For consideration for presentation at the Transportation Research Record conference

Transcription

1 Transit Commuting and the Built Environment in Station Areas across the United States For consideration for presentation at the Transportation Research Record conference Text Word Count: 3,286 Tables: 2 Figures: 1 John Renne, Ph.D., AICP Associate Provost for Urban Initiatives Director, Merritt C. Becker Jr. Transportation Institute Associate Professor, Department of Planning and Urban Studies University of New Orleans 368 Milneburg Hall University of New Orleans 2000 Lakeshore Drive New Orleans, LA Phone: jrenne@uno.edu Reid Ewing, Ph.D. Director, Metropolitan Research Center Professor of City and Metropolitan Planning University of Utah 375 South 1530 East Salt Lake City, UT Phone: ewing@arch.utah.edu Shima Hamidi Doctoral Student City and Metropolitan Planning University of Utah 375 South 1530 East Salt Lake City, UT Phone: shima.hamidi@gmail.com Total Word Count (including tables and figures [equivalent to 250 words each], but excluding references in accordance with TRB Info for Authors alternative manuscript length option, p.5): 4,036

2 ABSTRACT This study examines how the built environment impacts the mode share for transit commuting across station. Data is examined from 4,400 fixed- transit stations (91% rail) across 54 regions in the United States. The study utilizes a Multiple Level Model (MLM) that examines data at both the neighborhood- level and regional- level. This is one of the first studies to include a variable that measures the network accessibility of jobs and population within walking distance of the transit network. Significant variables in predicting the mode share for transit commuting at the neighborhood level include income, population and jobs intensity, nonwhite status, transit technology (ie. light rail/streetcar and heavy rail/metro rail), jobs- population balance, walkability, and transit service frequency. However, the strongest predictor was the regional network accessibility variable. The results of this study support policies that promote TOD at both the neighborhood and regional scales. Given that only 20% of stations across the United States achieve a minimum density of 15 units per acre, the findings of this study should guide planners and policy- makers to encourage TOD to boost the mode share of transit commuting in station areas where the infrastructure already exists, but where walkable TOD- style development is currently lacking. INTRODUCTION The United States is best know globally as a nation of roads, highways, and automobiles while few studies have examined the network of over 4,500 fixed- transit nodes, including the built environment around these stations. Transit- Oriented Development (TOD) is increasingly becoming a popular way to orient neighborhoods and regions for the goal of boosting mode shares of transit commuting. Studies of travel behavior and the built environment have disaggregated the building blocks of TOD into components such as density, land use mix, and urban design (1, 2). Most travel behavior studies of residents in close proximity to rail stations have been limited in scale and geography. From a design- scale perspective, previous studies tend to focus solely on household and/or neighborhood- scale variables, ignoring the importance of network accessibility of regional transit systems to access jobs and people. This study overcomes this limitation by including neighborhood and regional- level variables into a multiple- level model (MLM). Moreover, the study is national in scope and includes data from 4,400 station areas across 54 regions. Travel and the Built Environment As a share of all travel, commuting represents only 22% of all trips across the United States (3) and public transit commuting has remained relatively constant at approximately five percent of all workers from (4). However, these statistics mask the important role of commuting in the transportation system and the role that transit plays. Roadway congestion costs the average American commuter $818 in lost time and fuel in 2011 compared to an inflation- adjusted $342 per commuter in In total, congestion cost the American economy $121 billion in 2011, while public transit saved $20.8 billion (5). Most aggregate studies of commuting focus on regions where transit mode shares are high, such as New York, Chicago, Philadelphia, Washington, Baltimore, Boston, and San Francisco. Other aggregate studies examine the phenomenon of transit commuting through the lens of metropolitan size, central city versus suburbs, and population density (6), but have not taken into consideration the role of the built environment on transit use. Transportation and the built environment have a reciprocal relationship, with impacts in both directions (7). Research on this topic dates back to the Von Thunen model of agricultural land use (8), and extended to residential location choice by Alonso (9) and Muth (10). The Alonso- Muth model predicts higher land values near the city center, as transportation cost savings are

