ABSTRACT
Title of dissertation: THE PRICE SENSITIVITY OF DEMAND
FOR HIGHER EDUCATION
AMONG NON-TRADITIONAL STUDENTS
CHARLES BASCHNAGEL, 2015
Dissertation directed by: DR MELISSA KEARNEY
Department of ECONOMICS
This dissertation examines the price sensitivity of demand for higher education
among non-traditional students in the United States. Chapter 1 discusses the issues
related to the demand for higher education. It presents the recent trends and reviews
the literature addressing these issues. A major conclusion that emerges from this
chapter is that the price sensitivity of demand for higher education appears to
depend on the source of the variation in price and the characteristics of the students
who face the price change. The baseline estimate for the price sensitivity of demand
is that a $1,000 (in year 2000 dollars) decrease in tuition costs should result in
a 4 percentage-point increase in enrollment for the traditional 18- to 24-year-old
student.
Chapter 2 examines the price sensitivity of demand for higher education for
military spouses resulting from variation in tuition due to military-mandated moves
across states. The data suggest that a $1,000 (in year 2000 dollars) decrease in
the cost of 2-year schools is associated with a 1–1.5 percentage-point increase in
the probability of attending college. This estimate is less than half the previous
estimates due to in-state tuition price differences faced by the civilian 18- to 24-
year-old population on a percentage-point basis. However, this represents a 7–10
percent increase for this population, and the magnitude of this metric is in line with
previous estimates. This suggests tuition assistance can be an effective means of
increasing enrollment for military spouses, but other barriers to education for this
population may also need to be addressed.
Chapter 3 examines the impact of a change in the tax treatment of savings
set aside for higher education by those who decide to suspend their education and
enter the workforce. The taxation of these funds appears to have increased the rate
at which these funds are included in an employee’s initial contract and the quantity
of funds allocated. These results are counterintuitive if the tax preference was the
primary reason for the savings plan. However, these results suggest the rationale
for the savings plan was to offer targeted additional compensation to recruits with
greater negotiating power. Taxation of funds previously set aside did not appear
to have a statistically significant impact on their utilization. Point estimates of the
price sensitivity of demand from changes in the out-of-pocket costs for higher edu-
cation induced by the taxation of these funds were small and often not statistically
significant.
The results from this dissertation show responses to changes in the net cost of
college that differ by the source of price variation and the population experiencing
them. This is consistent with the previous literature. This dissertation contributes
to the literature by providing estimates for the price sensitivity of demand for higher
education to previously understudied non-traditional students.
THE PRICE SENSITIVITY OF DEMAND FOR HIGHER
EDUCATION AMONG NON-TRADITIONAL STUDENTS
by
CHARLES BASCHNAGEL
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
2015
Advisory Committee:
Professor of Economics MELISSA KEARNEY, Chair/Advisor
Emeritus Professor of Economics ROBERT SCHWAB
Associate Professor of Economics, Director of Graduate Studies JOHN SHEA
Assistant Professor of Economics LESLEY TURNER
Professor of Sociology JOAN KAHN, Dean’s Representative
c© Copyright by
CHARLES BASCHNAGEL
2015
Table of Contents
List of Tables v
List of Figures viii
List of Abbreviations viii
1 REVIEW OF ISSUES RELATED TO THE PRICE SENSITIVITY OF DE-
MAND FOR HIGHER EDUCATION IN THE UNITED STATES 4
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Trends in higher education . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Enrollment . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1.1 Increased Demand . . . . . . . . . . . . . . . . . . . 6
1.2.1.2 Supply Side Response . . . . . . . . . . . . . . . . . 7
1.2.2 Tuition Prices . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 Non-traditional Students . . . . . . . . . . . . . . . . . . . . . 13
1.3 Federal Aid Programs . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.1 GI Bills . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.2 Pell Grants . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3.3 Tax Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.3.4 Loans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.3.4.1 Repayment . . . . . . . . . . . . . . . . . . . . . . . 27
1.3.4.2 Data Availability . . . . . . . . . . . . . . . . . . . . 29
1.3.5 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4 Returns to Higher Education in the United States . . . . . . . . . . . 31
1.5 Price Sensitivity of Demand for Higher Education . . . . . . . . . . . 36
1.5.1 Changes in List Price . . . . . . . . . . . . . . . . . . . . . . . 37
1.5.2 Federal Programs . . . . . . . . . . . . . . . . . . . . . . . . . 39
1.5.2.1 Social Security . . . . . . . . . . . . . . . . . . . . . 39
1.5.2.2 GI Bills . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.5.2.3 Pell Grants . . . . . . . . . . . . . . . . . . . . . . . 40
1.5.2.4 Tax-Based Aid . . . . . . . . . . . . . . . . . . . . . 41
1.5.3 Non-Federal Programs . . . . . . . . . . . . . . . . . . . . . . 44
1.5.3.1 State Merit Aid Programs . . . . . . . . . . . . . . . 44
1.5.3.2 Place-Based Programs . . . . . . . . . . . . . . . . . 47
1.5.4 Other Sources of Variation . . . . . . . . . . . . . . . . . . . . 48
ii
1.5.4.1 Immigration Status . . . . . . . . . . . . . . . . . . . 48
1.5.5 Non-Monetary Sensitivities of Demand . . . . . . . . . . . . . 49
1.5.5.1 Credit Constraints . . . . . . . . . . . . . . . . . . . 53
1.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2 THE PRICE SENSITIVITY OF DEMAND FOR HIGHER EDUCATION
AMONG MILITARY SPOUSES 56
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.2 Background and Literature Review . . . . . . . . . . . . . . . . . . . 60
2.2.1 Previous Price Sensitivity Estimates . . . . . . . . . . . . . . 60
2.2.2 Tuition Prices Faced by Military Spouses . . . . . . . . . . . . 62
2.2.3 Characteristics of Military Spouses . . . . . . . . . . . . . . . 64
2.2.4 Distance to Nearest School . . . . . . . . . . . . . . . . . . . . 66
2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.3.1 Sample Restrictions . . . . . . . . . . . . . . . . . . . . . . . . 70
2.3.2 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4 Empirical Strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3 ESTIMATING THE IMPACT OF TAXING COLLEGE SAVINGS PLANS
OF NON-TRADITIONAL STUDENTS 91
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.2.1 Structure of Initial Contracts . . . . . . . . . . . . . . . . . . 94
3.2.2 Rationale for CSP program . . . . . . . . . . . . . . . . . . . 95
3.2.3 Major Changes in the Industry . . . . . . . . . . . . . . . . . 99
3.2.3.1 Changes to the Evaluation of Talent . . . . . . . . . 99
3.2.3.2 Collective Bargaining Agreements . . . . . . . . . . . 101
3.2.3.3 Changes to the Tax Treatment of CSP Funds . . . . 102
3.3 Data and Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . 104
3.3.1 Data Description . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.3.2 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . 105
3.4 Empirical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.4.1 Effect of Tax Change on Compensation Package . . . . . . . . 112
3.4.1.1 Sample Selection . . . . . . . . . . . . . . . . . . . . 112
3.4.1.2 Econometric Models . . . . . . . . . . . . . . . . . . 113
3.4.1.3 Estimation Results . . . . . . . . . . . . . . . . . . . 115
3.4.1.4 Heterogeneity of Responses to Tax . . . . . . . . . . 119
3.4.2 Effect of Tax on CSP Usage . . . . . . . . . . . . . . . . . . . 124
3.4.2.1 Sample Selection . . . . . . . . . . . . . . . . . . . . 124
3.4.2.2 Econometric Model . . . . . . . . . . . . . . . . . . . 127
3.4.2.3 Estimation Results . . . . . . . . . . . . . . . . . . . 130
3.4.2.4 Heterogeneity of Responses to Tax . . . . . . . . . . 132
3.4.3 Implied Price Sensitivity of Demand . . . . . . . . . . . . . . 133
iii
3.4.3.1 Sample Selection . . . . . . . . . . . . . . . . . . . . 135
3.4.3.2 Econometric Model . . . . . . . . . . . . . . . . . . . 136
3.4.3.3 Estimation Results . . . . . . . . . . . . . . . . . . . 138
3.4.3.4 Heterogeneity of Responses by Rank . . . . . . . . . 141
3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
B.6 Appendix B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
C.7 Appendix C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
Bibliography 179
iv
List of Tables
1.1 Percentage Distribution of Undergraduates, by Control and Level of
Institution and Selected Student Characteristics . . . . . . . . . . . . 15
2.1 Mean Price of Tuition and Fees for Military Spouses . . . . . . . . . . 71
2.2 Summary Statistics for 18- to 24-year-old Military Spouses . . . . . . 74
2.3 Summary Statistics for 25- to 40-year-old Military Spouses . . . . . . 76
2.4 Price Sensitivity of Military Spouses for Price Paid for College Tu-
ition, 18- to 40-year-olds . . . . . . . . . . . . . . . . . . . . . . . . . 81
2.5 Price Sensitivity of Military Spouses for Price Paid for College Tu-
ition, by Age Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
2.6 Price Sensitivity of Military Spouses for Price Paid for College Tu-
ition, by Educational Attainment . . . . . . . . . . . . . . . . . . . . 86
3.1 Summary Statistics for All Employees by Tax Status . . . . . . . . . 106
3.2 Summary Statistics for Employees With CSPs by Tax Status . . . . . 108
3.3 OLS Regressions of Compensation Package Characteristics on Knowl-
edge of Tax at Signing, 2004-2011 . . . . . . . . . . . . . . . . . . . . 116
3.4 OLS Regressions of Compensation Package Characteristics on Knowl-
edge of Tax at Signing Interacted with Rank, 2004-2011 . . . . . . . . 121
3.5 OLS Regressions of Compensation Package Characteristics on Knowl-
edge of Tax at Signing by Rank Group, 2004-2011 . . . . . . . . . . . 123
3.6 Linear Probability Model of CSP Usage on Tax Status of Funds,
2004-2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
3.7 OLS Regressions Estimating the Sensitivity of Demand, 2004-2011 . 140
3.8 OLS Regressions Estimating the Sensitivity of Demand by Rank
Group, 2004-2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
B.9 Price Paid by Military Spouses at 2-Year Public Institutions . . . . . 146
B.10 Published Price for Military Spouses at 2-Year Public Institutions . . 147
B.11 Price Paid by Military Spouses at 4-Year Public Institutions . . . . . 148
B.12 Published Price for Military Spouses at 4-Year Public Institutions . . 149
B.13 Military Spouses Exempted From Residency Requirements (2000-
2008) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
v
B.14 Effects of Average Price Paid on Attendance of 18- to 40-year-old
Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
B.15 Effects of Average Price Paid on Attendance of 18- to 24-year-old
Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
B.16 Effects of Average Price Paid on Attendance of 25- to 40-year-old
Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
B.17 Effects of Average Price Paid on Attendance of 18- to 40-year-old
Military Spouses Who Do Not Have an Associate’s Degree . . . . . . 155
B.18 Effects of Average Price Paid on Attendance of 18- to 40-year-old
Military Spouses Who Have an Associate’s Degree . . . . . . . . . . . 156
B.19 Effects of Average Published Price on Attendance of 18- to 40-year-
old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
B.20 Effects of Average Published Price on Attendance of 18- to 24-year-
old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
B.21 Effects of Average Published Price on Attendance of 25- to 40-year-
old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
B.22 Effects of Average Published Price on Attendance of 18- to 40-year-
old Military Spouses Who Do Not Have an Associate’s Degree . . . . 160
B.23 Effects of Average Published Price on Attendance of 18- to 40-year-
old Military Spouses Who Have an Associate’s Degree . . . . . . . . . 161
B.24 Effects of Log of Average Price Paid on Attendance of 18- to 40-year-
old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
B.25 Effects of Log of Average Price Paid on Attendance of 18- to 24-year-
old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
B.26 Effects of Log of Average Price Paid on Attendance of 25- to 40-year-
old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
B.27 Effects of Log of Average Price Paid on Attendance of 18- to 40-year-
old Military Spouses Who Do Not Have an Associate’s Degree . . . . 165
B.28 Effects of Log of Average Price Paid on Attendance of 18- to 40-year-
old Military Spouses Who Have an Associate’s Degree . . . . . . . . . 166
B.29 Effects of Log of Average Published Price on Attendance of 18- to
40-year-old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . 167
B.30 Effects of Log of Average Published Price on Attendance of 18- to
24-year-old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . 168
B.31 Effects of Log of Average Published Price on Attendance of 25- to
40-year-old Military Spouses . . . . . . . . . . . . . . . . . . . . . . . 169
B.32 Effects of Log of Average Published Price on Attendance of 18- to
40-year-old Military Spouses Who Do Not Have an Associate’s Degree 170
B.33 Effects of Log of Average Published Price on Attendance of 18- to
40-year-old Military Spouses Who Have an Associate’s Degree . . . . 171
C.34 Summary Statistics for All Employees by Year Recruited . . . . . . . 172
C.35 Summary Statistics for Employees With CSPs by Year Recruited . . 173
C.36 Summary Statistics for Employees in Rank Group 1 by Tax Status . . 173
C.37 Summary Statistics for Employees in Rank Group 2 by Tax Status . . 174
C.38 Summary Statistics for Employees in Rank Group 3 by Tax Status . . 174
vi
C.39 Summary Statistics for Employees in Rank Group 4 by Tax Status . . 174
C.40 Summary Statistics for Employees With CSPs in Rank Group 1 by
Tax Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
C.41 Summary Statistics for Employees With CSPs in Rank Group 2 by
Tax Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
C.42 Summary Statistics for Employees With CSPs in Rank Group 3 by
Tax Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
C.43 Summary Statistics for Employees With CSPs in Rank Group 4 by
Tax Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
vii
List of Figures
1.1 Increases in College Costs by Sector . . . . . . . . . . . . . . . . . . 11
1.2 Net Tuition, Fees, Room, and Board by Sector . . . . . . . . . . . . 12
1.3 Ratio of Male to Female College Enrollment in the U.S. Over Time . 19
3.1 Trends in Recruitment of and CSP Rates for High School Seniors . . 101
3.2 Histogram of the Difference Between the Date CSP Funds Were First
Used and an Employee’s Termination Date . . . . . . . . . . . . . . . 125
B.3 In-State Tuition 2000-2008 . . . . . . . . . . . . . . . . . . . . . . . . 150
B.4 Mean Cost of Tuition and Fees Paid (2008) . . . . . . . . . . . . . . . 151
B.5 Enrollment Rates of Military Spouses (2008) . . . . . . . . . . . . . . 151
C.6 Trends in CSP Contracts With Break Lines in 2005 and 2007 . . . . . 178
C.7 Histograms of Alternative Cut-offs . . . . . . . . . . . . . . . . . . . 178
viii
List of Abbreviations
ACS American Community Survey
AJCA American Jobs Creation Act of 2004, Public Law 108-357
AOTC American Opportunity Tax Credit
AP Advanced Placement
CBA Collective Bargaining Agreement
CPS Current Population Survey
CSP College Scholarship Plan
DCTAG District of Columbia’s Tuition Assistance Grant Program
FAFSA Free Application for Federal Student Aid
FTE full-time equivalent
HOPE Georgia Helping Outstanding Pupils Educationally
HTC Hope Learning Credit
IPEDS Integrated Postsecondary Education Data System
LLTC Lifetime Learning Tax Credit
LPM Linear Probability Model
MyCAA Military Spouse Career Advancement Accounts
NLSY National Longitudinal Survey of Youth
OLS Ordinary Least Squares
PLUS Parent Loans for Undergraduate Students
SIPP Survey of Income and Program Participation
YTV Year Terminated Variable
ix
Introduction
Chapter 1 focuses on reviewing the issues surrounding the demand for higher
education. This includes presenting recent trends in enrollment and price, an
overview of the current situation and understanding of non-traditional students,
what is known about the returns to higher education, and what is known about
how demand is influenced by price. This chapter presents the previous literature
necessary to understand the contribution of the next two chapters. One key result
of the previous literature is the baseline estimate for the price sensitivity of demand
for higher education. Although this estimate differs by the source of variation in
price, a $1,000 (in year 2000 dollars) decrease in tuition costs is associated with
a 4 percentage-point increase in enrollment for the traditional 18- to 24-year-old
student. Given the enrollment rate of this population, this translates into roughly
a 9 percent increase.
Chapter 2 examines the price sensitivity of demand for higher education of
military spouses, a subpopulation whose price sensitivity has not been examined
previously. However, this group has been the focus of several recent initiatives to
reduce their higher education costs. By exploiting the variation in price caused
by military-mandated moves, this chapter finds that military spouses’ college at-
tendance rates respond in a statistically significant way to changes in the cost of
1
2-year public colleges. The military-mandated moves allow for a unique identifica-
tion strategy using differences in tuition rates between states to estimate the price
sensitivity of demand for higher education. The data suggest that a $1,000 (in
year 2000 dollars) decrease in the cost of 2-year schools is associated with a 1–1.5
percentage-point increase in the probability of attending school. This estimate is
less than half the previous estimates for a percentage-point based on the civilian 18-
to 24-year-old population. However, the population studied had a much lower base-
line rate of attendance. The percentage-point increase in attendance translates into
a 7 to 10 percent increase, which is in line with previous estimates. These results
provide an initial estimate for the price sensitivity of demand for higher education
of military spouses, but given the unique circumstances military families face, this
estimate likely does not have external validity to non-military spouses.
Chapter 3 examines the question: Do tax breaks for higher education increase
college attendance? Settling this question empirically is difficult because eligibility
for tax breaks tends to be correlated with other factors that influence education
investment decisions. An exogenous shift in tax policy affecting the after-tax price
of education could help researchers address this question, and in late 2004, such a
change in U.S. tax policy occurred. Data were obtained from firms with a long his-
tory of offering delayed compensation in the form of education benefits as part of an
initial contract’s signing bonus. This chapter examines both the effect of the tax on
new savings for college and the use of existing funds following an unexpected change
in the tax status of these funds. The analysis indicates that the taxation of funds
that can only be used to pay for education increased the amount of new money set
2
aside on both a total and a per semester basis. The change in the treatment of new
funds from untaxed to taxed also appears to have increased the share of employees
who receive funds earmarked for education as part of their initial contract. These
results are most consistent with the economic rationale of these funds either being
used by employees as a commitment mechanism to return to school or as a means
for an employer to selectively offer additional compensation to potential employees
with greater negotiating power. This is in contrast to alternative rationales, where
the education benefits are primarily used as an insurance policy against being un-
successful in the industry or as a vehicle for tax avoidance. The analysis of the
unexpected tax on the use of funds previously set aside for education finds no sta-
tistically significant effect. Finally, estimates of the price sensitivity of demand for
higher education of this population are mixed, finding at most a 1 percentage-point
decrease in matriculation occurring for each $1,000 (in year 2000 dollars) increase
in out-of-pocket costs caused by the tax change, with most specifications and sam-
ples finding no statistically significant results. This finding falls in line with recent
estimates examining the impact of tax credits for higher education. Initial work
suggested sensitivities of around a 2 to 4 percentage-point response to $1,000 (in
year 2000 dollars) of tax-based aid for certain groups, but recent work with more
detailed data has found almost no impact on college enrollment from the expansion
of the higher education tax credits. A tax incentive for those returning to school
does not seem to have as much of an impact as an equivalent value change in tuition
costs for those attending college immediately after finishing high school.
3
Chapter 1: REVIEW OF ISSUES RELATED TO THE PRICE SEN-
SITIVITY OF DEMAND FOR HIGHER EDUCATION
IN THE UNITED STATES
1.1 Introduction
A student’s willingness to pay for higher education is determined by the asso-
ciated costs and benefits of their competing options as well as the resources available
to the student. The primary financial benefit of a college degree is the associated in-
crease in earnings from attending and/or completing college. This chapter reviews
the literature estimating the impact of higher education on wages and how that
impact has changed over time. The costs of a college degree include both the out-
of-pocket expenses and the opportunity cost of a student’s time in school when she
could otherwise be working and earning a wage. Meanwhile, the price of a college
education is driven by the supply and demand for higher education as well as insti-
tutional supports such as subsidies from state governments or direct aid and loans
from the federal government. This chapter begins with a review of historic trends
in these primary factors of demand, supply, and cost. It then provides a discussion
of what it means to be a non-traditional student and a snapshot of the distribution
4
of students across different schools by select characteristics usually used to classify
students as traditional or non-traditional.
After chronicling the changing costs and benefits of higher education over
time, this chapter then reviews the estimates for the price sensitivity of demand for
higher education that result, in part, from these factors. This section is organized by
the source of the variation in price from which the sensitivity was estimated. This
organization helps point out that the source of price variation appears to matter.
Changes in price also appear to affect various subpopulations differently. These
observations put the contributions of the second and third chapters, which focus
on specific subpopulations of non-traditional students, in context. This overview
presents (1) trends in total enrollment and costs, (2) a snapshot of non-traditional
students, (3) an overview of federal aid, and (4) a brief discussion on the returns to
higher education in the United States. The chapter then reviews (5) estimates of
the price sensitivity of demand for higher education.
1.2 Trends in higher education
1.2.1 Enrollment
As the fraction of the population attending college increases, the composition
of the average and marginal student can change. This new pool of students may
have different preferences for and returns to higher education. This section reports
on how the number and composition of those attending college has changed, how
the suppliers of higher education have responded, and how the resulting price has
5
fluctuated over time.
1.2.1.1 Increased Demand
During the early 1980s, some university administrators were concerned with
how they would maintain enrollment during a period when the college-aged popu-
lation was on the decline, and academics predicted severe contraction in enrollment
(Dresch, 1983).1 But those days are long past. The total number of students attend-
ing college in the United States has been increasing for decades, and this increase
has recently accelerated. The U.S. Department of Education, National Center for
Education Statistics (2013), reported that enrollment in degree-granting institutions
increased by 11 percent between 1991 and 2001 and then increased by 32 percent
from 2001 to 2011. This later increase from 15.9 million to 21.0 million was driven by
both a slight increase in the traditional college-aged population of 18- to 24-year-olds
and an increase in the share of the relevant population attending college.
Aside from the 21 million enrolled in accredited 2-year colleges, 4-year col-
leges, and universities in 2011, roughly 572,000 students also attended non-degree-
granting, Title IV eligible post-secondary institutions.2 These statistics come from
reporting requirements as part of participation in Title IV and are thus not read-
ily available for non-Title IV institutions. However, some estimates put the size of
1An increase in the fraction of the population attending college helped maintain the enrollment
numbers when the college-aged population contracted. Despite this increase in the share of the
population attending college, a short decrease in enrollment occurred from 10.8 million in 1983 to
10.6 million in 1985.
2These are predominantly organized to educate students for occupational, trade, and technical
careers, and include institutions that offer programs via distance education. These are typically
certificate programs.
6
the student body at non-Title IV eligible institutions at as large as a third of the
for-profit sector (Cellini & Goldin, 2014).
The increase in the share of the traditional and non-traditional college-aged
population attending college has been the main factor in increasing total enrollment,
as opposed to population growth in these age categories. The percentage of 18- to
24-year-olds enrolled in college rose from 25 percent in 1980 to 36 percent in 2001,
and then to 42 percent in 2011. The increase in the enrollment rate was not limited
to the traditional college-aged population. From 2000 to 2011, the enrollment rate
of students age 25 and over increased faster than the rate at which 18- to 24-year-
olds increased enrollment. This increase in the share of the population enrolling in
post-secondary institutions is frequently attributed to an increase in the perceived
returns to education in the United States spurring demand. The increase in the
college wage premium is discussed in more detail later in this chapter, but now I
turn to the supply-side response to this increase in demand for higher education.
1.2.1.2 Supply Side Response
Only a fraction of the revenue used to maintain most non-profit post-secondary
institutions comes from tuition and fees charged to students.3 Public institutions
rely heavily on subsidies from state and local governments, and these subsidies have
not kept pace with the recent increases in the demand for education. Also, private
institutions have not necessarily seen a corresponding increase in their non-tuition
3For public 2- and 4-year institutions in the 2012-13 academic year the percentage of revenue
from tuition and fees was 17 percent and 22 percent, respectively (NCES, 2015).
7
funding sources. The response to this shortfall of necessary funds has differed by
sector and the selectivity of an institution.
As the number of college applicants has grown in recent years, the number
of positions available at American 4-year colleges or universities has not kept pace.
Bound, Hershbein, and Long (2009) showed that the number of applicants to 4-year
schools increased by 44 percent, from 1.19 million in 1992 to 1.71 million in 2004,
while the size and number of these schools did not increase proportionally. Instead
of increasing quantity to meet demand, selective non-profit 4-year colleges became
more selective in whom they admit. As a result, high school seniors responded
by improving their credentials. Over the last 20 years, high school students have
become more likely to study calculus, take Advanced Placement (AP) courses, and
spend resources to prepare for and take the ACT or SAT (Bound, Hershbein, &
Long, 2009). This competition makes sense given that the most selective schools,
particularly those in the top 1 percent, spend considerably more per student while
also providing the largest average subsidy. Students at the top resource-intensive
schools end up paying on average the lowest share of student expenditures via tuition
and receive by far the best deal in higher education (Hoxby, 2009).
As selective, resource-intensive institutions increased their selectivity instead
of increasing their enrollment enough to keep up proportionally with demand, a
higher share of post-secondary students were pushed into less resource-intensive
institutions (Hoxby, 2009). These lower average resources per student have been
blamed for the decrease in the college completion rate (Bound & Turner, 2007;
Bound, Lovenheim, & Turner, 2010). Decreased college preparation also has been
8
cited as a reason for some of the decrease in college completion in these studies, but
resources per student has been acknowledged as being the dominant factor. It also
appears that male students may be more sensitive to the resource-intensity issue,
as a corresponding decrease in female students’ completion rate from attending less
resource-intensive institutions was not found (Bound, Lovenheim, & Turner, 2010).
This segment of the literature suggests that supply may not be able to respond
fully to changes in demand. This contradicts the assumption of perfectly elastic
supply often used in modeling the demand for higher education. With an upward-
sloping supply curve, one would also expect to see an increase in price as demand
increases. In fact, the increase in the cost of higher education is the second major
trend discussed in this section.
1.2.2 Tuition Prices
The increase in the real cost of college tuition and fees in the United States
is a well-known and well-documented phenomenon. Kane (1999) attributed the rise
in tuition prices to increases in the cost of the components to producing education,
especially the labor involved, and a reduction in the real subsidies provided by state
governments to public institutions of higher education. The reason for inflation ap-
peared to depend in part on sector. Ehrenberg (2012) showed that tuition increases
at private 4-year institutions were associated with higher spending per pupil, while
at public institutions the tuition increases were attempts to maintain the level of
expenditure per pupil in the face of decreasing state support. Meanwhile, others
9
critiqued the increase in the number of non-faculty staff and other potential rent-
seeking behavior as a reason for the increase in costs (Vedder, 2004). One such
rent-seeking behavior would be the capture of state or federal student aid by the
providers of higher education.4 Cellini and Goldin (2014) found that tuition is 78
percent higher at aid-eligible, sub-baccalaureate for-profit programs than at similar
non-aid-eligible for-profit institutions. This suggests there may be some rent-seeking
behavior in the form of aid capture, at least in the for-profit sector of higher educa-
tion, that is raising tuition costs.
It is worth noting that college costs have not increased uniformly by sector.
Figure 1.1 shows how the costs of tuition and fees have changed over time for public
4-year, public 2-year, and private non-profit 4-year institutions separately using the
1990-1991 academic year as the base year for each sector. The figure shows that 4-
year public institutions, which are both more reliant on state and local government
for funding than private non-profit schools and more resource-intensive then their
2-year public counterparts, have increased the most in price relative to their 1990-91
baseline.
A number of studies have documented how increases in the tuition and fees at
public institutions of higher education are correlated with decreases in the enroll-
ment rate, holding all else equal (Leslie & Brinkman, 1988; Kane, 1994; Cameron
& Heckman, 1998; Kane, 1999). However, as the overall number and percentage of
students enrolled in college continues to increase, the question as to the social costs
of these higher prices turns to a discussion of either the distributional impacts of
4This is often referred to as the Bennett hypothesis.
10
Figure 1.1: Increases in College Costs by Sector
.
7
1
1.
3
1.
6
1.
9
2.
2
2.
5
2.
8
Pr
ic
es
 R
el
at
iv
e 
to
 1
99
0−
91
1990 1995 2000 2005 2010 2015
Year
Public Two−Year Public Four−Year
Prviate Nonprofit Four−Year
SOURCES: The College Board, Annual Survey of Colleges; NCES, IPEDS.
NOTES: This figure shows published tuition and fees by sector, relative to 1990−91, adjusting for 
inflation using the Consumer Price Index. Average tuition and fee prices reflect in−district 
charges for public two−year institutions and in−state charges for public four−year institutions.
these increases in price or how much higher the college attendance and completion
rate would be in the absence of these increases in price. On the issue of distributional
effects, Kane (1995) found that tuition price increases disproportionately affect the
educational attainment of individuals from low-income families regardless of race.
The rise in the sticker price is often covered in the popular press, but it is
the net price after accounting for government and institutional aid that reflects true
out-of-pocket costs and a student’s ability or inability to pay. Figure 1.2 illustrates
the trend in net college costs by sector. Net total costs at public 4-year schools
have increased steadily since 2000.5 Meanwhile, the average net price of public
2-year schools has actually decreased over the last decade. The recent expansion
5Net tuition and fees at public 4-year schools held steady over the period from 1994 to 2015,
but rising room and board costs have helped push up total net costs.
11
Figure 1.2: Net Tuition, Fees, Room, and Board by Sector
50
00
10
00
0
15
00
0
20
00
0
25
00
0
Ye
ar
 2
01
4 
D
ol
la
rs
1990 1995 2000 2005 2010 2015
Year
Public Two−Year Public Four−Year
Prviate Nonprofit Four−Year
SOURCES: The College Board, Annual Survey of Colleges; NCES, IPEDS.
NOTES: This figure shows net tuition, fees, room, and board by sector, adjusted for inflation,
using the Consumer Price Index. Average tuition and fee prices reflect in−district charges for 
public two−year institutions and in−state charges for public four−year institutions.
of Pell Grant aid appears to have been particularly helpful to holding down net
costs at these schools. Both Pell Grant aid and increases in institutional aid at
private, non-profit 4-year institutions have helped push down the real net price of
these institutions beginning in 2008, after the net price rose in real terms for over
a decade. However, data from the last couple of years show a return to increasing
real net costs for this sector.
Some might find it curious that certain non-profit 4-year schools would have
both rising tuition and lower net costs due in part to increases in institutional
aid. Turner (2013) argued that redistribution may be one possible benefit of higher
tuition prices if large institutions are able to use the additional funds from tuition
paid by students from high-income families to fund more need-based institutional
12
aid to students from low-income families. She warned that this strategy can only
be effective if the colleges and universities can effectively communicate the lower
net cost to low-income students from this high-tuition/high-aid strategy. However,
recent research has shown that prospective students often overestimate the net price
of college they will face; this is particularly true for disadvantaged students (Scott-
Clayton, 2013). This calls into question the potential effectiveness of this strategy.
1.2.3 Non-traditional Students
The empirical chapters of this dissertation examine the price sensitivity of
demand on two sets of non-traditional students: military spouses and individuals
who start a career before potentially returning to school. These two populations
may be included in a broad definition of a non-traditional student. To clarify the
definition, I will begin with a definition of a “traditional” undergraduate.
The term “traditional” is typically used to refer to an undergraduate who has
a high school diploma (as opposed to a GED), enrolls full-time immediately after fin-
ishing high school, depends on his parents for financial support, and does not work
full-time during the school year. The research on the price sensitivity for higher
education has historically focused on the behavior of traditional undergraduates.
However, only 27 percent of undergraduates met all of these criteria in 1999-2000
(U.S. Department of Education, 2002).6 Therefore, extending the estimation of the
price sensitivity of demand for higher education to various non-traditional student
6This includes undergraduates at all types of post-secondary institutions (less-than-2-year, 2-
year, and 4-year).
13
populations, as this dissertation does, is necessary to fully understand the cur-
rent higher education landscape. As new legislation targets incentives at particular
segments of the non-traditional student population, such as military spouses, it is
important to have price sensitivity estimates for these subpopulations to accurately
forecast the impact of these proposals.
The definition of a “non-traditional” student varies (Bean & Metzner, 1985;
Horn, 1996; Schuetze & Slowey, 2002). However, older, non-continuously educated,
independent and/or married, non-resident, and previously or currently in the labor
force full-time are common defining characteristics for non-traditional students and
subsets of these characteristics apply to each of the samples in the following two
chapters. The second chapter consists of a married, non-resident, population and
the third chapter focuses on a non-continuously educated population. Horn (1996)
defined a non-traditional student as one who has any of the following characteristics:
Delays enrollment;
Attends part-time for at least part of the academic year;
Works full-time (35 hours or more per week) while enrolled;
Is considered financially independent for financial aid eligibility purposes;
Has dependents;
Is married; or
Does not have a high school diploma.
Horn (1996) defined students as “minimally non-traditional” if they have only
one non-traditional characteristic, “moderately non-traditional” if they have two or
three, and “highly non-traditional” if they have four or more. The samples examined
14
in the empirical portion of this dissertation would be classified as moderately to
highly non-traditional according to Horn’s scale. Also of note is that each of the
populations examined tend to be dominated by one gender, with the dominant
gender switching between the two chapters. As the price paid and the educational
outcomes achieved are correlated with the sector of higher education attended, Table
1.1 presents a snapshot of the percent of the student population attending each type
of school by gender and key characteristics that help define the samples in the later
chapters as non-traditional. This information is for the 2007-08 academic year, which
is situated in the middle of the period of observation of the empirical analyses.
Table 1.1: Percentage Distribution of Undergraduates, by Control and Level of Institution
and Selected Student Characteristics
Public Private nonprofit For Profit More
Less-than- Less-than- Less-than- 2-years than one
2-year 2-year 4-year 4-year 4-year 2-year or more institution
Sex
Male 0.4 40.6 31.7 0.4 13.3 1.1 5.3 7.3
Female 0.5 39.5 27.4 0.4 12.8 2.7 8.2 8.5
Dependency Status
Dependent 0.2 32.3 38.2 0.2 16.3 1.3 2.8 8.7
Independent 0.6 48.6 19.2 0.6 9.3 2.9 11.6 7.1
Unmarried, no dependents 0.5 45.1 25.6 0.6 9.2 2.2 9.0 7.8
Married, no dependents 0.6 49.1 21.0 0.4 10.3 1.8 9.1 7.6
Unmarried, with dependents 0.8 49.1 14.2 0.7 7.8 4.9 16.2 6.5
Married, with dependents 0.7 51.9 15.5 0.6 10.3 2.6 11.8 6.6
Age
18 years or younger 0.2 37.6 33.0 0.2 17.5 1.8 2.9 6.8
19 to 23 years 0.3 32.9 36.8 0.3 14.8 1.7 4.1 9.1
24 to 29 years 0.5 43.3 24.4 0.7 8.7 2.9 11.7 7.7
30 to 39 years 0.7 49.8 16.0 0.7 10.5 2.6 13.2 6.5
40 years or older 0.9 58.4 13.4 0.6 10.2 1.8 8.7 5.9
NOTE: Estimates include students enrolled in Title IV eligible post-secondary institutions in the 50 states, the District of Columbia, and
Puerto Rico.
SOURCE: U.S. Department of Education, National Center for Education Statistics, 2010. Table 1.3. Percentage distribution of undergraduates,
by control and level of institution and selected student characteristics: 2007-08.
Table 1.1 shows that men are more likely to attend public colleges, while
women are slightly more likely to attend for-profit schools. Meanwhile, dependent
students attend public and private nonprofit 4-year colleges at a higher rate, while
independent students, including married independent students, are more likely to
15
attend 2-year schools and for-profit schools. A similar pattern is seen when looking
at the distribution of undergraduates by age. Students 18 to 24 years old are more
likely to attend 4-year schools, while 24- to 40-year-olds attend 2-year public colleges
and for-profit colleges at higher rates than their younger counterparts.
1.3 Federal Aid Programs
1.3.1 GI Bills
The original GI Bill is often credited with kick-starting the golden age of federal
funding for higher education. The mid-century GI Bills created the largest direct
aid program in the history of American higher education. Meanwhile, the Post-9/11
GI Bill provided benefits to over 790,000 recipients in the 2013-2014 academic year,
with an average benefit per recipient of $14,107 (College Board, 2014).
The GI Bills have been credited with increasing the post-secondary educa-
tional attainment of their corresponding cohorts. The precise effects of the financial
incentives, however, have been difficult to measure (Bound & Turner, 2002; Staley,
2003).7 The reason it is hard to extract a sensitivity or an elasticity of demand
estimate from the tuition subsidies is that it is difficult to separate the effect of the
subsidy from the effect of military service. For example, while the financial effect
may be thought to dominate for the WWII and Korean War cohorts, research has
shown that the Vietnam War increased educational attainment among men through
both the process of draft avoidance during the war and utilization of GI benefits after
7Stanley (2003) attributed the WWII and Korean War GI Bills with increasing post-secondary
attainment of eligible cohorts of males by 15 to 20 percent.
16
the war. Card and Lemieux (2001) estimated that draft avoidance raised college en-
rollment by 4 to 6 percentage-points and college completion by 2 percentage-points
for men born in the 1940s. Bound and Turner (2003) found that the benefits of the
GI Bill did not appear to accrue evenly to everyone. Although white men across the
country and African-American men outside of the South saw gains in educational
attainment associated with the GI Bill, African-American veterans residing in the
South saw little improvement. The authors attributed this to a lack of educational
choices available to African-American men in that region of the country.
Although parsing out the effect of military service from the subsidy on edu-
cational attainment may prove difficult, other research has looked at the combined
effect of military service and the GI Bill on educational attainment to estimate
the effect of increased educational attainment on the returns to education. Angrist
(1993) utilized the 1987 Survey of Veterans to identify the effect of GI benefits on
education and earnings. Data from the survey allowed him to see the educational at-
tainment of individuals both before and after service. This enabled him to estimate
that educational benefits led to an increase of 1.4 years of schooling, on average.
Furthermore, the returns to education were estimated at 4.3 percent per year or 6
percent per veteran who used their educational benefits.8 Angrist also found that
this increase in earnings was limited to those who attended college or graduate
school and was not caused by enrollment in non-degree granting training programs.
Bennett, Lucchesi, and Vedder (2010) offered a possible explanation for the lack of
8In an update using 2000 census data, Angrist and Chen (2011) found returns to a year of
school instigated by the Vietnam-era GI Bill to be 7 percent.
17
return on training. They discussed the history of for-profit schools, which dominated
the educational training landscape at the time, and mentioned how these schools
were the subject of at least five federal reports investigating widespread accusations
of fraud.
The early GI Bills are also credited with temporarily counteracting the trend
of an increasing share of female enrollment in higher education. Figure 1.3 illustrates
the ratio of male to female college enrollment totals for each academic year. The
ratio fell from 2.5 in the late 1940s to below 1 by 1979. The ratio then continued
to drop smoothly and level off at around 0.75 by the early 2000s. The unevenness
in the early decline corresponds to cohorts where men came of age during times of
war in the United States and later qualified for GI college benefits. Thus, the GI
Bill has been credited with causing these deviations from the overall trend (Card &
Lemieux, 2001).
There are three reasons why more recent and future iterations of the GI Bill
might not have as significant an impact on the gender ratio at American colleges.
First, the share of the population participating in the military during the Iraq and
Afghanistan conflicts has not been as high as during past conflicts.9 Second, women
have increased as a share of the armed forces and, although still a minority of the
service, are also eligible for GI benefits.10 Finally, the Post-9/11 GI Bill allows for
the transfer of educational benefits from a member of the armed services to their
9There were over 25 million active duty military personnel in 1945 and roughly 3 million in
1955 and 1970, but less than 1.5 million in 1996. This number has not crossed that threshold since,
despite the wars in Iraq and Afghanistan and the rise in the U.S. population from 140 million in
1945 to over 300 million today (Sources: U.S. Census and U.S. Department of Defense).
