R ajamani, et al. 1 Assessing the impact of urban form measures in nonwork trip mode choice after controlling for demographic and level-of-service effects Jayanthi Rajamani jrajamani@mail.utexas.edu Chandra R. Bhat bhat@mail.utexas.edu Susan Handy handy@mail.utexas.edu Gerritt Knaap gknaap@ursp.umd.edu and Yan Song yansong@uiuc.edu TRB 2003: Submitted for Presentation and Publication Submission Date: August 1, 2002 Word Count: 6757 words+3 tables R ajamani, et al. 2 ABSTRACT The relationship between travel behavior and the local built environment continues to be a contentious issue, despite several research efforts in the area. The current paper investigates the significance and explanatory power of a variety of urban form measures on nonwork activity travel mode choice. The travel data used for analysis is the 1995 Portland Metropolitan Activity Survey conducted by Portland Metro. The database on the local built environment was developed by Song (2002) and includes a more extensive set of variables than previous studies that have examined the relationship between travel behavior and the local built environment using the Portland data. A multinomial logit mode choice model results indicate that higher residential densities and mixed-uses promote walking behavior for nonwork activities. R ajamani, et al. 3 1. INTRODUCTION The increasingly adverse effects of automobile use on traffic congestion and air pollution, combined with the limited financial ability to continually invest in transportation infrastructure, has led to the consideration of non-transportation strategies for managing and influencing travel demand. The paradigm shift toward non-transportation strategies to manage travel demand gained momentum, in particular, with the advent of the New Urbanism movement in the early 1990s (Duany and Plater-Zyberk, 1991). The New Urbanism movement is a manifestation of environmental determinism, wherein the urban planner’s role is to engineer and encourage socially-vibrant and environmentally-friendly modes of transportation such as walking and bicycling. The consideration of non-transportation strategies to manage demand, spurred by the New Urbanism movement, has led to a burgeoning literature at the interface of land use and transportation. In particular, there have been several studies in the past decade focusing on the influence of urban form and the built environment on travel behavior. While these studies have contributed substantially to our understanding of the interactions between urban form and travel behavior, there is still considerable research to be done in this area. The next few sections discuss some of the issues characterizing earlier studies, and position the current study in the broader context of the earlier studies. 1.1 Work versus Nonwork Travel Mode Choice The association between aspects of the built environment at the employment site or residence and workers’ commute choices has been studied by many researchers (for example, see Cervero, 1989; Cambridge Systematics, 1994; Kockelman, 1995; Cervero, 1996; Messenger and Ewing, 1996; Cervero and Wu, 1997; Levinson and Kumar, 1997). In contrast to the focus on the effect of the built environment on commute travel, there has been relatively lesser attention on the influence of the built environment on nonwork travel (for example, see Handy, 1992; Bhat et al., 1999; Boarnet and Sarmiento, 1998; Boarnet and Crane, 2001; Reilly, 2002). Nonwork trips constitute about three-quarters of urban trips and represent an increasingly large proportion of peak period trips (FHWA, 1995). Thus, it is important to analyze the impact of land use on nonwork travel. This study contributes toward this objective by examining the impact of the built environment on nonwork mode choice. The focus on the modal dimension of nonwork trips is motivated by the observation that the few earlier studies examining land use impact on nonwork travel have not focused on this dimension (for example, Handy, 1992 examines land use impacts on shopping trip frequency; Bhat et al., 1999 study land use and other variable impacts on shopping trips; Boarnet and Sarmiento, 1998 and Boarnet and Crane, 2001 examine land use impacts on nonwork automobile trips). Reilly’s (2002) study is the closest to the current research paper, although the empirical settings are different between his paper and ours. In addition, Reilly (2002) uses qualitative measures such as a Transit Access Index and proxies for streetscape in his San Francisco study, while the current paper attempts to use more direct measures of urban form. 1.2 Urban Form Measures Earlier research studies have used various kinds of urban form measures to capture the effect of the built environment on travel behavior. But in any particular study, it has been quite typical to consider only a handful of measures of urban form (and in most cases, just one measure). For R ajamani, et al. 4 example, a single measure of density has been used in several studies including Bhat and Singh (2000), Spillar and Rutherford (1990) and Dunphy and Fisher (1996). Some other studies such as Cervero et al. (1997), Handy (1993), Bhat and Pozsgay (2002), and Bhat and Zhao (2002) have focused on a single measure of accessibility to study the effect of urban form on travel and related behavior. A handful of studies have considered multiple urban form measures jointly. These multiple measures have typically been one of the two composite urban form measures discussed earlier (density or accessibility) and two additional characteristics of urban form. For instance, Frank and Pivo (1994) consider density and land use mix, Holtzclaw (1994) and Kitamura et al. (2001) use density and an accessibility measure, Kockelman (1996) considers accessibility, land use mix, and land use balance, and Greenwald and Boarnet (2001) use a composite pedestrian environment factor, population and retail densities, and proportion of gridiron streets. In this study, the focus is on capturing a multitude of urban form measures, some of which are composite indices (such as land use mix and accessibility) and others of which are direct, disaggregate measures of the built environment. Thus, for example, we consider not only the degree of mixing of different land uses, but also consider the actual kinds of land uses involved in the mixing. Hess et al. (2001) note that capturing the degree of mix may not suffice, and recommend including the actual kinds of land uses. Additionally, we examine the influence of the built environment, while controlling for the effects of sociodemographic and level-of- service variables on travel behavior. 