .152 .466 1.164 .245 
pov2 -.006 .012 -.595 -.478 .633 
pov3 .000 .000 .516 .327 .744 
pov4 -1.681E-6 .000 -.289 -.396 .692 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
189 
 
Table 57: Model fit results regressing homelessness rates by eviction filing rate 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .053a .003 .002 1.96953 .003 6.508 1 2345 .011 
2 .087b .008 .007 1.96527 .005 11.178 1 2344 .001 
3 .091c .008 .007 1.96486 .001 1.962 1 2343 .161 
4 .092d .009 .007 1.96509 .000 .459 1 2342 .498 
a. Predictors: (Constant), Eviction Filing Rate 
b. Predictors: (Constant), Eviction Filing Rate, evic2 
c. Predictors: (Constant), Eviction Filing Rate, evic2, evic3 
d. Predictors: (Constant), Eviction Filing Rate, evic2, evic3, evic4 
 
 
Figure 52: Homelessness rates by eviction filing rate quadratic regression 
  
190 
 
Table 58: ANOVA table regressing homelessness rates by eviction filing rate 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 25.243 1 25.243 6.508 .011b 
Residual 9096.346 2345 3.879   
Total 9121.590 2346    
2 Regression 68.416 2 34.208 8.857 .000c 
Residual 9053.174 2344 3.862   
Total 9121.590 2346    
3 Regression 75.989 3 25.330 6.561 .000d 
Residual 9045.600 2343 3.861   
Total 9121.590 2346    
4 Regression 77.761 4 19.440 5.034 .000e 
Residual 9043.828 2342 3.862   
Total 9121.590 2346    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Eviction Filing Rate 
c. Predictors: (Constant), Eviction Filing Rate, evic2 
d. Predictors: (Constant), Eviction Filing Rate, evic2, evic3 
e. Predictors: (Constant), Eviction Filing Rate, evic2, evic3, evic4 
 
  
191 
 
Table 59: Coefficient table regressing homelessness rates by eviction filing rate 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 2.046 .057  35.708 .000 
Eviction Filing Rate -.017 .007 -.053 -2.551 .011 
2 (Constant) 2.200 .073  29.967 .000 
Eviction Filing Rate -.073 .018 -.225 -4.053 .000 
evic2 .003 .001 .185 3.343 .001 
3 (Constant) 2.269 .089  25.642 .000 
Eviction Filing Rate -.113 .033 -.346 -3.366 .001 
evic2 .006 .003 .472 2.226 .026 
evic3 -9.156E-5 .000 -.182 -1.401 .161 
4 (Constant) 2.233 .104  21.567 .000 
Eviction Filing Rate -.081 .057 -.249 -1.415 .157 
evic2 .001 .008 .074 .118 .906 
evic3 .000 .000 .384 .454 .650 
evic4 -4.455E-6 .000 -.271 -.677 .498 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
192 
 
Table 60: Model fit results regressing homelessness rates by percentage of rent-
burdened households 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .285a .081 .081 1.92049 .081 277.258 1 3139 .000 
2 .287b .082 .082 1.91974 .001 3.445 1 3138 .064 
a. Predictors: (Constant), Percentage of Rent-Burdened Households 
b. Predictors: (Constant), Percentage of Rent-Burdened Households, burd2 
 
 
Figure 53: Homelessness rates by percentage of rent-burdened households linear 
regression 
  
193 
 
Table 61: ANOVA table regressing homelessness rates by percentage of rent-
burdened households 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 1022.607 1 1022.607 277.258 .000b 
Residual 11577.517 3139 3.688   
Total 12600.124 3140    
2 Regression 1035.304 2 517.652 140.460 .000c 
Residual 11564.820 3138 3.685   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Percentage of Rent-Burdened Households 
c. Predictors: (Constant), Percentage of Rent-Burdened Households, burd2 
 
Table 62: Coefficients table regressing homelessness rates by percentage of rent-
burdened households 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) -4.954 .417  -11.868 .000 
Percentage of Rent- .225 .014 .285 16.651 .000 
Burdened Households 
2 (Constant) 1.949 3.742  .521 .602 
Percentage of Rent- -.218 .239 -.276 -.911 .362 
Burdened Households 
burd2 .007 .004 .561 1.856 .064 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
194 
 
Table 63: Model fit results regressing homelessness rates by Gini index 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .268a .072 .071 1.88739 .072 216.052 1 2793 .000 
2 .277b .077 .076 1.88244 .005 15.710 1 2792 .000 
a. Predictors: (Constant), Gini Index 
b. Predictors: (Constant), Gini Index, gini2 
 
 
Figure 54: Homelessness rates by Gini Index linear regression 
  
195 
 
Table 64: ANOVA table regressing homelessness rates by Gini index 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 769.629 1 769.629 216.052 .000b 
Residual 9949.324 2793 3.562   
Total 10718.953 2794    
2 Regression 825.299 2 412.649 116.450 .000c 
Residual 9893.654 2792 3.544   
Total 10718.953 2794    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Gini Index 
c. Predictors: (Constant), Gini Index, gini2 
 
Table 65: Coefficient table regressing homelessness rates by Gini index 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) -5.911 .535  -11.044 .000 
Gini Index 17.383 1.183 .268 14.699 .000 
2 (Constant) 14.455 5.166  2.798 .005 
Gini Index -72.365 22.674 -1.116 -3.192 .001 
gini2 98.430 24.834 1.385 3.964 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
196 
 
Table 66: Model fit results regressing homelessness rates by vacancy rate 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .129a .017 .016 1.98673 .017 53.244 1 3139 .000 
2 .150b .022 .022 1.98130 .006 18.245 1 3138 .000 
3 .165c .027 .026 1.97673 .005 15.524 1 3137 .000 
4 .177d .031 .030 1.97280 .004 13.510 1 3136 .000 
a. Predictors: (Constant), Vacancy Rate 
b. Predictors: (Constant), Vacancy Rate, vac2 
c. Predictors: (Constant), Vacancy Rate, vac2, vac3 
d. Predictors: (Constant), Vacancy Rate, vac2, vac3, vac4 
 
 
Figure 55: Homelessness rates by vacancy rate linear regression 
 
  
197 
 
Table 67: ANOVA table regressing homelessness rates by vacancy rate 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 210.159 1 210.159 53.244 .000b 
Residual 12389.965 3139 3.947   
Total 12600.124 3140    
2 Regression 281.782 2 140.891 35.891 .000c 
Residual 12318.342 3138 3.926   
Total 12600.124 3140    
3 Regression 342.442 3 114.147 29.213 .000d 
Residual 12257.682 3137 3.907   
Total 12600.124 3140    
4 Regression 395.020 4 98.755 25.374 .000e 
Residual 12205.104 3136 3.892   
Total 12600.124 3140    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Vacancy Rate 
c. Predictors: (Constant), Vacancy Rate, vac2 
d. Predictors: (Constant), Vacancy Rate, vac2, vac3 
e. Predictors: (Constant), Vacancy Rate, vac2, vac3, vac4 
 
  
198 
 
Table 68: Coefficients table regressing homelessness rates by vacancy rate 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.482 .076  19.501 .000 
Vacancy Rate .039 .005 .129 7.297 .000 
2 (Constant) 1.982 .139  14.224 .000 
Vacancy Rate -.032 .017 -.103 -1.808 .071 
vac2 .002 .000 .245 4.271 .000 
3 (Constant) 1.132 .256  4.415 .000 
Vacancy Rate .148 .049 .487 3.036 .002 
vac2 -.008 .003 -1.058 -3.154 .002 
vac3 .000 .000 .766 3.940 .000 
4 (Constant) 2.645 .485  5.458 .000 
Vacancy Rate -.282 .127 -.925 -2.223 .026 
vac2 .030 .011 3.819 2.791 .005 
vac3 -.001 .000 -5.481 -3.204 .001 
vac4 1.478E-5 .000 2.777 3.676 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
199 
 
Table 69: Model fit summary regressing homelessness rates by mean temperature 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .301a .090 .090 1.85850 .090 301.778 1 3037 .000 
2 .315b .099 .098 1.85004 .009 28.826 1 3036 .000 
3 .324c .105 .104 1.84428 .006 20.006 1 3035 .000 
4 .329d .108 .107 1.84097 .004 11.926 1 3034 .001 
a. Predictors: (Constant), Mean Temperature in January in Fahrenheit 
b. Predictors: (Constant), Mean Temperature in January in Fahrenheit, temp2 
c. Predictors: (Constant), Mean Temperature in January in Fahrenheit, temp2, temp3 
d. Predictors: (Constant), Mean Temperature in January in Fahrenheit, temp2, temp3, temp4 
 
 
Figure 56: Homelessness rates by mean temperature in January linear regression 
  
200 
 
Table 70: ANOVA table regressing homelessness rates by mean temperature 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 1042.345 1 1042.345 301.778 .000b 
Residual 10489.841 3037 3.454   
Total 11532.186 3038    
2 Regression 1141.008 2 570.504 166.685 .000c 
Residual 10391.178 3036 3.423   
Total 11532.186 3038    
3 Regression 1209.055 3 403.018 118.487 .000d 
Residual 10323.131 3035 3.401   
Total 11532.186 3038    
4 Regression 1249.474 4 312.368 92.167 .000e 
Residual 10282.713 3034 3.389   
Total 11532.186 3038    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Mean Temperature in January in Fahrenheit 
c. Predictors: (Constant), Mean Temperature in January in Fahrenheit, temp2 
d. Predictors: (Constant), Mean Temperature in January in Fahrenheit, temp2, temp3 
e. Predictors: (Constant), Mean Temperature in January in Fahrenheit, temp2, temp3, temp4 
 
  
201 
 
Table 71: Coefficients table regressing homelessness rates by mean temperature 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) .381 .096  3.960 .000 
Mean Temperature in .045 .003 .301 17.372 .000 
January in Fahrenheit 
2 (Constant) 1.309 .198  6.625 .000 
Mean Temperature in -.014 .011 -.092 -1.220 .223 
January in Fahrenheit 
temp2 .001 .000 .403 5.369 .000 
3 (Constant) 2.238 .286  7.818 .000 
Mean Temperature in -.121 .027 -.814 -4.573 .000 
January in Fahrenheit 
temp2 .004 .001 2.145 5.409 .000 
temp3 -3.256E-5 .000 -1.054 -4.473 .000 
4 (Constant) 1.817 .311  5.850 .000 
Mean Temperature in -.011 .041 -.075 -.271 .786 
January in Fahrenheit 
temp2 -.003 .002 -1.319 -1.223 .221 
temp3 .000 .000 3.955 2.691 .007 
temp4 -1.133E-6 .000 -2.304 -3.453 .001 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
202 
 
Table 72: Model fit results regressing homelessness rates by total precipitation 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .160a .026 .025 1.92041 .026 79.918 1 3029 .000 
2 .211b .045 .044 1.90201 .019 59.892 1 3028 .000 
3 .224c .050 .049 1.89689 .005 17.364 1 3027 .000 
4 .224d .050 .049 1.89717 .000 .130 1 3026 .719 
a. Predictors: (Constant), Total Precipitation in January in Inches 
b. Predictors: (Constant), Total Precipitation in January in Inches, precip2 
c. Predictors: (Constant), Total Precipitation in January in Inches, precip2, precip3 
d. Predictors: (Constant), Total Precipitation in January in Inches, precip2, precip3, precip4 
 
 
Figure 57: Homelessness rates by total precipitation in January linear regression 
  
203 
 
Table 73: ANOVA table regressing homelessness rates by total precipitation 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 294.736 1 294.736 79.918 .000b 
Residual 11170.915 3029 3.688   
Total 11465.651 3030    
2 Regression 511.405 2 255.702 70.682 .000c 
Residual 10954.246 3028 3.618   
Total 11465.651 3030    
3 Regression 573.883 3 191.294 53.164 .000d 
Residual 10891.768 3027 3.598   
Total 11465.651 3030    
4 Regression 574.350 4 143.587 39.894 .000e 
Residual 10891.301 3026 3.599   
Total 11465.651 3030    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Total Precipitation in January in Inches 
c. Predictors: (Constant), Total Precipitation in January in Inches, precip2 
d. Predictors: (Constant), Total Precipitation in January in Inches, precip2, precip3 
e. Predictors: (Constant), Total Precipitation in January in Inches, precip2, precip3, precip4 
 
  
204 
 
Table 74: Coefficients table regressing homelessness rates by total precipitation 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.581 .053  29.562 .000 
Total Precipitation in .124 .014 .160 8.940 .000 
January in Inches 
2 (Constant) 1.950 .071  27.375 .000 
Total Precipitation in -.090 .031 -.116 -2.909 .004 
January in Inches 
precip2 .017 .002 .309 7.739 .000 
3 (Constant) 2.165 .088  24.654 .000 
Total Precipitation in -.280 .055 -.361 -5.086 .000 
January in Inches 
precip2 .051 .008 .916 6.064 .000 
precip3 -.001 .000 -.409 -4.167 .000 
4 (Constant) 2.185 .104  21.017 .000 
Total Precipitation in -.306 .091 -.395 -3.355 .001 
January in Inches 
precip2 .059 .023 1.055 2.539 .011 
precip3 -.002 .002 -.622 -1.038 .299 
precip4 1.868E-5 .000 .109 .360 .719 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
205 
 
Table 75: Model fit results regressing homelessness rates by Housing First index 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .003a .000 .000 2.00441 .000 .032 1 3134 .859 
2 .094b .009 .008 1.99594 .009 27.627 1 3133 .000 
3 .094c .009 .008 1.99625 .000 .045 1 3132 .831 
a. Predictors: (Constant), Housing First index 
b. Predictors: (Constant), Housing First index, hf2 
c. Predictors: (Constant), Housing First index, hf2, hf3 
 
 
Figure 58: Homelessness rates by Housing First index quadratic regression 
  
206 
 
Table 76: ANOVA table regressing homelessness rates by Housing First index 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression .128 1 .128 .032 .859b 
Residual 12591.294 3134 4.018   
Total 12591.421 3135    
2 Regression 110.189 2 55.094 13.830 .000c 
Residual 12481.233 3133 3.984   
Total 12591.421 3135    
3 Regression 110.369 3 36.790 9.232 .000d 
Residual 12481.052 3132 3.985   
Total 12591.421 3135    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), Housing First index 
c. Predictors: (Constant), Housing First index, hf2 
d. Predictors: (Constant), Housing First index, hf2, hf3 
 
Table 77: Coefficients table regressing homelessness rates by Housing First index 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.994 .126  15.775 .000 
Housing First index -.040 .226 -.003 -.178 .859 
2 (Constant) .744 .269  2.764 .006 
Housing First index 5.454 1.069 .432 5.101 .000 
hf2 -5.422 1.032 -.445 -5.256 .000 
3 (Constant) .677 .414  1.636 .102 
Housing First index 6.012 2.829 .476 2.125 .034 
hf2 -6.716 6.165 -.551 -1.089 .276 
hf3 .893 4.196 .064 .213 .831 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
207 
 
Table 78: Model fit results regressing homelessness rates by HUD CoC funding 
Model Summary 
Change Statistics 
R Adjusted R Std. Error of R Square F Sig. F 
Model R Square Square the Estimate Change Change df1 df2 Change 
1 .390a .152 .152 1.83842 .152 553.701 1 3089 .000 
2 .400b .160 .159 1.83023 .008 28.696 1 3088 .000 
3 .402c .161 .161 1.82884 .002 5.707 1 3087 .017 
4 .405d .164 .163 1.82666 .002 8.365 1 3086 .004 
a. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person 
b. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person, fund2 
c. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person, fund2, fund3 
d. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person, fund2, fund3, fund4 
 
 
Figure 59: Homelessness rates by CoC funding in the previous year linear 
regression 
 
  
208 
 
Table 79: ANOVA table regressing homelessness rates by HUD CoC funding 
ANOVAa 
Model Sum of Squares df Mean Square F Sig. 
1 Regression 1871.392 1 1871.392 553.701 .000b 
Residual 10440.172 3089 3.380   
Total 12311.564 3090    
2 Regression 1967.515 2 983.758 293.680 .000c 
Residual 10344.049 3088 3.350   
Total 12311.564 3090    
3 Regression 1986.604 3 662.201 197.988 .000d 
Residual 10324.960 3087 3.345   
Total 12311.564 3090    
4 Regression 2014.516 4 503.629 150.936 .000e 
Residual 10297.048 3086 3.337   
Total 12311.564 3090    
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
b. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person 
c. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person, fund2 
d. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person, fund2, fund3 
e. Predictors: (Constant), HUD CoC Funding in the Previous Year per Person, fund2, fund3, fund4 
 
  
209 
 
Table 80: Coefficients table regressing homelessness rates by HUD CoC funding 
Coefficientsa 
Standardized 
Unstandardized Coefficients Coefficients 
Model B Std. Error Beta t Sig. 
1 (Constant) 1.191 .047  25.473 .000 
HUD CoC Funding in the .151 .006 .390 23.531 .000 
Previous Year per Person 
2 (Constant) 1.415 .063  22.628 .000 
HUD CoC Funding in the .078 .015 .201 5.143 .000 
Previous Year per Person 
fund2 .003 .001 .209 5.357 .000 
3 (Constant) 1.529 .079  19.414 .000 
HUD CoC Funding in the .019 .029 .050 .673 .501 
Previous Year per Person 
fund2 .009 .002 .615 3.528 .000 
fund3 .000 .000 -.279 -2.389 .017 
4 (Constant) 1.373 .095  14.400 .000 
HUD CoC Funding in the .124 .046 .321 2.687 .007 
Previous Year per Person 
fund2 -.007 .006 -.518 -1.208 .227 
fund3 .001 .000 1.406 2.366 .018 
fund4 -9.960E-6 .000 -.822 -2.892 .004 
a. Dependent Variable: People per 1,000 Experiencing Homelessness 
 
  
210 
 
Appendix 5: Model Testing Results 
Table 81: Autoregressive residual covariance matrix for yearcoded repeating 
variable 
Residual Covariance (R) Matrixa 
[yearcoded [yearcoded [yearcoded [yearcoded [yearcoded [yearcoded [yearcoded 
 = 1.00] = 2.00] = 3.00] = 4.00] = 5.00] = 6.00] = 7.00] 
[yearcoded = 2.158026 1.881578 1.640544 1.430387 1.247151 1.087388 .948092 
1.00] 
[yearcoded = 1.881578 2.158026 1.881578 1.640544 1.430387 1.247151 1.087388 
2.00] 
[yearcoded = 1.640544 1.881578 2.158026 1.881578 1.640544 1.430387 1.247151 
3.00] 
[yearcoded = 1.430387 1.640544 1.881578 2.158026 1.881578 1.640544 1.430387 
4.00] 
[yearcoded = 1.247151 1.430387 1.640544 1.881578 2.158026 1.881578 1.640544 
5.00] 
[yearcoded = 1.087388 1.247151 1.430387 1.640544 1.881578 2.158026 1.881578 
6.00] 
[yearcoded = .948092 1.087388 1.247151 1.430387 1.640544 1.881578 2.158026 
7.00] 
First-Order Autoregressive 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 82: Information criteria for Model 1 measuring main effects of all variables 
Information Criteriaa 
-2 Restricted Log Likelihood 5117.713 
Akaike's Information Criterion (AIC) 5123.713 
Hurvich and Tsai's Criterion (AICC) 5123.725 
Bozdogan's Criterion (CAIC) 5143.484 
Schwarz's Bayesian Criterion (BIC) 5140.484 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
211 
 
Table 83: Type III tests of fixed effects for Model 1 measuring main effects of all 
variables 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 6859.020 .103 .748 
coccat 2 311.003 4.061 .018 
inczon 1 320.766 1.049 .307 
pop 0 . . . 
bach 1 810.553 3.016 .083 
medinc 1 803.556 .468 .494 
medren 1 922.549 18.395 .000 
medval 1 828.139 10.432 .001 
renfam 1 666.197 .540 .463 
renocc 1 545.749 .164 .686 
renwhi 1 585.687 3.854 .050 
renedu 1 746.652 5.821 .016 
unemp 1 1647.906 1.409 .235 
pov 1 1320.456 .779 .378 
evic 1 1422.967 .114 .736 
burd 1 1729.809 6.754 .009 
gini 1 1002.101 .046 .830 
vac 1 452.685 2.584 .109 
temp 1 1837.697 2.424 .120 
precip 1 1696.294 1.837 .176 
hf 1 1977.624 .299 .584 
fund 1 753.774 24.652 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 633.257 .508 .476 
evic2 1 1880.448 .146 .702 
hf2 1 1966.959 1.150 .284 
yearcoded 1 1339.173 23.222 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
212 
 
Table 84: Fixed effects estimates for Model 1 measuring main effects of all 
variables 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.626141 5.523480 6636.718 .294 .768 -9.201654 12.453937 
[coccat=1] .743383 .368418 307.441 2.018 .044 .018444 1.468322 
[coccat=2] .001605 .329376 316.705 .005 .996 -.646438 .649647 
[coccat=3] 0b 0 . . . . . 
[inczon=0] -.206832 .201961 320.766 -1.024 .307 -.604169 .190504 
[inczon=1] 0b 0 . . . . . 
pop -7.1892E-7 2.06991E-7 310.007 -3.473 .001 -1.12621E-6 -3.11637E-7 
bach -.038692 .022279 810.553 -1.737 .083 -.082423 .005040 
medinc -2.8937E-5 4.22782E-5 803.556 -.684 .494 -.000112 5.40513E-5 
medren .004398 .001025 922.549 4.289 .000 .002385 .006410 
medval 4.1306E-6 1.27893E-6 828.139 3.230 .001 1.62036E-6 6.64102E-6 
renfam .051218 .069706 666.197 .735 .463 -.085652 .188089 
renocc .007266 .017937 545.749 .405 .686 -.027968 .042499 
renwhi .014115 .007190 585.687 1.963 .050 -5.96438E-6 .028235 
renedu -.032513 .013476 746.652 -2.413 .016 -.058969 -.006058 
unemp .031997 .026951 1647.906 1.187 .235 -.020865 .084858 
pov .035939 .040715 1320.456 .883 .378 -.043935 .115813 
evic .007692 .022786 1422.967 .338 .736 -.037005 .052390 
burd -.078772 .030310 1729.809 -2.599 .009 -.138220 -.019323 
gini .827676 3.861156 1002.101 .214 .830 -6.749202 8.404555 
vac .024776 .015413 452.685 1.607 .109 -.005514 .055065 
temp .004907 .003151 1837.697 1.557 .120 -.001274 .011087 
precip -.011155 .008231 1696.294 -1.355 .176 -.027299 .004989 
hf .562213 1.027655 1977.624 .547 .584 -1.453188 2.577614 
fund .080640 .016241 753.774 4.965 .000 .048757 .112524 
pop2 7.911E-14 2.4860E-14 308.876 3.182 .002 3.0196E-14 1.2802E-13 
medinc2 -2.874E-10 2.5053E-10 629.693 -1.147 .252 -7.7940E-10 2.0457E-10 
renfam2 -.000700 .000982 633.257 -.713 .476 -.002629 .001229 
evic2 -.000287 .000749 1880.448 -.383 .702 -.001755 .001182 
hf2 -1.022895 .953825 1966.959 -1.072 .284 -2.893509 .847719 
yearcoded -.139791 .029008 1339.173 -4.819 .000 -.196697 -.082884 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
213 
 
Table 85: Information criteria for Model 2 removing gini 
Information Criteriaa 
-2 Restricted Log Likelihood 5122.298 
Akaike's Information Criterion (AIC) 5126.298 
Hurvich and Tsai's Criterion (AICC) 5126.304 
Bozdogan's Criterion (CAIC) 5139.480 
Schwarz's Bayesian Criterion (BIC) 5137.480 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 86: Type III tests of fixed effects for Model 2 removing gini 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 858.376 .483 .487 
inczon 1 322.018 1.068 .302 
coccat 2 310.346 4.075 .018 
pop 0 . . . 
bach 1 648.554 3.389 .066 
medinc 1 809.849 .513 .474 
medren 1 922.176 18.399 .000 
medval 1 777.154 11.255 .001 
renfam 1 660.682 .572 .450 
renocc 1 546.508 .166 .684 
renwhi 1 555.609 3.884 .049 
renedu 1 637.635 6.039 .014 
unemp 1 1648.679 1.411 .235 
pov 1 1321.576 .853 .356 
evic 1 1424.129 .108 .743 
burd 1 1732.914 6.713 .010 
vac 1 447.438 2.748 .098 
temp 1 1843.563 2.446 .118 
precip 1 1696.426 1.830 .176 
hf 1 1978.566 .293 .588 
fund 1 758.594 24.964 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 622.168 .551 .458 
evic2 1 1883.307 .145 .704 
214 
 
hf2 1 1967.651 1.142 .285 
yearcoded 1 1311.031 25.039 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 87: Fixed effects estimates for Model 2 removing gini 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.907563 2.989405 838.768 .638 .524 -3.960031 7.775156 
[inczon=0] -.208400 .201651 322.018 -1.033 .302 -.605120 .188320 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .750463 .366460 306.311 2.048 .041 .029365 1.471562 
[coccat=2] .009978 .326671 315.213 .031 .976 -.632753 .652708 
[coccat=3] 0b 0 . . . . . 
pop -7.14421E-7 2.05754E-7 308.807 -3.472 .001 -1.11927E-6 -3.09562E-7 
bach -.036486 .019818 648.554 -1.841 .066 -.075402 .002430 
medinc -3.00508E-5 4.19735E-5 809.849 -.716 .474 -.000112 5.233890E-5 
medren .004396 .001025 922.176 4.289 .000 .002384 .006407 
medval 4.190620E-6 1.24913E-6 777.154 3.355 .001 1.738549E-6 6.642692E-6 
renfam .052503 .069408 660.682 .756 .450 -.083785 .188790 
renocc .007308 .017921 546.508 .408 .684 -.027894 .042509 
renwhi .013751 .006978 555.609 1.971 .049 4.516017E-5 .027456 
renedu -.031681 .012892 637.635 -2.457 .014 -.056998 -.006365 
unemp .032005 .026941 1648.679 1.188 .235 -.020837 .084846 
pov .037186 .040272 1321.576 .923 .356 -.041817 .116190 
evic .007470 .022751 1424.129 .328 .743 -.037159 .052099 
burd -.078370 .030247 1732.914 -2.591 .010 -.137694 -.019045 
vac .025243 .015228 447.438 1.658 .098 -.004684 .055169 
temp .004927 .003150 1843.563 1.564 .118 -.001251 .011104 
precip -.011134 .008230 1696.426 -1.353 .176 -.027276 .005007 
hf .556092 1.026889 1978.566 .542 .588 -1.457805 2.569989 
fund .080918 .016195 758.594 4.996 .000 .049125 .112710 
pop2 7.86203E-14 2.4733E-14 307.157 3.179 .002 2.99523E-14 1.27288E-13 
medinc2 -2.8883E-10 2.5024E-10 626.396 -1.154 .249 -7.8024E-10 2.02585E-10 
renfam2 -.000724 .000975 622.168 -.742 .458 -.002639 .001192 
evic2 -.000285 .000748 1883.307 -.380 .704 -.001752 .001183 
hf2 -1.018889 .953330 1967.651 -1.069 .285 -2.888531 .850754 
yearcoded -.137819 .027542 1311.031 -5.004 .000 -.191850 -.083787 
215 
 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
216 
 