3 Renne, Ewing and Hamidi 3 capitalized in the value of land. Land values drop off with distance from a city center. Recent meta- analyses (11, 12) and literature review (13) focus on the positive impact that railway stations have on property values. A number of studies have examined the relationships between transit and urban form. An early pioneering study found that density had a significant influence on rail transit boardings, with light rail transit being more sensitive to residential density and commuter rail more sensitive to CBD employment density (14). Many since then have helped to develop a robust literature on this topic as summarized by a recent meta- analysis that examined various D variables as measures of the built environment (2). The Ds are development Density, land use Diversity, pedestrian- oriented Design, Destination accessibility, Distance to transit, Demand management, and Demographics. The Ds virtually define TOD, which seeks to maximize transit use by creating dense, walkable, and mixed- use communities. While this paper is broader than TOD, it is important to note that literature on this topic informs this paper and the results of this analysis sheds light on the debates within that discipline. Such debates are summarized in the findings section below. METHODOLOGY The conceptual modeling framework utilized in this study seeks to examine the average commute mode share for all fixed- transit station areas across the United States to better understand why some precincts generate higher shares of transit commuting than others. Data were collected from the National TOD Database, an open source dataset made available by the Center for Transit Oriented Development, with funding from the Federal Transit Administration. The dataset contains information on approximately 4,400 stations across 54 metropolitan areas. Data are available for quarter and half- mile buffers around the individual stations. For the purpose of this study, we utilized the half- mile buffer. This study utilizes MLM, otherwise known as hierarchical modeling, to explain variance in transit commute mode shares across regions and station areas. Essentially, MLM partitions variance between the station precincts and regional levels and then insofar as possible, explains variance at each level using variables specific to that level. MLM accounts for the fact that stations are nested within regions and share the characteristics of the region, violating the independence assumption of ordinary least squares (OLS) regression. Because it overcomes this serious limitation of OLS, MLM has long been in fields like education and public health to analyze nested data. MLM is just beginning to be used in planning research (15, 16). As shown in Figure 1 the commute mode share across the station area is hierarchical in nature and a function of both neighborhood level (Level 1) and regional level (Level 2) characteristics. Variables and data sources for Level 1 and Level 2 are depicted in Table 1. Level 1 variables are significant determinants of travel behavior and include population and employment density, demographics, transit service, land use mix, urban design characteristics and distance to the CBD. Some important data are not available at this scale for the dataset, such as parking availability, which has been shown significant in other studies (17). At the regional level, the accessibility of people to jobs via the transit network varies significantly from region to region. In regions like New York, where a relatively high percentage of jobs and population live within the network, one would expect such accessibility to positively influence the share of transit commuting. Alternatively, for a city like Houston with low accessibility of jobs and people within the railway network, we would expect that a significant percent of people living within its railway precincts to access jobs via car. This level of analysis seeks to capture network effects; the higher the share of jobs and people living within the fixed- transit network, the higher

4 Renne, Ewing and Hamidi 4 the predicted mode share for transit commuting at the station level. Other Level 2 variables include data on sprawl and traffic at the regional level. Figure 1: Conceptual Hierarchical Study Design Commute Mode Share across Station Area Characteristics of the Neighborhood (Level 1 Variables) Population density Employment density Demographics Transit service Land use mix Urban design characteristics Distance to the CBD Characteristics of the Region (Level 2 Variables) Percentage of regional population within all railway station precincts Percentage of regional jobs within all railway station precincts Sprawl Traffic congestion Data Sources The National TOD Database provides nearly 70,000 data points derived from the 2000 and 2010 Decennial Census, the 2009 American Community Survey, the 2000 Census Transportation Planning Package, and the Local Employment Dynamics data. As shown in Table 1, most of the data in this study is derived from the National TOD Database, however other data were joined to the database using Geographic Information Systems (GIS). At the neighborhood level (Level 1), data on the percentage of four- way intersections, Walk Score ratings, distance to the central business district, and transit frequency data were joined to the database. At the regional level (Level 2), data on sprawl and traffic were joined to the database. Self- Selection Bias Over the past decade or so a debate within the literature has questioned if the built environment has an influence over peoples decisions to use transit or if people with a desire to use transit self- select to live near railway stations. Such a debate is useful, but it requires data on individual attitudes and preferences that are not available on a national scale. This paper takes a different approach, one that is more aggregate in nature. It focuses on the characteristics of neighborhoods and regions that make transit mode share higher in one place than another. It seeks to explain why some station precincts generate high mode shares for commuting and others underperform. Dependent Variable The dependent variable in this study is the share of transit commuting for each railway, ferry, bus- rapid transit (BRT), and monorail/automatic guideway precinct in the United States; 91 percent of all stations are railway, 2.3 percent are ferry, 5.6 percent are BRT and 1.3 percent are monorail/automatic guideway. The United States has a network of over 4,400 stations, most of which have failed to attract transit- supportive development (18). The average share of transit commuting across all station precincts in the United States is 22 percent, but the average masks