10Women made up less than 2.3 percent of enlisted personnel in 1973 but made up over 14
percent in 2010 (Patten & Parker, 2011).
18
Figure 1.3: Ratio of Male to Female College Enrollment in the U.S. Over Time
.
5
1
1.
5
2
2.
5
M
al
e−
Fe
m
al
e 
Co
lle
ge
 R
at
io
19
47
19
52
19
57
19
62
19
67
19
72
19
77
19
82
19
87
19
92
19
97
20
02
20
07
20
12
Year
Source: U.S. Department of Education, National Center for Education Statistics,
Digest of Education Statistics 2013, Table 303.10
spouse if the individual reenlists for an additional term of service.
The second chapter of this dissertation estimates military spouses’ price sen-
sitivity of demand for higher education by utilizing price variation due to military-
mandated moves across states. The sample is drawn from the time period before GI
benefits were transferable. Military spouses appear to be less price sensitive than
their civilian counterparts on a percentage-point basis. This estimate can be used
to help project the potential impact of the transferability of the GI Bill. A proper
estimate would also require a model of the joint educational and career choices of
the members of military family units, and may be a candidate for further research.
The results from research on the early and mid-century GI Bills will help
predict how veterans returning from the Iraq and Afghanistan wars will respond to
19
the educational incentives in the Post-9/11 GI Bill. An interesting wrinkle from
other aid programs that might interact with the GI funds is that Title IV eligible
institutions now have to follow what is called the “90/10 rule,” where no more than
90 percent of the revenue for a Title IV school can come from the federal government.
Federal Pell Grants and federal student loans count toward this 90 percent, but GI
funds do not. This rule gives schools that are close to this threshold incentive to
recruit military veterans aggressively, as one military veteran could allow for as
many as nine additional students who fully rely on federal funds to enroll. This may
explain why Post-9/11 GI Bill funds are spent in the for-profit sector at a higher rate
than funds from any other government source.11 Forty-two percent of Post-9/11 GI
Bill funds are spent in the for-profit sector even though overall the for-profit sector
makes up less than 10 percent of the market for higher education.
1.3.2 Pell Grants
The Pell Grant program is the largest source of means-tested financial as-
sistance available to post-secondary students in the United States. Pell Grants
(previously called Basic Grants) were enacted in 1972 and were intended to be the
foundation on which all other forms of aid would build. There have been periodic
expansions of the Pell program, with the most recent expansion coming at the start
of the Obama administration.12
11The flexibility of for-profit online education and willingness of for-profits to set up campuses
near military bases and offer schedules that are more convenient for those trying to balance a
military career and school likely also play a role in the explanation.
12The period between academic years 1995–96 and 2002–03 also saw particularly large increases
in the maximum Pell Grant even if the total Pell Grant funds did not increase that much those
years.
20
Several studies examined the difference in the enrollment rates of low-income
students before and after the establishment of the Pell program in 1973 (Hansen,
1983; Manski, 1993; Kane, 1994). Using three distinct data sources, none of these
three studies found a disproportionate growth in enrollment by low-income 18- to
24-year-olds. This result is both puzzling and disconcerting to the advocates of the
program.
One positive result of the program, found by research, came from examining
a non-traditional student population: older students. Seftor and Turner (2002)
found a large positive response by 25- to 30-year-old students to the introduction
of the Pell program. The authors suggested that a possible explanation for the
difference in response to this program by older students is that older students may
have more experience filling out complex forms (e.g., filing income taxes) and thus
may not be as deterred by the complexity of the Pell Grant application. A similar
interaction between complexity and experience may explain why although Pell Grant
aid does not appear to increase the initial enrollment of low-income 18- to 24-
year-olds, Bettinger (2004) found some evidence that Pell Grants might increase
persistence in college after the first year.13 Once someone has paid the cognitive cost
of figuring out how to complete an annual Free Application for Federal Student Aid
(FAFSA) form, it is likely not as daunting the second time.14 Therefore, complexity
might be a barrier to entry, but once in college, the financial incentive from a Pell
Grant to continue college is not as sensitive to the cost of filling out the form again.
13The average ages of first-year students in the sample were 18.8 and 21.0 for those attending
4-year and 2-year college, respectively.
14The reduction in uncertainty over the amount of aid due to the ability to use the previous
year’s aid as an estimate for future aid, may alternatively be the mechanism at work.
21
The complexity that comes with the current application process is due in part
to the means-tested nature of the program. One would hope that paying the cost
of this added complexity would ensure that low-income students are indeed the
beneficiaries of the program. Burman et al. (2005) used a tax simulation model to
estimate that 98 percent of Pell funds flow to students in tax units with adjusted
gross income of less than $50,000 and over 60 percent of students receiving funds
are in households earning less than $20,000.
However, simply because the statutory recipients of Pell Grants are indeed
low-income students (also often older and independent) does not mean they enjoy
the full value of this aid. As higher education institutions set their price each year
and engage in price discrimination by varying the level of institutional aid they offer
students, they may be able to capture some of the economic benefits of the Pell
Grant program. This is in contrast to the stated intent of the Pell Grant program
to be the base on which other grant aid is added. Turner (2014) used both regression
kink and regression discontinuity designs to tease apart the two competing desires
of institutions to either attract more Pell Grant-eligible students or capture more of
the aid themselves. She found through the combination of these two techniques that
schools on average enjoy about 12 percent of the economic benefits of Pell Grants,
but that the results vary by institution type. Cellini and Goldin (2014) and Singell
and Stone (2007) also found evidence of colleges capturing federal aid. The results
from Singell and Stone (2007) varied by type of institution as well. They found that
private universities increase net and sticker prices nearly dollar for dollar, while
public institutions do not appear to change the in-state tuition rate in response to
22
Pell Grants. Meanwhile, the results from Cellini and Goldin (2014) were focused
on for-profit schools and showed these schools as very aggressive and successful at
capturing aid.
1.3.3 Tax Benefits
There are a slew of educational incentives in the tax code. Some of the more
prominent are the American Opportunity Tax Credit (formerly the Hope Scholarship
credit), the Lifetime Learning Credit, deductions for interest on student loans or
tuition and fees, and various tax-exempt savings plans. Tax benefits are the most
widely enjoyed federal educational benefit. In 2013, 13.8 million Americans received
one form of federal education tax benefit or another, with the value of the average
benefit being $1,355 (College Board, 2014).
One line of inquiry focuses on showing how most of the benefits of tax-based
aid accrue to middle- or high-income households as opposed to low-income house-
holds. Burman et al. (2005) described the distributional effects of each of the tax
credits and deductions available at the time of publication in 2005. Their microsim-
ulation model estimated that over 60 percent of the tax expenditures from both
the Hope credit and the tax deduction for student loan interest payments were dis-
tributed to tax units with income above $50,000 a year. The Lifetime Learning
credit was slightly less targeted toward middle- and high-income households, with
roughly 50 percent enjoyed by those earning more than $50,000. The tax deduction
for higher education expenses was the least progressive, with the majority of the
23
tax expenditure going to tax units earning between $100,000 and $200,000. Similar
distributional results were reported by Long (2004), who found that families with
incomes between $30,000 and $75,000 benefitted the most from the Hope and Life-
time Learning Tax Credits, in part because insufficient tax liability and interactions
with other aid programs limited the benefit to low-income families. Finally, Dy-
narski (2004b) examined the distributional effects of educational saving incentives.
She found the gains from these incentives are heavily skewed toward high-income
households.
There is also the issue of economic incidence as opposed to the statutory
incidence. Turner (2012) found that tax-based aid crowded out institutional aid
roughly dollar for dollar within his sample of 4-year colleges and universities. These
are troubling results for those concerned with income inequality, not only because
of the immediate issue of a lack of progressivity in these tax benefits, but also
for their potential long-term effects on inequality and intergenerational mobility
through educational attainment. These long-term effects could manifest themselves
if these tax breaks increase the middle- and high-income recipients’ willingness to
pay. This increase in demand could lead to further increases in tuition faced by all.
This price increase could deter some low-income students who do not benefit from
the tax breaks. Long (2004) found some suggestive but inconclusive evidence that
public institutions raised prices in response to the Hope and Lifetime Learning Tax
Credits.
The complexity of having various alternative tax-based aid options appears to
lead to sub-optimal decisions by some tax filers. Turner (2011b) found that between
24
2002 and 2008, one out of four tax returns claiming a educational tax benefit failed
to select the tax-minimizing option. He attributed inertia in program selection and
a lack of salience of both the federal and state tax implications of various options as
the reason why non-tax-minimizing selections are made. Regardless of the reason,
a failure to choose the tax-minimizing option will likely mitigate the potential of
tax-based student aid to influence educational outcomes.
Evidence on the effect of these programs on educational attainment outcomes
has been mixed. Long (2004) was unable to find evidence that these programs
increased educational attainment. Meanwhile, Turner (2011a) estimated the price
sensitivity of enrollment from $1,000 of tax-based aid increases college enrollment of
18- to 19-year-olds by 3 percentage-points. LaLamia (2012) did not find a response
to tax-based aid among adults unless the adults fell short of their own educational
expectations from youth. Finally, Bulman and Hoxby (2015) used both regression
kink design to examine the response of students from higher income families and
simulated instruments, an evaluation similar to difference-in-differences, to evaluate
students at various income levels and found no meaningful changes in enrollment or
other education outcomes from the expansion of the education tax credits in 2009.
The third chapter of this dissertation adds to the research examining the effect
of changes in the tax code on educational outcomes. I use a natural experiment that
came about from an unanticipated change in the tax treatment of funds previously
set aside for higher education. The data do not indicate a statistically significant
change in enrollment or matriculation from the change in the tax treatment of these
funds despite the sometimes large economic value of the tax change. Instead, the
25
only significant response appears to be an increase in the allocation of future funds
into the account vehicle that saw a change in its tax status from exempt to non-
exempt.
1.3.4 Loans
The Higher Education Act of 1965 created the Stafford loan program, which
has long been by far the largest federal student loan program. A student loan
program for parents was added in 1980, called the Parent Loans for Undergraduate
Students (PLUS) Program. In 1992, Congress created an unsubsidized Stafford pro-
gram for borrowers ineligible for the means-tested subsidized Stafford loans. These
three programs make up over 80 percent of all student loan debt.
There has been much recent discussion in the media on the total amount
of student debt overtaking credit card debt in June 2010 and then continuing to
grow to over $1 trillion dollars two years later.15 By 2015, the total outstanding
student loan debt in the United States exceeded $1.25 trillion. It is important,
however, to understand the composition of this debt when considering its impacts
on the higher education landscape. Baum (2013) provided a thorough breakdown
of recent trends in the acquisition and repayment of student loans by graduate vs.
undergraduate status, federal vs. private loans, and dependent vs. independent
individuals. She emphasized that although both the total amount of student loans
and the average loan amount per full-time equivalent (FTE) student have increased
15See http://topics.nytimes.com/top/reference/timestopics/subjects/s/student_
loans/index.html and http://www.forbes.com/sites/halahtouryalai/2014/02/21/
1-trillion-student-loan-problem-keeps-getting-worse/.
26
considerably, most of the increase between 2005 and 2012 was due to the number of
individuals taking out student loans rather than an increase in the average balance
per borrower. Furthermore, graduate students have seen a larger rise in borrowing
per student than undergraduates for the 5 and 10 years prior to 2012, so the overall
increase for undergraduates is even less when graduate students are excluded.
Even if the average loan per borrower remains manageable, questions still
remain about the distribution of the amount students borrow and whether a subset
of students are taking on too much debt. Baum and Steele (2010) reported on the
distribution of student debt and the frequency of high debt levels. They found high
debt levels to be more common among independent students and dependent students
from middle-income families. They also found high debt is more common among
African-American Bachelor’s degree recipients than among those from other racial/
ethnic groups. Additionally, students who graduated from for-profit institutions are
much more likely than similar students in other sectors to have high debt levels.
1.3.4.1 Repayment
The potential threat of increasing student debt is that those taking out the
loans may not be making sound investments in their own futures with the access the
loans permit. It may be difficult for the individual—let alone the government—to
know whether each loan issued will end up improving the life of the borrower and
the welfare of society more than the cost of the loan. One signal the government
does receive about the quality of the loans it issues after the fact is through the
27
default rate on student loans. Unfortunately, the fraction of student loan balances
delinquent for more than 90 days increased steadily from 6 percent in the first
quarter of 2003 to roughly 9 percent by the first quarter of 2012. This delinquency
rate then suddenly jumped to 12 percent by the third quarter of 2012 (Dai, 2013).
What might be even more disconcerting is the large amount of student loans that
are not yet in repayment and thus not eligible to be considered for delinquency. If
one only looks at loans in repayment, the delinquency rate as of the fourth quarter
of 2012 was over 30 percent (Brown, 2013).
Some researchers have tried to shift the focus of the debate from the volume
of debt to the ability to manage it. Dynarski and Kreisman (2013) looked at the
moderate median loan amount and the high default rate and concluded the country
is facing a repayment crisis, not a debt crisis. They proposed a simplified income-
based repayment system to replace the current complex repayment system. This
income-based repayment plan, called Loans for Educational Opportunity, would ad-
just deductions automatically from a borrower’s paycheck based on their income, like
Social Security. Those who wanted to pay off their loans more aggressively would
still have an option to do so, but the new system would be the default option. The
authors claimed this system would work efficiently for 98 percent of student loan bor-
rowers and recommended stronger consumer protections for the remaining 2 percent
who are currently more than $50,000 in debt. The proposed consumer protections
would include allowing private student loans to be dischargeable in bankruptcy and
requiring federal student loans to be exhausted before private loans can be issued.
Potentially addressing the repayment issue in a different manner, in 2014 Ore-
28
gon and Michigan both proposed a “Pay It Forward” program, where students would
agree to pay a fixed percentage of their income for a fixed number of years. However,
these programs are not mandatory and are likely to suffer from the same adverse se-
lection encountered by a similar program attempted by Yale in the 1970s. Dynarski
(2014) suggested some “tweaks” to these current programs to make them more sus-
tainable long-term. These suggestions would make these programs similar to her
proposed income-contingent repayment program and would allow borrowers to fully
pay their debt plus interest ahead of schedule and then be out of the program for
good to ensure that those who expect high incomes after college still participate.
Others also have championed a simplification of the student loan process.
Baum and Schwartz (2014) argued that the current set of options creates a “para-
dox of choice” and that a streamlined set of options would be preferable. This
streamlining would include the elimination of the in-school subsidy, thus negating
the need for a complicated financial aid determination. They also advocated for
private student loans to be dischargeable in bankruptcy and for improvements in
distinguishing federal and private student loans in the minds of potential borrowers.
1.3.4.2 Data Availability
There are still several unanswered questions about student loans. Part of
the reason is a lack of publicly available data that link together characteristics of
the borrower, the school they attended, their loan type and amount, and their
repayment history. Dynarski (2015) recently criticized this lack of available data for
29
researchers, pointing out that the Federal Reserve purchases data from Equifax to
track trends in student debt. Meanwhile, the data that are made available from the
government are not as detailed or as comprehensive as would be necessary to answer
several outstanding questions. However, this is slowly changing. In late 2015, the
government released detailed school-level data on student loan repayment that had
not previously been made public.
1.3.5 Interactions
It is important to understand not only the role and effect of each of these
federal, state, and institutional financial aid programs focused on higher education
in isolation, but also how they interact. Marx and Turner (2015) examined the
interaction between the federal aid and student loan programs and concluded that
the marginal dollar of Pell Grant aid crowds out more than a dollar of student
loans. The authors suggested this greater than dollar-for-dollar reduction in loans
could be explained by fixed costs to borrowing. This cost argument means that an
increase in Pell Grants might push a student below her borrowing threshold alto-
gether. For those who cross this threshold, the lower amount of funds could drive
down persistence and educational attainment. This potential decrease in educa-
tional attainment for some could explain why there does not appear to be a positive
effect on educational attainment from Pell aid overall despite the expectation that
increased aid should boost educational attainment.
Other research has focused not on whether an individual attends college but
30
on which college they attend. Avery and Hoxby (2004) examined the behavioral
response of potential students to different financial aid packages, albeit for a non-
representative, high-ability group. They found that these students responded to
prices; most of these responses are rational, but there are some systematic errors,
and these errors are less likely to be made by students from high-income families,
particularly if one of the parents went to a selective school. One of the most inter-
esting errors is that students responded to a $1,000 increase in grants, loans, and
work-study commitment in almost the exact same way. Given the difference in the
economic value of these three items, it appears that even these high-ability students
are overvaluing loans and work-study relative to scholarships. This apparent lack
of understanding about loans is troubling given the large and growing amount of
debt-financed investment in higher education.
1.4 Returns to Higher Education in the United States
Higher educational attainment is associated with greater earnings and lower
levels of unemployment. However, the fact that more educated individuals enjoy
better economic outcomes could also be due to other factors correlated with college
attendance and completion, such as ability or family income. If high-ability individ-
uals who would be paid more anyway are also more likely to go to college, this could
bias the estimate of the returns to education if one only looks at wage differences
between individuals with different levels of education. Alternatively, if those who
benefit the most from college are the ones self-selecting into college, then using the
31
increase in wages from this group to project what would happen if any high school
graduate were to enroll in college would overestimate the returns to education.
To identify causal effects between college and earnings, one would have to find
exogenous sources of variation in educational attainment and examine any corre-
sponding increase in wages. Card (1993) used geographic proximity to college and
Angrist and Chen (2011) used the Vietnam-era draft lottery to obtain an exogenous
increase in educational attainment; both found returns to education higher than
simple ordinary least squares (OLS) regressions would predict. Specifically, Card
(1993) estimated a return to a year of college to be between 10 and 14 percent, while
Angrist and Chen (2011) estimated a 9 percent increase in earning from a year of
college.
These estimates served as a baseline for the returns to education for some
time.16 However, these causal estimates are determined by those who switched
their education levels, known as “compliers,” due to the specific driver of change.
The effect is the local average treatment effect, or LATE, and may not apply to
other reasons for switching. For a LATE to be useful for policy implications, the
driver should be mutable from a policy perspective. Zimmerman (2011) found such
an instrument and a rich data set to test it on: a large sample of Florida students
at the cusp of academically qualifying for admission at a public 4-year university in
Florida. Those who did not meet the eligibility criteria were more likely to attend
a community college. Using this instrument, Zimmerman (2011) found the returns
to a year of education at a 4-year school to be 8.7 percent. This is in line with prior
16Angrist had related versions of this paper with various co-authors circulating for some time.
32
LATE estimates but has greater external validity given the nature of the instrument.
With the returns to education from various LATE estimates confirmed, I now turn
to the research focused on examining the changes to the returns to education over
time.
Goldin and Katz (2008) took the long view and chronicled the large educational
wage differentials starting in 1915 for both a high school and college degree. Both
these education premiums shrank from the beginning of their period of observation
until 1940 and then held steady through 1980. The big difference between these
two premiums came after 1980, when the high school wage premium had anemic
growth while the college wage premium increased, coming close to its prior 1915
highs by 2005. Goldin and Katz (2008) went on to show that a simple supply-
demand framework where the demand for college workers grew steadily and supply
fluctuated could explain the movements of the college wage premium over the entire
time frame pretty well. It is this later time frame, from 1980 onward, that will be
the primary focus of the rest of this section.
The deviation in the trends in the education wage premium for high school
and college since 1980 raises questions on how the other educational attainment cat-
egories responded over that time frame. In their examination of the reasons for the
increase in U.S. wage inequality over the last four decades, Autor, Katz, and Kear-
ney (2008) presented the real, experience-adjusted earnings of full-time, full-year
workers for different levels of educational attainment: high school drop-out, high
school graduate, some college, exactly college, and more than college. The authors
chronicled a near-uniform climb in log real wage levels from 1963 through 1973 for
33
five education categories spanning from high school drop-out to those with more
than a college degree, with exactly college and more than college growing slightly
faster. From 1973 to 1979, real wages leveled off for all five educational attainment
categories, but since 1979 real wages for some college, high school graduates, and
high school dropouts have all fallen, with high school dropouts falling below their
1963 levels. Meanwhile, wages of those with a college degree but no post-college for-
mal education held at their late 1970s level through the 1980s and early 1990s and
rose even higher starting in the late 1990s, while those with more than a college de-
gree experienced steady increases over time. An update of this breakdown in Autor
(2014) showed that those with more than a Bachelor’s degree have seen a continued
increase in their real wages while the wages of those with all other education levels
have leveled off or fallen since the start of the Great Recession. Women have seen
a similar increase in dispersion of wages by educational attainment, but real wages
for women with low education held steady instead of declining.
Not only has the dispersion of earnings by education increased over time, but
the dispersion of earnings among those with a college education also appears to
be increasing over time. Since 1970, the income distribution has widened among
college-educated workers. Card (1995) found not only that the return to a year
of college is particularly large for disadvantaged students but also that low-income
students had larger returns from attending selective colleges. Hoxby and Long (1999)
documented the increasing inequality in wages of college graduates. They found that
while college graduates at the 90th percentile of earnings saw their wages increase
between 1975 and 1992, those at the 10th percentile saw their wages decline in real
34
terms. They suggested increasing segregation by selectivity in higher education with
more selective schools spending more per student as a possible explanation. The
most selective institutions often spend more than three times as much as less selective
schools. Hoxby and Long (1999) found that nearly half of the explained growth in
this dispersion was due to the increasing concentration of peer and financial resources
at more selective colleges and universities relative to other institutions. There is
no consensus in this literature, however, because Dale and Krueger (2002) found
there was no return to college selectivity for most students when they compared the
earnings of students who were admitted to the same colleges but chose to attend
different ones. This negative result appeared to hold for the average college student,
but Dale and Krueger still found large returns to college selectivity for low-income
students.
Part of the mystery over the selectivity of and return to education from a
college may be explained by the students at the most selective schools having a
higher probability of going on to post-baccalaureate education. Lindley and Machin
(2011) showed that much of the variance in the earnings of college graduates could
be explained by those college graduates with professional degrees. The ability to go
on after college and receive a professional degree or doctorate can therefore be seen
as an option value of a college degree.
The returns to education appear substantial and seem to be growing over time.
A number of authors have attributed this growth to an increase in the demand
for skilled labor outstripping supply (Goldin and Katz, 2008; Autor, 2014). This
increased demand for skilled labor driving the returns to higher education may in
35
turn affect the price sensitivity of demand for higher education.
1.5 Price Sensitivity of Demand for Higher Education
There are many factors that enter into the determination of the price sensitivity
of demand for higher education. The ability and willingness of an individual to pay
for college will depend on the price she faces, her expected returns from additional
schooling, the consumption value of attending college, and her wealth or lack of
credit constraints (Becker, 1962 & 1975; Cameron & Taber, 2004). There are also
several other factors that relate more to social norms and default conditions that
appear to contribute to an individual’s decision to continue his or her education
(Field, 2009; Pallais, 2015).
The literature estimating the price sensitivity of higher education has been
thoroughly developed over the years. Multiple reviews of this literature have been
published over time (Leslie & Brinkman, 1988; Heller, 1997; Deming & Dynarski,
2010). This chapter presents a broad review and is organized by the source of
variation in price. This organization highlights a puzzle: the effect of a reduction in
the cost for college seems to depend on the form of the price reduction. A general
pattern emerges in that the changes in net cost that are more easily understood and
are known earlier in the application process seem to have the greatest effect.
36
1.5.1 Changes in List Price
Leslie and Brinkman (1988) summarized 25 studies from the 1970s and 1980s
that primarily relied on variation in college cost by state and over time. They
concluded that a $100 increase in tuition price (in year 1982-1983 dollars) would
result in a 0.7 percentage-point decrease in the matriculation of 18- to 24-year-olds.
This translates into a 4 percentage-point decrease for a $1,000 increase in price (in
year 2000 dollars).
Following the meta-analysis by Leslie and Brinkman, three additional studies
used differences in the tuition rate at public colleges between states and correspond-
ing differences in matriculation rates into college to estimate the price sensitivity
of demand for higher education. Kane (1994) and Cameron and Heckman (1998)
applied similar identification strategies using data from the October Current Popula-
tion Survey (CPS) and the National Longitudinal Survey of Youth (NLSY), respec-
tively. Both studies estimated roughly a 4 percentage-point increase in enrollment
for a $1,000 per year decrease in price (in year 2000 dollars). These results appear
promising, but one criticism of this analysis is possible endogeneity. States that
chose to provide greater subsidies to public colleges to lower in-state tuition costs
could also invest more in K-12 education, have a more extensive network of com-
munity colleges, or have a population that values education more highly and pushes
their children to attend college. In response to this criticism, Kane (1999) focused
on differences in tuition within states over time and found similar results to his 1994
paper using between-state differences.
37
Despite the potential endogeneity issues, most studies using this form of vari-
ation found the expected sign for the effect for most subgroups with one notable
exception. Avery and Hoxby (2004) found that for high-ability students, there was
an interestingly heterogeneous response to higher tuition by the selectivity of the
school their parents attended. High-ability students whose parents went to low- and
medium-selective colleges reduced their probability of attending a school as tuition
increased, but students of parents who attended highly selective college were more
likely to matriculate as tuition of a school increased. The authors believed this could
be due to the tuition prices signaling quality and/or per pupil spending on students.
Although the tuition prices faced by residents of states seldom vary much over
time, there is at least one notable exception. Abraham and Clark (2006) studied the
impact of the District of Columbia’s Tuition Assistance Grant Program (DCTAG),
which allows DC residents to attend public colleges and universities throughout
the country at considerably lower in-state tuition rates starting in 1999. They
found that the percentage of DC residents of high school graduating age enrolling as
freshman increased 3.6 percentage points for every $1,000 decrease in out-of-pocket
costs (in year 2000 dollars). They also found even greater movement over where the
DC residents went to school, with substantial shifts to public 4-year institutions,
particularly within neighboring states.
The second chapter of this dissertation also uses variation in tuition costs
across states and avoids many of these potential endogeneity issues by focusing on
a population whose state of residence is determined by military-mandated moves.
Using this identification strategy, I find a 1-1.5 percentage-point increase in college
38
attendance from a $1,000 decrease in tuition costs (in year 2000 dollar) for both
traditional and non-traditional aged students. This result is roughly half the esti-
mated price sensitivity (on a percentage-point basis) from the previous studies and
may be due to unique challenges military families face, or might reflect a change in
the price sensitivity of demand over time as the returns to college have increased.17
That said, this result could be used to estimate the expected response of programs
aimed at increasing the educational attainment of military families even if they are
not necessarily generalizable to the non-military public.
1.5.2 Federal Programs
1.5.2.1 Social Security
To get around the aforementioned endogeneity issues that arise when using
cross-state variation in the cost of college, an alternative source of changes in price
could be used. A researcher would ideally use some exogenous change in price
unrelated to a student’s tastes for college. The elimination of the Social Security
Student Benefit in 1986, which provided aid to 18- to 22-year-old college students
who were children of deceased, retired, or disabled Social Security beneficiaries,
provided such an exogenous price change. Dynarski (2003) estimated that a $1,000
(in year 2000 dollars) increase in the out-of-pocket costs of attending college from the
elimination of the benefit translated into a 4 percentage-point reduction in college
17This result translates into a 7 to 10 percent change in enrollment given the baseline level of 14
percent enrollment in the sample. This percent change in enrollment is consistent with previous
estimates, which traditionally examined populations exhibiting 40 to 50 percent enrollment.
39
attendance by the affected population.
1.5.2.2 GI Bills
Angrist (1993) and Bound and Turner (2002) showed that the GI Bills from
the Korean War and World War II increased educational attainment. While these
results are important, projecting the estimates from these studies to the current
educational environment may be irrelevant given increased returns to education and
tuition costs. Also, as discussed earlier, it is difficult to separate the financial impact
of the GI Bills from the effect of military service or draft avoidance behavior for a
cohort. Thus, any price sensitivity of demand estimates generated from the results
of these studies may be both inaccurate and out-of-date.
1.5.2.3 Pell Grants
Several studies have used the establishment of the Pell Grant program in 1973
to obtain an estimate of the price sensitivity of demand. Hansen (1983) examined
the rate at which students from low-income and high-income families attended col-
lege before and after the program started and found no significant difference. Due
to questions on the effect of the selection of particular years to limit the sample and
the potential contamination of the sample from the end of the Vietnam War, Kane
(1994) reexamined this question, focusing on women, and again found no significant
differences in enrollment rates for program-eligible females compared to those ineli-
gible. However, Sefton and Turner (2002) found that non-traditional aged students
40
did respond positively to Pell Grants. Their results translate into a 0.5 percentage-
point increase in matriculation from a $1,000 increase in aid (in year 2000 dollars).
The complexity of the Pell Grant program may be a reason for the lack of
an effect on initial enrollment. However, after enrolling and receiving a Pell Grant
a student may have a better understanding of the program and be able to more
easily estimate the value of further Pell Grants. Therefore, the Pell Grant program
could affect persistence even if it does not influence initial enrollment. Manski (1993)
questioned whether the additional funds from Pell aid might increase persistence and
lead to higher graduation rates but found no disproportionate growth in obtaining
a college degree for the eligible “college-aged” population. Bettinger (2004) found
some suggestive yet inconclusive evidence that Pell aid increases persistence.
1.5.2.4 Tax-Based Aid
The source of variation in price most relevant to the third chapter of this
dissertation is tax-based aid.18 The literature on the price sensitivity of demand for
higher education from tax-based federal student aid is fairly limited, as tax-based aid
for higher education in the United States is itself a fairly new concept, beginning in
1998. The Hope Learning Credit (HTC) and Lifetime Learning Tax Credit (LLTC)
were both introduced in the Tax Relief Act of 1997.19 While these programs clearly
subsidized the college education of the American middle class, it is less clear how
much these tax programs increased college attendance (Long, 2004; Burman et al.,
18The variation in chapter 3 is based on the tax-treatment of savings for higher education.
Therefore, tax-based aid is similar but not a perfect analog.
19The HTC was later modified and renamed the American Opportunity Tax Credit (AOTC)
beginning in 2009.
41
2005).
The first study that attempted to estimate the impact of these tax incentives on
college enrollment was that of Long (2004). She used October CPS data from 1990
to 2000 which had family background data that allowed her to estimate eligibility
for the tax benefits, but the data were not ideal. The incomplete data created
a noisy estimate for tax-based aid. A noisy measure makes it more difficult to
find statistically significant results. Long was not able to find an increase in total
enrollment, a shift to more 4-year schools, or a shift to more full-time enrollment for
either traditional 18- to 24-year-olds or 24- to 40-year-olds. Therefore, it appeared
that either the complexity of the aid and the time delay involved in receiving it
mitigated its predicted effect, or the measurement error in the income variable and
thus the eligibility for the tax credits obfuscated the true impact by biasing the
estimates toward zero.
Turner (2011a) used Survey of Income and Program Participation (SIPP) data,
which have more precise income information than the October CPS, to estimate the
impact of tax-based aid on the college enrollment of 18- to 19-year-olds.20 Unfortu-
nately, the SIPP does not include variables on tax-based aid or all the variables nec-
essary to calculate their value, but it does allow for a close approximation. Turner
found a $1,000 increase in tax-based aid increases full-time enrollment by either
3.2 percentage-points or 2.6 percentage-points depending on whether only cross-
sectional variation in award size among eligible youths or difference-in-differences
20The SIPP is a nationally representative survey designed to provide accurate and comprehensive
information on income and program use.
42
estimation using both eligible and ineligible youth is used for identification. This
translates into an 8.6 percent and 9.9 percent increase in full-time enrollment from
a $1,000 increase in tax-based aid. Turner (2011a) also found that tax-based aid
increased persistence.
SIPP data allowed for more detailed examination of 18- to 19-year-olds but
were not well suited for examining the effect of tax-based aid on the educational
outcomes of adults. Given the inherent complexity of tax-based aid in the United
States and the results from Sefton and Turner (2002), which could imply adults are
more responsive to complex forms of aid, a re-examination of the adult population’s
response to tax-based aid with an alternative data source to the October CPS was
desired. LaLumia (2012) used data from the NLSY79, which have more detailed
information on educational attainment and college enrollment than the October
CPS. The sample used in her study had an average age of 41, a majority were
married, and most had children. Given that the age range started at 33, all were
at least “minimally” non-traditional, and most of her sample was “moderately”
to “highly” non-traditional. LaLumia had difficulty finding statistically significant
impacts of tax-based aid on the college attendance or degree completion for this
population and could rule out a effect on attendance larger than 2 percentage-
points. One exception was for a subset of adults who failed to meet their education
expectation as reported in their youth. For this subset, a $1,000 increase in tax-based
aid was associated with a 3.3 percentage point increase in college attendance. The
sample in this study had a lower rate of college attendance than samples used in other
studies, so a percent change, instead of a percentage-point, is the more appropriate
43
measure to use to compare against other studies. Although the estimated sensitivity
is in line with percentage-point estimates from other populations, the percentage
increase this represents is uncharacteristically large. This raises questions about the
both the internal and external validity of this result.
The expansion of higher education tax credits with the enactment of the AOTC
in 2009 and de-identified data from an IRS database for a more detailed examina-
tion of the effects of tax credits on higher education outcomes. Bulman and Hoxby
(2015) used these data and both regression kink and simulated instruments to look
for a response to the expansion of the tax incentives. The AOTC did not have a
statistically significant impact on any of the educational outcomes examined, in-
cluding total enrollment. Over the wide range of household income and student age
categories examined, this study did not find any evidence that tax credits affected
educational outcomes. It appears tax credits, at least the expanded credits in the
AOTC, did not generate the same magnitude of response as an equivalent change
in the net price of higher education from other sources of variation.
1.5.3 Non-Federal Programs
1.5.3.1 State Merit Aid Programs
The most studied state merit program is the Georgia Helping Outstanding
Pupils Educationally (HOPE) Program, which also served as the inspiration for the
Hope Tax Credit and many similar state programs. The Georgia HOPE Scholarship
allowed free attendance at Georgia’s public colleges for state residents with at least a
44
B average in high school. The results from Dynarski (2004a) and Cornwell, Mustard,
and Sridhar (2006) suggested that Georgia’s program had large impacts on the
college attendance rates of children from middle- and high-income families. Using
a set of nearby states as a control group, Dynarski found that Georgia’s program
likely has increased the college attendance rate of all 18- to 19-year-olds by at least 7
percentage points. The results suggest that each $1,000 in aid (in year 1998 dollars)
increased the college attendance rate in Georgia by 3.7 to 4.2 percentage points.
Dynarski (2008) found similar results, with women responding more than their
male counterparts, when evaluating the first two large-scale merit-aid scholarships
in Arkansas and Georgia together. Although decomposing the increase in college
completion into the effect on increased initial enrollment and greater persistence is
difficult given the available data, Dynarski showed that both initial enrollment and
college completion increased and that the private benefits of the program appear to
far exceed the costs. However, Sjoquist and Winters (2012b), studying a broader
set of similar programs, did not find similar results. Furthermore, when Sjoquist
and Winters (2012a) tried to recreate the results from Dynarski (2004a) using the
Census 5 percent file instead of the 1 percent file, they did not find statistically
significant results.
The merit programs in Georgia and Arkansas are not the only ones to be
studied in depth. Under California’s Cal Grant program, those who met certain
academic and financial need requirements were eligible in 1999 for up to $9,400 per
year to attend a private 4-year institution in California or a grant to cover tuition
and fees at a California public college. Using regression discontinuity to exploit the
45
thresholds to qualify for the Cal Grant program, Kane (2003) found that Cal Grant
eligibility translated into a 3 to 4 percentage-point increase in enrollment among
financial aid applicants.
This result cannot easily be translated into an increase-per-dollar estimate, as
the net dollar value of aid is difficult to calculate given limited information on how
much Cal Grant assistance may have “crowded out” other forms of aid. Assuming
no “crowd-out” produces a lower bound estimate of a 1.2 percentage-point increase
in enrollment from $1,000 in Cal Grant funds. When attempting to control for
institutional aid crowd-out, Kane estimated that a net $1,000 increase in aid from
the Cal Grant program resulted in a 9.2 percentage-point increase in enrollment, an
estimate much higher than the rest of the literature has found.
All of the merit scholarships above were studied as quasi-experiments, and
the results varied. The STAR project, on the other hand, followed an experimen-
tal design and randomly assigned entering first-year undergraduates to one of three
treatment groups receiving either additional services, financial incentives, or both.
The additional services consisted of a peer-advising service and a supplemental in-
struction service. The financial incentives ranged from $1,000 to $5,000 dollars
for maintaining good grades. Angrist, Lang, and Oreopoulos (2009) reported on
the results of the experiment. They found that students offered services without
the financial incentives did no better than those in the control group. Women ap-
peared to respond the most to financial incentives with increased GPAs, and these
increases continued after the program ended. However, the Angrist, Oreopoulos,
and Williams (2014) examination of the Opportunity Knocks program, a program
46
similar to STAR, was unable to find positive results that persisted after the program
had completed.
1.5.3.2 Place-Based Programs
There also has been a recent emergence of place-based scholarship programs
that allow for the study of the effect of a sudden change in the expected cost of college
on educational outcomes. The most well-known of these programs is the Kalamazoo
Promise, which pays a portion of the tuition at any Michigan public college for
students graduating from the local school district. The amount of the scholarship
depends on how many years the student had gone to the local school. A longer term
follow-up of the effects of the Kalamazoo Promise on enrollment, persistence, and
completion using difference-in-differences and propensity score matching is due out
shortly (Bartik, Hershbein, & Lachowska, 2015). Preliminary results suggest the
Kalamazoo Promise increased college enrollment by 7 to 8 percentage points. These
increases in enrollment also appear to have translated into statistically significant
increases in college completion within 6 years.
In addition to differences in the effect of an additional dollar reduction in net
cost by the source of funding, some heterogeneous treatment effects by recipient
characteristics have been found. Work by Angrist, Lang, and Oreopoulos (2009)
and Dynarski (2008) both suggested greater sensitivity to merit-based tuition sup-
ports among women than men. Coupled with the long-term rise in the share of
women attending college, this suggests that many of the recent programs may have
47
been more effective at providing incentives for women than men. Why men remain
relatively unresponsive and have fallen behind their female counterparts in college
attendance and completion rates remains an area for further research.
1.5.4 Other Sources of Variation
1.5.4.1 Immigration Status
The recent legislative focus on either curtailing or extending in-state tuition
benefits to undocumented students has provided additional estimates for the price
sensitivity of demand for higher education, at least for this subgroup. Given that
undocumented youths have difficulty capitalizing on additional education in the
formal labor market and may struggle to fund their education without access to
federal student loans or aid, one might expect these students to be more responsive
to relatively small changes in the price they face.
Kaushal (2008) and Chin and Juhn (2010) used data from the CPS-ORG and
the ACS respectively and changes in state laws allowing in-state tuition to children of
undocumented immigrants to estimate the demand response to this change in price.
Their results showed a much lower elasticity of demand, if any, for this population
due to these policies as compared to responses to tuition changes of native-born
Americans.
The estimation strategy used in the aforementioned studies may be problem-
atic. Research has suggested that new immigrants’ location choices are unaffected
by state subsidies (Kaushal, 2005; Zavodny, 1997), preserving the identifying as-
48
sumption above. However, measurement error was an issue for both studies as
citizenship status is not explicitly asked in the data collected. Conger and Turner
(2015) overcame the measurement error problem by obtaining data from a large ur-
ban university system that collected data on the immigration status of their students
and underwent changes in the tuition rate these students faced. A series of papers
from this data source is the most convincing work to date and has shown strong ef-
fects of tuition increases causing a reduction in degree completion by undocumented
students.
1.5.5 Non-Monetary Sensitivities of Demand
Net tuition and fees are not the only costs students must pay to attend college.
