Impact of Metropolitan-level Built Environment on Travel Behavior Arefeh Nasri 1 and Lei Zhang 2,* 1. Graduate Research Assistant; 2. Assistant Professor (*Corresponding Author) Department of Civil and Environmental Engineering, University of Maryland 1173 Glenn Martin Hall, College Park, MD 20742, 301-405-2881, lei@umd.edu Abstract There has been a growing recognition about the significant impact of land use pattern on travel behavior and changes in built environment pattern could potentially be considered as a long-term solution to change people s travel behavior and especially vehicle miles traveled (VMT). However, the existing literature has been mainly focused on local and neighborhood characteristics of the built environment and little is known about the unique or relative influence of the metropolitan-level built environment. In this empirical analysis we use an extensive database for six major metropolitan areas in the U.S to employ multilevel mixed effect model highlighting the impact of built environment characteristics on travel behavior at different scales. Our findings show that changes in built environment measures not only at local and neighborhood levels but also at larger metropolitan/ regional levels could be very influential in changing people s travel behavior. Specifically, promoting compact, mixed-use built environment with well-connected street network can help to reduce VMT and thus provide better solutions to transportation-related issues. Key words: Built environment, Metropolitan level land use, Travel behavior, Vehicle miles traveled (VMT), Multilevel model, Mixed effect. Acknowledgement The authors would like to thank MPOs in Seattle, Baltimore, Washington DC, Atlanta, and State DOTs in Maryland, Virginia, and Georgia for providing several datasets for the analysis. The authors are solely responsible for all statements in this paper. 1. Introduction The linkage between built environment and travel behavior has been intensively studied since the 1980s and a lot of mixed findings have been published. Recent research suggests that people living in neighborhoods with pedestrian-oriented design, characterized by high street connectivity, mixed land use, and high population density, are encouraged to drive less. However, the existing research has mainly focused on built environment at local and neighborhood levels rather than at regional and higher levels.
The rapid process of urbanization in the U.S. is identified as a leading factor to increased automobile-based mobility (Berube, 2007; Ross, 2009). As a consequence of the increased mobility, individuals or households would no longer be restricted into opportunities located in their own residential neighborhoods and are more likely to travel to further destinations. Thus the physical form of metropolitan areas including their land use pattern, size, population, employment, and transportation infrastructure pattern could potentially have a significant impact on economic activities, housing, transportation, etc. Some empirical research suggests that metropolitan structure such as degree of concentration, decentralization, regional sub-centers, and job-housing balance could be more influential in travel than the typical neighborhood level built environment features (Newman and Kenworthy, 1999; Horner 2002; Shen 2000; Yang 2008; Yang and Ferreira 2008). Given the lack of empirical research on the connection between travel behavior and metropolitan-scale built environment, this paper attempts to examine the statistical association between the two using multilevel mixed effect regression models. 2. Data and Model Specification This project uses data for six metropolitan areas in the United States including Washington DC; Baltimore, MD; Seattle, WA; Atlanta, GA; Richmond-Petersburg, VA; and Norfolk-Virginia Beach, VA. For each metropolitan area, household travel information and the most recent land use data for the entire metro area obtained from various local MPOs, regional transportation planning agencies, and States DOTs has been used. There are five built-environment factors at neighborhood level used in our model, including residential and employment densities, mixed use development (entropy), average block size, and distance to the city center. Our built environment measures at the neighborhood level have been directly taken from the analysis of Zhang, et al. 2011. The density variables are calculated by dividing the residential/ employment population by the area size (acre) of each TAZ or census tract. Distance to the city center is simply the straight line from CBD and entropy is calculated using the following formula ranging from 0 to 1: Entropy = Where is the proportion of land use in the jth use category and J is the number of different land use type classes in the area. The factors we constituted to measure metropolitan level built environment capture residential and employment density, strength of city centers, and street network accessibility and connectivity. To measure each of these factors, we defined variables such as average population