1.3 Scale of Measurement and Level of Analysis The studies of land use and travel behavior may use urban form measures based on spatially aggregate units (such as city-level or urban/suburban level) or on much more disaggregate spatial units (such as the neighborhood level). Similarly, the analysis may be conducted at the level of an aggregate group of individuals or at the individual level. Therefore, four combinations of geographic scale and level of analysis are possible: (a) aggregate spatial data (at the traffic analysis zone or zip code level) and aggregate sociodemographics (for example, see San Diego Association of Governments, 1993; Handy, 1993; Hotlzclaw, 1994; Parsons Brinckerhoff Quade Douglas, 1994; McNally and Kulkarni, 1997), (b) aggregate spatial data and disaggregate sociodemographics (at the individual tripmaker’s level) (for example, see Ewing, 1995; Schimek, 1996; Kockelman, 1996; Boarnet and Crane, 2001; Greenwald and Boarnet, 2000), (c) disaggregate spatial data and aggregate sociodemographics, and (d) disaggregate spatial data and disaggregate sociodemographics (for example, see Cervero, 1996; Kitamura et al., 1997; Handy and Clifton, 2001; and Reilly, 2002). As should be obvious from above, there have been few studies that have employed urban form measures at a high level of spatial resolution and conducted the analysis at an individual level. In this paper, we use a GIS-based method to develop urban form measures at the neighborhood level of each household and conduct the analysis at an individual level. 1.4 Summary and Overview of Current Research This paper examines the impact of the built environment on travel behavior, with specific focus in the relatively lesser-researched area of nonwork travel. In addition, the paper uses a multitude of urban form measures, including composite indices and direct disaggregate measures. The analysis in the paper is based on a high spatial resolution for developing measures of urban form and is at the level of the decision-making unit (i.e. the individual tripmaker) and. A discrete R ajamani, et al. 5 choice methodology is used to examine the effect of household and individual sociodemographics, level-of-service of travel modes, and urban form measures on nonwork travel mode choice. The primary data source used for this study is the 1995 Portland Metropolitan Area Activity Survey, which collected travel information from members of a sample of households over a two-weekday period. The rest of this paper is organized as follows. The next section describes the data sources and the sample formation process. Section 3 discusses model specification issues. Section 4 presents the results of the empirical analysis. Finally, section 5 summarizes the significant findings from the research and identifies areas for future work. 2. DATA ASSEMBLY 2.1 Data Sources The primary source used in this study is the Portland Metropolitan Area Activity Survey conducted by Portland Metro in the Spring and Fall of 1995. This survey was a two-weekday travel diary of households. The information gathered in the activity survey included the travel mode used, start and end time of the trip and the activity, origin and destination locations (which were later mapped to traffic analysis zones) and individual and household socio-demographic information. In addition to the primary data source, three secondary data sources were used to generate the final sample for analysis. The first secondary data source is a zone-to-zone level-of-service file that includes mode-specific travel times and costs between each pair of zones in the Portland metropolitan region. This data includes a zone-to-zone transit fare matrix, and parking charges in each destination zone (besides the inter-zonal travel times by auto and transit). Travel times for the walk and bicycle modes are generated for each zonal pair based on the network distances and assuming an average walking and bicycling speed of 3 miles/hour and 10 miles/hour respectively, as suggested by Portland Metro. Walk and bike modes are assumed to entail no monetary cost to the trip maker. The second secondary data source used in the analysis is a zone- level land use file obtained from Portland Metro that provides the retail employment at each traffic analysis zone. The first and second data sources are used to develop a composite measure of accessibility to nonwork activity destinations for each travel mode and each zone. The third source is a neighborhood-level land use Geographic Information System (GIS) shape file developed by Portland Metro and enhanced by Song (2002) (to include several urban design and land use measures). This spatial land use file provides the following information for Washington County: (a) neighborhood land area, (b) area dedicated to each use (single-family housing, multi- family housing, commercial, industrial, and public open spaces), (c) population and housing