The largest cost is likely the opportunity cost of their time. There are also several
upfront costs that high school graduates face when applying to a school before they
can attend. Whether it is taking the SAT or ACT, writing an application essay,
filling out a FAFSA form for aid, or deciding which schools to apply to, there are
several cognitive, hassle, and small financial costs that are part of the application
process. Although the magnitudes of these costs are small, researchers have found
that the elimination or mitigation of these small costs can increase application to
and enrollment into college.
I begin with the literature on the costs associated with applying for financial
aid. Appling for financial aid includes several small tasks that involve paperwork and
uncertainty, and these seemingly low-cost activities may prevent individuals from
49
applying to college or for financial aid due to inertia or unfamiliarity with the system.
Kane (1995) was an early and vocal critic of the complex and backward-looking,
means-tested federal financial aid program in which the value of financial assistance
is not known with certainty until after acceptance into a college. In 2004, Avery
and Kane reported on a survey of student perceptions of college in the Boston area.
They found that both low-income students and high-income students in the city
perceived college as a sound investment at similar rates. They found some evidence
to support the idea that low-income students were disproportionately discouraged
by the complexity of the process of applying for financial aid and college admissions,
even if they were qualified and enthusiastic about going to college. They concluded
that interventions that only present information on tuition, financial aid, and likely
wages are unlikely to be successful at decreasing the gap in college attendance by
family income. Their results suggest that interventions that also target simplifying
or assisting with the application process may be more effective.
One point of note is that the survey was conducted in Boston, an urban area
with an uncommonly large number of colleges and universities. Although low-income
students in Boston may already be aware of the financial benefits of attending col-
lege, that does not mean all low-income students are as aware as their high-income
counterparts. Later in this section, I will discuss other research that specifically
tries to find, target, and inform isolated, high-achieving, low-income students who
the researchers believed would make different choices after receiving additional in-
formation.
In 2006, Dynarski and Scott-Clayton published a description of the complexity
50
of the aid system and argued that this complexity is a serious obstacle to both
efficiency and equity in the distribution of student aid. Bettinger et al. (2012)
tested this idea by randomly assigning H&R Block tax professionals to help low-
to moderate-income families complete the FAFSA. Their analysis suggested that
individuals who received assistance with the FAFSA and information about aid were
substantially more likely to submit the aid application, enroll in college the following
fall, and receive more financial aid than those who only received the personalized
information about aid. Specifically, enrollment rates for high school seniors increased
by 7 percentage points (or 30 percent) while increasing 20 percent for young adults
who had already graduated high school or who had a GED. The program was
targeted at households with less than $45,000 dollars in income with a 17- to 30-
year-old not already in college, but the effects seem to be driven by families with
income less than $22,000.21
Aside from applying for financial aid, there are also cognitive costs of deciding
which schools to apply to, and schools also often charge an application fee. Carrell
and Sacerdote (2013) found that providing high school students with college ap-
plication guidance and waivers for college application fees increased matriculation
into college, particularly for students attending disadvantaged high schools. Hoxby
and Turner (2013) found that sending high-achieving, low-income students appli-
cation fee waivers and information about colleges and optimal application strate-
gies induced them to attend more selective colleges and, given their high academic
21For further reading on the political and legislative reaction to this line of research and potential
further improvements, one can refer to Dynarski and Wiederspan (2012).
51
achievement, these more selective colleges were populated with intellectual peers,
representing a better school match for these students. Hoxby and Turner (2015)
presented information gathered from surveys on the understanding of the aforemen-
tioned high-achieving, low-income students on their collegiate options and prefer-
ences. They found that universities with a reputation for being a “party school” or
known for athletics were unappealing to high-achieving students. Customized infor-
mation about net cost and colleges’ characteristics appeared to have influenced the
students’ perceptions and resulted in a stronger set of schools selected during the ap-
plication stage. Hoxby and Turner also found that these high-achieving, low-income
students tended to be particularly misinformed about liberal arts colleges.
Even after getting into college, many students are unaware of the full implica-
tions of the financial decisions they have made. Akers and Chingos (2014) reported
on survey results showing that college freshman often cannot correctly identify the
net cost they are paying to attend school or the amount of loans they are taking
out. Finally, of those students who had federal student debt, 28 percent did not
think they had federal student debt and 14 percent reported not knowing they had
any student debt.
A general theme that has emerged from this research is that small fixed costs
- financial, cognitive, or informational - seem to have unexpectedly large effects on
the rate at which students apply or matriculate to college. These costs also appear
to affect low-income students more than their middle- and high-income peers. This
might help explain the persistent and growing gap in higher education by family
income, which is an area of great policy and academic interest, but is tangential to
52
the main issues being considered in this dissertation.
1.5.5.1 Credit Constraints
Given the array of federal (and now private) student loans available to poten-
tial students, are some students who would otherwise go to college still credit con-
strained? Shea (2000) examined the effect of “luck” or random changes in parental
income on the educational attainment of their children. He generally found negli-
gible effects on the human capital acquisition of the children except for those from
low-income families. This result suggests that credit market constraints lead low-
income households to make suboptimal investments in their children. However,
Cameron and Taber (2004), using NLSY data and a variety of identification strate-
gies, found no evidence that borrowing constraints impeded educational attainment.
Cameron and Taber concluded that credit access is not a driver of under-investment
in education, given the array of current programs available. Carneiro and Heckman
(2002) came to a similar conclusion; they argued that given the array of current loan
options, credit constraints were likely not binding in most cases. However, they did
identify a group of people (at most 8 percent of the population) who seemed to
face short-run credit constraints. But overall, long-run family factors that were
correlated with family income, like the home environment or investments in early
childhood education, appeared to dominate.
Most of the studies that concluded credit constraints have little effect on college
attendance were conducted using data on students who would have been deciding
53
whether or not to go to college in the 1990s. Belley and Lochner (2007) wondered
whether increases in tuition may have made borrowing constraints more binding.
They found an increased correlation between college attendance and income in later
years, which could reflect a reemergence of credit constraints.
1.6 Conclusion
With growing demand for skills driving the returns to education higher, the
benefits from a college degree have increased. Meanwhile, the supply of college
instruction, particularly at resource-intensive schools, has not kept pace, thereby
escalating the costs of obtaining a college degree. These two factors may cause fluc-
tuations in the price sensitivity of demand for higher education over time. Therefore,
estimates for this sensitivity should be updated periodically to check for such fluc-
tuations. Additionally, the sensitivity of demand for higher education appears to
vary by the source of aid and the population receiving it.
• Changes in tuition prices appear to generate the largest and most consistent
sensitivity estimates.
• Changes in price from the Pell grant program, which is complex, has not been
found to increase enrollment except among older, non-traditional students.
• There is some evidence that the Pell grant program increases persistence.
Perhaps the experience with the program mitigates some of the issues of its
complexity.
54
• Studies of tax-based aid have struggled to find consistent statistically signifi-
cant effects on enrollment.
• Many but not all studies of state merit aid and place-based programs find
results similar in magnitude to changes in tuition prices, but there is some
variation in results by the age and gender of potential students.
As the proportion of non-traditional students attending college grows, research ex-
tending estimates for the sensitivity of various non-traditional groups of students
becomes more valuable. The next two chapters estimate the price sensitivity of
demand for two previously understudied groups of non-traditional students.
55
Chapter 2: THE PRICE SENSITIVITY OF DEMAND FOR HIGHER
EDUCATION AMONG MILITARY SPOUSES
2.1 Introduction
The U.S. government recently passed legislation aimed at improving the edu-
cational outcomes of military spouses. The new legislation aims to lower the costs
of attending school in all states through direct aid. Congress is also attempting to
ease the transition following a cross-state Permanent Change of Station move by
mandating that military spouses and dependents qualify for in-state tuition rates
immediately upon moving into a state. Prior to July 1, 2009, the exemption of mili-
tary spouses and dependents from any residency requirements of public institutions
of higher education was left to the discretion of the states or in some cases individual
institutions.
The expected magnitude of response in college attendance by military spouses
to these price changes is uncertain. The existing literature has provided several
estimates for the price sensitivity of demand for higher education of civilian 18- to
24-year-olds. These estimates consistently have found that a $1,000 (in year 2000
dollars) decrease in tuition costs is associated with a 3 to 5 percentage-point increase
56
in attendance.1 However, there are systematic differences between the military and
civilian populations that likely lead to differences in price sensitivity between these
populations. For example, military families have to deal with the unique stress of
deployment. Military families also move frequently, often over large distances. The
distance and frequency of these moves affect the socioeconomic outcomes of military
spouses. Military spouses have reported that these frequent moves are an impedi-
ment to furthering their education. Due to these differences and recent legislation
targeting spouses and dependents of military personnel, it is beneficial to identify
the price sensitivity of higher education of this specific group. Fortunately, the fre-
quent moves undertaken by military spouses also allow for a unique identification
strategy of their price sensitivity of demand for higher education.
Some previous studies use cross-state variation in the cost of in-state 2- and
4-year public colleges, along with cross-state variation in high school graduates’
rate of matriculation to college, to estimate the price sensitivity of demand for
higher education. However, this method may be subject to omitted variable bias
due to other characteristics of states correlated with both costs at public colleges
and the state’s matriculation rate. For example, states that provide more funds to
subsidize public colleges and universities, resulting in lower tuition costs, may also
spend more money on public elementary and secondary education, thus preparing a
greater percentage of their population for college. Furthermore, relative tuition costs
between states have remained fairly constant over time.2 As a result, it is difficult
1There is some variation in demand sensitivity from other sources of changes in price.
2See the Appendix for detailed tuition estimates from 2000 to 2008. It is important to note
that there were substantial spikes in college tuition in 2009 due to the financial crisis. However,
this date is outside the time frame used for the analysis in this paper.
57
to control for individual state effects without losing much of the variation in costs.
Other researchers have attempted to control for these issues and other omitted
variable bias problems by using other variations in costs, such as the introduction
or elimination of tax breaks, grants, or scholarships that change the tuition prices
paid by particular groups across multiple states. This would allow researchers to
control for state-specific effects and mitigate the omitted variable bias.
This paper takes a different methodological approach due to a natural exper-
iment unique to the subpopulation studied. Most military personnel experience a
mandated move every 24 to 48 months. These moves are often across state lines,
and military personnel have little input into which installation they are assigned
to.3 With the existing variation in tuition costs across states, examining military
families experiencing a Permanent Change of Station will constitute a natural ex-
periment. Additionally, four states have changed their laws to exempt military
spouses from the residency requirements for in-state tuition during the time period
examined. These changes lead to drastic reductions in the cost of attending public
colleges, therefore providing additional variations in cost.4 However, the majority
of the variation in cost used to identify the price sensitivity of demand for higher
education among military spouses comes from variation between the states rather
then within states over time.
The key assumption of this paper is that the cross-state moves by military
3The one common exception is if a member of the military has a dependent currently enrolled
in his or her senior year of high school. If this is the case, the family is able to delay the move up
to a year to allow the child to finish his or her education.
4For example, when Virginia changed their policy in 2007, the enrollment-weighted average cost
of attending a 4-year public school dropped from $13,899 to $5,676.
58
families as a result of Permanent Change of Station orders result in exogenous vari-
ation in the price military spouses face for higher education. The military requires
that enlisted personnel relocate at least once every 3 years, but not more than once a
year. Within rank and occupation, the military asserts that all members are equally
likely to be relocated to a particular base. If so, military spouses would be moved to
high- or low-cost states independent of their socioeconomic characteristics. Lleras-
Muney (2010) tested this claim and found that, at least for enlisted servicemen,
individual characteristics observed at the time of relocation were uncorrelated with
base of relocation. This systematic placement of military personnel is not random
assignment, but may be exogenous with respect to the military spouses’ educational
preferences.
This chapter uses individual-level data matched at the state level with tuition
data for the period 2000-2008. I find that military spouses appear to be responsive
to changes in the cost of 2-year public institutions, but not to changes in the cost
of 4-year colleges. Specifically, a $1,000 decrease in the cost of a 2-year school is
associated with a 1–1.5 percentage-point increase in the probability of a military
spouse attending college.5
5Unfortunately, the data do not inform us if the schools being attended are 2-year or 4-year
institutions.
59
2.2 Background and Literature Review
2.2.1 Previous Price Sensitivity Estimates
Several studies provide estimates for the price sensitivity of a year of college
education. The price sensitivity appears to vary by the source of variation in price,
but sensitivity from a change in the in-state tuition price, the source of variation
examined in this chapter, is consistently estimated at around a 4 percentage-point
increase in matriculation for a $1,000 decrease in price (in year 2000 dollars). I refer
the reader to the first chapter of this dissertation for a detailed review.
There are two other findings from the price sensitivity literature that are rel-
evant to this chapter. The first pertains to which price matters. Kane (1995) used
between-state differences in public tuition at 2-year and 4-year colleges and found
that the marginal price, the price that determines whether or not someone attends
college at all, is that of tuition at a 2-year college and not a 4-year college. Second,
given the age range of the population studied here, Seftor and Turner’s (2002) re-
sults from examining the effect of the Pell Grant program are germane. Seftor and
Turner found non-traditional students, defined as 25- to 30-year-olds, had roughly
twice the price sensitivity of demand as their younger counterparts from this vari-
ation in price. Accordingly, this study looks for heterogeneous responses by age
among military spouses.
Prior to this paper, the literature examining the socioeconomic outcomes for
military spouses has focused on the work prospects of military spouses based on
60
different policies or the level of education of the spouse.6 However, little work has
been done examining the responsiveness of military spouses to policies aimed at
increasing their educational attainment. Examining military spouses’ price sensi-
tivity of demand for college will add to the prior estimates of this sensitivity that
traditionally focus on 18- to 24-year-olds without differentiating by marital status
or said spouse’s membership in the armed forces.7 It would not be appropriate to
extend the estimate of the price sensitivity of other populations to military spouses
due to the unique challenges military families face. Thus, the primary addition of
this work will be to provide an initial estimate of price sensitivity of demand for
college among military spouses.
The importance of the price sensitivity of higher education of this particular
group has recently increased due to policy changes in the Post-9/11 GI Bill that
was passed in 2002, the National Defense Authorization Act for Fiscal Year 2009,
and the Higher Education Opportunity Act of 2008. The Post-9/11 GI Bill allows
service members to transfer educational benefits to their spouses if they have served
at least 6 years and agree to serve an additional 4 years. On August 1, 2009, the first
military personnel who enlisted under the new GI Bill will had served the required
6 years and had the option to transfer benefits upon re-enlistment. Additionally,
the National Defense Authorization Act for Fiscal Year 2009 set up Military Spouse
Career Advancement Accounts (MyCAA), beginning in 2010. The MyCAA program
6The RAND Corporation has published a monograph, “Working Around the Military,” which
provides a summary of some of the challenges military spouses face in the workforce.
7Several studies use an individual’s own military service and his corresponding relationship to
the draft or eligibility for a GI Bill benefits to examine the relationships between these factors and
an individual’s own educational outcomes, but not the outcomes of his family members.
61
provides up to $2,000 a year for up to 2 years to help cover the costs of tuition
and fees incurred by military spouses furthering their education.8 Meanwhile, the
Higher Education Opportunity Act of 2008 stipulates that all public institutions that
receive federal funding must exempt military spouses from residency requirements
for in-state tuition as of July 1, 2009. Furthermore, once a spouse is enrolled and
paying in-state tuition, she will continue to pay the in-state tuition rate as long
as she remains continuously enrolled at the institution, even if the service member
is reassigned outside the state. This law ensures that the military spouses in the
handful of states that have not already passed similar legislation will now be eligible
for lower in-state rates when moving with their spouses across state lines for military-
mandated moves. An accurate measure of the price sensitivity of higher education
of military spouses is necessary to understand the extent to which these additional
benefits will induce more members of military families to pursue a college degree.
This paper aims to estimate this measure and predict whether the policy changes
set forth in the Post-9/11 GI Bill and the National Defense Authorization Act for
Fiscal Year 2009 will entice more military spouses to further their education.
2.2.2 Tuition Prices Faced by Military Spouses
Before estimating the response of military spouses to the price of higher ed-
ucation, it is important to determine the prices these individuals actually face. As
of 2000, 41 states exempted military spouses from the waiting period to qualify for
8Initially the cap for the scholarship was $3,000 a year for up to two years, but the program
did not have sufficient funds to maintain that funding level. There are also additional eligibility
criteria and restrictions on the usage of MyCAA funds which limit the use towards certificates,
licenses, or an Associate’s degree.
62
in-state tuition at public colleges upon moving into the state as a result of a Per-
manent Change of Station. Over the time frame used, 2000-2008, four more states
extended this exemption to military spouses.9,10 Military spouses who attend college
tend to do so at public institutions.11 As a result, the tuition and fees at public
institutions in the current state of residence can serve as a fairly accurate measure of
the financial cost to attend college as long as the proper rate (in-state/out-of-state)
is assigned.
In determining the price faced, it is also important to examine any possible
sources of financial aid available specifically to military spouses. If individual states
offered scholarships targeted at military spouses, the use of public college tuition and
fees would not accurately represent the costs military spouses face. According to
the U.S. Department of Defense, there were no regular tuition assistance programs
for the spouses or dependents of Army personnel stationed in the United States as
of 2007 (Department of Defense, 2007). There was a Spouse Education Assistance
Program through the Army Emergency Relief program, but it was only for spouses
of soldiers stationed and living in Europe, Japan, and Korea (Department of De-
fense, 2007). There are some branch-specific scholarships for military spouses, but
eligibility for these scholarships does not appear to depend on the state of residence
and thus would not affect relative cost differences between states.12
The data used for this analysis are from 2000 to 2008, and thus the estimates
9Specifically, Georgia in 2003, Virginia in 2006, and Indiana and Kansas in 2007.
10Figure B.3 in the Appendix graphically illustrates the legislative landscape on granting in-state
tuition to military spouses between 2000 and 2008.
11In the sample of military spouses used in this paper, 85 percent of those who attend college
report attending a public institution.
12Financial Aid Finder, retrieved from: http://www.financialaidfinder.com
63
for the tuition costs faced are not influenced by the Post-9/11 GI bill or the National
Defense Authorization Act for Fiscal Year 2009. Data from 2009 onward would
have to correct for the eligibility and possible take-up of benefits from these pieces
of legislation moving forward. The previous GI Bill had programs that offered
financial aid for family members of military personnel who were killed or disabled
during service. However, since the data examined are limited to families with an
active duty service member currently in the family, this program does not apply to
the individuals in this study.13 As a result, the tuition and fees at public institutions
in the current state of residence appear to remain a fairly accurate measure of the
financial cost military spouses face in order to attend college.
2.2.3 Characteristics of Military Spouses
Military spouses tend to be more highly educated than their civilian coun-
terparts. Ninety-seven percent of spouses have completed at least a high school
education, compared to 84 percent of all adults 25 years or older in the nation.14
The typical spouse of a service member had at least some college experience (71
percent). Furthermore, the majority of spouses indicated that furthering their edu-
cation was a goal (78–95 percent depending on branch of service).15 This evidence
suggests that military spouses may have different preferences for education than the
13In 2008, the National Military Family Association began awarding scholarships of up to $1,000
to spouses of Uniformed Service members (active duty, National Guard and Reserve, retirees and
survivors) to obtain professional certification or to attend post-secondary or graduate school. There
were 350 scholarships awarded in 2008 and 293 scholarships awarded in 2009, but as the awards
are to be used in the following academic year, none of the observations in this study would have
received one of these scholarships yet.
14U.S. Census, 2005
152006 Survey of Active Duty Spouses: Fact Sheet
64
traditional 18- to 24-year-olds who have traditionally been the focus of estimates
for the price sensitivity of higher education. Married individuals in general are a
selected group even before the added stresses of being married to a spouse with a
job in the military.
The U.S. Department of Defense periodically surveys the spouses of military
personnel in order to better address their specific needs. The data provided by the
2006 Department of Defense Survey of Active-Duty Spouses also give preliminary
insight into the enrollment decisions of military spouses (DMDC, 2007). According
to the results of the survey, over 18 percent of military spouses were in school and
over half of those not in school expressed a desire for additional education. Of those
who were not currently enrolled but expressed a desire to further their education,
80 percent cited the cost of education as a reason for not attending school. This
evidence suggests that military spouses will be price sensitive. Family responsi-
bilities were the second most common response, cited by just over 70 percent of
respondents. Thus, controlling for household characteristics that increase family re-
sponsibilities of military spouses is important when modeling the college attendance
decision. Factors that increase family responsibilities include the size of the fam-
ily, the presence of a newborn, or the deployment of the military service member.
Fortunately, data are available to control for the first two factors listed. However,
deployment information is not available in the data used in this paper.
There are also unique costs and challenges that military spouses face if they
decide to return to school. The frequent moves required by the military make it dif-
ficult to finish a degree at one institution. Often when transferring between schools,
65
not all credits are accepted and courses have to be retaken. Additionally, the degree
or certification they are pursuing may not be offered at schools close to their new
location. Deployment of the military spouse also leaves the civilian partner as the
sole adult in charge of the household, a responsibility civilian spouses do not tradi-
tionally face. These differences between military spouses and the civilian population
suggest that previous estimates for the price sensitivity of higher education based on
the responsiveness of 18- to 24-year-old civilians may not be applicable to military
spouses.
2.2.4 Distance to Nearest School
Previous research has found that men who were raised in labor markets with
a nearby 4-year college have statistically significant higher levels of education and
earnings (Card, 1993). The estimated effect of distance on community college en-
rollment of adults in urban areas is particularly high (Jepsen & Montgomery, 2009).
Given that the distance to an institution of higher education is associated with edu-
cational outcomes, one may worry that the distance military spouses have to travel
may influence their enrollment rates. Furthermore, if the average distance between a
military installation and the closest college campus is correlated with college tuition
prices at the state level, then (if not properly controlled for) this correlation may
bias the results of this study. Unfortunately, the Census data used in this study
only list the state of residence of the respondent and do not identify the military
installation or more specific location information of the respondent, which would
66
be necessary to calculate the distance each military spouse might have to travel to
attend the closest 2-year or 4-year school. However, I obtained data that allowed me
to estimate the minimum distance a military spouse might have to travel to attend
a public college in person.
I started with the five-digit ZIP code for each of the 195 military installations in
the 50 states and the District of Columbia large enough to justify an adult education
center to service the military personnel on base. I then obtained the five-digit ZIP
code from the address of record for each public 2-year and 4-year college campus for
the 2008 academic year from the Integrated Postsecondary Education Data System
(IPEDS) administered by the National Center for Education Statistics, a part of
the Institute for Education Sciences within the U.S. Department of Education. I
then calculated the distance from each military installation to the closest 2-year and
4-year public school based on the latitude and longitude of the centroid of each ZIP
code.
Based on these calculations, the average distance from one of these 195 military
installations to the closest 2-year or 4-year public school was 8.7 miles, with a me-
dian distance of just under 12 miles. Furthermore, over 98 percent of these military
installations were within 60 miles of a public 2-year or 4-year school. Comparing this
to the average distance from home an independent student or a married indepen-
dent student lives from the school she attends—105 and 114 miles, respectively—it
appears that the distance military spouses might have to travel to attend a public
institution of higher education should not be an issue.16 Furthermore, many of the
16The source for the average distance traveled by independent students is the Profile of Under-
67
adult education centers at these military installations allow military spouses to use
the services if there is excess capacity after all active duty service personnel have
been accommodated.17 This includes use of the computer centers to facilitate on-
line courses and in-person classes offered on the base itself by some higher education
providers.
2.3 Data
This analysis is based on data from two main sources: the U.S. Census and
the Integrated Postsecondary Education Data System (IPEDS). Two different U.S.
Census surveys are used in this analysis: the annual American Community Survey
(ACS) from 2000 to 2008 and the 2000 Decennial Census. The ACS is based on the
census long-form survey, with minor modifications to allow the survey to provide
annual updates. These sources provide information on the location, income, type,
and number of people in the surveyed household, as well as age, gender, race, ed-
ucational attainment, enrollment status, employment status, industry, occupation,
income, fertility, and migration information for each individual.
The primary difference of interest between the two surveys pertains to the
migration variables. The ACS records whether a respondent has moved within the
last year and, if so, where she resided in the previous year. The decennial survey
graduate Students: 2007–08, produced by the U.S Department of Education. This report also
estimates the median distance traveled for all independent and married independent students to
be 14 miles, which is more than the median distance from a military installation to a public college
reported above.
17A military spouse’s ability to commute to school with her active duty spouse on his way to
work may mitigate the potentially high effect of distance on adult community college enrollment
in urban areas found by Jepsen and Montgomery (2009).
68
reports whether the individual has moved in the last 5 years and, if so, what state
she lived in 5 years ago. The decennial survey also measures how long ago an
individual moved into their current house, so I can tell if she have moved within the
last year. However, if there was more than one move in the last 5 years, it would
not be clear whether the most recent move was across state lines. For the purposes
of sample selection, only individuals who moved across state lines in the last year
were included. I assume that if the respondent moved in the last year and reported
residing in a different state 5 years ago, then their last move was across state lines
and they are included in the sample. This leaves a possibility for measurement
error.18 Individuals may have simply moved within the state in the last year. Such
individuals would be assigned the out-of-state rate if they reside in states that had
a 1- or 2-year residency restriction for in-state tuition without a military spouse
exemption, even though they would actually be subject to the in-state tuition rate.
IPEDS is compiled by the U.S. Department of Education and managed by the
National Center for Educational Statistics. This database provides a wide variety of
self-reported information on academic institutions in the United States. For the pur-
pose of this paper, information about tuition, enrollment, location, and sector was
extracted for all degree-granting, post-secondary institutions in the United States.19
With this information, it is possible to calculate the average tuition and fees charged
18Removing the 2000 Decennial Census data from the analysis to mitigate this measurement error
would come at a high cost. The full implementation of the ACS, sampling 2.9 million housing units
annually, did not begin until 2005. From 2000 to 2004, only 740,000 to 900,000 housing units were
sampled annually.
19The poor quality of IPEDS data is well known. Many colleges and universities place a low
priority on accurately fulfilling their reporting requirements, and discrepancies are pervasive. How-
ever, these are the best data available and are widely used despite their flaws.
69
to in-state and out-of-state undergraduate students in each state, weighted by en-
rollment and calculated by the sector in which the institution participates. The
sector variable makes it possible to define a school as public or private and as either
a 2-year or 4-year institution. I calculated the enrollment-weighted state averages
of tuition and fees at 1,027 2-year and 655 4-year public degree-granting, post-
secondary institutions. Both price paid and price published were extracted and
used as explanatory variables in separate regressions and are summarized below in
Table 2.1.20 Price paid is the preferred variable for models shown in this chapter, but
results using price published are also included in the Appendix. Figures B.4 and B.5
in the Appendix also graphically display the variation in enrollment-weighted state
average price paid and average enrollment rates of military spouses in 2008.21,22
2.3.1 Sample Restrictions
An individual is categorized as a military spouse if he or she is in a married
family household and indicates that his or her spouse is present. Furthermore,
an individual’s spouse must belong to a military industry. Finally, the individual
studied, henceforth known as the military spouse, must not report his or her own
industry as military.23 The sample was then restricted to only include individuals
20Table 2.1 reflects the average prices in each state across years, properly assigning in-state and
out-of-state prices in each state-year combination and weighting by the sample within each year.
21Vermont is the only state without a major military base or installation, so observations were
omitted from this state. This unique characteristic of Vermont may explain its status as an outlier
in both figures.
22Meanwhile, Tables B.9 through B.12 in the Appendix provide the enrollment-weighted average
price paid and price published at 2- and 4-year public institutions for each state, year combination.
23A military industry code is classified as the industry codes for the five branches of the armed
forces or “Armed Forces, branch not specified.”
70
Table 2.1: Mean Price of Tuition and Fees for Military Spouses
Two-Year Costs Four-Year Costs
State Sample Size Percent Enrolled Price Paid Published Price Price Paid Published Price
Alabama 165 17.6% 449 486 3291 3098
Alaska 224 13.8% 2177 2080 3188 2942
Arizona 186 18.3% 1460 1344 2752 2610
Arkansas 31 12.9% 1531 1396 3346 3101
California 772 17.0% 381 386 2898 2797
Colorado 240 14.2% 1783 1798 3289 3017
Connecticut 67 7.5% 5932 5789 12174 12120
Delaware 31 12.9% 1790 1704 5224 4962
District of Columbia 26 23.1% N/A N/A 2078 2008
Florida 463 14.5% 1496 1363 2151 1994
Georgia 498 12.9% 2667 2533 7222 6811
Hawaii 369 12.7% 1103 1063 2871 2681
Idaho 25 20.0% 2334 2260 2828 2752
Illinois 98 10.2% 5140 4782 4891 4593
Indiana 24 8.3% 3741 3706 10527 10167
Iowa 19 10.5% 2357 2204 3868 3646
Kansas 144 11.8% 2925 2880 8476 8377
Kentucky 231 13.9% 1577 1441 3493 3183
Louisiana 176 15.3% 982 942 2907 2558
Maine 39 20.5% 2615 2483 4620 4256
Maryland 227 9.7% 4241 4000 4880 4744
Massachusetts 38 7.9% 6928 6918 11475 11429
Michigan 33 21.2% 3478 3141 12121 11820
Minnesota 10 10.0% 2600 2407 4157 3852
Mississippi 94 18.1% 1275 1114 3269 3153
Missouri 119 20.2% 2300 2097 4390 4062
Montana 21 23.8% 2454 2349 3479 3280
Nebraska 53 18.9% 1476 1405 3431 3228
Nevada 98 16.3% 1416 1286 1948 1806
New Hampshire 9 22.2% 4437 4164 6959 6606
New Jersey 69 10.1% 3502 3396 5997 5552
New Mexico 85 15.3% 1048 970 2724 2462
New York 158 13.9% 2836 2775 3883 3797
North Carolina 658 14.7% 975 876 2613 2425
North Dakota 51 13.7% 2216 2100 3219 3002
Ohio 77 14.3% 2561 2559 5309 5073
Oklahoma 135 20.0% 1448 1430 2472 2310
Oregon 21 9.5% 1910 1789 4011 3860
Pennsylvania 38 5.3% 4473 4292 6235 5904
Rhode Island 36 2.8% 1906 1823 4718 4494
South Carolina 213 16.4% 4469 4305 12167 10434
South Dakota 44 11.4% 2855 2578 3836 3589
Tennessee 182 12.6% 1651 1527 3349 3071
Texas 749 14.0% 1522 1450 3216 3006
Utah 43 20.9% 1647 1586 2327 2228
Vermont 5 40.0% 6574 6261 16855 17011
Virginia 720 13.1% 4600 4374 11509 11230
Washington 430 14.7% 1929 1845 3545 3294
West Virginia 20 15.0% 1863 1786 2862 2694
Wisconsin 22 22.7% 2456 2314 3997 3792
Wyoming 38 7.9% 1454 1357 2620 2475
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education Statistics, Integrated Postsecondary
Education Data System (IPEDS), 2000-2008, U.S. Census Bureau, Integrated Public Use Microdata Series (IPUMS), 2000 5% sample
and 2000-2008 ACS samples and U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
Notes:
(1) All values begin by calculating the enrollment-weighted average of degree-granting, public institutions within each state and year.
The in-state or out-of-state prices are used in accordance with the prices a military spouse who moved into the state as part of a
military mandated move within the last year would have faced in each state-year combination. Each value is then indexed to year
2000 dollars using the Bureau of Labor Statistics CPI-U. Finally, the sample-weighted values of each state are reported
(2) The Community College of the District of Columbia was not established until after the time period analyzed. Thus, there were
not appropriate 2-year tuition costs for residents of the District of Columbia.
71
that had moved between states in the past year. This restriction was performed
to ensure that individuals had moved at least once and that their current state
of residence, and thus the price that they faced to attend college, were exogenous
with respect to the educational characteristics of the state they grew up in. If a
military spouse had not yet experienced a Permanent Change of Station across state
lines, she could still be residing in her home state of record where she received her
primary and secondary education. By requiring at least one observable cross-state
move, this paper assumes the military has provided a natural experiment, even if
the most recent move happens to place the individual back into their original home
state of record.
The sample was also restricted to individuals between the ages of 18 and 40
with at least a GED or high school diploma and who had not yet completed a Bach-
elor’s degree. This was to ensure that the individuals were eligible to apply for an
undergraduate degree. The price sensitivity for graduate school was not considered
in this study. Table 2.1 lists the resulting sample in each state and the share of that
sample enrolled in school. To ensure that the cost of tuition might influence the
enrollment decision of the military spouse, I removed observations from the state
of Vermont, which does not have a major military base or installation. It is likely
that military spouses who are enrolled in these schools are there for reasons other
than price and distance. Additionally, observations for which the income variable
appeared to be an outlier were removed.24 These sample restrictions resulted in
a final sample of 8,252 military spouses. This sample was further separated into
24Individuals with negative household income or household income above $300,000 were removed.
72
two age groups: 18- to 24-year-olds, the age group traditionally studied, and 25- to
40-year-olds, resulting in 2,969 and 5,319 observations respectively. For each sample
individual, I calculated the enrollment-weighted average cost of tuition and fees for
2- and 4-year degree-granting, public institutions in her current state of residence.
Each individual was assigned either the average in-state or out-of-state rate based
on the residency requirements for the state and year corresponding to the observa-
tion.25 All tuition and income variables were indexed to year 2000 dollars using the
Consumer Price Index-Urban.
2.3.2 Summary Statistics
Table 2.2 presents the summary statistics of the variables of interest for mili-
tary spouses 18 to 24 years old, broken out by school attendance. This table shows
that for the younger military spouses, those who decided to attend school faced
slightly lower costs to attend a 2- or 4-year public school. However, these differences
were not statistically significant at the α = 0.05 level. Those 18- to 24-year-olds not
in school faced a lower average unemployment rate, 4.37 percent as compared to the
4.61 percent for those who decided to attend school. This difference is statistically
significant at the 99 percent confidence level and has a t-statistic of 4.23. Those
in school were also 10 percentage-points more likely to have previously obtained an
Associate’s degree. This difference is statistically significant, with a t-statistic of
5.70.
25All states except for Tennessee have at least a 1-year residency restriction for civilians, while
the residency requirements for spouses and dependents of military personnel have changed over
time. Table B.13 in the Appendix illustrates the state-year combinations for which military spouses
immediately qualified for in-state tuition rates.
73
Table 2.2: Summary Statistics for 18- to 24-year-old Military Spouses
Not in School In School
Variable Mean Std. Dev. N Mean Std. Dev. N
Two-Year Costs 2047 1415.95 2438 1946 1354.99 531
Four-Year Costs 4569 3083.12 2444 4405 2960.26 534
Other Household Income 23650 16859.78 2444 23310 10941.95 534
Age 21.64 1.742 2444 21.60 1.763 534
Male 4.6 % 0.210 2444 4.7 % 0.211 534
White 89.6 % 0.305 2444 91.6 % 0.278 534
Unemployment Rate 4.37 % 0.011 2444 4.61 % 0.012 534
Associate’s Degree 7.4 % 0.263 2444 17.2 % 0.378 534
Newborn 22.9 % 0.420 2444 10.9 % 0.311 534
Number of Children Under 5 0.665 0.754 2444 0.451 0.671 534
Looking for Work 15.9 % 0.366 2444 16.1 % 0.368 534
Married to an Officer 2.6 % 0.160 2444 3.4 % 0.181 534
Spouse– High School Graduate 46.9 % 0.499 2444 33.5 % 0.473 534
Spouse– Some College 45.3 % 0.498 2444 55.1 % 0.498 534
Spouse– College Degree 5.1 % 0.220 2444 8.4 % 0.278 534
Spouse– Graduate Degree 0.5 % 0.067 2444 1.1 % 0.106 534
Enrolled in a Public Institution N/A N/A 2444 86.1 % 0.346 534
NOTE: Table is restricted to individuals with at least a GED and less than a Bachelor’s degree.
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census
Bureau, Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS
samples and U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and
Unemployment Data.
Those not in school also had significantly higher rates of a newborn child in
the household, as well as a higher average total number of children under age 5
in the household. These differences have t-statistics of 7.54 and 6.52 respectively.
Those not in school were more likely to have a spouse with a high school education
as their final educational attainment and less likely to have a spouse with some
college or a college degree. These differences have t-statistics of 5.87, 4.12, and 2.57
respectively. Finally, of those in school, 18- to 24-year-old military spouses attend
public, degree-granting institutions more than 86 percent of the time. This high
rate of attendance at public institutions by military spouses validates the use of the
average enrollment-weighted cost of tuition and fees at public institutions as the
primary measures for tuition costs.
74
Table 2.3 presents the summary statistics of the variables of interest for mili-
tary spouses 25 to 40 years old, broken out by school attendance. This table shows
that for older military spouses, those who decided to attend school faced slightly
lower costs to attend a 2- or 4-year public college. Those not in school faced an
average cost at a 2-year public school of $2,141, as opposed to $1,966 for those who
attended college. The $175 difference is statistically significant, with a t-statistic of
3.03. Those not in school faced an average cost at a 4-year public school of $4,635,
as opposed to $4,372 for those who attended college. This $263 difference is also
statistically significant, with a t-statistic of 2.14. Those in school were also, on av-
erage, a year younger. This difference is statistically significant, with a t-statistic of
5.51. Within the the 25- to 40-year-old sample, 18.9 percent of those not in school
had an Associate’s degree, compared to 30.2 percent of those in school. This dif-
ference is statistically significant, with a t-statistic of 6.03. As was the case with
the younger sample, those not in school in this age group had a newborn child in
the household at significantly higher rates, as well as a higher average total number
of children under age 5 in the household. These differences have t-statistics of 2.90
and 4.47 respectively. Greater family obligations appear to be correlated with lower
school attendance. Those not in school were more likely to have a spouse with a
high school education as their final educational attainment and less likely to have
a spouse with some college. These differences have t-statistics of 4.56 and 3.60 re-
spectively. The differences between the rates of those with a spouse with a college
or graduate degree were not statistically significant. Finally, those in this age group
75
who are in school attend public institutions more than 83 percent of the time.26
Table 2.3: Summary Statistics for 25- to 40-year-old Military Spouses
Not in School In School
Variable Mean Std. Dev. N Mean Std. Dev. N
Two-Year Costs 2141 1473.57 4657 1966 1381.01 662
Four-Year Costs 4635 3171.13 4671 4372 2936.85 665
Other Household Income 35763 20966.82 4671 35109 18789.50 665
Age 31.35 4.394 4671 30.39 4.182 665
Male 6.9 % 0.253 4671 9.5 % 0.293 665
White 87.9 % 0.326 4671 88.4 % 0.320 665
Unemployment Rate 4.29 % 0.011 4671 4.35 % 0.011 665
Associate’s Degree 18.9 % 0.391 4671 30.2 % 0.460 665
Newborn 12.4 % 0.330 4671 8.9 % 0.285 665
Number of Children Under 5 0.657 0.756 4671 0.523 0.718 665
Looking for Work 9.8 % 0.297 4671 9.9 % 0.299 665
Married to an Officer 4.3 % 0.204 4671 3 % 0.171 665
Spouse– High School Graduate 20.7 % 0.406 4671 14 % 0.347 665
Spouse– Some College 56.6 % 0.496 4671 63.8 % 0.481 665
Spouse– College Degree 13.9 % 0.346 4671 14 % 0.347 665
Spouse– Graduate Degree 7.4 % 0.262 4671 7.2 % 0.259 665
Enrolled in a Public Institution N/A N/A 4671 83.2 % 0.375 665
NOTE: Table is restricted to individuals with at least a GED and less than a Bachelor’s degree.
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census
Bureau, Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS
samples and U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and
Unemployment Data.
2.4 Empirical Strategy
The empirical strategy in this chapter is designed to utilize the variation across
states in the cost of 2- and 4-year public institutions to identify the price sensitivity
for higher education of military spouses.