Table 88: Information criteria for Model 3 adding inczon * gini interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 5112.382 
Akaike's Information Criterion (AIC) 5116.382 
Hurvich and Tsai's Criterion (AICC) 5116.388 
Bozdogan's Criterion (CAIC) 5129.562 
Schwarz's Bayesian Criterion (BIC) 5127.562 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 89: Type III tests of fixed effects for Model 3 adding inczon * gini interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 821.793 .313 .576 
inczon 1 545.271 .218 .641 
coccat 2 309.384 4.017 .019 
pop 0 . . . 
bach 1 805.087 3.110 .078 
medinc 1 814.394 .574 .449 
medren 1 916.386 17.939 .000 
medval 1 834.274 10.158 .001 
renfam 1 664.252 .558 .455 
renocc 1 541.821 .155 .694 
renwhi 1 586.157 3.650 .057 
renedu 1 745.521 6.059 .014 
unemp 1 1649.571 1.343 .247 
pov 1 1314.204 .792 .374 
evic 1 1419.282 .105 .746 
burd 1 1746.575 6.510 .011 
vac 1 450.352 2.538 .112 
temp 1 1837.088 2.425 .120 
precip 1 1695.448 1.841 .175 
hf 1 1976.750 .288 .592 
fund 1 748.994 24.489 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 631.713 .529 .467 
evic2 1 1876.625 .140 .708 
217 
 
hf2 1 1966.200 1.127 .288 
yearcoded 1 1332.298 23.069 .000 
inczon * gini 2 718.713 .181 .834 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 90: Fixed effects estimates for Model 3 adding inczon * gini interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.049835 3.421242 755.317 .307 .759 -5.666439 7.766108 
[inczon=0] 1.034062 2.215126 545.271 .467 .641 -3.317165 5.385288 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .743894 .368245 305.771 2.020 .044 .019278 1.468510 
[coccat=2] .007511 .329397 314.975 .023 .982 -.640587 .655608 
[coccat=3] 0b 0 . . . . . 
pop -7.058659E-7 2.081809E-7 310.349 -3.391 .001 -1.115490E-6 -2.962414E-7 
bach -.039340 .022308 805.087 -1.763 .078 -.083130 .004449 
medinc -3.232463E-5 4.268087E-5 814.394 -.757 .449 -.000116 5.145285E-5 
medren .004355 .001028 916.386 4.235 .000 .002337 .006372 
medval 4.084599E-6 1.281562E-6 834.274 3.187 .001 1.569134E-6 6.600063E-6 
renfam .052093 .069708 664.252 .747 .455 -.084781 .188967 
renocc .007053 .017935 541.821 .393 .694 -.028178 .042283 
renwhi .013779 .007213 586.157 1.910 .057 -.000387 .027945 
renedu -.033382 .013561 745.521 -2.462 .014 -.060005 -.006759 
unemp .031272 .026984 1649.571 1.159 .247 -.021655 .084199 
pov .036247 .040718 1314.204 .890 .374 -.043633 .116128 
evic .007398 .022792 1419.282 .325 .746 -.037312 .052107 
burd -.077536 .030389 1746.575 -2.551 .011 -.137139 -.017933 
vac .024554 .015412 450.352 1.593 .112 -.005734 .054842 
temp .004909 .003152 1837.088 1.557 .120 -.001274 .011091 
precip -.011172 .008233 1695.448 -1.357 .175 -.027320 .004977 
hf .551450 1.028047 1976.750 .536 .592 -1.464720 2.567621 
fund .080398 .016246 748.994 4.949 .000 .048504 .112292 
pop2 7.728713E- 2.50577E-14 309.619 3.084 .002 2.798212E- 1.265921E-
14 14 13 
medinc2 -2.57821E-10 2.55852E-10 654.649 -1.008 .314 -7.60211E-10 2.44569E-10 
renfam2 -.000715 .000982 631.713 -.728 .467 -.002644 .001214 
evic2 -.000280 .000749 1876.625 -.374 .708 -.001749 .001189 
hf2 -1.013185 .954198 1966.200 -1.062 .288 -2.884531 .858160 
218 
 
yearcoded -.139366 .029016 1332.298 -4.803 .000 -.196289 -.082444 
[inczon=0] * -.287419 4.338840 959.590 -.066 .947 -8.802128 8.227291 
gini 
[inczon=1] * 2.467155 4.838609 718.919 .510 .610 -7.032337 11.966646 
gini 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
219 
 
Table 91: Information criteria for Model 4 adding coccat * gini interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 5103.325 
Akaike's Information Criterion (AIC) 5107.325 
Hurvich and Tsai's Criterion (AICC) 5107.331 
Bozdogan's Criterion (CAIC) 5120.504 
Schwarz's Bayesian Criterion (BIC) 5118.504 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 92: Type III tests of fixed effects for Model 4 adding coccat * gini interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 765.047 .002 .968 
inczon 1 319.073 1.053 .306 
coccat 2 532.163 .653 .521 
pop 0 . . . 
bach 1 801.793 3.079 .080 
medinc 1 799.070 .459 .498 
medren 1 918.671 17.572 .000 
medval 1 822.203 10.426 .001 
renfam 1 666.796 .721 .396 
renocc 1 547.614 .217 .641 
renwhi 1 582.998 3.905 .049 
renedu 1 744.557 6.524 .011 
unemp 1 1638.611 1.348 .246 
pov 1 1315.617 .600 .439 
evic 1 1412.195 .118 .731 
burd 1 1726.639 6.339 .012 
vac 1 451.325 2.709 .100 
temp 1 1836.695 2.454 .117 
precip 1 1695.319 1.829 .176 
hf 1 1975.925 .313 .576 
fund 1 759.144 23.090 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 630.619 .634 .426 
evic2 1 1874.421 .153 .696 
220 
 
hf2 1 1966.318 1.165 .281 
yearcoded 1 1319.671 23.237 .000 
coccat * gini 3 645.694 .615 .605 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 93: Fixed effects estimates for Model 4 adding coccat * gini interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept -.549985 6.059427 670.969 -.091 .928 -12.447705 11.347735 
[inczon=0] -.206973 .201726 319.073 -1.026 .306 -.603854 .189908 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -1.094950 5.946213 553.302 -.184 .854 -12.774862 10.584962 
[coccat=2] 2.599959 5.106646 590.786 .509 .611 -7.429430 12.629348 
[coccat=3] 0b 0 . . . . . 
pop -7.059128E-7 2.068568E-7 308.370 -3.413 .001 -1.112942E-6 -2.988834E-7 
bach -.039045 .022252 801.793 -1.755 .080 -.082724 .004634 
medinc -2.864934E-5 4.228225E-5 799.070 -.678 .498 -.000112 5.434805E-5 
medren .004308 .001028 918.671 4.192 .000 .002291 .006324 
medval 4.132776E-6 1.279918E-6 822.203 3.229 .001 1.620485E-6 6.645067E-6 
renfam .059327 .069845 666.796 .849 .396 -.077816 .196470 
renocc .008386 .017986 547.614 .466 .641 -.026944 .043716 
renwhi .014222 .007197 582.998 1.976 .049 8.713853E-5 .028356 
renedu -.034597 .013545 744.557 -2.554 .011 -.061187 -.008007 
unemp .031287 .026944 1638.611 1.161 .246 -.021560 .084135 
pov .031617 .040811 1315.617 .775 .439 -.048446 .111679 
evic .007834 .022771 1412.195 .344 .731 -.036836 .052503 
burd -.076410 .030348 1726.639 -2.518 .012 -.135932 -.016888 
vac .025357 .015405 451.325 1.646 .100 -.004917 .055631 
temp .004939 .003153 1836.695 1.566 .117 -.001245 .011122 
precip -.011139 .008236 1695.319 -1.352 .176 -.027293 .005016 
hf .576151 1.029992 1975.925 .559 .576 -1.443835 2.596136 
fund .078568 .016350 759.144 4.805 .000 .046470 .110665 
pop2 7.55962E-14 2.49863E-14 308.852 3.026 .003 2.64312E-14 1.24761E-13 
medinc2 -2.90236E-10 2.50575E-10 626.639 -1.158 .247 -7.82306E-10 2.01832E-10 
renfam2 -.000782 .000983 630.619 -.796 .426 -.002712 .001147 
evic2 -.000293 .000749 1874.421 -.391 .696 -.001761 .001175 
hf2 -1.030999 .955343 1966.318 -1.079 .281 -2.904591 .842593 
yearcoded -.140427 .029131 1319.671 -4.821 .000 -.197575 -.083279 
221 
 
[coccat=1] * 9.202982 7.707563 551.714 1.194 .233 -5.936777 24.342741 
gini 
[coccat=2] * -.235590 3.937730 968.957 -.060 .952 -7.963052 7.491871 
gini 
[coccat=3] * 5.698489 11.996953 641.467 .475 .635 -17.859556 29.256534 
gini 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
222 
 
Table 94: Information criteria for Model 21 adding gini * hf interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 5104.665 
Akaike's Information Criterion (AIC) 5108.665 
Hurvich and Tsai's Criterion (AICC) 5108.671 
Bozdogan's Criterion (CAIC) 5121.845 
Schwarz's Bayesian Criterion (BIC) 5119.845 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 95: Type III tests of fixed effects for Model 21 adding gini * hf interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 880.455 .622 .431 
inczon 1 320.257 1.042 .308 
coccat 2 311.205 3.879 .022 
pop 0 . . . 
bach 1 803.696 2.478 .116 
medinc 1 813.309 .555 .457 
medren 1 923.838 18.948 .000 
medval 1 817.850 10.662 .001 
renfam 1 665.983 .423 .516 
renocc 1 549.408 .096 .756 
renwhi 1 587.121 3.283 .071 
renedu 1 730.259 5.385 .021 
unemp 1 1650.557 1.501 .221 
pov 1 1313.197 .883 .348 
evic 1 1425.128 .147 .701 
burd 1 1729.257 6.877 .009 
vac 1 455.182 2.322 .128 
temp 1 1837.875 2.389 .122 
precip 1 1696.618 1.646 .200 
hf 1 1648.346 1.267 .260 
fund 1 771.272 26.934 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 630.977 .424 .515 
evic2 1 1886.152 .235 .628 
223 
 
hf2 1 1910.449 2.611 .106 
yearcoded 1 1351.521 22.217 .000 
gini * hf 1 1588.479 1.351 .245 
gini * hf2 1 1890.754 2.953 .086 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 96: Fixed effects estimates for Model 21 adding gini * hf interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.233044 3.020365 861.996 .739 .460 -3.695085 8.161174 
[inczon=0] -.206180 .201990 320.257 -1.021 .308 -.603575 .191215 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .705063 .368709 308.550 1.912 .057 -.020440 1.430566 
[coccat=2] -.025791 .329185 317.703 -.078 .938 -.673450 .621867 
[coccat=3] 0b 0 . . . . . 
pop -7.039148E-7 2.069620E-7 309.041 -3.401 .001 -1.111148E-6 -2.966818E-7 
bach -.034615 .021990 803.696 -1.574 .116 -.077779 .008550 
medinc -3.141696E-5 4.217935E-5 813.309 -.745 .457 -.000114 5.137626E-5 
medren .004463 .001025 923.838 4.353 .000 .002451 .006476 
medval 4.159867E-6 1.273953E-6 817.850 3.265 .001 1.659265E-6 6.660470E-6 
renfam .045389 .069767 665.983 .651 .516 -.091602 .182379 
renocc .005571 .017952 549.408 .310 .756 -.029691 .040833 
renwhi .012982 .007165 587.121 1.812 .071 -.001090 .027054 
renedu -.031090 .013398 730.259 -2.321 .021 -.057394 -.004787 
unemp .033006 .026937 1650.557 1.225 .221 -.019827 .085840 
pov .038212 .040666 1313.197 .940 .348 -.041565 .117990 
evic .008736 .022762 1425.128 .384 .701 -.035915 .053388 
burd -.079420 .030286 1729.257 -2.622 .009 -.138821 -.020019 
vac .023472 .015403 455.182 1.524 .128 -.006797 .053742 
temp .004866 .003148 1837.875 1.546 .122 -.001308 .011039 
precip -.010555 .008227 1696.618 -1.283 .200 -.026691 .005581 
hf -8.146859 7.236424 1648.346 -1.126 .260 -22.340411 6.046693 
fund .084935 .016366 771.272 5.190 .000 .052808 .117061 
pop2 7.766000E-14 2.486717E- 307.858 3.123 .002 2.872888E-14 1.265911E-13 
14 
medinc2 -2.869738E- 2.505918E- 627.922 -1.145 .253 -7.790732E- 2.051256E-10 
10 10 10 
renfam2 -.000640 .000983 630.977 -.651 .515 -.002570 .001290 
224 
 
evic2 -.000363 .000749 1886.152 -.485 .628 -.001832 .001106 
hf2 14.264418 8.827970 1910.449 1.616 .106 -3.049054 31.577890 
yearcoded -.135554 .028759 1351.521 -4.713 .000 -.191971 -.079138 
gini * hf 18.546846 15.959503 1588.479 1.162 .245 -12.757057 49.850749 
gini * hf2 -33.330962 19.395476 1890.754 -1.718 .086 -71.369745 4.707822 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
225 
 
Table 97: Information criteria for Model 24 removing evic 
Information Criteriaa 
-2 Restricted Log Likelihood 5102.639 
Akaike's Information Criterion (AIC) 5106.639 
Hurvich and Tsai's Criterion (AICC) 5106.645 
Bozdogan's Criterion (CAIC) 5119.822 
Schwarz's Bayesian Criterion (BIC) 5117.822 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 98: Type III tests of fixed effects for Model 24 removing evic 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 853.017 .523 .470 
inczon 1 321.528 1.052 .306 
coccat 2 309.352 4.161 .016 
pop 0 . . . 
bach 1 637.159 3.345 .068 
medinc 1 810.103 .542 .462 
medren 1 920.021 18.501 .000 
medval 1 755.674 11.321 .001 
renfam 1 653.185 .589 .443 
renocc 1 547.264 .151 .698 
renwhi 1 493.350 4.197 .041 
renedu 1 634.575 5.995 .015 
unemp 1 1572.607 1.510 .219 
pov 1 1266.642 .808 .369 
burd 1 1734.324 6.673 .010 
vac 1 435.264 2.661 .104 
temp 1 1856.445 2.496 .114 
precip 1 1697.344 1.831 .176 
hf 1 1980.516 .294 .588 
fund 1 755.492 25.088 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 610.599 .577 .448 
hf2 1 1969.468 1.144 .285 
yearcoded 1 1316.086 24.955 .000 
226 
 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 99: Fixed effects estimates for Model 24 removing evic 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.977640 2.979387 833.433 .664 .507 -3.870343 7.825623 
[inczon=0] -.206028 .200884 321.528 -1.026 .306 -.601241 .189186 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .759296 .365347 304.481 2.078 .039 .040371 1.478220 
[coccat=2] .013780 .326140 314.778 .042 .966 -.627909 .655469 
[coccat=3] 0b 0 . . . . . 
pop -7.132101E-7 2.049724E-7 306.642 -3.480 .001 -1.116541E-6 -3.098796E-7 
bach -.035996 .019682 637.159 -1.829 .068 -.074646 .002654 
medinc -3.084698E-5 4.189878E-5 810.103 -.736 .462 -.000113 5.139600E-5 
medren .004401 .001023 920.021 4.301 .000 .002393 .006409 
medval 4.153871E-6 1.234579E-6 755.674 3.365 .001 1.730259E-6 6.577484E-6 
renfam .052953 .068977 653.185 .768 .443 -.082491 .188397 
renocc .006946 .017872 547.264 .389 .698 -.028161 .042052 
renwhi .013451 .006565 493.350 2.049 .041 .000551 .026351 
renedu -.031492 .012862 634.575 -2.448 .015 -.056749 -.006235 
unemp .032652 .026568 1572.607 1.229 .219 -.019460 .084764 
pov .035803 .039837 1266.642 .899 .369 -.042350 .113956 
burd -.078047 .030213 1734.324 -2.583 .010 -.137304 -.018790 
vac .024573 .015063 435.264 1.631 .104 -.005032 .054178 
temp .004950 .003133 1856.445 1.580 .114 -.001194 .011095 
precip -.011129 .008224 1697.344 -1.353 .176 -.027258 .005001 
hf .556283 1.026406 1980.516 .542 .588 -1.456666 2.569232 
fund .080959 .016163 755.492 5.009 .000 .049229 .112689 
pop2 7.852766E-14 2.46744E-14 306.159 3.183 .002 2.997477E-14 1.270806E-13 
medinc2 -2.85862E-10 2.49711E-10 631.231 -1.145 .253 -7.76229E-10 2.045035E-10 
renfam2 -.000734 .000967 610.599 -.760 .448 -.002632 .001164 
hf2 -1.019180 .952890 1969.468 -1.070 .285 -2.887958 .849598 
yearcoded -.137352 .027495 1316.086 -4.996 .000 -.191291 -.083413 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
227 
 
Table 100: Information criteria for Model 45 removing renocc 
Information Criteriaa 
-2 Restricted Log Likelihood 6545.698 
Akaike's Information Criterion (AIC) 6549.698 
Hurvich and Tsai's Criterion (AICC) 6549.702 
Bozdogan's Criterion (CAIC) 6563.439 
Schwarz's Bayesian Criterion (BIC) 6561.439 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 101: Type III tests of fixed effects for Model 45 removing renocc 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 1116.982 3.506 .061 
inczon 1 376.766 2.557 .111 
coccat 2 367.762 4.937 .008 
pop 0 . . . 
bach 1 823.679 11.374 .001 
medinc 1 1124.670 .143 .705 
medren 1 1071.900 13.835 .000 
medval 1 899.543 40.561 .000 
renfam 1 784.191 .000 .988 
renwhi 1 585.361 .968 .326 
renedu 1 847.895 14.862 .000 
unemp 1 2023.764 .393 .531 
pov 1 1408.741 1.713 .191 
burd 1 2241.307 4.442 .035 
vac 1 490.377 2.074 .151 
temp 1 2525.336 3.607 .058 
precip 1 2315.761 .131 .717 
hf 1 2605.326 3.382 .066 
fund 1 903.709 18.390 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 732.033 .035 .851 
hf2 1 2602.020 5.501 .019 
yearcoded 1 1515.123 23.509 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
228 
 
Table 102: Fixed effects estimates for Model 45 removing renocc 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 4.170769 2.308064 1083.461 1.807 .071 -.358013 8.699550 
[inczon=0] -.282972 .176973 376.766 -1.599 .111 -.630951 .065007 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .748136 .344982 360.639 2.169 .031 .069708 1.426565 
[coccat=2] -.017133 .310948 374.213 -.055 .956 -.628558 .594291 
[coccat=3] 0b 0 . . . . . 
pop -7.472261E-7 1.886218E-7 362.596 -3.962 .000 -1.118156E-6 -3.762961E-7 
bach -.056484 .016748 823.679 -3.373 .001 -.089358 -.023610 
medinc -1.303725E-5 3.441816E-5 1124.670 -.379 .705 -8.056828E-5 5.449377E-5 
medren .002994 .000805 1071.900 3.720 .000 .001415 .004573 
medval 6.373217E-6 1.000697E-6 899.543 6.369 .000 4.409243E-6 8.337190E-6 
renfam -.000866 .059636 784.191 -.015 .988 -.117932 .116199 
renwhi .005312 .005399 585.361 .984 .326 -.005293 .015917 
renedu -.043351 .011245 847.895 -3.855 .000 -.065422 -.021280 
unemp .014400 .022959 2023.764 .627 .531 -.030625 .059425 
pov .041000 .031321 1408.741 1.309 .191 -.020442 .102441 
burd -.050741 .024076 2241.307 -2.108 .035 -.097955 -.003527 
vac .016709 .011604 490.377 1.440 .151 -.006090 .039508 
temp .005179 .002727 2525.336 1.899 .058 -.000168 .010526 
precip -.002402 .006631 2315.761 -.362 .717 -.015405 .010600 
hf 1.572970 .855274 2605.326 1.839 .066 -.104115 3.250054 
fund .057326 .013368 903.709 4.288 .000 .031090 .083562 
pop2 8.146299E-14 2.28432E-14 360.985 3.566 .000 3.654052E-14 1.263855E-13 
medinc2 -3.57139E-10 2.02525E-10 872.808 -1.763 .078 -7.54633E-10 4.035510E-11 
renfam2 -.000158 .000844 732.033 -.187 .851 -.001815 .001498 
hf2 -1.867677 .796329 2602.020 -2.345 .019 -3.429179 -.306175 
yearcoded -.114866 .023690 1515.123 -4.849 .000 -.161335 -.068397 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
229 
 
Table 103: Information criteria for Model 50 adding renocc * medinc interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6615.417 
Akaike's Information Criterion (AIC) 6619.417 
Hurvich and Tsai's Criterion (AICC) 6619.422 
Bozdogan's Criterion (CAIC) 6633.157 
Schwarz's Bayesian Criterion (BIC) 6631.157 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 104: Type III tests of fixed effects for Model 50 adding renocc * medinc 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 1107.256 3.724 .054 
inczon 1 372.980 1.588 .208 
coccat 2 364.076 4.253 .015 
pop 0 . . . 
bach 1 788.516 11.224 .001 
medinc 1 1032.278 2.456 .117 
medren 1 1130.052 14.112 .000 
medval 1 968.391 43.452 .000 
renfam 1 796.033 .065 .799 
renwhi 1 567.637 2.235 .135 
renedu 1 833.537 12.672 .000 
unemp 1 2014.625 .613 .434 
pov 1 1521.812 .838 .360 
burd 1 2508.877 5.318 .021 
vac 1 495.429 2.751 .098 
temp 1 2513.022 3.504 .061 
precip 1 2319.971 .079 .778 
hf 1 2606.167 3.215 .073 
fund 1 856.531 19.972 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 732.207 .003 .955 
hf2 1 2602.516 5.269 .022 
yearcoded 1 1461.397 22.625 .000 
230 
 
medinc * renocc 0 . . . 
renocc * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 105: Fixed effects estimates for Model 50 adding renocc & medinc 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 4.347970 2.339241 1072.257 1.859 .063 -.242039 8.937979 
[inczon=0] -.222939 .176920 372.980 -1.260 .208 -.570826 .124947 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .695526 .341734 359.316 2.035 .043 .023475 1.367577 
[coccat=2] -.013667 .307455 371.273 -.044 .965 -.618240 .590905 
[coccat=3] 0b 0 . . . . . 
pop -6.778096E-7 1.889143E-7 362.585 -3.588 .000 -1.049315E-6 -3.063043E-7 
bach -.056158 .016763 788.516 -3.350 .001 -.089063 -.023254 
medinc -6.107715E-5 3.897385E-5 1032.278 -1.567 .117 -.000138 1.539986E-5 
medren .003201 .000852 1130.052 3.757 .000 .001529 .004873 
medval 6.798069E-6 1.031295E-6 968.391 6.592 .000 4.774238E-6 8.821900E-6 
renfam -.015415 .060435 796.033 -.255 .799 -.134046 .103216 
renwhi .008496 .005682 567.637 1.495 .135 -.002665 .019657 
renedu -.040141 .011276 833.537 -3.560 .000 -.062273 -.018008 
unemp .017937 .022902 2014.625 .783 .434 -.026977 .062852 
pov .029919 .032681 1521.812 .915 .360 -.034185 .094023 
burd -.059020 .025593 2508.877 -2.306 .021 -.109206 -.008834 
vac .020310 .012245 495.429 1.659 .098 -.003748 .044368 
temp .005115 .002733 2513.022 1.872 .061 -.000243 .010474 
precip -.001870 .006639 2319.971 -.282 .778 -.014889 .011149 
hf 1.532378 .854685 2606.167 1.793 .073 -.143551 3.208308 
fund .060275 .013487 856.531 4.469 .000 .033803 .086747 
pop2 7.17998E-14 2.29997E-14 362.062 3.122 .002 2.65699E-14 1.17029E-13 
medinc2 2.49045E-10 3.02010E-10 724.537 .825 .410 -3.43874E-10 8.41965E-10 
renfam2 4.816458E-5 .000851 732.207 .057 .955 -.001622 .001718 
hf2 -1.827069 .795995 2602.516 -2.295 .022 -3.387916 -.266222 
yearcoded -.112494 .023650 1461.397 -4.757 .000 -.158886 -.066102 
medinc * renocc 1.298502E-6 5.782189E-7 631.404 2.246 .025 1.630369E-7 2.433967E-6 
renocc*medinc2 -1.87060E-11 6.60921E-12 642.368 -2.830 .005 -3.16849E-11 -5.72841E-12 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
231 
 
b. This parameter is set to zero because it is redundant. 
Table 106: Information criteria for Model 62 adding renocc * hf interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6540.710 
Akaike's Information Criterion (AIC) 6544.710 
Hurvich and Tsai's Criterion (AICC) 6544.715 
Bozdogan's Criterion (CAIC) 6558.450 
Schwarz's Bayesian Criterion (BIC) 6556.450 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 107: Type III tests of fixed effects for Model 62 adding renocc * hf 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 1101.999 3.801 .051 
inczon 1 375.569 2.650 .104 
coccat 2 366.234 5.233 .006 
pop 0 . . . 
bach 1 800.208 10.592 .001 
medinc 1 1108.939 .246 .620 
medren 1 1173.583 15.078 .000 
medval 1 914.049 40.452 .000 
renfam 1 790.692 .016 .900 
renwhi 1 592.356 .855 .355 
renedu 1 845.962 14.933 .000 
unemp 1 2013.547 .436 .509 
pov 1 1578.951 2.033 .154 
burd 1 2433.495 4.557 .033 
vac 1 513.495 .924 .337 
temp 1 2512.942 3.498 .062 
precip 1 2316.159 .085 .771 
hf 1 1513.141 .527 .468 
fund 1 952.560 23.076 .000 
pop2 0 . . . 
medinc2 0 . . . 
renfam2 1 731.044 .011 .915 
hf2 1 1986.031 2.887 .089 
232 
 
yearcoded 1 1480.980 22.378 .000 
renocc * hf 1 1116.453 2.154 .143 
renocc * hf2 1 1665.789 6.792 .009 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 108: Fixed effects estimates for Model 62 adding renocc * hf interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 4.397914 2.324846 1068.700 1.892 .059 -.163866 8.959695 
[inczon=0] -.287563 .176639 375.569 -1.628 .104 -.634887 .059762 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .738453 .343283 360.550 2.151 .032 .063365 1.413541 
[coccat=2] -.059382 .309286 373.268 -.192 .848 -.667544 .548779 
[coccat=3] 0b 0 . . . . . 
pop -7.282665E-7 1.888141E-7 363.390 -3.857 .000 -1.099572E-6 -3.569610E-7 
bach -.054579 .016770 800.208 -3.255 .001 -.087498 -.021660 
medinc -1.699864E-5 3.428388E-5 1108.939 -.496 .620 -8.426724E-5 5.026996E-5 
medren .003269 .000842 1173.583 3.883 .000 .001617 .004921 
medval 6.427512E-6 1.010589E-6 914.049 6.360 .000 4.444169E-6 8.410856E-6 
renfam -.007519 .059896 790.692 -.126 .900 -.125093 .110056 
renwhi .005193 .005615 592.356 .925 .355 -.005834 .016220 
renedu -.043553 .011271 845.962 -3.864 .000 -.065675 -.021431 
unemp .015114 .022881 2013.547 .661 .509 -.029760 .059987 
pov .046906 .032900 1578.951 1.426 .154 -.017626 .111439 
burd -.053232 .024937 2433.495 -2.135 .033 -.102132 -.004332 
vac .011736 .012210 513.495 .961 .337 -.012252 .035724 
temp .005101 .002727 2512.942 1.870 .062 -.000247 .010449 
precip -.001929 .006624 2316.159 -.291 .771 -.014919 .011062 
hf -1.478815 2.036926 1513.141 -.726 .468 -5.474313 2.516682 
fund .065478 .013631 952.560 4.804 .000 .038729 .092228 
pop2 7.857862E-14 2.28956E-14 361.899 3.432 .001 3.355336E-14 1.236039E-13 
medinc2 -3.62595E-10 2.04594E-10 924.921 -1.772 .077 -7.64117E-10 3.892711E-11 
renfam2 -8.983819E-5 .000844 731.044 -.106 .915 -.001747 .001567 
hf2 3.847674 2.264345 1986.031 1.699 .089 -.593066 8.288415 
yearcoded -.111763 .023626 1480.980 -4.731 .000 -.158106 -.065419 
renocc * hf .085694 .058392 1116.453 1.468 .143 -.028876 .200263 
renocc * hf2 -.171087 .065648 1665.789 -2.606 .009 -.299848 -.042327 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
233 
 
b. This parameter is set to zero because it is redundant. 
  