unit densities in each neighborhood (developed by Song, 2002), and (d) local street network characteristics (number of blocks per housing unit, length of linear street network, fraction of cul-de-sac streets, street connectivity index etc. developed by Song, 2002). The definition of “neighborhood” adopted by Portland Metro is synonymous with census block group boundaries, except when an arterial road cuts across census block groups; in which case the block group is further divided with the arterial road representing a boundary. R ajamani, et al. 6 2.2 Sample Formation The process of developing the sample involved several steps. First, the composite (travel and non-travel) activity file was converted into a corresponding trip file. In doing so, information on the type of activity pursued at, and the zone identifier for, the origin and destination ends of each trip were retained. Second, nonwork trips originating at home were selected from the trip file based on the activity type designations at each end of the trip. Third, the travel mode chosen for each trip was identified and assigned one of the following categories: drive-alone, shared-ride, transit, walk, and bike. Fourth, the availability of modes for each trip was determined as follows: (a) Driving alone is considered as being available if the individual making the trip has a vehicle in the household and is a licensed driver, (b) Ridesharing is designated as being available to all individuals in the sample, (c) Transit is designated as being available to individuals whose origin-destination zonal pair has existing bus routes (this information was provided by Portland Metro), (d) Walking is designated as being available to individuals whose trip distance is less than the maximum distance walked by an individual in the sample, and (e) Biking is designated as being available to individuals whose trip distance is less than the maximum distance biked by an individual in the sample. Fifth, the trip data were matched with the appropriate sociodemographic characteristics of the individual pursuing the trip and his/her household. Sixth, the mode-specific level-of-service data were appended to each record based on the origin- destination zones and time-of-day of the trip. Seventh, the zone-level composite accessibility measures were matched with each trip record based on the origin zone. This provides the overall accessibility by each mode from the home zone of the individual pursuing the trip. Eighth, the location of the household of each traveler’s trip was geocoded, using the ArcGIS spatial tool, to the neighborhood level at which the urban form measures are available. This enables the assignment of urban form measures associated with the residence of the individual to each nonwork trip pursued by the individual. Since the urban form measures were available only for Washington County, the final trip sample includes only the trips of individuals residing in this county. The final sample for analysis includes 2,500 individual home-based1 nonwork trips (originating from 369 different households in 131 neighborhoods in Washington county), out of which 706 are shopping trips, and 763 are recreational trips. The overall mode shares for nonwork trips is as follows: drive-alone (48.1%), shared-ride (42.4%), transit (2.7%), walk (5.7%), and bicycle (1.1%). 3. EMPIRICAL ANALYSIS 3.1 Model Variables We considered four sets of variables for inclusion – household sociodemographics, individual sociodemographics, trip characteristics, and urban form measures. 1 Home-based trips, for the purpose of this study, refer to trips that originate at home; i.e. trips that have the tripmaker’s residential zone as the origin TAZ. Rajamani, et al. 7 3.1.1 Household Sociodemographic Characteristics The household sociodemographic attributes considered in our analysis included household income, number of vehicles per adult (above the age of 16) in the household, and number of children (below the age of 16) in the household. 3.1.2 Individual Sociodemographic Characteristics The individual sociodemographic characteristics that were explored included sex and age of the individual, ethnicity, student status, presence of a physical handicap and employment status. 3.1.3 Trip Characteristics The trip characteristics represent level-of-service variables. The in-vehicle travel time and out- of-vehicle travel time variables were combined into a single time variable, as initial testing of specifications rejected that their effects were different. Separate time coefficients were estimated for motorized and non-motorized modes to accommodate the differential time sensitivities based on travel mode. Travel cost was also included in the group of level-of-service variables. 3.1.4 Urban Form Measures The urban form measures incorporated in the model belonged to four categories: land use type and mix, accessibility, residential density, and local street network. The next four paragraphs discuss each of these categories in greater detail. The first category of urban form measures comprises the distribution quotients and a land use mix diversity variable. The distribution quotients are ratios of acreage in each land use type to the number of housing units in the neighborhood (see Bendavid Val, 1991 for use of distribution quotients). These ratios have been incorporated to overcome Hess et al.’s (2001) critique that land use-travel behavior models fail to test the effect of the types of uses mixed. The degree of land use mix is captured by the land use mix diversity measure. This measure varies between 0 and 1 and has been computed as follows: Land use mix diversity = 1-  r 1 c 1 i 1 o 1 − + − + − + −  (1)  T 4 T 4 T 4 T 4   3  2    where r = acres in residential use (single and multi-family housing), c = acres in commercial use, i = acres in industrial use, o = acres in other land uses, and T = r + c + i + o. A value of 0 for this measure means that the land in the neighborhood is exclusively dedicated to a single use, while a value of 1 indicates perfect mixing of the four land uses. The second category of urban form measures represents accessibility variables. Three accessibility variables are considered in the analysis: accessibility index, percentage of households within walking distance from commercial establishments, and percentage of households within walking distance from bus stops. While the first measure is an indicator of regional accessibility, the other two measures are associated with local accessibility (see Handy, 1993). The accessibility index has been computed using Levinson and Kumar’s (1995) gravity- type functional form as follows: R ajamani, et al. 8 JA (2) im = ∑[ f (Cijm ) * Rj ] j=1 where f(Cijm) is the friction factor between zones i and j by mode m (computed using the same functional form as used in Levinson and Kumar, 1995), Rj is the retail employment in zone j, J is the total number of zones in the Portland metropolitan area, i is the zone for which the accessibility index is being computed, and m is the mode for which the accessibility index is being computed Residential density is the next land use variable included in the model specification. It is defined as the number of housing units per unit area of a neighborhood (see Song, 2002 for development of this measure). Density has often been used to proxy a large number of excluded urban form measures. By including the residential density measure alongside other land use characteristics, this paper attempts to isolate the “true” impact of density. Finally, the local street network variables (developed by Song, 2002) capture the suitability of neighborhood streets for pedestrian and bicycle use. This set of variables includes a connectivity index and the percentage of cul-de-sac streets in the neighborhood. The connectivity index is defined as the ratio of the number of links to the number of nodes in the neighborhood. A greater value of this measure indicates a larger number of routes available for a given pair of point locations. Cul-de-sacs are common features of post World war-II type developments and are often considered the converse of traditional grid-like street geometry. The final model specification was developed through a systematic process of adding variables to the constants-only model, eliminating statistically insignificant variables, and combining statistically similar variables. This process was guided by intuitive reasoning and parsimony in the representation of variable effects. 3.2 Estimation Results The results of the multinomial logit estimation of the final model specification are presented in Table 1. The closed form of the MNL makes it straightforward to estimate, interpret, and use. The parameter estimates in the MNL model indicate the effects of exogenous variables on the latent utilities of the four modes relative to the shared-ride alternative. The log-likelihood value at convergence of the market share model is –2067.1, while the log-likelihood value of the final model specification is –1784.88. The likelihood ratio test value to compare the final model with the market share model is 564.44, which exceeds the critical chi- square value corresponding to 35 degrees of freedom at any reasonable level of significance. Thus, the hypothesis of no observed variable effects is rejected. In the subsequent sections, we present and explain the effects of variables, considering one set of variables at a time. The variables have been included with the coefficient on the shared-ride mode as the base. Three sets of constants are estimated, one for non-shopping and non-recreational trips, another for shopping trips, and a third for recreational trips. This is done to accommodate mode share differences across trip purposes. 3.2.1 Household Sociodemographic Characteristics Among the household sociodemographic variables, the annual household income was tested for a quadratic effect (see Boarnet and Sarmiento, 1998 and Boarnet and Crane, 2001). The results indicate that households below the income level of $50,475 per annum derive diminishing benefits from driving alone with an increase in income. On the other hand, individuals from higher-income households tend to drive more as their income rises. In other words, carpooling, R ajamani, et al. 9 riding transit, walking and biking are more attractive options relative to driving for households in the middle-income category. This may be a reflection of the increased availability of time for households that are rising from the lower-income to the middle-income category, which can be spent on a leisurely stroll in the park or on a social visit. Higher-income households, on the contrary, may have stringent time constraints and so prefer faster modes. An increase in the number of vehicles per adult in the household significantly increases the likelihood of choosing to drive-alone, as one would expect. The results also indicate that non-motorized modes are particularly unlikely to be used as vehicle availability increases. The next household characteristic is the number of children present in the household. A household with a higher number of children (persons below 16 years of age) is more likely to rideshare, as indicated by the negative and statistically significant coefficient for non-rideshare modes. This is a reflection of the mobility dependence of children on the adults in the household. Finally, in the category of household sociodemographics, the effect of the number of adults indicates a lower propensity to walk for individuals with several adults in the household. Further exploration of this effect is an area for future study. 