## Built Environment and Nonmotorized Travel: Evidence from Baltimore City Using the NHTS

## Built Environment and Nonmotorized Travel: Evidence from Baltimore City Using the NHTS

**FELIPE TARGA** ^{1}**KELLY J. CLIFTON** ^{2 }*

### ABSTRACT

The relationship between land use and travel behavior is a topic of debate among researchers and practitioners seeking to find land-use planning interventions to manage travel demand. This paper presents an empirical analysis of the effects of several land-use, urban form, and neighborhood-level design attributes, as well as traveler attitudes/perceptions of the urban system, on the frequency of walking, and the share of walking trips relative to total trips. Using the 2001 National Household Travel Survey add-on for the Baltimore metropolitan region, the paper estimates Poisson regression models at the person-level for the number of walking trips and a linear regression model for the share of walking trips made during a single travel day. The results suggest that neighborhoods with higher densities, more diverse land-use mixes, better street connectivity, and better access to bus transit lines are associated with persons who walk more frequently and make more walking trips with respect to trips made by other modes. Among the built environment variables, street network connectivity had the largest elasticity with respect to frequency of walking. Potential limitations of the analytical approach, as well as the degree of generalization of the results and their policy implications, are discussed.

KEYWORDS: Built environment, urban form, land use, nonmotorized travel, trip generation, NHTS, Poisson regression.

### INTRODUCTION

In the last decade, transportation trends for American cities have been generally characterized by increasing automobile ownership and use. These trends, on the one hand, have originated or aggravated existing transportation problems in urban areas, such as traffic congestion, air quality, energy consumption, livability, and public health. On the other hand, these trends have also motivated planning policy initiatives aimed at reducing some of these negative impacts by managing the demand for travel. Among the set of transportation demand management policy measures recently proposed, one of the most controversial relies on the connection between transportation and land use.

Planning practitioners and researchers who advocate for neotraditional and transit-oriented development rely on a hypothesized relationship between land use and travel activity behavior. Recent initiatives along this line of thinking consider land-use policies as a means to ease congestion, improve air quality, curb automobile demand, and contribute to improved quality of life by making urban areas more livable. Although these benefits may be expected if the urban built environment becomes more accessible, success of these policies hinges on understanding and anticipating travelers' responses.

Over the last decade, a substantial body of research has been accumulated that focuses on empirically testing the effect of several measures of urban form and neighborhood-level design characteristics on travel demand. The widely dissimilar approaches to measuring urban form attributes, the use of different levels of aggregation for both land-use and travel data, and the focus on different transportation-related outcomes have resulted in mixed, and sometimes, insignificant evidence. The empirical analysis conducted in this paper contributes to the general understanding of the relationship between land use and travel behavior by testing the effects of several land-use and urban form characteristics, as well as traveler attitudes/perceptions of the urban system, on two travel-related outcomes measured at the individual levelfrequency of walking trips and share of walking trips with respect to total all-mode trips.

### CONCEPTUAL FRAMEWORK

From a theoretical perspective, it is unclear what the net effect on the intensity of travel should be if urban areas are made more accessible. For example, improvements in the transportation system are expected to decrease travel times and thus lower the cost of travel. If the cost of travel goes down, consumption of travel can be expected to go up. Some studies use this rationale to argue that in urban environments where destinations are close by or more accessible, the cost per trips will be lower and higher trip generation rates or vehicle-miles travel can be expected (Crane 1996).

This is just a theoretical expectation, however, because the actual effect will depend on the elasticities of the demand with respect to price and the availability and feasibility of alternative modes. Indeed, depending on contextual attributes of the urban and transportation systems, the cost of travel may not be lower. The rationale for this argument is that the cost of a trip is a function of time, which in turn is a function of distance and speed and other travel-related attributes, such as out-of-pocket expenses, safety, and comfort. In a compact, clustered, and mixed-use development, we might expect that origins and destinations would be closer to each other, but the effect on speeds will depend on other factors. Therefore, the total impact on travel times and travel costs will depend on whether or not speed impacts overcome distance effects.