234 
 
Table 109: Information criteria for Model 65 removing renfam 
Information Criteriaa 
-2 Restricted Log Likelihood 6522.660 
Akaike's Information Criterion (AIC) 6526.660 
Hurvich and Tsai's Criterion (AICC) 6526.665 
Bozdogan's Criterion (CAIC) 6540.401 
Schwarz's Bayesian Criterion (BIC) 6538.401 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 110: Type III tests of fixed effects for Model 65 removing renfam 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 1043.836 4.477 .035 
inczon 1 376.549 2.684 .102 
coccat 2 366.746 5.371 .005 
pop 0 . . . 
bach 1 663.866 10.033 .002 
medinc 1 1034.146 .382 .537 
medren 1 1168.096 14.313 .000 
medval 1 918.117 41.087 .000 
renwhi 1 528.889 1.938 .164 
renedu 1 843.079 14.807 .000 
unemp 1 2008.031 .310 .578 
pov 1 1407.189 1.625 .203 
burd 1 2400.480 4.161 .041 
vac 1 509.998 1.239 .266 
temp 1 2517.001 3.455 .063 
precip 1 2317.289 .070 .792 
hf 1 1501.363 .644 .422 
fund 1 853.135 27.610 .000 
pop2 0 . . . 
medinc2 0 . . . 
hf2 1 1983.460 3.072 .080 
yearcoded 1 1426.635 21.905 .000 
renocc * hf 1 1098.124 2.498 .114 
renocc * hf2 1 1661.268 7.203 .007 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
235 
 
Table 111: Fixed effects estimates for Model 65 removing renfam 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.932061 1.937938 1000.043 2.029 .043 .129169 7.734952 
[inczon=0] -.288589 .176155 376.549 -1.638 .102 -.634959 .057781 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .753021 .341144 361.877 2.207 .028 .082147 1.423895 
[coccat=2] -.048626 .308162 374.930 -.158 .875 -.654568 .557317 
[coccat=3] 0b 0 . . . . . 
pop -7.315078E-7 1.879995E-7 365.136 -3.891 .000 -1.101205E-6 -3.618102E-7 
bach -.046945 .014821 663.866 -3.167 .002 -.076047 -.017843 
medinc -2.094065E-5 3.387484E-5 1034.146 -.618 .537 -8.741191E-5 4.553062E-5 
medren .003134 .000828 1168.096 3.783 .000 .001509 .004759 
medval 6.463353E-6 1.008342E-6 918.117 6.410 .000 4.484430E-6 8.442276E-6 
renwhi .007237 .005198 528.889 1.392 .164 -.002976 .017449 
renedu -.043272 .011246 843.079 -3.848 .000 -.065345 -.021200 
unemp .012621 .022678 2008.031 .557 .578 -.031853 .057096 
pov .040768 .031986 1407.189 1.275 .203 -.021977 .103514 
burd -.050545 .024779 2400.480 -2.040 .041 -.099135 -.001955 
vac .013417 .012053 509.998 1.113 .266 -.010263 .037097 
temp .005069 .002727 2517.001 1.859 .063 -.000278 .010416 
precip -.001748 .006622 2317.289 -.264 .792 -.014734 .011237 
hf -1.622359 2.021155 1501.363 -.803 .422 -5.586945 2.342227 
fund .069012 .013134 853.135 5.254 .000 .043233 .094790 
pop2 7.920646E-14 2.28176E-14 363.138 3.471 .001 3.433506E-14 1.240779E-13 
medinc2 -3.53577E-10 2.03701E-10 902.835 -1.736 .083 -7.53360E-10 4.620516E-11 
hf2 3.952964 2.255433 1983.460 1.753 .080 -.470302 8.376230 
yearcoded -.108670 .023219 1426.635 -4.680 .000 -.154216 -.063123 
renocc * hf .091228 .057725 1098.124 1.580 .114 -.022035 .204492 
renocc * hf2 -.175215 .065284 1661.268 -2.684 .007 -.303263 -.047167 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
236 
 
Table 112: Information criteria for Model 84 removing precip 
Information Criteriaa 
-2 Restricted Log Likelihood 6514.533 
Akaike's Information Criterion (AIC) 6518.533 
Hurvich and Tsai's Criterion (AICC) 6518.537 
Bozdogan's Criterion (CAIC) 6532.275 
Schwarz's Bayesian Criterion (BIC) 6530.275 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 113: Type III tests of fixed effects for Model 84 removing precip 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 1046.570 4.420 .036 
inczon 1 377.082 2.697 .101 
coccat 2 367.063 5.378 .005 
pop 0 . . . 
bach 1 663.393 9.967 .002 
medinc 1 1037.884 .364 .546 
medren 1 1166.278 14.431 .000 
medval 0 . . . 
renwhi 1 528.840 1.965 .162 
renedu 1 842.239 14.746 .000 
unemp 1 1999.449 .336 .562 
pov 1 1408.426 1.633 .202 
burd 1 2400.444 4.188 .041 
vac 1 510.360 1.263 .262 
temp 1 2521.466 3.391 .066 
hf 1 1503.083 .674 .412 
fund 1 853.672 27.617 .000 
pop2 0 . . . 
medinc2 0 . . . 
hf2 1 1985.683 3.136 .077 
yearcoded 1 1428.148 22.247 .000 
renocc * hf 1 1099.333 2.536 .112 
renocc * hf2 1 1662.538 7.271 .007 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
237 
 
Table 114: Fixed effects estimates for Model 84 removing precip 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.895140 1.932368 1002.210 2.016 .044 .103190 7.687091 
[inczon=0] -.289186 .176083 377.082 -1.642 .101 -.635414 .057043 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .752877 .341033 362.270 2.208 .028 .082224 1.423531 
[coccat=2] -.049100 .308058 375.264 -.159 .873 -.654837 .556637 
[coccat=3] 0b 0 . . . . . 
pop -7.308587E-7 1.879231E-7 365.538 -3.889 .000 -1.100405E-6 -3.613126E-7 
bach -.046625 .014768 663.393 -3.157 .002 -.075622 -.017627 
medinc -2.040573E-5 3.380304E-5 1037.884 -.604 .546 -8.673581E-5 4.592436E-5 
medren .003143 .000827 1166.278 3.799 .000 .001520 .004767 
medval 6.413091E-6 9.899162E-7 913.255 6.478 .000 4.470316E-6 8.355866E-6 
renwhi .007282 .005194 528.840 1.402 .162 -.002922 .017486 
renedu -.043096 .011223 842.239 -3.840 .000 -.065123 -.021068 
unemp .013107 .022599 1999.449 .580 .562 -.031214 .057427 
pov .040859 .031977 1408.426 1.278 .202 -.021868 .103586 
burd -.050685 .024768 2400.444 -2.046 .041 -.099254 -.002116 
vac .013531 .012041 510.360 1.124 .262 -.010125 .037187 
temp .004987 .002708 2521.466 1.841 .066 -.000324 .010298 
hf -1.655346 2.016741 1503.083 -.821 .412 -5.611272 2.300580 
fund .069002 .013130 853.672 5.255 .000 .043231 .094774 
pop2 7.919169E-14 2.28102E-14 363.530 3.472 .001 3.433509E-14 1.240483E-13 
medinc2 -3.56670E-10 2.03301E-10 906.035 -1.754 .080 -7.55666E-10 4.232626E-11 
hf2 3.986779 2.251261 1985.683 1.771 .077 -.428301 8.401860 
yearcoded -.109150 .023141 1428.148 -4.717 .000 -.154545 -.063755 
renocc * hf .091831 .057664 1099.333 1.592 .112 -.021314 .204975 
renocc * hf2 -.175872 .065222 1662.538 -2.697 .007 -.303797 -.047946 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
238 
 
Table 115: Information criteria for Model 102 removing unemp 
Information Criteriaa 
-2 Restricted Log Likelihood 6542.904 
Akaike's Information Criterion (AIC) 6546.904 
Hurvich and Tsai's Criterion (AICC) 6546.909 
Bozdogan's Criterion (CAIC) 6560.653 
Schwarz's Bayesian Criterion (BIC) 6558.653 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 116: Type III tests of fixed effects for Model 102 removing unemp 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 1075.410 4.050 .044 
inczon 1 377.119 2.788 .096 
coccat 2 366.765 5.443 .005 
pop 0 . . . 
bach 1 627.907 11.924 .001 
medinc 1 1041.223 .198 .656 
medren 1 1245.362 15.785 .000 
medval 0 . . . 
renwhi 1 538.048 2.250 .134 
renedu 1 843.573 16.194 .000 
pov 1 1620.216 3.076 .080 
burd 1 2386.811 4.255 .039 
vac 1 509.455 1.379 .241 
temp 1 2528.979 3.267 .071 
hf 1 1490.177 .583 .445 
fund 1 850.180 27.284 .000 
pop2 0 . . . 
medinc2 0 . . . 
hf2 1 1974.894 2.971 .085 
yearcoded 1 1766.658 31.280 .000 
renocc * hf 1 1091.990 2.399 .122 
renocc * hf2 1 1655.565 7.088 .008 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
239 
 
Table 117: Fixed effects estimates for Model 102 removing unemp 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.675016 1.914129 1027.374 1.920 .055 -.081033 7.431065 
[inczon=0] -.293252 .175614 377.119 -1.670 .096 -.638558 .052053 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .774246 .340232 362.372 2.276 .023 .105169 1.443323 
[coccat=2] -.025563 .307082 374.706 -.083 .934 -.629382 .578256 
[coccat=3] 0b 0 . . . . . 
pop -7.210015E-7 1.874189E-7 365.292 -3.847 .000 -1.089557E-6 -3.524461E-7 
bach -.049622 .014370 627.907 -3.453 .001 -.077842 -.021402 
medinc -1.489229E-5 3.342889E-5 1041.223 -.445 .656 -8.048797E-5 5.070338E-5 
medren .003228 .000813 1245.362 3.973 .000 .001634 .004822 
medval 6.295051E-6 9.498994E-7 1039.363 6.627 .000 4.431112E-6 8.158990E-6 
renwhi .007770 .005180 538.048 1.500 .134 -.002406 .017945 
renedu -.044871 .011150 843.573 -4.024 .000 -.066757 -.022985 
pov .052340 .029844 1620.216 1.754 .080 -.006196 .110877 
burd -.050422 .024443 2386.811 -2.063 .039 -.098354 -.002489 
vac .014113 .012020 509.455 1.174 .241 -.009501 .037727 
temp .004910 .002716 2528.979 1.808 .071 -.000416 .010236 
hf -1.540210 2.017216 1490.177 -.764 .445 -5.497095 2.416675 
fund .068400 .013095 850.180 5.223 .000 .042698 .094102 
pop2 7.823612E-14 2.27541E-14 363.300 3.438 .001 3.348971E-14 1.229825E-13 
medinc2 -3.78383E-10 2.02109E-10 905.919 -1.872 .062 -7.75039E-10 1.827364E-11 
hf2 3.885492 2.254086 1974.894 1.724 .085 -.535145 8.306130 
yearcoded -.116595 .020847 1766.658 -5.593 .000 -.157483 -.075708 
renocc * hf .089347 .057685 1091.990 1.549 .122 -.023838 .202533 
renocc * hf2 -.173894 .065315 1655.565 -2.662 .008 -.302003 -.045784 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
240 
 
Table 118: Information criteria for Model 119 removing medinc 
Information Criteriaa 
-2 Restricted Log Likelihood 6504.198 
Akaike's Information Criterion (AIC) 6508.198 
Hurvich and Tsai's Criterion (AICC) 6508.203 
Bozdogan's Criterion (CAIC) 6521.948 
Schwarz's Bayesian Criterion (BIC) 6519.948 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 119: Type III tests of fixed effects for Model 119 removing medinc 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 984.043 7.064 .008 
inczon 1 374.362 4.177 .042 
coccat 2 362.226 6.038 .003 
pop 0 . . . 
bach 1 566.920 38.784 .000 
medren 1 1291.687 1.050 .306 
medval 0 . . . 
renwhi 1 517.055 3.118 .078 
renedu 1 837.137 23.988 .000 
pov 1 1184.668 7.434 .006 
burd 1 2165.696 .298 .585 
vac 1 442.241 9.551 .002 
temp 1 2579.779 5.335 .021 
hf 1 1532.747 6.221 .013 
fund 1 868.915 22.923 .000 
pop2 0 . . . 
hf2 1 2039.367 8.755 .003 
yearcoded 1 1755.105 42.082 .000 
renocc * hf 1 1049.049 13.638 .000 
renocc * hf2 1 1684.436 16.971 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
241 
 
Table 120: Fixed effects estimates for Model 119 removing medinc 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.352458 1.375845 894.532 2.437 .015 .652199 6.052718 
[inczon=0] -.367112 .179617 374.362 -2.044 .042 -.720298 -.013927 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .963730 .348353 358.962 2.767 .006 .278660 1.648800 
[coccat=2] .141471 .314161 369.605 .450 .653 -.476296 .759237 
[coccat=3] 0b 0 . . . . . 
pop -6.079172E-7 1.914875E-7 356.864 -3.175 .002 -9.845029E-7 -2.313315E-7 
bach -.081513 .013089 566.920 -6.228 .000 -.107221 -.055804 
medren .000642 .000626 1291.687 1.025 .306 -.000587 .001871 
medval 5.871492E-6 9.593658E-7 1055.730 6.120 .000 3.989011E-6 7.753972E-6 
renwhi .009276 .005253 517.055 1.766 .078 -.001044 .019597 
renedu -.054717 .011172 837.137 -4.898 .000 -.076645 -.032789 
pov .066609 .024430 1184.668 2.727 .006 .018678 .114540 
burd -.012724 .023291 2165.696 -.546 .585 -.058399 .032952 
vac .035065 .011346 442.241 3.090 .002 .012766 .057365 
temp .006232 .002698 2579.779 2.310 .021 .000941 .011523 
hf -4.823383 1.933914 1532.747 -2.494 .013 -8.616781 -1.029986 
fund .063659 .013296 868.915 4.788 .000 .037563 .089756 
pop2 6.86214E-14 2.33380E-14 356.526 2.940 .003 2.272397E-14 1.145189E-13 
hf2 6.546338 2.212491 2039.367 2.959 .003 2.207360 10.885316 
yearcoded -.125518 .019349 1755.105 -6.487 .000 -.163467 -.087568 
renocc * hf .199988 .054154 1049.049 3.693 .000 .093726 .306250 
renocc * hf2 -.261987 .063596 1684.436 -4.120 .000 -.386723 -.137252 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
242 
 
Table 121: Information criteria for Model 122 adding pop * medinc interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6791.405 
Akaike's Information Criterion (AIC) 6795.405 
Hurvich and Tsai's Criterion (AICC) 6795.410 
Bozdogan's Criterion (CAIC) 6809.152 
Schwarz's Bayesian Criterion (BIC) 6807.152 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 122: Type III tests of fixed effects for Model 122 adding pop * medinc 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 956.595 4.356 .037 
inczon 1 374.011 3.046 .082 
coccat 2 372.415 5.626 .004 
pop 1 763.971 5.214 .023 
bach 1 585.131 30.601 .000 
medren 1 1239.719 4.293 .038 
medval 0 . . . 
renwhi 1 517.278 3.147 .077 
renedu 1 838.192 20.611 .000 
pov 1 1417.453 11.817 .001 
burd 1 2343.625 1.578 .209 
vac 1 444.798 7.370 .007 
temp 1 2567.800 4.866 .027 
hf 1 1590.193 2.997 .084 
fund 1 853.250 26.842 .000 
pop2 0 . . . 
hf2 1 2059.937 5.680 .017 
yearcoded 1 1730.465 46.356 .000 
renocc * hf 1 1151.451 7.208 .007 
renocc * hf2 1 1737.925 11.465 .001 
pop * medinc 0 . . . 
pop * medinc2 0 . . . 
medinc * pop2 0 . . . 
pop2 * medinc2 0 . . . 
243 
 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 123: Fixed effects estimates for Model 122 adding pop * medinc interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.635092 1.389876 864.085 1.896 .058 -.092836 5.363019 
[inczon=0] -.313140 .179429 374.011 -1.745 .082 -.665956 .039675 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .903681 .355281 373.689 2.544 .011 .205080 1.602281 
[coccat=2] .093858 .321120 382.705 .292 .770 -.537522 .725239 
[coccat=3] 0b 0 . . . . . 
pop -3.960482E-6 1.734423E-6 763.971 -2.283 .023 -7.365282E-6 -5.556819E-7 
bach -.073449 .013277 585.131 -5.532 .000 -.099526 -.047371 
medren .001398 .000675 1239.719 2.072 .038 7.430328E-5 .002721 
medval 6.038826E-6 9.713118E-7 1070.628 6.217 .000 4.132935E-6 7.944717E-6 
renwhi .009290 .005237 517.278 1.774 .077 -.000999 .019578 
renedu -.050875 .011206 838.192 -4.540 .000 -.072871 -.028879 
pov .091739 .026687 1417.453 3.438 .001 .039388 .144090 
burd -.030121 .023979 2343.625 -1.256 .209 -.077143 .016900 
vac .031107 .011458 444.798 2.715 .007 .008588 .053626 
temp .005964 .002704 2567.800 2.206 .027 .000663 .011266 
hf -3.427219 1.979656 1590.193 -1.731 .084 -7.310228 .455791 
fund .069115 .013340 853.250 5.181 .000 .042932 .095299 
pop2 9.95547E-13 6.34477E-13 1065.921 1.569 .117 -2.49418E-13 2.240513E-12 
hf2 5.342444 2.241645 2059.937 2.383 .017 .946318 9.738571 
yearcoded -.135644 .019923 1730.465 -6.809 .000 -.174719 -.096569 
renocc * hf .150959 .056227 1151.451 2.685 .007 .040640 .261279 
renocc * hf2 -.219755 .064902 1737.925 -3.386 .001 -.347049 -.092462 
pop * medinc 1.21219E-10 5.24051E-11 834.079 2.313 .021 1.835833E-11 2.240811E-10 
pop * medinc2 -1.08159E-15 4.07763E-16 922.221 -2.653 .008 -1.88184E-15 -2.81341E-16 
medinc * pop2 -3.40856E-17 2.16264E-17 1123.509 -1.576 .115 -7.65183E-17 8.347157E-18 
pop2 * medinc2 3.09848E-22 1.86616E-22 1195.566 1.660 .097 -5.62836E-23 6.759797E-22 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
244 
 
Table 124: Information criteria for Model 124 adding medinc * medren interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6552.442 
Akaike's Information Criterion (AIC) 6556.442 
Hurvich and Tsai's Criterion (AICC) 6556.446 
Bozdogan's Criterion (CAIC) 6570.190 
Schwarz's Bayesian Criterion (BIC) 6568.190 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 125: Type III tests of fixed effects for Model 124 adding medinc * medren 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 852.263 .610 .435 
inczon 1 377.255 2.096 .149 
coccat 2 367.558 5.079 .007 
pop 0 . . . 
bach 1 605.668 9.990 .002 
medren 1 953.433 31.290 .000 
medval 0 . . . 
renwhi 1 523.517 3.491 .062 
renedu 1 809.718 12.965 .000 
pov 1 1313.767 3.111 .078 
burd 1 2421.152 8.580 .003 
vac 1 494.869 .188 .665 
temp 1 2531.297 2.102 .147 
hf 1 1494.886 .148 .700 
fund 1 823.365 31.778 .000 
pop2 0 . . . 
hf2 1 1959.888 2.090 .148 
yearcoded 1 1730.941 32.432 .000 
renocc * hf 1 1086.965 1.145 .285 
renocc * hf2 1 1636.739 5.349 .021 
medinc * medren 0 . . . 
medren * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
245 
 
Table 126: Fixed effects estimates for Model 124 adding medinc * medren 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .963199 1.367818 799.091 .704 .482 -1.721741 3.648139 
[inczon=0] -.246588 .170344 377.255 -1.448 .149 -.581531 .088355 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .678949 .329980 363.469 2.058 .040 .030039 1.327859 
[coccat=2] -.079344 .297473 374.122 -.267 .790 -.664273 .505584 
[coccat=3] 0b 0 . . . . . 
pop -7.19822E-7 1.810207E-7 364.316 -3.976 .000 -1.075799E-6 -3.638458E-7 
bach -.043229 .013677 605.668 -3.161 .002 -.070090 -.016368 
medren .011423 .002042 953.433 5.594 .000 .007416 .015431 
medval 6.978933E-6 9.457942E-7 1073.163 7.379 .000 5.123117E-6 8.834748E-6 
renwhi .009352 .005006 523.517 1.868 .062 -.000482 .019186 
renedu -.039540 .010981 809.718 -3.601 .000 -.061096 -.017985 
pov .047517 .026938 1313.767 1.764 .078 -.005330 .100363 
burd -.072622 .024792 2421.152 -2.929 .003 -.121238 -.024006 
vac .005067 .011681 494.869 .434 .665 -.017884 .028017 
temp .003944 .002720 2531.297 1.450 .147 -.001390 .009277 
hf -.760853 1.977095 1494.886 -.385 .700 -4.639028 3.117323 
fund .072404 .012844 823.365 5.637 .000 .047193 .097615 
pop2 7.59309E-14 2.19893E-14 362.990 3.453 .001 3.26883E-14 1.19173E-13 
hf2 3.216780 2.225220 1959.888 1.446 .148 -1.147266 7.580826 
yearcoded -.112892 .019823 1730.941 -5.695 .000 -.151772 -.074012 
renocc * hf .060303 .056344 1086.965 1.070 .285 -.050253 .170859 
renocc * hf2 -.148934 .064396 1636.739 -2.313 .021 -.275241 -.022627 
medinc * medren -1.59748E-7 3.886288E-8 975.923 -4.111 .000 -2.360132E-7 -8.348429E-8 
medren*medinc2 5.94686E-13 2.11004E-13 972.317 2.818 .005 1.80608E-13 1.00876E-12 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
246 
 
Table 127: Information criteria for Model 125 adding medinc * medval interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6549.851 
Akaike's Information Criterion (AIC) 6553.851 
Hurvich and Tsai's Criterion (AICC) 6553.856 
Bozdogan's Criterion (CAIC) 6567.600 
Schwarz's Bayesian Criterion (BIC) 6565.600 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 128: Type III tests of fixed effects for Model 125 adding medinc * medval 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 845.920 .340 .560 
inczon 1 382.656 .814 .367 
coccat 2 372.881 5.443 .005 
pop 0 . . . 
bach 1 582.444 13.241 .000 
medren 1 1163.845 23.117 .000 
medval 1 1040.850 55.838 .000 
renwhi 1 524.055 4.473 .035 
renedu 1 787.379 15.048 .000 
pov 1 1313.910 14.065 .000 
burd 1 2362.361 10.352 .001 
vac 1 463.599 1.289 .257 
temp 1 2593.208 2.821 .093 
hf 1 1550.848 .033 .856 
fund 1 796.202 41.629 .000 
pop2 0 . . . 
hf2 1 2001.588 .701 .403 
yearcoded 1 1699.567 35.013 .000 
renocc * hf 1 1130.133 .225 .636 
renocc * hf2 1 1691.249 2.805 .094 
medinc * medval 0 . . . 
medval * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
247 
 
Table 129: Fixed effects estimates for Model 125 adding medinc * medval 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .559136 1.337505 790.112 .418 .676 -2.066348 3.184620 
[inczon=0] -.150460 .166742 382.656 -.902 .367 -.478306 .177387 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .780096 .319618 368.617 2.441 .015 .151592 1.408601 
[coccat=2] .041190 .288395 380.996 .143 .887 -.525855 .608235 
[coccat=3] 0b 0 . . . . . 
pop -5.920855E-7 1.755308E-7 368.958 -3.373 .001 -9.372517E-7 -2.469193E-7 
bach -.046710 .012836 582.444 -3.639 .000 -.071921 -.021499 
medren .003250 .000676 1163.845 4.808 .000 .001923 .004576 
medval 3.697498E-5 4.948150E-6 1040.850 7.472 .000 2.726549E-5 4.668447E-5 
renwhi .010346 .004892 524.055 2.115 .035 .000736 .019956 
renedu -.041361 .010662 787.379 -3.879 .000 -.062291 -.020431 
pov .093487 .024928 1313.910 3.750 .000 .044584 .142390 
burd -.077909 .024215 2362.361 -3.217 .001 -.125394 -.030425 
vac .012258 .010795 463.599 1.136 .257 -.008955 .033471 
temp .004517 .002689 2593.208 1.680 .093 -.000756 .009790 
hf .353554 1.954149 1550.848 .181 .856 -3.479500 4.186608 
fund .081945 .012701 796.202 6.452 .000 .057014 .106876 
pop2 5.57066E-14 2.14325E-14 370.171 2.599 .010 1.35617E-14 9.78514E-14 
hf2 1.854333 2.214878 2001.588 .837 .403 -2.489374 6.198040 
yearcoded -.113607 .019200 1699.567 -5.917 .000 -.151265 -.075950 
renocc * hf .026268 .055418 1130.133 .474 .636 -.082465 .135001 
renocc * hf2 -.107203 .064012 1691.249 -1.675 .094 -.232754 .018349 
medinc * medval -6.24685E-10 1.14401E-10 1000.842 -5.460 .000 -8.49179E-10 -4.00191E-10 
medval*medinc2 2.68310E-15 6.53710E-16 1051.848 4.104 .000 1.40037E-15 3.96583E-15 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
248 
 
Table 130: Information criteria for Model 126 adding medinc * renwhi interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6571.396 
Akaike's Information Criterion (AIC) 6575.396 
Hurvich and Tsai's Criterion (AICC) 6575.401 
Bozdogan's Criterion (CAIC) 6589.145 
Schwarz's Bayesian Criterion (BIC) 6587.145 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 131: Type III tests of fixed effects for Model 126 adding medinc * renwhi 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 969.090 2.670 .103 
inczon 1 376.607 2.635 .105 
coccat 2 365.573 5.742 .004 
pop 0 . . . 
bach 1 614.615 19.980 .000 
medren 1 1313.630 6.918 .009 
medval 0 . . . 
renwhi 1 741.001 .487 .486 
renedu 1 861.431 17.780 .000 
pov 1 1510.941 9.828 .002 
burd 1 2354.652 2.070 .150 
vac 1 472.781 5.078 .025 
temp 1 2569.130 4.810 .028 
hf 1 1523.079 2.043 .153 
fund 1 861.325 25.482 .000 
pop2 0 . . . 
hf2 1 2024.881 4.877 .027 
yearcoded 1 1718.301 41.885 .000 
renocc * hf 1 1105.008 5.411 .020 
renocc * hf2 1 1697.695 10.313 .001 
medinc * renwhi 0 . . . 
renwhi * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
249 
 
Table 132: Fixed effects estimates for Model 126 adding medinc * renwhi 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.079370 1.412168 892.002 1.472 .141 -.692189 4.850929 
[inczon=0] -.290950 .179223 376.607 -1.623 .105 -.643354 .061454 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .868570 .346278 360.120 2.508 .013 .187588 1.549551 
[coccat=2] .047284 .313683 374.683 .151 .880 -.569515 .664084 
[coccat=3] 0b 0 . . . . . 
pop -7.001548E-7 1.920340E-7 365.089 -3.646 .000 -1.077786E-6 -3.225233E-7 
bach -.063375 .014178 614.615 -4.470 .000 -.091219 -.035532 
medren .001916 .000729 1313.630 2.630 .009 .000487 .003346 
medval 6.015638E-6 9.568300E-7 1039.533 6.287 .000 4.138099E-6 7.893176E-6 
renwhi -.010800 .015483 741.001 -.698 .486 -.041195 .019595 
renedu -.047566 .011280 861.431 -4.217 .000 -.069707 -.025426 
pov .087397 .027879 1510.941 3.135 .002 .032712 .142082 
burd -.034572 .024029 2354.652 -1.439 .150 -.081693 .012549 
vac .026381 .011707 472.781 2.253 .025 .003376 .049386 
temp .005924 .002701 2569.130 2.193 .028 .000627 .011220 
hf -2.864622 2.004345 1523.079 -1.429 .153 -6.796191 1.066946 
fund .066820 .013237 861.325 5.048 .000 .040840 .092801 
pop2 7.68318E-14 2.32800E-14 362.436 3.300 .001 3.10508E-14 1.22612E-13 
hf2 4.964873 2.248074 2024.881 2.209 .027 .556094 9.373653 
yearcoded -.132690 .020503 1718.301 -6.472 .000 -.172902 -.092477 
renocc * hf .132868 .057122 1105.008 2.326 .020 .020789 .244947 
renocc * hf2 -.208869 .065040 1697.695 -3.211 .001 -.336437 -.081301 
medinc * renwhi 9.675904E-7 4.537043E-7 822.674 2.133 .033 7.703619E-8 1.858145E-6 
renwhi*medinc2 -1.05562E-11 3.39710E-12 820.421 -3.107 .002 -1.72243E-11 -3.88824E-12 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
250 
 
Table 133: Information criteria for Model 127 adding medinc * renedu interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6571.481 
Akaike's Information Criterion (AIC) 6575.481 
Hurvich and Tsai's Criterion (AICC) 6575.486 
Bozdogan's Criterion (CAIC) 6589.230 
Schwarz's Bayesian Criterion (BIC) 6587.230 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 134: Type III tests of fixed effects for Model 127 adding medinc * renedu 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 958.768 3.117 .078 
inczon 1 375.296 3.694 .055 
coccat 2 364.745 5.690 .004 
pop 0 . . . 
bach 1 608.630 21.727 .000 
medren 1 1249.013 6.917 .009 
medval 0 . . . 
renwhi 1 531.923 2.280 .132 
renedu 1 946.531 5.962 .015 
pov 1 1710.192 4.868 .027 
burd 1 2353.514 1.113 .292 
vac 1 488.117 3.290 .070 
temp 1 2551.307 4.425 .036 
hf 1 1544.773 1.859 .173 
fund 1 863.141 24.525 .000 
pop2 0 . . . 
hf2 1 2025.009 4.567 .033 
yearcoded 1 1780.903 34.527 .000 
renocc * hf 1 1142.548 5.081 .024 
renocc * hf2 1 1712.275 9.795 .002 
medinc * renedu 0 . . . 
renedu * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
251 
 