5 Renne, Ewing and Hamidi 5 sharp differences from station to station. The standard deviation of mode share18.6. The dependent variable was transformed to a natural log in order to achieve a more normal distribution of the dependent variables. The independent variables were also logged. Log- log transformations have the added advantage of allowing regression coefficients to be directly interpreted as arc elasticities. Table 1: Variables and Data Sources Dependent Variable Description Source Transit Share of transit commuting across half mile station area Regional Variables (Level 2) Population Share Share of the total regional population (2010) living within a half mile catchment of all of the rail stations within the region Jobs Share Share of the total number of jobs (2009) located within a half mile catchment of all of the rail stations within the region Jobs+Population Share Share of the total number of jobs (2009) and people (2010) located within a half mile catchment of all of the rail stations within the region Sprawl Measure of regional sprawl developed by Ewing and Hamadi (2014) Ewing and Hamadi 2014 Traffic Average annual delay per commuter Texas Tranpsortation Institute Neighborhood Variables (Level 1) Density Activity Density Number of people (2010) plus the number of jobs (2009) per acre within rail precinct Population Density People per acre (2010) Emloyment Density Jobs per acre (2009) Demographics and Socioeconomics Nonwhite Share of nonwhite population within rail precinct (2010) Hispanic Share all households within the rail precinct that are Hispanic (2010) Income Median household income Professional Workers Percent of workers employed in profssional occupations Service Workers Percent of workers employed in service and support occupations Vehicles Average number of vehicles per household Renters Share of households that rent within the rail precinct (2010) Urban Design Block Size Average block size in acres Intersection Density Number of 4-way intersections per square mile within rail precinct Percentage of Four-Way Intersection Percentage of four-way intersections Census Walkscore (station point) Walkscore rating at the station Walk Score Inc. Walkscore (station area average) Average Walkscore for the station area Walk Score Inc. Land Use Diversity Job/Pop Balance Index that measures balance between employmentand resident population within the rail precinct. Index ranges from 0, where only jobs or residents are present in a rail precinct not both, to 1 where the ratio of jobs to residents is optimal from the standpoint of trip generation. Values are intermediate when rail precincts have both jobs and residents, but one predominates. 1 Entropy (Land Mix) Another diversity index that captures the variety of land uses within the precinct. Entropy calculation based on net acreage in land-use categories likely to exchange trips. The entropy index varies in value from 0, where all developed land is in one of these categories, to 1, where developed land is evenly divided among these categories. Destination Accessibility CBD Distance Distance from the railway station to a central point within the central business district, as the crow flies. Calculated by Authors Transit Service/Mode Transit Frequency Aggregate frequency of transit service in station area per hour during the evening peak period Environmental Protection Agency Smart Location Database LRT/Streetcar Dummy variable of light rail or streetcar service at rail station Heavy Rail Dummy variable of heavy rail (ie. subway or metro) service at rail station Commuter Rail Bus Rapid Transit Ferry Notes: 1. JOBPOP = 1 - [ABS (employment * population)/(employment * population)]; ABS is the absolute value of the expression in parentheses. The value 0.2, representing a balance of employment and population, was found through trail and error to maximize the explanatory power of the variable (see Ewing et al. 2011). 2. All variables were transformed to the natural log so beta values repreesnt elasticities. Neighborhood Level Independent Variables (Level 1) Precinct level variables are divided into six categories: demographics and socioeconomics, development density, land use diversity, urban design, destination accessibility, and transit service/mode. Measures of demographics and socioeconomics used in this study include the share of nonwhite, Hispanics, measures of income, the share of professional and service workers, and housing tenure or the share of renters. Measures of density include population and employment density as well as the combination of both, which is listed as activity density. A study of long- term data around the globe indicated that a minimum activity density of 35 jobs or people per hectare,

