17

Does TOD Need the T?

On the Importance of Factors Other Than Rail Access
Daniel G. Chatman

Problem, research strategy, and ndings: Transit-oriented developments (TODs) often consist of new housing near rail stations. Channeling urban growth into such developments is intended in part to reduce the climate change, pollution, and congestion caused by driving. But new housing might be expected to attract more afuent households that drive more, and rail access might have smaller effects on auto ownership and use than housing tenure and size, parking availability, and the neighborhood and subregional built environments. I surveyed households in northern New Jersey living within two miles of 10 rail stations about their housing age and type, access to off-street parking, work and non-work travel patterns, demographics, and reasons for choosing their neighborhoods. The survey data were geocoded and joined to on-street parking data from a eld survey, along with neighborhood and subregional built environment measures. I analyzed how these factors were correlated with automobile ownership and use as reported in the survey. Auto ownership, commuting, and grocery trip frequency were substantially lower among households living in new housing near rail stations compared to those in new households farther away. But rail access does little to explain this fact. Housing type and tenure, local and subregional density, bus service, and particularly off- and on-street parking availability, play a much more important role. Takeaway for practice: Transportation and land use planners should broaden their efforts to develop dense, mixed-use, lowparking housing beyond rail station areas. This could be both more inuential and less

ransit-oriented development (TOD) is a common urban planning strategy; in practice, it often means developing new housing near rail stations. The term TOD can refer to buildings near transit, clusters of buildings near transit, or larger areas of up to a half-mile radius around a rail stop that are high-density and mixed-use, with walk-accessible shopping, pedestrian amenities, lower parking supply, and physical designs that are thought to encourage households to walk, bicycle, and take transit instead of driving (e.g., Belzer & Autler, 2002; Calthorpe, 1993). One of the main objectives of TOD policies is to reduce the regional and global environmental impacts of auto use. Pursuing environmental goals through TOD has two important premises: rst, that households occupying newly constructed housing units near rail stations drive less than those in older housing near rail or those living farther from rail; and second, that the proximity to rail, as opposed to other attributes of TOD, is a critical part of the equation. There are reasons to doubt these premises. New housing might attract more afuent residents who drive more than those living in older housing near rail. Higher development density, less parking, and the presence of more shops and services nearby could all induce households to drive less, with or without rail access. While studies have long found that households living near rail stations have substantially higher rates of transit use, particularly rail ridership (see review in Cervero, Ferrell, & Murphy, 2002), there are fewer studies of whether those households also own and use personal vehicles less. A study of selected transit-oriented housing developments in California in 2003 found that 72% of survey respondents commuted in personal vehicles, lower than the Census rate for surrounding cities of 90% in 1999 (Lund, Cervero, & Willson, 2004). A study of 17 transit-oriented developments in four U.S. urban areas, using vehicle counters in driveways, found 44% fewer vehicle
expensive than a development policy oriented around rail. Keywords: transit-oriented development, rail transit, auto use, parking, sustainability Research support: Data collection and initial research were funded under contract with the New Jersey Department of Transportation. About the author: Daniel G. Chatman ([email protected]) is assistant professor of city and regional planning at the University of California, Berkeley.
Journal of the American Planning Association, Vol. 79, No. 1, Winter 2013 DOI 10.1080/01944363.2013.791008 American Planning Association, Chicago, IL.

18

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

trips than the published rates in the Institute of Transportation Engineers manual (Arrington & Cervero, 2008). Because neither of these studies included a control group, the magnitude of the reported differences may not be generalizable. The nature of non-response to the TOD survey, the use of a different survey instrument, and the timing of the survey (a four-year difference) could all inuence the lower observed auto use in comparison to Census rates; and lower vehicle trip counts in comparison to the Institute of Transportation Engineers estimates could be partly because those estimates are inated (Shoup, 2005). Well-controlled statistical studies about the impacts on auto travel of the built environment are relevant because they control for many of the factors that comprise TOD. However, compared to built environment factors like population density, there are relatively few studies that include rail or transit access. A recent meta-analysis of more than 200 studies in the built environment-travel literature found just six studies at the household or individual level that used vehicle distance traveled as a dependent variable along with distance to rail or bus as an independent variable (Ewing & Cervero, 2010). The average elasticity of vehicle use with respect to transit proximity was very small, at 0.05, and likely not statistically signicant. Some research has found that rail access has either little association or a positive relationship with auto ownership or use. A study of San Diego and the San Francisco Bay Area found that proximity to heavy rail was associated with higher vehicle miles traveled when controlling for a large set of neighborhood-level built environment features (Chatman, 2008), and a study of Manhattan and Hong Kong found that rail station ridership was positively associated with the auto ownership of households living nearby (Loo, Chen, & Chan, 2010). A study of 370 metropolitan areas in the United States using structural equation modeling found that rail access was only weakly associated with auto distance traveled per capita (Cervero & Murakami, 2010). A simulation model conducted for Austin (TX) estimated that there was very little change in travel mode associated with increasing the share of new development near rail stations, although projected vehicle mileage was lower because auto trip distances were shortened (Zhang, 2010). A slightly larger set of studies has found that rail access is associated with lower auto use. A study of both commute mode and auto distance traveled using data from a subset of 114 U.S. metropolitan areas in the National Household Travel Survey found that rail access, bus access, and urban form were all associated with lower auto use (Bento,