3.2.2 Individual Sociodemographic Attributes In the class of individual sociodemographics, age has a significant impact on mode utilities. Non- linear effects were explored in our analysis, but the simple linear representation performed as well as non-linear forms. The magnitudes of the coefficients on age indicate that older individuals most prefer to rideshare for their non-work trips. This may reflect fewer time constraints among older individuals, and the consequent use of nonwork trips as socializing opportunities. The impacts of other individual sociodemographic traits indicate that physically handicapped individuals are more likely to travel by non drive-alone modes, and Caucasians have a greater aversion to walking than individuals from other ethnic backgrounds (see Greenwald and Boarnet (2001) for similar results). 3.2.3 Level-of-Service Variables The model estimation results indicate the usual negative impacts of travel time and travel cost. The results indicate that an additional minute of walk or bike time is marginally more onerous than a minute of driving or transit time. The implied money values of travel time by motorized and non-motorized modes are $17.90 per hour and $18.84 per hour, respectively. It is indeed interesting to note the similarity in valuation of travel time across modes. 3.2.4 Urban Form Measures As indicated earlier, four subgroups of urban form measures were considered in the current analysis. The effects of urban form measures are discussed by subgroup below. In the subgroup of land use type and mix, the effect of the ratio of park area per housing unit indicates, as expected, that availability of parks in the neighborhood increases the propensity to walk for recreation. The coefficient on the land use mix diversity index specific to transit is not significant, indicating that New Urbanists may be overestimating the impacts of transit- oriented development on transit ridership. On the other hand, the positive coefficient specific to the walk mode indicates that mixed-uses may have the potential to substitute motorized trips with walk trips. Overall, the effect of the land use mix diversity index corroborates the effectiveness of mixed-uses in encouraging walking as a mode for nonwork travel. Rajamani, et al. 10 The results of the accessibility index indicate that residents of a zone with higher regional accessibility by a given mode have a greater preference for recreational trips for that mode. The coefficient on the percentage of households within walking distance from bus stops (a measure of local accessibility) indicates a higher propensity to ride transit for individuals who live close to bus stops. Although this finding is consistent with New Urbanist concepts, the possibility of self-selection into transit-oriented neighborhoods cannot be discounted. The results of the impacts of residential density (captured by single-family housing unit density) indicate that denser neighborhoods increase the likelihood of walking for all nonwork trips. However, the negative and highly significant effect of housing density on transit use propensity is surprising. This finding needs to be explored in more detail; it may be related to the incorporation of density along with other urban form measures. The positive effect of density on transit use found in some earlier studies may be simply a manifestation of the effect of urban form measures that are positively correlated with density. In addition to the impacts of density on all nonwork trips, the coefficient on single-family housing unit density specific to shopping indicates a lower propensity to drive in dense neighborhoods. This is reasonable since more houses within the same area imply more traffic in local neighborhood streets, which may cause people to avoid driving to shopping destinations. The final variable in Table 1 is the percentage of cul-de-sacs. The direction of the influence of cul-de-sac streets is intuitive. A large number of cul-de-sac streets in a neighborhood (reminiscent of post World War-II type developments) might make walking very difficult because of curvilinear and seemingly longer routes. 4. POLICY IMPLICATIONS The results of the multinomial logit mode choice model have several important policy implications. The aggregate-level elasticity effects of variables are often used to assess the potential impacts of policy actions. In the following sections, the computation procedure for elasticity effects is explained and their implications are discussed. 4.1 Computation of Elasticity Effects The self and cross elasticity effects of mode-related variables like travel time and the accessibility index have been computed using the standard multinomial logit formulations. The aggregate-level elasticity effect of a continuous exogenous variable such as age may be computed by weighting the disaggregate-level elasticities by mode choice probabilities. The disaggregate-level elasticities with respect to variables with linear effects (e.g. age, land use mix diversity index, cul-de-sacs) may be computed as follows: E Pn ( i ) = x [γ − ∑ P (i)γ ] xk k ki n kjj where x is the kth attribute, P (i) is the probability that individual n chooses alternative i, and γ k n kj is the parameter estimate for the kth attribute specific to the jth alternative. The aggregate-level elasticity is then computed as follows: R ajamani, et al. 11 N ∑ P Pn ( i ) P ( i ) n (i)Ex η = n=1 k x Nk ∑ Pn (i) n=1 To compute an aggregate-level elasticity of an ordinal exogenous variable (such as the number of children in the household), we increase the value of the variable by 1 unit for each observation and compute new mode shares. We then compute the aggregate-level elasticity as follows: S l (i) − S (i) E S ( i ) = xk S (i) where S(i) and S l (i) are the shares of alternative i before and after the ordinal variable is increased by 1 unit respectively. Finally, to compute the aggregate-level elasticity of a dummy exogenous variable (such as ethnic status, and presence of a handicap), we divide the entire sample into two subsamples based on the value of the dummy variable. We then compute shifts in mode shares in each subsample due to a zero to one and one to zero change in the dummy variable respectively. The ratio of the difference of the shifts between the two subsamples to the aggregate mode shares in the entire sample yields the aggregate-level elasticity of the dummy variable. 