To investigate the impact of changes in costs for military spouses on their
college attendance rates, I run Ordinary Least Squares regressions, exploiting the
variation in costs at 2- and 4-year public colleges across states.
26The school characteristics data from the Census only provides information on whether a school
attended is public or private. As the sample is limited to those with at least a high school diploma
or GED but who have yet to complete a Bachelor’s degree, the school can be assumed to be post-
secondary. However, the rate of attendance at 2-year vs. 4-year schools for this sample cannot be
determined directly from these data.
76
The model takes the following form:
(1)Attendanceist = α + β1(2− Y earCosts)st + β2(4− Y earCosts)st + β3Xi
+β4(UnemploymentRate)st + β5Zi + β6Mi + τt + ςs + εist
In this model, Attendanceist is an indicator of whether an individual i is currently
attending school in state s in year t. The regressors of interest are 2− Y earCostsst
and 4−Y earCostsst, which are the enrollment-weighted average prices paid by full-
time undergraduates at public institutions in state s in year t.27 All dollar values
are adjusted to year 2000 dollars using the Bureau of Labor Statistics’ Consumer
Price Index-Urban and divided by 1,000.28
The vector Xi consists of demographic controls for age, gender, race, and
household income not attributed to the military spouse.29 This income variable
measures the income support the individual has to draw upon when deciding whether
or not to attend college. I control for age and its square. Indicator variables are also
included, measuring if the spouse is male or white to control for gender and race.
The yearly average of the unemployment rate in state s and year t was also included
as a control for labor demand. A strong labor market could indicate higher forgone
earnings while an individual attends college, thus discouraging enrollment.
27Price published instead of price paid is substituted into the regression, and the tables showing
those results are included in Tables B.19 through B.23.
28Using the log of dollar values instead of the level in thousands does not appear to have much
impact on our primary regressors of interest. The tables presenting the results with those specifi-
cations are located in Tables B.24 through B.33.
29To calculate this variable, an individual’s reported income was subtracted from total household
income and divided by 1,000.
77
The vector Zi consists of controls that are possibly endogenous and are in-
cluded in only certain specifications of the model. These controls include an indi-
cator variable for having a newborn child under the age of 1 in the household. An
indicator variable measuring if the individual reports that they are currently search-
ing for work and a discrete variable measuring the number of children under age 5
in the household are also included. These variables control for family responsibili-
ties that may place time constraints on the military spouse, reducing their college
attendance rates.
Vector Mi consists of a series of indicator variables that control for the educa-
tional attainment and officer status of the military spouse. Relocation to a particular
base may be independent of individual characteristics after controlling for rank and
occupation. However, exact rank or occupation within the military is not specified
in Census data. Only the distinction between an officer and an enlisted individual is
available to distinguish rank. Specific Military Occupational Specialty (MOS) is not
available, but most MOSs have educational requirements. Educational attainment
indicator variables are included to mitigate the absence of MOS data. Finally, τt
is a binary indicator for the year and ςs is a binary indicator that captures fixed
effects associated with state s; these indicators are included in certain specifica-
tion of the model. Unfortunately, because the average costs of college in individual
states do not change much over time, the inclusion of this variable will eliminate
most of the variation in the data, as seen in previous studies. An advantage of the
identification strategy is the movement of the spouses into states is assumed to be
exogenous with respect to their preferences for higher education. Therefore, the pre-
78
ferred specification of the model does not include state fixed effects, but the results
of the regressions including state fixed effects are shown to illustrate the advantage
of this identification strategy and what the results would be if the between state
variation in price was not used and only the variation over time, particularly from
those states that changed the residency requirements for military spouses during
the period of observation, was available for identification. In all specifications the
standard errors are clustered at the state*year level.
The identifying assumption of equation (1) is that the average costs of higher
education in each state during this time period are not correlated with other state-
level characteristics that are not controlled for in the regression but that might affect
the college enrollment decision of military spouses. A possible source of omitted
variable bias would be different deployment rates for service members stationed
in different states. Additionally, the average distance between the bases to which
active-duty military members are assigned and the public colleges in the state could
be a concern in specification of the model in which ςs is not included. Unfortunately,
data limitations do not allow for the identification of military families for which a
spouse is deployed or the specific military installation or location within a state an
individual is assigned to in order to measure the distance to local colleges.
79
2.5 Results
Table 2.4 displays the estimation results using the full sample.30 Column 1
displays the results of estimating attendance as a function of only the 2- and 4-year
enrollment-weighted average price paid at public institutions. The cost at 4-year
schools does not enter the model with statistical significance, but the estimated
coefficient on 2-year costs suggests that military spouses are responsive to this price.
Note that this finding matches the finding of Kane (1995) that the price which
determines whether or not someone attends college at all is that of tuition at a
2-year college rather than a 4-year college. This is especially true because this
sample contains non-traditional students, a population that is targeted by 2-year
institutions. All other regressions control for the state unemployment rate and the
demographic characteristics of age, age2, gender, race, and other household income.
Column 2 adds the aforementioned unemployment rate and demographic char-
acteristics as independent variables. The coefficient estimate on 2-year costs is now
-0.012 instead of -0.014 and is still statistically significant at the 0.05 level instead
of the 0.01 level. A Chi-Square test concludes that there is no statistically signifi-
cant difference between these coefficient estimates on 2-year costs. The coefficient
estimate on 4-year costs still lacks statistical significance.
The specification reported in column 3 adds having a newborn, looking for
work, and counting the number of children under age 5 as independent variables.
30Tables reporting the estimated coefficients for the collapsed variables and specifications using
price published instead of price paid or the log of monetary variables instead of the level are
included in Tables B.19 through B.33.
80
Table 2.4: Price Sensitivity of Military Spouses for Price Paid for College Tuition, 18- to
40-year-olds
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.014∗∗ -0.012∗ -0.013∗ -0.014∗∗ -0.013∗∗ -0.003
(0.005) (0.005) (0.005) (0.005) (0.005) (0.015)
Four-Year Costs 0.003 0.004 0.004 0.004 0.003 -0.006
(0.002) (0.002) (0.002) (0.002) (0.002) (0.007)
Additional Controls
Unemployment Rate X X X X X
Demographic Variables X X X X X
Familial Responsibilities X X X X
Spousal Characteristics X X X
Year X X
State X
N 8288 8288 8288 8288 8288 8288
Adj. R2 0.001 0.010 0.020 0.027 0.029 0.029
Standard errors in parentheses. The standard errors are clustered at the state*year level.
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
NOTE: Demographic Variables consist of Age, Age2, Male, White, and Other Household Income.
Familial Responsibilities consist of Having a Newborn, Number of Children Under Age 5, and
Looking for Work. Spousal Characteristics consist of Officer, Spouse–High School Grad, Spouse–
Some College, Spouse–College Degree, and Spouse–Graduate Degree.
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census
Bureau, Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS
samples and U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and
Unemployment Data.
81
Column 4 adds the indicator variables for whether the spouse serving in the military
is an officer as well as that spouse’s educational attainment grouped into high school
graduate, some college, college graduate, and graduate degree. Column 5 adds year
fixed effects. Again, in each of the specifications, including the preferred specification
in column 5, the data demonstrate that school attendance is negatively related to
the cost of 2-year colleges and not responsive in a statistically significant way to the
cost of 4-year colleges. These specifications yield estimates that a $1,000 (in year
2000 dollars) increase in the costs of 2-year public colleges is associated with a 1.3
to 1.4 percentage-point decrease in attendance. Given the low attendance rate of
the sample, 14 percent, this translates to roughly a 9 percent change, which is line
with previous estimates for a similar change in direct costs.31
Column 6 adds state fixed effects into the model. This estimation controls
for differences across states that are not explicitly captured by the regressors of
the model. However, as the 2- and 4-year cost variables do not change much for
individual states over time, the inclusion of these fixed effects eliminates most of
the variation in the cost variables. In this specification, the coefficient estimate
on 2-year costs is now -0.003 and is not statistically significant. The coefficient
estimate on 4-year costs still does not enter the model with statistical significance.
31To further translate this estimate into an elasticity term may add unnecessary complication.
To do so would first involve a definition of total cost. Whether this should include only the relevant
tuition costs or also the cost of forgone earnings is unclear. The results from Kane (1995) imply
an elasticity of demand of -0.2 with respect to tuition prices alone, but of -3.9 if including the
cost of 9 months of forgone earnings. Meanwhile, results from Kane (1999) which incorporate
the opportunity costs of lost wages imply an elasticity of roughly -1. Furthermore, calculating
the opportunity costs of forgone earnings, given the unique characteristics of military spouses and
the uncertain labor market conditions around military installations, would be unlikely to produce
reliable estimates given available data.
82
This specification shows the advantage of the identification strategy that leverages
exogenous variation in state tuition costs, thereby allowing the preferred model
specification to forgo the inclusion of state fixed effects.
Table 2.5 displays the estimation results broken out by the 18- to 24-year-
old and 25- to 40-year-old samples. The coefficient estimates for the specifications
presented in columns 1 through 5 for each age subgroup are nearly identical to those
of the full 18- to 40-year-old sample. The coefficient estimates on 2-year costs for
the larger, 25- to 40-year-old sample presented in bottom panel of Table 2.5 remain
statistically significant at the 0.05 level in the specifications presented in columns
1–5. However, for the smaller, 18- to 24-year-old sample presented in the top panel
of Table 2.5, the coefficient estimates on 2-year costs no longer enter into the model
with statistical significance. This is most likely due to the reduced sample size,
which would increase the standard errors. Thus, if the sample were larger, I would
expect their estimates to be statistically significant because the coefficient estimate
remains the same as in the full sample. It is also interesting that the point estimates
for the price sensitivity for the two different age groups are almost the same. With
fewer years to use their education, the cost-benefit calculation would be different for
an older individual, yet the point estimates for the price sensitivity are almost the
same.
Table 2.6 presents the results of a specification check of the model. If an
individual already has an Associate’s degree, then she may be less interested in
attending a 2-year school because she already has the training needed for a positive
gain in the labor market. It is possible that an individual with an Associate’s
83
Table 2.5: Price Sensitivity of Military Spouses for Price Paid for College Tuition, by Age
Group
(1) (2) (3) (4) (5) (6)
18- to 24-year-olds (N= 2969)
Two-Year Costs -0.012 -0.012 -0.012 -0.014 -0.013 -0.013
(0.010) (0.010) (0.010) (0.010) (0.010) (0.027)
Four-Year Costs 0.002 0.006 0.006 0.007 0.004 -0.001
(0.005) (0.005) (0.005) (0.005) (0.005) (0.012)
25- to 40-year-olds (N= 5319)
Two-Year Costs -0.013∗ -0.013∗ -0.013∗ -0.014∗ -0.014∗ -0.000
(0.006) (0.006) (0.006) (0.006) (0.006) (0.019)
Four-Year Costs 0.002 0.003 0.003 0.003 0.002 -0.008
(0.003) (0.003) (0.003) (0.003) (0.003) (0.008)
Additional Controls
Unemployment Rate X X X X X
Demographic Variables X X X X X
Familial Responsibilities X X X X
Spousal Characteristics X X X
Year X X
State X
Standard errors are in parentheses. The standard errors are clustered at the state*year level.
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
NOTE: Demographic Variables consist of Age, Age2, Male, White, and Other Household Income.
Familial Responsibilities consist of Having a Newborn, Number of Children Under Age 5, and
Looking for Work. Spousal Characteristics consist of Officer, Spouse–High School Grad, Spouse–
Some College, Spouse–College Degree, and Spouse–Graduate Degree.
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census
Bureau, Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS
samples and U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and
Unemployment Data.
84
degree may be taking courses simply for consumption or working towards a second
Associate’s degree in a different field, but it seems more likely that an individual who
completed an Associate’s degree would primarily be interested in a 4-year school
to obtain a Bachelor’s degree if she were to continue her education.32 The top
panel of Table 2.6 shows the various model specifications for all 18- to 40-year-olds
who have not yet obtained an Associate’s degree. In columns 1–5 the coefficient
estimates for both cost variables are similar to those seen in our full sample in Table
2.4. Additionally, the coefficient estimate on 2-year costs enters in the model with
statistical significance, which indicates that those without an Associate’s degree are
responsive to the costs of 2-year schools. Column 6 includes state fixed effects. As
before, in this specification, the coefficient estimate on 2-year costs does not enter
into the model with statistical significance, but unlike in the model containing the
full sample, the coefficient estimate remains close to those appearing in columns
1–5. Meanwhile, the coefficient on 4-year costs is estimated as an imprecise zero.
The bottom panel of Table 2.6 displays the estimation results using the sample
of military spouses who have already obtained an Associate’s degree. With a much
smaller sample size, the standard errors have increased across all specifications. The
coefficient estimates on 2-year costs do not enter the model with statistical signif-
icance in any specification, but the coefficients are larger. Meanwhile, the point
estimates for the 4-year costs have remained statistically insignificant, except in col-
umn 6. In this specification, where state fixed effects are included, the coefficient
32Kane (1995) used the enrollment rates of 18- to 19-year-olds as his outcome of interest when
determining 2-year public tuition costs were driving college attendance and did not examine the
effect of 2-year public tuition costs on those with an Associate’s degree.
85
Table 2.6: Price Sensitivity of Military Spouses for Price Paid for College Tuition, by
Educational Attainment
(1) (2) (3) (4) (5) (6)
Associate’s = No (N= 6938)
Two-Year Costs -0.013∗ -0.010∗ -0.011∗ -0.012∗ -0.012∗ -0.014
(0.005) (0.005) (0.005) (0.005) (0.005) (0.016)
Four-Year Costs 0.003 0.004 0.004 0.004 0.003 0.000
(0.002) (0.002) (0.002) (0.002) (0.003) (0.007)
Associate’s = Yes (N= 1350)
Two-Year Costs -0.018 -0.018 -0.019 -0.019 -0.019 0.059
(0.015) (0.015) (0.015) (0.015) (0.015) (0.045)
Four-Year Costs 0.001 0.001 0.000 0.001 -0.002 -0.045∗
(0.007) (0.007) (0.007) (0.007) (0.007) (0.019)
Additional Controls
Unemployment Rate X X X X X
Demographic Variables X X X X X
Familial Responsibilities X X X X
Spousal Characteristics X X X
Year X X
State X
Standard errors in parentheses. The standard errors are clustered at the state*year level.
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
NOTE: Demographic Variables consist of Age, Age2, Male, White, and Other Household Income.
Familial Responsibilities consist of Newborn, Number of Children Under Age 5, and Looking for
Work. Spousal Characteristics consist of Officer, Spouse–High School Grad, Spouse–Some College,
Spouse–College Degree, and Spouse–Graduate Degree.
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census
Bureau, Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS
samples and U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and
Unemployment Data.
86
estimate on 4-year costs enters into the model negatively and with statistical sig-
nificance. The estimated coefficient of -0.045 is in line with prior estimates of the
civilian population.33 This is the only instance in which military spouses appear to
be responsive to 4-year costs. Further investigation with variation in costs not tied
to individual states may assist in unraveling these anomalies.
Given the apparent importance of the cost at 2-year public schools in deter-
mining enrollment rates for military spouses, an estimate of the share of military
spouses that are students who are enrolled within a 2-year institution would be of
interest. Unfortunately, the school characteristics data from the Census only pro-
vides information on whether a school is public or private. The rate of attendance
of military spouses at 2-year vs. 4-year schools is not available from this source.
However, a program started in 2009 that offers military spouses tuition assistance
does track the schools the recipients of these funds attend.
The MyCAA program provides up to $4,000 of tuition assistance to eligible
military spouses over 2 years. To be eligible one must be married to an active duty
service member in one of the lower pay grades.34 The spouse must be able to start
and complete their program within 3 years. The funds can only be used towards
a certification, a license, or an Associate’s degree, but does not limit the spouse
from attending a 4-year school. Data from fiscal year 2011 through 2013 of the
program show 14.4 percent of MyCAA scholarships recipients attended community
33In this instance the percentage-point change from a $1,000 change in direct cost would be
accurate, but the percent change would be uncharacteristically high given the low rate of college
attendance among this population.
34Specifically, grades E1–E5, W1–W2, and O1–O2, including the spouses of activated Guard
and Reserve members within those ranks.
87
colleges.35 Unfortunately, given the eligibility and use restrictions on the program,
these data may not be representative of the types of schools the overall military
spouse population attends. The one check available from the Census data compares
the rate of public school to private school attendance. The Census data from 2000
to 2008 reported military spouses attended public college in over 80 percent of
cases, but only 37 percent of MyCAA awardees from 2011 to 2013 attended public
institutions. This calls into question the extent to which the MyCAA sample is
representative of military spouses.
2.5.1 Conclusion
This paper offers an initial estimate of the price sensitivity of higher education
among military spouses. It uses the variation across states in the costs of 2- and
4-year public colleges to estimate this sensitivity. This analysis is based on Census
and IPEDS data from 2000 to 2008.
I find that a $1,000 (in year 2000 dollars) increase in the costs of 2-year pub-
lic colleges is associated with a 1.0 to 1.5 percentage-point decrease in attendance
by military spouses. This figure implies a price sensitivity less than half that of
the civilian 18- to 24-year-old population as defined as a percentage-point change.
However, this translates into a 7 to 10 percent change in enrollment given the base-
line level of 14 percent enrollment. This percent change in enrollment is consistent
with previous estimates which traditionally examined populations exhibiting 40 to
35During the same time period 59.8 percent of MyCAA recipients attended 4-year schools while
25.9 percent attended ”Other” forms of schools.
88
50 percent enrollment. This response is robust to various model specification.36
The primary advantage of this paper is its ability to examine the price sensi-
tivity of demand for higher education by leveraging exogenous variation in tuition
costs from military-mandated moves. This negates many of the potential omitted
variable bias issues in previous research that called for the inclusion of state fixed ef-
fects, a control that could not be included without eliminating much of the variation
in price needed for identification and was often omitted in the previous literature.37
This paper shows the challenge of trying to identify the effects of tuition changes on
enrollment relying only on within-state variation after including state fixed effects.
However, the strength of the identification strategy is also tied to this paper’s great-
est weakness: the unique characteristics and challenges faced by military spouses,
including the military-mandated moves, make the external validity of these results
to other populations suspect.
Some additional points should be kept in mind when interpreting the results.
First, the analysis has identified average effects but, due to data limitations, can-
not sufficiently examine the heterogeneity of responses by military spouses. With
additional data, breaking out the sample by branch of military service or additional
demographic variables may be of interest, along with re-examining the effect on 18-
to 24-year-old military spouses, or those who have already attained an Associate’s
36Aside from the inclusion of state fixed effects, which is unnecessary given the process by which
residency is determined for this sample and is instead done to illustrate the advantage of the
identification strategy.
37Kane (1999) included a section leveraging over a decade’s worth of data to examine the effect
of within-state variation in tuition and included state fixed effects. That section found similar
estimates for the effect of tuition changes on enrollment, but this example was not representative
of the identification strategy commonly used in the literature.
89
degree separately but with larger sample sizes. Second, there are aspects of military
life for which control variables were not available that may affect the decision to
attend school. Specifically, no variable for whether the military member was cur-
rently deployed was available, and it is possible that deployment rate could vary by
state and may even be correlated with state tuition costs. Finally, the tuition and
location variables used were noisy. Data limitations precluded being able to deter-
mine which installation the military household was assigned to or which school was
attended. This resulted in having to assign the enrollment-weighted state average
cost to each individual within the state. With more detailed information, assigning
the correct price faced and the distance required to travel may generate more accu-
rate and efficient estimates. Despite these limitations, this paper provides an initial
estimate of the price sensitivity of demand for higher education of military spouses.
Further research should leverage data sources that are specifically designed for
and focus on military spouses. The Survey of Active Duty Spouses conducted by
the Defense Manpower Data Center could be a vehicle for this further research, but
the detailed address information necessary to leverage this resource is not currently
made available to the public, and some additional refinements to the survey to focus
on education costs faced or aid received would be beneficial.
90
Chapter 3: ESTIMATING THE IMPACT OF TAXING COLLEGE
SAVINGS PLANS OF NON-TRADITIONAL STUDENTS
3.1 Introduction
The federal government spends nearly $40 billion through tax expenditures on
higher education each year.1,2 Tax expenditure on higher education subsidizes those
who would have gone to college even in the absence of the subsidy and potentially
incentivizes those at the margin to enroll and therefore increase college attendance.
Considering how much money is spent on higher education through the tax code, it
is important to evaluate the effects of these measures.
This chapter examines the effect of a change in the tax incentives for edu-
cational savings on a traditionally understudied group, those who return to school
after participating in the labor market. The literature on the price sensitivity of
demand for higher education in the United States shows that sensitivity appears to
differ by the source of variation in price and by the characteristics of the students.
1The tax benefits (i.e., credits, exemptions, and deductions) in 2012 totaled over $33.2 billion,
and the refundable portion of a higher education tax credit accounted for an additional $6.6 billion.
2These tax expenditures include the Hope Tax Credit, the American Opportunity Tax Credit,
the Lifetime Learning Tax Credit, Section 529 College Saving Plans, Coverdell Educational IRA
accounts, and finally the tax code’s provision for the deductibility of interest expenses on student
loans and the exclusion of scholarship and fellowship income from taxable income.
91
Previous work on the price sensitivity of higher education due to changes in the
tax code have been limited and mixed. Turner (2011a) found that a $1,000 (in
year 2000 dollars) increase in tax-based federal student aid increased enrollment for
18- to 19-year-olds by 3 percentage-points. However, LaLumia (2012) found only
limited evidence of any effect of the Lifetime Learning tax credit or the tuition de-
duction on adult college attendance or degree completion. The one exception were
a statistically significant impact was detected was for those in the cohort whose
educational attainment in adulthood fell short of their initial expectations. That
select group appeared to respond to the tax incentives. Finally, using more detailed
data, Bulman and Hoxby (2015) did not detect any statistically significant responses
to expansion of higher education tax credits from the enactment of the American
Opportunity Tax Credit (AOTC).
These results measuring the tax-based aid sensitivity of demand for higher
education seem slightly at odds with the prior literature on the price sensitivity of
higher education with respect to the interaction of the magnitude of the response
with the age of the student for a complex form of aid. That literature found the
more opaque or complex the form of aid, the smaller the apparent impact of an
equivalent dollar value reduction in net price. It also found that the response of
adults was less mitigated than that of the young to more opaque forms of financial
aid.3 One proposed explanation is that older students have more experience filling
out complex government forms, such as those for income taxes, and are therefore less
3Several studies of the introduction of the Pell grant program failed to find a corresponding
increase in enrollment of the 18- to 24-year-old population, but Seftor and Turner (2002) found a
large positive response in matriculation by 25- to 30-year-old students to the introduction of the
Pell program.
92
deterred by the complexity of means-tested financial aid. Therefore, it is curious
that for such a complex form of aid as that administered through the tax code,
Turner (2011a) found an impact on the young almost as large as that seen due to
changes in tuition price, while LaLumia (2012) found almost no effect for adults.4
Finally, Bulman and Hoxby (2015) did not find any statistically significant impacts
for any age group or income level. It is therefore unclear how much of an effect, if
any, to expect from the change in the tax treatment of savings set aside for higher
education by individuals who interrupt their education with some time in the labor
force.
I obtained data from a network of firms that have a long-standing program
that offers delayed compensation which employees can only use for continuing their
own education. The firms call this pool of delayed compensation their College
Scholarship Plan (CSP). The CSP funds are only available as part of an employee’s
initial contract and represent a form of signing bonus.
A law passed in late 2004 changed the tax status of delayed compensation
allocated after January 5, 2005. However, the firms and their employees were not
aware of the implications of the law changing the tax status of CSP funds until
2007. This means the terms of the contracts signed in 2005 and 2006 were not
influenced by the tax change. For these two cohorts, the tax change could only
influence the decision of whether to use the funds after 2007. However, employees
who signed after 2007—and were aware of the tax change at signing—could modify
their compensation packages in response to the tax change.
4Data quality and measurement error may explain this issue.
93
This chapter investigates both the changes in the composition of compensation
packages for new employees who signed after the tax was announced and the change
in the usage rate of CSP funds by those surprised by the tax. Additionally, this
chapter calculates the sensitivity of demand for higher education to variations in
out-of-pocket costs from the interaction of the amount of CSP funds allocated and
the change in the tax status of those funds. This paper tests the hypothesis that
those who have already chosen to set aside funds for college will have a lower-than-
average sensitivity of demand, and that the change in the tax status of these funds
will decrease future allocations towards savings for college.
3.2 Background
3.2.1 Structure of Initial Contracts
The firms are in an industry that uses highly skilled physical labor to generate
considerable revenue. The firms require new employees to sign initial contracts that
last several years and include a non-compete clause giving whichever firm hires them
exclusive rights over the employees’ labor within the industry. An employee’s pay in
those initial years is determined on an industry-wide basis by the employee’s tenure
and the level to which the employee has advanced within the firm. Most pay during
the period examined was between $550 and $2,500 a month unless the employee
reaches the top tier in the industry, in which case the employee’s salary jumps to a
minimum of almost half a million dollars a year. However, less than 10 percent of
all employees who sign with a firm make it to this level. After the initial contract
94
is fulfilled, an employee earns the right under the collective bargaining agreement
(CBA) between the firms and the union to negotiate and sign with any firm within
the industry. At this stage, if the employee has reached the top tier, his annual
salary can be in the millions of dollars.5
To compensate an employee for the acceptance of the non-compete clause,
initial contracts typically include a signing bonus. These signing bonuses can be
quite large. It is customary for those who have not yet finished a college degree
and who are valued enough by a firm to receive a signing bonus to take a portion
of their signing bonus as CSP funds. These funds can only be distributed to cover
tuition and living expenses if the employee returns to college. The CSP funds are
negotiated for a set number of semesters with an allowance per semester for living
expenses and tuition. If tuition and living expenses are less than the amount set
aside in the CSP for a particular semester, then the firm keeps the balance. Also,
if CSP funds are not used within a certain time period, the employee forfeits those
funds. Typically, remaining funds are forfeited if the employee has not either worked
for the firm or attended college at least part-time within the past 2 years.
3.2.2 Rationale for CSP program
There are several possible reasons for the existence of the CSP program. The
reasons are not mutually exclusive, but the anticipated reaction of participants
in the CSP program to the changes in the tax treatment differ by rationale. By
5All employees in the sample are male, so I will use the masculine pronoun when referring to
an employee without bias.
95
examining how the volume and value of CSP contracts react to the tax, evidence of
the dominant rationale(s) for the CSP contracts may be found. There are four main
reasons why an employee would forgo a larger cash signing bonus for CSP funds.
First, CSP funds historically had a tax advantage.6 If this were the only reason for
CSPs, then the change in tax status would mark the end of the CSP program.
Second, CSP funds could be seen as an insurance policy by the employee. If
an employee signs with a firm, he is hoping he will be part of the roughly 10 percent
that get promoted to the top tier, after which time he may earn enough money that
he never needs to work again, let alone invest in marketable skills such as a college
degree. However, if he is part of the 90 percent who do not reach the top tier, he will
be let go by his firm after a few years without great wealth.7 New employees may
recognize the benefit of setting aside funds to pay for education to help transition
to a new career if they are unsuccessful with the firm. Furthermore, from the
employer’s perspective, since not every employee uses all of their CSP money, a
firm knows a dollar in CSP funds does not cost them the same in expectation as
a dollar of signing bonus. For example, if a firm expects that only half of all CSP
funds will ever be used, an employee may face the choice between receiving $100,000
cash compensation or compensation consisting of $80,000 cash and $40,000 in CSP
funds.8 Therefore, if an employee only plans to return to college if he fails to advance
6If a firm was willing to pay an employee a $100,000 signing bonus and the employee planned to
eventually attend or go back to college and did not receive CSP funds, then all of the employee’s
education purchases would be with post-tax dollars. However, if an employee anticipated spending
$40,000 in educational expenses, he could take $60,000 cash in the signing bonus and $40,000 in
CSP funds and only need to pay tax on $60,000 instead of the full $100,000.
7The average and median career lengths of all employees who signed between 1990 and 2000
are 3.70 and 2.89 years respectively.
8This example of 50 percent expected use matches the historic CSP usage data from contracts
96
to the top tier, he can shift additional funds into that possible future state by trading
signing bonus funds for more than dollar-for-dollar CSP funds in his initial contract.
The historic tax advantage would only have added to the favorability of the terms
of this insurance contract. Given this rationale, economic theory would predict that
the elimination of the tax benefit may reduce—but likely not eliminate—the volume
and value of CSP contracts.
CSP funds could also serve as a commitment mechanism for the employee to
attend college after his time in the industry. A potential employee may recognize
that an opportunity within this industry is preferable to attending college directly
after graduation. He may also recognize that if he does not succeed, attending college
would be his best option. The prospective employee may also anticipate the future
temptation to stay in the labor force and continue earning money. By accepting CSP
funds, an employee commits to continuing his education because he only receives
those funds while enrolled at least half-time in an undergraduate institution. Thus,
if an employee is concerned about having time-varying preferences, then accepting
CSP funds when signing an initial contract could act as a commitment mechanism
to return to school. Given this rationale, the tax wedge announced in 2007 should
translate into fewer employees opting for CSPs, but those who do would increase
their CSP funds to maintain the post-tax value of the commitment incentive.
An additional rationale for the program is that CSP funds allow a firm to
selectively provide additional compensation to potential employees with greater ne-
gotiating power. The set of skills the firms value often has little economic value
offered between 1990 and 2000.
97
outside the industry. Through the CBA with the union, the firms are given consid-
erable strength in negotiating contracts with new employees for 1 full year. A firm
has near monopsony power once all the firms within the industry agree not to com-
pete against one another. Even with the large signing bonuses, new employees are
offered compensation far below their expected marginal value to a firm because their
economic value outside the industry is usually much lower. Often, an employee’s
best outside option is to wait a year to see if he is offered a better deal during the
next hiring season. Potential employees who are offered contracts within the indus-
try coming directly from high school or after their junior year of college are often
simultaneously in possession of or have been offered an alternative scholarship from
an academic institution.9 This gives those potential employees greater bargaining
power than other recruits who have already finished college. CSP funds could be a
targeted response by the firms to this leverage. This rationale sets up CSP funds
as an addition to cash signing bonuses, not as a trade-off in exchange for less cash
at signing.10 Economic theory predicts that CSP funds per contract should increase
in response to the tax in order to maintain the same level of post-tax compensation
to nullify the outside option. The number of CSP contracts would not be greatly
affected by the tax given this rationale.
9These scholarships become void and an individual becomes ineligible for any similar future
scholarships if he ever signs with a firm within the industry.
10Quote from industry recruiter: “As time has passed, the number of (employees) who have
received CSPs has increased. It has gotten to the point where the general consensus is that the
CSP is now just a ‘throw-in.’ All the CFOs believe that the CSP should lower the (employee’s)
signing bonus, but that does not happen often in practice.”
98
3.2.3 Major Changes in the Industry
A series of important events affected how this industry recruits its employees
and who is recruited. It is important to understand these events and how they affect
the employers’ preferences, the decisions prospective employees make and the resul-
tant employee pool. What follows is a selection and interpretation of the important
events that shaped the industry in ways that are pertinent to this chapter. This
section discusses how those events break down the recent timeline of the industry
into periods in which similar decisions were made by both sides of the recruitment
transaction. It is this breakdown which guides sample selection, specifically the
years included in the pre-tax control group, in order to best identify the effect of
the tax change.
3.2.3.1 Changes to the Evaluation of Talent
For quite some time it has been informally understood by industry participants
that employees recruited out of high school were much less likely to reach the top
level of the firm and be as productive as similarly ranked employees selected with
some college experience. However, this fact had not historically been used to drive
the hiring decisions by firms within the industry.
During the 2002 hiring season, one firm allowed this observation to drive its
hiring during that recruitment year. In that same year, the firm had a much larger
number of positions to fill than usual. College applicants were heavily favored, as
the firm felt it could project the value of college applicants with greater certainty
99
than it could project that of high school applicants. The firm believed this allowed
it to obtain a pool of employees with similar or higher expected value with much
lower risk.11 The firm also used a modified measure of performance that was widely
ignored by other firms to rank both existing employees and potential recruits. The
firm implementing the two new strategies greatly exceeded expectations in 2002.
Shortly before the hiring season in 2003, a report was published in which
the two major strategies employed by the aforementioned firm were discussed and
credited, in part, for the firm’s success the previous year. This report was widely
circulated and the two strategies openly debated for the months leading up to,
during, and following the 2003 hiring season.12
Initially, some firms openly disagreed with the soundness of the strategies.
However, continued success by the few firms that implemented these strategies
through 2003 proved an effective argument for their merits. For this chapter, I
treat both the 2002 and 2003 recruitment seasons as transition periods for the in-
dustry, which stabilized by 2004. Figure 3.1a shows the percentage of individuals
offered initial contracts from all firms who were high school seniors by year recruited.
The figure illustrates the decline in the share of high school seniors from historic
levels during the transition years of 2002 and 2003. Figure 3.1b shows that the rate
of the remaining high school seniors receiving a CSP increased to nearly 100 percent
after the transition period ended. This change in the share of new employees and
the rate of CSP award for recruits who were high school graduates at signing is why
11This information is from a 2003 book on the industry; citation omitted to maintain the
anonymity of the data source.
12This information is from a 2003 book on the industry; citation omitted to maintain the
anonymity of the data source.
100
Figure 3.1: Trends in Recruitment of and CSP Rates for High School Seniors
(a) Change in Recruitment Pattern
10
15
20
25
30
Pe
rc
en
ta
ge
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
20
09
20
10
20
11
20
12
20
13
Year
Transition Years Percentage High School Seniors
High School Seniors as Share of Firm Recruits
(b) Increase in Rate of CSP
70
80
90
10
0
Pe
rc
en
ta
ge
19
95
19
96
19
97
19
98
19
99
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
20
09
20
10
20
11
20
12
20
13
Year
Transition Years Percentage with CSP
Share of High School Seniors with CSPs
Source: Author’s tabulation of data provided by the firms who wish to remain unidentified.
the pre-2004 cohorts are excluded from the pre-tax control group.
3.2.3.2 Collective Bargaining Agreements
The firms’ top tier employees are unionized, requiring the firms to periodically
negotiate industry-wide CBAs that determine not only compensation guidelines for
current employees, but also the manner in which the firms can recruit, sign, and pay
new employees. Each CBA is typically a 5-year contract. Many rules in the CBA
vary by the educational attainment of the recruit. For example, the length in years
of an employee’s initial contract before the employee can negotiate freely with other
firms differs by that employee’s education level. CBAs were signed in 1990, 1997,
2002, 2006, and 2011.
There are two major ways a CBA can affect the recruitment of new employ-
ees. First and foremost, it can change the rules and terms under which potential
employees can be recruited and what contract they can be offered. Second, even
101
if the CBA does not change these rules, the threat of a strike or an actual strike
could impact the pool of new hires. To eliminate potential bias in the sample, this
analysis only examines employees who signed under the same or very similar CBAs
and whose decision to sign was not influenced by a potential strike.
The CBA reached at the end of 2006 did not create much uncertainty through
the threat of a strike, as it was signed long before the 2007 hiring season began and it
did not substantively change the rules governing the recruitment of new employees.
However, the 2011 CBA drastically changed the rules governing the recruitment of
employees starting in 2012. Therefore, the 2011 recruitment class serves as a logical
end point to the timeframe of the analysis on the effects of the tax changes on the
recruitment and selection of new employees.
3.2.3.3 Changes to the Tax Treatment of CSP Funds
The change to the tax code examined in this chapter is the American Jobs
Creation Act of 2004, Public Law 108-357 (AJCA). The AJCA changed the tax
treatment of many forms of delayed compensation. After passage, some funds that
were traditionally not taxed either in the year they were earned or in the year they
were later received became subject to tax in the year they were received. The new
law applies to amounts deferred after January 5, 2005. Compensation deferred be-
fore 2005 but received afterward is subject to special grandfather provisions. These
amounts would only be subject to the new rules if the deferred compensation is
“materially modified” after October 3, 2004.
102
The firms initially thought that only delayed compensation not related to
education was subject to these new rules and that the CSP program would be found
exempt.13 As a result, employees who signed initial contracts that included CSP
funds in 2005 and 2006 did so believing the money for tuition, fees, and books would
be tax free, but those funds—if used—are in fact subject to taxes. On June 6, 2007,
the industry informed all employees who signed contracts as of January 5, 2005, of
the changes in the tax treatment of their CSP funds. Additionally, the industry
informed all new recruits of the tax status of any new CSP funds moving forward.
Consequently, the CSP funds from all contracts signed after 2004 are subject to
taxation, but only those who signed after June 6, 2007, knew about the tax when
negotiating their contract.
To summarize, employees recruited prior to 2004 are excluded from the analysis
due to the major reform in the industry occurring in 2002 and 2003. Those recruited
in 2004 who signed their initial contract before January 5, 2005, are part of the pre-
tax control group. Those who signed between January 5, 2005, and June 6, 2007,
are the “unknowingly taxed” group. Finally, those who signed after June 6, 2007,
but before the reforms of the 2011 CBA are part of the “knowingly taxed” group.
Those recruited after 2011 are excluded from the analysis.
13There is an exemption for some forms of educational assistance, but the CSP program did not
meet all of the final requirements.
103
3.3 Data and Descriptive Statistics
3.3.1 Data Description
This analysis is based on data from firms that wish to remain anonymous and
is augmented with Consumer Price Index data from the Bureau of Labor Statistics
to adjust dollar values over time. Data were made available on employees who
were offered contracts from 1990 to 2013. Firms take turns selecting domestically
recruited employees and thereby ranking them as they enter the firm. The data
contain this ranking, year of ranking, date of birth, race, primary language, and
highest education level attained at the time they were ranked. These data are
available for each individual offered an employment contract.
Foreign-born employees are subject to a different recruitment process, seldom
receive a CSP, and are excluded from this analysis.14 Additionally, employees with
educational attainment levels less than a high school senior or with this information
missing from their records are excluded from the analysis as they are few in number,
are typically outliers on other variables, and have unique barriers to overcome before
they can use any CSP funds.15
For those who signed with a firm, the firms provided additional data on when
the employees signed with the firm, when and why they left the firm, how high in
the firm the employees advanced, and the composition of the initial signing bonuses.
Of those who received CSP funds in their initial contract, data on the amount per
14This eliminated 1,701 employees from the data, 135 of which had a CSP.
15Those with educational attainment missing or less than a high school senior combined for only
67 employees, 40 of which had a CSP.
104
academic period and number of periods employees can receive funds are available.
Additionally, invoice records tell when and how much employees use their CSP funds.
Data and analysis from the Tax Policy Center and the National Center for
Educational Statistics are used to convert the value of the CSP and the tax status
of those funds into an implied out-of-pocket cost for a semester of college. This
out-of-pocket cost is then used to calculate estimates for the price sensitivity of
demand.
3.3.2 Descriptive Statistics
Tables 3.1 and 3.2 present the summary statistics for those who signed in
2004 before the tax change, in 2005 and 2006 after the tax change was in effect
but before the change was announced, and in 2007 through 2011 after the tax was
announced.16 Table 3.1 provides this information for all employees, while Table 3.2
consists of summary statistics on only those employees that signed an initial contract
that included a CSP. Pairwise t-tests are performed to determine if the differences
between the values from 2004 and the values from 2005 and 2006 are statistically
significant. The differences between 2004 through 2006 and 2007 through 2011 are
also tested for statistical significance.
Table 3.1 presents the summary statistics for all employees who signed with
16The recruiting process for domestic hires occurs in June of each year, but in the past employees
were allowed several months to decide on an offer from the firm. Some employees recruited in 2004
did not actually sign their initial contracts until after January 5, 2005. Thus they were assigned
to the “unknowingly taxed” group. As a result, Tables 3.1 and 3.2 are not strictly organized by
year recruited because they are organized by date signed and tax status. The first two tables in
the appendix show the summary statistics tables organized by year recruited. These tables show
that the differences between 2004 and 2005 and 2006 are not as stark as shown below. Thus the
assumption that the industry stabilized by 2004 looks more plausible.