Additionally, the net effect on travel consumption (i.e., making more or fewer trips) might not only depend on the elasticity of the demand, but also on the degree of mode substitution and regional accessibility aspects. This net effect could be also moderated by particular travel activity attributes, such as trip purpose or mode of travel, as well as individual and household sociodemographic characteristics. In general, if there are cases where differences in travel-related outcomes can be attributed to differences in the built environment, they might also depend on the elasticity of the demand for travel with respect to person- and trip-related attributes.

The conceptual structure developed in this paper focuses on the empirical examination of observed travel outcomes, in a single point of the time (i.e., cross-sectional design), for households located in neighborhoods with different built environment attributes. Although the analytical framework allows for statistical association of these effects, it also has some limitations for inference about causality and net substitution effects. On the one hand, self-selected travelers (i.e., those who decided to locate in dense, mixed-use, and accessible neighborhoods as a result of their desire to live in these urban environments in order to walk more frequently) might confound the ability to make causality inferences. On the other hand, statistical associations between observed travel outcomes and households' built environment characteristics do not allow for formal inference about net travel substitution effects (e.g., more walking trips are substituting for car trips). These two aspects are crucial for understanding the implications of any land-use policy aiming to leverage the demand for auto travel and promoting more competitive and sustainable modes of travel.

These critical issues could be addressed with a more comprehensive study design, particularly with longitudinal or panel data structures, which could contain more detailed data (e.g., the travel decisionmaking process on consecutive days of travel, including short-term travel activity behavior, and long-term decisions, such as auto ownership and residential location choices). Based on the cross-sectional data available for this paper, the analytical framework estimates measures of statistical association between walking trip generation rates and households' built environment attributes, including land-use, urban form, and other neighborhood-level design characteristics. In addition to trip generation rates, statistical associations are also estimated for the share of walking trips with respect to all trips. If more accessible neighborhoods are associated with households where travelers walk more frequently, the set of associations based on modal share can provide additional insights about the possible trip substitution effects.

Finally, this paper advances the limited knowledge of built environment effects on nonmotorized and nonwork travel. Likewise, instead of using only a handful of built environment attributes, this study captures the association effects of a full complement of land-use and urban form measures, including disaggregate measures and composite indexes. In addition to objective or direct measures of the built environment attributes, the analytical framework also captures the association effects of attitudes and perceptions of travelers with respect to the urban system. Recent studies of the relationship between land use and travel behavior (Targa and Clifton 2004) have recognized that travel-related choices are expected to not depend exclusively on objective measures of the transportation system or the land-use characteristics but also on the perceived subjective attributes of the system.

### LITERATURE REVIEW

Over the last decade, researchers have focused on empirically testing the effect of several measures of land-use, urban form, and neighborhood-level characteristics on travel behavior or travel-related outcomes (Badoe and Miller (2000) and Ewing and Cervero (2001) provide a detailed review of these studies). Overall, results from the most disaggregated and carefully controlled studies suggest that effects on trip generation rates depend mainly on household socioeconomic characteristics and that travel demand is inelastic with respect to accessibility (Ewing and Cervero 2001). Likewise, one common finding that comes from these studies is that the built environment has a greater impact on trip lengths than on trip frequencies. Nonetheless, some studies have also shown that urban environments with higher densities, a mix of land uses, and grid-style street configurations are associated with higher frequencies of walking/biking and other nonwork-based trips (Handy 1993, 1995, 1996; Friedman et al. 1994; Cervero and Gorham 1995; Kulkarni et al. 1995; and Cervero and Radisch 1996). Studies focusing on mode choice have found that this decision depends as much on built environment attributes as on socioeconomic characteristics. The association effects of built environment attributes with other travel-related outcomes, such as vehicle-miles traveled, have been documented as small but statistically significant.

Within the existing empirical studies, questions remain about the degree of trip substitution effects among different modes of travel and issues of self-selectivity (e.g., people who prefer walking/biking choose to live in built environments that facilitate that behavior as opposed to the urban form influencing their behaviors). Few studies have provided formal evidence of the underlying direction of causality, and among these studies, the results are mixed. Using cross-sectional data and controlling for preferences and attitudes, some studies have found that observed associations between travel behavior and neighborhood characteristics are largely explained by the self-selection of residents with certain attitudes (Bagley and Mokhtarian 2002), while others have not found such an impact after accounting for attitudes (Schwanen and Mokhtarian 2005). A recent study found that characteristics of the built environment influence walking behavior after accounting for a preference for walking-friendly neighborhoods (Cao et al. In press).