4.2 Elasticity Effects of Exogenous Variables Table 2 presents the elasticity effects of all statistically significant non-mode related exogenous variables. The elasticity values indicate that vehicle ownership has a substantial impact on mode choice decision-making. Among individual sociodemographic characteristics, ethnicity (whether the individual is Caucasian or not) is the single most important determinant of the likelihood to walk. Although urban form measures do not seem as critical to mode choice, there are some important implications for integrated transportation-land use modeling. First, mixed-uses can lead to considerable substitution between motorized modes and walk modes. However, its influence on transit is quite contrary to expectation. The preference of people to walk in mixed- use neighborhoods for shopping and social-recreational trips suggests that residents “value the choice to walk”, as concluded by Handy (1996). Third, higher densities considerably improve the chances of walking as compared to other modes. Fourth, cul-de-sacs, characteristic of planned unit developments (PUD), can increase the resistance to walking and lead to greater automobile dependence. Table 3 presents the self and cross elasticity effects of mode-related variables. The self- elasticity results indicate that the share of walking (transit) for nonwork trips is very sensitive to increases in walk time (transit time). On the other hand, drive-alone share is relatively insensitive to travel time and cost increases associated with the private automobile. Further, the self- elasticity values show a substantial increase in the walk/bicycle share for recreational trips due to an improvement in accessibility by the walk/bicycle modes. This reinforces the idea that individuals prefer to walk or bike to accessible destinations for recreational purposes. Thus, providing a safe and comfortable pedestrian/bicycle environment, along with bringing shopping and recreational activity sites closer to residential neighborhoods, may be an effective way to reduce automobile dependency. However, another issue that deserves attention is the high cross- elasticity between the walk and bicycle modes associated with accessibility (see the last two R ajamani, et al. 12 rows in Table 3). That is, improvement in walk (bicycle) accessibility draws most share away from bicycle (walk) modes. This suggests that it is important to improve both the accessibilities by bicycle and walk modes simultaneously to increase non-motorized mode share. 5. CONCLUSIONS This paper presents a multinomial logit mode choice model for nonwork activity travel. The choice set comprises drive-alone, shared-ride, transit, walk and bike. The motivating factors for this study are: (a) the growing proportion of nonwork travel, and (b) the desire to explore the urban form influence on mode choice. The empirical analysis in this paper is based on the 1995 Portland Metropolitan Area Activity Survey. The model specifications tested several variables describing household sociodemographics, individual demographics, level-of-service variables and land use characteristics such as land use mix, accessibility, residential density and cul-de- sacs. Some salient results of the analysis are as follows: • Individuals in higher-income households have a greater tendency to drive-alone to nonwork activity sites as compared to individuals in lower-income households. However, there is no difference in their propensity to choose among other modes. Likewise, individuals in households that own more vehicles are more likely to drive-alone and less likely to walk or bike as compared to individuals in households owning fewer vehicles. Further, households with a large number of children are more likely to rideshare, because of the mobility dependence of children. • Older individuals tend to carpool or vanpool more and ride transit less than younger individuals for nonwork travel, while physically challenged individuals are less likely to drive and Caucasians are less likely to walk, for nonwork travel. • People tend to value walk and bike time only marginally more than they value travel time by motorized modes. • Mixed-uses and higher residential densities encourage walking for nonwork travel, while a large number of cul-de-sacs in local streets discourage walking. An increase in regional accessibility has the greatest positive influence on the propensity to walk/bike for recreation. The results indicate a clear relationship between mode choice decision-making and urban form characteristics. Specifically, mixed-uses and higher residential densities promote walking behavior. Thus, improving the diversity of uses in neighborhoods and increasing housing densities through flexible zoning can reduce automobile dependence. Land use planners should therefore envision denser, and diverse, neighborhoods as their goal for the next decade. Our exploration of accessibility measures yield results concurrent with previous studies. If destinations are easily accessible by walk/bicycle, people are more likely