Table 135: Fixed effects estimates for Model 127 adding medinc * renedu 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.292077 1.410453 883.738 1.625 .105 -.476151 5.060306 
[inczon=0] -.343699 .178819 375.296 -1.922 .055 -.695312 .007915 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .857618 .346098 360.597 2.478 .014 .176995 1.538242 
[coccat=2] .039736 .312431 373.482 .127 .899 -.574608 .654080 
[coccat=3] 0b 0 . . . . . 
pop -6.963734E-7 1.913839E-7 364.388 -3.639 .000 -1.072729E-6 -3.200178E-7 
bach -.064856 .013914 608.630 -4.661 .000 -.092181 -.037531 
medren .001931 .000734 1249.013 2.630 .009 .000491 .003372 
medval 5.994227E-6 9.547048E-7 1044.166 6.279 .000 4.120869E-6 7.867586E-6 
renwhi .007941 .005259 531.923 1.510 .132 -.002391 .018272 
renedu -.060088 .024608 946.531 -2.442 .015 -.108381 -.011795 
pov .064413 .029195 1710.192 2.206 .027 .007151 .121674 
burd -.025173 .023863 2353.514 -1.055 .292 -.071967 .021622 
vac .021806 .012022 488.117 1.814 .070 -.001815 .045427 
temp .005690 .002705 2551.307 2.104 .036 .000386 .010995 
hf -2.760570 2.024667 1544.773 -1.363 .173 -6.731957 1.210817 
fund .065527 .013232 863.141 4.952 .000 .039557 .091497 
pop2 7.64942E-14 2.32215E-14 361.939 3.294 .001 3.08280E-14 1.22160E-13 
hf2 4.842355 2.265920 2025.009 2.137 .033 .398577 9.286132 
yearcoded -.121582 .020691 1780.903 -5.876 .000 -.162164 -.081000 
renocc * hf .130629 .057950 1142.548 2.254 .024 .016928 .244330 
renocc * hf2 -.205757 .065743 1712.275 -3.130 .002 -.334703 -.076812 
medinc * renedu 1.056058E-6 7.105381E-7 897.518 1.486 .138 -3.384520E-7 2.450567E-6 
renedu*medinc2 -1.47271E-11 5.83451E-12 842.331 -2.524 .012 -2.61790E-11 -3.27524E-12 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
252 
 
Table 136: Information criteria for Model 128 adding medinc * pov interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6556.349 
Akaike's Information Criterion (AIC) 6560.349 
Hurvich and Tsai's Criterion (AICC) 6560.353 
Bozdogan's Criterion (CAIC) 6574.098 
Schwarz's Bayesian Criterion (BIC) 6572.098 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 137: Type III tests of fixed effects for Model 128 adding medinc * pov 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 933.477 1.092 .296 
inczon 1 377.836 2.174 .141 
coccat 2 363.200 5.522 .004 
pop 0 . . . 
bach 1 622.696 20.224 .000 
medren 1 1173.783 11.814 .001 
medval 0 . . . 
renwhi 1 519.072 5.090 .024 
renedu 1 841.350 14.636 .000 
pov 1 771.910 4.031 .045 
burd 1 2361.943 3.278 .070 
vac 1 471.300 3.678 .056 
temp 1 2557.363 4.122 .042 
hf 1 1574.388 1.885 .170 
fund 1 829.335 31.107 .000 
pop2 0 . . . 
hf2 1 2038.227 4.494 .034 
yearcoded 1 1789.253 37.538 .000 
renocc * hf 1 1151.200 5.102 .024 
renocc * hf2 1 1723.350 9.604 .002 
medinc * pov 1 1132.128 16.237 .000 
pov * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
253 
 
Table 138: Fixed effects estimates for Model 128 adding medinc * pov interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.294006 1.422899 867.000 .909 .363 -1.498724 4.086736 
[inczon=0] -.261414 .177301 377.836 -1.474 .141 -.610035 .087206 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .821879 .340424 359.516 2.414 .016 .152407 1.491351 
[coccat=2] .025252 .306927 370.098 .082 .934 -.578288 .628792 
[coccat=3] 0b 0 . . . . . 
pop -6.527653E-7 1.868220E-7 359.790 -3.494 .001 -1.020166E-6 -2.853651E-7 
bach -.061068 .013579 622.696 -4.497 .000 -.087734 -.034401 
medren .002646 .000770 1173.783 3.437 .001 .001135 .004156 
medval 6.418933E-6 9.526470E-7 1064.767 6.738 .000 4.549654E-6 8.288211E-6 
renwhi .011644 .005161 519.072 2.256 .024 .001505 .021783 
renedu -.043129 .011273 841.350 -3.826 .000 -.065257 -.021002 
pov -.118026 .058786 771.910 -2.008 .045 -.233427 -.002626 
burd -.043767 .024175 2361.943 -1.810 .070 -.091173 .003640 
vac .022182 .011567 471.300 1.918 .056 -.000546 .044911 
temp .005494 .002706 2557.363 2.030 .042 .000188 .010800 
hf -2.700150 1.966854 1574.388 -1.373 .170 -6.558079 1.157780 
fund .073645 .013204 829.335 5.577 .000 .047727 .099563 
pop2 6.954577E-14 2.272134E- 358.226 3.061 .002 2.486179E-14 1.142298E-13 
14 
hf2 4.725247 2.229008 2038.227 2.120 .034 .353875 9.096618 
yearcoded -.127061 .020739 1789.253 -6.127 .000 -.167736 -.086387 
renocc * hf .126076 .055818 1151.200 2.259 .024 .016559 .235593 
renocc * hf2 -.199709 .064442 1723.350 -3.099 .002 -.326103 -.073316 
medinc * pov 9.881022E-6 2.452150E-6 1132.128 4.030 .000 5.069752E-6 1.469229E-5 
pov * medinc2 -1.115757E- 2.433721E- 1290.497 -4.585 .000 -1.593206E- -6.383091E-
10 11 10 11 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
254 
 
Table 139: Information criteria for Model 129 adding medinc * burd interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6555.636 
Akaike's Information Criterion (AIC) 6559.636 
Hurvich and Tsai's Criterion (AICC) 6559.641 
Bozdogan's Criterion (CAIC) 6573.385 
Schwarz's Bayesian Criterion (BIC) 6571.385 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 140: Type III tests of fixed effects for Model 129 adding medinc * burd 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 911.344 1.583 .209 
inczon 1 374.587 2.974 .085 
coccat 2 363.776 5.359 .005 
pop 0 . . . 
bach 1 627.470 12.573 .000 
medren 1 1164.639 16.153 .000 
medval 0 . . . 
renwhi 1 530.489 2.620 .106 
renedu 1 846.441 14.564 .000 
pov 1 1641.195 4.949 .026 
burd 1 1647.991 .243 .622 
vac 1 512.056 1.328 .250 
temp 1 2524.471 3.185 .074 
hf 1 1478.531 .391 .532 
fund 1 844.752 27.526 .000 
pop2 0 . . . 
hf2 1 1954.228 2.548 .111 
yearcoded 1 1814.500 34.294 .000 
renocc * hf 1 1090.961 1.881 .171 
renocc * hf2 1 1639.004 6.274 .012 
medinc * burd 1 1030.324 .137 .711 
burd * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
255 
 
Table 141: Fixed effects estimates for Model 129 adding medinc * burd interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.604093 1.393811 846.370 1.151 .250 -1.131639 4.339824 
[inczon=0] -.302578 .175442 374.587 -1.725 .085 -.647553 .042397 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .774047 .339344 359.550 2.281 .023 .106698 1.441395 
[coccat=2] -.015802 .306039 371.388 -.052 .959 -.617587 .585984 
[coccat=3] 0b 0 . . . . . 
pop -7.228696E-7 1.869683E-7 362.249 -3.866 .000 -1.090549E-6 -3.551901E-7 
bach -.050496 .014241 627.470 -3.546 .000 -.078462 -.022531 
medren .003333 .000829 1164.639 4.019 .000 .001706 .004961 
medval 6.290463E-6 9.481405E-7 1043.959 6.635 .000 4.429984E-6 8.150941E-6 
renwhi .008346 .005156 530.489 1.619 .106 -.001783 .018475 
renedu -.042763 .011205 846.441 -3.816 .000 -.064757 -.020770 
pov .064294 .028900 1641.195 2.225 .026 .007609 .120979 
burd -.021544 .043671 1647.991 -.493 .622 -.107200 .064113 
vac .013867 .012032 512.056 1.153 .250 -.009772 .037507 
temp .004856 .002721 2524.471 1.785 .074 -.000480 .010192 
hf -1.268679 2.029812 1478.531 -.625 .532 -5.250297 2.712938 
fund .068598 .013075 844.752 5.247 .000 .042935 .094261 
pop2 7.827590E-14 2.269918E- 360.254 3.448 .001 3.363636E-14 1.229154E-13 
14 
hf2 3.612328 2.263207 1954.228 1.596 .111 -.826225 8.050880 
yearcoded -.120501 .020577 1814.500 -5.856 .000 -.160858 -.080144 
renocc * hf .079808 .058194 1090.961 1.371 .171 -.034377 .193994 
renocc * hf2 -.164490 .065669 1639.004 -2.505 .012 -.293295 -.035685 
medinc * burd 3.810254E-7 1.029536E-6 1030.324 .370 .711 -1.639200E-6 2.401251E-6 
burd * medinc2 -1.866524E- 6.587901E- 881.393 -2.833 .005 -3.159504E- -5.735434E-
11 12 11 12 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
256 
 
Table 142: Information criteria for Model 130 adding medinc * vac interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6559.869 
Akaike's Information Criterion (AIC) 6563.869 
Hurvich and Tsai's Criterion (AICC) 6563.874 
Bozdogan's Criterion (CAIC) 6577.618 
Schwarz's Bayesian Criterion (BIC) 6575.618 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 143: Type III tests of fixed effects for Model 130 adding medinc * vac 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 978.910 3.333 .068 
inczon 1 376.142 2.675 .103 
coccat 2 365.972 5.583 .004 
pop 0 . . . 
bach 1 616.222 23.652 .000 
medren 1 1423.419 7.585 .006 
medval 0 . . . 
renwhi 1 523.468 3.242 .072 
renedu 1 851.613 19.400 .000 
pov 1 1534.098 6.527 .011 
burd 1 2360.745 1.771 .183 
vac 1 867.835 1.534 .216 
temp 1 2568.086 4.323 .038 
hf 1 1578.628 3.292 .070 
fund 1 856.029 25.731 .000 
pop2 0 . . . 
hf2 1 2059.226 6.297 .012 
yearcoded 1 1755.922 39.856 .000 
renocc * hf 1 1132.374 7.889 .005 
renocc * hf2 1 1730.240 12.766 .000 
medinc * vac 1 961.169 6.762 .009 
vac * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
257 
 
Table 144: Fixed effects estimates for Model 130 adding medinc * vac interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.347813 1.390763 896.835 1.688 .092 -.381716 5.077342 
[inczon=0] -.290980 .177907 376.142 -1.636 .103 -.640798 .058838 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .800572 .344546 362.782 2.324 .021 .123013 1.478131 
[coccat=2] -.018239 .311654 375.635 -.059 .953 -.631044 .594566 
[coccat=3] 0b 0 . . . . . 
pop -7.190135E-7 1.901573E-7 365.027 -3.781 .000 -1.092955E-6 -3.450722E-7 
bach -.065442 .013456 616.222 -4.863 .000 -.091868 -.039016 
medren .001895 .000688 1423.419 2.754 .006 .000545 .003245 
medval 6.176346E-6 9.530435E-7 1033.186 6.481 .000 4.306225E-6 8.046468E-6 
renwhi .009365 .005201 523.468 1.801 .072 -.000853 .019582 
renedu -.049141 .011157 851.613 -4.405 .000 -.071039 -.027243 
pov .071732 .028078 1534.098 2.555 .011 .016656 .126807 
burd -.031701 .023818 2360.745 -1.331 .183 -.078408 .015005 
vac -.095019 .076715 867.835 -1.239 .216 -.245588 .055550 
temp .005614 .002700 2568.086 2.079 .038 .000319 .010910 
hf -3.544358 1.953493 1578.628 -1.814 .070 -7.376072 .287356 
fund .066912 .013191 856.029 5.073 .000 .041022 .092803 
pop2 7.839020E-14 2.307199E- 361.721 3.398 .001 3.301813E-14 1.237623E-13 
14 
hf2 5.574422 2.221432 2059.226 2.509 .012 1.217935 9.930910 
yearcoded -.127574 .020208 1755.922 -6.313 .000 -.167208 -.087940 
renocc * hf .155100 .055222 1132.374 2.809 .005 .046751 .263449 
renocc * hf2 -.229056 .064108 1730.240 -3.573 .000 -.354792 -.103319 
medinc * vac 6.578615E-6 2.529785E-6 961.169 2.600 .009 1.614077E-6 1.154315E-5 
vac * medinc2 -7.789349E- 2.222740E- 1115.007 -3.504 .000 -1.215057E- -3.428124E-
11 11 10 11 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
258 
 
Table 145: Information criteria for Model 132 adding medinc * hf interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6603.937 
Akaike's Information Criterion (AIC) 6607.937 
Hurvich and Tsai's Criterion (AICC) 6607.941 
Bozdogan's Criterion (CAIC) 6621.684 
Schwarz's Bayesian Criterion (BIC) 6619.684 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 146: Type III tests of fixed effects for Model 132 adding medinc * hf 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 945.312 3.200 .074 
inczon 1 368.348 3.082 .080 
coccat 2 357.120 5.672 .004 
pop 0 . . . 
bach 1 638.426 21.134 .000 
medren 1 1454.157 8.994 .003 
medval 0 . . . 
renwhi 1 519.455 3.069 .080 
renedu 1 833.044 19.053 .000 
pov 1 1650.952 8.434 .004 
burd 1 2419.057 2.938 .087 
vac 1 492.445 3.637 .057 
temp 1 2543.228 4.580 .032 
hf 1 1549.892 3.262 .071 
fund 1 837.930 26.411 .000 
pop2 0 . . . 
hf2 1 2067.783 5.883 .015 
yearcoded 1 1814.730 39.470 .000 
renocc * hf 1 1120.877 4.940 .026 
renocc * hf2 1 1685.018 10.349 .001 
medinc * hf 1 1477.479 3.773 .052 
medinc * hf2 1 1975.790 4.337 .037 
hf * medinc2 0 . . . 
medinc2 * hf2 0 . . . 
259 
 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 147: Fixed effects estimates for Model 132 adding medinc * hf interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.264449 1.391422 869.004 1.627 .104 -.466491 4.995389 
[inczon=0] -.310666 .176954 368.348 -1.756 .080 -.658632 .037300 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .861643 .342094 353.495 2.519 .012 .188846 1.534439 
[coccat=2] .057952 .308472 364.434 .188 .851 -.548656 .664561 
[coccat=3] 0b 0 . . . . . 
pop -6.732020E-7 1.887162E-7 356.354 -3.567 .000 -1.044339E-6 -3.020646E-7 
bach -.063234 .013755 638.426 -4.597 .000 -.090245 -.036223 
medren .002251 .000751 1454.157 2.999 .003 .000779 .003723 
medval 5.959137E-6 9.600909E-7 1057.094 6.207 .000 4.075237E-6 7.843038E-6 
renwhi .009110 .005200 519.455 1.752 .080 -.001106 .019326 
renedu -.048602 .011135 833.044 -4.365 .000 -.070457 -.026747 
pov .082035 .028248 1650.952 2.904 .004 .026630 .137441 
burd -.041570 .024254 2419.057 -1.714 .087 -.089131 .005991 
vac .022477 .011786 492.445 1.907 .057 -.000681 .045634 
temp .005802 .002711 2543.228 2.140 .032 .000486 .011119 
hf -10.061591 5.571077 1549.892 -1.806 .071 -20.989235 .866052 
fund .067900 .013212 837.930 5.139 .000 .041967 .093833 
pop2 7.39970E-14 2.29541E-14 355.135 3.224 .001 2.885400E-14 1.191402E-13 
hf2 16.789217 6.922060 2067.783 2.425 .015 3.214282 30.364151 
yearcoded -.129237 .020571 1814.730 -6.283 .000 -.169582 -.088892 
renocc * hf .126201 .056783 1120.877 2.223 .026 .014788 .237614 
renocc * hf2 -.208547 .064827 1685.018 -3.217 .001 -.335697 -.081396 
medinc * hf .000303 .000156 1477.479 1.942 .052 -2.989829E-6 .000609 
medinc * hf2 -.000424 .000204 1975.790 -2.083 .037 -.000823 -2.470244E-5 
hf * medinc2 -3.009604E-9 1.111687E-9 1322.890 -2.707 .007 -5.190467E-9 -8.28741E-10 
medinc2 * hf2 3.682995E-9 1.512114E-9 1817.664 2.436 .015 7.173309E-10 6.648659E-9 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
260 
 
Table 148: Information criteria for Model 133 adding medinc * fund interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6562.545 
Akaike's Information Criterion (AIC) 6566.545 
Hurvich and Tsai's Criterion (AICC) 6566.549 
Bozdogan's Criterion (CAIC) 6580.294 
Schwarz's Bayesian Criterion (BIC) 6578.294 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 149: Type III tests of fixed effects for Model 133 adding medinc * fund 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 957.665 4.630 .032 
inczon 1 374.375 2.506 .114 
coccat 2 362.998 5.439 .005 
pop 0 . . . 
bach 1 564.804 38.087 .000 
medren 1 1272.533 3.159 .076 
medval 0 . . . 
renwhi 1 513.837 4.265 .039 
renedu 1 821.403 23.279 .000 
pov 1 1299.363 13.343 .000 
burd 1 2244.577 1.417 .234 
vac 1 442.724 9.349 .002 
temp 1 2583.038 5.139 .023 
hf 1 1562.299 3.904 .048 
fund 1 928.388 2.073 .150 
pop2 0 . . . 
hf2 1 2053.962 6.703 .010 
yearcoded 1 1766.790 45.406 .000 
renocc * hf 1 1086.310 9.426 .002 
renocc * hf2 1 1710.207 13.733 .000 
medinc * fund 1 1093.832 6.913 .009 
fund * medinc2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
261 
 
Table 150: Fixed effects estimates for Model 133 adding medinc * fund interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.670148 1.372539 874.553 1.945 .052 -.023707 5.364003 
[inczon=0] -.282274 .178307 374.375 -1.583 .114 -.632883 .068335 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .884656 .343970 359.782 2.572 .011 .208211 1.561100 
[coccat=2] .105859 .309802 370.041 .342 .733 -.503335 .715053 
[coccat=3] 0b 0 . . . . . 
pop -5.924213E-7 1.890843E-7 359.016 -3.133 .002 -9.642734E-7 -2.205693E-7 
bach -.079882 .012944 564.804 -6.171 .000 -.105305 -.054458 
medren .001127 .000634 1272.533 1.777 .076 -.000117 .002370 
medval 6.531240E-6 9.872575E-7 1133.350 6.616 .000 4.594182E-6 8.468298E-6 
renwhi .010747 .005204 513.837 2.065 .039 .000523 .020971 
renedu -.053412 .011070 821.403 -4.825 .000 -.075141 -.031683 
pov .092357 .025284 1299.363 3.653 .000 .042755 .141958 
burd -.028063 .023577 2244.577 -1.190 .234 -.074297 .018171 
vac .034325 .011226 442.724 3.058 .002 .012262 .056388 
temp .006106 .002694 2583.038 2.267 .023 .000824 .011388 
hf -3.826815 1.936847 1562.299 -1.976 .048 -7.625909 -.027721 
fund -.136927 .095094 928.388 -1.440 .150 -.323550 .049697 
pop2 6.416077E-14 2.30598E-14 359.209 2.782 .006 1.881153E-14 1.095100E-13 
hf2 5.725348 2.211370 2053.962 2.589 .010 1.388586 10.062110 
yearcoded -.132699 .019693 1766.790 -6.738 .000 -.171323 -.094075 
renocc * hf .166942 .054377 1086.310 3.070 .002 .060247 .273637 
renocc * hf2 -.235802 .063630 1710.207 -3.706 .000 -.360602 -.111002 
medinc * fund 7.667060E-6 2.916012E-6 1093.832 2.629 .009 1.945451E-6 1.338867E-5 
fund * medinc2 -6.37810E-11 2.02370E-11 1332.346 -3.152 .002 -1.03480E-10 -2.40812E-11 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
262 
 
Table 151: Information criteria for Model 134 adding medinc * yearcoded 
interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6566.855 
Akaike's Information Criterion (AIC) 6570.855 
Hurvich and Tsai's Criterion (AICC) 6570.859 
Bozdogan's Criterion (CAIC) 6584.603 
Schwarz's Bayesian Criterion (BIC) 6582.603 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 152: Type III tests of fixed effects for Model 134 adding medinc * yearcoded 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 972.595 4.911 .027 
inczon 1 373.680 3.386 .067 
coccat 2 362.524 5.801 .003 
pop 0 . . . 
bach 1 588.533 35.388 .000 
medren 1 1279.397 1.438 .231 
medval 0 . . . 
renwhi 1 519.493 3.596 .058 
renedu 1 835.292 22.680 .000 
pov 1 1412.772 13.926 .000 
burd 1 2290.118 .858 .354 
vac 1 452.944 9.537 .002 
temp 1 2573.840 5.309 .021 
hf 1 1584.499 3.967 .047 
fund 1 860.810 23.961 .000 
pop2 0 . . . 
hf2 1 2072.832 6.550 .011 
yearcoded 1 1878.093 14.937 .000 
renocc * hf 1 1115.112 9.548 .002 
renocc * hf2 1 1728.872 13.476 .000 
medinc * yearcoded 1 1813.655 9.675 .002 
medinc2 * yearcoded 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
263 
 
Table 153: Fixed effects estimates for Model 134 adding medinc * yearcoded 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Lower 
Parameter Estimate Std. Error df t Sig. Bound Upper Bound 
Intercept 2.804722 1.396521 889.579 2.008 .045 .063862 5.545581 
[inczon=0] -.330741 .179734 373.680 -1.840 .067 -.684157 .022676 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .941061 .347434 359.328 2.709 .007 .257803 1.624320 
[coccat=2] .138048 .313812 369.557 .440 .660 -.479033 .755130 
[coccat=3] 0b 0 . . . . . 
pop -6.19839E-7 1.91098E-7 358.096 -3.244 .001 -9.95655E-7 -2.44023E-7 
bach -.078547 .013204 588.533 -5.949 .000 -.104480 -.052615 
medren .000837 .000698 1279.397 1.199 .231 -.000533 .002208 
medval 6.172096E-6 9.69017E-7 1025.121 6.369 .000 4.270611E-6 8.073580E-6 
renwhi .009950 .005247 519.493 1.896 .058 -.000358 .020258 
renedu -.053543 .011243 835.292 -4.762 .000 -.075611 -.031475 
pov .098148 .026301 1412.772 3.732 .000 .046555 .149741 
burd -.021982 .023731 2290.118 -.926 .354 -.068519 .024554 
vac .035385 .011458 452.944 3.088 .002 .012867 .057902 
temp .006212 .002696 2573.840 2.304 .021 .000926 .011499 
hf -3.895937 1.956175 1584.499 -1.992 .047 -7.732900 -.058973 
fund .065369 .013354 860.810 4.895 .000 .039158 .091580 
pop2 6.84472E-14 2.3272E-14 357.233 2.941 .003 2.26792E-14 1.14215E-13 
hf2 5.693711 2.224647 2072.832 2.559 .011 1.330935 10.056488 
yearcoded -.534461 .138288 1878.093 -3.865 .000 -.805674 -.263247 
renocc * hf .170222 .055089 1115.112 3.090 .002 .062133 .278311 
renocc * hf2 -.235185 .064065 1728.872 -3.671 .000 -.360839 -.109531 
medinc * yearcoded 1.235730E-5 3.97276E-6 1813.655 3.111 .002 4.565631E-6 2.014897E-5 
medinc2 * -8.7904E-11 2.7536E-11 1824.373 -3.192 .001 -1.4191E-10 -3.3897E-11 
yearcoded 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
264 
 
Table 154: Information criteria for Model 135 removing burd 
Information Criteriaa 
-2 Restricted Log Likelihood 6498.810 
Akaike's Information Criterion (AIC) 6502.810 
Hurvich and Tsai's Criterion (AICC) 6502.815 
Bozdogan's Criterion (CAIC) 6516.561 
Schwarz's Bayesian Criterion (BIC) 6514.561 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 155: Type III tests of fixed effects for Model 135 removing burd 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 887.019 6.856 .009 
inczon 1 372.049 3.992 .046 
coccat 2 360.604 6.083 .003 
pop 0 . . . 
bach 1 563.148 38.647 .000 
medren 1 1170.144 .800 .371 
medval 0 . . . 
renwhi 1 523.214 3.084 .080 
renedu 1 833.008 23.868 .000 
pov 1 1030.524 7.741 .005 
vac 1 446.843 9.430 .002 
temp 1 2582.354 5.370 .021 
hf 1 1491.068 6.764 .009 
fund 1 874.045 23.022 .000 
pop2 0 . . . 
hf2 1 2024.655 9.229 .002 
yearcoded 1 1661.592 44.233 .000 
renocc * hf 1 1031.913 14.447 .000 
renocc * hf2 1 1679.789 17.625 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
265 
 
Table 156: Fixed effects estimates for Model 135 removing burd 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.131270 1.315059 813.743 2.381 .017 .549963 5.712578 
[inczon=0] -.355458 .177911 372.049 -1.998 .046 -.705296 -.005620 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .945704 .346063 360.318 2.733 .007 .265147 1.626261 
[coccat=2] .113853 .309303 362.042 .368 .713 -.494404 .722110 
[coccat=3] 0b 0 . . . . . 
pop -6.100333E-7 1.909559E-7 361.572 -3.195 .002 -9.855571E-7 -2.345095E-7 
bach -.080971 .013025 563.148 -6.217 .000 -.106554 -.055388 
medren .000525 .000588 1170.144 .894 .371 -.000627 .001678 
medval 5.919344E-6 9.539859E-7 1050.821 6.205 .000 4.047410E-6 7.791278E-6 
renwhi .009202 .005240 523.214 1.756 .080 -.001092 .019495 
renedu -.054420 .011139 833.008 -4.885 .000 -.076284 -.032556 
pov .060745 .021833 1030.524 2.782 .005 .017903 .103588 
vac .034714 .011304 446.843 3.071 .002 .012498 .056931 
temp .006253 .002698 2582.354 2.317 .021 .000962 .011544 
hf -4.971713 1.911653 1491.068 -2.601 .009 -8.721527 -1.221899 
fund .063694 .013275 874.045 4.798 .000 .037640 .089748 
pop2 6.86170E-14 2.32793E-14 361.346 2.948 .003 2.283709E-14 1.143970E-13 
hf2 6.676000 2.197537 2024.655 3.038 .002 2.366330 10.985670 
yearcoded -.122312 .018391 1661.592 -6.651 .000 -.158383 -.086241 
renocc * hf .203803 .053618 1031.913 3.801 .000 .098589 .309016 
renocc * hf2 -.265436 .063227 1679.789 -4.198 .000 -.389448 -.141425 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
266 
 
Table 157: Information criteria for Model 141 adding medval * burd interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6524.730 
Akaike's Information Criterion (AIC) 6528.730 
Hurvich and Tsai's Criterion (AICC) 6528.735 
Bozdogan's Criterion (CAIC) 6542.480 
Schwarz's Bayesian Criterion (BIC) 6540.480 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 158: Type III tests of fixed effects adding medval * burd interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 870.418 8.552 .004 
inczon 1 375.900 2.825 .094 
coccat 2 361.086 6.378 .002 
pop 0 . . . 
bach 1 572.550 34.607 .000 
medren 1 1154.472 .094 .760 
medval 1 1337.048 .002 .964 
renwhi 1 516.915 2.897 .089 
renedu 1 817.352 23.596 .000 
pov 1 1055.431 3.631 .057 
vac 1 440.255 9.211 .003 
temp 1 2585.655 5.472 .019 
hf 1 1487.777 7.786 .005 
fund 1 847.771 24.937 .000 
pop2 0 . . . 
hf2 1 2004.482 10.112 .001 
yearcoded 1 1638.989 37.085 .000 
renocc * hf 1 1017.199 15.836 .000 
renocc * hf2 1 1652.094 18.791 .000 
medval * burd 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
267 
 
Table 159: Fixed effects estimates for Model 141 adding medval * burd interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.559247 1.320525 801.219 2.695 .007 .967150 6.151344 
[inczon=0] -.298441 .177573 375.900 -1.681 .094 -.647601 .050719 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .882171 .342673 358.304 2.574 .010 .208268 1.556074 
[coccat=2] .014430 .308674 366.288 .047 .963 -.592567 .621426 
[coccat=3] 0b 0 . . . . . 
pop -6.127592E- 1.885128E-7 357.813 -3.250 .001 -9.834914E-7 -2.420270E-7 
7 
bach -.076722 .013042 572.550 -5.883 .000 -.102338 -.051106 
medren .000185 .000605 1154.472 .306 .760 -.001002 .001372 
medval -1.309192E- 2.920383E-6 1337.048 -.045 .964 -5.859951E-6 5.598113E-6 
7 
renwhi .008833 .005189 516.915 1.702 .089 -.001362 .019028 
renedu -.053714 .011058 817.352 -4.858 .000 -.075419 -.032009 
pov .043968 .023073 1055.431 1.906 .057 -.001306 .089242 
vac .033930 .011180 440.255 3.035 .003 .011957 .055902 
temp .006312 .002698 2585.655 2.339 .019 .001021 .011603 
hf -5.330299 1.910207 1487.777 -2.790 .005 -9.077284 -1.583313 
fund .065861 .013189 847.771 4.994 .000 .039975 .091748 
pop2 6.682861E- 2.299212E- 357.758 2.907 .004 2.161192E-14 1.120453E-13 
14 14 
hf2 6.980433 2.195102 2004.482 3.180 .001 2.675513 11.285353 
yearcoded -.114016 .018723 1638.989 -6.090 .000 -.150739 -.077293 
renocc * hf .212564 .053415 1017.199 3.979 .000 .107748 .317379 
renocc * hf2 -.273326 .063054 1652.094 -4.335 .000 -.397000 -.149652 
medval * burd 1.997317E-7 9.125471E-8 1334.543 2.189 .029 2.071343E-8 3.787500E-7 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
268 
 