and employment densities for each county in the metropolitan area, proportion of population living within 5 miles from CBD, and proportion of employment locating within 5 miles from CBD, average block size for each county in the metropolitan area, and average entropy (mixed use) also at the county level. The dependent variable VMT- is measured also by using the same method as Zhang et al. 2011 and calculated the weighted distance by dividing total travel distance for each reported trip by the number of people in the vehicle used for the trip. Table 1 shows the descriptive statistics for the households in each of the study area. Table 1 Socioeconomic Characteristics for Metropolitan Areas of Study Metropolitan Atlanta Baltimore Norfolk- & Seattle Washington Area Richmond Sample Size 2079 3991 4352 3945 8537 Household Size Number of Workers Number of vehicles Household Income Mean 2.28 2.29 2.49 2.26 2.23 Std. 1.18 1.22 1.21 1.22 1.22 Mean 1.12 1.24 1.11 1.17 1.28 Std. 0.73 0.87 0.89 0.84 0.81 Mean 1.91 1.89 2.32 1.92 1.77 Std. 0.93 1.04 1.13 1.04 0.99 Mean $50,000- $60,000- $60,000- $70,000 $75,000-59,999 74,999 64,999-80,000 99,999 Std. - - - - - Table 2 Summary Statistics of Built Environment Measures Metropolitan Area Atlanta Baltimore Norfolk- & Richmond Mean Std. Mean Std. Mean Std. Seattle Mean Std. Washington Mean Std. Population density 4.37 2.47 8.83 9.32 3.14 2.82 6.64 7.45 11.95 14.58 Employment density 2.26 2.47 4.72 17.22 1.25 1.84 3.71 14.83 7.22 22.62 Entropy 0.71 0.24 0.47 0.21 0.60 0.16 0.33 0.15 0.42 0.22 Average block size 0.087 0.082 0.097 0.145 0.143 0.167 0.078 0.132 0.107 0.177 Distance to CBD 14.53 7.73 13.18 8.75 18.15 12.19 15.12 10.26 14.79 12.80
Table 2 reports means and standard deviations of the TAZ and census tract level variables for all the cases represented in this study. These statistics indicated considerable variation in neighborhood characteristics. The mixed model is well suited for our analysis as on the one hand, there are households of the same category (i.e., live in the same TAZ or census tract), but on the other hand, they have different characteristics (i.e., different socioeconomic status). The mixed effect model allows us to have different coefficients by subject groups. Subjects in the same level/group are likely to be similar to each other in terms of their observable characteristics. Consequently, there are two sources of variation: variation between different TAZs or census tracts (inter-subject variance) and variation within a particular TAZ or census tract (intra-subject variance). The mixed model can be represented as: Where: y = Xβ+ Zu+ ε y is a vector of observations, with mean E(y) = Xβ β is a vector of fixed effects u is a vector of independent identically-distributed random effects with mean E(u) = 0 and variance-covariance matrix var(u )=G ε is a vector of random error terms with mean E(ε) = 0 and variance var(ε)=r X and Z are matrices of regressors relating the observations y to β and u 3. Results The results from the multilevel mixed-effect model show a strong association between VMT and built environment characteristics at both neighborhood and metropolitan levels. Table 3 attempts to summarize the empirical evidence for all six metropolitan areas included in our model which was discussed earlier in more details. The table is divided into three main categories: Households socio-economic variables impacts, neighborhood-level land use factors impacts, and metropolitan-level built environment impacts. We controlled for the potential effects of socioeconomic status using households size, annual income, number of workers, and vehicle ownership in the model. The results show that except household size, other socio-demographic variables proved to be significantly influencing VMT with a positive direction, though the impacts of number of workers and car ownership are much more considerable compared to households annual income. It is reasonable as larger households with higher income tend to have more cars and thus drive more.