Cropper, Mobarak, & Vinha, 2005). Another study of National Household Travel Survey data at the national level, using structural equations, found that rail accessibility, measured in terms of walking distance, was associated with lower vehicle miles traveled, both directly, presumably by substituting for auto use, and indirectly, via an association with higher population density (Bailey, Mokhtarian, & Little, 2008). A study of travel diary data from New York City found that subway lines near home and work were correlated with lower auto use and more walking, while noting that subway lines might be a proxy for walkable neighborhoods (Salon, 2009). Two international studies also found the expected relationship. A study of Santiago de Chile found that distance to urban rail stations was associated with higher levels of auto commuting, primarily via a direct effect on mode choice rather than any strong effect on auto ownership (Zegras, 2010). A study of national data from Germany, focusing on licensed drivers owning cars, found that walking distance to transit was highly correlated with vehicle distance traveled (Vance & Hedel, 2007). An important missing factor in all of the above studies is the availability of vehicle parking. Off- and on-street parking has been studied even less than rail access, largely because data are not readily available. A case study of two neighborhoods in New York City argued that differences among them in auto use were likely caused by parking availability and not by transit access, highway access, or demographics (Weinberger, Seaman, & Johnson, 2009). A Census tract level study of New York data from 1998 found that both transit accessibility and an imputed measure of off-street parking availability were positively associated with auto commuting to Manhattan (Weinberger, 2012). A recent New York study, using the same dataset, restricted to units for which Google observations of parking could be made, found that both subway distance and off-street parking supply were signicant predictors of auto ownership (Guo, 2013). Studies of how auto use might be affected by on-street parking availability are even scarcer; one study shows that that street cleaning requirements in New York City are associated with more driving for households without off-street parking, and less driving for housing units with it (Guo & Xu, 2012). Almost all of these studies have limited applicability to the research question here because they omit potentially important covariates of rail access. In addition to parking availability, these include neighborhood scale and subregional built environment measures, and the age and type of housing. Few of them test for the importance of being within walking distance of rail.

Chatman: Does TOD Need the T?

19

Study Design
I conducted a mail survey of households within a two-mile radius of 10 rail stations in New Jersey, some of them living in purpose-built TODs as well as those living in new and older housing nearby and farther away from rail. I selected two-mile radius areas, rather than sampling the entire state, in order to balance the need to control for spatially correlated inuences on auto use with the need to observe travel behavior near and far from rail stops. Since transit use tends to drop off signicantly beyond a half mile from the nearest transit stop (e.g., Dill, 2003; Pushkarev & Zupan, 1977), and since TOD is dened as being within walking distance of rail, households outside walking distance serve as controls. Restricting the sample frame to 10 station areas made it possible to collect on-street parking data for many of the respondents. These consisted of on-foot observations of on-street parking supply and use for a quarter-mile airline radius around the 10 stations. The analysis dataset was constructed by merging household survey and on-street parking data, then joining to that dataset neighborhood and subregional spatial measures constructed near respondent households using secondary data sources in a geographical information system. Only households nearest the rail stations had observations of on-street parking supply. These data assembly stages are described briey below; more details are available elsewhere (Chatman & DiPetrillo, 2010). The stations selected were Morristown and South Orange on the Morris & Essex Line, Perth Amboy and South Amboy on the North Jersey Coast Line, Rahway and Trenton on the Northeast Corridor Line, Westeld and Cranford on the Raritan Valley Line, and 2nd Street and Essex stations on the Hudson-Bergen Light Rail line (Figure 1). These stations provide excellent access to downtown Manhattan and can be characterized as a mix of light rail, heavy rail, and high-frequency commuter rail with very good transit accessibility. The two-mile-radius area around these 10 stations includes about 740,000 people, or about 9% of the population of New Jersey, with generally better transit access and higher population density than the remainder of the state. I constructed a sample of 5,000 housing units, including 1,073 units in recently built or substantially renovated multifamily housing developments within walking distance of the stations. The remainder of the sample was drawn from a list of households based on U.S. postal service addresses in zip codes within two miles of the stations. This list was geocoded, and I randomly sampled 2,427 housing units within a quarter-mile airline