105
the firm in the given tax cohorts. The average signing bonus increased from $140,560
(year 2000 dollars) in 2004 before the tax change to $158,530 (year 2000 dollars) for
the 2007-2011 cohort who knew about the tax change. Although there may appear
to be a trend over time, the differences in the values are not statistically significant
at the α = 0.05 level for either t-test.
Table 3.1: Summary Statistics for All Employees by Tax Status
Untaxed Unknowingly Taxed Knowingly Taxed
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 140.56 333.89 151.18 370.52 158.53 435.64
Age When Recruited by Firm 21.28 1.50 21.01 1.60 21.19 1.52
Fraction High School Seniors 0.17 0.37 0.15 0.35 0.17 0.37
Fraction From Junior College 0.10 0.30 0.19 0.39 0.11 0.31
Fraction College Fr. or So. 0.01 0.12 0.01 0.09 0.01 0.11
Fraction 4-Year College Juniors 0.31 0.46 0.29 0.45 0.33 0.47
Fraction 4-Year College Seniors 0.41 0.49 0.36 0.48 0.38 0.49
Fraction With a CSP 0.55 0.50 0.61 0.49 0.62 0.49
Observations 798 1714 4582
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample
includes all employees recruited domestically who signed with the firm between 2004 and 2011 with at least a high school degree.
(1) All monetary values are indexed to year 2000 dollars.
The average age of an employee who signs with the firm in the baseline group
is 21.28 and drops to 21.01 in the second cohort before rising again to 21.19 in
the third cohort. Both the difference between the first and second cohort and the
difference between the third and the first two cohorts combined are statistically
significant at the α = 0.01 level with t-statistics of 4.0863 and -2.5191, respectively.
It appears there was a drop in the average age of employees between 2004 and 2007.
This variable is accounted for in the regression analyses.
This drop in age can be attributed to the changes in the educational attainment
of recruits. Those who were recruited from junior colleges increased from 10 percent
for the baseline period to 19 percent for the “unknowingly taxed” cohort, before
dropping back down to 11 percent for the third cohort. These statistically significant
106
changes were paired with decreases in the number of recruits who had 4-year college
senior listed as their educational attainment at the time of signing from 41 percent
in the first cohort to only 36 percent in the second cohort. A statistically significant
increase in the fraction of recruits joining the firm after their junior year in college
in the third cohort to 33 percent coincides with the decline of recruits from junior
colleges in that cohort.
It appears that after the firms stabilized their preferences for high school se-
niors in 2004 at a new lower level, the high school seniors who previously would
have signed with a firm instead entered both junior colleges and 4-year colleges.
Those high school seniors who entered a junior college were eligible to sign with a
firm in 2005 and 2006 and apparently did so, replacing the marginal college junior
or senior. Meanwhile, those displaced high school seniors from 2004 who entered
a 4-year college were not eligible to sign with a firm until 2007, their junior year.
This delayed stabilization of the other educational attainment levels is justification
for examining only those who signed out of high school from 2004 through 2011 as
a separate subgroup. It also justifies educational attainment at signing as an im-
portant control variable included in the regression analysis when examining samples
including all educational attainment levels.
Finally, the percentage of employees who signed an initial contract with a CSP
rose from 55 percent in the first cohort to 61 percent and 62 percent in the second and
third cohorts. This change over baseline is statistically significant in both cases. The
analysis of the composition of compensation packages includes a linear time trend
to adjust for this pattern of increased CSP inclusion over time. This difference may
107
also be explained by the change in the composition of the educational attainment
of new employees over time.
Table 3.2 presents the summary statistics for selected characteristics of em-
ployees who had a CSP as part of their initial contract. These data are presented by
cohorts grouped by tax status. The differences in the average signing bonuses shown
in Table 3.2 are not statistically significant. The difference in age when recruited
by the three cohorts is statistically significant, first falling and then rising. This
is likely due to the statistically significant increase and then decrease in the share
from junior college recruits at the expense of the share of 4-year college juniors.
This mirrors the pattern seen in employees both with and without CSPs.
Table 3.2: Summary Statistics for Employees With CSPs by Tax Status
Untaxed Unknowingly Taxed Knowingly Taxed
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 245.93 416.02 237.63 444.61 249.41 529.70
Age When Recruited by Firm 20.47 1.44 20.27 1.48 20.48 1.41
Fraction High School Seniors 0.28 0.45 0.23 0.42 0.26 0.44
Fraction From Junior College 0.14 0.35 0.26 0.44 0.15 0.36
Fraction College Fr. or So. 0.02 0.13 0.01 0.11 0.02 0.14
Fraction 4-Year College Juniors 0.48 0.50 0.43 0.50 0.49 0.50
Fraction 4-Year College Seniors 0.08 0.27 0.07 0.25 0.07 0.26
Number of Semesters in CSP 4.46 2.46 4.47 2.31 4.29 2.44
CSP Funds Per Semester1 $6,931.92 $4,394.98 $6,683.32 $4,127.82 $8,744.81 $5,386.51
Total Value of CSP1 $30,860.53 $28,676.77 $28,978.98 $23,654.71 $37,385.71 $32,828.58
Proportion of Signing Bonus 0.28 0.22 0.26 0.19 0.29 0.20
Taken as CSP Funds
Observations 441 1040 2840
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample includes all employees
recruited domestically, who signed with the firm between 2004 and 2011, with at least a high school degree, and had a CSP included in their initial contract.
The total value of a CSP is calculated by the amount of money available per semester, multiplied by the number of semesters specified in the contract. The
proportion of the signing bonus taken in CSP funds is equal to the total value of CSP funds divided by the sum of CSP funds and the cash signing bonus.
(1) All monetary values are indexed to year 2000 dollars.
The average number of semesters provided for by the CSPs did not have a
statistically significant difference between the first and second cohorts, but the third
cohort experienced a significant decline. This is paired with a large and statistically
significant increase in the funds allocated per semester in the CSPs from an average
of $6,757 for the entire period before the tax was known to $8,745 per semester after
108
the tax was announced. No statistically significant change is seen in this value after
the tax was in effect but before it was known.
The total value of a CSP is calculated by multiplying the funds per academic
period by the number of periods the CSP is available according to the initial contract.
This value increased from an average of $29,539 per CSP contract for the first two
cohorts to $37,385 for the third cohort. Again, no statistically significant change
was seen in this variable after the tax was in effect but before it was known. This
increase in funds after the tax was announced suggests that the CSP acts more like
a commitment mechanism or as compensation for increased negotiating power than
as a mechanism for tax avoidance or as an insurance policy against not making the
top level within the firm.
The rank an employee is assigned the year he is recruited by the firm is a fair
proxy for the value and thus negotiating power of this individual. Individuals with
higher or lower rank may respond differently when made aware of the change to the
tax code. This potential cross-sectional variation in the response to the tax change
is explored in two ways. First, I interact the primary explanatory variable with the
rank variable. Second, I segment the employee population into groups based on their
rank to examine any non-monotonic interactions between rank and the response to
the tax change. Based off of conversations with industry experts, the employees
were broken into four rank groups. Those assigned a rank of 1 to 90 in the year they
were recruited are in the top-ranked group with the most negotiating power, those
ranked 91 to 300 are assigned to the second-ranked group, those ranked 301 to 600
are in the third, and those ranked above 600 make up the forth-ranked group.
109
Tables C.36 through C.43 in the appendix present the summary statistics for
the different cohorts by each of these rank groups. They show that the likelihood
of having a CSP included in their initial contract, as well as the amount of the cash
signing bonus, CSP funds per semester, and total CSP funds all decline uniformly
as the rank group decreases, as one would expect. Of note is that 99 percent of
those in the top-ranked group already have a CSP included in their initial contract,
leaving little room for improvement. The share of signing bonus taken in the form of
CSP funds for those with a CSP increased from roughly 7 percent for the top-ranked
group to over 40 percent for the fourth-ranked group.
The demographic characteristics of age and educational attainment also vary
by rank group. The top-ranked group is a year younger on average than the entire
employee pool and has a disproportionate number of high school juniors and 4-year
college juniors. These two educational attainment categories make up less than half
of all those recruited, but more than 90 percent of those in the top-ranked group.
The other major demographic pattern of note by rank group is that the increase in
recruits from junior colleges seen in the 2005 and 2006 cohorts was concentrated in
the third- and fourth-ranked groups.
3.4 Empirical Framework
This chapter first examines some of the ways new employees may have re-
sponded to the tax through changes in their initial contracts. These potential re-
sponses include: whether a CSP is included in the initial contract, changes in the
110
amount of CSP funds per semester, or changes in the total value of the CSP. Re-
gression analysis estimates the response after the tax was announced in 2007 while
controlling for other factors that changed over time.17 Next, I examine changes in
the usage rate of CSP funds due to the unexpected taxation of funds allocated in
2005 and 2006. Finally, I construct an estimate for the expected out-of-pocket costs
of attending college based on the characteristics of each employee’s initial contract
and when he left employment with the firm to yield estimates of the sensitivity of
demand for higher education.
In each of the first two models, the primary explanatory variables of interest
are essentially year indicator variables. Interpreting the coefficients of these variables
as causal estimates of the effect of the change in the tax treatment requires certain
assumptions, specifically that there are no other factors that changed over time that
affect the outcomes of interest, that are not controlled for by other variables in the
regression.
The background section describes major changes to the industry that could
affect the outcomes of interest including changes to the evaluation of talent and
new collective bargaining agreements. The section also describes the censoring of
the window of observation to mitigate these potential threats to the assumptions of
the model. However, the censoring does not appear to have resulted in a complete
absence of statistically significant differences in observable characteristics between
employees selected in the different time periods. Including these observed variables
17I also use regression analysis to confirm the true level of information diffusion, by performing
a falsification test, examining whether there was a change that occurred between contracts signed
before and after January 5, 2005.
111
as controls should alleviate the direct threat of these factors, but there is the concern
that unobserved differences may also bias the results.
The most significant event that roughly coincides with the change in the tax
treatment of CSP funds that may affect the outcomes of interest and that does not
have readily available controls is the Great Recession beginning in 2008. This event
threatens the identification strategy and may bias the coefficients for the variables
of interest. An attempt to forecast the direction of the potential bias will be made
for the outcome of each model separately.
3.4.1 Effect of Tax Change on Compensation Package
To explore the effect of the knowledge of the tax change on the initial contract,
I compare the contracts of those who signed with the firm from 2004 through 2006
to those who signed in 2007 through 2011. Analysis of the contracts signed in 2004
against those signed in 2005 and 2006 is also performed in order to check if there
was any anticipatory behavior or early information diffusion about the tax change.
3.4.1.1 Sample Selection
All employees with at least a high school degree recruited domestically between
2004 and 2011 are included in the analysis of whether an initial contract included a
CSP, but only the subset of employees who had a CSP as part of their initial contract
are included in the analysis of the other compensation package characteristics.
112
3.4.1.2 Econometric Models
To investigate the impact of the knowledge of the tax change on selected char-
acteristics of the initial contract, I run Ordinary Least Squares (OLS) regressions.
The model is a straightforward pre-post specification taking the following form:
(1) yicft = β0 + β1Taxc + β2Xi + β3Y eart + γf + icft
In this model, yicft is the outcome of interest for an individual i in tax cohort
c who signed an initial contract with firm f in recruitment year t. The outcomes
of interest include: an indicator variable of whether CSP funds are included in the
initial contract, the natural log of CSP funds per semester, the natural log of the
total value of CSP funds, and the share of compensation packages delivered in the
CSP.18 The regressor of interest, Taxc, is an indicator variable equal to one if the
individual was in the cohort that knew their CSP funds would be taxed and zero
otherwise. The knowledge of the tax would make the value of CSP funds decrease.
The expected direction and magnitude of the effect of knowledge of the tax on the
various outcomes of interest would depend on the dominant rationale for the CSP,
as addressed previously.
The vector Xi consists of controls including race, age and level of educational
attainment at signing, and the rank of the employee in the year he signed. The
race variable is an indicator variable equal to one if the employee is White. The
18Both CSP funds per semester and the total value of CSP funds are indexed to year 2000 dollars
using the CPI-U index.
113
educational attainment variables are a set of indicator variables for whether the
employee’s highest educational attainment level at signing is junior college, some
4-year college, 4-year college junior, or 4-year college senior/graduate, with high
school senior as the reference category.19
The rank value assigned to an employee during the annual recruitment process
is an interesting control variable. The rank is associated with the perceived skill
within the industry and the expected value an employee can bring to the firm that
offers the individual a contract. The lower a recruit’s rank score, the higher his
perceived value and the stronger his negotiating position.20 If negotiating power is
the driving force for the terms of CSP contracts, then this greater negotiating power
would make an employee more likely to receive a CSP, have more CSP funds per
semester, and have more total CSP funds. The effect of rank on the share of the
compensation package delivered in CSP funds is unclear, as increases in both forms
of compensation—cash and CSP funds—would be expected with lower rank.21 The
low-ranked employees are also more likely to make the highest level in the firm and
not need funds set aside for retraining after their tenure with the firm is over. If an
insurance contract is the dominant motivation for a CSP, then a lower rank score
would make the employee less likely to be released before making the top tier within
the firm and therefore the employee would have a lower demand for CSP funds as
an insurance policy. It follows that lower ranked employees would be associated
19The “some 4-year college” variable includes 4-year college freshmen and sophomores.
20I follow the industry convention that the first recruit selected, with a rank score of 1, has the
lowest ranking or rank score.
21However, since CSP funds can only be spent on college, the total potential useful value of CSP
funds is capped. Thus top-ranked employees who max out their CSPs and receive all additional
funds as cash would be expected to have a lower share of their compensation in their CSP.
114
with fewer CSP contracts, less CSP funds per semester, lower total CSP value,
and a smaller share of total compensation taken in the form of CSP funds. Given
the varying association this variable can have with the outcomes of interest, the
coefficient for this variable is shown and analyzed in the results.
The regression model also includes controls for firm fixed effects, γf . The firm
that selects an employee may have different preferences for offering a CSP or the
level of funds included in a CSP. Firm fixed effects control for the possibility that
firms with stronger preferences for offering CSPs may make up a larger or smaller
share of the contracts offered in any particular year. Year fixed effects would harm
the identification strategy because there is hardly any variation in the knowledge of
tax status within each recruitment year. However, a linear time trend is included
as a control in the model.
3.4.1.3 Estimation Results
Table 3.3 indicates that the knowledge of the end of the tax break increased
the rate at which CSP funds were awarded, the amount of CSP funds per semester,
the total CSP award, and CSP funds’ share of total compensation. These results
are inconsistent with the “tax avoidance” and “insurance policy” rationales for the
CSP program.
The increase in total CSP funds, CSP funds per semester, and CSP funds as
a share of total compensation after 2007 is consistent with the commitment mech-
anism rationale, but the increase in the share of contracts that include CSP funds
115
Table 3.3: OLS Regressions of Compensation Package Characteristics on Knowledge of
Tax at Signing, 2004-2011
CSP Funds Included Log of CSP Funds Log of Total Value CSP as Share of
in Initial Contract Per Semester of CSP Funds Total Signing Bonus
2004-2006 vs 2007-2011
Elimination of Tax Break 0.0553∗∗∗ 0.0752∗∗ 0.0758∗∗ 0.0275∗∗∗
Known at Signing (0.0128) (0.0320) (0.0337) (0.00874)
Rank in the Year -0.0197∗∗∗ -0.0458∗∗∗ -0.0604∗∗∗ 0.0362∗∗∗
Contract Was Offered (x 100) (0.000943) (0.00239) (0.00252) (0.000654)
N 7094 4321 4321 4321
2004 vs 2005-2006
Elimination of Tax Break 0.00880 -0.000820 -0.00310 -0.0370∗∗∗
Unknown at Signing (0.0129) (0.0317) (0.0334) (0.00882)
Rank in the Year -0.0222∗∗∗ -0.0439∗∗∗ -0.0607∗∗∗ 0.0368∗∗∗
Contract Was Offered (x 100) (0.00180) (0.00443) (0.00467) (0.00123)
N 2507 1478 1478 1478
Notes: The regression of whether CSP funds were included in the initial contract is run on all domestically recruited employees
who signed contracts and had at least a high school degree when joining the firm. The sample for the other three regressions is a
subset of the first including only those who signed an initial contract with a CSP. In each model controls for age at signing, race,
educational attainment at signing, rank, and fixed effects for the firm who signed the employee are included. For the regressions
using the full sample through 2011, a linear year trend is included and the key outcome variable is an indicator variable equal to 1
if CSP funds were known to be taxed at signing and 0 otherwise. For the regressions using the sample through 2006, a linear year
trend is not included given the short time frame of the sample, and the key outcome variable is an indicator variable equal to 1 if
CSP funds were subject to taxation (albeit unknowingly) at signing and 0 otherwise.
Standard errors in parentheses∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01
is inconsistent with that rationale. Economic theory would predict the taxation of
CSP funds drives out the marginal employee wishing to commit to returning to col-
lege. Recruits would either go directly into college instead of signing with the firm
or would sign with the firm but take the full amount of CSP funds as cash. However,
it is possible that there was an overall increase in demand for college and CSP funds
over time and the linear time trend does not fully account for this impact.22
The rationale that CSP funds are a means for a firm to target additional
compensation to individuals with a stronger negotiating position is consistent with
the results pertaining to the characteristics of the CSP contracts. If a firm uses
the CSP to compensate recruits who have the outside option of attending college
22The appendix has figures that illustrate the change in each of the four outcomes. Each of these
graphs shows a stark change beginning in 2007, making the misspecification of the year effects less
likely.
116
on a university scholarship, then the firm would need to provide additional funds
after 2007 to maintain the after-tax value of that compensation. Additionally, if it
is shown that the end of the tax break for CSP funds means that CSP funds are
used less frequently, then each CSP dollar is less costly for a firm to offer. This
could explain the increase in the rate at which employees receive CSP funds. If CSP
funds are less costly, the firm may be more willing to offer CSP funds as targeted
compensation to the marginal employee.
The sign on the coefficient for the rank variable is suggestive of which rationale
is dominant. Employees with a lower rank value are more likely to make the top level
in the firm, and thus would be less likely to use the CSP as an insurance contract or
commitment mechanism. Therefore, we would expect employees with lower rank to
receive fewer CSPs if their primary rationale is as insurance or commitment. How-
ever, the coefficient on the rank variable is negative and statistically significant for
the outcomes of whether a CSP is awarded, the amount of CSP funds per semester,
and the total value of the CSP. This implies that employees with higher expected
value and greater negotiating power are more likely to have a CSP plan and have
more CSP funds per contract, suggesting targeted compensation as the dominant
rationale for CSPs. The coefficient on rank in the regression for the share of the
total signing bonus made up by CSP funds is positive and significant. This is not
inconsistent with the theory that negotiating power drives the CSP contract. Due
to the in-kind nature of the CSP compensation, only so much money can be of use
in that contract vehicle. However, there is no limit to the desirability of additional
cash compensation. Thus, once enough CSP funds to pay for college have been
117
allocated, employees with extremely high negotiating power associated with a very
low rank value will demand additional compensation in the form of cash, pushing
down CSP’s share of total compensation.
The bottom half of Table 3.3 indicates that the unknown end of the tax break
on CSP funds in 2005 did not have an effect on the rate at which CSP funds were
included in an initial contract, the amount of CSP funds awarded per semester,
or the average total value of a CSP award before the change was announced in
2007. The rank variable continues to have a negative and statistically significant
coefficient for the first three regressions, while the CSP’s share of the total signing
bonus appears to be higher for employees with a high rank value as before in the full
sample. This result was unanticipated, but may be explained by changes to the cash
signing bonus. The regressions in the bottom half of Table 3.3 do not include a year
trend; however, including the trend does not change the results in any meaningful
way.
Each of the outcomes of interest in this section may have been affected by the
Great Recession. However, as the Great Recession weakened the demand for labor,
the potential bias from this event on the coefficient of interest for the first three
outcomes would be negative. As the coefficients for the variable of interest for each
of these three outcomes are positive, the conclusions of this section are reinforced by
the potential bias that may be present if the assumptions required to interpret the
coefficients as causal estimates are violated by the timing of the Great Recession.
The impact of a weaker economy on the share of an initial signing bonus
taken in the form of CSP funds is unclear. A weaker economy and labor market
118
should mean less negotiating power for the potential employee along all forms of
compensation including both cash signing bonus and CSP funds. As it is not clear
whether the affect of a weaker economy would be greater in percentage terms on the
cash compensation or CSP funds, it is difficult to forecast the sign of this potential
bias on the outcome of the share of signing bonus taken as CSP funds.
3.4.1.4 Heterogeneity of Responses to Tax
Given the impact of rank on the outcomes of interest and to further explore the
rationale for CSP funds, this section examines the effect of the interaction between
the rank of employees and the change in the tax code on the characteristics of
employee compensation packages. This cross-sectional variation in the response to
the tax change is explored in two ways: (1) by interacting the primary explanatory
variable with the rank variable and (2) by segmenting the employee population into
groups based on their rank to look for any non-monotonic interactions between rank
and the tax change. Following conversations with industry experts on what may
be meaningful rank thresholds, the employees were broken into four rank groups.
The first 90 employees recruited in any given year, with corresponding ranks of
1 through 90, are assigned to the top-ranked group. This group is labeled rank
group 1 in Table 3.5 and is believed to have the most negotiating power. Those
with ranks of 91 through 300 are assigned to the second-ranked group. The third
group is composed of employees ranked 301 to 600 in the year they were recruited.
Those ranked above 600 make up the forth-ranked group. A brief description of the
119
characteristics of each group is included in the summary statistics section of this
chapter.
Table 3.4 shows the results when a variable interacting rank with the key tax
variables for each sample is included in the model for each compensation package
characteristic outcome. For the 2004–2011 sample, the coefficients for the original
tax variable for the first three outcomes maintain their sign and statistical signifi-
cance while the interaction terms are not statistically significant at the 0.05 level.
The estimates for coefficient of the tax variable for the the first three outcomes using
the 2004–2006 sample remain not statistically significant when the interaction term
is added to the specification of each model. Furthermore, none of the coefficients
for the interaction term are statistically significant for the first three outcomes in
this sample. The inclusion of the interaction term changes the results for the fourth
outcome, the share of the signing bonus taken as CSP funds, in both samples. The
interaction term is positive and statistically significant in the 2004–2011 sample,
but negative and statistically significant in the 2004–2006 sample. As the 2004–
2006 sample was included as a check on the assumptions of the model, perhaps
these results indicate some other factor related to rank, which has not been fully
controlled for in the model, is biasing the results for this outcome.
As for the first three outcomes, it may be the case that the impact of the tax
change did differ by rank but not in a monotonic way. As seen in the summary
statistics, those in the first-tanked group almost all had CSP funds as part of their
initial contract, and the average CSP funds per semester of that group was over
$9,000. Therefore, although the change in the tax code appeared to increase these
120
Table 3.4: OLS Regressions of Compensation Package Characteristics on Knowledge of
Tax at Signing Interacted with Rank, 2004-2011
CSP Funds Included Log of CSP Funds Log of Total Value CSP as Share of
in Initial Contract Per Semester of CSP Funds Total Signing Bonus
2004-2006 vs 2007-2011
Elimination of Tax Break 0.0379∗∗ 0.112∗∗∗ 0.0979∗∗ 0.0163
Known at Signing (0.0161) (0.0380) (0.0400) (0.0104)
Rank in the Year -0.0218∗∗∗ -0.0398∗∗∗ -0.0568∗∗∗ 0.0344∗∗∗
Contract Was Offered (x 100) (0.00154) (0.00407) (0.00428) (0.00111)
Interaction of Rank (x 100) with 0.00320∗ -0.00882∗ -0.00525 0.00266∗∗
Elimination of Tax Break (0.00181) (0.00488) (0.00513) (0.00133)
Known at Signing
N 7094 4321 4321 4321
2004 vs 2005-2006
Elimination of Tax Break -0.00651 -0.0413 -0.0385 0.0259∗∗
Unknown at Signing (0.0226) (0.0476) (0.0502) (0.0131)
Rank in the Year -0.0243∗∗∗ -0.0517∗∗∗ -0.0674∗∗∗ 0.0488∗∗∗
Contract Was Offered (x 100) (0.00308) (0.00813) (0.00857) (0.00223)
Interaction of Rank (x 100) with 0.00291 0.0107 0.00932 -0.0166∗∗∗
Elimination of Tax Break (0.00352) (0.00935) (0.00986) (0.00256)
Unknown at Signing
N 2507 1478 1478 1478
Notes: The regression of whether CSP funds were included in the initial contract is run on all domestically recruited employees
who signed contracts and had at least a high school degree when joining the firm. The sample for the other three regressions is a
subset of the first including only those who signed an initial contract with a CSP. In each model controls for age at signing, race,
educational attainment at signing, rank, and fixed effects for the firm who signed the employee are included. For the regressions
using the full sample through 2011, a linear year trend is included and the key outcome variable is an indicator variable equal to 1
if CSP funds were known to be taxed at signing and 0 otherwise. The interaction term for this sample multiplies the rank variable
by the ”Known at Signing” tax variable. For the regressions using the sample through 2006, a linear year trend is not included
given the short timeframe of the sample, and the key outcome variable is an indicator variable equal to 1 if CSP funds were subject
to taxation (albeit unknowingly) at signing and 0 otherwise. The interaction term for this sample multiplies the rank variable by
the ”Unknown at Signing” tax variable.
Standard errors in parentheses∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01
121
outcome measures, there might not have been as much room to grow in the top-
ranked group, which was close to being maxed out already. Therefore, there could
be non-monotonic impacts of the tax by rank that are masked within the simple
interaction term. To examine this possibility, I break out the sample by rank group
and run the original model on each sample separately.
Table 3.5 shows the results of the original model for each outcome (without
the interaction term), for both the entire sample and each rank group separately.
Two patterns emerge from this table. First, all rank groups, besides rank group 1,
which was already receiving CSP funds in 99 percent of initial contracts before the
tax change, saw an increase in the share of their initial contracts that included CSP
funds following the announcement of the tax change. Second, the majority of the
impact for the other outcomes appear to be driven primarily by rank group 2.
The rationale where the purpose of CSP funds is to offer additional compensa-
tion to recruits with more negotiating power can explain this second pattern, where
overall results appear to be driven by the second rank group. After the change in
the tax treatment is announced, those with more leverage demand additional com-
pensation so that their after-tax compensation remains the same. However, if those
from rank group 1 are already maximizing the potential useful value of their CSP
funds, then there would be no point in increasing the money in the CSP portion of
their signing bonus. They may instead demand more cash to offset the tax on their
CSP funds. That would leave those in rank group 2 with the combination of (1)
room in the CSP portion of their signing bonus to grow and (2) strong leverage to
demand such an increase. Therefore, the results by rank group reinforce the conclu-
122
Table 3.5: OLS Regressions of Compensation Package Characteristics on Knowledge of
Tax at Signing by Rank Group, 2004-2011
CSP Funds Included Log of CSP Funds Log of Total Value CSP as Share of
in Initial Contract Per Semester of CSP Funds Total Signing Bonus
Full 2004-2006 vs 2007-2011
Elimination of Tax Break 0.0553∗∗∗ 0.0752∗∗ 0.0758∗∗ 0.0275∗∗∗
Known at Signing (0.0128) (0.0320) (0.0337) (0.00874)
N 7094 4321 4321 4321
Rank Group 1
Elimination of Tax Break 0.0104 0.0562 0.0521 0.00863
Known at Signing (0.0203) (0.0740) (0.0750) (0.00546)
N 675 662 662 662
Rank Group 2
Elimination of Tax Break 0.0417∗ 0.262∗∗∗ 0.262∗∗∗ 0.0296∗∗∗
Known at Signing (0.0246) (0.0569) (0.0573) (0.00952)
N 1471 1268 1268 1268
Rank Group 3
Elimination of Tax Break 0.0755∗∗∗ 0.00760 0.0152 -0.000983
Known at Signing (0.0271) (0.0594) (0.0631) (0.0167)
N 1799 1157 1157 1157
Rank Group 4
Elimination of Tax Break 0.0635∗∗∗ 0.00541 0.0321 0.0407∗
Known at Signing (0.0206) (0.0637) (0.0679) (0.0209)
N 3149 1234 1234 1234
Notes: The regression of whether CSP funds were included in the initial contract is run on all domestically recruited employees
who signed contracts and had at least a high school degree when joining the firm. The sample for the other three regressions is a
subset of the first including only those who signed an initial contract with a CSP. In each model controls for age at signing, race,
educational attainment at signing, rank, and fixed effects for the firm who signed the employee are included. All regressions shown
use the full sample through 2011 and include a linear year trend. The key outcome variable is an indicator variable equal to 1 if
CSP funds were known to be taxed at signing and 0 otherwise. Rank Group 1 includes only those ranked in the top 90 for the year
they were recruited. Rank Group 2 includes those selected between 91 and 300 in their recruitment year. Rank Group 3 includes
those between 301 and 600, and Rank Group 4 includes those selected after the first 600 in a given year.
Standard errors in parentheses∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01
123
sion that the rationale for the CSP funds is to offer targeted compensation to those
with more negotiating power.
3.4.2 Effect of Tax on CSP Usage
3.4.2.1 Sample Selection
Given that CSP funds are typically available for initial use up to 2 years after
termination and given that the average career length is around 3 years, many CSP
contracts signed before 2012 are still active.23 Furthermore, many employees wait
until after termination from employment with the firm to start using their CSP
funds. Thus the timing of termination is a sensitive predictor of when CSP funds
are first used. Those who signed CSP contracts as part of their initial contract in
2004 had a longer time to use their CSP funds than those who signed once the tax
was in effect due to right censoring. To compare like samples, earlier cohorts are
pared down to resemble the information available for later cohorts. This rephrasing
of the question of “has the CSP ever been used” to “has the CSP been used by a
certain cut-off” requires selecting a cut-off.
I considered age, time since signing the initial contract, and time since being
released by the firm as potential cut-offs. Of these, I use the cut-off of 2 years
after being released by the firm, because only this outcome has a clear break point
in the timing of CSP usage. That break point is around the second or third year
after termination and was expected given the way the CSP contracts are typically
23The distribution of career length is not symmetric, having a long right tail.
124
Figure 3.2: Histogram of the Difference Between the Date CSP Funds Were First Used
and an Employee’s Termination Date
0
5
10
15
20
Pe
rc
en
t
−16 −14 −12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12
Difference in Years
Source: Author’s tabulation of data provided by the firms who wish to remain unidentified.
written. Figure 3.2 illustrates this break. The histograms for the alternative cut-offs
are included in the appendix.
In simple regressions using the date education funds are first used as the
outcome variable and either the date an employee was terminated (if available),
the date he signed, or his date of birth as the explanatory variable, I found that the
termination date had a slightly higher R-squared when using the full sample.
I compare those who signed in 2005 and 2006 and thus were affected by the
end of the tax break to those who signed in 2004. Given this sample, the maximum
window of observation is determined by the last year included in the “unknowingly
taxed” cohort. If the 2006 recruitment cohort is included, then that cohort had 7
125
years after signing to start using their CSP funds at the time of data extraction. To
provide a 2-year post-termination window of observation, the sample is limited to
those who left the firm within 5 years. I assign all cohorts the same cut-off to allow
for comparison of similar samples. The outcome variable is equal to one if CSP
funds were first used either before termination or within 2 years of termination.
The indicator variable for those with CSP funds subject to taxation has a
slightly muddled interpretation without further sample restriction. The indicator
variable is equal to one if the contract was signed after January 5, 2005. However,
those who signed after January 5, 2005, but used their CSP before the announcement
in 2007 did so without knowledge of the tax. This means individuals who used their
CSP early could not have been influenced by the tax change because they did not
know about it. Only those who did not use their CSP before June 6, 2007, could
be prevented from ever doing so because of the tax.24 This may bias the estimate
of the impact of the tax on CSP usage.
Obtaining an unbiased estimate of the effect of the tax on usage for those who
signed in 2005 and 2006 requires separate sample restrictions for each treatment
cohort. To examine the effect of the tax on those who signed in 2006, all those who
used their CSP in the first year after signing in both 2004 and 2006 are eliminated
from the sample. Similarly, to estimate the effect of the tax on those who signed
in 2005, the analysis limits the sample to those who did not use their CSP funds
within 2 years of signing their contract. However, if only employees from 2004 and
24For those who learned about the tax change in 2007 who had already started using their CSP
funds, knowledge of the tax may cause them to be less likely to continue using their CSP and
perhaps cause them to be be less likely to complete their degree, but these questions are outside
the scope of this chapter.
126
2005 are analyzed, then an additional year of observation is available for the full
sample. Accordingly, employees who were terminated in their sixth year with the
firm can be included while still allowing a 2-year run-off post-termination before
the data extraction date. I examine the impact using the full 2004 through 2006
sample under two conditions. In one sample I exclude anyone who started using
their CSP funds within 2 years of signing an initial contract. In the other, I increase
my sample by including all employees who were terminated within 5 years of signing
with a firm, acknowledging the potential mitigation of the estimate of the coefficient
through measurement error.
Employees who reach the top level within the firm before leaving the firm are
excluded from the analysis. The rationale is that earnings for those rare few who
reach the top level are so large that upon leaving the firm those individuals may
not need to invest in marketable skills to sustain the standard of living they desire.
Their view of college may change from an investment to a consumption good.
3.4.2.2 Econometric Model
I perform OLS analysis to examine if there is any statistically significant effect
of the tax change on the likelihood of using CSP funds to attend college. An indicator
variable for those who were affected by the tax change who did not know about the
loss of the tax benefit when signing their initial contract is included as the primary
explanatory variable in the model. The model takes the following form:
(2) yicft = β0 + β1Taxc + β2Xi + γf + icft
127
In this Linear Probability Model (LPM), yicft is an indicator if an individual
i from tax cohort c who signed an initial contract with firm f in recruitment year
t used their CSP funds before the cut-off date of the sample. The regressor of
interest, Taxc, is an indicator variable equal to one if the individual was in a cohort
who thought they had the tax break but later found out they did not, and zero
otherwise. The knowledge of the tax would make the value of CSP funds decrease.
The expected sign of the tax coefficient is negative.
The vector Xi consists of controls including the rank of the employee in the
year they signed, race, the level of educational attainment of the employee at signing,
age at signing, the natural log of the cash signing bonus, and the natural log of CSP
funds per semester. The race variable is an indicator variable equal to one if the
employee is White. The educational attainment variables are a set of indicator
variables for each categorical response.
The rank assigned is associated with perceived skill, and the earlier analysis
finds that a lower rank number is predictive of higher CSP funds per semester. A
low rank from being selected earlier is also positively correlated with a larger cash
signing bonus. However, I do not have a prediction for how an employee’s rank
in the year they were recruited affects CSP usage when controlling for these other
factors.
The log of the cash signing bonus proxies as a control for income. Actual
household wealth is not available, but the larger the signing bonus, the more likely
the employee has the resources to overcome credit constraints imposed by the tax.
The coefficient for the log of CSP funds per semester is expected to be non-negative,
128
as larger values reduce the probability of credit constraints deterring matriculation.
The regression model also includes controls for firm fixed effects, γf . The
firm that selects an employee may have a work culture that either encourages or
discourages college attendance. Firm fixed effects control for the possibility that
firms with different cultures may make up larger or smaller shares of the contracts
offered in any particular year.
The identification assumption is that the usage rate of CSP funds would not
be systematically correlated with signing after January 5, 2005, in the absence of the
change in tax treatment of CSP funds. As the key independent variable is essentially
a year indicator, any other shift that influences the employees’ tastes for college over
time may add bias to the estimate.
The greatest potential shift in the preferences for college may come from the
different alternative opportunities available to those who were terminated in different
years. Attending college might have appeared more attractive to those terminated
after the Great Recession began as the opportunity cost of forgone earnings of
attending college likely decreased for those facing a more difficult job market. Those
recruited and signed in 2005 and 2006 are more likely to have their careers with the
firm last until after the start of the recession and therefore may have been more
likely to attend school even with the knowledge that their CSP funds would be
taxed. This could counteract the expected impact of the tax and positively bias the
results for the coefficient on the tax variable towards zero.
Given the employee is seldom the one who decides when their career is over
with the firm and the length of an individual’s career varies, individuals recruited in
129
different years can be terminated in the same year and face the same macroeconomic
impacts on their opportunity costs of attending school. Therefore, models specifica-
tions including year terminated fixed effects or the unemployment rate when they
were terminated will be examined in addition to the specification above.
3.4.2.3 Estimation Results
The first column of Table 3.6 examines the effect of the tax on the cohorts that
signed in 2005 against those who signed in 2004 where the sample is restricted to
those terminated within the first 6 years of employment and who did not use their
CSP in the first 2 years after signing. The only statistically significant predictor
of CSP usage is the log of CSP funds per semester, and the coefficient has the
expected sign. The coefficient for the effect of the unexpected loss of the tax break
for CSP funds is negative, as expected, but not statistically significant. Similar
results are found using the 2006 cohort as the treatment group where the employees
are only included in the sample if they left the firm within 5 years of signing but are
excluded if they started using their CSP funds within the first 2 years after signing.
The same holds true for the combined 2005 and 2006 treatment group under both
sample restriction criteria.
These results suggest that the taxation of CSP funds does not have a measur-
able impact on the rate at which employees use those funds for college. Meanwhile,
the positive and statistically significant coefficient for the amount of CSP funds
available per semester suggests that the out-of-pocket costs faced by the employee
130
Table 3.6: Linear Probability Model of CSP Usage on Tax Status of Funds, 2004-2006
Initial CSP Usage Rate
Treatment Group 2005 2006 2005 and 2006
Exclusion Criteria 6,2 5,1 5,2 5,0
CSP Funds Taxed -0.00664 -0.00974 -0.0310 -0.0193
(0.0459) (0.0451) (0.0414) (0.0357)
Natural Log of Cash 0.0390∗ 0.00604 0.0316 0.00563
Signing Bonus1 (0.0237) (0.0233) (0.0202) (0.0176)
Natural Log of CSP 0.128∗∗∗ 0.0845∗ 0.0963∗∗ 0.114∗∗∗
Funds Per Semester1 (0.0466) (0.0449) (0.0383) (0.0326)
Ranking in the Year 0.00803 0.00247 0.00557 0.00693
Contract Was Offered (x 100) (0.00842) (0.00831) (0.00712) (0.00631)
R2 0.155 0.103 0.121 0.107
N 520 544 727 886
Notes: All regressions include race, age and educational attainment at signing, firm fixed
effects, and the variables for which coefficients are reported. Given the limited number
of years in each sample and the correlation between year and the treatment, neither year
fixed effect nor a linear time trend are included as controls. All regressions use the 2004
recruitment class as the control group while the treatment cohorts are listed in the associated
column. All regressions exclude foreign recruits, employees who reached the top level within
the firm, and employees who did not include a CSP as part of their initial contract. The
additional exclusion criteria are indicated in each column. The first number in the pair
reports the number of years before which an employee had to be terminated to be included
in the sample, while the second number reports the number of years after signing where
if the employee first used their CSP funds before that date they were excluded from the
analysis. The key outcome of interest is an indicator variable that is equal to one if CSP
funds were subject to taxation and zero otherwise.
Standard errors in parentheses
∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01
(1) All monetary values are indexed to year 2000 dollars.
131
are negatively correlated with college attendance. However, this correlation can be
driven by those with a stronger preference for college negotiating for more money
in their CSP at signing.25
3.4.2.4 Heterogeneity of Responses to Tax
Alternative model specifications leveraging the fact that employees were re-
cruited in different years and thus were subject to different tax rules but were
released in the same year and thus faced similar macroeconomic conditions also
failed to find a statistically significant impact from the change in the tax treatment
of CSP funds. These specifications included year terminated fixed effects or the
unemployment rate in the year terminated.