Table 3 Results for Multilevel Mixed Effect Regression Model Variables Coefficient Standard error p-value Socioeconomic factors Household size 0.007056 0.006903 0.3067 Household income 0.006640 0.000949 <.0001 # of vehicles 0.3135 0.008323 <.0001 # of workers 0.3151 0.009978 <.0001 Neighborhood-level built environment Population density (logged) -0.1566 0.02162 <.0001 Employment density (logged) -0.03610 0.01924 0.0606 Land use mix (entropy) -0.1171 0.1123 0.2970 Distance from CBD (logged) 0.2447 0.01827 <.0001 Average block size -0.5893 0.08915 <.0001 Metropolitan-level built environment Average residential density (logged) -0.01031 0.04644 0.8242 Average employment density (logged) -0.06790 0.03413 0.0466 Average block size -0.7437 0.1016 <.0001 Average entropy -1.2378 0.1675 <.0001 Population within 5 miles of CBD (logged) -0.9022 0.03412 <.0001 Employment within 5 miles of CBD (logged) -0.4959 0.04632 <.0001 Covariance parameter estimates TAZ or Census Tract 0.1228 Residual 1.0638 As it is shown in Table 3, all land use variables at the neighborhood and local level have negative relationship with VMT except the distance from CBD, which is positively linked to VMT. This result is consistent with the previous studies as well. The results of the model at neighborhood level show that people living in areas with compact development pattern, higher employment opportunities, and better mixed neighborhoods tend to drive less. Distance to central business district has a negative association with VMT meaning people living farther from CBD have to drive more to reach various destinations. The average block size as a measure of street network connectivity has a positive significant relationship with households VMT also because with lower block size - higher street connectivity- distance to various types of destinations would be lower and as a result people drive less to reach those destinations. At the metropolitan level, the influence of built environment measures is much lower than that at the neighborhood level in terms of magnitude, though the directions of the relationship are the same. However, the impact of metropolitan-level average entropy seems to be much higher than
that of local level which is not statistically significant either. These results claim that the metropolitan-level population and employment density could also be very influential on driving behavior of the households in addition to these factors at the neighborhood and local level. Average entropy seems to be the most important land use factor at the metropolitan level, which is correlated to households VMT. Better mixed used development through the whole metropolitan area influences VMT and driving behavior of the residents. Furthermore, the proportion of population and employment locating in the 5-mile buffer zone of the downtown area is a significant factor in predicting driving behavior and households VMT. This measure of metropolitan built environment could be interpreted as an indicator of strength of city centers as downtowns with higher population and employment concentrated within the five-mile buffer zone are considered stronger and could potentially influence the driving behavior of the people living the whole metropolitan area. This impact is larger in terms of magnitude compared to the average population and employment densities which confirms the significant role of city centers in the metropolitan areas. These results clearly confirm the previous findings saying that people living in more compact/ mixed-development neighborhoods with higher street connectivity tend to drive less. However, providing additional case study areas is needed in order to assess a better and more reliable model on the potential impacts of metropolitan level built environment on travel behavior and VMT. 4. Conclusion The relationships found between local and metropolitan-level built environment and VMT show that transportation-related problems would be much less in better mixed areas with higher population density, better street connectivity and less sprawl even when controlling for income, household size, and other households socioeconomic characteristics. While these findings may seem obvious, this is one of the first studies to explicitly measure metropolitan-level built environment and relate it to household s VMT. Thus, looking at not only the land use characteristics of the immediate neighborhoods of residence in the analysis, but also considering the position of the neighborhoods in the large metro areas has been the main contribution of this analysis. Our findings clearly show that measures of urban form have a significant association with VMT among households of all studied metropolitan areas, adjusting for households sociodemographic characteristics. Households living in neighborhoods with higher population and employment densities, better mixed land development, and closer to city centers tend to drive significantly lower.