### DATA DESCRIPTION

The primary data source for this study is the 2001 National Household Travel Survey (NHTS), in particular, the additional 3,446 households surveyed from June 2001 through July 2002 in the Baltimore metropolitan region. Households were randomly selected for participation in the Baltimore add-on sample. The survey was gathered through computer-assisted telephone interviews. In order to be consistent with the national data, the 2001 NHTS add-on survey was conducted following basically the same definitions and procedures of the 2001 NHTS national sample.

Land-use and urban form/design attributes used in the empirical examination were computed from several archived sources, such as census and county TIGER-enhanced files for the year 2000. Household locations were geocoded based on the respondent-provided closer location place of residence. Using geographic information systems (GIS), land-use, urban form, and other neighborhood-level design characteristics were assigned to each household record based on its geographic location. Most of these measures were operationalized consistently with previous efforts focusing on the characterization of built environment attributes (Galster et al. 2001; Song and Knaap In press). GIS and the increasing availability of land-use and transportation data in electronic format aided in the production of these secondary data. Census 2000 sociodemographic information was also obtained for the area of study. The geographic area of analysis consisted of the city of Baltimore, including 1,539 surveyed households (figure 1) or 2,934 persons with reported travel-day data.

Among the 2,934 travelers, 2,061 (70.25%) did not make any walking trips during the reported travel day, and 580 travelers (19.77%) reported one or two walking trips. Figure 2 depicts the spatial distribution of the frequency of walking trips on the reported day of travel. Analyzing the trip purpose variable from the trip-level data (not shown here), we confirmed that the majority of these walking trips (91.5%) were generated from nonwork-related activities (38.48% were home-related). In terms of modal share, walking trips accounted for 100% of the total trips on the reported day of travel for 10.87% of the surveyed travelers, while the average modal share for walking trips was 19.24%.

In addition, to control for traditional socioeconomic and demographic characteristics and trip-related attributes, the analytical framework developed in this paper uses attitudinal and perceptual data as proxies for sociopsychological factors influencing travel activity behavior. Perceptual data include attitudes toward traffic accidents, highway congestion, the presence of drunk drivers on the road, lack of sidewalks and walkways, and the price of gasoline. The existence of a medical condition that impedes the mobility of the respondent is also expected to influence travel behavior by limiting driving, the use of transit, or reducing the amount of travel made.

The set of explanatory variables of interest consists of urban form, neighborhood design, and land-use attributes associated with the geographic location of each traveler's household. Although several variables were constructed using GIS-based data, household unit density at the census block-level, street connectivity (measured as the perimeter of the census block), the diversity of land-use mix at the census block group-level (measured as an 0 to 1 index indicating the degree of land-use mixing), and distance to the nearest bus transit stop were the variables finally selected. The selection of explanatory variables was based on statistical (e.g., the most parsimonious model specification) and study-specific considerations (e.g., different levels of spatial aggregations will impact the effects of some land-use or urban form attributes). Neighborhood sociodemographic characteristics were also obtained at the census block-level for 2000.

The possibility exists that some land-use variables could be correlated with household variables, or that some built environment attributes may be correlated with specific socioeconomic variables. However, a correlation analysis (not shown here) confirmed that all pair correlations were low, except for density and street connectivity (p = 0.51), which were part of the same set of built environment attributes. Table 1 presents summary statistics for the set of dependent and explanatory variables for the area of analysis (i.e., Baltimore City).

Conceptually, we expected that neighborhoods with higher densities, fine land-use mixes, better street connectivity, and generally better access to transit would be associated with persons making more walking trips and with a higher walking modal share. Because shopping trips and other nonwork-based trips tend to be more elastic with respect to accessibility and are more likely to be done by nonmotorized modes than work trips, we expected differences in urban form and design attributes to be more influential for these trips.