Table 160: Information criteria for Model 150 removing medren 
Information Criteriaa 
-2 Restricted Log Likelihood 6486.568 
Akaike's Information Criterion (AIC) 6490.568 
Hurvich and Tsai's Criterion (AICC) 6490.573 
Bozdogan's Criterion (CAIC) 6504.320 
Schwarz's Bayesian Criterion (BIC) 6502.320 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 161: Type III tests of fixed effects for Model 150 removing medren 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 737.155 11.404 .001 
inczon 1 375.552 5.025 .026 
coccat 2 360.392 5.945 .003 
pop 0 . . . 
bach 1 576.206 37.998 .000 
medval 0 . . . 
renwhi 1 434.444 2.330 .128 
renedu 1 817.899 26.027 .000 
pov 1 902.463 7.016 .008 
vac 1 442.591 9.879 .002 
temp 1 2586.171 5.431 .020 
hf 1 1486.460 6.417 .011 
fund 1 870.768 22.613 .000 
pop2 0 . . . 
hf2 1 2019.339 8.847 .003 
yearcoded 1 1811.159 55.362 .000 
renocc * hf 1 1033.016 14.043 .000 
renocc * hf2 1 1679.445 17.201 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
269 
 
Table 162: Fixed effects estimates for Model 150 removing medren 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.636592 1.188484 683.169 3.060 .002 1.303072 5.970112 
[inczon=0] -.389735 .173858 375.552 -2.242 .026 -.731593 -.047878 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .962887 .345800 360.014 2.785 .006 .282845 1.642929 
[coccat=2] .153862 .306276 360.196 .502 .616 -.448451 .756175 
[coccat=3] 0b 0 . . . . . 
pop -5.946172E-7 1.903056E-7 362.748 -3.125 .002 -9.688581E-7 -2.203763E-7 
bach -.080117 .012997 576.206 -6.164 .000 -.105644 -.054589 
medval 6.448734E-6 7.482612E-7 659.010 8.618 .000 4.979471E-6 7.917997E-6 
renwhi .007369 .004828 434.444 1.526 .128 -.002120 .016858 
renedu -.056063 .010989 817.899 -5.102 .000 -.077633 -.034493 
pov .056462 .021316 902.463 2.649 .008 .014628 .098296 
vac .035458 .011281 442.591 3.143 .002 .013286 .057630 
temp .006287 .002698 2586.171 2.330 .020 .000997 .011577 
hf -4.825482 1.904925 1486.460 -2.533 .011 -8.562108 -1.088855 
fund .063078 .013265 870.768 4.755 .000 .037044 .089113 
pop2 6.697828E- 2.322309E- 361.672 2.884 .004 2.130903E-14 1.126475E-13 
14 14 
hf2 6.514962 2.190307 2019.339 2.974 .003 2.219466 10.810459 
yearcoded -.113029 .015191 1811.159 -7.441 .000 -.142822 -.083235 
renocc * hf .200522 .053510 1033.016 3.747 .000 .095522 .305523 
renocc * hf2 -.261708 .063101 1679.445 -4.147 .000 -.385473 -.137943 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
270 
 
Table 163: Information criteria for Model 155 adding medren * medval interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6517.769 
Akaike's Information Criterion (AIC) 6521.769 
Hurvich and Tsai's Criterion (AICC) 6521.774 
Bozdogan's Criterion (CAIC) 6535.520 
Schwarz's Bayesian Criterion (BIC) 6533.520 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 164: Type III tests of fixed effects for Model 155 adding medren * medval 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 732.362 7.999 .005 
inczon 1 376.664 4.576 .033 
coccat 2 359.586 5.620 .004 
pop 0 . . . 
bach 1 567.959 35.864 .000 
medval 1 1316.579 36.446 .000 
renwhi 1 437.600 1.589 .208 
renedu 1 803.106 27.142 .000 
pov 1 881.767 9.263 .002 
vac 1 439.687 10.075 .002 
temp 1 2592.698 5.158 .023 
hf 1 1485.332 4.496 .034 
fund 1 841.050 26.502 .000 
pop2 0 . . . 
hf2 1 2015.985 6.744 .009 
yearcoded 1 1777.758 36.493 .000 
renocc * hf 1 1048.024 10.559 .001 
renocc * hf2 1 1688.765 13.843 .000 
medren * medval 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
271 
 
Table 165: Fixed effects estimates for Model 155 adding medren * medval 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.020641 1.195742 682.652 2.526 .012 .672866 5.368415 
[inczon=0] -.367516 .171796 376.664 -2.139 .033 -.705316 -.029717 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .951206 .341271 359.286 2.787 .006 .280066 1.622345 
[coccat=2] .188172 .302490 360.051 .622 .534 -.406697 .783041 
[coccat=3] 0b 0 . . . . . 
pop -5.597886E- 1.881715E-7 362.737 -2.975 .003 -9.298326E-7 -1.897447E-7 
7 
bach -.077269 .012903 567.959 -5.989 .000 -.102612 -.051927 
medval 1.155605E-5 1.914180E-6 1316.579 6.037 .000 7.800875E-6 1.531123E-5 
renwhi .006045 .004795 437.600 1.261 .208 -.003380 .015470 
renedu -.056779 .010899 803.106 -5.210 .000 -.078173 -.035386 
pov .064924 .021332 881.767 3.044 .002 .023057 .106791 
vac .035387 .011148 439.687 3.174 .002 .013476 .057297 
temp .006126 .002697 2592.698 2.271 .023 .000837 .011415 
hf -4.053821 1.911879 1485.332 -2.120 .034 -7.804090 -.303552 
fund .068182 .013244 841.050 5.148 .000 .042186 .094177 
pop2 6.22685E-14 2.29726E-14 362.386 2.711 .007 1.70921E-14 1.07445E-13 
hf2 5.709165 2.198379 2015.985 2.597 .009 1.397833 10.020497 
yearcoded -.097235 .016096 1777.758 -6.041 .000 -.128803 -.065666 
renocc * hf .174939 .053837 1048.024 3.249 .001 .069298 .280580 
renocc * hf2 -.235941 .063414 1688.765 -3.721 .000 -.360319 -.111563 
medren*medval -3.15287E-9 1.090251E-9 1297.199 -2.892 .004 -5.291720E-9 -1.014023E-9 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
272 
 
Table 166: Information criteria for Model 156 adding medren * renwhi interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6501.315 
Akaike's Information Criterion (AIC) 6505.315 
Hurvich and Tsai's Criterion (AICC) 6505.320 
Bozdogan's Criterion (CAIC) 6519.066 
Schwarz's Bayesian Criterion (BIC) 6517.066 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 167: Type III tests of fixed effects for Model 156 adding medren * renwhi 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 736.935 10.417 .001 
inczon 1 370.593 3.145 .077 
coccat 2 358.975 6.139 .002 
pop 0 . . . 
bach 1 559.338 41.454 .000 
medval 0 . . . 
renwhi 1 562.308 .838 .360 
renedu 1 820.274 20.461 .000 
pov 1 940.300 9.217 .002 
vac 1 443.806 8.386 .004 
temp 1 2576.464 5.015 .025 
hf 1 1490.419 7.873 .005 
fund 1 861.331 24.749 .000 
pop2 0 . . . 
hf2 1 2016.881 10.113 .001 
yearcoded 1 1532.268 60.861 .000 
renocc * hf 1 1026.904 15.966 .000 
renocc * hf2 1 1668.707 18.657 .000 
medren * renwhi 1 819.964 6.863 .009 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
273 
 
Table 168: Fixed effects estimates for Model 156 adding medren * renwhi 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.450487 1.181137 682.776 2.921 .004 1.131390 5.769583 
[inczon=0] -.309757 .174676 370.593 -1.773 .077 -.653237 .033723 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .911919 .342479 359.570 2.663 .008 .238406 1.585432 
[coccat=2] .075174 .304395 358.830 .247 .805 -.523449 .673797 
[coccat=3] 0b 0 . . . . . 
pop -6.010333E- 1.882444E-7 361.296 -3.193 .002 -9.712256E-7 -2.308410E-7 
7 
bach -.083331 .012943 559.338 -6.439 .000 -.108754 -.057909 
medval 5.210784E-6 8.805831E-7 822.753 5.917 .000 3.482330E-6 6.939237E-6 
renwhi -.006564 .007171 562.308 -.915 .360 -.020649 .007521 
renedu -.050372 .011136 820.274 -4.523 .000 -.072231 -.028514 
pov .064988 .021407 940.300 3.036 .002 .022978 .106998 
vac .032505 .011225 443.806 2.896 .004 .010444 .054565 
temp .006045 .002699 2576.464 2.239 .025 .000752 .011338 
hf -5.353025 1.907777 1490.419 -2.806 .005 -9.095239 -1.610812 
fund .065651 .013197 861.331 4.975 .000 .039750 .091553 
pop2 6.90614E-14 2.29841E-14 359.972 3.005 .003 2.386148E-14 1.14261E-13 
hf2 6.965900 2.190436 2016.881 3.180 .001 2.670147 11.261654 
yearcoded -.135461 .017364 1532.268 -7.801 .000 -.169521 -.101402 
renocc * hf .213476 .053426 1026.904 3.996 .000 .108640 .318312 
renocc * hf2 -.271980 .062967 1668.707 -4.319 .000 -.395483 -.148477 
medren * renwhi 2.219553E-5 8.472744E-6 819.964 2.620 .009 5.564710E-6 3.882635E-5 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
274 
 
Table 169: Information criteria for Model 158 adding medren * pov interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6495.428 
Akaike's Information Criterion (AIC) 6499.428 
Hurvich and Tsai's Criterion (AICC) 6499.432 
Bozdogan's Criterion (CAIC) 6513.179 
Schwarz's Bayesian Criterion (BIC) 6511.179 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 170: Type III tests of fixed effects for Model 158 adding medren * pov 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 748.769 9.224 .002 
inczon 1 373.058 1.988 .159 
coccat 2 360.598 6.028 .003 
pop 0 . . . 
bach 1 589.488 35.619 .000 
medval 0 . . . 
renwhi 1 465.202 5.507 .019 
renedu 1 826.320 20.678 .000 
pov 1 700.687 .559 .455 
vac 1 447.275 7.844 .005 
temp 1 2583.158 5.035 .025 
hf 1 1473.046 5.350 .021 
fund 1 855.001 24.680 .000 
pop2 0 . . . 
hf2 1 2000.798 8.453 .004 
yearcoded 1 1563.814 64.561 .000 
renocc * hf 1 1040.385 11.210 .001 
renocc * hf2 1 1665.943 15.790 .000 
medren * pov 1 1004.633 9.773 .002 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
275 
 
Table 171: Fixed effects estimates for Model 158 adding medren * pov interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.261887 1.180979 690.817 2.762 .006 .943148 5.580625 
[inczon=0] -.249553 .176988 373.058 -1.410 .159 -.597572 .098465 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .837612 .342598 360.952 2.445 .015 .163874 1.511351 
[coccat=2] -.006589 .305855 362.996 -.022 .983 -.608058 .594881 
[coccat=3] 0b 0 . . . . . 
pop -6.569069E-7 1.884754E-7 362.179 -3.485 .001 -1.027551E-6 -2.862632E-7 
bach -.076883 .012882 589.488 -5.968 .000 -.102184 -.051583 
medval 5.347256E-6 8.200185E-7 826.970 6.521 .000 3.737693E-6 6.956818E-6 
renwhi .011591 .004939 465.202 2.347 .019 .001885 .021298 
renedu -.050239 .011048 826.320 -4.547 .000 -.071925 -.028554 
pov -.025187 .033691 700.687 -.748 .455 -.091335 .040961 
vac .031363 .011198 447.275 2.801 .005 .009356 .053370 
temp .006054 .002698 2583.158 2.244 .025 .000763 .011344 
hf -4.387754 1.896939 1473.046 -2.313 .021 -8.108745 -.666764 
fund .065303 .013145 855.001 4.968 .000 .039503 .091103 
pop2 7.076270E- 2.289858E- 360.112 3.090 .002 2.573096E-14 1.157944E-13 
14 14 
hf2 6.339456 2.180492 2000.798 2.907 .004 2.063184 10.615728 
yearcoded -.141480 .017608 1563.814 -8.035 .000 -.176017 -.106942 
renocc * hf .179054 .053479 1040.385 3.348 .001 .074115 .283992 
renocc * hf2 -.249730 .062846 1665.943 -3.974 .000 -.372997 -.126464 
medren * pov .000123 3.927192E-5 1004.633 3.126 .002 4.570660E-5 .000200 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
276 
 
Table 172: Information criteria for Model 159 adding medren * vac interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6496.511 
Akaike's Information Criterion (AIC) 6500.511 
Hurvich and Tsai's Criterion (AICC) 6500.516 
Bozdogan's Criterion (CAIC) 6514.262 
Schwarz's Bayesian Criterion (BIC) 6512.262 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 173: Type III tests of fixed effects for Model 159 adding medren * vac 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 731.017 10.247 .001 
inczon 1 370.459 3.340 .068 
coccat 2 358.777 6.309 .002 
pop 0 . . . 
bach 1 567.519 38.995 .000 
medval 0 . . . 
renwhi 1 452.256 5.032 .025 
renedu 1 804.002 21.964 .000 
pov 1 978.303 11.100 .001 
vac 1 692.624 2.914 .088 
temp 1 2576.312 4.665 .031 
hf 1 1467.766 6.315 .012 
fund 1 849.387 25.282 .000 
pop2 0 . . . 
hf2 1 1999.958 9.052 .003 
yearcoded 1 1666.435 65.213 .000 
renocc * hf 1 1016.894 13.133 .000 
renocc * hf2 1 1654.524 16.773 .000 
medren * vac 1 707.213 9.298 .002 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
277 
 
Table 174: Fixed effects estimates for Model 159 adding medren * vac interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.416004 1.175257 677.533 2.907 .004 1.108421 5.723587 
[inczon=0] -.315321 .172524 370.459 -1.828 .068 -.654570 .023928 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .903861 .340010 358.375 2.658 .008 .235194 1.572527 
[coccat=2] .056515 .302442 358.997 .187 .852 -.538265 .651296 
[coccat=3] 0b 0 . . . . . 
pop -6.285754E-7 1.872585E-7 360.187 -3.357 .001 -9.968326E-7 -2.603181E-7 
bach -.080024 .012815 567.519 -6.245 .000 -.105194 -.054853 
medval 5.724649E-6 7.762778E-7 671.692 7.374 .000 4.200426E-6 7.248872E-6 
renwhi .010954 .004884 452.256 2.243 .025 .001357 .020552 
renedu -.051451 .010978 804.002 -4.687 .000 -.073000 -.029901 
pov .072196 .021670 978.303 3.332 .001 .029672 .114721 
vac -.052941 .031015 692.624 -1.707 .088 -.113835 .007953 
temp .005839 .002703 2576.312 2.160 .031 .000538 .011141 
hf -4.751981 1.890942 1467.766 -2.513 .012 -8.461218 -1.042743 
fund .066042 .013135 849.387 5.028 .000 .040262 .091822 
pop2 7.076686E- 2.284331E- 359.300 3.098 .002 2.584347E-14 1.156903E-13 
14 14 
hf2 6.556069 2.179028 1999.958 3.009 .003 2.282666 10.829472 
yearcoded -.128228 .015879 1666.435 -8.075 .000 -.159372 -.097083 
renocc * hf .192248 .053048 1016.894 3.624 .000 .088151 .296345 
renocc * hf2 -.256817 .062707 1654.524 -4.095 .000 -.379810 -.133823 
medren * vac 9.134685E-5 2.995657E-5 707.213 3.049 .002 3.253240E-5 .000150 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
278 
 
Table 175: Information criteria for Model 164 removing renwhi 
Information Criteriaa 
-2 Restricted Log Likelihood 6480.055 
Akaike's Information Criterion (AIC) 6484.055 
Hurvich and Tsai's Criterion (AICC) 6484.060 
Bozdogan's Criterion (CAIC) 6497.808 
Schwarz's Bayesian Criterion (BIC) 6495.808 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 176: Type III tests of fixed effects for Model 164 removing renwhi 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 744.116 25.514 .000 
inczon 1 378.980 4.777 .029 
coccat 2 367.144 5.342 .005 
pop 0 . . . 
bach 1 559.092 42.153 .000 
medval 0 . . . 
renedu 1 782.603 31.688 .000 
pov 1 910.644 5.854 .016 
vac 1 442.221 10.751 .001 
temp 1 2612.282 4.437 .035 
hf 1 1531.778 5.138 .024 
fund 1 877.764 22.667 .000 
pop2 0 . . . 
hf2 1 2061.418 7.641 .006 
yearcoded 1 1832.813 56.569 .000 
renocc * hf 1 1055.946 12.080 .001 
renocc * hf2 1 1724.433 15.440 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
279 
 
Table 177: Fixed effects estimates for Model 164 removing renwhi 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 4.661186 .983429 691.081 4.740 .000 2.730318 6.592053 
[inczon=0] -.381389 .174498 378.980 -2.186 .029 -.724495 -.038283 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .841094 .337620 370.695 2.491 .013 .177203 1.504985 
[coccat=2] .048808 .299666 368.667 .163 .871 -.540460 .638077 
[coccat=3] 0b 0 . . . . . 
pop -6.755968E-7 1.835839E-7 367.950 -3.680 .000 -1.036602E-6 -3.145914E-7 
bach -.083441 .012852 559.092 -6.493 .000 -.108684 -.058197 
medval 6.286443E-6 7.429514E-7 671.283 8.461 .000 4.827655E-6 7.745231E-6 
renedu -.060117 .010679 782.603 -5.629 .000 -.081081 -.039153 
pov .050990 .021074 910.644 2.420 .016 .009630 .092350 
vac .036987 .011281 442.221 3.279 .001 .014817 .059157 
temp .005614 .002665 2612.282 2.107 .035 .000388 .010841 
hf -4.229334 1.865812 1531.778 -2.267 .024 -7.889151 -.569517 
fund .063296 .013295 877.764 4.761 .000 .037203 .089390 
pop2 7.58203E-14 2.25899E-14 363.840 3.356 .001 3.139699E-14 1.202437E-13 
hf2 5.979891 2.163359 2061.418 2.764 .006 1.737295 10.222488 
yearcoded -.114235 .015188 1832.813 -7.521 .000 -.144023 -.084447 
renocc * hf .180717 .051995 1055.946 3.476 .001 .078693 .282742 
renocc * hf2 -.244017 .062100 1724.433 -3.929 .000 -.365817 -.122218 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
280 
 
Table 178: Information criteria for Model 167 adding pop * renwhi interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6575.869 
Akaike's Information Criterion (AIC) 6579.869 
Hurvich and Tsai's Criterion (AICC) 6579.874 
Bozdogan's Criterion (CAIC) 6593.620 
Schwarz's Bayesian Criterion (BIC) 6591.620 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 179: Type III tests of fixed effects for Model 167 adding pop * renwhi 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 740.140 19.380 .000 
inczon 1 376.608 4.926 .027 
coccat 2 359.599 8.270 .000 
pop 0 . . . 
bach 1 556.950 41.335 .000 
medval 0 . . . 
renedu 1 779.500 28.362 .000 
pov 1 895.101 7.143 .008 
vac 1 437.330 11.941 .001 
temp 1 2608.978 5.566 .018 
hf 1 1500.707 6.613 .010 
fund 1 865.453 22.628 .000 
pop2 0 . . . 
hf2 1 2036.674 8.979 .003 
yearcoded 1 1810.052 55.774 .000 
renocc * hf 1 1032.252 14.634 .000 
renocc * hf2 1 1695.862 17.486 .000 
pop * renwhi 0 . . . 
renwhi * pop2 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
281 
 
Table 180: Fixed effects estimates for Model 167 adding pop * renwhi interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.873056 1.034192 677.422 3.745 .000 1.842449 5.903664 
[inczon=0] -.383559 .172810 376.608 -2.220 .027 -.723353 -.043764 
[inczon=1] 0b 0 . . . . . 
[coccat=1] 1.341180 .377252 358.718 3.555 .000 .599277 2.083083 
[coccat=2] .425116 .327059 357.955 1.300 .195 -.218082 1.068314 
[coccat=3] 0b 0 . . . . . 
pop -1.435312E-6 3.483569E-7 384.044 -4.120 .000 -2.120237E-6 -7.503865E-7 
bach -.082101 .012770 556.950 -6.429 .000 -.107184 -.057018 
medval 6.585060E-6 7.447818E-7 661.379 8.842 .000 5.122639E-6 8.047482E-6 
renedu -.057062 .010715 779.500 -5.326 .000 -.078094 -.036029 
pov .056247 .021045 895.101 2.673 .008 .014943 .097550 
vac .038710 .011202 437.330 3.456 .001 .016693 .060727 
temp .006305 .002672 2608.978 2.359 .018 .001065 .011544 
hf -4.819447 1.874097 1500.707 -2.572 .010 -8.495574 -1.143319 
fund .062886 .013220 865.453 4.757 .000 .036939 .088833 
pop2 1.330398E-13 4.92221E-14 371.469 2.703 .007 3.625085E-14 2.298288E-13 
hf2 6.497532 2.168398 2036.674 2.996 .003 2.245022 10.750042 
yearcoded -.113087 .015142 1810.052 -7.468 .000 -.142786 -.083389 
renocc * hf .200061 .052298 1032.252 3.825 .000 .097438 .302684 
renocc * hf2 -.260307 .062251 1695.862 -4.182 .000 -.382403 -.138211 
pop * renwhi 1.390840E-8 6.645377E-9 376.621 2.093 .037 8.417069E-10 2.697509E-8 
renwhi * pop2 -9.36675E-16 1.19174E-15 362.941 -.786 .432 -3.28026E-15 1.406910E-15 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
282 
 
Table 181: Information criteria for Model 168 adding bach * renwhi interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6491.177 
Akaike's Information Criterion (AIC) 6495.177 
Hurvich and Tsai's Criterion (AICC) 6495.181 
Bozdogan's Criterion (CAIC) 6508.928 
Schwarz's Bayesian Criterion (BIC) 6506.928 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 182: Type III tests of fixed effects for Model 168 adding bach * renwhi 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 778.209 16.654 .000 
inczon 1 375.876 5.136 .024 
coccat 2 361.661 6.289 .002 
pop 0 . . . 
bach 1 460.850 44.832 .000 
medval 0 . . . 
renedu 1 824.003 23.953 .000 
pov 1 894.239 7.277 .007 
vac 1 440.069 10.002 .002 
temp 1 2596.311 5.728 .017 
hf 1 1499.406 6.927 .009 
fund 1 868.166 22.985 .000 
pop2 0 . . . 
hf2 1 2031.052 9.196 .002 
yearcoded 1 1810.851 54.846 .000 
renocc * hf 1 1037.899 14.942 .000 
renocc * hf2 1 1689.527 17.780 .000 
bach * renwhi 1 440.346 4.486 .035 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
283 
 
Table 183: Fixed effects estimates for Model 168 adding bach * renwhi interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.860291 1.049031 718.376 3.680 .000 1.800758 5.919825 
[inczon=0] -.392792 .173328 375.876 -2.266 .024 -.733605 -.051978 
[inczon=1] 0b 0 . . . . . 
[coccat=1] 1.010164 .344858 362.180 2.929 .004 .331989 1.688340 
[coccat=2] .191376 .305050 361.569 .627 .531 -.408520 .791272 
[coccat=3] 0b 0 . . . . . 
pop -5.624230E-7 1.898616E-7 364.223 -2.962 .003 -9.357856E-7 -1.890604E-7 
bach -.099703 .014891 460.850 -6.696 .000 -.128965 -.070441 
medval 6.507764E-6 7.463386E-7 659.371 8.720 .000 5.042278E-6 7.973251E-6 
renedu -.053990 .011031 824.003 -4.894 .000 -.075643 -.032337 
pov .057110 .021171 894.239 2.698 .007 .015561 .098660 
vac .035507 .011227 440.069 3.163 .002 .013442 .057573 
temp .006435 .002688 2596.311 2.393 .017 .001163 .011706 
hf -4.991809 1.896663 1499.406 -2.632 .009 -8.712203 -1.271414 
fund .063458 .013236 868.166 4.794 .000 .037480 .089437 
pop2 6.380560E- 2.312494E- 362.856 2.759 .006 1.832987E-14 1.092813E-13 
14 14 
hf2 6.612463 2.180548 2031.052 3.032 .002 2.336120 10.888807 
yearcoded -.112408 .015178 1810.851 -7.406 .000 -.142177 -.082639 
renocc * hf .205289 .053109 1037.899 3.865 .000 .101077 .309502 
renocc * hf2 -.264354 .062693 1689.527 -4.217 .000 -.387318 -.141390 
bach * renwhi .000352 .000166 440.346 2.118 .035 2.536694E-5 .000679 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
284 
 
Table 184: Information criteria for Model 169 adding medval * renwhi interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6497.534 
Akaike's Information Criterion (AIC) 6501.534 
Hurvich and Tsai's Criterion (AICC) 6501.538 
Bozdogan's Criterion (CAIC) 6515.285 
Schwarz's Bayesian Criterion (BIC) 6513.285 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 185: Type III tests of fixed effects for Model 169 adding medval * renwhi 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 761.758 13.777 .000 
inczon 1 378.145 5.697 .017 
coccat 2 365.798 6.918 .001 
pop 0 . . . 
bach 1 564.210 38.736 .000 
medval 1 557.147 3.357 .067 
renedu 1 815.466 20.156 .000 
pov 1 882.332 8.402 .004 
vac 1 444.824 7.848 .005 
temp 1 2610.711 6.197 .013 
hf 1 1523.128 9.313 .002 
fund 1 860.080 24.479 .000 
pop2 0 . . . 
hf2 1 2041.057 11.353 .001 
yearcoded 1 1793.999 51.352 .000 
renocc * hf 1 1054.511 18.966 .000 
renocc * hf2 1 1697.860 21.126 .000 
medval * renwhi 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
285 
 
Table 186: Fixed effects estimates for Model 169 adding medval * renwhi 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.348557 1.022108 706.289 3.276 .001 1.341823 5.355291 
[inczon=0] -.408712 .171242 378.145 -2.387 .017 -.745419 -.072006 
[inczon=1] 0b 0 . . . . . 
[coccat=1] 1.103097 .337592 367.395 3.268 .001 .439242 1.766953 
[coccat=2] .302086 .300399 367.150 1.006 .315 -.288631 .892804 
[coccat=3] 0b 0 . . . . . 
pop -4.86025E-7 1.858774E-7 369.336 -2.615 .009 -8.515362E-7 -1.205149E-7 
bach -.079021 .012697 564.210 -6.224 .000 -.103960 -.054083 
medval 2.260786E-6 1.233925E-6 557.147 1.832 .067 -1.629284E-7 4.684501E-6 
renedu -.048978 .010909 815.466 -4.490 .000 -.070392 -.027564 
pov .060699 .020941 882.332 2.899 .004 .019600 .101799 
vac .031279 .011165 444.824 2.801 .005 .009336 .053221 
temp .006642 .002668 2610.711 2.489 .013 .001410 .011874 
hf -5.770223 1.890818 1523.128 -3.052 .002 -9.479106 -2.061341 
fund .064954 .013128 860.080 4.948 .000 .039187 .090722 
pop2 5.97054E-14 2.24911E-14 365.623 2.655 .008 1.547718E-14 1.039338E-13 
hf2 7.334592 2.176773 2041.057 3.369 .001 3.065663 11.603520 
yearcoded -.108540 .015146 1793.999 -7.166 .000 -.138246 -.078833 
renocc * hf .230370 .052898 1054.511 4.355 .000 .126573 .334167 
renocc * hf2 -.287560 .062563 1697.860 -4.596 .000 -.410268 -.164851 
medval * renwhi 8.058153E-8 1.986940E-8 488.249 4.056 .000 4.154146E-8 1.196216E-7 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
286 
 