distance from the stations and an additional 1,500 units between a quarter mile and two miles away. The survey questionnaire focused on housing unit characteristics, on- and off-street parking, work and non-work travel, household characteristics, and residential location criteria (see Chatman & DiPetrillo, 2010). The questionnaire was pretested, and revised, prior to elding from June 3 to August 26, 2009. Five recruitment mailings were sent: an invitation letter with questionnaire, a reminder postcard, two subsequent letters with replacement questionnaires to non-respondents, and a nal last chance contact letter, in a modied version of the Dillman total design method mail survey protocol (Dillman, 1978; Dillman, Dillman, & Makela, 1984). In total, 1,143 completed surveys were received, for a response rate of 25.4%. See Table 1 for a summary of data from the survey. On-street parking observations were recorded for blocks tting at least 50% within a quarter-mile airline buffer of the stations. Blocks were equally divided among three trained student surveyors. Field workers observed on foot during the evening peak parking period, between 5 p.m. and 8:30 p.m., collecting data on the number of on-street spaces by type (marked and unmarked), whether the spaces were occupied, parking duration limitations, space type (including limitations for disabled use and other permit holders), time restrictions, street cleaning, and noparking periods, for 6,237 parking spaces on 818 street segments. The parking data were collected a year prior to the household survey (the delay was due to an interruption in research funding). The parking observations were merged with a street segment map and later aggregated in a GIS to construct measures of overnight parking spaces per road mile for a quarter-mile radius around the homes of the 532 households living within a quarter-mile airline distance of the stations. The population density in Census blocks within a quarter mile of each respondents home was calculated from data on population and land area of the blocks from the 2000 Census, using GIS. Local retail and total employment density were similarly calculated using the Census Bureaus 2008 Longitudinal Employer-Household Dynamics dataset (U.S. Census Bureau, 2008). Data on grocery stores, using NAICS code 445110, were downloaded from referenceusa. com, geocoded at the address level, and aggregated to the quarter-mile radius around respondent homes. The density of bus stops within a mile of home was calculated using bus stop locations from NJ Transit provided as of 2010. Network distance to the Manhattan central business district (CBD), dened as the nearer of Grand Central Station or Penn Station, was calculated using a street le and network

20

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

Figure 1. Selected stations with two-mile and quarter-mile buffers.

Chatman: Does TOD Need the T?

21

Table 1. Descriptive statistics (selected variables). Variable Distance to nearest rail station (miles) New housing near raila Older housing near rail Older housing farther from rail Less than one off-street parking space per adult in household On-street overnight parking spaces (100s) per road mile within mile Scarce on- and off-street parkingb On-street parking not observed Duplex or triplex Rowhouse or townhouse Apartment or condominium Other housing unit type Missing housing unit information Rental unit Home owned without mortgage Unknown unit tenure (owned or rented) Population per square mile (000s) in Census blocks within 18 mile Employment per square mile (000s) in Census blocks within mile Retail employment per square mile (000s) in Census blocks within mile Bus stops, 1-mile radius Subregional employment density (000s per square mile in home PUMA) Subregional bus stop density (10s per square mile in home PUMA) Network distance to Manhattan CBD (miles, from home) Household income ($10,000s, coded at category midpoints) Household income not reported Household size Children in household Single-parent household Hispanic African American Asian American Native American Race not reported Full-time worker Part-time worker Worker in management occupation Worker in nancial occupation Worker in sales occupation Worker in clerical occupation Worker in craftsman occupation Worker in laborer occupation Worker in service occupation Worker in unknown occupation (not reported) Retired Chose neighborhood based on access to friends/family Chose neighborhood based on access to leisure opportunities Chose neighborhood based on access to job Chose neighborhood based on access to transit Chose neighborhood based on access to childrens schools Chose neighborhood based on quality of public services Chose neighborhood based on design Chose neighborhood based on distance to school Chose neighborhood based on distance to shops Chose neighborhood based on distance to highway Chose neighborhood based on house characteristics Chose neighborhood based on other characteristics Obs 1,143 1,143 1,143 1,143 1,089 532 508 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,133 1,143 1,143 1,143 1,143 1,143 1,143 1,031 1,143 1,141 1,131 1,131 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143 1,143

Mean
0.63 0.16 0.33 0.38 0.34 1.67 0.15 0.53 0.08 0.08 0.51 0.01 0.01 0.37 0.13 0.02 12.6 8.5 0.5 103.7 4.1 3.8 21.2 11.6 0.10 2.3 0.24 0.03 0.14 0.13 0.06 0.01 0.04 0.71 0.07 0.12 0.08 0.06 0.04 0.02 0.02 0.05 0.02 0.17 0.31 0.11 0.46 0.42 0.16 0.02 0.28 0.05 0.18 0.09 0.22 0.15

SD
0.60 0.37 0.47 0.49 0.47 0.67 0.36 0.50 0.27 0.27 0.50 0.08 0.08 0.48 0.34 0.15 12.2 14.7 0.5 118.7 5.5 6.0 12.1 8.4 0.30 1.3 0.43 0.17 0.34 0.34 0.24 0.10 0.19 0.45 0.26 0.33 0.27 0.23 0.20 0.15 0.15 0.21 0.14 0.38 0.46 0.31 0.50 0.49 0.37 0.15 0.45 0.23 0.39 0.29 0.41 0.36

Min
0.03

Max
3.38 ind. var. ind. var. ind. var. ind. var. 3.02 ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. 87.6 89.6 4.8 622 19.6 23.7 58.1 32.5 ind. var. 9 ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var. ind. var.