### METHODOLOGY

Walking trip generation rates and modal share proportions were calculated at the person-level for all household members with reported travel-day data (a 24-hour period). Given the count-type nature of the data for the number of walking trips, the methodological approach consisted, initially, of specifying and estimating a Poisson regression model. In a Poisson model specification, a random variable indicates the number of events (i.e., walking trips) during an interval of time (i.e., reported travel day). In the regression model, the number of events *y* has a Poisson distribution with a conditional mean that depends on household or travelers' characteristics, trip characteristics, and land-use/urban form attributes according to the following structural model:

*μ _{i}* =

*E*(

*y*|

_{i}*x*) = exp (

_{i}*)*

**x**_{i}**β**where * x_{i}* is a row vector with observations of the explanatory variables for each person, and

*is a column vector of estimated coefficients associated with each explanatory variable. This structural model is estimated by means of maximum likelihood (ML) estimation techniques. Asymptotic tests of the coefficient estimates and calculation of marginal effects are used to evaluate the statistical significance and relative magnitude of the effects of land-use/urban form measures on frequency of walking.*

**β**McFadden's likelihood ratio index (McFadden 1973) and adjusted McFadden's *R* ^{2} for the number of parameters (Ben-Akiva and Lerman 1985) are used as scalar measures of fit for the Poisson models. These measures are the most popular approximations to the coefficient of determination *R* ^{2} in linear regression models. In particular, the log-likelihood of the model without regressors is thought of as the total sum of squares, while the log-likelihood of the model with regressors is thought of as the residual sum of squares.

The second step in the methodological approach consisted of specifying and estimating a linear regression model by means of the ordinary least squares (OLS) for the walking modal share variable. This regression model uses the same set of explanatory variables specified in the Poisson models and follows the structural model below:

*y _{i}* =

*c*+

*+*

**x**_{i}**β***u*

_{i}### MODEL ESTIMATION

This section presents the estimation results for the models discussed in the preceding section. The coefficients of the explanatory variables included in the Poisson model specification are estimated by means of ML and represent the relative effect of the associated variable on the frequency of walking. Coefficients are estimated without expansion factors or analysis weights commonly used to avoid bias in the statistical analysis. Particular attention is devoted to the estimates of the built environment attributes, the primary interest of this paper.

The Poisson model was estimated for three different specifications (table 2). Model 1 includes only traditional household and person socioeconomic characteristics. Model 2 includes all variables used in model 1, along with attitudinal and perceptual data of the urban and transportation system. Model 3 includes all variables used in model 2, as well as all the built environment attributes and the neighborhood sociodemographic characteristics. The OLS model (model 4) for walking modal share is estimated with all of the explanatory variables used in the Poisson model 3.

Table 2 summarizes the corresponding coefficient estimates, *t* statistics, and the statistical significance test for each estimated coefficient. All models were statistically significant at the 99% confidence level (*p*< 0.001 for the *χ*^{2} test). The model specification with traditional explanatory variables for trip generation rates (model 1) helps to explain some of the variability of frequency of walking compared with a model without regressors (McFadden's adjusted *R*^{2}= 0.098). Among household characteristics, lower number of vehicles and higher number of bicycles per household member, college dorm home type, and lower household income were characteristics associated with a higher frequency of walking, as expected. Traveler characteristics associated with a higher frequency of walking included young, nonlicensed driver, temporarily absent from a job or looking for work, full-time workers, professional or managerial occupation category if working, healthy, graduate-level education, and people who frequently walk for exercise and have their work location closer to home.

Adding attitudinal variables to the model specification (McFadden's adjusted *R*^{2}= 0.110) increased the statistical explanatory power of the model with respect to model 1 (likelihood ratio test; *χ*^{2}_{(8)}= 107.7 and *p*< 0.001). Among attitudinal variables, the individual estimated coefficients suggest that people who are more concerned with traffic accidents, highway congestion, and drunk drivers are likely to walk more frequently than people less concerned with these urban system characteristics. Interestingly, people who drive frequently and express more concern about the price of gasoline were less likely to walk. Those who indicated that sidewalk conditions presented "a little" and "somewhat" of a problem tended to walk more frequently.