Models examining potential cross-section variation in response to the tax
change on CSP usage were also explored. To examine variation by rank, one spec-
ification interacted the tax variable with the rank variable, and another segmented
the population by rank group, running the original four models on each group. Nei-
ther effort discovered a statistically significant coefficient on the tax variable. The
same held when segmenting the population by educational attainment at signing
and running on each group.
25There is no meaningful difference in the results if each sample is further restricted to only
include employees who were high school seniors when recruited.
132
3.4.3 Implied Price Sensitivity of Demand
This section calculates an estimate for the price sensitivity of demand for
higher education using an estimate for the out-of-pocket price faced by each indi-
vidual. The estimate of an employee’s expected out-of-pocket cost per year of college
is a function of the total tuition and fees cost of college attendance, the funds per
semester in an employee’s CSP contract, and the tax status of the CSP funds.26
To obtain an estimate of the out-of-pocket costs faced by the employee, I begin
with the average cost of tuition, fees, room, and board at a 2- or 4-year institution
in the United States for the year an employee was terminated.27 Then the after-tax
value of an employee’s CSP funds per semester is subtracted from the average cost
of attending college in the year terminated. The resulting number is the estimated
out-of-pocket cost for an employee attending college.
To estimate the effective marginal tax rate on CSP funds, additional assump-
tions are necessary given data limitations. The income bracket, marital status, and
state of residence are unknown for most employees in the data set. As a result, the
estimate for the effective marginal tax faced by employees who signed after Jan-
uary 5, 2005, can only be a rough approximation of the true effective marginal tax
rate they face. Estimates of the effective marginal tax rate during this period were
gathered from analysis by the Tax Policy Center (2007). These estimates take into
account federal and state individual income taxes and the federal payroll tax, and
26Forgone earnings are an important opportunity cost of college attendance but do not factor
into the out-of-pocket cost calculation in this section.
27The national average is used because the state of residence of the employee is frequently
unknown.
133
are available by earning decile. I use the median effective marginal tax rate on earn-
ings in the third decile of all earners for each year. The third decile is used due to the
below-average expected income of the individuals in my sample given their young
age. With estimates for the education costs per semester and the effective marginal
tax rates per year, the projected out-of-pocket cost of a semester is calculated for
each employee as:
OutOfPocketCostsi = CostPerSemestert − (1− (Ic ∗Ratet)) ∗CSPFundsi
Where CostPerSemestert is the national average cost of a semester of college
in the year the employee was terminated, Ic is an indicator variable for whether CSP
funds were subject to taxation,28 Ratet is an estimate of effective marginal tax rate,
and CSPFundsi are the CSP funds per semester of the employee.
This specification of out-of-pocket costs exploits variation in the financial im-
pact of the tax due to the tax change and variations in the amount of CSP funds per
semester to help estimate a price sensitivity of demand for higher education. The
main sources of variation in the OutOfPocketCostsi variable is from the CSP funds
per semester of the employee, CSPFundsi. When a simple regression is run with
OutOfPocketCostsi on the left-hand side and CSPFundsi on the right-hand side,
the R-squared ranges between 0.83 and 0.87 depending on the sample. R-squared
increased by an additional 0.1 to 0.13 when a variable representing the estimate of
the value of the taxes due from the use of CSP funds, if any, is included. This can
28Employees who signed in 2005 and 2006 have their tax status assigned according to the infor-
mation available to the employee on the day of their termination. So if an employee who signed in
2005 was terminated in 2006, the out-of-pocket cost they expected to pay if they attended college
does not include the taxation on CSP funds. This more accurately reflects the information set at
the time the decision was likely made, even if unexpected back taxes would eventually be due.
134
be interpreted to mean that 10 to 13 percent of the variation in the out-of-pocket
costs variable is due to the change in the tax code.29
In the previous section, the impact of the tax was modeled as uniform across
all employees with a CSP. However, even if two employees faced the same tax rate,
the increase in out-of-pocket cost would be larger for the employee who negotiated
more CSP funds per semester in their initial contract. Meanwhile, including CSP
funds per semester for both those whose funds are taxed and those whose funds
are not taxed controls for the correlation between CSP funds and an employee’s
preference for higher education. The residual variation in the OutOfPocketCostsi
variable, primarily from the estimate of the taxes due, will be used to identify the
impact of prices on demand for college.
3.4.3.1 Sample Selection
I include domestically recruited employees who signed in 2004 or later with a
CSP in their initial contract and who were terminated by 2011 in this analysis. As in
the previous section, I exclude employees who reach the highest level in the firm be-
cause their increased wealth from reaching the top tier may change their motivation
for attending college from an investment to purely consumption. I present results
for four different samples. There may be reactions to the announcement of the tax
that are unobserved, so those who signed in 2004–2006 are examined separately to
look at those who could not modify their initial contract or change their decision to
29When the variable representing the value of the taxes due is included by itself in the simple
regression, the R-squared value ranges from 0.25 to 0.29. However, as the value of the CSP funds
per semester is a component in the estimate for taxes due, then including this variable by itself
likely results in an overestimate of this variable’s effect on the variation in out-of-pocket costs.
135
join the firm based on the knowledge of the tax change. The examination of this
subgroup is an attempt to mitigate potential omitted variable bias. High school
seniors are the traditional group studied in the literature. They are examined in
isolation for both the full sample and the 2004-2006 cohort subsample to compare
the results of this chapter to historical estimates and to check for potential heteroge-
neous treatment effects by educational attainment at signing. It is conceivable that
the closer one is to finishing a degree, the less price sensitive he or she is for any
one year of college. Therefore, high school seniors could be the most price-sensitive
subgroup.
3.4.3.2 Econometric Model
I estimate a linear probability model with OLS to calculate the price sensitiv-
ity of demand for higher education. This modeling is consistent with the existing
literature and allows for comparison of the results with estimates from previous
studies. The specification of the econometric model is:
(3) CSPUsedift = β0 + β1OutOfPocketCostsi + β2Xi + β3Tt + γf + ift
In this Linear Probability Model (LPM), CSPUsedift is an indicator vari-
able equal to one if an individual i who signed an initial contract with firm f
and was released by the firm in year t has used their CSP funds within the win-
dow of observation. The definition of the primary independent variable of interest,
OutOfPocketCostsi, is provided above. The out-of-pocket costs are indexed to
136
year 2000 dollars, and per semester costs were doubled and then divided by 1,000
to allow for comparison with results from earlier academic work.
Xi includes controls for individual-level characteristics of the employee. These
controls include the natural log of the signing bonus, the natural log of CSP funds
per semester, the employee’s ranking in the year they signed, and indicator variables
for his educational attainment at signing and race.
The major difficulty with this analysis is the potential endogeneity of the out-
of-pocket price for college with respect to preferences for education. Individuals
who value education may be more likely to negotiate for a CSP and/or for more
funds in their CSP than those with a lower desire to return to school. CSP funds
per semester themselves may be a function of both preferences for education and
the negotiating power of the employee at signing. The rank of the employee in the
year signed and the natural log of the cash signing bonus attempt to control for
the negotiating power of the employee, allowing the residual variance in CSP funds
per semester to hopefully capture and control for the preference for education more
precisely. Accordingly, CSP funds per semester, rank, and the value of the cash
signing bonus are included in all regressions. The variation in the out-of-pocket
costs coming from the interaction of the tax status with the CSP funds allocated
per semester is what identifies the estimated price sensitivity in the model.
Meanwhile, Tt includes variables determined by the year an employee was
terminated. This set of controls includes the age of the employee when they are
terminated and the national unemployment rate in the year the employee was ter-
137
minated.30 Economic theory predicts a high unemployment rate in the year an
employee was terminated would make them more likely to use their CSP funds,
because a higher unemployment rate would translate into a lower opportunity cost
of returning to school from forgone wages. The age at termination controls for an
employee’s potential aversion to returning to school due to feeling out of place be-
cause he is older than the regular college population. Also, an older CSP holder’s
potential returns to education would be limited by the number of working years
available post-graduation.
The appropriateness of these two variables depends on the assumption that
termination from the firm is an event which triggers the decision by the employee of
whether or not to even use their CSP funds. This is observably true for many employ-
ees, but not all; therefore, specifications excluding these year terminated variables
(YTV) are performed to examine the results without relying on this assumption.
3.4.3.3 Estimation Results
The price sensitivity of demand results consistently fail to find a statistically
significant relationship between college attendance and the variation in out-of-pocket
costs created by the interaction between the amount of CSP funds allocated and
the change in their tax status. This result holds across various samples and model
specifications. This is consistent with the findings of Bulman and Hoxby (2015).
The one exception is the model with the full sample where all the cohorts from 2004
30Although some employees use their CSP before being terminated, the most common year to
start using a CSP is the year terminated. That is why the age and unemployment rate in the year
terminated, as opposed to some other year, such as year signed, were used.
138
through 2011 are eligible, employees of all educational attainment levels are included,
and the specification does not include controls whose values are determined by the
year an employee is terminated. Table 3.7 presents the results of estimating Equation
(3). Column 1 reports the results for the most inclusive sample but without controls
for labor market conditions or age in the year an employee was terminated. This
sparse specification yields a point estimate on β1 of -0.0118, which is statistically
significant at the 1 percent level.
This estimate implies a 1 percentage-point reduction in the use of CSP funds
for a $1,000 (year 2000 dollars) increase in out-of-pocket costs. This estimate implies
a much lower sensitivity of demand than the 3 percentage-point increase in college
attendance from $1,000 in tax-based aid found in previous studies of 18- to 19-year-
olds (Turner, 2011) or adults whose educational attainment fell short of early life
educational expectations (LaLumia, 2012).31 Furthermore, this result does not hold
in any other sample or specification of the model. In all other samples and model
specifications, the point estimate on β1 is still negative, often smaller in magnitude,
and never statistically significant.32
There are several possible explanations for the deviation of this result from
the existing literature. These possible explanations include: measurement error
from the crudeness of the out-of-pocket cost estimate; uncertainty, confusion, or a
31LaLumia did not find a statistically significant result for the entire adult population, only those
who fell short of their original education expectations, but those who set aside money for further
education can certainly be said to have an expectation to eventually achieve more education.
32This result is in line with main result from LaLumia (2012), which also found no effect on adult
education from college aid through the tax code for the general adult population. However, as the
population LaLumia studied was in their 30s and 40s, while the population is this study is closer
in age to that of Turner (2011), the absence of a statistically significant result was unexpected.
139
Table 3.7: OLS Regressions Estimating the Sensitivity of Demand, 2004-2011
2004-2011 2004-2006
All HS Seniors All HS Seniors
No YTVs YTVs No YTVs YTVs No YTVs YTVs No YTVs YTVs
(1) (2) (3) (4) (5) (6) (7) (8)
Out-of-Pocket Costs1 -0.0118∗∗∗ -0.00647 -0.0156 -0.00364 -0.00644 -0.00167 -0.0146 -0.00549
(0.00419) (0.00458) (0.0112) (0.0124) (0.00516) (0.00570) (0.0128) (0.0145)
Natural Log of CSP -0.0468 0.00821 0.109 0.221∗ 0.0527 0.0990 0.0786 0.160
Funds Per Semester1 (0.0446) (0.0484) (0.115) (0.124) (0.0569) (0.0612) (0.146) (0.158)
Natural Log of 0.0205 0.0233∗ -0.0446 -0.0456 0.0144 0.0217 0.0363 0.0505
the Signing Bonus1 (0.0132) (0.0134) (0.0394) (0.0399) (0.0163) (0.0167) (0.0486) (0.0502)
Rank in the Year 0.00732∗ 0.00920∗∗ 0.0166 0.0233∗∗ 0.00817 0.00765 0.0389∗∗ 0.0381∗∗
Contract Was Offered (x 100) (0.00443) (0.00449) (0.0111) (0.0111) (0.00583) (0.00589) (0.0150) (0.0151)
Education Dummies Yes Yes No No Yes Yes No No
Termination Year Variables No Yes No Yes No Yes No Yes
R2 0.080 0.086 0.153 0.182 0.104 0.108 0.255 0.263
N 1611 1611 322 322 967 967 213 213
The Out-of-Pocket Costs variable is calculate by taking the average annual college tuition and fees in the year an employee was terminated and
subtracting the after-tax value of two semesters’ worth of CSP funds. This value is then divided by $1,000 to more readily compare the value
of the point estimate to those from the existing literature. All regressions include control variables for race and indicator variables for the firm
that initially signed the employee. Indicator variables for educational attainment are included in any regression that did not have a sample
limited to only employees who were in high school when recruited by the firm. Termination year variables include the age of an employee and
the national unemployment rate in the year the employee was terminated. These variables are included in the specification with YTV in the
column label. The sample for columns (1) through (4) consists of employees signed between 2004 and 2011. Employees recruited from overseas,
who advanced to the top level of the firm, or who were not terminated by the end of 2011 are not included in the analysis. The sample for
columns (5) through (8) is similar but only includes employees who signed before 2007 as opposed to 2011.
Standard errors in parentheses
∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01
lack of salience of the true after-tax price at the time when the decision was made;
a truly different sensitivity of demand for these non-traditional students; or a sense
of commitment given that some funds have already been set aside for education.
Each of these reasons may explain the low price sensitivity of demand estimate.
Data limitations, particularly not knowing the state of residence upon termination
or variation in the tax status of CSP funds within cohorts, hinder the parsing out
of the potential reasons.
Also of note: this model specification yields a coefficient on the natural log
of CSP funds per semester that is similar in magnitude and sign to the results
from estimating Equation (2), but is seldom statistically significant. One possible
explanation is that the definition of out-of-pocket costs may rely too heavily on
the value of CSP funds per semester for its own variation. With near or almost
140
perfect collinearity, an increase in the standard errors of the collinear parameters is
expected and would increase the probability of making a type II error of accepting
the null hypothesis when it is false for both variables. This could explain the lack
of statistical significance for the point estimate of β1.
The coefficient for the variable controlling for the rank of an employee at
signing was consistently positive and was statistically significant in over half of the
various model specifications and sample selection combinations. The coefficients
associated with the rank variable were largest for the sample of high school seniors
who all signed before the tax was announced. Thus the youngest group was more
likely to return to school after termination if their prospects with the firm were
worse. With a worse rank, they likely had a lower expectation of success with the
firm and a higher expectation of furthering their education. This result mirrors that
of LaLumia (2012), who found higher expectation of educational attainment at a
younger age was associated with the decision to return to school.
3.4.3.4 Heterogeneity of Responses by Rank
Given the apparent effect of rank on CSP usage, this section investigates po-
tential heterogeneous effects of the change in the tax code by rank. This is done by
segmenting the population and running each model specification by rank group.33
Table 3.8 shows the results of the sensitivity of demand regressions by rank
group. Certain subgroups suffer from small sample size, but a general story does
33A model interacting the Out-of-Pocket Costs variable with the Rank variable was also explored.
This interaction variable was not statistically significant at the 0.05 level in any specification and
the inclusion of this variable did not meaningfully change the coefficient on the Out-of-Pocket
Costs variable itself.
141
appear to emerge. Those from rank group 4 consistently have large negative co-
efficients for the Out-of-Pocket Costs variable. The coefficient for the 2004–2011
sample including all educational attainment levels has a larger magnitude than that
of the full sample including all rank groups and is statistically significant at the
0.05 level. As with the contract characteristics outcomes, rank group 2 also has
a statistically significant outcome when using the 2004–2011 sample including all
educational attainment levels. Rank group 1 has the smallest sample size and does
not exhibit a statistically significant result, and the sign on the coefficient fluctuates
by specification and sample. Meanwhile, the coefficients for rank group 3 are in-
consistent with the rest of the sample. The overall conclusion is those who had the
weakest negotiating position, were offered the lowest average cash signing bonuses,
and were likely the ones most expecting to fail with the firm are those who appear
to be the most sensitive to the change in the out-of-pocket costs due to the taxation
of CSP funds.
142
Table 3.8: OLS Regressions Estimating the Sensitivity of Demand by Rank Group, 2004-
2011
2004-2011 2004-2006
All HS Seniors All HS Seniors
No YTVs YTVs No YTVs YTVs No YTVs YTVs No YTVs YTVs
(1) (2) (3) (4) (5) (6) (7) (8)
Full Sample
Out-of-Pocket Costs1 -0.0118∗∗∗ -0.00647 -0.0156 -0.00364 -0.00644 -0.00167 -0.0146 -0.00549
(0.00419) (0.00458) (0.0112) (0.0124) (0.00516) (0.00570) (0.0128) (0.0145)
N 1611 1611 322 322 967 967 213 213
Rank Group 1
Out-of-Pocket Costs1 -0.0235 -0.0162 0.0874 0.0413 -0.00991 -0.00855 0.0608 0.128
(0.0226) (0.0258) (0.0860) (0.0910) (0.0213) (0.0238) (0.0937) (0.0750)
N 92 92 39 39 76 76 32 32
Rank Group 2
Out-of-Pocket Costs1 -0.0186∗∗ -0.0127 -0.0160 -0.00769 -0.0125 -0.0104 -0.0156 -0.00983
(0.00816) (0.00910) (0.0210) (0.0233) (0.00912) (0.0102) (0.0230) (0.0267)
N 380 380 97 97 277 277 75 75
Rank Group 3
Out-of-Pocket Costs1 0.000719 0.00874 -0.00893 0.0109 0.0162 0.0312∗∗ 0.0412 0.0326
(0.00885) (0.00975) (0.0267) (0.0366) (0.0122) (0.0136) (0.0396) (0.0501)
N 499 499 99 99 297 297 65 65
Rank Group 4
Out-of-Pocket Costs1 -0.0174∗∗ -0.0128∗ -0.0666∗ -0.0533 -0.0158 -0.0127 -0.0306 -0.0610
(0.00719) (0.00775) (0.0339) (0.0397) (0.0111) (0.0121) (0.0704) (0.0839)
N 640 640 87 87 317 317 41 41
Education Dummies Yes Yes No No Yes Yes No No
Termination Year Variables No Yes No Yes No Yes No Yes
The Out-of-Pocket Costs variable is calculate by taking the average annual college tuition and fees in the year an employee was terminated and
subtracting the after-tax value of two semesters’ worth of CSP funds. This value is then divided by $1,000 to more readily compare the value
of the point estimate to those from the existing literature. All regressions include control variables for race and indicator variables for the firm
that initially signed the employee. Indicator variables for educational attainment are included in any regression that did not have a sample
limited to only employees who were in high school when recruited by the firm. Termination year variables include the age of an employee and
the national unemployment rate in the year the employee was terminated. These variables are included in the specification with YTV in the
column label. The sample for columns (1) through (4) consists of employees signed between 2004 and 2011. Employees recruited from overseas,
who advanced to the top level of the firm, or who were not terminated by the end of 2011 are not included in the analysis. The sample for
columns (5) through (8) is similar but only includes employees who signed before 2007 as opposed to 2011. The rank groups are determined by
the same 90/300/600 splits used in the rest of the chapter.
Standard errors in parentheses
∗ p < .1, ∗∗ p < .05, ∗∗∗ p < .01
143
3.5 Conclusion
This chapter examines the effect of an unexpected loss of an exemption from
taxes on savings set aside for college on both future savings for education and further
college attendance. The analysis indicates that the knowledge of the tax on CSP
funds increased the amount of money set aside on both a total and a per semester
basis. These results suggest that the primary justification for the CSP program is
not as an insurance policy or a vehicle for tax avoidance, but as either a means by
which firms can target additional compensation to employees with higher expected
value and greater negotiating power or a commitment mechanism to return to school.
The change in the treatment of funds from untaxed to taxed also appears to have
increased the share of employees who receive a CSP as part of their initial contract.
This result suggests the dominant rationale for CSPs is targeted compensation,
rather than a commitment mechanism to return to school. The results analyzing
the effect of the tax on CSP usage are insufficient to reject the hypothesis that
taxation of CSP funds do not affect CSP usage rates. This evidence is fairly robust
to model specification and sample selection.
The estimated sensitivity of demand for higher education of these non-traditional
students is lower than the traditional estimate of 3 to 5 percentage-points per $1,000
(year 2000 dollars) change in tuition or the 2 to 3 percentage-points per $1,000 (year
2000 dollars) change in tax-based aid found by Turner (2011a). However, these re-
sults are consistent with the work of Long (2004) and Bulman and Hoxby (2015),
who did not find statistically significant impacts on education outcomes associated
144
with tax-based aid. Several possibilities including data limitations, the source of
variation, and different preferences for education by this population could explain
this result. If the result is not population specific but instead due to the source of
variation in price, then the continuation of tax incentives that encourage saving for
college would appear to be a less effective means of raising the matriculation rate into
college than lowering the cost of college directly through subsidies to universities.
145
B.6 Appendix B
Table B.9: Price Paid by Military Spouses at 2-Year Public Institutions
State 2000 2001 2002 2003 2004 2005 2006 2007 2008 Number of Schools
Alabama 304 352 2057 2341 2532 446 428 444 463 27
Alaska 2028 2009 2022 2142 2113 2319 2735 2902 2922 2
Arizona 935 2899 3244 3250 3337 3420 3580 3649 1297 21
Arkansas 1399 1480 1688 1776 1757 1759 1785 1757 1879 22
California 316 303 298 449 660 636 575 490 468 113
Colorado 1770 1738 1745 1764 1770 1856 1810 1793 1834 15
Connecticut 5460 5589 5630 6404 6529 6684 6787 7046 7127 12
Delaware 1680 1740 1784 1852 1900 1978 2012 2065 2146 3
District of Columbia N/A N/A N/A N/A N/A N/A N/A N/A N/A 0
Florida 1431 1426 1430 1526 1595 1651 1716 1553 1685 20
Georgia 3057 2967 3007 3114 1279 1401 1417 1518 1490 48
Hawaii 1069 1034 1018 1043 1070 1082 1186 1299 1405 6
Idaho 2269 2283 2369 2428 2517 2511 2516 2536 2542 3
Illinois 5130 4768 4896 4870 5259 5002 5087 5254 5553 48
Indiana 3705 3896 4258 4376 4477 4574 4625 4688 2343 14
Iowa 2144 2290 2444 2517 2638 2702 2686 2710 2736 16
Kansas 2966 2814 3048 3109 3122 3106 3056 3023 1787 24
Kentucky 1212 1337 1699 2125 2291 2090 2240 2299 2343 16
Louisiana 844 908 911 1101 1249 1319 1310 1350 1338 34
Maine 2614 2572 2622 2576 2554 2689 2595 2714 2618 7
Maryland 4209 3911 3989 4324 4503 4370 4400 4397 4310 16
Massachusetts 6968 7219 7209 7244 7193 7062 6851 6376 6661 16
Michigan 3375 3278 3244 3382 3466 3723 3730 3796 3853 30
Minnesota 2520 2680 2921 3218 3496 3581 3678 3749 3678 31
Mississippi 1170 1358 1410 1287 1361 1462 1451 1431 1414 15
Missouri 2090 2066 2409 2546 2718 2794 2669 2885 2877 21
Montana 2312 2372 2495 2704 2668 2732 2734 2810 2771 12
Nebraska 1407 1436 1481 1548 1605 1670 1687 1758 1763 8
Nevada 1395 1392 1411 1429 1447 1450 1442 1462 1535 1
New Hampshire 3949 4162 4248 4488 4857 5079 4741 4943 4788 7
New Jersey 3432 3414 3439 3582 3662 3743 3853 3865 3922 19
New Mexico 1002 993 995 1021 1113 1213 1236 1226 1196 20
New York 2801 2572 2787 2841 2891 2909 2924 2940 2921 35
North Carolina 902 984 1056 1085 1137 1146 1107 1144 1120 59
North Dakota 1925 1938 2135 2361 2776 2883 3140 3217 3247 6
Ohio 2405 2422 2634 2770 2899 2977 2976 2879 2741 30
Oklahoma 1292 1199 1285 1595 1668 1842 1909 1958 2023 12
Oregon 1650 1690 1882 2291 2378 2382 2478 2440 2377 17
Pennsylvania 4389 4378 4288 4422 4730 4956 4951 4940 4869 21
Rhode 1806 1793 1913 1971 2102 2182 2286 2360 2471 1
South Carolina 3786 4185 4646 5312 5433 5483 5551 5622 5669 20
South Dakota 2868 2883 3012 2635 2618 2820 3019 3080 3138 5
Tennessee 1440 1593 1663 1930 2010 2115 2122 2182 2221 13
Texas 1361 1403 1469 1612 1644 1768 1864 1886 1911 64
Utah 1588 1646 1735 1832 1916 1982 1931 2019 2071 7
Vermont 5764 5805 6615 6632 6816 6999 7070 7248 7412 1
Virginia 4799 4678 5071 5483 5501 5469 5834 2061 2130 24
Washington 1764 1825 1987 2062 2145 2228 2243 2269 2260 30
West Virginia 1671 1616 1651 2653 2679 2216 2268 2232 2218 11
Wisconsin 2292 2224 2477 2397 2540 2605 2676 2736 2770 17
Wyoming 1432 1437 1478 1496 1522 1560 1543 1587 1601 7
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education Statistics, Inte-
grated Postsecondary Education Data System (IPEDS), 2000-2008.
Notes:
(1) All values represent an enrollment-weighted average of degree-granting, public institutions within each state.
Each value is then indexed to year 2000 dollars using the Bureau of Labor Statistics CPI-U.
(2) The Community College of the District of Columbia was not established until after the time period analyzed.
Thus, there were no appropriate 2-year tuition costs for residents of the District of Columbia.
146
Table B.10: Published Price for Military Spouses at 2-Year Public Institutions
State 2000 2001 2002 2003 2004 2005 2006 2007 2008 Number of Schools
Alabama 271 271 1819 342 2039 2138 376 390 417 27
Alaska 1994 1914 1924 1875 1804 1899 2323 2394 2480 2
Arizona 891 762 2767 3004 2972 2998 3015 3120 1069 21
Arkansas 1286 1311 1381 1428 1637 1604 1550 1585 1555 22
California 327 304 292 285 416 591 556 513 435 113
Colorado 1834 1640 1610 1598 1650 1602 1655 1594 1593 15
Connecticut 5460 5454 5487 6067 6098 6188 6205 6427 6423 12
Delaware 1616 1585 1666 1654 1693 1724 1743 1788 1795 3
District of Columbia N/A N/A N/A N/A N/A N/A N/A N/A N/A 0
Florida 1295 1279 1323 1285 1372 1370 1390 1451 1349 20
Georgia 2907 2853 2801 2814 1142 1133 1211 1247 1320 48
Hawaii 1052 1008 990 944 953 961 953 1055 1129 6
Idaho 2209 2143 2179 2197 2231 2262 2221 2234 2178 3
Illinois 4699 4776 4593 4549 4550 4754 4428 4585 4607 48
Indiana 3705 3635 3730 3976 4020 4024 4028 4111 2032 14
Iowa 2015 1975 2145 2208 2281 2346 2360 2333 2353 16
Kansas 2936 2734 2723 2753 2829 2817 2746 2733 1583 24
Kentucky 1126 1146 1349 1532 1911 2066 1838 1984 1998 16
Louisiana 813 885 900 975 1056 1109 1145 1154 1158 34
Maine 2431 2539 2527 2393 2333 2282 2327 2319 2387 7
Maryland 3908 3832 3856 3803 3941 3899 3892 3903 3822 16
Massachusetts 7020 6919 6545 6711 6575 6377 6139 5647 5792 16
Michigan 2999 2943 3056 2864 3084 3066 3172 3302 3267 30
Minnesota 2353 2373 2554 2700 2911 3136 3146 3264 3251 31
Mississippi 978 988 1147 1192 1178 1214 1266 1260 1223 15
Missouri 1882 1881 1915 2175 2292 2427 2373 2356 2512 21
Montana 2229 2110 2215 2480 2607 2505 2400 2435 2378 12
Nebraska 1350 1326 1366 1369 1401 1429 1456 1493 1523 8
Nevada 1230 1316 1333 1308 1306 1300 1277 1282 1271 1
New Hampshire 3662 3700 3938 3879 4036 4366 4257 4224 4260 7
New Jersey 3360 3202 3215 3163 3230 3248 3259 3351 3356 19
New Mexico 926 926 943 933 928 1000 1034 1048 1055 20
New York 2753 2635 2542 2587 2590 2609 2567 2588 2571 35
North Carolina 781 844 928 972 971 1007 1000 967 989 59
North Dakota 1909 1837 1873 1998 2160 2477 2502 2628 2772 6
Ohio 2495 2204 2253 2348 2443 2536 2537 2546 2433 30
Oklahoma 1333 1153 1250 1290 1449 1603 1630 1687 1686 12
Oregon 1505 1525 1577 1723 2077 2175 2156 2170 2063 17
Pennsylvania 4217 4106 4070 3980 4040 4288 4357 4363 4177 21
Rhode Island 1746 1704 1716 1774 1802 1889 1922 2032 2051 1
South Carolina 3702 3813 3948 4327 4866 4881 4829 4946 4887 20
South Dakota 2484 2483 2506 2406 2449 2401 2471 2495 2685 5
Tennessee 1320 1358 1515 1542 1764 1802 1863 1879 1912 13
Texas 1298 1280 1346 1393 1497 1539 1615 1668 1658 64
Utah 1537 1499 1576 1609 1674 1727 1685 1736 1782 7
Vermont 5510 5438 5558 6135 6061 6126 6165 6284 6299 1
Virginia 4614 4498 4484 4668 4979 4902 4854 1664 1682 24
Washington 1689 1656 1735 1815 1886 1932 1959 2013 2004 30
West Virginia 1655 1573 1553 1742 1848 1916 1948 1959 1938 11
Wisconsin 2145 2125 2149 2142 2165 2271 2299 2373 2394 17
Wyoming 1314 1331 1376 1363 1367 1368 1377 1381 1379 7
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education Statistics, Inte-
grated Postsecondary Education Data System (IPEDS), 2000-2008.
Notes:
(1) All values represent an enrollment-weighted average of degree-granting, public institutions within each state.
Each value is then indexed to year 2000 dollars using the Bureau of Labor Statistics CPI-U.
(2) The Community College of the District of Columbia was not established until after the time period analyzed.
Thus, there were no appropriate 2-year tuition costs for residents of the District of Columbia.
147
Table B.11: Price Paid by Military Spouses at 4-Year Public Institutions
State 2000 2001 2002 2003 2004 2005 2006 2007 2008 Number of Schools
Alabama 2999 3151 3345 3692 3986 4019 3981 4032 4390 14
Alaska 2940 2961 3004 3182 3430 3581 3763 3937 4003 3
Arizona 2347 2406 2458 3335 3709 3917 3991 4125 4471 3
Arkansas 2988 3231 3499 3723 3896 4074 4177 4487 4590 11
California 2503 2570 2573 3422 3810 3813 3711 3957 4138 36
Colorado 2970 3039 3129 3137 3273 3740 3742 4123 4307 13
Connecticut 11587 11186 11681 12432 13133 13295 13378 13393 13466 11
Delaware 4755 4863 5138 5711 6035 6248 6312 6483 6644 2
District of Columbia 2070 2002 1966 1925 1883 1828 2272 2604 2511 2
Florida 2034 2110 2128 2196 2349 2401 2402 2342 2524 20
Georgia 8484 8418 8811 9347 2883 3012 3074 3228 3344 26
Hawaii 2692 2671 2691 2724 2761 2766 3243 3719 4122 4
Idaho 2612 2748 2875 3082 3259 3457 3534 3632 3685 4
Illinois 4129 4362 4836 5177 5832 6252 6756 7359 7809 12
Indiana 10553 10527 11290 12077 12461 12627 12725 13022 5452 15
Iowa 3160 3359 3936 4641 4918 4963 5125 5158 5146 3
Kansas 8261 8189 8773 9232 9766 9996 9945 10322 4512 8
Kentucky 2924 3081 3238 3594 4102 4534 4949 5245 5458 8
Louisiana 2836 2724 2713 2946 3190 3235 3190 3161 3244 17
Maine 4238 4590 4360 4631 5020 5254 5504 5933 6327 8
Maryland 4533 4638 4946 5527 5722 5814 5645 5574 5453 14
Massachusetts 11053 10690 11860 11520 12939 12832 12677 12839 12699 14
Michigan 11176 11343 12133 12280 12693 13428 13725 14492 15000 15
Minnesota 4000 4337 4785 5353 5938 6143 6330 6330 6563 11
Mississippi 2929 3301 3525 3486 3623 3688 3790 3948 3958 9
Missouri 3964 4035 4427 5016 5327 5134 5395 5510 5538 13
Montana 3082 3358 3734 3865 4103 4369 4573 4490 4364 6
Nebraska 3093 3091 3486 3926 4242 4299 4398 4555 4682 7
Nevada 1826 1829 1854 1924 2049 2132 2175 2301 2603 6
New Hampshire 6308 6354 6642 6915 7182 7407 7597 7879 8068 5
New Jersey 5542 5820 6314 6779 7232 7616 7919 7635 7930 14
New Mexico 2545 2655 2760 2833 3111 3249 3333 3400 3486 8
New York 3758 3666 3684 4423 4346 4277 4147 4074 3974 45
North Carolina 2257 2509 2886 2977 3207 3194 3421 3556 3491 16
North Dakota 2819 2918 3112 3428 3924 4210 4405 4518 4577 8
Ohio 4572 4770 5412 5902 6996 7241 7433 6599 6355 31
Oklahoma 2097 2143 2305 2757 2977 3272 3442 3639 3938 17
Oregon 3636 3727 3812 4335 4686 4711 4753 4926 5037 9
Pennsylvania 5908 6091 6703 7065 7562 7694 7737 7954 8092 44
Rhode Island 4448 4494 4774 4955 5275 5522 5642 5854 6063 2
South Carolina 11074 11407 12806 14487 13065 13309 13627 14254 14619 13
South Dakota 3520 3603 3799 4049 4207 4303 4291 4590 4608 7
Tennessee 2940 3220 3400 3756 3919 4206 4258 4446 4541 9
Texas 2665 2738 2987 3171 3871 4054 4280 4487 4718 45
Utah 2193 2276 2469 2686 2885 3034 3185 3334 3362 7
Vermont 16328 16092 16420 16497 16783 16930 17003 17650 18067 5
Virginia 11982 11774 12572 13390 13571 13705 13899 5676 5901 15
Washington 3168 3238 3602 3807 3971 4108 4239 4375 4475 13
West Virginia 2594 2560 2754 2951 3248 3371 3456 3601 3731 12
Wisconsin 3284 3425 3608 4137 4571 4749 4877 5045 5153 14
Wyoming 2575 2714 2847 2873 2476 2538 2511 2480 2445 1
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education Statistics, Inte-
grated Postsecondary Education Data System (IPEDS), 2000-2008.
Notes:
(1) All values represent an enrollment-weighted average of degree-granting, public institutions within each state.
Each value is then indexed to year 2000 dollars using the Bureau of Labor Statistics CPI-U.
(2) The Community College of the District of Columbia was not established until after the time period analyzed.
Thus, there were no appropriate 2-year tuition costs for residents of the District of Columbia.
148
Table B.12: Published Price for Military Spouses at 4-Year Public Institutions
State 2000 2001 2002 2003 2004 2005 2006 2007 2008 Number of Schools
Alabama 2803 2995 3176 3285 3621 3959 3906 3847 3938 14
Alaska 2793 2626 2909 2940 3113 3003 3272 3477 3600 3
Arizona 2262 2257 2364 2406 3260 3601 3774 3888 3993 3
Arkansas 2772 2840 3177 3415 3621 3805 3962 4258 4324 11
California 2494 2421 2524 2609 3346 3526 3674 3606 3833 36
Colorado 2760 2771 2838 2951 3175 3116 3581 3646 3988 13
Connecticut 11218 11018 11305 11864 12818 13716 13847 14272 14393 11
Delaware 4598 4598 4777 5084 5543 5812 5974 6144 6267 2
District of Columbia 2070 2002 1966 1925 1883 1828 1762 1791 2511 2
Florida 1890 1887 2017 2053 2149 2212 2259 2265 2196 20
Georgia 7987 8191 8449 8835 2736 2787 2891 2997 3254 26
Hawaii 2575 2500 2534 2616 2697 2713 2696 3138 3540 4
Idaho 2588 2688 2859 3007 3016 3164 3331 3443 3500 4
Illinois 3993 4022 4319 4827 5133 5698 6035 6650 7175 12
Indiana 10106 10152 10618 11708 12662 13013 13200 13460 5220 15
Iowa 2998 3056 3300 3854 4541 4775 4781 4993 4973 3
Kansas 8064 8299 8385 8824 9679 10199 10403 10584 4245 8
Kentucky 2713 2790 2970 3159 3500 4011 4368 4812 5061 8
Louisiana 2327 2535 2592 2577 2881 3078 3074 3071 3057 17
Maine 3884 4028 4221 4435 4684 4867 5055 5371 5745 8
Maryland 4491 4356 4653 4722 5236 5463 5570 5520 5378 14
Massachusetts 10894 10699 10899 10905 12008 13449 13432 13556 13580 14
Michigan 10737 10744 11332 12107 12966 13154 13955 14653 15569 15
Minnesota 3750 3868 4260 4711 5223 5822 5919 6114 6092 11
Mississippi 2873 2864 3248 3453 3415 3521 3555 3688 3976 9
Missouri 3620 3713 3826 4275 4817 5119 5132 5243 5263 13
Montana 2930 2943 3251 3620 3756 3901 4178 4339 4267 6
Nebraska 2938 2884 3019 3424 3846 4134 4214 4327 4435 7
Nevada 1674 1776 1818 1839 1899 2005 2072 2138 2391 6
New Hampshire 6069 6097 6240 6500 6763 6973 7136 7396 7597 5
New Jersey 5143 5387 5713 6177 6640 7001 7344 7145 7374 14
New Mexico 2264 2461 2607 2701 2767 2960 3127 3232 3273 8
New York 3666 3607 3599 3598 4266 4174 4087 4020 3956 45
North Carolina 2068 2218 2471 2822 2916 3127 3118 3370 3456 16
North Dakota 2702 2737 2867 3062 3354 3809 4047 4289 4335 8
Ohio 4356 4415 4679 5319 5812 6794 6992 7220 6391 31
Oklahoma 2001 1943 2044 2164 2852 3069 3263 3437 3642 17
Oregon 3538 3481 3662 3691 4198 4504 4485 4564 4729 9
Pennsylvania 5600 5665 5936 6564 6997 7341 7415 7564 7673 44
Rhode Island 4268 4301 4415 4674 4848 5121 5321 5524 5662 2
South Carolina 8457 8605 9287 10608 11981 12955 13799 14456 15074 13
South Dakota 3275 3401 3536 3699 3955 4076 4055 4304 4321 7
Tennessee 2681 2843 3163 3331 3661 3790 4052 4134 4287 9
Texas 2494 2632 2745 2956 3194 3800 4010 4223 4454 45
Utah 2114 2124 2236 2440 2628 2801 2927 3119 3175 7
Vermont 15834 15819 16202 16494 17055 17480 17620 18219 18726 5
Virginia 11623 11580 11810 12643 13629 14083 14302 5325 5460 15
Washington 2929 2934 3062 3486 3696 3834 3990 4149 4232 13
West Virginia 2453 2469 2514 2703 2900 3164 3264 3395 3462 12
Wisconsin 3188 3175 3347 3525 4048 4441 4576 4721 4862 14
Wyoming 2416 2490 2666 2787 2353 2403 2446 2915 2692 1
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education Statistics, Inte-
grated Postsecondary Education Data System (IPEDS), 2000-2008.
Notes:
(1) All values represent an enrollment-weighted average of degree-granting, public institutions within each state.
Each value is then indexed to year 2000 dollars using the Bureau of Labor Statistics CPI-U.