0.42

0.13 0 0 0 0.40 0.31 2.50 0.50 1

Notes: ind.var. indicator (01) variable. a. New housing dened as seven or fewer years old at the time of the survey. Near rail is within walking distance, dened as 0.4 miles measured along the road network. b. Scarce on- and off-street parking dened as having less than the median value for on-street parking space availability and less than one off-street parking space per adult in the household.

22

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

analysis routine in a GIS. Subregional measures of population density, employment density, and bus stop density were created with the 20052007 pooled American Community Survey Public Use Microdata Sample for the Public Use Microdata Areas (PUMAs) within which the households lived. I constructed residential location criteria variables using answers to the question, Please rate the top three factors that attracted you to this neighborhood. A dummy variable was set equal to 1 for any of a dozen such factors ranked by a respondent, regardless of rank value. I set an indicator of off-street parking scarcity equal to 1 if the respondent reported having less than one off-street parking space per adult in the household, and 0 otherwise. I also constructed a variable representing the interaction between on- and off-street parking. If there is little offstreet parking but ample on-street parking, or if there is plenty of off-street parking but no parking on the street, there should be no difculty in parking a car. The variable was set equal to 1 if the household had less than one offstreet parking space per adult and if on-street overnight parking availability was below the observed median value of 138 overnight parking spaces per road mile. In the data description and analysis, I distinguish new from older units, and those within walking distance to rail from those farther away. New housing was dened as housing that had been built within seven years of the survey, based on respondent reports as well as independently collected information about selected buildings near the stations.1 I dened walking distance as being within 0.4 miles of any rail station, as measured along the local street network, along which sidewalks were universally available in the study area. This is a bit shorter than Calthorpes (1993) 2,000-foot denition of walking distance for TODs. For most houses, it was roughly equivalent to a quarter-mile airline distance. Table 1 shows means and standard deviations for the main variables used in the analysis.
Table 2. Auto ownership and use by age of housing and distance to rail. Vehicles per household 1.14 ** 1.40 ** 1.77 1.67 1,118

Observed Differences by Rail Distance and Housing Age

Respondents living in new housing within walking distance of rail stations reported lower auto ownership, less auto commuting, and fewer weekly personal vehicle grocery trips than those living in new or older housing farther away (Table 2). They also had a lower rate of auto commuting and grocery trip frequency than those living in older housing near rail, a remarkable result given that this group also reported substantially higher household income. A number of factors associated with proximity to rail and age of housing may play a role in inuencing auto ownership and use. Both rental housing and smaller housing units may attract households who use autos less because they are younger, of lower income, and have fewer children. In these areas, new housing near rail is much more likely to be for rent, and almost all consists of smaller units; in fact, even new housing farther from rail is much more likely to consist of smaller units (Table 3, columns 1 and 2). Off-street parking availability is lower in new housing near rail than in housing farther from rail, although newer units have more on-street parking available to them (Table 3, columns 3 and 4). Although a higher share of older housing near rail has combined low on- and off-street parking, the difference is not statistically signicant (Table 3, column 5). The larger neighborhood spatial context could also play a role. Population density for both new housing and old housing near rail, and, notably, for older housing farther from rail, is much higher than for new housing farther from rail (Table 3, column 6). New housing near rail averages more than 150 bus stops within a mile, which is much higher than the other subgroups (Table 3, column 7). There are other possible explanations for the observed lower auto ownership and use of residents of new housing

Subgroupa New housing near rail Older housing near rail Older housing farther from rail New housing farther from rail Complete responses

Vehicles per adult 0.73 * 0.81 * 0.86 * 0.96 1,118

Commuted via SOV (indicator variable) 0.36 ** 0.59 0.67 0.63 810

Grocery trips via auto, per week 1.47 ** 1.84 ** 2.44 2.45 878

Notes: SOV singly occupied vehicle. a. New housing is seven or fewer years old at the time of the survey. Near rail is within walking distance, dened as 0.4 miles measured along road network. * Statistically signicant difference from new housing farther from rail at the 95% level. **Value is also signicantly different from the value for the category below it, at the 95% level.

Chatman: Does TOD Need the T?

23

Table 3. Housing, parking, and spatial characteristics by age of housing and distance to rail. 1 2 Apartment/ condo/ townhouse/ rowhouse 0.98 ** 0.62 ** 0.37 ** 0.71 1,135 3 Scarce off-street parkingb 0.47 * 0.39 ** 0.30 ** 0.19 1,089 4 On-street parking per road mile 193 ** 152 [183] [149] 532 *d
d

5 Low onand off-street parkingc 0.12 0.17 [0.07] d [0.25] d 508

6 Population density (000s per square mile, 1 -mile radius) 8 13,200 * 12,800 * 13,400 * 7,810 1,143

7 Bus stops (1-mile radius) 152 ** 93 101 *

Subgroupa New housing near rail Older housing near rail Older housing farther from rail New housing farther from rail Complete responses