A particularly notable finding of our analysis is the statistically significant association between built environment attributes and the frequency of walking. Comparing the overall performance of model 3 (McFadden's adjusted *R*^{2}= 0.141), the explanatory power of the model increases statistically with respect to model 2 (likelihood ratio test; *χ*^{2}_{(6)}= 256.5 and *p*< 0.001) and with respect to model 1 (likelihood ratio test; *χ*^{2}_{(14)}= 364.3 and *p*< 0.001). The same set of explanatory variables explain 13.5% (adjusted *R*^{2}= 0.135) of the variability of the proportion of walking trips with respect to all trips made on the surveyed day of travel (model 4). Overall, these results show that both attitudinal variables and built environment attributes increase the statistical explanatory power of the models, and consequently, help to better explain the variance of the dependent variables (i.e., frequency of walking and walking modal share). However, this study's primary focus is the relative effect that specific land-use, urban form, and neighborhood-level design variables have on walking trip generation rates and on the share of walking trips.

In particular, people living in denser urban settings, measured as the number of household units per square mile in the corresponding household census block, tend to walk more frequently on the surveyed day of travel, all else being equal. We hypothesize that the sign of the *Density* coefficient is positive and statistically different from zero based on a one-tail test (*p*= 0.093). The marginal effect of *Density* (table 3) suggests that an increase of 1% in the number of household units per square mile (within the census block where the traveler's household is located) is associated with an increase of the expected number walking trips of 0.026, all else being equal. This translates into an elasticity of 0.033, evaluated at the mean value of the walking trip generation rate.^{1} In other words, a 1% increase in the number of household units per unit of area is associated with a 0.033% increase in the expected number of walking trips on a given day. This elasticity is even lower than the average density elasticity of vehicle-miles traveled (0.05) estimated in previous studies (Ewing and Cervero 2001).

Likewise, people living in neighborhoods with higher street connectivity or with more grid-like street networks, measured as the perimeter of the corresponding household census block, are likely to walk more frequently, as reported on the surveyed day of travel. The marginal effect of the street connectivity variable (table 3) suggests that a one mile decrease in the perimeter of the corresponding census block (i.e., more connected street networks) is associated with an increase of the expected number of walking trips of 0.587, all else being equal. Evaluated at the mean of the walking trip generation rate, this translates into an elasticity of 0.258.^{2} This means that a 1% decrease in street network connectivity (i.e., length of a census block perimeter) is associated with a 0.258% increase in the expected number of walking trips made on the reported day of travel, all else being equal.

Figure 3 depicts the probability of making one or more walking trips on the reported day of travel as the length of the census block perimeter varies from 0 to 3 miles, holding the rest of the variables at their means. Moreover, the coefficient of street connectivity in the walking share model (model 4) suggests that the same one mile decrease in street network connectivity is associated with an increase of 8.9% in the proportion of walking trips with respect to all other trips made by other modes, including car, on the reported day of travel.

The degree of land-use mix was captured by an index of land-use mix diversity ranging from 0 to 1 (Song and Knapp In press). If the land use in the census block group associated with the traveler's household is dedicated exclusively to a single use, the diversity index variable takes a value of 0. Conversely, a value of 1 indicates perfect mixing of the land uses considered in this study (i.e., residential, commercial, industrial, institutional, and open urban space). Evaluated at the mean of the walking trip generation rate, the marginal effect of this land-use mix index (table 3) translates into an elasticity of 0.065.

Figure 4 depicts the probability of making one or more walking trips on the reported travel day as the land-use mix diversity index varies from 0 to 0.75, holding the rest of the variables at their means. The slope of the fitted line in figure 4 shows the low elasticity value for the land-use mix index. Likewise, the coefficient of the land-use mix index in the walking share model (model 4) suggests the same significant, but marginal, association of this index for the proportion of walking trips with respect to all trips made on the reported day of travel.

Access to bus transit lines is also statistically associated with higher walking frequency, as expected. In particular, the marginal effect suggests that people living one mile closer to a bus stop are expected to make 0.647 more walking trips on the reported day of travel. Evaluated at the mean of the walking trip generation rate, this effect translates into an elasticity of 0.070.