to walk/bike for recreational activities. It is therefore critical from a policy standpoint to make walking and biking attractive propositions through appropriate urban design. Our empirical analysis also suggests the validity of the New Urbanist contention that traditional neighborhood street design with few cul-de-sacs and a grid-like geometry has the potential to encourage walking (see Handy (1992) for a similar result about frequency of walking trips). However, various other factors such as street landscape and safety of sidewalks must be analyzed in conjunction with street geometry to comprehensively examine the New Urbanist contention. R ajamani, et al. 13 The current study may be extended in a number of ways. First, modeling mode choice and residential location choice decisions jointly would aid in disentangling the “true” causal effect of urban form on mode choice from the “spurious” effect of individuals selecting neighborhoods that support their intrinsic mode choice preferences. Second, panel data may be used instead of cross-sectional data to better understand the land use-transportation link. Third, since most nonwork travel occurs during the weekend, data on weekend travel would facilitate a more comprehensive examination of the effect of the built environment on nonwork travel. Finally, the inclusion of landscape variables such as patch density (see Crepeau, 2001) might improve the explanatory power of the models. Rajamani, et al. 14 Table 1 Effect of Exogenous Variables on Mode Choice Variable Drive-alone Transit Walk Bike Parameter t-statistic Parameter t-statistic Parameter t-statistic Parameter t-statistic Constants Non-work trips 1.1175 2.9388 3.3896 4.6195 3.5483 4.4097 -0.2377 -0.4614 Shopping 1.9018 2.3304 1.7175 -3.4371 2.4736 -2.6596 -0.2377 -0.4614 Recreation 0.6366 -3.9163 0.8905 -3.9277 4.2490 1.9157 -0.0913 0.2621 Household Sociodemographics Income in 10,000 dollars -0.3715 -2.8346 Square of Income 0.0368 2.8036 Vehicles per adult 0.7217 4.4129 -1.2987 -4.0020 -1.2987 -4.0020 Number of kids (<16 years) -0.3874 -7.8176 -0.3874 -7.8176 -0.3874 -7.8176 -0.3874 -7.8176 Number of Adults -0.6531 -4.1409 Individual Sociodemographics Age -0.0106 -3.3650 -0.0688 -6.8939 -0.0153 -3.5315 -0.0555 -3.5840 Handicap dummy -1.1786 -2.3104 Caucasian dummy -1.2210 -3.0818 Trip Characteristics Total Travel time in minutes -0.0513 -5.5614 -0.0513 -5.5614 -0.0540 -9.4097 -0.0540 -9.4097 Travel cost in dollars -0.1720 -1.6137 -0.1720 -1.6137 -0.1720 -1.6137 -0.1720 -1.6137 Urban Form Measures Park area per housing unit (recreational trips) 0.0001 1.5404 Land-use Mix Diversity Index 1.3825 2.9065 Natural Log of Accessibility Index (recreational trips) 0.1282 1.4289 0.1282 1.4289 0.1282 1.4289 0.1282 1.4289 Percentage of households within walking distance of bus stops 1.3061 2.2464 Single-family housing unit density -0.5892 -4.0765 0.1721 2.0345 Single-family housing unit density (shopping trips) -0.1642 -2.2069 Percentage of cul-de-sac streets -1.0246 -3.1503 Percentage of cul-de-sac streets (shopping trips) 0.5736 3.3555 Rajamani, et al. 15 Table 2 Aggregate Level Elasticity Effects of Non-Mode Related Exogenous Variables on Mode Shares Variable Drive-alone Shared-ride Transit Walk Bike Household Sociodemographics Annual Household Income (in 10,000 dollars) 0.1260 -0.0739 -0.0273 -0.0364 -0.0360 Vehicles per Adult 0.3859 -0.1224 0.0433 -0.9913 -1.0070 Number of Kids (<16 years) -0.1582 0.1533 -0.2078 -0.1665 -0.2090 Number of Adults (>=16 years) 0.0233 0.0409 0.0605 -0.3701 0.1250 Individual Sociodemographic Characteristics Age in years -0.1877 0.1980 -0.7468 -0.2741 -0.8219 Handicap Dummy -0.9037 0.5000 0.2462 0.4762 0.3115 Caucasian Race Dummy 0.0773 0.1462 0.3395 -2.1556 0.4929 Urban Form Measures Park Area per housing unit (recreational trips) -0.0631 -0.1007 -0.0120 0.0871 -0.0219 Land Use Mix Diversity Index -0.0179 -0.0269 -0.0370 0.3610 -0.0785 Percentage of Households within walking distance from bus stops -0.2795 -0.0182 0.4599 -0.0187 -0.0397 Single-family Housing unit density (all non-work trips) -0.0088 0.0223 -1.7527 0.6420 0.0362 Cul-de-sac streets (all non-work trips) 0.0002 0.0004 0.0005 -0.0048 0.0011 Rajamani, et al. 16 Table 3 Aggregate Level Self and Cross-Elasticity Effects of Mode-Related Variables Total Travel Time Drive-alone Shared-ride Transit Walk Bike Drive-alone -0.0837 0.0785 0.0784 0.0583 0.0773 Shared-ride 0.0693 -0.0939 0.0659 0.0562 0.0717 Transit 0.0120 0.0111 -0.9973 0.0092 0.0191 Walk 0.0854 0.0935 0.0779 -0.8623 0.1392 Bike 0.0022 0.0023 0.0035 0.0004 -0.3982 Travel Cost Drive-alone Shared-ride Transit Walk Bike Drive-alone -0.0177 0.0152 0.0218 0.0177 0.0773 Shared-ride 0.0067 -0.0103 0.0121 0.0099 0.0083 Transit 0.0014 0.0013 -0.1186 0.0016 0.0023 Walk - - - - - Bike - - - - - Natural log of accessibility (recreational trips) Drive-alone Shared-ride Transit Walk Bike Drive-alone 0.1322 -0.1161 -0.0308 -0.1377 -0.1346 Shared-ride -0.1029 0.1571 -0.0309 -0.1668 -0.1564 Transit -0.0006 -0.0006 0.0558 -0.0011 -0.0014 Walk -0.0174 -0.0232 -0.0076 0.1930 -0.0351 Bike -0.0014 -0.0019 -0.0009 -0.0033 0.3126 Rajamani, et al. 17 REFERENCES Bhat, C.R., Carini, J.P., and Misra, R. (1999) Modeling the Generation and Organization of Household Activity Stops. Transportation Research Record, No. 1676, 153-161 Bhat, C.R. and Singh, S.K. (2000) A Comprehensive Daily Activity-Travel Generation Model System for Workers. Transportation Research A, Vol. 34, No. 1, 1-22 Bhat, C.R., and Pozsgay, M.A. (2002) Destination Choice Modeling for Home-Based Recreational Trips: Analysis and Implications for Land-use, Transportation, and Air Quality Planning. Transportation Research Record, No.1777, 47-54 Bhat, C.R., and Zhao, H. (2002) The Spatial Analysis of Activity Stop Generation. Transportation