Rental unit 0.57 ** 0.48 ** 0.29 ** 0.16 1,116

79 1,143

Notes: a. New housing dened as seven or fewer years old at the time of the survey. Near rail is within walking distance, dened as 0.4 miles measured along road network. b. Off-street parking scarcity dened as less than one off street space per adult in the household. c. Below median on-street parking + less than one off-street parking space per adult (see text). d. Brackets denote very small subsample sizes. On-street parking data was gathered primarily for housing units within walking distance of rail. *Statistically signicant difference from new housing farther from rail at the 95% level. **Value is also signicantly different from the value for the category below it, at the 95% level.

near rail, but for these, data are harder to come by. For example, perhaps recent movers to TODs optimize their commutes around transit in the short run, but in later years as their work locations shift, they begin to drive. It is also possible that changing lifestyle preferences among younger people explain some of the correlation of new TOD housing and lower auto use, or that shifts in the housing and labor markets, and the recent economic downturn, are more keenly felt by those recent movers who are more likely to save money by owning and using autos less. To investigate some of these potential explanations, I carried out a series of multivariate regressions for auto ownership, auto commuting, and auto grocery trip frequency.2 For each of the three measures I rst carried out a regression with only rail proximity and age of housing. In the second regression I added other housing unit, parking, and spatial characteristics; in the third, I added demographic characteristics and residential choice criteria.3 Different houses and neighborhoods may attract households with different levels of and preferences for auto ownership and use. The second model in each of the tables implicitly includes these residential choice effects, while the third model is meant to estimate effects independent of those choices. The variation in coefcients denotes a range depending on how much of the effects associated with preferences and residential choice can be expected to occur in the future. The fourth model consists of a regression restricted to households within walking distance of a rail station, to test for the interaction of rail proximity and

other factors such as parking availability. Finally, for auto commuting and grocery trip frequency, I carried out a fth model including auto ownership as an (endogenous) explanatory variable, as explained below.

Auto Ownership
I dened per capita auto ownership as the number of reported vehicles divided by the number of adults in the household. In the rst model, per capita auto ownership was regressed on distance to rail and the housing age and walking distance threshold variables, using ordinary least squares. Each additional mile from a rail station is associated with an additional 0.09 vehicles per adult in the household (Table 4, column 1). Older housing, whether within walking distance of a rail station or farther away, is associated with fewer cars per capita (the omitted category is new housing outside walking distance). The coefcients together suggest that new housing near rail is associated with 27% lower per capita auto ownership than new housing farther away. The correlation of vehicle ownership with both rail proximity and housing age markedly decreased when housing, parking and built environment measures were controlled (Table 4, column 2). Neither rail proximity nor housing age is a statistically signicant predictor of per capita auto ownership, and, in fact, the coefcient on new housing near rail turns positive. Off-street parking scarcity, and low on- and off-street parking availability, are among the most powerful variables in this model. Houses with fewer than one off-street parking space per adult have

24

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

Table 4. Vehicles per adult in household as a function of distance to rail and other factors (OLS regressions). 1 Housing age and distance to rail Distance to rail (miles) New housing near raila Older housing near rail Older housing farther from rail Scarce off-street parking On-street overnight parking spaces Scarce on- and off-street parking Apartment/condo/row\townhouse Unit type unknown Rental unit Job density, mile (000s) Bus stops, 1-mile radius Household income ($10,000s) Owned home without mortage Household size Single-parent household Hispanic African American Service occupation Neighborhood choice: friends Neighborhood choice: leisure Neighborhood choice: access to job Neighborhood choice: near transit Neighborhood choice: public services Neighborhood choice: looks/design Neighborhood choice: near school Neighborhood choice: near highway Constant Observations Adjusted R2 0.9 1118 0.0245 *** 1.11 1071 0.1871 *** 0.091 *** 0.18 *** 0.11 ** 0.14 *** 2 Add housing, parking, and spatial variables 0.0034 0.01 0.029 0.048 0.16 0.011 0.13 0.35 0.13 0.0023 0.0008 *** *** ** 0.065 * *** 3 Add demographics and preferences 0.018 0.045 0.0017 0.019 0.11 0.0077 0.11 0.13 0.4 0.1 * *** * *** *** 0.12 0.011 0.24 0.027 0.23 0.15 0.0013 0.0004 *** *** ** 4 Near-station households; same variables as Model 2 0.16 0.041

0.003 ** 0.0007 ** 0.006 *** 0.074 * 0.065 *** 0.29 0.07 0.16 0.1 *** * *** ** 0.075 **

0.055 ** 0.051 * 0.098 *** 0.2 0.13 0.11 1.03 1063 0.2776 ** ** *** *** 1.23 *** 525 0.1644 0.081 ***