6 Renne, Ewing and Hamidi 6 which equals approximately 7,000 people or jobs within a half- mile station precinct, where automobile dependence is significantly reduced (19). Intersection density measures walkable urban design and land use diversity variables include job/population balance and entropy (15, 16). Distance from the station to the central business district (CBD) is our destination accessibility measure and is often used as a proxy for accessibility to regional jobs (2). Finally, transit mode variables include dummy variables for light rail (LRT)/streetcar, heavy rail (subway and metro rail), commuter rail, bus rapid transit (BRT) and ferry service (not all were used in the final model). Regional Independent Variables (Level 2) This study presents new measures of regional network accessibility not found in previous studies. Since transit commuting involves access and egress, it makes sense to see how accessible the regional population is to the transit network. This study includes a measure of the total share of regional jobs located within all station precincts. It also includes a measure of the share of jobs plus population within station areas as a share of total jobs and population across the region. This study includes a regional sprawl index developed by Ewing and Hamidi (21) to see if the urban form of the region as a whole affects the share of transit commuting. We do not account, however, for the share of the regional population that is able to access the network via park- and- ride or transfer from another transit service, such as a feeder bus line, in regions with less accessibility to transit. Finally, this study utilizes data on the average annual delay per commuter at the regional level, reported by the Texas Transportation Institute s Urban Mobility Report. RESULTS This section discusses the overall fit of the best- fit MLM model as well as the significance and elasticity of each individual variable (see Table 2). Overall Model Fit The best- fit model chosen for this paper includes a pseudo R- squared value of (McFaddan s formulation). Unexplained variance can be categorized into two groups including those that we know about but cannot easily measure for this national database. Such variables include parking availability, fuel price, transit fare cost, impacts of transit passes, perception of cleanliness, crime and safety, and accessibility for non- drivers. The second category of unexplained variance arises in variables that we have not yet identified theoretically and thus may never be measured. It is important to note that since the data represent aggregated station areas (as opposed to individuals), we would expect to see a relatively high pseudo R- squared value. Nevertheless, we feel that the pseudo R- squared value derived in this model is an indicator of a strong model that can shed new light for planners and policy- makers. Best- Fit Model Results When examining the best- fit model results, all of the independent variables are significant at the 0.10, 0.05, or 0.01 probability levels. At the neighborhood level (Level 1), average household income, significant at the 0.10 level, indicates that a doubling of the average household income across the station area results in a 17.7% decline in transit commuting for the neighborhood. On the other hand, the results indicate that a doubling of the intensity of jobs and people within the neighborhood, significant at the 0.05, yields a 17.5% increase in transit commuting. Also significant at this same probability level, a doubling in the share of nonwhites across the station area yields a 14.4% premium for transit commuting. When examining the impact of transit technology, dummy variables representing light rail/streetcars and heavy rail/metro rail are both significant at the level and yield a bonus of

