Built Environment Impacts on Individual Mode Choice: An Empirical Study of the Houston-Galveston Metropolitan Area

ABSTRACT

This study examines the impacts of the built environment measures based on two geographic scales, i.e., traffic analysis zone and one quarter-mile buffer on individual mode choice in the Houston metropolitan area. It is confirmed that they have significant impacts on mode choice in varying degrees. The models including the buffer-based measures are more reasonable than those with conventional zone-based variables for both home-based work and other trips. Finally, the elasticity estimates suggest the built environments are undervalued in the conventional transportation practices. Both land use and transport pricing measures should be considered complementary to control the demand for driving.

Key Words:

• built environment
• elasticity
• Houston-Galveston Area Council
• mode choice
• multinomial logit

ACKNOWLEDGMENTS

The authors are grateful to Mr. Charlie Hall of the Texas Department of Transportation and to Dr. David Pearson and Mr. Edwin Hard of the Texas Transportation Institute for providing the HGAC travel survey data. The authors appreciate the generous help of Mr. Chris Van Slyke, Ms. Heng Wang, and Ms. Sharon Ju of the HGAC, who made other data available. The authors also thank Dr. Ming-Han Li of Texas A&M University and the anonymous reviewers for providing throrough and helpful comments.

Notes

¹Because the areas are computed based on GIS data that contains total area of each object such as town, city and county, they would be a little different from other sources of information.

^a Modal splits of 1990, 2000, and 2005 for HGAC regional transportation planning area are obtained from HGAC (2007).

^b Mode share of work trips in the study area computed based on the 2007 HGAC regional travel survey. Only 84% of total planned samples are used as done in this study.

^c 2007 national average of principal mode share of commuting trips (Bureau of Transportation Statistics 2008).

²Discrete choice theory is similar to the economic consumer theory in that the consumer choice for utility maximization is still effective; however, it employs different functional specifications due to discrete dependent variables (Ben-Akiva and Lerman 1985; Meyer and Miller 2001).

³The MNL model has the fundamental property of the independence from irrelevant alternatives (IIA). It indicates that the ratio of the probabilities of making two choices depend only on their attributes; it is independent of the existence of any other alternative. It leads to overestimating the probabilities of choosing similar alternatives and also underestimating the chances of choosing the distinct options (Ben-Akiva and Lerman 1985; Meyer and Miller 2001).

^a Because the disaster caused by the hurricane Ike in September 2009 affected residents’ travel behavior in the metropolitan region, 84% of total samples that have been collected before the natural hazard are used for the study.

⁴Travel cost in this study represents short-term operating cost. Generalized cost is a broader concept (Lee 2006). In order to estimate actual travel cost, a variety of vehicle years, models, makes and types were fully considered referring to the fuel economy estimates published by U.S. DOE and U.S. EPA (2009). In addition, 2008 average gas price in Texas, and 2008 average maintenance and tires costs are also taken into account (American Automobile Association 2008).

^a Built environment variables are measured for 17 households located in TAZ number of 1427.

^a For the non-motorized mode, estimated trip time = 1/[1 + |age −30| / 30] × trip distance/speed, assuming 3 miles per hour for walking and 9 miles per hour for biking (Zhang 2004).

^b For the automobile trip cost, estimated trip cost ($) = [gas price + maintenance + tires] × distance (U.S. DOE and U.S. EPA).

^c For the shared-ride mode, estimated trip cost = total automobile trip cost/number of riders.

^d Regional accessibility (RA _i ) = , where i and j are TAZs, and t _ij is travel time between TAZ i and j. β is a parameter of Bessel function (HBW = 0.00156, and HBO = 0.0042) (Lee 2006).

^e O = trip origin, D = trip destination.

^a DA = driving-alone, SR = shared-ride, TR = transit, WB = walk/bike.

^b O = trip origin, D = trip destination.

*p < 0.10, **p < 0.05, ***p < 0.01.

⁵Travel costs specific to auto modes are highly correlated with travel time variables because they represent short-term operating costs. It is found that the travel times worked better in the models for HBW trips.

⁶It is partially because connectivity gets much higher in TAZs located in downtown Houston and some suburban centers (Lee 2009). It seems to be associated with negative externalities such as high traffic congestion and commuting cost primarily in downtown Houston.

^a DA = driving-alone, SR = shared-ride, TR = transit, WB = walk/bike.

^b O = trip origin, D = trip destination.

*p < 0.10, **p < 0.05, ***p < 0.01.

⁷For detailed explanations and mathematical expression of the probability-weighted average elasticity, refer to Ben-Akiva and Lerman (1985).

^a DA = driving alone, SR = shared ride, TR = transit, WB = walk/bike. For non-driving-alone modes, the effects of travel time indicate the cross-elasticity estimates of modechoice probabilities.

^b O = trip origin, D = trip destination.

^a DA = driving alone, SR = shared ride, TR = transit, WB = walk/bike. For non-driving alone modes, the effects of travel cost indicate the cross-elasticity estimates of mode choice probabilities.

^b O = trip origin, D = trip destination.

^c Elasticity of roadway is calculated by a 1% decrease of roadway measure.

⁸The elasticity estimates are additive because they are partial elasticities (Ben-Akiva and Lerman 1985; Zhang 2004). Based on the result of the buffer-based model in Table 7, the combined elasticities of driving-alone are −0.011 for diversity and −0.007 for density.

⁹Based on the result of the buffer-based model in Table 8, the combined elasticities of driving-alone for density, diversity, and design dimensions are −0.024, −0.034, and −0.054, respectively.

¹⁰This study does not control for residential self-selection, which could lead to biased estimates of coefficients and elasticities. However, this limitation should not affect the major findings of this study.

¹¹Some experts only support feasible market reforms, arguing that limited choices and growing automobile dependence due to transportation market distortions result in economic inefficiency, social inequality, and environmental disruption (Litman 2000). Others, on the other hand, claim that land use policy reform is required to address the problems and its negative consequences with skepticism about the role of zoning and other land use regulation measures (Levine 2006). However, they are beyond the scope of this article.