Notes: Included, statistically insignicant, not shown: [Models 24] duplex/triplex, unit type missing, tenure unknown, population density (1 8 mile), retail employment density ( mile), distance to Manhattan central business district, subregional bus stop density, subregional employment density; [Model 3] household income missing, children in household, Asian American, Native American, race unknown, occupation indicator variables (management, nancial, sales, clerical, craft, labor, unknown), full-time worker, part-time worker, retired, neighborhood choice criteria indicator variables (school district, near shops/services, house characteristics, other). a. New housing is seven or fewer years old at the time of the survey. Near rail is within walking distance, dened as 0.4 miles measured along road network. * p .10 ** p .05 *** p .01

0.16 fewer vehicles per adult, all else equal, while those with both low on- and off-street parking availability have an additional reduction of 0.13 vehicles per adult. Rental housing is also associated with 0.065 fewer vehicles per

adult. Of the built environment variables, the most signicant is the number of bus stops within a mile of the home. The coefcient of 0.0008 implies that a one-standard-deviation increase in bus service (the

Chatman: Does TOD Need the T?

25

equivalent of 118 bus stops in the mile radius around home) is associated with 0.09 fewer vehicles per adult. The third model in this set adds in additional controls for demographics and preferences of households, accounting both for the fact that TODs may attract previous transit users as well as the fact that they may enable households moving in to use alternative modes more (Table 4, Model 3). A number of coefcients on the newly entered demographic and preference variables are large and signicant in this model, but I focus on the housing unit and spatial characteristics, as they are the most policy relevant. The distance from rail coefcients remain insignicant and small. The coefcients on off-street parking scarcity and the combination of low onand off-street parking are reduced from 0.16 to 0.11 and from 0.13 to 0.11 vehicles per adult respectively, but remain substantive, each representing a 13% reduction in auto ownership at the mean. The coefcient on townhomes and apartments doubles, from -0.065 to -0.13; the increase appears to be due to household size being controlled, since larger households have fewer cars per adult. Townhomes and apartments might also have off-street parking that is farther from the unit. In short, this model suggests that sorting by income, household size, and housing preferences apparently does explain a signicant share of the correlation of auto ownership with on- and off-street parking availability, the tenure and type of unit, bus access, and job density, but those measures remain signicantly associated with lower auto ownership, in marked contrast to rail proximity. Limiting the analysis to households near stations provides a test of how rail access may interact with other factors (Table 4, column 4). Low on- and off-street parking availability apparently has stronger effects combined with rail station proximity: there are 0.24 fewer vehicles per capita when the analysis is restricted to near-station households, almost double the relationship in Model 2.

Auto Commuting
Of the dataset of 1,134 respondents, 810 reported that they worked part or full time in the previous week, and of those, all reported their commute mode. A logit model of the decision to commute by auto (singly occupied vehicle) is presented in Table 5. Exponentiated coefcients, or odds ratios, are shown; the increment greater or less than 1 can be interpreted as a percentage change in the probability of auto commuting. Before controlling for non-rail factors, each mile from a rail station is associated with a 74% increase in the odds of commuting via auto, and households living in new housing within walking distance of a rail station are only

43% as likely to commute via auto compared to households in new housing farther away (Table 5, column 1). New and old housing are statistically indistinguishable from each other in this initial model. When housing unit, parking availability, and built environment variables are introduced (Table 5, column 2), the effect on auto commuting of being within walking distance of rail vanishes entirely, while the continuous distance-to-rail coefcient shrinks from 1.72 to 1.32 and becomes statistically insignicant. Off-street parking, job density, subregional bus stop density, and distance to downtown are all highly associated with auto commuting. Households living in older housing are more likely to commute via car when controlling for housing, parking, and built environment factors. Since all households living in new housing have recently moved, those occupying older housing are perhaps more likely to have experienced changes in the location of work or other chained activity locations since their last move, and driving to work may have become a more attractive choice. When controlling for demographic characteristics and residential location criteria, the positive association between older housing and auto commuting loses statistical signicance, although it remains relatively large in magnitude (Table 5, column 3). Having scarce off-street parking remains very signicantly associated with lower probability of commuting via auto, with the odds decreasing from 63% to 57%. Rail access becomes more insignicant still. The fourth auto commuting model is restricted to commuters within walking distance of rail to test for interactions between the presence of rail and other factors (Table 5, column 4). Households in new housing are less likely to commute via auto in this model, consistent with Model 2. While off-street parking is no longer independently signicant, near-station households with both low on- and offstreet parking commute by auto just 40% as much as other households. Few of the remaining variables in Model 2 are signicant, with the exception of local population density. Finally, I estimated an auto commuting model like Model 2 but with the addition of a single explanatory variable, the number of vehicles per adult. Since auto ownership is intimately tied to the commuting decision, adding it will tend to bias the coefcient estimates for the other independent variables. But it does illustrate how parking supply, housing characteristics, and transit proximity are directly correlated with auto commuting and indirectly correlated via auto ownership. The number of vehicles per adult has an odds ratio of 7.59 while off-street parking loses statistical signicance, suggesting that its effects on auto commuting are felt primarily via the auto ownership link (Table 5, column 5).