(2) The Community College of the District of Columbia was not established until after the time period analyzed.
Thus, there were no appropriate 2-year tuition costs for residents of the District of Columbia.
149
Table B.13: Military Spouses Exempted From Residency Requirements (2000-2008)
State Always Never Sometimes Year Began State Always Never Sometimes Year Began
Alabama X Montana X
Alaska X Nebraska X
Arizona X Nevada X
Arkansas X New Hampshire X
California X New Jersey X
Colorado X New Mexico X
Connecticut X New York X
Delaware X North Carolina X
District of Columbia X North Dakota X
Florida X Ohio X
Georgia X 2004 Oklahoma X
Hawaii X Oregon X
Idaho X Pennsylvania X
Illinois X Rhode Island X
Indiana X 2008 South Carolina X
Iowa X South Dakota X
Kansas X 2008 Tennessee X
Kentucky X Texas X
Louisiana X Utah X
Maine X Vermont X
Maryland X Virginia X 2007
Massachusetts X Washington X
Michigan X West Virginia X
Minnesota X Wisconsin X
Mississippi X Wyoming X
Missouri X
Note: All states receiving federal funds for education are required to charge military spouses in-state tuition rates as of 2010.
SOURCE: Department of Defence, In-State Tuition State-By-State Summary, https://www.hrc.army.mil/SITE/Education/InState/index.htm
Figure B.3: In-State Tuition 2000-2008
Green: In-state rates apply
Yellow: State law changed during time period
Red: Out-of-state rates apply
150
Figure B.4: Mean Cost of Tuition and Fees Paid (2008)
Figure B.5: Enrollment Rates of Military Spouses (2008)
151
Table B.14: Effects of Average Price Paid on Attendance of 18- to 40-year-old Military
Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.014∗∗ -0.012∗ -0.013∗ -0.014∗∗ -0.013∗∗ -0.003
(0.005) (0.005) (0.005) (0.005) (0.005) (0.015)
Four-Year Costs 0.003 0.004 0.004 0.004 0.003 -0.006
(0.002) (0.002) (0.002) (0.002) (0.002) (0.007)
Other Household Income 0.000 0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.004 0.007 -0.001 -0.000 -0.001
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.031 0.028 0.024 0.025 0.024
(0.016) (0.016) (0.016) (0.016) (0.016)
White 0.012 0.012 0.011 0.009 0.008
(0.012) (0.012) (0.012) (0.012) (0.012)
Unemployment Rate 0.012∗∗ 0.012∗∗ 0.012∗∗ 0.003 0.015
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.041∗∗∗ -0.041∗∗∗ -0.040∗∗∗ -0.040∗∗
(0.012) (0.012) (0.012) (0.012)
Number of Children Under 5 -0.036∗∗∗ -0.035∗∗∗ -0.036∗∗∗ -0.036∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.004 -0.002 -0.005 -0.005
(0.012) (0.012) (0.012) (0.012)
Officer -0.017 -0.013 -0.012
(0.021) (0.021) (0.021)
Spouse- High School -0.014 -0.017 -0.014
(0.031) (0.031) (0.031)
Spouse- Some College 0.051 0.048 0.051
(0.031) (0.031) (0.031)
Spouse- College Degree 0.059 0.056 0.057
(0.032) (0.032) (0.032)
Spouse- Graduate Degree 0.071∗ 0.069∗ 0.072∗
(0.035) (0.035) (0.035)
Additional Controls
Year X X
State X
N 8288 8288 8288 8288 8288 8288
Adj. R2 0.001 0.010 0.020 0.027 0.029 0.029
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
152
Table B.15: Effects of Average Price Paid on Attendance of 18- to 24-year-old Military
Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.012 -0.012 -0.012 -0.014 -0.013 -0.013
(0.010) (0.010) (0.010) (0.010) (0.010) (0.027)
Four-Year Costs 0.002 0.006 0.006 0.007 0.004 -0.001
(0.005) (0.005) (0.005) (0.005) (0.005) (0.012)
Other Household Income -0.000 -0.000 -0.001 -0.001 -0.001∗
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.040 -0.020 -0.036 -0.027 -0.029
(0.100) (0.099) (0.099) (0.099) (0.099)
Age2 0.001 0.001 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
Male 0.002 -0.007 -0.001 0.006 0.001
(0.034) (0.034) (0.033) (0.033) (0.034)
White 0.028 0.025 0.024 0.023 0.021
(0.023) (0.023) (0.023) (0.023) (0.023)
Unemployment Rate 0.027∗∗∗ 0.026∗∗∗ 0.026∗∗∗ 0.013 0.033∗
(0.007) (0.006) (0.006) (0.008) (0.016)
Newborn -0.069∗∗∗ -0.075∗∗∗ -0.076∗∗∗ -0.072∗∗∗
(0.021) (0.021) (0.021) (0.021)
Number of Children Under 5 -0.036∗∗ -0.028∗ -0.027∗ -0.030∗
(0.012) (0.012) (0.012) (0.012)
Looking for Work -0.003 -0.000 -0.005 -0.002
(0.019) (0.019) (0.019) (0.019)
Officer 0.019 0.022 0.029
(0.043) (0.043) (0.043)
Spouse- High School -0.033 -0.030 -0.032
(0.049) (0.049) (0.049)
Spouse- Some College 0.044 0.046 0.044
(0.049) (0.049) (0.049)
Spouse- College Degree 0.098 0.103 0.099
(0.056) (0.056) (0.056)
Spouse- Graduate Degree 0.210∗ 0.222∗ 0.206
(0.106) (0.106) (0.107)
Additional Controls
Year X X
State X
N 2969 2969 2969 2969 2969 2969
Adj. R2 0.000 0.005 0.020 0.031 0.035 0.037
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
153
Table B.16: Effects of Average Price Paid on Attendance of 25- to 40-year-old Military
Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.013∗ -0.013∗ -0.013∗ -0.014∗ -0.014∗ -0.000
(0.006) (0.006) (0.006) (0.006) (0.006) (0.019)
Four-Year Costs 0.002 0.003 0.003 0.003 0.002 -0.008
(0.003) (0.003) (0.003) (0.003) (0.003) (0.008)
Other Household Income 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.008 -0.006 -0.008 -0.006 -0.004
(0.016) (0.016) (0.016) (0.016) (0.016)
Age2 0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.042∗ 0.041∗ 0.036∗ 0.036∗ 0.036∗
(0.018) (0.018) (0.018) (0.018) (0.018)
White 0.004 0.004 0.003 0.002 0.003
(0.014) (0.014) (0.014) (0.014) (0.014)
Unemployment Rate 0.002 0.003 0.002 -0.004 0.000
(0.004) (0.004) (0.004) (0.006) (0.010)
Newborn -0.016 -0.014 -0.014 -0.013
(0.016) (0.016) (0.016) (0.016)
Number of Children Under 5 -0.036∗∗∗ -0.037∗∗∗ -0.038∗∗∗ -0.037∗∗∗
(0.007) (0.007) (0.007) (0.007)
Looking for Work -0.005 -0.004 -0.006 -0.006
(0.015) (0.015) (0.015) (0.015)
Officer -0.032 -0.028 -0.030
(0.023) (0.023) (0.023)
Spouse- High School 0.002 -0.003 0.003
(0.041) (0.041) (0.041)
Spouse- Some College 0.056 0.050 0.057
(0.040) (0.040) (0.040)
Spouse- College Degree 0.051 0.045 0.050
(0.041) (0.041) (0.041)
Spouse- Graduate Degree 0.063 0.059 0.065
(0.043) (0.043) (0.043)
Additional Controls
Year X X
State X
N 5319 5319 5319 5319 5319 5319
Adj. R2 0.001 0.006 0.013 0.017 0.019 0.018
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
154
Table B.17: Effects of Average Price Paid on Attendance of 18- to 40-year-old Military
Spouses Who Do Not Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.013∗ -0.010∗ -0.011∗ -0.012∗ -0.012∗ -0.014
(0.005) (0.005) (0.005) (0.005) (0.005) (0.016)
Four-Year Costs 0.003 0.004 0.004 0.004 0.003 0.000
(0.002) (0.002) (0.002) (0.002) (0.003) (0.007)
Other Household Income 0.000 0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.013 -0.004 -0.010 -0.010 -0.011
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 0.000 -0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.033∗ 0.031 0.026 0.027 0.025
(0.016) (0.016) (0.016) (0.016) (0.016)
White -0.000 0.001 -0.001 -0.002 -0.002
(0.013) (0.013) (0.013) (0.013) (0.013)
Unemployment Rate 0.013∗∗∗ 0.014∗∗∗ 0.013∗∗∗ 0.007 0.021∗
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.045∗∗∗ -0.046∗∗∗ -0.044∗∗∗ -0.043∗∗∗
(0.013) (0.013) (0.013) (0.013)
Number of Children Under 5 -0.029∗∗∗ -0.029∗∗∗ -0.030∗∗∗ -0.029∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.004 -0.002 -0.004 -0.004
(0.012) (0.012) (0.012) (0.012)
Officer -0.014 -0.012 -0.010
(0.022) (0.022) (0.023)
Spouse- High School -0.019 -0.020 -0.017
(0.032) (0.032) (0.032)
Spouse- Some College 0.041 0.039 0.041
(0.032) (0.032) (0.032)
Spouse- College Degree 0.056 0.053 0.055
(0.034) (0.034) (0.034)
Spouse- Graduate Degree 0.069 0.067 0.071
(0.037) (0.037) (0.037)
Additional Controls
Year X X
State X
N 6938 6938 6938 6938 6938 6938
Adj. R2 0.001 0.010 0.019 0.025 0.028 0.027
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
155
Table B.18: Effects of Average Price Paid on Attendance of 18- to 40-year-old Military
Spouses Who Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.018 -0.018 -0.019 -0.019 -0.019 0.059
(0.015) (0.015) (0.015) (0.015) (0.015) (0.045)
Four-Year Costs 0.001 0.001 0.000 0.001 -0.002 -0.045∗
(0.007) (0.007) (0.007) (0.007) (0.007) (0.019)
Other Household Income -0.000 -0.000 -0.000 -0.000 -0.000
(0.001) (0.001) (0.001) (0.001) (0.001)
Age -0.044 -0.025 -0.028 -0.029 -0.031
(0.023) (0.024) (0.024) (0.024) (0.024)
Age2 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.039 0.035 0.040 0.039 0.043
(0.050) (0.049) (0.050) (0.050) (0.051)
White 0.079∗ 0.074∗ 0.074∗ 0.072∗ 0.067
(0.034) (0.034) (0.034) (0.034) (0.035)
Unemployment Rate -0.009 -0.009 -0.009 -0.018 -0.017
(0.010) (0.010) (0.010) (0.013) (0.024)
Newborn -0.022 -0.017 -0.020 -0.019
(0.038) (0.038) (0.038) (0.039)
Number of Children Under 5 -0.048∗∗ -0.049∗∗ -0.045∗ -0.054∗∗
(0.018) (0.018) (0.018) (0.018)
Looking for Work 0.005 0.003 0.002 0.009
(0.036) (0.036) (0.036) (0.037)
Officer -0.024 -0.016 -0.021
(0.050) (0.050) (0.051)
Spouse- High School 0.027 0.019 0.031
(0.099) (0.099) (0.100)
Spouse- Some College 0.090 0.082 0.091
(0.096) (0.096) (0.097)
Spouse- College Degree 0.036 0.028 0.035
(0.097) (0.097) (0.098)
Spouse- Graduate Degree 0.062 0.055 0.068
(0.100) (0.100) (0.101)
Additional Controls
Year X X
State X
N 1350 1350 1350 1350 1350 1350
Adj. R2 0.001 0.040 0.047 0.048 0.057 0.058
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
156
Table B.19: Effects of Average Published Price on Attendance of 18- to 40-year-old Mili-
tary Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.015∗∗ -0.013∗ -0.014∗ -0.015∗∗ -0.015∗∗ -0.007
(0.005) (0.005) (0.005) (0.005) (0.005) (0.017)
Four-Year Costs 0.003 0.004 0.004 0.005 0.003 -0.004
(0.002) (0.003) (0.003) (0.002) (0.003) (0.007)
Other Household Income 0.000 0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.004 0.007 -0.001 -0.000 -0.001
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.031 0.028 0.024 0.025 0.024
(0.016) (0.016) (0.016) (0.016) (0.016)
White 0.012 0.012 0.011 0.009 0.008
(0.012) (0.012) (0.012) (0.012) (0.012)
Unemployment Rate 0.012∗∗ 0.012∗∗ 0.012∗∗ 0.004 0.015
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.041∗∗∗ -0.041∗∗∗ -0.041∗∗∗ -0.040∗∗
(0.012) (0.012) (0.012) (0.012)
Number of Children Under 5 -0.036∗∗∗ -0.035∗∗∗ -0.036∗∗∗ -0.036∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.005 -0.002 -0.005 -0.005
(0.012) (0.012) (0.012) (0.012)
Officer -0.017 -0.013 -0.012
(0.021) (0.020) (0.021)
Spouse- High School -0.014 -0.017 -0.014
(0.031) (0.031) (0.031)
Spouse- Some College 0.051 0.048 0.051
(0.031) (0.031) (0.031)
Spouse- College Degree 0.059 0.055 0.057
(0.032) (0.032) (0.032)
Spouse- Graduate Degree 0.071∗ 0.069∗ 0.073∗
(0.035) (0.035) (0.035)
Additional Controls
Year X X
State X
N 8288 8288 8288 8288 8288 8288
Adj. R2 0.001 0.011 0.020 0.027 0.029 0.029
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
157
Table B.20: Effects of Average Published Price on Attendance of 18- to 24-year-old Mili-
tary Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.013 -0.013 -0.013 -0.015 -0.015 -0.024
(0.011) (0.011) (0.011) (0.011) (0.011) (0.029)
Four-Year Costs 0.002 0.006 0.006 0.007 0.005 0.004
(0.005) (0.005) (0.005) (0.005) (0.005) (0.013)
Other Household Income -0.000 -0.000 -0.001 -0.001 -0.001∗
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.039 -0.019 -0.036 -0.026 -0.028
(0.100) (0.099) (0.099) (0.099) (0.099)
Age2 0.001 0.000 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
Male 0.002 -0.007 -0.001 0.006 0.001
(0.034) (0.034) (0.033) (0.033) (0.034)
White 0.028 0.025 0.024 0.023 0.022
(0.023) (0.023) (0.023) (0.023) (0.023)
Unemployment Rate 0.027∗∗∗ 0.026∗∗∗ 0.026∗∗∗ 0.013 0.033∗
(0.007) (0.007) (0.007) (0.008) (0.016)
Newborn -0.069∗∗∗ -0.076∗∗∗ -0.076∗∗∗ -0.072∗∗∗
(0.021) (0.021) (0.021) (0.021)
Number of Children Under 5 -0.036∗∗ -0.028∗ -0.027∗ -0.030∗
(0.012) (0.012) (0.012) (0.012)
Looking for Work -0.004 -0.000 -0.005 -0.003
(0.019) (0.019) (0.019) (0.019)
Officer 0.019 0.021 0.028
(0.043) (0.043) (0.043)
Spouse- High School -0.033 -0.031 -0.032
(0.049) (0.049) (0.049)
Spouse- Some College 0.044 0.045 0.044
(0.049) (0.049) (0.049)
Spouse- college Degree 0.098 0.103 0.099
(0.056) (0.056) (0.056)
Spouse- Graduate Degree 0.210∗ 0.221∗ 0.206
(0.106) (0.106) (0.107)
Additional Controls
Year X X
State X
N 2969 2969 2969 2969 2969 2969
Adj. R2 0.000 0.005 0.020 0.031 0.035 0.037
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
158
Table B.21: Effects of Average Published Price on Attendance of 25- to 40-year-old Mili-
tary Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.014∗ -0.014∗ -0.015∗ -0.015∗ -0.015∗ -0.000
(0.006) (0.006) (0.006) (0.006) (0.006) (0.021)
Four-Year Costs 0.003 0.003 0.003 0.004 0.003 -0.008
(0.003) (0.003) (0.003) (0.003) (0.003) (0.009)
Other Household Income 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.008 -0.006 -0.008 -0.006 -0.004
(0.016) (0.016) (0.016) (0.016) (0.016)
Age2 0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.043∗ 0.041∗ 0.037∗ 0.036∗ 0.036∗
(0.018) (0.018) (0.018) (0.018) (0.018)
White 0.004 0.004 0.003 0.002 0.003
(0.014) (0.014) (0.014) (0.014) (0.014)
Unemployment Rate 0.002 0.002 0.002 -0.004 0.000
(0.004) (0.004) (0.004) (0.006) (0.010)
Newborn -0.016 -0.014 -0.014 -0.013
(0.016) (0.016) (0.016) (0.016)
Number of Children Under 5 -0.036∗∗∗ -0.037∗∗∗ -0.038∗∗∗ -0.037∗∗∗
(0.007) (0.007) (0.007) (0.007)
Looking for Work -0.005 -0.004 -0.006 -0.006
(0.015) (0.015) (0.015) (0.015)
Officer -0.032 -0.028 -0.030
(0.023) (0.023) (0.023)
Spouse- High School 0.002 -0.003 0.003
(0.041) (0.041) (0.041)
Spouse- Some College 0.056 0.050 0.057
(0.040) (0.040) (0.040)
Spouse- College Degree 0.051 0.045 0.050
(0.041) (0.041) (0.041)
Spouse- Graduate Degree 0.063 0.059 0.065
(0.043) (0.043) (0.043)
Additional Controls
Year X X
State X
N 5319 5319 5319 5319 5319 5319
Adj. R2 0.001 0.007 0.013 0.017 0.019 0.018
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
159
Table B.22: Effects of Average Published Price on Attendance of 18- to 40-year-old Mili-
tary Spouses Who Do Not Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.013∗ -0.010 -0.011∗ -0.012∗ -0.012∗ -0.012
(0.006) (0.006) (0.006) (0.006) (0.006) (0.018)
Four-Year Costs 0.003 0.004 0.004 0.004 0.003 0.000
(0.003) (0.003) (0.003) (0.003) (0.003) (0.008)
Other Household Income 0.000 0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Age -0.013 -0.004 -0.010 -0.010 -0.011
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 0.000 -0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.033∗ 0.031 0.026 0.027 0.025
(0.016) (0.016) (0.016) (0.016) (0.016)
White -0.000 0.001 -0.001 -0.002 -0.002
(0.013) (0.013) (0.013) (0.013) (0.013)
Unemployment Rate 0.014∗∗∗ 0.014∗∗∗ 0.013∗∗∗ 0.007 0.021∗
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.045∗∗∗ -0.046∗∗∗ -0.044∗∗∗ -0.043∗∗∗
(0.013) (0.013) (0.013) (0.013)
Number of Children Under 5 -0.029∗∗∗ -0.029∗∗∗ -0.029∗∗∗ -0.029∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.004 -0.002 -0.004 -0.004
(0.012) (0.012) (0.012) (0.012)
Officer -0.015 -0.012 -0.010
(0.022) (0.022) (0.023)
Spouse- High School -0.019 -0.020 -0.017
(0.032) (0.032) (0.032)
Spouse- Some College 0.041 0.039 0.041
(0.032) (0.032) (0.032)
Spouse- College Degree 0.056 0.053 0.055
(0.034) (0.034) (0.034)
Spouse- Graduate Degree 0.069 0.067 0.071
(0.037) (0.037) (0.037)
Additional Controls
Year X X
State X
N 6938 6938 6938 6938 6938 6938
Adj. R2 0.001 0.010 0.019 0.025 0.028 0.027
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
160
Table B.23: Effects of Average Published Price on Attendance of 18- to 40-year-old Mili-
tary Spouses Who Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.026 -0.026 -0.027 -0.027 -0.028 0.028
(0.016) (0.016) (0.016) (0.016) (0.016) (0.050)
Four-Year Costs 0.004 0.003 0.003 0.004 0.001 -0.033
(0.007) (0.007) (0.007) (0.007) (0.007) (0.022)
Other Household Income -0.000 -0.000 -0.000 -0.000 -0.000
(0.001) (0.001) (0.001) (0.001) (0.001)
Age -0.043 -0.025 -0.028 -0.029 -0.032
(0.023) (0.024) (0.024) (0.024) (0.024)
Age2 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.039 0.035 0.041 0.040 0.042
(0.050) (0.049) (0.050) (0.050) (0.051)
White 0.080∗ 0.074∗ 0.075∗ 0.073∗ 0.067
(0.034) (0.034) (0.034) (0.034) (0.035)
Unemployment Rate -0.010 -0.010 -0.009 -0.018 -0.016
(0.010) (0.010) (0.010) (0.013) (0.024)
Newborn -0.022 -0.016 -0.020 -0.020
(0.038) (0.038) (0.038) (0.039)
Number of Children Under 5 -0.049∗∗ -0.049∗∗ -0.045∗ -0.054∗∗
(0.018) (0.018) (0.018) (0.018)
Looking for Work 0.005 0.003 0.002 0.008
(0.036) (0.036) (0.036) (0.037)
Officer -0.024 -0.016 -0.021
(0.050) (0.050) (0.051)
Spouse- High School 0.028 0.020 0.034
(0.099) (0.099) (0.100)
Spouse- Some College 0.091 0.084 0.092
(0.096) (0.096) (0.097)
Spouse- College Degree 0.038 0.030 0.036
(0.097) (0.097) (0.098)
Spouse- Graduate Degree 0.063 0.057 0.069
(0.100) (0.100) (0.102)
Additional Controls
Year X X
State X
N 1350 1350 1350 1350 1350 1350
Adj. R2 0.002 0.041 0.048 0.049 0.058 0.056
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
161
Table B.24: Effects of Log of Average Price Paid on Attendance of 18- to 40-year-old
Military Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.018∗ -0.015 -0.015∗ -0.016∗ -0.017∗ 0.003
(0.008) (0.008) (0.008) (0.008) (0.008) (0.035)
Four-Year Costs 0.004 0.008 0.010 0.011 0.001 -0.044
(0.011) (0.011) (0.011) (0.011) (0.011) (0.037)
Other Household Income 0.014 0.015 0.006 0.001 0.003
(0.008) (0.008) (0.008) (0.008) (0.009)
Age -0.005 0.005 -0.001 -0.000 -0.001
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.036∗ 0.033∗ 0.028 0.028 0.028
(0.016) (0.016) (0.016) (0.016) (0.016)
White 0.013 0.013 0.012 0.011 0.009
(0.012) (0.012) (0.012) (0.012) (0.012)
Unemployment Rate 0.011∗∗ 0.011∗∗ 0.011∗∗ 0.002 0.014
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.042∗∗∗ -0.042∗∗∗ -0.041∗∗∗ -0.040∗∗
(0.012) (0.012) (0.012) (0.012)
Number of Children Under 5 -0.036∗∗∗ -0.035∗∗∗ -0.035∗∗∗ -0.036∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.003 -0.001 -0.004 -0.004
(0.012) (0.012) (0.012) (0.012)
Officer -0.016 -0.012 -0.012
(0.021) (0.021) (0.021)
Spouse- High School -0.009 -0.011 -0.008
(0.031) (0.031) (0.031)
Spouse- Some College 0.056 0.054 0.057
(0.030) (0.030) (0.031)
Spouse- College Degree 0.059 0.057 0.058
(0.032) (0.032) (0.032)
Spouse- Graduate Degree 0.069∗ 0.068 0.070∗
(0.035) (0.035) (0.035)
Additional Controls
Year X X
State X
N 8288 8242 8242 8242 8242 8242
Adj. R2 0.001 0.011 0.021 0.027 0.029 0.029
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
162
Table B.25: Effects of Log of Average Price Paid on Attendance of 18- to 24-year-old
Military Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.018 -0.011 -0.010 -0.013 -0.015 -0.029
(0.014) (0.015) (0.014) (0.014) (0.014) (0.059)
Four-Year Costs 0.009 0.018 0.015 0.019 0.004 -0.011
(0.020) (0.020) (0.020) (0.020) (0.020) (0.064)
Other Household Income 0.006 0.007 -0.004 -0.013 -0.011
(0.015) (0.014) (0.015) (0.015) (0.015)
Age -0.039 -0.020 -0.036 -0.028 -0.032
(0.100) (0.100) (0.099) (0.099) (0.100)
Age2 0.001 0.000 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
Male 0.006 -0.004 0.002 0.008 0.004
(0.034) (0.034) (0.034) (0.034) (0.034)
White 0.031 0.028 0.027 0.026 0.024
(0.023) (0.023) (0.023) (0.023) (0.023)
Unemployment Rate 0.026∗∗∗ 0.025∗∗∗ 0.024∗∗∗ 0.011 0.032∗
(0.006) (0.006) (0.006) (0.008) (0.016)
Newborn -0.070∗∗∗ -0.077∗∗∗ -0.077∗∗∗ -0.072∗∗∗
(0.021) (0.021) (0.021) (0.021)
Number of Children Under 5 -0.037∗∗ -0.029∗ -0.027∗ -0.030∗
(0.012) (0.012) (0.012) (0.012)
Looking for Work -0.002 0.001 -0.004 -0.002
(0.019) (0.019) (0.019) (0.019)
Officer 0.021 0.024 0.031
(0.043) (0.043) (0.043)
Spouse- High School -0.024 -0.020 -0.021
(0.049) (0.049) (0.049)
Spouse- Some College 0.053 0.057 0.055
(0.049) (0.048) (0.049)
Spouse- College Degree 0.102 0.109 0.105
(0.056) (0.056) (0.056)
Spouse- Graduate Degree 0.211∗ 0.224∗ 0.206
(0.106) (0.106) (0.107)
Additional Controls
Year X X
State X
N 2969 2952 2952 2952 2952 2952
Adj. R2 -0.000 0.004 0.020 0.030 0.035 0.036
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
163
Table B.26: Effects of Log of Average Price Paid on Attendance of 25- to 40-year-old
Military Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.017 -0.017 -0.018∗ -0.018∗ -0.019∗ 0.006
(0.009) (0.009) (0.009) (0.009) (0.009) (0.045)
Four-Year Costs 0.001 0.003 0.005 0.006 -0.003 -0.051
(0.012) (0.012) (0.012) (0.012) (0.013) (0.046)
Other Household Income 0.018 0.018 0.015 0.011 0.012
(0.009) (0.009) (0.010) (0.010) (0.010)
Age -0.007 -0.005 -0.007 -0.004 -0.002
(0.016) (0.016) (0.016) (0.016) (0.016)
Age2 0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.048∗∗ 0.047∗∗ 0.041∗ 0.041∗ 0.041∗
(0.018) (0.018) (0.018) (0.018) (0.018)
White 0.004 0.004 0.003 0.002 0.003
(0.014) (0.014) (0.014) (0.014) (0.014)
Unemployment Rate 0.002 0.003 0.002 -0.005 -0.001
(0.004) (0.004) (0.004) (0.005) (0.010)
Newborn -0.016 -0.014 -0.014 -0.014
(0.016) (0.016) (0.016) (0.016)
Number of Children Under 5 -0.035∗∗∗ -0.036∗∗∗ -0.037∗∗∗ -0.037∗∗∗
(0.007) (0.007) (0.007) (0.007)
Looking for Work -0.003 -0.003 -0.005 -0.005
(0.015) (0.015) (0.015) (0.015)
Officer -0.031 -0.027 -0.030
(0.023) (0.023) (0.023)
Spouse- High School 0.007 0.002 0.007
(0.041) (0.041) (0.041)
Spouse- Some College 0.060 0.055 0.060
(0.040) (0.040) (0.040)
Spouse- College Degree 0.050 0.045 0.050
(0.041) (0.041) (0.041)
Spouse- Graduate Degree 0.061 0.057 0.064
(0.043) (0.043) (0.043)
Additional Controls
Year X X
State X
N 5319 5290 5290 5290 5290 5290
Adj. R2 0.001 0.007 0.014 0.017 0.019 0.018
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
164
Table B.27: Effects of Log of Average Price Paid on Attendance of 18- to 40-year-old
Military Spouses Who Do Not Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.019∗ -0.014 -0.014 -0.015 -0.016∗ -0.056
(0.008) (0.008) (0.008) (0.008) (0.008) (0.038)
Four-Year Costs 0.007 0.011 0.013 0.013 0.006 0.023
(0.011) (0.011) (0.011) (0.011) (0.011) (0.040)
Other Household Income 0.015 0.016∗ 0.007 0.002 0.003
(0.008) (0.008) (0.009) (0.009) (0.009)
Age -0.014 -0.005 -0.011 -0.010 -0.011
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 0.000 -0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.037∗ 0.035∗ 0.029 0.030 0.028
(0.017) (0.017) (0.017) (0.017) (0.017)
White 0.001 0.002 0.000 -0.001 -0.001
(0.013) (0.013) (0.013) (0.013) (0.013)
Unemployment Rate 0.013∗∗∗ 0.013∗∗∗ 0.013∗∗∗ 0.006 0.021∗
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.046∗∗∗ -0.046∗∗∗ -0.045∗∗∗ -0.044∗∗∗
(0.013) (0.013) (0.013) (0.013)
Number of Children Under 5 -0.029∗∗∗ -0.029∗∗∗ -0.029∗∗∗ -0.029∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.002 -0.001 -0.003 -0.003
(0.012) (0.012) (0.012) (0.012)
Officer -0.014 -0.011 -0.009
(0.022) (0.022) (0.023)
Spouse- High School -0.014 -0.014 -0.012
(0.032) (0.032) (0.032)
Spouse- Some College 0.045 0.044 0.046
(0.032) (0.031) (0.032)
Spouse- College Degree 0.055 0.053 0.055
(0.034) (0.034) (0.034)
Spouse- Graduate Degree 0.066 0.064 0.067
(0.037) (0.037) (0.037)
Additional Controls
Year X X
State X
N 6938 6899 6899 6899 6899 6899
Adj. R2 0.001 0.011 0.020 0.025 0.028 0.028
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
165
Table B.28: Effects of Log of Average Price Paid on Attendance of 18- to 40-year-old
Military Spouses Who Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.018 -0.023 -0.024 -0.026 -0.026 0.251∗∗
(0.023) (0.023) (0.023) (0.023) (0.023) (0.094)
Four-Year Costs -0.014 -0.014 -0.016 -0.014 -0.034 -0.372∗∗∗
(0.030) (0.030) (0.030) (0.030) (0.031) (0.101)
Other Household Income -0.015 -0.018 -0.004 -0.007 0.008
(0.023) (0.023) (0.026) (0.026) (0.027)
Age -0.040 -0.023 -0.027 -0.027 -0.029
(0.023) (0.024) (0.024) (0.024) (0.025)
Age2 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.047 0.041 0.049 0.048 0.057
(0.050) (0.050) (0.051) (0.051) (0.052)
White 0.079∗ 0.074∗ 0.075∗ 0.073∗ 0.065
(0.034) (0.034) (0.035) (0.034) (0.035)
Unemployment Rate -0.009 -0.009 -0.009 -0.020 -0.020
(0.010) (0.010) (0.010) (0.013) (0.024)
Newborn -0.024 -0.019 -0.022 -0.019
(0.038) (0.038) (0.038) (0.039)
Number of Children Under 5 -0.047∗∗ -0.047∗∗ -0.043∗ -0.050∗∗
(0.018) (0.018) (0.018) (0.018)
Looking for Work 0.006 0.003 0.002 0.008
(0.036) (0.036) (0.036) (0.037)
Officer -0.024 -0.016 -0.023
(0.050) (0.050) (0.051)
Spouse- High School 0.031 0.025 0.051
(0.098) (0.098) (0.099)
Spouse- Some College 0.095 0.090 0.108
(0.096) (0.095) (0.096)
Spouse- College Degree 0.040 0.034 0.048
(0.097) (0.097) (0.098)
Spouse- Graduate Degree 0.065 0.060 0.079
(0.100) (0.100) (0.101)
Additional Controls
Year X X
State X
N 1350 1343 1343 1343 1343 1343
Adj. R2 0.001 0.040 0.046 0.047 0.057 0.062
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
166
Table B.29: Effects of Log of Average Published Price on Attendance of 18- to 40-year-old
Military Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.019∗ -0.016∗ -0.016∗ -0.017∗ -0.019∗ -0.011
(0.008) (0.008) (0.008) (0.008) (0.008) (0.033)
Four-Year Costs 0.005 0.009 0.010 0.012 0.002 -0.029
(0.011) (0.011) (0.011) (0.011) (0.011) (0.037)
Other Household Income 0.014 0.015 0.006 0.001 0.003
(0.008) (0.008) (0.008) (0.008) (0.009)
Age -0.005 0.005 -0.001 -0.000 -0.001
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 -0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.036∗ 0.033∗ 0.028 0.028 0.028
(0.016) (0.016) (0.016) (0.016) (0.016)
White 0.013 0.013 0.012 0.011 0.009
(0.012) (0.012) (0.012) (0.012) (0.012)
Unemployment Rate 0.011∗∗ 0.011∗∗ 0.011∗∗ 0.002 0.014
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.042∗∗∗ -0.042∗∗∗ -0.041∗∗∗ -0.040∗∗
(0.012) (0.012) (0.012) (0.012)
Number of Children Under 5 -0.036∗∗∗ -0.035∗∗∗ -0.035∗∗∗ -0.036∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.003 -0.001 -0.004 -0.004
(0.012) (0.012) (0.012) (0.012)
Officer -0.016 -0.012 -0.012
(0.021) (0.021) (0.021)
Spouse- High School -0.009 -0.011 -0.008
(0.031) (0.031) (0.031)
Spouse- Some College 0.056 0.054 0.057
(0.030) (0.030) (0.031)
Spouse- College Degree 0.059 0.057 0.058
(0.032) (0.032) (0.032)
Spouse- Graduate Degree 0.069∗ 0.068 0.070∗
(0.035) (0.035) (0.035)
Additional Controls
Year X X
State X
N 8288 8242 8242 8242 8242 8242
Adj. R2 0.001 0.011 0.021 0.027 0.030 0.029
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
167
Table B.30: Effects of Log of Average Published Price on Attendance of 18- to 24-year-old
Military Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.016 -0.010 -0.008 -0.011 -0.014 -0.018
(0.015) (0.015) (0.015) (0.015) (0.015) (0.059)
Four-Year Costs 0.007 0.017 0.014 0.018 0.003 -0.013
(0.020) (0.020) (0.020) (0.020) (0.020) (0.065)
Other Household Income 0.006 0.007 -0.004 -0.013 -0.011
(0.015) (0.014) (0.015) (0.015) (0.015)
Age -0.039 -0.020 -0.036 -0.028 -0.030
(0.100) (0.100) (0.099) (0.099) (0.100)
Age2 0.001 0.000 0.001 0.001 0.001
(0.002) (0.002) (0.002) (0.002) (0.002)
Male 0.006 -0.004 0.002 0.008 0.004
(0.034) (0.034) (0.034) (0.034) (0.034)
White 0.031 0.028 0.027 0.026 0.024
(0.023) (0.023) (0.023) (0.023) (0.023)
Unemployment Rate 0.026∗∗∗ 0.025∗∗∗ 0.024∗∗∗ 0.012 0.033∗
(0.006) (0.006) (0.006) (0.008) (0.016)
Newborn -0.070∗∗∗ -0.077∗∗∗ -0.077∗∗∗ -0.072∗∗∗
(0.021) (0.021) (0.021) (0.021)
Number of Children Under 5 -0.037∗∗ -0.028∗ -0.027∗ -0.030∗
(0.012) (0.012) (0.012) (0.012)
Looking for Work -0.003 0.000 -0.004 -0.002
(0.019) (0.019) (0.019) (0.019)
Officer 0.021 0.024 0.031
(0.043) (0.043) (0.043)
Spouse- High School -0.024 -0.020 -0.021
(0.049) (0.049) (0.049)
Spouse- Some College 0.053 0.057 0.056
(0.049) (0.048) (0.049)
Spouse- College Degree 0.102 0.109 0.105
(0.056) (0.056) (0.056)
Spouse- Graduate Degree 0.211∗ 0.224∗ 0.207
(0.106) (0.106) (0.107)
Additional Controls
Year X X
State X
N 2969 2952 2952 2952 2952 2952
Adj. R2 -0.000 0.004 0.019 0.030 0.035 0.036
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
168
Table B.31: Effects of Log of Average Published Price on Attendance of 25- to 40-year-old
Military Spouses
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.019∗ -0.020∗ -0.020∗ -0.020∗ -0.022∗ -0.023
(0.009) (0.009) (0.009) (0.009) (0.009) (0.041)
Four-Year Costs 0.003 0.005 0.007 0.007 -0.000 -0.026
(0.012) (0.012) (0.012) (0.012) (0.012) (0.045)
Other Household Income 0.018 0.019∗ 0.015 0.011 0.012
(0.009) (0.009) (0.010) (0.010) (0.010)
Age -0.007 -0.005 -0.007 -0.004 -0.002
(0.016) (0.016) (0.016) (0.016) (0.016)
Age2 0.000 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.049∗∗ 0.047∗∗ 0.042∗ 0.041∗ 0.041∗
(0.018) (0.018) (0.018) (0.018) (0.018)
White 0.004 0.005 0.003 0.002 0.003
(0.014) (0.014) (0.014) (0.014) (0.014)
Unemployment Rate 0.002 0.002 0.002 -0.005 -0.001
(0.004) (0.004) (0.004) (0.005) (0.010)
Newborn -0.016 -0.014 -0.014 -0.014
(0.016) (0.016) (0.016) (0.016)
Number of Children Under 5 -0.035∗∗∗ -0.036∗∗∗ -0.037∗∗∗ -0.037∗∗∗
(0.007) (0.007) (0.007) (0.007)
Looking for Work -0.003 -0.003 -0.005 -0.006
(0.015) (0.015) (0.015) (0.015)
Officer -0.031 -0.027 -0.030
(0.023) (0.023) (0.023)
Spouse- High School 0.007 0.002 0.007
(0.041) (0.041) (0.041)
Spouse- Some College 0.060 0.055 0.060
(0.040) (0.040) (0.040)
Spouse- College Degree 0.050 0.045 0.050
(0.041) (0.041) (0.041)
Spouse- Graduate Degree 0.061 0.057 0.064
(0.043) (0.043) (0.043)
Additional Controls
Year X X
State X
N 5319 5290 5290 5290 5290 5290
Adj. R2 0.001 0.007 0.014 0.017 0.019 0.018
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
169
Table B.32: Effects of Log of Average Published Price on Attendance of 18- to 40-year-old
Military Spouses Who Do Not Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.019∗ -0.014 -0.014 -0.015 -0.016∗ -0.054
(0.008) (0.008) (0.008) (0.008) (0.008) (0.035)
Four-Year Costs 0.007 0.011 0.012 0.013 0.006 0.024
(0.011) (0.011) (0.011) (0.011) (0.011) (0.039)
Other Household Income 0.015 0.016∗ 0.007 0.002 0.003
(0.008) (0.008) (0.009) (0.009) (0.009)
Age -0.014 -0.005 -0.011 -0.010 -0.011
(0.007) (0.007) (0.007) (0.007) (0.007)
Age2 0.000 -0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.037∗ 0.035∗ 0.029 0.030 0.029
(0.017) (0.017) (0.017) (0.017) (0.017)
White 0.001 0.002 0.000 -0.001 -0.001
(0.013) (0.013) (0.013) (0.013) (0.013)
Unemployment Rate 0.013∗∗∗ 0.013∗∗∗ 0.013∗∗∗ 0.006 0.020∗
(0.004) (0.004) (0.004) (0.005) (0.009)
Newborn -0.046∗∗∗ -0.046∗∗∗ -0.045∗∗∗ -0.044∗∗∗
(0.013) (0.013) (0.013) (0.013)
Number of Children Under 5 -0.029∗∗∗ -0.029∗∗∗ -0.029∗∗∗ -0.029∗∗∗
(0.006) (0.006) (0.006) (0.006)
Looking for Work -0.002 -0.001 -0.003 -0.003
(0.012) (0.012) (0.012) (0.012)
Officer -0.014 -0.011 -0.009
(0.022) (0.022) (0.023)
Spouse- High School -0.014 -0.014 -0.012
(0.032) (0.032) (0.032)
Spouse- Some College 0.045 0.044 0.046
(0.032) (0.031) (0.032)
Spouse- College Degree 0.055 0.053 0.055
(0.034) (0.034) (0.034)
Spouse- Graduate Degree 0.066 0.064 0.068
(0.037) (0.037) (0.037)
Additional Controls
Year X X
State X
N 6938 6899 6899 6899 6899 6899
Adj. R2 0.001 0.011 0.020 0.025 0.028 0.028
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
170
Table B.33: Effects of Log of Average Published Price on Attendance of 18- to 40-year-old
Military Spouses Who Have an Associate’s Degree
(1) (2) (3) (4) (5) (6)
Two-Year Costs -0.025 -0.030 -0.031 -0.033 -0.033 0.216∗
(0.023) (0.023) (0.023) (0.023) (0.023) (0.096)
Four-Year Costs -0.007 -0.007 -0.008 -0.006 -0.026 -0.337∗∗
(0.030) (0.030) (0.030) (0.030) (0.031) (0.107)
Other Household Income -0.015 -0.018 -0.004 -0.007 0.008
(0.023) (0.023) (0.026) (0.026) (0.027)
Age -0.040 -0.022 -0.027 -0.027 -0.031
(0.023) (0.024) (0.024) (0.024) (0.025)
Age2 0.000 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000) (0.000)
Male 0.047 0.041 0.049 0.048 0.055
(0.050) (0.050) (0.051) (0.051) (0.052)
White 0.079∗ 0.074∗ 0.075∗ 0.073∗ 0.066
(0.034) (0.034) (0.035) (0.034) (0.035)
Unemployment Rate -0.010 -0.010 -0.009 -0.020 -0.020
(0.010) (0.010) (0.010) (0.013) (0.024)
Newborn -0.024 -0.018 -0.021 -0.019
(0.038) (0.038) (0.038) (0.039)
Number of Children Under 5 -0.047∗∗ -0.047∗∗ -0.044∗ -0.051∗∗
(0.018) (0.018) (0.018) (0.018)
Looking for Work 0.006 0.003 0.002 0.010
(0.036) (0.036) (0.036) (0.037)
Officer -0.025 -0.016 -0.022
(0.050) (0.050) (0.051)
Spouse- High School 0.031 0.026 0.052
(0.098) (0.098) (0.099)
Spouse- Some College 0.096 0.092 0.108
(0.096) (0.095) (0.097)
Spouse- College Degree 0.042 0.035 0.048
(0.097) (0.097) (0.098)
Spouse- Graduate Degree 0.066 0.061 0.079
(0.100) (0.100) (0.101)
Additional Controls
Year X X
State X
N 1350 1343 1343 1343 1343 1343
Adj. R2 0.001 0.040 0.046 0.048 0.057 0.060
Standard errors in parentheses
∗ p < 0.05, ∗∗ p < 0.01, ∗∗∗ p < 0.001
SOURCE: Author’s calculations of U.S. Department of Education, National Center for Education
Statistics, Integrated Postsecondary Education Data System (IPEDS), 2000-2008, U.S. Census Bureau,
Integrated Public Use Microdata Series (IPUMS), 2000 5% sample and 2000-2008 ACS samples and
U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index and Unemployment
Data.