Table 187: Information criteria for Model 173 adding renwhi * temp interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6491.668 
Akaike's Information Criterion (AIC) 6495.668 
Hurvich and Tsai's Criterion (AICC) 6495.672 
Bozdogan's Criterion (CAIC) 6509.419 
Schwarz's Bayesian Criterion (BIC) 6507.419 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 188: Type III tests of fixed effects for Model 173 adding renwhi * temp 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 778.752 20.291 .000 
inczon 1 374.093 5.178 .023 
coccat 2 365.619 6.191 .002 
pop 0 . . . 
bach 1 559.498 40.236 .000 
medval 0 . . . 
renedu 1 805.940 27.414 .000 
pov 1 898.094 7.304 .007 
vac 1 437.495 10.232 .001 
temp 1 1081.964 1.463 .227 
hf 1 1543.350 6.575 .010 
fund 1 862.782 23.049 .000 
pop2 0 . . . 
hf2 1 2061.583 8.938 .003 
yearcoded 1 1808.202 55.367 .000 
renocc * hf 1 1074.664 14.643 .000 
renocc * hf2 1 1724.779 17.561 .000 
renwhi * temp 1 1037.217 5.119 .024 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
287 
 
Table 189: Fixed effects estimates for Model 173 adding renwhi * temp interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 4.130677 1.003996 728.304 4.114 .000 2.159606 6.101748 
[inczon=0] -.393097 .172751 374.093 -2.276 .023 -.732782 -.053411 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .979393 .339942 371.377 2.881 .004 .310941 1.647844 
[coccat=2] .164771 .300938 368.935 .548 .584 -.426998 .756541 
[coccat=3] 0b 0 . . . . . 
pop -5.739947E-7 1.870400E-7 376.997 -3.069 .002 -9.417671E-7 -2.062223E-7 
bach -.081117 .012788 559.498 -6.343 .000 -.106236 -.055999 
medval 6.466177E-6 7.415153E-7 663.620 8.720 .000 5.010178E-6 7.922175E-6 
renedu -.056300 .010753 805.940 -5.236 .000 -.077407 -.035193 
pov .057018 .021098 898.094 2.703 .007 .015611 .098426 
vac .035775 .011184 437.495 3.199 .001 .013794 .057757 
temp -.008000 .006615 1081.964 -1.209 .227 -.020979 .004979 
hf -4.815341 1.877889 1543.350 -2.564 .010 -8.498824 -1.131858 
fund .063427 .013211 862.782 4.801 .000 .037497 .089357 
pop2 6.52578E-14 2.28246E-14 370.682 2.859 .004 2.037573E-14 1.101400E-13 
hf2 6.487344 2.169949 2061.583 2.990 .003 2.231823 10.742865 
yearcoded -.112755 .015153 1808.202 -7.441 .000 -.142475 -.083035 
renocc * hf .200917 .052505 1074.664 3.827 .000 .097894 .303941 
renocc * hf2 -.261328 .062360 1724.779 -4.191 .000 -.383638 -.139018 
renwhi * temp .000220 9.703434E-5 1037.217 2.262 .024 2.912719E-5 .000410 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
288 
 
Table 190: Information criteria for Model 174 adding renwhi * hf interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6487.493 
Akaike's Information Criterion (AIC) 6491.493 
Hurvich and Tsai's Criterion (AICC) 6491.498 
Bozdogan's Criterion (CAIC) 6505.244 
Schwarz's Bayesian Criterion (BIC) 6503.244 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 191: Type III tests of fixed effects for Model 174 adding renwhi * hf 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 796.322 20.360 .000 
inczon 1 377.781 5.364 .021 
coccat 2 364.133 6.016 .003 
pop 0 . . . 
bach 1 577.526 38.660 .000 
medval 0 . . . 
renedu 1 821.264 26.363 .000 
pov 1 916.475 6.626 .010 
vac 1 442.586 10.668 .001 
temp 1 2582.622 5.259 .022 
hf 1 1320.991 10.766 .001 
fund 1 870.612 22.779 .000 
pop2 0 . . . 
hf2 1 2153.406 12.678 .000 
yearcoded 1 1813.895 57.330 .000 
renocc * hf 1 1123.306 17.528 .000 
renocc * hf2 1 1887.796 21.018 .000 
renwhi * hf 1 1183.009 5.651 .018 
renwhi * hf2 1 2122.614 5.780 .016 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
289 
 
Table 192: Fixed effects estimates for Model 174 adding renwhi * hf interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 4.224825 1.023350 739.215 4.128 .000 2.215807 6.233844 
[inczon=0] -.403036 .174018 377.781 -2.316 .021 -.745202 -.060870 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .958725 .344401 367.683 2.784 .006 .281483 1.635968 
[coccat=2] .143168 .305971 364.545 .468 .640 -.458522 .744857 
[coccat=3] 0b 0 . . . . . 
pop -6.176076E-7 1.908799E-7 369.073 -3.236 .001 -9.929562E-7 -2.422590E-7 
bach -.080579 .012960 577.526 -6.218 .000 -.106032 -.055125 
medval 6.369170E-6 7.451415E-7 665.602 8.548 .000 4.906059E-6 7.832281E-6 
renedu -.056318 .010969 821.264 -5.134 .000 -.077847 -.034788 
pov .054381 .021126 916.475 2.574 .010 .012920 .095842 
vac .036793 .011265 442.586 3.266 .001 .014653 .058932 
temp .006171 .002691 2582.622 2.293 .022 .000894 .011448 
hf -9.628638 2.934521 1320.991 -3.281 .001 -15.385467 -3.871809 
fund .063304 .013264 870.612 4.773 .000 .037272 .089337 
pop2 6.90089E-14 2.32911E-14 365.652 2.963 .003 2.320746E-14 1.148103E-13 
hf2 13.144117 3.691461 2153.406 3.561 .000 5.904917 20.383317 
yearcoded -.115411 .015243 1813.895 -7.572 .000 -.145306 -.085516 
renocc * hf .240863 .057531 1123.306 4.187 .000 .127982 .353743 
renocc * hf2 -.327298 .071392 1887.796 -4.585 .000 -.467313 -.187283 
renwhi * hf .052455 .022067 1183.009 2.377 .018 .009160 .095749 
renwhi * hf2 -.067565 .028104 2122.614 -2.404 .016 -.122678 -.012452 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
290 
 
Table 193: Information criteria for Model 177 reintroducing gini 
Information Criteriaa 
-2 Restricted Log Likelihood 6474.585 
Akaike's Information Criterion (AIC) 6478.585 
Hurvich and Tsai's Criterion (AICC) 6478.589 
Bozdogan's Criterion (CAIC) 6492.336 
Schwarz's Bayesian Criterion (BIC) 6490.336 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 194: Type III tests of fixed effects for Model 177 reintroducing gini 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 688.914 9.205 .003 
inczon 1 379.012 4.738 .030 
coccat 2 368.829 5.105 .007 
pop 0 . . . 
bach 1 692.888 41.892 .000 
medval 0 . . . 
renedu 1 842.042 33.297 .000 
pov 1 1083.274 2.769 .096 
vac 1 468.033 8.232 .004 
temp 1 2600.918 4.062 .044 
hf 1 1499.135 4.284 .039 
fund 1 889.898 21.538 .000 
pop2 0 . . . 
hf2 1 2033.561 6.784 .009 
yearcoded 1 1780.220 58.100 .000 
renocc * hf 1 1044.308 10.623 .001 
renocc * hf2 1 1707.239 14.150 .000 
gini 1 1058.805 1.485 .223 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
291 
 
Table 195: Fixed effects estimates for Model 177 reintroducing gini 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 3.705489 1.256802 664.269 2.948 .003 1.237705 6.173272 
[inczon=0] -.378997 .174122 379.012 -2.177 .030 -.721363 -.036631 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .783621 .340041 372.629 2.304 .022 .114981 1.452261 
[coccat=2] -.002874 .301966 372.294 -.010 .992 -.596646 .590899 
[coccat=3] 0b 0 . . . . . 
pop -7.051038E-7 1.847778E-7 372.231 -3.816 .000 -1.068443E-6 -3.417647E-7 
bach -.089997 .013905 692.888 -6.472 .000 -.117297 -.062696 
medval 6.221050E-6 7.437323E-7 670.674 8.365 .000 4.760726E-6 7.681374E-6 
renedu -.063228 .010957 842.042 -5.770 .000 -.084735 -.041721 
pov .038806 .023319 1083.274 1.664 .096 -.006950 .084562 
vac .033389 .011637 468.033 2.869 .004 .010521 .056256 
temp .005387 .002673 2600.918 2.015 .044 .000146 .010628 
hf -3.897778 1.883250 1499.135 -2.070 .039 -7.591862 -.203693 
fund .061896 .013337 889.898 4.641 .000 .035720 .088071 
pop2 7.869913E- 2.266352E- 366.727 3.473 .001 3.413236E-14 1.232659E-13 
14 14 
hf2 5.669132 2.176627 2033.561 2.605 .009 1.400481 9.937783 
yearcoded -.117878 .015465 1780.220 -7.622 .000 -.148209 -.087547 
renocc * hf .171148 .052511 1044.308 3.259 .001 .068108 .274187 
renocc * hf2 -.235041 .062484 1707.239 -3.762 .000 -.357594 -.112487 
gini 3.574074 2.932974 1058.805 1.219 .223 -2.181027 9.329176 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
292 
 
Table 196: Information criteria for Model 180 reintroducing inczon * gini 
interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6467.642 
Akaike's Information Criterion (AIC) 6471.642 
Hurvich and Tsai's Criterion (AICC) 6471.647 
Bozdogan's Criterion (CAIC) 6485.393 
Schwarz's Bayesian Criterion (BIC) 6483.393 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 197: Type III tests of fixed effects for Model 180 reintroducing inczon * gini 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 666.781 8.232 .004 
inczon 1 647.607 1.677 .196 
coccat 2 369.378 4.959 .007 
pop 0 . . . 
bach 1 697.624 41.118 .000 
medval 0 . . . 
renedu 1 840.038 34.232 .000 
pov 1 1128.254 3.431 .064 
vac 1 467.469 7.938 .005 
temp 1 2600.262 4.142 .042 
hf 1 1496.650 4.138 .042 
fund 1 888.532 20.999 .000 
pop2 0 . . . 
hf2 1 2030.576 6.722 .010 
yearcoded 1 1777.559 58.462 .000 
renocc * hf 1 1044.212 10.156 .001 
renocc * hf2 1 1705.620 13.858 .000 
gini 1 1017.501 1.767 .184 
inczon * gini 1 657.085 2.262 .133 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
293 
 
Table 198: Fixed effects estimates for Model 180 reintroducing inczon * gini 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.104439 1.645533 573.972 1.279 .201 -1.127561 5.336440 
[inczon=0] 2.425475 1.872809 647.607 1.295 .196 -1.252036 6.102987 
[inczon=1] 0b 0 . . . . . 
[coccat=1] .786946 .339514 372.572 2.318 .021 .119342 1.454550 
[coccat=2] .017198 .301788 373.445 .057 .955 -.576220 .610616 
[coccat=3] 0b 0 . . . . . 
pop -6.80966E-7 1.851799E-7 374.744 -3.677 .000 -1.045088E-6 -3.168443E-7 
bach -.089139 .013901 697.624 -6.412 .000 -.116432 -.061846 
medval 6.124457E-6 7.456400E-7 676.299 8.214 .000 4.660409E-6 7.588504E-6 
renedu -.064133 .010961 840.038 -5.851 .000 -.085648 -.042618 
pov .043548 .023510 1128.254 1.852 .064 -.002581 .089676 
vac .032762 .011628 467.469 2.817 .005 .009912 .055611 
temp .005439 .002673 2600.262 2.035 .042 .000199 .010680 
hf -3.829186 1.882391 1496.650 -2.034 .042 -7.521591 -.136781 
fund .061112 .013336 888.532 4.582 .000 .034938 .087286 
pop2 7.50873E-14 2.27541E-14 369.403 3.300 .001 3.03435E-14 1.19831E-13 
hf2 5.640449 2.175525 2030.576 2.593 .010 1.373955 9.906943 
yearcoded -.118180 .015456 1777.559 -7.646 .000 -.148495 -.087866 
renocc * hf .167377 .052521 1044.212 3.187 .001 .064319 .270436 
renocc * hf2 -.232535 .062465 1705.620 -3.723 .000 -.355052 -.110019 
gini 7.023717 3.719118 708.979 1.889 .059 -.278086 14.325519 
[inczon=0] * -6.236651 4.146816 657.085 -1.504 .133 -14.379259 1.905956 
gini 
[inczon=1] * 0b 0 . . . . . 
gini 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
294 
 
Table 199: Information criteria for Model 181 reintroducing coccat * gini 
interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6452.868 
Akaike's Information Criterion (AIC) 6456.868 
Hurvich and Tsai's Criterion (AICC) 6456.872 
Bozdogan's Criterion (CAIC) 6470.617 
Schwarz's Bayesian Criterion (BIC) 6468.617 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 200: Type III tests of fixed effects for Model 181 reintroducing coccat * gini 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 674.262 1.176 .279 
inczon 1 653.575 1.100 .295 
coccat 2 674.223 .876 .417 
pop 0 . . . 
bach 1 701.735 42.692 .000 
medval 0 . . . 
renedu 1 846.681 35.920 .000 
pov 1 1138.646 3.098 .079 
vac 1 467.615 7.936 .005 
temp 1 2599.230 4.187 .041 
hf 1 1498.072 4.185 .041 
fund 1 925.987 18.435 .000 
pop2 0 . . . 
hf2 1 2023.631 6.922 .009 
yearcoded 1 1761.292 59.161 .000 
renocc * hf 1 1043.887 10.328 .001 
renocc * hf2 1 1698.141 14.232 .000 
gini 1 855.488 2.203 .138 
inczon * gini 1 662.515 1.563 .212 
coccat * gini 2 675.115 1.219 .296 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
295 
 
Table 201: Fixed effects estimates for Model 181 reintroducing coccat * gini 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.126017 4.885059 741.876 .435 .664 -7.464168 11.716202 
[inczon=0] 1.986411 1.893561 653.575 1.049 .295 -1.731785 5.704608 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -3.543472 5.631671 714.277 -.629 .529 -14.600080 7.513136 
[coccat=2] .784106 4.781196 767.519 .164 .870 -8.601667 10.169879 
[coccat=3] 0b 0 . . . . . 
pop -6.61636E-7 1.851775E-7 373.965 -3.573 .000 -1.025756E-6 -2.975169E-7 
bach -.090970 .013923 701.735 -6.534 .000 -.118305 -.063635 
medval 6.153670E-6 7.445956E-7 672.928 8.264 .000 4.691660E-6 7.615680E-6 
renedu -.066035 .011018 846.681 -5.993 .000 -.087661 -.044409 
pov .041495 .023576 1138.646 1.760 .079 -.004761 .087752 
vac .032723 .011616 467.615 2.817 .005 .009897 .055548 
temp .005470 .002673 2599.230 2.046 .041 .000228 .010712 
hf -3.855919 1.884795 1498.072 -2.046 .041 -7.553036 -.158802 
fund .057917 .013489 925.987 4.294 .000 .031444 .084390 
pop2 7.05928E-14 2.28871E-14 369.774 3.084 .002 2.55875E-14 1.15598E-13 
hf2 5.725803 2.176339 2023.631 2.631 .009 1.457703 9.993903 
yearcoded -.119291 .015509 1761.292 -7.692 .000 -.149709 -.088872 
renocc * hf .168834 .052536 1043.887 3.214 .001 .065745 .271922 
renocc * hf2 -.235664 .062467 1698.141 -3.773 .000 -.358186 -.113143 
gini 7.351234 11.363781 783.869 .647 .518 -14.955811 29.658279 
[inczon=0] * -5.242680 4.193912 662.515 -1.250 .212 -13.477642 2.992281 
gini 
[inczon=1] * 0b 0 . . . . . 
gini 
[coccat=1] * gini 9.126791 12.651132 721.335 .721 .471 -15.710647 33.964228 
[coccat=2] * gini -1.718744 10.979403 768.839 -.157 .876 -23.271909 19.834421 
[coccat=3] * gini 0b 0 . . . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
296 
 
Table 202: Information criteria for Model 182 reintroducing hf * gini interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6437.414 
Akaike's Information Criterion (AIC) 6441.414 
Hurvich and Tsai's Criterion (AICC) 6441.418 
Bozdogan's Criterion (CAIC) 6455.162 
Schwarz's Bayesian Criterion (BIC) 6453.162 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 203: Type III tests of fixed effects for Model 182 reintroducing hf * gini 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 2229.664 .002 .967 
inczon 1 652.970 1.075 .300 
coccat 2 677.705 .921 .399 
pop 0 . . . 
bach 1 700.394 42.043 .000 
medval 0 . . . 
renedu 1 841.761 35.209 .000 
pov 1 1132.054 2.833 .093 
vac 1 467.011 7.926 .005 
temp 1 2596.682 4.167 .041 
hf 1 2621.042 .042 .837 
fund 1 926.510 18.412 .000 
pop2 0 . . . 
hf2 1 2608.204 .016 .901 
yearcoded 1 1759.847 58.821 .000 
renocc * hf 1 1178.408 8.391 .004 
renocc * hf2 1 1954.718 9.860 .002 
gini 1 2291.777 1.662 .197 
inczon * gini 1 662.007 1.525 .217 
coccat * gini 2 678.418 1.270 .282 
hf * gini 1 2627.816 .217 .641 
hf2 * gini 1 2623.328 .076 .783 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
  
297 
 
Table 204: Fixed effects estimates for Model 182 reintroducing hf * gini interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept -.069637 6.125768 1262.991 -.011 .991 -12.087439 11.948165 
[inczon=0] 1.966274 1.896016 652.970 1.037 .300 -1.756750 5.689297 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -3.860185 5.659817 722.410 -.682 .495 -14.971839 7.251468 
[coccat=2] .595674 4.800396 776.289 .124 .901 -8.827621 10.018970 
[coccat=3] 0b 0 . . . . . 
pop -6.57799E-7 1.855347E-7 372.795 -3.545 .000 -1.022625E-6 -2.929740E-7 
bach -.090478 .013954 700.394 -6.484 .000 -.117874 -.063082 
medval 6.130822E-6 7.462926E-7 674.628 8.215 .000 4.665487E-6 7.596158E-6 
renedu -.065534 .011044 841.761 -5.934 .000 -.087212 -.043856 
pov .039915 .023716 1132.054 1.683 .093 -.006617 .086446 
vac .032817 .011656 467.011 2.815 .005 .009911 .055722 
temp .005458 .002674 2596.682 2.041 .041 .000215 .010701 
hf 2.892496 14.082878 2621.042 .205 .837 -24.722189 30.507181 
fund .057974 .013511 926.510 4.291 .000 .031459 .084490 
pop2 6.98884E-14 2.29421E-14 368.696 3.046 .002 2.47745E-14 1.15002E-13 
hf2 1.646841 13.201689 2608.204 .125 .901 -24.240007 27.533689 
yearcoded -.119087 .015527 1759.847 -7.669 .000 -.149541 -.088633 
renocc * hf .164272 .056710 1178.408 2.897 .004 .053007 .275536 
renocc * hf2 -.222953 .071004 1954.718 -3.140 .002 -.362204 -.083701 
gini 12.332136 14.125235 1294.553 .873 .383 -15.378724 40.042996 
[inczon=0] * -5.186814 4.199500 662.007 -1.235 .217 -13.432757 3.059130 
gini 
[inczon=1] * 0b 0 . . . . . 
gini 
[coccat=1] * gini 9.795723 12.710788 729.167 .771 .441 -15.158384 34.749830 
[coccat=2] * gini -1.305364 11.021802 777.158 -.118 .906 -22.941393 20.330666 
[coccat=3] * gini 0b 0 . . . . . 
hf * gini -15.101695 32.410010 2627.816 -.466 .641 -78.653419 48.450028 
hf2 * gini 8.483660 30.790369 2623.328 .276 .783 -51.892210 68.859530 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
298 
 
Table 205: Information criteria for Model 201 adding medval * pov interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6434.321 
Akaike's Information Criterion (AIC) 6438.321 
Hurvich and Tsai's Criterion (AICC) 6438.326 
Bozdogan's Criterion (CAIC) 6452.068 
Schwarz's Bayesian Criterion (BIC) 6450.068 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 206: Type III tests of fixed effects for Model 201 adding medval * pov 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 2187.093 .340 .560 
inczon 1 641.926 .073 .786 
coccat 2 654.751 .659 .518 
pop 0 . . . 
bach 1 730.132 26.559 .000 
medval 1 755.368 .652 .420 
renedu 1 828.789 30.936 .000 
pov 1 912.212 4.307 .038 
vac 1 460.104 9.174 .003 
temp 1 2609.082 4.320 .038 
hf 1 2625.225 .000 .995 
fund 1 889.576 22.178 .000 
pop2 0 . . . 
hf2 1 2616.628 .053 .818 
yearcoded 1 1720.050 62.493 .000 
renocc * hf 1 1180.402 2.234 .135 
renocc * hf2 1 1925.857 4.834 .028 
gini 1 2251.968 .551 .458 
inczon * gini 1 651.234 .025 .874 
coccat * gini 2 655.251 .927 .396 
hf * gini 1 2623.944 .011 .916 
hf2 * gini 1 2626.166 .001 .981 
medval * pov 0 . . . 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
299 
 
 
Table 207: Fixed effects estimates for Model 201 adding medval * pov interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.613423 6.007234 1226.030 .435 .664 -9.172173 14.399020 
[inczon=0] -.510351 1.882488 641.926 -.271 .786 -4.206930 3.186228 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -1.587286 5.503787 698.475 -.288 .773 -12.393236 9.218663 
[coccat=2] 1.892098 4.665346 743.310 .406 .685 -7.266725 11.050921 
[coccat=3] 0b 0 . . . . . 
pop -6.01305E-7 1.779338E-7 373.535 -3.379 .001 -9.511833E-7 -2.514285E-7 
bach -.071758 .013924 730.132 -5.154 .000 -.099095 -.044422 
medval -1.16035E-6 1.436570E-6 755.368 -.808 .420 -3.980495E-6 1.659793E-6 
renedu -.060024 .010792 828.789 -5.562 .000 -.081207 -.038842 
pov -.059666 .028752 912.212 -2.075 .038 -.116093 -.003239 
vac .033973 .011217 460.104 3.029 .003 .011931 .056015 
temp .005532 .002662 2609.082 2.078 .038 .000313 .010752 
hf .080928 14.012457 2625.225 .006 .995 -27.395650 27.557506 
fund .061959 .013157 889.576 4.709 .000 .036137 .087780 
pop2 5.17442E-14 2.21769E-14 371.254 2.333 .020 8.13608E-15 9.53524E-14 
hf2 3.026579 13.138286 2616.628 .230 .818 -22.735905 28.789062 
yearcoded -.120701 .015268 1720.050 -7.905 .000 -.150647 -.090754 
renocc * hf .085116 .056946 1180.402 1.495 .135 -.026611 .196844 
renocc * hf2 -.155766 .070843 1925.857 -2.199 .028 -.294703 -.016829 
gini 6.779120 13.850624 1257.164 .489 .625 -20.393766 33.952006 
[inczon=0] * .661590 4.184780 651.234 .158 .874 -7.555700 8.878881 
gini 
[inczon=1] * 0b 0 . . . . . 
gini 
[coccat=1] * gini 4.561341 12.363352 704.268 .369 .712 -19.712099 28.834781 
[coccat=2] * gini -4.340798 10.712360 744.014 -.405 .685 -25.370849 16.689253 
[coccat=3] * gini 0b 0 . . . . . 
hf * gini -3.411824 32.256446 2623.944 -.106 .916 -66.662473 59.838825 
hf2 * gini .736026 30.643310 2626.166 .024 .981 -59.351451 60.823503 
medval * pov 7.223034E-7 1.229466E-7 770.656 5.875 .000 4.809535E-7 9.636533E-7 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
300 
 
Table 208: Information criteria for Model 215 adding pov * hf interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6427.524 
Akaike's Information Criterion (AIC) 6431.524 
Hurvich and Tsai's Criterion (AICC) 6431.528 
Bozdogan's Criterion (CAIC) 6445.269 
Schwarz's Bayesian Criterion (BIC) 6443.269 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 209: Type III tests of fixed effects for Model 215 adding pov * hf interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 2384.775 .196 .658 
inczon 1 644.121 .002 .961 
coccat 2 654.172 .545 .580 
pop 0 . . . 
bach 1 729.719 26.456 .000 
medval 1 767.012 .288 .592 
renedu 1 827.625 31.331 .000 
pov 1 2359.611 .765 .382 
vac 1 464.201 7.874 .005 
temp 1 2606.699 4.623 .032 
hf 1 2614.834 .015 .902 
fund 1 893.979 24.929 .000 
pop2 0 . . . 
hf2 1 2604.372 .287 .592 
yearcoded 1 1711.001 57.435 .000 
renocc * hf 1 1165.649 3.691 .055 
renocc * hf2 1 1889.296 8.169 .004 
gini 1 2492.948 .486 .486 
inczon * gini 1 653.270 .003 .954 
coccat * gini 2 654.640 .793 .453 
hf * gini 1 2620.920 .015 .902 
hf2 * gini 1 2611.112 .208 .648 
medval * pov 0 . . . 
pov * hf 1 2624.954 .633 .426 
pov * hf2 1 2622.138 2.452 .118 
301 
 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 210: Fixed effects estimates for Model 215 adding pov * hf interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.970560 6.288587 1406.097 .313 .754 -10.365463 14.306583 
[inczon=0] -.090920 1.882386 644.121 -.048 .961 -3.787274 3.605435 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -1.389442 5.491205 698.456 -.253 .800 -12.170688 9.391804 
[coccat=2] 1.760257 4.655106 741.701 .378 .705 -7.378495 10.899010 
[coccat=3] 0b 0 . . . . . 
pop -5.75348E-7 1.776524E-7 374.037 -3.239 .001 -9.246713E-7 -2.260262E-7 
bach -.071469 .013895 729.719 -5.143 .000 -.098748 -.044190 
medval -7.71360E-7 1.438445E-6 767.012 -.536 .592 -3.595116E-6 2.052395E-6 
renedu -.060284 .010770 827.625 -5.597 .000 -.081423 -.039144 
pov -.058238 .066601 2359.611 -.874 .382 -.188840 .072365 
vac .031512 .011230 464.201 2.806 .005 .009443 .053580 
temp .005714 .002657 2606.699 2.150 .032 .000503 .010925 
hf -1.908128 15.553459 2614.834 -.123 .902 -32.406465 28.590209 
fund .065765 .013172 893.979 4.993 .000 .039914 .091616 
pop2 4.87694E-14 2.21396E-14 371.812 2.203 .028 5.23472E-15 9.23040E-14 
hf2 7.666855 14.308993 2604.372 .536 .592 -20.391295 35.725005 
yearcoded -.115986 .015304 1711.001 -7.579 .000 -.146004 -.085969 
renocc * hf .110307 .057414 1165.649 1.921 .055 -.002339 .222954 
renocc * hf2 -.206536 .072264 1889.296 -2.858 .004 -.348262 -.064810 
gini 8.179126 15.236624 1577.244 .537 .591 -21.707043 38.065295 
[inczon=0] * -.240543 4.184048 653.270 -.057 .954 -8.456348 7.975261 
gini 
[inczon=1] * 0b 0 . . . . . 
gini 
[coccat=1] * gini 4.156960 12.335249 704.152 .337 .736 -20.061311 28.375231 
[coccat=2] * gini -4.050209 10.689020 742.391 -.379 .705 -25.034514 16.934097 
[coccat=3] * gini 0b 0 . . . . . 
hf * gini 4.840743 39.488575 2620.920 .123 .902 -72.591201 82.272687 
hf2 * gini -16.519346 36.232305 2611.112 -.456 .648 -87.566292 54.527601 
medval * pov 7.041679E-7 1.227874E-7 773.460 5.735 .000 4.631318E-7 9.452039E-7 
pov * hf -.182334 .229242 2624.954 -.795 .426 -.631849 .267180 
pov * hf2 .328103 .209549 2622.138 1.566 .118 -.082795 .739000 
302 
 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
303 
 