The last set of estimated coefficients suggests that people living in neighborhoods with a lower median age and higher white population proportion are likely to walk more frequently as reported on the surveyed day of travel. Only neighborhoods with a greater proportion of white residents are associated with a larger walking share with respect to all trips in the surveyed day of travel.

### CONCLUSIONS AND LIMITATIONS

The conceptual structure and the empirical results presented in this paper advance the understanding of the relationship between land use and travel behavior. In particular, our findings contribute to the general understanding of this relationship by testing the effects of several land-use and urban form characteristics, as well as traveler attitudes and perceptions of the urban system, on two nonmotorized travel-related outcomes. Based on a cross-sectional study design, the paper estimated measures of statistical association between walking trip generation rates and households' built environment attributes, including land use, urban form, and other neighborhood-level design characteristics. In addition to trip generation rates, statistical associations were also estimated for the share of walking trips with respect to total travel.

Using the 2001 NHTS for the City of Baltimore (2,630 travelers), the paper estimated three Poisson regression models at the person-level for the number of walking trips during a single surveyed day of travel and a linear regression model for the share of walking trips made on the same travel day. The results suggest that neighborhoods with higher densities, fine land-use mixes (i.e., more diverse), better street connectivity, and generally better access to bus transit lines were associated with persons who walk more frequently and have a higher proportion of walking trips with respect to all trips. These results were expected given the theoretical elasticity of nonmotorized travel with respect to accessibility.

Results from the model for walking share with respect to total travel suggest that more accessible neighborhoods are not only associated statistically with households where travelers walk more frequently, but also with households where the proportion of walking trips is higher on the same day of travel. These results provide some insights into possible trip substitution effects in these neighborhoods but are restricted to inference limitations discussed later in this section.

The 2001 NHTS Baltimore add-on survey used random selection, and previous analysis of this dataset has shown that the sample is representative of the population (Battelle and Morpace 2002). However, caution should be taken when trying to transfer the results here to different locations. Indeed, the degree of generalization of the results and the general external validity of the empirical findings is limited to the context of travel and urban setting characteristics in the geographic area of study (i.e., Baltimore City).

Nonetheless, two critical issues could not be addressed comprehensively given the limitations of the study design and data availability. In particular, the analytical framework allowed for statistical association between land-use and travel behavior effects, but it had some limitations for inference on causality and net travel substitution effects. Without longitudinal or panel data structures containing more detailed information (e.g., the travel decisionmaking process on consecutive days of travel, including short-term travel activity behavior and long-term decisions, such as auto ownership and residential location choices), we were unable to formally evaluate possible confounded effects under conditions of self-selected travelers (e.g., those located in dense, mix-used, accessible neighborhoods as a result of their desire to live in those urban environments).

One of the complications of using cross-sectional data is that the model specification cannot capture the endogenous processes typically found in travel decisionmaking. In general, the presence of endogeneity in the model estimation might yield inconsistent and biased estimates of the relationships. However, few studies have provided formal empirical evidence of the underlying direction of causality, and among these studies the results are mixed. While some studies found that observed associations between travel behavior and neighborhood characteristics are largely explained by the self-selection of residents with certain attitudes, others have not found such an impact after accounting for attitudes toward travel and location preferences. Future research will benefit from more detailed travel data, particularly longer periods of observed or surveyed travel (i.e., panel or longitudinal studies).

Despite the potential limitations of the analytical approach, the results of this paper are highly relevant for transportation planning practitioners and researchers and improve our understanding of the relationship between land use and travel behavior. Ultimately, the empirical evidence provided in this paper is expected to contribute to the growing body of literature focusing on the interaction between land use and travel demand.

### ACKNOWLEDGMENT

The opinions expressed in this paper are those of the authors and do not necessarily reflect the official position of the Inter-American Development Bank.

### REFERENCES

Badoe, D.A. and E.J. Miller. 2000. Transportation-Land-Use Interaction: Empirical Findings in North America and Their Implications for Modeling. *Transportation Research Part D: Transport and Environment*5(4):235263.

Bagley, M.N. and P.L. Mokhtarian. 2002. The Impact of Residential Neighborhood Type on Travel Behavior: A Structural Equations Modeling Approach. *Annals of Regional Science*36:279297.