171
C.7 Appendix C
Tables C.34 and C.35 provide summary statistics for the cohorts organized
by year recruited. These tables vary slightly from the summary statistics in the
main body of the report, which are broken out by tax status, because those who
were recruited in 2004 but waited until after January 5, 2005, to sign their initial
contract are in the “taxed but did not know at time of signing” group. These
tables show that the differences between 2004 and 2005 and 2006 are not as stark,
supporting the assumption that the industry stabilized by 2004. However, this does
not strengthen the assumption that there were no unobserved differences between
the baseline group and the group that was taxed but did not know about it when
they signed with the firm.
Table C.34: Summary Statistics for All Employees by Year Recruited
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 142.57 349.59 150.50 364.59 158.52 435.45
Age When Recruited by Firm 21.17 1.55 21.06 1.59 21.19 1.52
Fraction High School Seniors 0.15 0.36 0.15 0.36 0.17 0.37
Fraction From Junior College 0.15 0.36 0.17 0.37 0.11 0.31
Fraction College Fr. or So. 0.01 0.11 0.01 0.09 0.01 0.11
Fraction 4-Year College Junior 0.29 0.45 0.30 0.46 0.33 0.47
Fraction 4-Year College Senior 0.39 0.49 0.37 0.48 0.38 0.49
Fraction With a CSP 0.57 0.50 0.60 0.49 0.62 0.49
Observations 854 1653 4587
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample
includes all employees recruited domestically who signed with the firm between 2004 and 2011 with at least a high school degree.
(1) All monetary values are indexed to year 2000 dollars.
172
Table C.35: Summary Statistics for Employees With CSPs by Year Recruited
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 237.05 416.71 241.59 446.03 249.40 529.47
Age When Recruited by Firm 20.37 1.45 20.32 1.48 20.48 1.41
Fraction High School Senior 0.26 0.44 0.24 0.43 0.26 0.44
Fraction From Junior College 0.22 0.41 0.23 0.42 0.16 0.36
Fraction College Fr. or So. 0.02 0.13 0.01 0.11 0.02 0.14
Fraction 4-Year College Juniors 0.44 0.50 0.45 0.50 0.49 0.50
Fraction 4-Year College Seniors 0.07 0.26 0.07 0.26 0.07 0.26
Number of Semesters in CSP 4.49 2.38 4.45 2.34 4.29 2.44
CSP Funds Per Semester1 6753.61 4303.41 6757.84 4168.21 8743.17 5384.64
Total Value of CSP1 30175.34 27682.77 29219.97 24029.98 37380.44 32813.00
Proportion of Signing Bonus 0.29 0.22 0.26 0.19 0.29 0.20
Taken as CSP Funds
Observations 485 993 2843
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample includes all
employees recruited domestically, who signed with the firm between 2004 and 2011, with at least a high school degree, and had a CSP included in their
initial contract. The total value of a CSP is calculated by the amount of money available per semester, multiplied by the number of semesters specified in
the contract. The proportion of the signing bonus taken is CSP funds is equal to the total value of CSP funds divided by the sum of CSP funds and the
cash signing bonus.
(1) All monetary values are indexed to year 2000 dollars.
Tables C.36 through C.43 provide summary statistics for the cohorts organized
by rank group and tax status.
Table C.36: Summary Statistics for Employees in Rank Group 1 by Tax Status
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 875.59 556.88 957.55 741.50 1044.66 1006.00
Age When Recruited by Firm 20.23 1.40 20.15 1.49 19.94 1.50
Fraction High School Seniors 0.36 0.48 0.40 0.49 0.46 0.50
Fraction From Junior College 0.05 0.22 0.03 0.18 0.03 0.16
Fraction College Fr. or So. 0.02 0.15 0.00 0.00 0.02 0.13
Fraction 4-Year College Junior 0.54 0.50 0.53 0.50 0.47 0.50
Fraction 4-Year College Senior 0.02 0.15 0.03 0.17 0.03 0.17
Fraction With a CSP 0.99 0.11 0.97 0.18 0.99 0.12
Observations 83 172 420
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample
includes all employees recruited domestically who signed with the firm between 2004 and 2011 with at least a high school degree who were ranked
90 of higher in the year they were recruited.
(1) All monetary values are indexed to year 2000 dollars.
173
Table C.37: Summary Statistics for Employees in Rank Group 2 by Tax Status
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 147.71 182.15 143.55 121.91 192.57 215.43
Age When Recruited by Firm 20.76 1.59 20.75 1.57 20.62 1.44
Fraction High School Seniors 0.27 0.45 0.24 0.43 0.24 0.43
Fraction From Junior College 0.10 0.30 0.09 0.29 0.11 0.31
Fraction College Fr. or So. 0.01 0.10 0.01 0.11 0.02 0.13
Fraction 4-Year College Junior 0.39 0.49 0.42 0.49 0.48 0.50
Fraction 4-Year College Senior 0.23 0.42 0.23 0.42 0.15 0.36
Fraction With a CSP 0.81 0.39 0.82 0.38 0.89 0.31
Observations 185 379 907
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample
includes all employees recruited domestically who signed with the firm between 2004 and 2011 with at least a high school degree who were ranked
between 90 and 300 in the year they were recruited.
(1) All monetary values are indexed to year 2000 dollars.
Table C.38: Summary Statistics for Employees in Rank Group 3 by Tax Status
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 40.90 148.09 49.52 110.34 62.85 115.68
Age When Recruited by Firm 21.41 1.39 21.10 1.52 21.15 1.44
Fraction High School Seniors 0.12 0.33 0.13 0.34 0.16 0.37
Fraction From Junior College 0.10 0.31 0.18 0.39 0.12 0.33
Fraction College Fr. or So. 0.01 0.12 0.01 0.09 0.01 0.10
Fraction 4-Year College Junior 0.34 0.47 0.30 0.46 0.35 0.48
Fraction 4-Year College Senior 0.42 0.50 0.38 0.49 0.36 0.48
Fraction With a CSP 0.55 0.50 0.60 0.49 0.68 0.47
Observations 220 441 1138
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample
includes all employees recruited domestically who signed with the firm between 2004 and 2011 with at least a high school degree who were
ranked between 301 and 600 in the year they were recruited.
(1) All monetary values are indexed to year 2000 dollars.
Table C.39: Summary Statistics for Employees in Rank Group 4 by Tax Status
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 10.22 26.09 25.18 56.46 19.57 52.99
Age When Recruited by Firm 21.77 1.30 21.29 1.61 21.71 1.36
Fraction High School Seniors 0.08 0.27 0.04 0.21 0.08 0.27
Fraction From Junior College 0.11 0.32 0.28 0.45 0.12 0.33
Fraction College Fr. or So. 0.01 0.11 0.01 0.09 0.01 0.11
Fraction 4-Year College Junior 0.17 0.37 0.16 0.37 0.22 0.42
Fraction 4-Year College Senior 0.63 0.48 0.50 0.50 0.56 0.50
Fraction With a CSP 0.28 0.45 0.41 0.49 0.40 0.49
Observations 310 722 2117
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample
includes all employees recruited domestically who signed with the firm between 2004 and 2011 with at least a high school degree who were
ranked below 600 in the year they were recruited.
(1) All monetary values are indexed to year 2000 dollars.
174
Table C.40: Summary Statistics for Employees With CSPs in Rank Group 1 by Tax Status
Untaxed Unknowingly Taxed Knowingly Taxed
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 870.70 558.51 961.26 733.05 1045.57 1007.75
Age When Recruited by Firm 20.25 1.39 20.13 1.47 19.95 1.50
Fraction High School Seniors 0.35 0.48 0.40 0.49 0.45 0.50
Fraction From Junior College 0.05 0.22 0.04 0.19 0.03 0.16
Fraction College Fr. or So. 0.02 0.16 0.00 0.00 0.02 0.13
Fraction 4-Year College Juniors 0.55 0.50 0.54 0.50 0.48 0.50
Fraction 4-Year College Seniors 0.02 0.16 0.02 0.13 0.03 0.16
Number of Semesters in CSP 5.11 2.36 5.17 2.44 5.28 2.60
CSP Funds Per Semester1 $9,123.99 $4,448.86 $9,513.52 $4,776.41 $12,222.25 $6,415.07
Total Value of CSP1 $45,277.29 $3,0852.68 $47,762.34 $32,276.03 $63,889.84 $44,033.32
Proportion of Signing Bonus 0.06 0.03 0.06 0.05 0.08 0.06
Taken as CSP Funds
Observations 82 166 414
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample includes all employees
recruited domestically, who signed with the firm between 2004 and 2011, with at least a high school degree, who were ranked in the top 90 the year they were
recruited, and had a CSP included in their initial contract. The total value of a CSP is calculated by the amount of money available per semester, multiplied by
the number of semesters specified in the contract. The proportion of the signing bonus taken in CSP funds is equal to the total value of CSP funds divided by
the sum of CSP funds and the cash signing bonus.
(1) All monetary values are indexed to year 2000 dollars.
Table C.41: Summary Statistics for Employees With CSPs in Rank Group 2 by Tax Status
Untaxed Unknowingly Taxed Knowingly Taxed
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 172.39 192.82 162.84 123.69 208.91 222.05
Age When Recruited by Firm 20.40 1.53 20.48 1.49 20.42 1.37
Fraction High School Seniors 0.33 0.47 0.28 0.45 0.27 0.45
Fraction From Junior College 0.11 0.32 0.09 0.29 0.12 0.33
Fraction College Fr. or So. 0.01 0.12 0.02 0.13 0.02 0.14
Fraction 4-Year College Juniors 0.47 0.50 0.50 0.50 0.53 0.50
Fraction 4-Year College Seniors 0.07 0.26 0.11 0.32 0.06 0.24
Number of Semesters in CSP 4.70 2.49 4.45 2.46 4.43 2.38
CSP Funds Per Semester1 $7,459.03 $4,933.12 $7,359.13 $4,248.49 $9,782.07 $5,300.94
Total Value of CSP1 $33,275.96 $31,200.40 $31,431.50 $23,447.61 $41,956.58 $31,487.63
Proportion of Signing Bonus 0.19 0.11 0.18 0.10 0.20 0.10
Taken as CSP Funds
Observations 150 311 807
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample includes all employees
recruited domestically, who signed with the firm between 2004 and 2011, with at least a high school degree, were ranked between 91 and 300 in the year they were
recruited, and had a CSP included in their initial contract. The total value of a CSP is calculated by the amount of money available per semester, multiplied by
the number of semesters specified in the contract. The proportion of the signing bonus taken in CSP funds is equal to the total value of CSP funds divided by
the sum of CSP funds and the cash signing bonus.
(1) All monetary values are indexed to year 2000 dollars.
175
Table C.42: Summary Statistics for Employees With CSPs in Rank Group 3 by Tax Status
Untaxed Unknowingly Taxed Knowingly Taxed
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 70.33 194.17 76.52 134.98 89.27 132.40
Age When Recruited by Firm 20.74 1.36 20.43 1.47 20.61 1.38
Fraction High School Seniors 0.20 0.41 0.20 0.40 0.23 0.42
Fraction From Junior College 0.14 0.35 0.27 0.44 0.16 0.37
Fraction College Fr. or So. 0.01 0.09 0.01 0.09 0.01 0.12
Fraction 4-Year College Juniors 0.53 0.50 0.44 0.50 0.50 0.50
Fraction 4-Year College Seniors 0.11 0.32 0.08 0.27 0.09 0.29
Number of Semesters in CSP 3.83 2.35 4.21 2.33 4.05 2.39
CSP Funds Per Semester1 $6,065.14 $3,553.98 $5,908.18 $3,547.78 $7,732.03 $4,450.30
Total Value of CSP1 $24,344.02 $25,834.78 $23,118.39 $18,340.38 $30,472.09 $25,217.70
Proportion of Signing Bonus 0.37 0.17 0.33 0.17 0.32 0.16
Taken as CSP Funds
Observations 122 265 770
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample includes all employees
recruited domestically, who signed with the firm between 2004 and 2011, with at least a high school degree, were ranked between 301 and 600 in the year they were
recruited, and had a CSP included in their initial contract. The total value of a CSP is calculated by the amount of money available per semester, multiplied by
the number of semesters specified in the contract. The proportion of the signing bonus taken in CSP funds is equal to the total value of CSP funds divided by the
sum of CSP funds and the cash signing bonus.
(1) All monetary values are indexed to year 2000 dollars.
Table C.43: Summary Statistics for Employees With CSPs in Rank Group 4 by Tax Status
Untaxed Unknowingly Taxed Knowingly Taxed
2004 2005 - 2006 2007 - 2011
Mean Standard deviation Mean Standard deviation Mean Standard deviation
Signing Bonus in $1,000s1 30.09 41.63 55.87 77.43 44.93 76.36
Age When Recruited by Firm 20.44 1.42 20.01 1.42 20.66 1.37
Fraction High School Seniors 0.24 0.43 0.10 0.31 0.19 0.39
Fraction From Junior College 0.29 0.46 0.55 0.50 0.24 0.43
Fraction College Fr. or So. 0.03 0.18 0.02 0.13 0.02 0.16
Fraction 4-Year College Juniors 0.36 0.48 0.29 0.45 0.46 0.50
Fraction 4-Year College Seniors 0.08 0.27 0.04 0.20 0.08 0.28
Number of Semesters in CSP 4.32 2.47 4.32 1.95 3.90 2.31
CSP Funds Per Semester1 $5,172.48 $3,314.48 $5,090.77 $2,978.06 $6,981.67 $4,631.93
Total Value of CSP1 $22,245.90 $18,480.95 $21,167.85 $14,509.02 $26,386.99 $24,846.35
Proportion of Signing Bonus 0.54 0.21 0.39 0.20 0.46 0.21
Taken as CSP Funds
Observations 87 298 849
SOURCE: Author’s calculations of firm data and the U.S. Department of Labor, Bureau of Labor Statistics, Consumer Price Index. Sample includes all employees
recruited domestically, who signed with the firm between 2004 and 2011, with at least a high school degree, were ranked below the top 600 in the year they were
recruited, and had a CSP included in their initial contract. The total value of a CSP is calculated by the amount of money available per semester, multiplied by
the number of semesters specified in the contract. The proportion of the signing bonus taken in CSP funds is equal to the total value of CSP funds divided by
the sum of CSP funds and the cash signing bonus.
(1) All monetary values are indexed to year 2000 dollars.
176
The panels in Figure C.6 show changes in the share of initial contracts that
include CSP funds and the characteristics of the CSP funds over time. There does
not appear to be much action on the year 2005 line break, which is included to check
if there was any leakage of the information of the loss of the tax break before the
announcement in 2007. However, the line break in 2007 shows a dramatic increase
in the share of contracts with a CSP. This initial uptick dissipates over time. The
change in the rate of growth in CSP funds per semester and the total value of CSP
funds in Figures C.6b and C.6c, respectively, do appear distinct and sustained.
177
Figure C.6: Trends in CSP Contracts With Break Lines in 2005 and 2007
(a) Trends in Share With
CSP Funds
42
44
46
48
50
52
Pe
rc
en
ta
ge
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
20
09
20
10
20
11
Year
Share of All Recruits with CSPs by Year
(b) Trends in CSP Funds
per Semester
60
00
70
00
80
00
90
00
10
00
0
In
de
xe
d 
to
 Y
ea
r 2
00
0 
$
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
20
09
20
10
20
11
Year
Average CSP Funds Per Semester by Year
(c) Trends in CSP Total
Value
25
00
0
30
00
0
35
00
0
40
00
0
In
de
xe
d 
to
 Y
ea
r 2
00
0 
$
20
00
20
01
20
02
20
03
20
04
20
05
20
06
20
07
20
08
20
09
20
10
20
11
Year
Average Total CSP Funds Per CSP Contract by Year
Figure C.7: Histograms of Alternative Cut-offs
(a) Age CSP Funds Were First Used
0
5
10
15
20
Pe
rc
en
t
18 20 22 24 26 28 30 32 34
Age
(b) Years After Signing CSP First Used
0
5
10
15
20
Pe
rc
en
t
0 2 4 6 8 10 12 14 16
Difference in Years
Figures C.7a and C.7b show the lack of a clear break point if the age when
an employee first uses their CSP or the time from signing an initial contract is
considered as a cut-off for the outcome of interest when examining CSP usage.
178
Bibliography
[1] Abraham, K., & Clark, M. (2006). Financial aid and students’ college de-
cisions: Evidence from the District of Columbia Tuition Assistance Grant
Program. Journal of Human Resources, 41(3), 578–610.
[2] Akers, E. J., & Chingos, M. M. (2014). Are college students borrowing blindly?
Brown Center on Education Policy at Brookings.
[3] Angrist, J. D. (1993). The effect of veterans benefits on education and earnings.
Industrial and Labor Relations Review, 46(4), 637–52.
[4] Angrist, J. D., & Chen S. H., (2011) Schooling and the Vietnam-Era GI Bill:
Evidence from the Draft Lottery, American Economic Journal: Applied Eco-
nomics 3 : 96–119.
[5] Angrist, J., Lang, D., & Oreopoulos, P. (2009). Incentives and services for
college achievement: Evidence from a randomized trial. American Economic
Journal: Applied Economics, 1(1), 136–163.
[6] Angrist, J., Oreopoulos, P., & Williams, T. (2014). When opportunity knocks,
who answers? New evidence on college achievement awards. Journal of Human
Resources, 49(3), 572–610.
[7] Avery, C., & Kane, T. J. (2004). Student perceptions of college opportuni-
ties: The Boston COACH Program. In C. M. Hoxby (Ed.), College choices:
The economics of where to go, when to go, and how to pay for it (239–302).
Chicago: University of Chicago Press.
[8] Avery, C., & Hoxby, C. M. (2004). Do and should financial aid packages affect
students’ college choices? In C. M. Hoxby (Ed.), College choices: The eco-
nomics of where to go, when to go, and how to pay for it (239–302). Chicago:
University of Chicago Press.
179
[9] Autor, D. H. (2014). Skills, education, and the rise of earnings inequality
among the other 99 percent. Science, 344 (6186), 843–851.
[10] Autor, D. H., Katz, L. F., & Kearney, M. S. (2008). Trends in U.S. wage
inequality: Revising the revisionists. The Review of Economics and Statistics,
90(2), 300–323.
[11] Bartik, T. J., Hershbein, B., & Lachowska, M. (2015). The effects of the
Kalamazoo Promise Scholarship on college enrollment, persistence, and com-
pletion (Upjohn Institute Working Paper 15-229). Kalamazoo, MI: W.E. Up-
john Institute for Employment Research. Retrieved from: http://research.
upjohn.org/up_workingpapers/229
[12] Baum, S. (2013). The Evolution of Student Debt in the U.S.: An Overview.
The Urban Institute, George Washington University Graduate School of Edu-
cation and Human Development. Retrieved from: http://www.upjohn.org/
stuloanconf/Baum.pdf
[13] Baum, S., & Schwartz, S. (2014). Student aid, student behavior, and educa-
tional attainment (Discussion Paper ).
[14] Baum, S., & Steele, P. (2010). Who borrows most? Bache-
lor’s degree recipients with high levels of student debt. Trends in
Higher Education Series. College Board Advocacy & Policy Center. Re-
trieved from: https://trends.collegeboard.org/sites/default/files/
trends-2010-who-borrows-most-brief.pdf
[15] Bean, J., and Metzner, M. (1985). A Conceptual Model of Nontraditional Un-
dergraduate Student Attrition. Review of Educational Research, 55(4), 485–
540.
[16] Becker, G. S. (1962). Investment in human capital: A theoretical analysis.
Journal of Political Economy, 70( 5), Part 2: Investment in Human Beings,
9–49.
[17] Becker, G. S. (1975). Investment in human capital: Effects on earnings. In G.
S. Becker, Human capital: A theoretical and empirical analysis, with special
reference to education, Second Edition, NBER 1975 (13–44). Retrieved from:
http://www.nber.org/chapters/c3733
[18] Belley, P., & Lochner, L. (2007). The changing role of family income and
ability in determining educational achievement (NBER Working Paper 13527).
National Bureau of Economic Research.
180
[19] Bennett, D. L., Lucchesi, A. R., & Vedder, R. (2010). For-profit
higher education: Growth, innovation, and regulation. Washing-
ton, DC: Center for College Affordability and Productivity. Retrieved
from: http://heartland.org/sites/all/modules/custom/heartland_
migration/files/pdfs/29010.pdf
[20] Bettinger, E. (2004). How financial aid affects persistence. In C. Hoxby (Ed.),
College choices: The economics of where to go, when to go, and how to pay
for it. Chicago: University of Chicago Press.
[21] Bettinger, E., Long, B. T., Oreopoulos, P., & Sanbonmatsu, L. (2012) The
role of application assistance and information in college decisions: Results
from the H&R Block FAFSA experiment. Quarterly Journal of Economics,
127(3).
[22] Bound, J., Hershbein, B., & Long, B. T. (2009). Playing the admissions game:
Student reactions to increasing college competition. Journal of Economic Per-
spectives, American Economic Association, 23(4), 119–46.
[23] Bound, J., Lovenheim, M. F., & Turner, S. (2010). Why have college comple-
tion rates declined? An analysis of changing student preparation and collegiate
resources. American Economic Journal: Applied Economics, 2(3), 129–157.
[24] Bound, J., & Turner, S. (2002). Going to war and going to college: Did the
World War II and the G.I. Bill increase educational attainment for returning
veterans? Journal of Labor Economics, 20(4), 784–815.
[25] Bound, J., & Turner, S. (2003). Closing the gap or widening the divide: The
effects of the G.I. Bill and World War II on the educational outcomes of black
Americans. The Journal of Economic History, 63(1), 145–177.
[26] Bound, J., & Turner, S. (2007). Cohort crowding: How resources affect colle-
giate attainment. Journal of Public Economics, 91(5–6), 877–899.
[27] Brown, M. (2013). Student Debt Overview. Federal Reserve Bank of
New York, Retrieved from: http://www.newyorkfed.org/regional/Brown_
presentation_GWU_2013Q2.pdf
[28] Bulman, G. B. and Hoxby, C. M. (2015). The Returns to the Federal Tax Cred-
its for Higher Education (NBER Working Paper 20833). National Bureau of
Economic Research. Retrieved from: http://www.nber.org/papers/w20833
[29] Burman, L. E., Maag, E., Orszag, P., Rohaly, J., & O’Hare, J. (2005). The
distributional consequences of federal assistance for higher education: The
181
intersection of tax and spending programs (Discussion Paper 26). Washington,
DC: The UrbanBrookings Tax Policy Center.
[30] Cameron, S. V., & Heckman, J. J. (1998). Life cycle schooling and dynamic
selection bias: Models and evidence for five cohorts of American males. Journal
of Political Economy, 106(2), 262–333.
[31] Cameron, S. V., & Taber, C. (2004). Estimation of educational borrowing
constraints using returns to schooling. Journal of Political Economy, 112(1),
132–182.
[32] Card, D. (1995). Earnings, schooling, and ability revisited. In S. Polachek
(Ed.), Research in Labor Economics, 14. Greenwich Connecticut: JAI Press.
[33] Card, D. (1993). Using geographic variation in college proximity to estimate
the return to schooling (NBER Working Paper 4483). National Bureau of
Economic Research. Retrieved from: http://www.nber.org/papers/w4483
[34] Card, D., & Lemieux, T. (2001). Going to college to avoid the draft: The un-
intended legacy of the Vietnam War. The American Economic Review, 91(2),
Papers and Proceedings of the Hundred Thirteenth Annual Meeting of the
American Economic Association. 97–102.
[35] Carneiro, P., & Heckman, J. J. (2002, October). The evidence on credit con-
straints in post-secondary schooling. Economic Journal, 112(482), 705–734.
[36] Carrell, S. E., & Sacerdote, B. (2013). Why do college going interventions
work? (NBER Working Paper 19031). National Bureau of Economic Research.
Retrieved from: http://www.nber.org/papers/w19031
[37] Cellini, S. R., & Goldin, C. (2014). Does federal student aid raise tuition?
New evidence on for-profit colleges. American Economic Journal: Economic
Policy, 6(4), 174–206.
[38] Chin, A., & Juhn, C. (2010). Does reducing college costs improve educational
outcomes for undocumented immigrants? Evidence from state laws permit-
ting undocumented immigrants to pay in-state tuition at state colleges and
universities (NBER Working Paper 15932). National Bureau of Economic Re-
search.
[39] Conger, D., & Turner, L. J. (2015). The impact of tuition increases on undocu-
mented college students’ attainment (NBER Working Paper 21135). National
Bureau of Economic Research.
182
[40] College Board. (2014). Trends in student aid 2014. Retrieved
from: http://trends.collegeboard.org/sites/default/files/
2014-trends-student-aid-final-web.pdf
[41] Cornwell, C., Mustard, D. B., & Sridhar, D. J. (2006). The enrollment effects
of merit-based financial aid: Evidence from Georgia’s HOPE Program. Journal
of Labor Economics, 24(4), 761–786.
[42] Dai, E. (2013). Student Loan Delinquencies Surge. Federal Reserve Bank of
St. Loius, Retrieved from: https://www.stlouisfed.org/publications/
inside-the-vault/spring-2013/student-loan-delinquencies-surge
[43] Dale, S. B., & Krueger, A. B. (2002). Estimating the payoff of attending a more
selective college: An application of selection on observables and unobservables.
Quarterly Journal of Economics, 107(4), 1491–1527.
[44] Deming, D., & Dynarski, S. (2010). Into college, out of poverty? Policies to
increase the postsecondary attainment of the poor (NBER Working Paper
15387). National Bureau of Economic Research.
[45] Dresch, S. P. (1983). College enrollment. Proceedings of the Academy of Po-
litical Science, 35(2), The Crisis in Higher Education, 108–118.
[46] Dynarski, S. M. (2003). Does aid matter? Measuring the effect of student
aid on college attendance and completion. American Economic Review, 93(1),
279–288.
[47] Dynarski, S. M. (2004a). The new merit aid. In C. M. Hoxby (Ed.), College
choices: The economics of which college, when college, and how to pay for it
(63–97). Chicago: University of Chicago Press.
[48] Dynarski, S. M. (2004b). Who benefits from the college saving incentives?
Income, educational expectations and the value of the 529 and Coverdell.
National Tax Journal, LVII(2), Part 2.
[49] Dynarski, S. M. (2008). Building the stock of college-educated labor. Journal
of Human Resources, 43(3), 576–610.
[50] Dynarski, S. M. (2014). The (fixable) problem with Pay It Forward. The
Hamilton Project.
[51] Dynarski, S. M. (2015). We’re frighteningly in the dark about
student debt. The New York Times, March 20, 2015. Re-
183
trieved from: http://www.nytimes.com/2015/03/22/upshot/
were-frighteningly-in-the-dark-about-student-debt.html?_r=2
[52] Dynarski, S. M., & Kreisman, D. (2013). Loans for educational opportunity:
Making borrowing work for today’s students (Discussion Paper 2013-05). The
Hamilton Project.
[53] Dynarski, S. M., & Scott-Clayton, J. E. (2006). The cost of complexity in fed-
eral student aid: Lessons from optimal tax theory and behavioral economics.
National Tax Journal, 59(2), 319–356.
[54] Dynarski, S. M., & Wiederspan, M. (2012). Student aid simplification: Look-
ing back and looking ahead. National Tax Journal, 65(1), 211–234.
[55] Ehrenberg, R. G. (2012). American higher education in transition. Journal of
Economic Perspectives, 26(1), 193–216.
[56] Field, E. (2009). Educational debt burden and career choice: Evidence from
a financial aid experiment at NYU Law School. American Economic Journal:
Applied Economics, 1(1), 1–21.
[57] Goldin, C., & Katz, L. F. (2007). The race between education and technology:
The evolution of U.S. educational wage differentials, 1890 to 2005 (NBER
Working Paper 12984). National Bureau of Economic Research.
[58] Hansen, W. L. (1983). The impact of student financial aid on access. In J.
Froomkin (Ed.), The crisis in higher education. New York: Academy of Polit-
ical Science.
[59] Heckman, J. J., Lochner, L. J., & Todd, P. E. (2006). Earnings equations and
rates of return: The Mincer equation and beyond. In E. A. Hanushek & F.
Welch (Eds.), Handbook of the Economics of Education (307–458). Amster-
dam: Elsevier.
[60] Heller, D. E. (1997). Student price response in higher education: An update
to Leslie and Brinkman. Journal of Higher Education, 68(6), 624–659.
[61] Horn, L. (1996). Nontraditional Undergraduates, Trends in Enrollment From
1986 to 1992 and Persistence and Attainment Among 1989-90 Beginning Post-
secondary Students (NCES 97-578). U.S. Department of Education, NCES.
Washington, DC: U.S. Government Printing Office.
[62] Hoxby, C. M. (2009). The changing selectivity of American colleges. Journal
of Economic Perspectives, 23(4), 95–118.
184
[63] Hoxby, C. M., & Long, B. T. (1999). Explaining rising income and wage
inequality among the college-educated (NBER Working Paper 6873). National
Bureau of Economic Research.
[64] Hoxby, C. M., & Turner, S. (2013). Expanding college opportunities for high-
achieving, low income students. Stanford Institute for Economic Policy Re-
search.
[65] Hoxby, C. M., & Turner, S. (2015). What high-achieving low-income students
know about college (NBER Working Paper 20861). Retrieved from: http:
//www.nber.org/papers/w20861
[66] Jepsen, C., & Montgomery, M. (2009). Miles to Go Before I Learn: The Effect
of Travel Distance on the Mature Person’s Choice of a Community College.
Journal of Urban Economics 65 (1), 6473.
[67] Kane, T. J. (1994). College entry by blacks since 1970: The role of college cost,
family background and the returns to education. Journal of Political Economy,
102(5), 878–911.
[68] Kane, T. J. (1995). Rising public college tuition and college entry: How well
do public subsidies promote access to college? (NBER Working Paper 5164).
National Bureau of Economic Research.
[69] Kane, T. J. (1999). The price of admission: Rethinking how Americans pay
for college. Washington, DC: Brookings.
[70] Kane, T. J. (2003). A quasi-experimental estimate of the impact of finan-
cial aid on college-going (NBER Working Paper 9703). National Bureau of
Economic Research.
[71] Kaushal, N. (2005). New immigrants’ location choices: Magnets without wel-
fare. Journal of Labor Economics, 23(1), 59–80.
[72] Kaushal, N. (2008). In-state tuition for the undocumented: Education effects
on Mexican young adults. Journal of Policy Analysis and Management, 27(4),
771–92.
[73] LaLumia, S. (2012). Tax Preferences for Higher Education and Adult College
Enrollment National Tax Journal, March 2012, 65 (1), 59–90
[74] Leslie, L., & Brinkman, P. (1988). The economic value of higher education.
New York: Macmillan.
185
[75] Lindley, J., & Machin, S. (2011). Rising wage inequality and postgraduate
education (CEP Discussion Paper 1075).
[76] Lleras-Muney, A. (2010). “The needs of the Army: using compulsory relo-
cation in the military to estimate the effect of environmental pollutants on
children’s health,” Journal of Human Resources, 35(3):549–590.
[77] Long, B. T. (2004). The Impact of Federal Tax Credits for Higher Education
Expenses. In C. M. Hoxby (Ed.), College choices: The economics of where to
go, when to go, and how to pay for it (10168). Chicago: University of Chicago
Press.
[78] Manski, C. F. (1993). Income and higher education. Focus, 14(3), 14–19.
[79] Marx, B. M., & Turner, L. J. (2015). Borrowing trouble? Student loans, the
cost of borrowing, and implications for the effectiveness of need-based grant
aid (NBER Working Paper 20850).
[80] Pallais, A. (2015). Small differences that matter: Mistakes in applying to
college. Journal of Labor Economics, 33(2):493–520.
[81] Patten, Eileen & Parker, Kim (2011). Women in the U.S. Military:
Growing Share, Distinctive Profile. PEW SOCIAL & DEMOGRAPHIC
TRENDS. Retrieved from: http://www.pewsocialtrends.org/files/
2011/12/women-in-the-military.pdf
[82] Schuetze, H. G., and M. Slowey. 2002. Participation and exclusion: A com-
parative analysis of nontraditional students and lifelong learners in higher
education. Higher Education 44: 309–327.
[83] Scott-Clayton, J. (2013). Information constraints and financial aid policy. In
D. E. Heller & C. Callender (Eds.), Student financing of higher education: A
comparative perspective.
[84] Seftor, N., & Turner, S. (2002). Back to school: Federal student aid policy
and adult college enrollment. Journal of Human Resources, 37(2), 336–352.
[85] Shea, J. (2000). Does parents’ money matter? Journal of Public Economics,
77, 155–184.
[86] Singell, L. D., Jr., & Stone, J. A. (2007). For whom the Pell tolls: The response
of university tuition to federal grants-in-aid. Economics of Education Review,
26(3), 285–295.
186
[87] Sjoquist, D. L., & Winters, J. V. (2012a). Building the stock of college-
educated labor revisited. Journal of Human Resources, 47(1), 270–85.
[88] Sjoquist, D. L., & Winters, J. V. (2012b). State merit-based financial aid pro-
grams and college attainment (IZA Discussion Paper 6801). Bonn, Germany:
Institute for the Study of Labor.
[89] Stanley, M. (2003). College education and the mid-century G.I. Bills. The
Quarterly Journal of Economics, 118(2), 671–708.
[90] Tax Policy Center. (2007). Effective marginal tax rates, by earnings level.
Urban Institute and Brookings Institution. Retrieved from: http://www.
taxpolicycenter.org/taxfacts/displayafact.cfm?Docid=453
[91] Turner, N. (2011a). The Effect of Tax-based Federal Student Aid on College
Enrollment. National Tax Journal, 64 (3), 839–862
[92] Turner, N. (2011b). Why Don’t Taxpayers Maximize their Tax-Based Student
Aid? Salience and Inertia in Program Selection, The B.E. Journal of Economic
Analysis & Policy: Vol. 11: Iss.1 (Contributions), Article 75.
[93] Turner, N. (2012). Who benefits from student aid? The economic incidence of
tax-based federal student aid. Economics of Education Review, 31, 463–481.
[94] Turner, L. J. (2013). Rethinking institutional aid: Implications for affordabil-
ity, access, and the effectiveness of federal student aid (AEI Memo). Prepared
for the American Enterprise Institute Conference, The Trillion-Dollar Ques-
tion: Reinventing Student Financial Aid for the 21st Century.
[95] Turner, L. J. (2014). The road to Pell is paved with good intentions: The
economic incidence of federal student grant aid (Working Paper).
[96] U.S. Department of Defense, In-State Tuition State-By-State Summary.
Retrieved from: https://www.hrc.army.mil/SITE/Education/InState/
FAQs.htm
[97] U.S. Department of Education, National Center for Education Statistics.
(2002). Non-traditional undergraduates (NCES 2002-012), by S. Choy. Wash-
ington, DC.
[98] U.S. Department of Education, National Center for Education Statistics.
(2010). Web tables: Profile of undergraduate students: 2007-08. Retrieved
from: https://nces.ed.gov/pubs2010/2010205.pdf
187
[99] U.S. Department of Education, National Center for Education Statistics.
(2013). Digest of Education Statistics, 2012 (NCES 2014-015).
[100] U.S. Department of Education, National Center for Education Statistics.
(2015) The Condition of Education 2015 (NCES 2015-144), by Kena, G.,
Musu-Gillette, L., Robinson, J., Wang, X., Rathbun, A., Zhang, J., Wilkinson-
Flicker, S., Barmer, A., and Dunlop Velez, E. Washington, DC. Retrieved
November 2015 from: http://nces.ed.gov/pubsearch specifically http:
//nces.ed.gov/pubs2015/2015144.pdf
[101] Vedder, R. (2004). Going broke by degree: Why college costs too much. Wash-
ington, DC: AEI Publications.
[102] Webber, D. A., & Ehrenberg, R. G. (2010). Do expenditures other than in-
structional expenditures affect graduation and persistence rates in American
higher education? Economics of Education Review, 29, 947–958.
[103] Weiss, M. J., Mayer, A., Cullinan, D., Ratledge, A., Sommo, C., & Diamond,
J. (2014). A random assignment evaluation of learning communities at Kings-
borough Community College: Seven years later. New York: MDRC.
[104] Zavodny, M. (1997). Welfare and the Locational Choices of New Immigrants.
Economic and Financial Policy Review, Federal Reserve Bank of Dallas, (Q
II), 2–10.
188