7 Renne, Ewing and Hamidi % and 30.6%, respectively. Also significant at the level, land use mix is captured by the jobs- population balance index, which indicates that neighborhoods with more balanced land uses exhibit higher shares of transit commuting. A doubling of the values in this index yields a 23.2% increase in the mode share for transit commuting. Two of the variables that depict walkability, each significant at the 0.05 level, include the percentage of four- way intersections and the average Walk Score for the station area. A doubling in each of these variables yields an increase in transit commuting by 9% and 27.6%, respectively. The last variable at the neighborhood level measures transit service frequency. Significant at the 0.01 level, a doubling in transit service frequency yields an 18.6% bonus for the mode share of transit commuting. The best- fit model also includes a significant variable at the regional level (Level 2). A doubling in the share of regional jobs and population within a half- mile of all stations in the region, also known as the network effect, yields a 38.6% increase in the mode share of transit commuting in station areas. This variable is significant at the 0.10 probability level. It is not surprising that the significance level is lower for this variable since many fewer degrees of freedom at this level than at the station- area level. Table 2: Best- Fit MLM Model Results Predicting Mode Share for Transit Commuting across Half- Mile Station Areas Level 1 Variables Description Coefficient p-value LNINCOME Natural Log of the average household income across the 1/2 mile station area LNINTENS Natural log of the Intensity (job + people per acre) across the 1/2 mile station area LNNONWHI Natural Log of the share of nonwhites across the 1/2 mile station area LRT_STRE Dummy variable of Light Rail and Streetcar HEAVY_RA Dummy Variable of Heavy and Metro Rail (not Commuter rail) LNJOB_PO Natural log of the jobs-population balance for the 1/2 mile station area LNPCTFOU Natural log of the percentage of 4-way intersections across the 1/2 mile station area LNAVGWAL Natural log of the average walk score for the 1/2 mile area around the station LNSERVFR Natural log of the transit service frequency Level 2 Variables LNREGION Natural log of the share of regional jobs and popualtion within 1/2 mile of all stations in the region Puesdo R-Squared Other Statistically Significant Variables The authors also conducted a number of other model runs. In all of the other runs, the average share of Hispanic residents at the neighborhood level was negatively related to the dependent variable. For a doubling in the share of Hispanic residents, transit commuting decreases by approximately 12%. At the regional level, the annual traffic delay per commuter yielded statistically significant impacts to the dependent variable. For each doubling in annual travel delay, the average share of transit commuting increased by 91% - 124%, depending upon the model. The reason this variable was not included in the final model was due to multicollinearity with the LNREGION variable.

8 Renne, Ewing and Hamidi 8 Non- significant Variables Some variables did not prove to be statistically significant. At the regional level, the sprawl index was not significant. This implies that development patterns outside the sphere of the network of station areas across the region, regardless of how well or poorly planned, have an insignificant impact on the mode share of transit commuting within station precincts. At the neighborhood level, the share of professional workers, distance to the CBD and entropy were not significant. This is not to say that these variables are not important, but after controlling for other variables, they do not reach conventional significance levels. DISCUSSION The findings of this study corroborate with other national studies that examine the impact of the built environment on transit usage. However, this is one of the first comprehensive national studies to include the vast majority of all fixed- transit stations across the United States. Because the study includes data at both the neighborhood- level and regional- level, this is one of the first studies to include a variable that measures the network benefits of access to jobs and population within walking distance of the network. Not surprisingly, the regional network variable of the share of jobs and population accessible to the fixed- transit network was correlated with traffic congestion in the region, another variable that was significant, but not chosen in the final model because planners and policy- makers typically work to encourage TODs around stations. (It would not make sense for planners to create traffic congestion with the goal of boosting transit commuting.) This study finds that the concepts of TOD, at the local and regional levels, make an important difference in determining if residents actually use transit for commuting or not. A recent study found that only 20% of stations across the United States achieve a minimum gross density of 15 units per acre (7,500 units within a half- mile of the station). Moreover, only six regions, including New York, San Francisco, Chicago, Philadelphia, Portland and Eugene have more than 25% of jobs accessible within walking distance of the fixed- transit network (18). As a nation, we have a tremendous opportunity to boost densities of people and jobs around existing rail infrastructure. Stakeholders at the local and regional levels each play an important part in ensuring that transit commuting is a viable option. Planners, local elected officials and developers that can boost the density of people and jobs within a half- mile of fixed- transit stations, create mixed- use and walkable projects can have an major difference on the transit mode share. For example, consider a neighborhood with express bus service and a 7.0% transit mode share that is able to double the land use intensity, jobs- population balance, the percentage of 4- way intersections and the Walk Score. That neighborhood would experience a boost in the mode share of transit commuting of 77.3% to a transit mode share of 12.4%. If the transit agency were to upgrade the express bus service to light rail and double the transit frequency, the mode share would increase from 12.4% to 17.7%. If this pattern were repeated across many stations in the region, resulting in a doubling in the network accessibility of the system (measured by the share of jobs and population within walking distance of all stations) the mode share in that neighborhood would rise to 24.2%, a level that mimics levels in many European cities. Other benefits not addressed here but currently under development by the authors is an examination of the ancillary walking and bicycling benefits associated with TOD and increased transit network connectivity. Future studies should also seek to capture variables in the model, such as cost and parking availability, where were not possible in this study.