Research B, Vol. 36, No. 7, 593-616 Bendavid-Val, A. (1991) Regional and Local Economic Analysis for Practitioners (Fourth Edition). Westport, Conn.: Praeger. Boarnet, M.G., and Crane, R. (2001) The Influence of Land Use on Travel Behavior: Specification and Estimation Strategies. Transportation Research A, Vol. 35, 823-845 Boarnet, M.G., and Sarmiento, S. (1998) Can Land Use Policy Really Affect Travel Behavior? A Study of the Link between Nonwork Travel and Land Use Characteristics. Urban Studies, Vol. 35, No. 7, 1155-1169 Cambridge Systematics. (1994) The Effects of Land Use and Travel Demand Strategies on Commuting Behavior. Washington, D.C. Federal Highway Administration Cervero, R. (1989) America’s Suburban Centers: The Land Use-Transportation Link. London: Allen and Unwin Cervero, R. (1996) Mixed Land Uses and Commuting: Evidence from the American Housing Survey. Transportation Research A, Vol. 30, 361-377 Cervero, R., and Kockelman, K.M. (1997) Travel Demand and the 3 Ds: Density, Diversity, and Design. Transportation Research D, Vol. 2, 199-219 Cervero, R., Rood, T., and Appleyard, B. (1997) Job Accessibility as a Performance Indicator: An Analysis of Trends and their Social Policy Implications in the San Francisco Bay Area. Working Paper 692. University of California at Berkeley. Cervero, R., and Wu, K. (1997) Influence of Land Use Environments on Commuting Choices: An Analysis of Large U.S. Metropolitan Areas using the 1985 American Housing Survey. Working Paper 683. University of California at Berkeley. Crepeau, R.J. (2001) A Spatial-Temporal Analysis of Transit-Oriented Development, Land Use Patterns and Travel Behavior. Presented at the 2001 Meeting of the Association of Collegiate Schools of Planning, Cleveland, Ohio Duany, A. and Plater-Zyberk E. (1992) The Second Coming of the American Small Town. Wilson, Quarterly 16 (Winter), 19-51 Dunphy, R.T., and Fisher, K. (1996) Transportation, Congestion, and Density: New Insights. Transportation Research Record, No. 1552, 89-96 Federal Highway Administration. Our Nation’s Travel: 1995 NPTS Early Results Report. Washington D.C., 1995. http://www-cta.ornl.gov/npts/1995/doc/index.shtml. Accessed on April 20, 2002 Frank, L.D., and Pivo, G. (1994) Impacts of Mixed Use and Density on Utilization of Three Modes of Travel: Single-Occupant Vehicle, Transit, and Walking. Transportation Research Record, No. 1466, 44-52 R ajamani, et al. 18 Greenwald, M.J., and Boarnet, M.G (2001) Built Environment as Determinant of Walking Behavior: Analyzing Nonwork Pedestrian Travel in Portland, Oregon. Transportation Research Record, No. 1780, 33-42 Handy, S.L. (1992) Regional Versus Local Accessibility: Neo-Traditional Development and its Implications for Nonwork Travel. Built Environment, Vol. 18, 253-267. Handy, S. L. (1993) Regional Versus Local Accessibility: Implications for Nonwork Travel. Transportation Research Record, No. 1400, 58-66 Handy, S. L. (1996) Urban Form and Pedestrian Choices: Study of Austin Neighborhoods. Transportation Research Record, No. 1552, 135-144 Handy, S.L., and Clifton, K.J. (2001) Local Shopping as a Strategy for reducing Automobile Travel. Transportation, Vol. 28, 317-346 Hess, P.M., Moudon, A.V., and Logsdon, M.G. (2001) Measuring Land Use Patterns for Transportation Research. Transportation Research Record, No. 1780, 17-24 Holtzclaw, J. (1994) Using Residential Patterns and Transit to decrease Auto Dependence and Costs. Unpublished Report, Natural Resources Defense Council. Kitamura, R., Mokhtarian, P.L., and Laidet, L. (1997) A Micro-analysis of Land Use and Travel in Five Neighborhoods in San Francisco Bay Area. Transportation, Vol. 24, 125-158 Kitamura, R., Akiyama, T., Yamamoto, T., and Golob, T.F. (2001) Accessibility in a Metropolis. Transportation Research Record, No. 1780, 64-75 Kockelman, K.M. (1995) Which Matters More in Mode Choice: Density or Income? Institute of Transportation Engineers 1995, Compendium of Technical Papers, 844-867 Kockelman, K.M. (1996) Travel Behavior as a function of Accessibility, Land Use Mixing and Land Use Balance: Evidence from the San Francisco Bay Area. Master’s thesis. University of California at Berkeley. Levinson, D., and Kumar, A. (1995) A Multimodal Trip Distribution Model: Structure and Application. Transportation Research Record, No. 1466, 124-131 Levinson, D.M., and Kumar, A. (1997) Density and the Journey to Work. Growth and Change, Vol. 28, 147-173 McNally, M.G., and Kulkarni, A. (1997) Assessment of Influence of the Land Use- Transportation System on Travel Behavior. Transportation Research Record, No. 1607, 105-115 Messenger, T. and Ewing, R. (1996) Transit-Oriented Development in the Sunbelt. Transportation Research Record, No.1552, 145-152 Parsons Brinckerhoff Quade Douglas. (1994) Building Orientation – A Supplement to “The Pedestrian Environment”. 1000 Friends of Oregon, Portland, 9-14 Reilly, M.K. (2002) The Influence of Urban Form and Land Use on Mode Choice: Evidence from the 1996 Bay Area Travel Survey. Presented at the 2002 Transportation Research Board, Washington, D.C. San Diego Association of Governments (1993) Trip Making in Traditional San Diego Communities. Working Paper. Schimek, P. (1996) Household Motor Vehicle Ownership and Use: How Much Does Residential Density Matter? Transportation Research Record, No. 1552, 120-125 Song, Y. (2002) Valuing the Impacts of New Urbanism Features on Prices of Single-Family Homes: A Case Study of Portland, Oregon. Dissertation. Graduate College of the University of Illinois at Urbana-Champaign Spillar, R.J., and Rutherford, G.S. (1990) The Effects of Population Density and Income on Per Capita Transit Ridership in Western American Cities. Institute of Transportation Engineers 1990, Compendium of Technical Papers¸ 327-331