As a next step, we are expanding this analysis by adding about 15 more case study areas across the country and also improving our model by adding a measure of transit accessibility in the neighborhoods of study. This makes our model even more comprehensive and thus can help policy and decision-makers to evaluate land use plans and policies according to their impact on vehicle miles traveled. References: Albright, J. J. and Marinova, D. M. (2010). Estimating Multilevel Models using SPSS, Stata, SAS, and R. available online at: http://www.indiana.edu/~statmath/stat/all/hlm/hlm.pdf Berube, A. (2007). Metro Nation: How U.S. Metropolitan Areas Fuel American Prosperity, research report at Washington, DC: Brooking Institute. Cao, X., Mokhtarian, P. L., & Handy, S. L. (2006). Neighborhood design and vehicle type choice: Evidence from Northern California. Transportation Research Part D, 11(2), 133-145. Cervero, R. (1996). Mixed land-uses and commuting: Evidence from the American Housing Survey. Transportation Research Part A: Policy and Practice, 30(5), 361-377 Cervero, R. (1998). The transit metropolis: a global inquiry. Ahont Island Press. Demidenko, E. (2004). Mixed Models, Theory and Applications. Wiley Interscience. El Nasser, H. and Overberg, P. (2001). A comprehensive look at sprawl in America, USA TODAY. Ewing, R., Pendall, R., and Chen, D. (2002). Measuring Sprawl and Its Impact. Smart Growth America. Available online at: http://www.smartgrowthamerica.org/research/measuring sprawl-and-its-impact/ Frank, L. D. (2000). Land Use and Transportation Interaction: Implications on Public Health and Quality of Life. Journal of Planning Education and Research, 20(6), 6-22. Hess, P. M., et al. (2001). Measuring land use patterns for transportation research. Transportation Research Record 1780. Horner, M. W. (2002). Extensions to the concept of excess commuting. Environment and Planning A 34:543-566. Kelly-Schwartz, A. C., Stockard, J., Doyle, S., and Schlossberg, M. (2004). Is Sprawl Unhealthy? A Multilevel Analysis of the Relationship of Metropolitan Sprawl to the Health of Individuals, Journal of Planning Education and Research 24:184-196 Kitamura, R., Mokhtarian, P. L., & Daidet, L. (1997). A micro-analysis of land use and travel in five neighborhoods in the San Francisco Bay Area. Transportation, 24(2), 125-158. McCann, B. A., and Ewing, R., Measuring the Health Effects of Sprawl; A National Analysis of Physical Activity, Obesity and Chronic Disease, 2003 National Academy of Science (2009). Driving and the Built Environment: The Effects of Compact Development on Motorized Travel, Energy Use, and CO2 Emissions. Available online at: http://www.nap.edu/catalog/12747.html Newman, Peter and Kenworthy, Jeffrey, 1999, Sustainability and Cities: Overcoming Auto Dependence, Island Press.
Ross, Catherine L. (2009). Megaregions: Planning for Global Competitiveness. Island Press. Rubin T, Lowe M, Aytekin B, Gereffi G (2010) Does Bus Transit Reduce Greenhouse Gas Emissions? Retrieved from http://reason.org/news/show/does-bus-transit-reducegreenhouse Rundle, A. et al. (2007). The Urban Built Environment and Obesity in New York City: A Multilevel Analysis. American Journal of Health Promotion. 21(4) 326-334. Shen, Q. (2000). Spatial and social dimentions of commuting, Journal of the American Planning Association 66 68-82. Yang, Jiawen, and Joseph Ferreira. 2008. Choices versus choice sets: a commuting spectrum method for representing job-housing possibilities. Environment and Planning B-Planning & Design 35:364-378. Yang, Jiawen. (2008). Policy implications of excess commuting: Examining the impacts of changes in US metropolitan spatial structure. Urban Studies 45:391-405. Zhang, L., et al. (2011). How Built Environment Affects Travel Behavior: A Comparative Analysis of the Connections between Land Use and Vehicle Miles Traveled in U.S. Cities.