Battelle and Morpace. 2002. 2001 National Household Travel Survey Add-On Program: Baltimore Regional Transportation Board, final report and data codebook. October.

Ben-Akiva, M.E. and S.R. Lerman. 1985. *Discrete Choice Analysis: Theory and Application to Travel Demand.* Cambridge, MA: MIT Press.

Cao, X., S. Handy, and P.L. Mokhtarian. In press. The Influences of the Built Environment and Residential Self-Selection on Pedestrian Behavior. *Transportation.*

Cervero, R. and R. Gorham. 1995. Commuting in Transit Versus Automobile Neighborhoods. *Journal of the American Planning Association*61:210225.

Cervero, R. and C. Radisch. 1996. Travel Choices in Pedestrian Versus Automobile Oriented Neighborhoods. *Transport Policy*3:127141.

Crane, R. 1996. Cars and Drivers in the New Suburbs: Linking Access to Travel in Neotraditional Planning. *Journal of the American Planning Association*62:5165, Winter.

Ewing, R. and R. Cervero. 2001. Travel and the Built Environment. *Transportation Research Record*1780:87114.

Friedman, B., S.P. Gordon, and J.B. Peers. 1994. Effect of Neotraditional Neighborhood Design on Travel Characteristics. *Transportation Research Record*1466:6370.

Galster, G., R. Hanson, M.R. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage. 2001. Wrestling Sprawl to the Ground: Defining and Measuring an Elusive Concept. *Housing Policy Debate*12(4):681717.

Handy, S. 1993. Regional Versus Local Accessibility: Implications for Non-Work Travel. *Transportation Research Record*1400:5866.

______. 1995. Understanding the Link Between Urban Form and Travel Behavior, presented at the 74th Annual Meetings of the Transportation Research Board, Washington, DC, 1995.

______. 1996. Urban Form and Pedestrian Choices: Study of Austin Neighborhoods. *Transportation Research Record *1552:135144.

Kulkarni, A., R. Wang, and M.G. McNally. 1995. Variation of Travel Behavior in Alternative Network and Land Use Structures. *ITE 1995 Compendium of Technical Papers.* Washington, DC: ITE, pp. 372375.

McFadden, D. 1973. Conditional Logit Analysis of Qualitative Choice Behavior. *Frontiers of Econometrics.* Edited by P. Zarembka. New York, NY: Academic Press.

Schwanen, T. and P.L. Mokhtarian. 2005. What Affects Commute Mode Choice: Neighborhood Physical Structure or Preferences Toward Neighborhoods? *Journal of Transport Geography*13:8399.

Song, Y. and G.J. Knaap. In press. Measuring Urban Form: Is Portland Winning the War on Sprawl? *Journal of the American Planning Association.*

Targa, F. and K.J. Clifton. 2004. Integrating Social and Psychological Processes into the Land-Use Travel Behavior Research Agenda: Theories, Concepts and Empirical Study Design, presented at the 7th International Conference on Travel Survey Methods, Los Sueños, Costa Rica, Aug. 16, 2004.

### END NOTES

1. Because Density was transformed with a logarithmic function, a change in the transformed variable is associated with a 1% change in the units of the untransformed variable. Moreover, the elasticity is calculated at the mean of the walking trip generation rate; an increase of 0.026 trips is equivalent to an increase of 3.3% with respect to the mean value (0.80 trips).

2. A one mile decrease with respect to the mean of the perimeter of the census blocks is equivalent to a 296% decrease, and an increase of 0.587 trips is equivalent to an increase of 76.4% with respect to the mean value (0.80 trips). This translates into an elasticity of 0.258.

### ADDRESSES FOR CORRESPONDENCE

F. Targa, Inter-American Development Bank, Finance and Infrastructure Division, RE1/FI1, 1300 New York Avenue, Washington, DC 20577. E-mail: felipet@iadb.org

Corresponding author: K. Clifton, Assistant Professor, Urban Studies and Planning, National Center for Smart Growth Research and Education, Preinkert Fieldhouse, Suite 1112, University of Maryland-College Park, College Park, MD 20742. E-mail: kclifton@umd.edu