Table 211: Information criteria for Model 221 adding vac * yearcoded interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6425.769 
Akaike's Information Criterion (AIC) 6429.769 
Hurvich and Tsai's Criterion (AICC) 6429.774 
Bozdogan's Criterion (CAIC) 6443.514 
Schwarz's Bayesian Criterion (BIC) 6441.514 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 212: Type III tests of fixed effects for Model 221 adding vac * yearcoded 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 2379.592 .114 .735 
inczon 1 641.006 .000 .994 
coccat 2 650.803 .564 .569 
pop 0 . . . 
bach 1 726.552 26.436 .000 
medval 1 762.797 .163 .687 
renedu 1 822.563 31.545 .000 
pov 1 2354.967 .666 .415 
vac 1 779.315 19.666 .000 
temp 1 2607.112 5.106 .024 
hf 1 2614.625 .011 .918 
fund 1 886.398 24.270 .000 
pop2 0 . . . 
hf2 1 2604.475 .232 .630 
yearcoded 1 2083.569 1.076 .300 
renocc * hf 1 1167.368 5.179 .023 
renocc * hf2 1 1890.427 10.703 .001 
gini 1 2489.412 .494 .482 
inczon * gini 1 650.053 .013 .908 
coccat * gini 2 651.274 .802 .449 
hf * gini 1 2620.440 .009 .926 
hf2 * gini 1 2611.092 .152 .697 
medval * pov 0 . . . 
pov * hf 1 2623.986 .831 .362 
304 
 
pov * hf2 1 2621.343 3.128 .077 
vac * yearcoded 1 2210.959 12.422 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 213: Fixed effects estimates for Model 221 adding vac * yearcoded 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .868503 6.274823 1399.037 .138 .890 -11.440573 13.177578 
[inczon=0] .014288 1.873956 641.006 .008 .994 -3.665547 3.694124 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -.460531 5.473061 694.589 -.084 .933 -11.206258 10.285195 
[coccat=2] 2.512242 4.639571 737.435 .541 .588 -6.596100 11.620584 
[coccat=3] 0b 0 . . . . . 
pop -5.808386E- 1.766663E-7 373.536 -3.288 .001 -9.282239E-7 -2.334534E-7 
7 
bach -.071128 .013834 726.552 -5.142 .000 -.098287 -.043969 
medval -5.782818E- 1.433348E-6 762.797 -.403 .687 -3.392056E-6 2.235492E-6 
7 
renedu -.060233 .010724 822.563 -5.616 .000 -.081283 -.039183 
pov -.054206 .066436 2354.967 -.816 .415 -.184484 .076072 
vac .065357 .014738 779.315 4.435 .000 .036426 .094287 
temp .005995 .002653 2607.112 2.260 .024 .000793 .011197 
hf -1.594517 15.521384 2614.625 -.103 .918 -32.029959 28.840925 
fund .064648 .013123 886.398 4.926 .000 .038893 .090403 
pop2 4.975633E- 2.201752E- 371.274 2.260 .024 6.461647E- 9.305100E-
14 14 15 14 
hf2 6.874591 14.281530 2604.475 .481 .630 -21.129708 34.878889 
yearcoded -.029885 .028814 2083.569 -1.037 .300 -.086393 .026623 
renocc * hf .130882 .057510 1167.368 2.276 .023 .018048 .243716 
renocc * hf2 -.237475 .072589 1890.427 -3.272 .001 -.379838 -.095112 
gini 9.607927 15.191920 1570.663 .632 .527 -20.190652 39.406505 
[inczon=0] * -.483759 4.165452 650.053 -.116 .908 -8.663124 7.695606 
gini 
[inczon=1] * 0b 0 . . . . . 
gini 
[coccat=1] * gini 2.086779 12.294364 700.259 .170 .865 -22.051452 26.225010 
[coccat=2] * gini -5.763916 10.653184 738.129 -.541 .589 -26.678066 15.150234 
305 
 
[coccat=3] * gini 0b 0 . . . . . 
hf * gini 3.660029 39.406775 2620.440 .093 .926 -73.611522 80.931580 
hf2 * gini -14.078276 36.164284 2611.092 -.389 .697 -84.991841 56.835290 
medval * pov 6.823041E-7 1.224265E-7 767.939 5.573 .000 4.419739E-7 9.226343E-7 
pov * hf -.208697 .228870 2623.986 -.912 .362 -.657481 .240086 
pov * hf2 .370452 .209451 2621.343 1.769 .077 -.040253 .781157 
vac * yearcoded -.006905 .001959 2210.959 -3.525 .000 -.010746 -.003063 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
306 
 
Table 214: Information criteria for Model 224 adding temp * yearcoded interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6421.655 
Akaike's Information Criterion (AIC) 6425.655 
Hurvich and Tsai's Criterion (AICC) 6425.659 
Bozdogan's Criterion (CAIC) 6439.399 
Schwarz's Bayesian Criterion (BIC) 6437.399 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 215: Type III tests of fixed effects for Model 224 adding temp * yearcoded 
interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 2377.865 .030 .862 
inczon 1 640.378 .007 .934 
coccat 2 649.760 .689 .503 
pop 0 . . . 
bach 1 725.781 25.916 .000 
medval 1 761.026 .316 .574 
renedu 1 818.457 29.275 .000 
pov 1 2351.246 1.021 .312 
vac 1 785.799 15.399 .000 
temp 1 2611.651 21.338 .000 
hf 1 2613.567 .001 .974 
fund 1 888.094 22.006 .000 
pop2 0 . . . 
hf2 1 2603.142 .135 .714 
yearcoded 1 2321.795 4.523 .034 
renocc * hf 1 1165.676 5.364 .021 
renocc * hf2 1 1892.020 10.191 .001 
gini 1 2488.109 .596 .440 
inczon * gini 1 649.436 .037 .847 
coccat * gini 2 650.222 .961 .383 
hf * gini 1 2619.408 .000 .997 
hf2 * gini 1 2609.883 .072 .788 
medval * pov 0 . . . 
pov * hf 1 2622.969 .614 .433 
307 
 
pov * hf2 1 2620.157 2.619 .106 
vac * yearcoded 1 2218.165 7.496 .006 
temp * yearcoded 1 2543.898 16.267 .000 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 216: Fixed effects estimates for Model 224 adding temp * yearcoded 
interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .586368 6.256898 1398.599 .094 .925 -11.687548 12.860283 
[inczon=0] .155095 1.868775 640.378 .083 .934 -3.514573 3.824763 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -1.548664 5.463649 692.923 -.283 .777 -12.275958 9.178629 
[coccat=2] 1.965403 4.627939 736.322 .425 .671 -7.120126 11.050932 
[coccat=3] 0b 0 . . . . . 
pop -5.78212E-7 1.761453E-7 373.157 -3.283 .001 -9.245744E-7 -2.318506E-7 
bach -.070228 .013795 725.781 -5.091 .000 -.097311 -.043145 
medval -8.03407E-7 1.430230E-6 761.026 -.562 .574 -3.611072E-6 2.004256E-6 
renedu -.057938 .010708 818.457 -5.411 .000 -.078956 -.036919 
pov -.066999 .066318 2351.246 -1.010 .312 -.197048 .063049 
vac .058094 .014805 785.799 3.924 .000 .029033 .087156 
temp .022728 .004920 2611.651 4.619 .000 .013080 .032376 
hf -.512685 15.478773 2613.567 -.033 .974 -30.864579 29.839209 
fund .061489 .013108 888.094 4.691 .000 .035763 .087214 
pop2 4.96229E-14 2.19524E-14 370.920 2.260 .024 6.45606E-15 9.27898E-14 
hf2 5.230909 14.246023 2603.142 .367 .714 -22.703771 33.165589 
yearcoded .086664 .040749 2321.795 2.127 .034 .006756 .166573 
renocc * hf .132808 .057344 1165.676 2.316 .021 .020299 .245316 
renocc * hf2 -.231112 .072395 1892.020 -3.192 .001 -.373094 -.089130 
gini 9.320590 15.147782 1569.337 .615 .538 -20.391432 39.032612 
[inczon=0] * gini -.801552 4.153960 649.436 -.193 .847 -8.958365 7.355261 
[inczon=1] * gini 0b 0 . . . . . 
[coccat=1] * gini 4.473250 12.272527 698.578 .364 .716 -19.622208 28.568708 
[coccat=2] * gini -4.486783 10.626633 736.986 -.422 .673 -25.348863 16.375296 
[coccat=3] * gini 0b 0 . . . . . 
hf * gini .143090 39.302349 2619.408 .004 .997 -76.923709 77.209889 
hf2 * gini -9.708852 36.075868 2609.883 -.269 .788 -80.449061 61.031357 
medval * pov 6.819152E-7 1.220673E-7 767.255 5.586 .000 4.422897E-7 9.215407E-7 
308 
 
pov * hf -.178890 .228327 2622.969 -.783 .433 -.626609 .268829 
pov * hf2 .338219 .208997 2620.157 1.618 .106 -.071596 .748035 
vac * yearcoded -.005440 .001987 2218.165 -2.738 .006 -.009336 -.001543 
temp * yearcoded -.003777 .000937 2543.898 -4.033 .000 -.005614 -.001941 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
309 
 
Table 217: Information criteria for Model 225 adding hf * fund interaction 
Information Criteriaa 
-2 Restricted Log Likelihood 6420.719 
Akaike's Information Criterion (AIC) 6424.719 
Hurvich and Tsai's Criterion (AICC) 6424.724 
Bozdogan's Criterion (CAIC) 6438.462 
Schwarz's Bayesian Criterion (BIC) 6436.462 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 218: Type III tests of fixed effects for Model 225 adding hf * fund interaction 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 2418.720 .027 .870 
inczon 1 640.351 .005 .941 
coccat 2 649.456 .715 .490 
pop 0 . . . 
bach 1 725.968 26.376 .000 
medval 1 766.174 .164 .686 
renedu 1 821.725 29.785 .000 
pov 1 2405.731 .776 .378 
vac 1 783.550 15.926 .000 
temp 1 2610.541 21.485 .000 
hf 1 2609.294 .006 .940 
fund 1 2604.726 2.183 .140 
pop2 0 . . . 
hf2 1 2600.501 .033 .856 
yearcoded 1 2325.827 5.028 .025 
renocc * hf 1 1281.250 2.649 .104 
renocc * hf2 1 2087.905 5.256 .022 
gini 1 2528.481 .505 .477 
inczon * gini 1 649.461 .035 .852 
coccat * gini 2 649.916 .994 .371 
hf * gini 1 2611.723 .003 .956 
hf2 * gini 1 2601.779 .020 .888 
medval * pov 0 . . . 
pov * hf 1 2619.522 .530 .467 
pov * hf2 1 2614.020 2.196 .139 
310 
 
vac * yearcoded 1 2217.889 8.033 .005 
temp * yearcoded 1 2543.442 16.210 .000 
hf * fund 1 2595.756 .139 .709 
fund * hf2 1 2583.599 .009 .924 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
 
Table 219: Fixed effects estimates for Model 225 adding hf * fund interaction 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .742409 6.345850 1452.748 .117 .907 -11.705600 13.190418 
[inczon=0] .138394 1.871855 640.351 .074 .941 -3.537321 3.814109 
[inczon=1] 0b 0 . . . . . 
[coccat=1] -1.878982 5.471649 692.746 -.343 .731 -12.621987 8.864023 
[coccat=2] 1.758901 4.633629 737.670 .380 .704 -7.337770 10.855572 
[coccat=3] 0b 0 . . . . . 
pop -5.81714E-7 1.765098E-7 372.198 -3.296 .001 -9.287958E-7 -2.346327E-7 
bach -.070947 .013814 725.968 -5.136 .000 -.098068 -.043827 
medval -5.82164E-7 1.439746E-6 766.174 -.404 .686 -3.408480E-6 2.244151E-6 
renedu -.058541 .010727 821.725 -5.458 .000 -.079595 -.037486 
pov -.059866 .067942 2405.731 -.881 .378 -.193097 .073365 
vac .059292 .014857 783.550 3.991 .000 .030127 .088456 
temp .022798 .004918 2610.541 4.635 .000 .013154 .032442 
hf 1.206022 16.101985 2609.294 .075 .940 -30.367934 32.779978 
fund .155998 .105583 2604.726 1.477 .140 -.051038 .363034 
pop2 4.88791E-14 2.20076E-14 370.382 2.221 .027 5.60342E-15 9.21548E-14 
hf2 2.684197 14.837162 2600.501 .181 .856 -26.409648 31.778042 
yearcoded .091465 .040792 2325.827 2.242 .025 .011473 .171458 
renocc * hf .097365 .059825 1281.250 1.627 .104 -.020002 .214732 
renocc * hf2 -.175611 .076599 2087.905 -2.293 .022 -.325829 -.025393 
gini 8.583556 15.522266 1659.301 .553 .580 -21.861735 39.028847 
[inczon=0] * gini -.778145 4.160574 649.461 -.187 .852 -8.947945 7.391656 
[inczon=1] * gini 0b 0 . . . . . 
[coccat=1] * gini 5.137486 12.289497 698.534 .418 .676 -18.991293 29.266265 
[coccat=2] * gini -4.069268 10.639258 738.242 -.382 .702 -24.956074 16.817538 
[coccat=3] * gini 0b 0 . . . . . 
hf * gini -2.295082 41.348162 2611.723 -.056 .956 -83.373564 78.783400 
hf2 * gini -5.355943 37.912504 2601.779 -.141 .888 -79.697669 68.985783 
medval * pov 6.652342E-7 1.232425E-7 777.083 5.398 .000 4.233066E-7 9.071619E-7 
311 
 
pov * hf -.169765 .233253 2619.522 -.728 .467 -.627143 .287613 
pov * hf2 .315860 .213155 2614.020 1.482 .139 -.102110 .733830 
vac * yearcoded -.005637 .001989 2217.889 -2.834 .005 -.009537 -.001737 
temp * yearcoded -.003770 .000936 2543.442 -4.026 .000 -.005606 -.001934 
hf * fund -.125295 .335575 2595.756 -.373 .709 -.783316 .532726 
fund * hf2 -.025853 .270699 2583.599 -.096 .924 -.556662 .504956 
a. Dependent Variable: People per 1,000 Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
 
 
  
312 
 
Appendix 6: Final Model Results for Subsets of Homelessness 
Table 220: Information criterion for sheltered homelessness model 
Information Criteriaa 
-2 Restricted Log Likelihood 2631.088 
Akaike's Information Criterion (AIC) 2635.088 
Hurvich and Tsai's Criterion (AICC) 2635.093 
Bozdogan's Criterion (CAIC) 2648.839 
Schwarz's Bayesian Criterion (BIC) 2646.839 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Sheltered Homelessness. 
 
Table 221: Type III tests of fixed effects for sheltered homelessness model 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 703.947 7.790 .005 
coccat 2 368.023 12.969 .000 
inczon 1 377.736 3.984 .047 
pop 0 . . . 
bach 1 557.223 .014 .907 
medval 0 . . . 
renwhi 1 454.299 10.143 .002 
renedu 1 751.529 .630 .428 
pov 1 904.207 3.046 .081 
vac 1 666.860 10.069 .002 
temp 1 2602.589 5.816 .016 
hf 1 1437.334 73.844 .000 
fund 1 810.006 106.664 .000 
pop2 0 . . . 
hf2 1 1945.737 74.317 .000 
yearcoded 1 1618.058 39.833 .000 
renocc * hf 1 980.365 71.307 .000 
renocc * hf2 1 1594.910 88.135 .000 
medren * vac 1 679.627 14.697 .000 
a. Dependent Variable: People per 1,000 Experiencing Sheltered Homelessness. 
  
313 
 
Table 222: Fixed effects estimates for sheltered homelessness model 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 1.315003 .539504 657.748 2.437 .015 .255646 2.374360 
[coccat=1] .571874 .152957 367.958 3.739 .000 .271095 .872653 
[coccat=2] .021806 .136070 369.337 .160 .873 -.245764 .289375 
[coccat=3] 0b 0 . . . . . 
[inczon=0] -.155089 .077697 377.736 -1.996 .047 -.307862 -.002316 
[inczon=1] 0b 0 . . . . . 
pop -3.069454E-7 8.425347E-8 369.279 -3.643 .000 -4.726222E-7 -1.412686E-7 
bach -.000683 .005852 557.223 -.117 .907 -.012177 .010812 
medval 1.314303E-7 3.562310E-7 648.801 .369 .712 -5.680745E-7 8.309350E-7 
renwhi .007052 .002214 454.299 3.185 .002 .002701 .011404 
renedu -.004018 .005062 751.529 -.794 .428 -.013955 .005919 
pov .017530 .010044 904.207 1.745 .081 -.002182 .037242 
vac -.045202 .014245 666.860 -3.173 .002 -.073172 -.017232 
temp -.003139 .001301 2602.589 -2.412 .016 -.005691 -.000587 
hf -7.617765 .886479 1437.334 -8.593 .000 -9.356696 -5.878833 
fund .062659 .006067 810.006 10.328 .000 .050750 .074568 
pop2 4.21048E-14 1.02766E-14 367.836 4.097 .000 2.189640E-14 6.231329E-14 
hf2 8.880153 1.030096 1945.737 8.621 .000 6.859946 10.900359 
yearcoded -.047133 .007468 1618.058 -6.311 .000 -.061782 -.032485 
renocc * hf .207966 .024628 980.365 8.444 .000 .159637 .256295 
renocc * hf2 -.276794 .029484 1594.910 -9.388 .000 -.334625 -.218963 
medren * vac 5.277754E-5 1.376685E-5 679.627 3.834 .000 2.574686E-5 7.980821E-5 
a. Dependent Variable: People per 1,000 Experiencing Sheltered Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
314 
 
Table 223: Information criterion for unsheltered homelessness model 
Information Criteriaa 
-2 Restricted Log Likelihood 5869.934 
Akaike's Information Criterion (AIC) 5873.934 
Hurvich and Tsai's Criterion (AICC) 5873.939 
Bozdogan's Criterion (CAIC) 5887.685 
Schwarz's Bayesian Criterion (BIC) 5885.685 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Unsheltered Homelessness. 
 
Table 224: Type III tests of fixed effects for unsheltered homelessness model 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 721.536 4.471 .035 
coccat 2 361.591 .937 .393 
inczon 1 372.638 1.103 .294 
pop 0 . . . 
bach 1 563.817 50.032 .000 
medval 0 . . . 
renwhi 1 452.670 1.050 .306 
renedu 1 785.646 24.186 .000 
pov 1 952.653 8.561 .004 
vac 1 683.574 .169 .681 
temp 1 2585.314 14.134 .000 
hf 1 1457.306 2.810 .094 
fund 1 835.497 .310 .578 
pop2 0 . . . 
hf2 1 1982.470 1.307 .253 
yearcoded 1 1650.497 34.263 .000 
renocc * hf 1 1004.302 .062 .803 
renocc * hf2 1 1634.960 .064 .800 
medren * vac 1 697.482 2.613 .106 
a. Dependent Variable: People per 1,000 Experiencing Unsheltered Homelessness. 
  
315 
 
Table 225: Fixed effects estimates for unsheltered homelessness model 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept 2.032401 1.028680 670.589 1.976 .049 .012578 4.052223 
[coccat=1] .315075 .295610 361.295 1.066 .287 -.266257 .896407 
[coccat=2] .035727 .262954 362.155 .136 .892 -.481381 .552835 
[coccat=3] 0b 0 . . . . . 
[inczon=0] -.157607 .150052 372.638 -1.050 .294 -.452663 .137448 
[inczon=1] 0b 0 . . . . . 
pop -3.136161E-7 1.628145E-7 362.939 -1.926 .055 -6.337944E-7 6.562282E-9 
bach -.079203 .011197 563.817 -7.073 .000 -.101197 -.057209 
medval 5.538393E-6 6.793991E-7 663.532 8.152 .000 4.204362E-6 6.872424E-6 
renwhi .004362 .004257 452.670 1.025 .306 -.004004 .012728 
renedu -.047329 .009624 785.646 -4.918 .000 -.066220 -.028437 
pov .055678 .019029 952.653 2.926 .004 .018335 .093022 
vac -.011166 .027152 683.574 -.411 .681 -.064478 .042145 
temp .009034 .002403 2585.314 3.760 .000 .004322 .013746 
hf 2.793430 1.666299 1457.306 1.676 .094 -.475171 6.062032 
fund .006412 .011520 835.497 .557 .578 -.016199 .029024 
pop2 2.77769E-14 1.98607E-14 361.872 1.399 .163 -1.12801E-14 6.683402E-14 
hf2 -2.200800 1.925391 1982.470 -1.143 .253 -5.976803 1.575203 
yearcoded -.081991 .014007 1650.497 -5.853 .000 -.109465 -.054518 
renocc * hf -.011599 .046602 1004.302 -.249 .803 -.103047 .079849 
renocc * hf2 .014021 .055312 1634.960 .253 .800 -.094470 .122511 
medren * vac 4.240173E-5 2.623062E-5 697.482 1.616 .106 -9.098714E-6 9.390218E-5 
a. Dependent Variable: People per 1,000 Experiencing Unsheltered Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
316 
 
Table 226: Information criteria for family homelessness model 
Information Criteriaa 
-2 Restricted Log Likelihood 4219.289 
Akaike's Information Criterion (AIC) 4223.289 
Hurvich and Tsai's Criterion (AICC) 4223.294 
Bozdogan's Criterion (CAIC) 4237.040 
Schwarz's Bayesian Criterion (BIC) 4235.040 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 in Families Experiencing Homelessness. 
 
Table 227: Type III tests of fixed effects for family homelessness model 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 598.061 1.491 .223 
coccat 2 393.413 .908 .404 
inczon 1 393.469 4.529 .034 
pop 0 . . . 
bach 1 503.569 5.667 .018 
medval 0 . . . 
renwhi 1 448.450 3.292 .070 
renedu 1 580.060 1.835 .176 
pov 1 660.014 3.711 .054 
vac 1 565.277 .219 .640 
temp 1 2566.127 1.473 .225 
hf 1 1271.225 6.409 .011 
fund 1 673.193 23.250 .000 
pop2 0 . . . 
hf2 1 1614.917 12.161 .001 
yearcoded 1 1356.360 28.501 .000 
renocc * hf 1 820.291 10.293 .001 
renocc * hf2 1 1271.760 18.856 .000 
medren * vac 1 576.250 .649 .421 
a. Dependent Variable: People per 1,000 in Families Experiencing Homelessness. 
  
317 
 
Table 228: Fixed effects estimates for family homelessness model 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .680624 .582971 572.305 1.168 .243 -.464400 1.825649 
[coccat=1] .190378 .153379 395.208 1.241 .215 -.111164 .491919 
[coccat=2] .068397 .136650 400.495 .501 .617 -.200243 .337038 
[coccat=3] 0b 0 . . . . . 
[inczon=0] -.166109 .078056 393.469 -2.128 .034 -.319568 -.012651 
[inczon=1] 0b 0 . . . . . 
pop -1.481427E-7 8.450224E-8 394.782 -1.753 .080 -3.142733E-7 1.798796E-8 
bach -.014730 .006188 503.569 -2.380 .018 -.026886 -.002573 
medval 1.224828E-6 3.839673E-7 562.718 3.190 .002 4.706434E-7 1.979012E-6 
renwhi .004144 .002284 448.450 1.814 .070 -.000344 .008632 
renedu -.007479 .005522 580.060 -1.355 .176 -.018325 .003366 
pov .021572 .011198 660.014 1.926 .054 -.000416 .043561 
vac -.007209 .015391 565.277 -.468 .640 -.037440 .023021 
temp .002132 .001756 2566.127 1.214 .225 -.001312 .005576 
hf -2.692322 1.063451 1271.225 -2.532 .011 -4.778634 -.606010 
fund .032435 .006727 673.193 4.822 .000 .019227 .045643 
pop2 2.43348E-14 1.02944E-14 390.929 2.364 .019 4.095516E-15 4.457425E-14 
hf2 4.462446 1.279634 1614.917 3.487 .001 1.952529 6.972364 
yearcoded -.048380 .009062 1356.360 -5.339 .000 -.066157 -.030602 
renocc * hf .089977 .028045 820.291 3.208 .001 .034929 .145026 
renocc * hf2 -.154652 .035614 1271.760 -4.342 .000 -.224521 -.084782 
medren * vac 1.201098E-5 1.491394E-5 576.250 .805 .421 -1.728132E-5 4.130328E-5 
a. Dependent Variable: People per 1,000 in Families Experiencing Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
318 
 
Table 229: Information criteria for chronic homelessness model 
Information Criteriaa 
-2 Restricted Log Likelihood 593.224 
Akaike's Information Criterion (AIC) 597.224 
Hurvich and Tsai's Criterion (AICC) 597.229 
Bozdogan's Criterion (CAIC) 610.975 
Schwarz's Bayesian Criterion (BIC) 608.975 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: People per 1,000 Experiencing Chronic Homelessness. 
 
Table 230: Type III tests of fixed effects for chronic homelessness model 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 649.164 4.895 .027 
coccat 2 370.580 7.213 .001 
inczon 1 376.828 2.330 .128 
pop 0 . . . 
bach 1 526.736 32.017 .000 
medval 0 . . . 
renwhi 1 443.942 4.077 .044 
renedu 1 665.760 21.361 .000 
pov 1 786.640 19.817 .000 
vac 1 614.560 5.188 .023 
temp 1 2629.508 1.004 .316 
hf 1 1366.500 .014 .904 
fund 1 739.842 12.628 .000 
pop2 0 . . . 
hf2 1 1830.077 .047 .829 
yearcoded 1 1514.736 32.841 .000 
renocc * hf 1 905.892 .928 .336 
renocc * hf2 1 1472.184 1.051 .305 
medren * vac 1 625.590 10.213 .001 
a. Dependent Variable: People per 1,000 Experiencing Chronic Homelessness. 
  
319 
 
Table 231: Fixed effects estimates for chronic homelessness model 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .675977 .340103 612.683 1.988 .047 .008068 1.343886 
[coccat=1] .247164 .093641 371.106 2.639 .009 .063031 .431297 
[coccat=2] -.007651 .083330 373.743 -.092 .927 -.171505 .156203 
[coccat=3] 0b 0 . . . . . 
[inczon=0] -.072700 .047626 376.828 -1.526 .128 -.166346 .020945 
[inczon=1] 0b 0 . . . . . 
pop -1.606095E-7 5.158842E-8 371.743 -3.113 .002 -2.620512E-7 -5.916775E-8 
bach -.020709 .003660 526.736 -5.658 .000 -.027898 -.013519 
medval 1.702557E-6 2.243914E-7 602.004 7.587 .000 1.261872E-6 2.143242E-6 
renwhi .002768 .001371 443.942 2.019 .044 7.381208E-5 .005463 
renedu -.014824 .003207 665.760 -4.622 .000 -.021122 -.008526 
pov .028563 .006416 786.640 4.452 .000 .015968 .041158 
vac -.020462 .008984 614.560 -2.278 .023 -.038104 -.002819 
temp .000888 .000886 2629.508 1.002 .316 -.000850 .002626 
hf -.069531 .578772 1366.500 -.120 .904 -1.204910 1.065848 
fund .013719 .003861 739.842 3.554 .000 .006140 .021299 
pop2 1.422031E- 6.290513E- 369.468 2.261 .024 1.850607E-15 2.659001E-14 
14 15 
hf2 .147401 .681313 1830.077 .216 .829 -1.188833 1.483634 
yearcoded -.028073 .004899 1514.736 -5.731 .000 -.037682 -.018464 
renocc * hf .015230 .015811 905.892 .963 .336 -.015800 .046260 
renocc * hf2 -.019814 .019323 1472.184 -1.025 .305 -.057717 .018089 
medren * vac 2.777082E-5 8.689786E-6 625.590 3.196 .001 1.070614E-5 4.483550E-5 
a. Dependent Variable: People per 1,000 Experiencing Chronic Homelessness. 
b. This parameter is set to zero because it is redundant. 
  
320 
 
Table 232: Information criteria for veteran homelessness model 
Information Criteriaa 
-2 Restricted Log Likelihood -3954.029 
Akaike's Information Criterion (AIC) -3950.029 
Hurvich and Tsai's Criterion (AICC) -3950.024 
Bozdogan's Criterion (CAIC) -3936.553 
Schwarz's Bayesian Criterion (BIC) -3938.553 
The information criteria are displayed in smaller-is-better form. 
a. Dependent Variable: Veterans Experiencing Homelessness per 1,000 People. 
 
Table 233: Type III tests of fixed effects for veteran homelessness 
Type III Tests of Fixed Effectsa 
Source Numerator df Denominator df F Sig. 
Intercept 1 710.036 .772 .380 
coccat 2 373.265 6.961 .001 
inczon 1 383.275 .116 .733 
pop 0 . . . 
bach 1 562.390 9.930 .002 
medval 0 . . . 
renwhi 1 457.690 8.734 .003 
renedu 1 753.297 9.651 .002 
pov 1 884.631 9.020 .003 
vac 1 646.749 5.960 .015 
temp 1 2239.023 13.475 .000 
hf 1 1400.959 .019 .889 
fund 1 804.266 12.179 .001 
pop2 0 . . . 
hf2 1 1831.529 .363 .547 
yearcoded 1 1794.751 84.559 .000 
renocc * hf 1 980.374 6.378 .012 
renocc * hf2 1 1552.246 6.337 .012 
medren * vac 1 644.954 13.103 .000 
a. Dependent Variable: Veterans Experiencing Homelessness per 1,000 People. 
  