26

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

Table 5. Probability of commuting by singly occupied vehicle as a function of distance to rail and other factors (logit regressions). 1 2 Add housing, parking, and spatial variables 1.34 1.00 1.68 * 1.79 ** 0.63 ** 1.30 0.60 5.71 *
1 8

4 Nearstation HHs only, same variables as Model 2 2.83 0.61 *

5 All HHs, add vehicles per adult to Model 2 1.22 1.02 1.83 * 1.93 **

Housing age and distance to rail Distance to rail (miles) New housing near raila Older housing near rail Older housing farther from rail Scarce off-street parking On-street overnight parking spaces Scarce on- and off-street parking Tenure unknown Population density, mile (000s) Job density, mile (000s) Subregional bus stop density (10s) Distance to downtown (mile) Household income $25,000 Race unknown Labor occupation Neighborhood choice: leisure Neighborhood choice: access to job Neighborhood choice: near transit Neighborhood choice: school district Neighborhood choice: near school Neighborhood choice: near highway Neighborhood choice: other Vehicles per adult in household Observations Pseudo R2 810 0.0446 1.74 *** 0.43 *** 1.06 1.00

Add demographics and preferences 1.20 1.00 1.41 1.61 0.57 ** 1.10 0.62 6.60 * 0.99 0.99 * 0.95 ** 1.02 2.43 * 0.35 * 3.12 ** 3.26 *** 2.06 *** 0.39 *** 1.75 ** 2.70 ** 1.96 ** 1.68 *

0.85 1.13 0.40 ** 2.89 0.97 ** 0.99 0.97 1.03

0.83 1.51 0.75 7.64 ** 0.98 0.99 0.97 1.03 **

0.98 ** 0.99 * 0.95 * 1.02 **

7.59 *** 785 0.121 782 0.2239 400 0.1296 773 0.1805

Notes: Included, statistically insignicant, not shown: [Models 25] on-street parking not observed, housing type dummy variables (duplex/triplex, apartment/condominium/rowhouse/townhouse, mobile home, other home, unit type unknown), rental unit, retail employment density (-mile); [Model 3] household income, household income missing, owned home without mortage, household size, children in household, single-parent household, Hispanic, African American, Asian American, Native American, occupation dummy variables (management, nancial, sales, clerical, craft, service, unknown), part-time worker, neighborhood choice criteria dummy variables (friends, public services, looks/design, house important). a. New housing is seven or fewer years old at the time of the survey. Near rail is within walking distance, dened as 0.4 miles measured along road network. Exponentiated coefcients. * p .10 ** p .05 *** p .01

Grocery Auto Trip Frequency

Rail access could directly and indirectly reduce driving to the grocery store by reducing auto ownership; by lowering the rate of auto commuting, and subsequent autobased grocery trips chained into those commutes, or by

encouraging the use of rail for the grocery trip itself. In the most recent National Household Transportation Survey, the category grocery/hardware/clothes shopping was the most common trip purpose, exceeding even commute trips in frequency (Federal Highway Administration, 2009). Grocery trips may be among the most routine because food

Chatman: Does TOD Need the T?

27

is a basic necessity; they may, therefore, be relatively easily to remember and report accurately. I constructed a measure of weekly auto-based grocery trip frequency using answers to a question about the timing and mode of the last three grocery trips, and dividing the weeks elapsed since the longest-ago reported grocery trip by the number of those trips that were conducted via a personal vehicle, either singly or jointly occupied. The variable was constructed only for the 878 respondents (77% of the pool) who reported full information on at least two grocery trips. I estimated these regressions using ordinary least squares. The variable is continuous, ranging from 0 (in about 5% of cases) to as

high as 10.5 trips per week, with a mean of 2.07 trips per week. The initial regression found an additional 0.51 autobased grocery trips per week for every mile farther from a rail station, while new housing near rail has 0.73 fewer such trips than other new housing (Table 6, column 1). When controlling for parking supply, housing, and built environment characteristics, the signicance of being within walking distance of rail and of housing age both disappear, although the distance-to-rail variable coefcient remains statistically signicant as it decreases in size (Table 6, column 2). Each additional grocery store within a quarter mile of home is associated with a reduction of

Table 6. Weekly auto grocery trips as a function of distance to rail and other factors (OLS regressions). 1 Housing age and distance to rail Distance to rail (miles) New housing near raila Older housing near rail Older housing farther from rail Scarce off-street parking On-street overnight parking spaces Scarce on- and off-street parking On-street parking not observed Grocery stores,
1 4

2 Add housing, parking, and spatial variables 0.33 0.011 0.099 0.14 0.2 0.14 0.57 0.08 0.098 *** 0.0023 ** 0.07 ** 0.077 *** 0.034 *** ** ***

3 Add demographics and preferences 0.28 0.065 0.25 0.22 0.13 0.16 0.48 0.04 0.11 0.0014 0.045 0.057 *** 0.03 0.41 0.31 *** ** * 0.013 * *** * **