9 Renne, Ewing and Hamidi 9 REFERENCES 1- Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: density, diversity, and design. Transportation Research Part D: Transport and Environment, 2(3), Ewing, R., & Cervero, R. (2010). Travel and the built environment: a meta- analysis. Journal of the American Planning Association, 76(3), Santos, A., McGuckin, N., Nakamoto, H., Gray, D., and Liss, S. (2011) Summary of Travel Trends: 2009 National Household Travel Survey. Washington, D.C.: U.S. Department of Transportation, Federal Highway Administration. Accessed 28 July 2014, Available at: 4- McKenzie, B. S. (2010) Public Transportation Usage Among U.S. Workers: 2008 and American Community Survey Briefs. Washington, D.C.: U.S. Department of Commerce, Economics and Statistics Administration, U.S. Census Bureau. Accessed 28 July 2014, Available at: 5- Schrank, D., Eisele, B. and Lomax, T. (2012) TTI s 2012 Urban Mobility Report: Powered by INRIX Traffic Data. College Station, Texas: Texas A&M Transportation Institute, Texas A&M University System. Accessed 28 July 2014, Available at: report pdf 6- Pisarski, A. (2006). Commuting in America III: The third national report on commuting patterns and trends. Washington, D.C.: National Cooperative Highway Research Program Report 550 & Transit Cooperative Research Program Report 110, Transportation Research Board. 7- Boarnet, M. G., & Crane, R. (2001). Travel by design the influence of urban form on travel. Oxford: Oxford University Press. 8- Von Thuned, J. (1826) Der Isolierte Staat in Beziehung ant Landswirtschaft and Nationalekomie. Hamburt. 9- Alonso, W. (1964) Location and land use. Toward a general theory of land rent. Cambridge, MA: Harvard University Press. 10- Muth, R. (1969) Cities and Housing. Chicago: University of Chicago Press. 11- Debrezion, G., Pels, E., & Rietveld, P. (2007) The impact of railway stations on residential and commercial property value: a meta- analysis. The Journal of Real Estate Finance and Economics, 35(2), Mohammad, S. I., Graham, D. J., Melo, P. C., & Anderson, R. J. (2013). A meta- analysis of the impact of rail projects on land and property values.transportation Research Part A: Policy and Practice, 50, Bartholomew, K., & Ewing, R. (2011). Hedonic price effects of pedestrian- and transit- oriented development. Journal of Planning Literature, 26(1), Seskin, S., Cervero, R., and Zupan, J. (1996) Transit and Urban Form, Volume 1. Washington, D.C.: Transit Cooperative Research program 16, Transportation Research Board. Accessed 28 July 2014, Available at:

10 Renne, Ewing and Hamidi Ewing, R., Greenwald, M., Zhang, M., Walters, J., Feldman, M., Cervero, R., Frank, L., & Thomas, J. (2011). Traffic Generated by Mixed- Use Developments A Six- Region Study Using Consistent Built Environmental Measures. Journal of the Urban Planning and Development. 137(3), Ewing, R., Greenwald, M. J., Zhang, M., Bogaerts, M., & Greene, W. (2013). Predicting transportation outcomes for LEED projects. Journal of Planning Education and Research, X Chatman, D. G. (2013). Does TOD need the T? On the importance of factors other than rail access. Journal of the American Planning Association, 79(1), Renne, J. L. (2013). Transport Beyond Oil: Policy Choices for a Multimodal Future. Island Press. 19- Newman, P., & Kenworthy, J. (2006). Urban design to reduce automobile dependence. Opolis, 2(1). 20- Ewing, R. & Hamidi, S Measuring Urban Sprawl and Validating Sprawl Measures, Technical Report Prepared for the National Cancer Institute, National Institutes of Health, the Ford Foundation, and Smart Growth America. sprawl Accessed July 28, 2014.