321 
 
Table 234: Fixed effects estimates for veteran homelessness model 
Estimates of Fixed Effectsa 
95% Confidence Interval 
Parameter Estimate Std. Error df t Sig. Lower Bound Upper Bound 
Intercept .094683 .143823 664.352 .658 .511 -.187720 .377085 
[coccat=1] .099358 .040965 373.162 2.425 .016 .018808 .179908 
[coccat=2] -.011628 .036406 374.534 -.319 .750 -.083213 .059958 
[coccat=3] 0b 0 . . . . . 
[inczon=0] -.007094 .020819 383.275 -.341 .733 -.048028 .033841 
[inczon=1] 0b 0 . . . . . 
pop -5.709531E-8 2.250945E-8 374.623 -2.537 .012 -1.013560E-7 -1.283461E-8 
bach -.004932 .001565 562.390 -3.151 .002 -.008007 -.001858 
medval 3.571368E-7 9.926381E-8 607.737 3.598 .000 1.621950E-7 5.520785E-7 
renwhi .001755 .000594 457.690 2.955 .003 .000588 .002922 
renedu -.004215 .001357 753.297 -3.107 .002 -.006878 -.001551 
pov .008136 .002709 884.631 3.003 .003 .002819 .013453 
vac -.009498 .003891 646.749 -2.441 .015 -.017138 -.001859 
temp .001224 .000333 2239.023 3.671 .000 .000570 .001878 
hf -.033651 .241785 1400.959 -.139 .889 -.507951 .440649 
fund .005612 .001608 804.266 3.490 .001 .002456 .008769 
pop2 5.31243E-15 2.74166E-15 372.676 1.938 .053 -7.86402E-17 1.070352E-14 
hf2 .167337 .277686 1831.529 .603 .547 -.377278 .711952 
yearcoded -.018605 .002023 1794.751 -9.196 .000 -.022574 -.014637 
renocc * hf .016632 .006586 980.374 2.526 .012 .003709 .029556 
renocc * hf2 -.019731 .007838 1552.246 -2.517 .012 -.035105 -.004357 
medren * vac 1.355517E-5 3.744674E-6 644.954 3.620 .000 6.201949E-6 2.090840E-5 
a. Dependent Variable: Veterans Experiencing Homelessness per 1,000 People. 
b. This parameter is set to zero because it is redundant. 
  
322 
 
Glossary 
Emergency shelter means any facility, the primary purpose of which is to provide 
temporary or transitional shelter for the homeless in general or for specific 
populations of the homeless. 
Housing First is a homelessness alleviation philosophy rooted in giving people 
experiencing homelessness immediate access to housing and support services. 
Other permanent housing means a program that must provide long-term housing 
that is not otherwise considered permanent supportive housing or rapid re-
housing, such as SRO or VA programs that provide permanent housing. 
Permanent housing means either permanent supportive housing, rapid re-housing, 
or other permanent housing in which people experiencing homelessness are 
placed in housing without a limit on how long they may stay, though there 
may be time limits placed on assistance. 
Permanent supportive housing means a program that must provide long-term 
housing to individuals with disabilities experiencing homelessness or families 
experiencing homelessness in which one member of the household has a 
disability and supportive services that are designed to meet the needs of the 
program participants must be available to the household. 
Rapid re-housing means a program that must provide short-term or medium-term 
assistance (up to 24 months), the lease for units must be between the landlord 
and the program participant, the program participant must be able to select 
the unit they lease, and the provider cannot impose a restriction on how long 
the person may lease the unit, though the provider can impose a maximum 
length of time that grant funds will be used to assist the program participant 
in the unit. 
Transitional housing means a project that has as its purpose facilitating the 
movement of homeless individuals and families to permanent housing within a 
reasonable amount of time (usually 24 months). Transitional housing includes 
housing primarily designed to serve deinstitutionalized homeless individuals 
and other homeless individuals with mental or physical disabilities and 
homeless families with children.  
323 
 
References 
American Planning Association. (1992). Ethical Principles in Planning.   Retrieved 
from https://www.planning.org/ethics/ethicalprinciples/ 
Anderson, L., Snow, D., & Cress, D. (1994). Negotiating the Public Realm: Stigma 
Management and Collective Action among the Homeless. Research in 
Community Sociology, 1, 121-143.  
Appelbaum, R. P., Dolny, M., Dreier, P., & Gilderbloom, J. I. (1991). Scapegoating 
Rent Control Masking the Causes of Homelessness. Journal of the American 
Planning Association, 57.  
Aubry, T., Goering, P., Veldhuizen, S., Adair, C. E., Bourque, J., Distasio, J., . . . 
Tsemberis, S. (2016). A Multiple-City Rct of Housing First with Assertive 
Community Treatment for Homeless Canadians with Serious Mental Illness. 
Psychiatric services (Washington, D.C.), 67(3), 275-281. 
doi:10.1176/appi.ps.201400587 
Bao, W.-N., Whitbeck, L. B., & Hoyt, D. R. (2000). Abuse, Support, and Depression 
among Homeless and Runaway Adolescents. Journal of Health and Social 
Behavior, 41(4), 408-420.  
Bassuk, E. L., Rubin, L., & Lauriat, A. (1984). Is Homelessness a Mental Health 
Problem? American Journal of Psychiatry, 141, 1546-1550.  
Benston, E. A. (2015). Housing Programs for Homeless Individuals with Mental 
Illness: Effects on Housing and Mental Health Outcomes. Psychiatric 
Services, 66(8), 806-816. doi:10.1176/appi.ps.201400294 
Berkovec, J. A., Canner, G. B., Hannan, T. H., & Gabriel, S. A. (1997). Race, 
Redlining, and Residential Mortgage Defaults: Evidence from the Fha-Insured 
Single-Family Loan Program. In The Urban Institute (Ed.), Mortgage 
Lending, Racial Discrimination, and Federal Policy. Farnham, United 
Kingdom: Ashgate Publishing. 
Bohanon, C. (1991). The Economic Correlates of Homelessness in Sixty Cities. 
Social Science Quarterly, 72, 817-825.  
324 
 
Boston, D. (2012). Unsheltered Homelessness in Maryland: Impact and Spatial 
Change During the Foreclosure Crisis. (Master of Community Planning), 
University of Maryland, College Park, College Park, MD.    
Bowen, E. A., & Irish, A. (2018). 'Hello, You're Not Supposed to Be Here': Homeless 
Emerging Adults' Experiences Negotiating Food Access. Public health 
nutrition, 21(10), 1943-1951. doi:10.1017/S1368980018000356 
Brooks, R., & Simon, R. (2007, December 3, 2007). Subprime Debacle Traps Even 
Very Credit-Worthy: As Housing Boomed, Industry Pushed Loans to a 
Broader Market. Wall Street Journal. 
Burt, M. R. (1992). Over the Edge: The Growth of Homelessness in the 1980s: 
Russell Sage Foundation. 
Burt, M. R., & Cohen, B. S. (1989). America?s Homeless: Numbers, Characteristics 
and Programs That Serve Them. Published in Washington, DC. 
Burt, M. R., Wilkins, C., Spellman, B., D'alanno, T., White, M., Henry, M., . . . Abt 
Associates Inc. (2016). Rapid Re-Housing for Homeless Families 
Demonstration Programs Evaluation Report: Part I: How They Worked---
Process Evaluation. Published in Washington, DC. 
Byrne, T., Munley, E. A., Fargo, J. D., Montgomery, A. E., & Culhane, D. P. (2013). 
New Perspectives on Community-Level Determinants of Homelessness. 
Journal of Urban Affairs, 35(5), 607-625.  
Calsyn, R. J., & Roades, L. A. (1994). Predictors of Past and Current Homelessness. 
Journal of Community Psychology, 22, 272-278.  
Children's Defense Fund. (1988). Vanishing Dreams: The Growing Economic Plight 
of America's Young Families. Published in Washington, DC. 
Collins, S. E., Malone, D. K., & Clifasefi, S. L. (2013). Housing Retention in Single-
Site Housing First for Chronically Homeless Individuals with Severe Alcohol 
Problems. American Journal of Public Health. 
doi:10.2105/AJPH.2013.301312 
325 
 
Collinson, R., & Reed, D. (2018). The Effects of Evictions on Low-Income 
Households. Published in  
Corinth, K. (2017). The Impact of Permanent Supportive Housing on Homeless 
Populations. Journal of Housing Economics, 35, 69-84. 
doi:https://doi.org/10.1016/j.jhe.2017.01.006 
Corinth, K., & Lucas, D. S. (2018). When Warm and Cold Don?t Mix: The 
Implications of Climate for the Determinants of Homelessness. Journal of 
Housing Economics, 41, 45-56. doi:https://doi.org/10.1016/j.jhe.2018.01.001 
Culhane, D. P., Lee, C. M., & Wachter, S. M. (1996). Where the Homeless Come 
From: A Study of the Prior Address Distribution of Families Admitted to 
Public Shelters in New York City and Philadelphia. Housing Policy Debate, 7, 
327-365.  
Culhane, D. P., & Metraux, S. (2008). Rearranging the Deck Chairs or Reallocating 
the Lifeboats? Homelessness Assistance and Its Alternatives. Journal of the 
American Planning Association, 74(1), 111-121. 
doi:10.1080/01944360701821618 
Culhane, D. P., Metraux, S., & Hadley, T. (2002). Public Service Reductions 
Associated with Placement of Homeless Persons with Severe Mental Illness in 
Supportive Housing. Housing Policy Debate, 13(1), 107-163. 
doi:10.1080/10511482.2002.9521437 
Desmond, M. (2016). Evicted: Poverty and Profit in the American City. New York, 
NY: Broadway Books. 
Desmond, M., Gromis, A., Edmonds, L., Hendrickson, J., Krywokulski, K., Leung, 
L., & Porton, A. (2018a). Eviction Lab Methodology Report: Version 1.0. 
Published in Princeton. 
Desmond, M., Gromis, A., Edmonds, L., Hendrickson, J., Krywokulski, K., Leung, 
L., & Porton, A. (2018b). Eviction Lab National Database: Version 1.0   
Retrieved from www.evictionlab.org. from Princeton University 
www.evictionlab.org 
326 
 
Early, D., & Olsen, E. (2002). Subsidized Housing, Emergency Shelters, and 
Homelessness: An Empirical Investigation Using Data from the 1990 Census. 
Advances in Economic Analysis & Policy, 2(1).  
Edin, K. (1992). Counting Chicago?s Homeless: An Assessment of the Census 
Bureau?s ?Street and Shelter Night?. Evaluation Review, 16(4), 365-375.  
Elliot, M., & Krivo, L. J. (1991). Structural Determinants of Homelessness in the 
United States. Social problems, 38, 113-131.  
Fargo, J. D., Munley, E. A., Byrne, T. H., Montgomery, A. E., & Culhane, D. P. 
(2013). Community-Level Characteristics Associated with Variation in Rates 
of Homelessness among Families and Single Adults. American Journal of 
Public Health, 103, S340-S347. doi:10.2105/AJPH.2013.301619 
Filer, R. K. (1990). What We Really Know About the Homeless. Wall Street Journal, 
10.  
Finkel, M., Henry, M., Matthews, N., Spellman, B., Culhane, D., & Abt Associates 
Inc. (2016). Rapid Re-Housing for Homeless Families Demonstration 
Programs Evaluation Report: Part Ii: Demonstration Findings---Outcomes 
Evaluation. Published in Washington, DC. 
Freeman, R. B., & Hall, B. (1987). Permanent Homelessness in America? Population 
Research and Policy Review, 6(1), 3-27. doi:10.1007/bf00124800 
Galster, G. C. (1987). The Ecology of Racial Discrimination in Housing: An 
Exploratory Model. Urban Affairs Quarterly, 23, 84-107.  
Goldstein, I., & Urevick-Ackelsberg, D. (2008). Subprime Lending, Mortgage 
Foreclosures and Race: How Far Have We Come and How Far Have We to 
Go? Published in Philadelphia, PA. 
Grimes, P. W., & Chressanthis, G. A. (1997). Assessing the Effect of Rent Control on 
Homelessness. Journal of Urban Economics, 41, 23-37. 
doi:10.1006/juec.1996.1085 
Grounded Solutions Network. (2018). Inclusionary Housing Database Map (Beta 
Version). Retrieved from: 
327 
 
https://gsn.maps.arcgis.com/apps/webappviewer/index.html?id=331f8a985a24
4e8fb6e6a2ad23731179 
Gulcur, L., Stefancic, A., Shinn, M., Tsemberis, S., & Fischer, S. (2003). Housing, 
Hospitalization, and Cost Outcomes for Homeless Individuals with 
Psychiatric Disabilities Participating in Continuum of Care and Housing First 
Programs. Journal of Community and Applied Psychology, 13, 171-186.  
Hanratty, M. (2017). Do Local Economic Conditions Affect Homelessness? Impact 
of Area Housing Market Factors, Unemployment, and Poverty on Community 
Homeless Rates. Housing Policy Debate, 27, 640-655. 
doi:10.1080/10511482.2017.1282885 
Hartman, C. (1986). The Housing Part of the Homelessness Problem. In E. L. Bassuk 
(Ed.), The Mental Health Needs of Homeless Persons (pp. 71-85). San 
Francisco: Jossey-Bass. 
Honig, M., & Filer, R. K. (1993). Causes of Intercity Variation in Homelessness. The 
American Economic Review, 83(1), 248-255.  
Hopper, K. (1992). Counting the Homeless: S-Night in New York. Evaluation 
Review, 16(4), 376-388.  
Hopper, K., & Hamberg, J. (1986). The Making of America's Homeless: From Skid 
Row to New Poor, 1945-1984. In R. G. Bratt, C. Hartman, & A. Myerson 
(Eds.), Critical Perspectives on Housing (pp. 12-40). Philadelphia: Temple 
University Press. 
Hudson, C. G. (1993). The Homeless of Massachusetts: An Analysis of the 1990 U.S. 
Census S-Night Data. New England Journal of Public Policy, 9(1), 79-100.  
Huey, L. (2010). False Security or Greater Social Inclusion? Exploring Perceptions of 
Cctv Use in Public and Private Spaces Accessed by the Homeless. The British 
Journal of Sociology, 61(1), 63-82. doi:10.1111/j.1468-4446.2009.01302.x 
Ibm. (2019). How Can I Incorporate Quadratic or Higher Order Terms for Covariates 
in a Linear Mixed Model?   Retrieved from 
https://www.ibm.com/support/pages/how-can-i-incorporate-quadratic-or-
higher-order-terms-covariates-linear-mixed-model 
328 
 
Jencks, C. (1994). The Homeless. Cambridge: Harvard University Press. 
Jones, R. E. (1983). Street People and Psychiatry: An Introduction. Hospital and 
Community Psychiatry, 34, 807-811.  
Karl, T. R., Jr., C. N. W., Young, P. J., & Wendland, W. M. (1986). A Model to 
Estimate the Time of Observation Bias Associated with Monthly Mean 
Maximum, Minimum and Mean Temperatures for the United States. Journal 
of Climate and Applied Meteorology, 25(2), 145-160. doi:10.1175/1520-
0450(1986)025<0145:amtett>2.0.co;2 
Katz, A. S., Zerger, S., & Hwang, S. W. (2017). Housing First the Conversation: 
Discourse, Policy and the Limits of the Possible. Critical Public Health, 
27(1), 139-147. doi:10.1080/09581596.2016.1167838 
Kertesz, S. G., & Johnson, G. (2017). Housing First: Lessons from the United States 
and Challenges for Australia Housing First. Australian Economic Review, 
50(2), 220-228. doi:10.1111/1467-8462.12217 
Lee, B. A., Price-Spratlen, T., & Kanan, J. W. (2003). Determinants of Homelessness 
in Metropolitan Areas. Journal of Urban Affairs, 25(3), 335-356.  
Lee, B. A., & Schreck, C. J. (2005). Danger on the Streets: Marginality and 
Victimization among Homeless People. American Behavioral Scientist, 48(8), 
1055-1081.  
Mahoney, P. E., & Zorn, P. M. (1996). The Promise of Automated Underwriting: 
Providing a Simpler, Fairer, More Inclusive Mortgage-Lending System. 
Published in Washington, DC. 
Mares, A., & Rosenheck, R. A. (2007). Hud/Hhs/Va Collaborative Initiative to Help 
End Chronic Homelessness: National Performance Outcomes Assessment 
Preliminary Client Outcomes Report, Washington, DC. 
Martin, E. (1992). Assessment of S-Night Street Enumeration in the 1990 Census. 
Evaluation Review, 16(4), 418-438.  
Massey, D. S., & Denton, N. A. (1993). American Apartheid : Segregation and the 
Making of the Underclass. Cambridge, Mass. :: Harvard University Press. 
329 
 
Massey, D. S., Rugh, J. S., Steil, J. P., & Albright, L. (2016). Riding the Stagecoach 
to Hell: A Qualitative Analysis of Racial Discrimination in Mortgage 
Lending. City & Community, 15(2), 118-136. doi:10.1111/cico.12179 
Meanwell, E. (2012). Experiencing Homelessness: A Review of Recent Literature. 
Sociology Compass, 6(1), 72-85.  
Menne, M. J., & Jr., C. N. W. (2009). Homogenization of Temperature Series Via 
Pairwise Comparisons. Journal of Climate, 22(7), 1700-1717. 
doi:10.1175/2008jcli2263.1 
M?ller, S., Scealy, J. L., & Welsh, A. H. (2013). Model Selection in Linear Mixed 
Models. Statistical Science, 28(2), 135-167.  
Munnell, A., Tootell, G. M. B., Browne, L., & Mceneaney, J. (1996). Mortgage 
Lending in Boston: Interpreting Hmda Data. The American Economic Review, 
86(1), 25-53.  
Mutchler, J. E., & Krivo, L. J. (1989). Availability and Affordability: Household 
Adaptation to a Housing Squeeze. Social Forces, 68, 241-261.  
O'Flaherty, B. (1996). Making Room : The Economics of Homelessness. Cambridge, 
Mass. :: Harvard University Press. 
O'Flaherty, B. (2004). Wrong Person and Wrong Place: For Homelessness, the 
Conjunction Is What Matters. Journal of Housing Economics, 13(1), 1-15.  
O'Flaherty, B. (2010). Homelessness as Bad Luck: Implications for Research and 
Policy. In I. G. Ellen & B. O?Flaherty (Eds.), How to House the Homeless (pp. 
143-182). New York, NY: Russell Sage Foundation. 
O'Flaherty, B. (2012). Homelessness in the United States. In N. Brooks, K. Donaghy, 
& G. Knaap (Eds.), The Oxford Handbook of Urban Economics and Planning. 
New York, NY: Oxford University Press. 
O'Flaherty, B. (2019). Homelessness Research: A Guide for Economists (and 
Friends). Journal of Housing Economics, 44, 1-25. 
doi:https://doi.org/10.1016/j.jhe.2019.01.003 
330 
 
Padgett, D., Henwood, B. F., & Tsemberis, S. J. (2016). Housing First: Ending 
Homelessness, Transforming Systems, and Changing Lives: Oxford University 
Press, USA. 
Pearson, C., Montgomery, A. E., & Locke, G. (2009). Housing Stability among 
Homeless Individuals with Serious Mental Illness Participating in Housing 
First Programs. Journal of Community Psychology. doi:10.1002/jcop.20303 
Quigley, J. M., & Portney, P. R. (1990). Does Rent Control Cause Homelessness? 
Taking the Claim Seriously. Journal of Policy Analysis & Management, 9, 89-
93.  
Quigley, J. M., Raphael, S., & Smolensky, E. (2001). Homeless in America, 
Homeless in California. The Review of Economics and Statistics, 83(1), 37-51.  
Redburn, F. S., & Buss, T. F. (1986). Responding to America's Homeless: Public 
Policy Alternatives. New York: Praeger. 
Ribisl, K. M., Walton, M. A., Mowbray, C. T., Luke, D. A., Davidson, W. S., & 
Bootsmiller, B. J. (1996). Minimizing Participant Attrition in Panel Studies 
through the Use of Effective Retention and Tracking Strategies: Review and 
Recommendations. Evaluation and Program Planning, 19(1), 1-25. 
doi:10.1016/0149-7189(95)00037-2 
Roschelle, A. R., & Kaufman, P. (2004). Fitting in and Fighting Back: Stigma 
Management Strategies among Homeless Kids. Symbolic Interaction, 27(1), 
23-46.  
Rossi, P. H. (1989). Down and out in America : The Origins of Homelessness. 
Chicago :: University of Chicago Press. 
Rossi, P. H., & Wright, J. D. (1987). The Determinants of Homelessness. Health 
Affairs, 6, 19-32.  
Roth, D., & Bean, G. J. J. (1986). New Perspectives on Homelessness: Findings from 
a Statewide Epidemiological Study. Hospital Community Psychiatry, 37, 712-
719.  
331 
 
Roth, D., Bean, J., Lust, N., & Saveanu, T. (1985). Homeless in Ohio: A Study of 
People in Need. Columbus, Ohio: Ohio Department of Mental Health. 
Rugh, J. S., & Massey, D. S. (2010). Racial Segregation and the American 
Foreclosure Crisis. American Sociological Review, 75(5), 629-651.  
Sadowski, L. S., Kee, R. A., Vanderweele, T. J., & Buchanan, D. (2009). Effect of a 
Housing and Case Management Program on Emergency Department Visits 
and Hospitalizations among Chronically Ill Homeless Adults: A Randomized 
Trial. JAMA, 301(17), 1771-1778. doi:10.1001/jama.2009.561 
Schafer, R., & Ladd, H. F. (1981). Discrimination in Mortgage Lending (H.-M. J. C. 
f. U. Studies Ed. Illustrated ed.). Cambridge, MA: MIT Press. 
Schwartz, A. F. (2010). Housing Policy in the United States (2nd ed. ed.). New York 
:: Routledge. 
Shlay, A. B., & Rossi, P. H. (1992). Social Science Research and Contemporary 
Studies of Homelessness. Annual Review of Sociology, 18, 129-160.  
Spss Inc. (2005). Linear Mixed-Effects Modeling in Spss: An Introduction to the 
Mixed Procedure. Published in Chicago, IL. 
Stefancic, A., Henwood, B. F., Melton, H., Shin, S. M., Lawrence-Gomez, R., & 
Tsemberis, S. (2013). Implementing Housing First in Rural Areas: Pathways 
Vermont. American Journal of Public Health, 103, 206-209.  
Stefancic, A., & Tsemberis, S. (2007). Housing First for Long-Term Shelter Dwellers 
with Psychiatric Disabilities in a Suburban County: A Four-Year Study of 
Housing Access and Retention. Journal of Primary Prevention. 
doi:10.1007/s10935-007-0093-9 
Stergiopoulos, V., Hwang, S. W., Gozdzik, A., Nisenbaum, R., Latimer, E., Rabouin, 
D., . . . Investigators, F. T. a. H. C. S. (2015). Effect of Scattered-Site Housing 
Using Rent Supplements and Intensive Case Management on Housing 
Stability among Homeless Adults with Mental Illness: A Randomized Trial. 
JAMA, 313(9), 905-915. doi:10.1001/jama.2015.1163 
332 
 
Troutman, W. H., Jackson, J. D., & Ekelund, R. B. J. (1999). Public Policy, Perverse 
Incentives, and the Homeless Problem. Public Choice, 98, 195-212. 
doi:10.2307/30024475 
Tsemberis, S. (2010). Housing First: Ending Homelessness, Promoting Recovery, and 
Reducing Costs. In I. G. Ellen & B. O?Flaherty (Eds.), How to House the 
Homeless (pp. 37-56): Russell Sage Foundation. 
Tsemberis, S., & Eisenberg, R. F. (2000). Pathways to Housing: Supported Housing 
for Street-Dwelling Homeless Individuals with Psychiatric Disabilities. 
Psychiatric Services, 51, 487-493.  
Tsemberis, S., Gulcur, L., & Nakae, M. (2004). Housing First, Consumer Choice, and 
Harm Reduction for Homeless Individuals with a Dual Diagnosis. American 
Journal of Public Health. doi:10.2105/AJPH.94.4.651 
Tsemberis, S., Kent, D., & Respress, C. (2012). Housing Stability and Recovery 
among Chronically Homeless Persons with Co-Occuring Disorders in 
Washington, Dc. American Journal of Public Health. 
doi:10.2105/AJPH.2011.300320 
Tucker, W. (1987, September 25). Where Do the Homeless Come From? The 
National Review, 32-43. 
United States Census Bureau. (2013-2017). American Community Survey 5-Year 
Estimates. Retrieved from: 
https://factfinder.census.gov/faces/nav/jsf/pages/index.xhtml 
United States Census Bureau. (2017). 2017 Tiger/Line Shapefiles [GIS Shapefiles]. 
Retrieved from: https://www.census.gov/cgi-
bin/geo/shapefiles/index.php?year=2017&layergroup=Counties+%28and+equ
ivalent%29 
United States Census Bureau. (2019). When to Use 1-Year, 3-Year, or 5-Year 
Estimates. American Community Survey (ACS).  Retrieved from 
https://www.census.gov/programs-surveys/acs/guidance/estimates.html 
United States Department of Housing and Urban Development. (1984). Report to the 
Secretary on the Homeless & Emergency Shelters. Published in Washington, 
DC. 
333 
 
United States Department of Housing and Urban Development. (1994). Continuum-
of-Care and Hopwa Applications. Published in Washington, DC. 
United States Department of Housing and Urban Development. (1999). Guide to 
Continuum of Care Planning and Implementation. Published in Washington, 
DC. 
United States Department of Housing and Urban Development. (2011). Homeless 
Emergency Assistance and Rapid Transition to Housing: Defining 
?Homeless.? Federal Register: Rules and Regulations, 76(233), 75994-
76019. Published in Washington, DC. 
United States Department of Housing and Urban Development. (2012). Introductory 
Guide to the Continuum of Care (Coc) Program. Published in Washington, 
DC. 
United States Department of Housing and Urban Development. (2013). Notice of 
Funding Availability (Nofa) for the Fiscal Years 2013 and 2014 Continuum of 
Care Program Competition. Published in Washington, DC. 
United States Department of Housing and Urban Development (Producer). (2014a, 
January 10, 2018). Continuum of Care Program Roadmap.  
United States Department of Housing and Urban Development. (2014b). Point-in-
Time Count Methodology Guide. Published in Washington, DC. 
United States Department of Housing and Urban Development. (2017a). 2017 Coc 
Gis Shapefiles [GIS Shapefiles]. Retrieved from: 
https://www.hudexchange.info/programs/coc/gis-
tools/?filter_ToolType=Tool&filter_Year=2017&filter_State=&program=Co
C&group=GIS 
United States Department of Housing and Urban Development. (2017b). Notice for 
Housing Inventory Count (Hic) and Point-in-Time (Pit) Data Collection for 
Continuum of Care (Coc) Program and the Emergency Solutions Grants (Esg) 
Program. (CPD-17-08). Washington, DC. 
United States Department of Housing and Urban Development. (2018a). The 2018 
Annual Homeless Assessment Report (Ahar) to Congress. Published in 
Washington, DC. 
334 
 
United States Department of Housing and Urban Development. (2018b). Pit and Hic 
Data since 2007. Retrieved from: 
https://www.hudexchange.info/resource/3031/pit-and-hic-data-since-2007/ 
United States Department of Housing and Urban Development. (2019a). Hud User 
Glossary.   Retrieved from 
https://archives.huduser.gov/portal/glossary/glossary_all.html 
United States Department of Housing and Urban Development. (2019b). Notice of 
Funding Availability (Nofa) for the Fiscal Year (Fy) 2019 Continuum of Care 
Program Competition. Published in Washington, DC. 
Vose, R. S., Applequist, S., Squires, M., Durre, I., Menne, M. J., Jr., C. N. W., . . . 
Arndt, D. (2014a). Improved Historical Temperature and Precipitation Time 
Series for U.S. Climate Divisions. Journal of Applied Meteorology and 
Climatology, 53(5), 1232-1251. doi:10.1175/jamc-d-13-0248.1 
Vose, R. S., Applequist, S., Squires, M., Durre, I., Menne, M. J., Jr., C. N. W., . . . 
Arndt, D. (2014b). Noaa's Gridded Climate Divisional Dataset (Climdiv). 
Retrieved from: ftp://ftp.ncdc.noaa.gov/pub/data/cirs/climdiv/ 
West, B. T. (2009). Analyzing Longitudinal Data with the Linear Mixed Models 
Procedure in Spss. Evaluation & the Health Professions, 32(3), 207-228.  
West, B. T., Welch, K. B., & Galecki, A. T. (2015). Linear Mixed Models: A 
Practical Guide Using Statistical Software (2 ed.). Boca Raton, FL: CRC 
Press, Taylor & Francis Group. 
Wienk, R. E., Reid, C. E., Simonson, J. C., & Eggers, F. J. (1979). Measuring Racial 
Discrmination in American Housing Markets. Published in Washington, DC. 
Write, J. D., & Devine, J. A. (1992). Counting the Homeless: The Census Bureau?s 
?S-Night? in Five U.S. Cities. Evaluation Review, 16(4), 355-364.  
Yurieff, K. (2019, March 28, 2019). Hud Charges Facebook with Housing 
Discrimination in Ads.   Retrieved from 
https://www.cnn.com/2019/03/28/tech/facebook-hud-ad-
discrimination/index.html 
 
335