4 Near-station HHs only; same variables as Model 2 0.6 0.053

5 All HHs, add vehicles per adult to Model 2 0.33 0.059 0.081 0.13 ***

0.51 0.73 0.39 0.22

*** *** **

0.16 0.094 0.6 0.14 0.14 0.0001 0.014 0.068 0.013 *** **

0.22 0.14 0.45 0.11 0.097 *** 0.0026 ** 0.068 ** 0.074 *** 0.035 *** *

mile

Bus stops, 1 mile radius Job density, subregion (000s) Bus stop density, subregion (10s) Distance to downtown (miles) Household income ($10,000s) Full-time worker Neighborhood choice: school district Vehicles per adult in household Constant Observations Adjusted R 2 2.09 878 0.0757 ***

0.4 3.42 855 0.1614 *** 3.99 851 0.1662 *** 2.84 428 0.1342 *** 2.98 843 0.1687

*** ***

Notes: Included, statistically insignicant, not shown: Housing type dummy variables (duplex/triplex, apartment/condominium/rowhouse/townhouse, mobile home, other home, unit type unknown), housing tenure (rental unit, tenure unknown), population density (18 mile), employment density ( mile), retail employment density ( mile), household income missing, owned home without mortage, household size, children in household, singleparent household, Hispanic, African American, Asian American, Native American, race/ethnicity unknown, occupation dummy variables (management, nancial, sales, clerical, craft, labor, service, unknown), part-time worker, retired, neighborhood choice criteria dummy variables (friends, leisure, access to job, near transit, public services, looks/design, near school, near shops/services, near highway, house important, other). a. New housing is seven or fewer years old at the time of the survey. Near rail is within walking distance, dened as 0.4 miles measured along road network. * p .10 ** p .05 *** p .01

28

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

0.098 auto-based grocery trips per week. Low on- and off-street parking has a coefcient of 0.57, implying a 25% reduction in auto-based grocery trips. Neither onstreet nor off-street parking is independently signicant, suggesting that for non-work trips requiring goods carrying, the auto is doubly attractive and only signicant impediments to its use may have an inuence. Housing type and tenure, local population density, and local job density are not signicant in these models, while subregional bus stop and employment density are negatively associated as expected. There are two puzzling coefcients: distance from the Manhattan CBD is associated with fewer auto-based grocery trips, and the number of bus stops within a mile is associated with more (although this latter effect declines and becomes insignicant once demographic characteristics are controlled). Perhaps there are more but also shorter auto trips in places that have high bus accessibility and are nearer to Manhattan. Trip distance is not measured in the dataset. When demographic and residential location criteria variables are added, the implied effect of low on- and off-street parking remains large, at 0.48 fewer grocery trips per week, although it is now signicant only at the 90% condence level; the coefcients on subregional bus stop density, the number of grocery stores, and distance to Manhattan are slightly smaller but still signicant; and subregional employment density and bus stops within one mile are no longer signicant (Table 6, column 3). Worker status is associated with 0.41 fewer trips to the grocery store, which could be caused by time scarcity relative to non-workers. Of all of the stated residential choice criteria, only seeking good schools is associated with grocery store trip frequency. When restricting the sample to households near rail stations, the distance to rail variable becomes statistically insignicant (Table 6, Model 4), suggesting that whatever role distance to rail plays in the use of autos for groceries, it is indirect. Perhaps it is a proxy for road congestion, which is not observed. The coefcient on low on- and off-street parking stays about the same as in Model 2 and the number of grocery stores nearby becomes again larger and more signicant, while the subregional built environment measures are no longer signicant. Finally, when the number of vehicles per adult is added as an endogenous explanatory variable (Table 6, Model 5), each additional vehicle per adult in the household is associated with an additional 0.4 auto-based grocery trips per week, and the independent inuence of low on- and offstreet parking declines a bit but remains large and statistically signicant at the 90% level. In contrast to the auto commuting models, this result implies that on- and off-

street parking availability may affect auto-based grocery trip frequency, even for people with high auto ownership.

Conclusions
Developing high-density, mixed-use housing near rail stations may reduce regional road congestion and auto pollution while slowing the growth in greenhouse gas emissions caused by auto use. But those benets may not depend very much on rail access. In these data, the lower auto ownership and use in TODs is not from the T (transit), or at least, not from the R (rail), but from lower onand off-street parking availability; better bus service; smaller and rental housing; more jobs, residents, and stores within walking distance; proximity to downtown; and higher subregional employment density. Previous disaggregate studies testing the inuence of rail access on auto ownership and use have typically controlled for only a subset of neighborhood or subregional built environment measures, rarely included housing type and tenure, and even more rarely controlled for on- or off-street parking supply. As others have argued, rail access and population density could be highly correlated with auto use due to unobserved variables like parking availability and walkability (e.g., Salon, 2009). In contrast to the results here, a study of 1998 survey data from New York matched to current Google observations of off-street parking found that walking distance to subway stations in New York remained signicant in predicting auto ownership when off-street parking was controlled (Guo, 2013). The analysis did not control for distance to downtown, subregional job and employment density, bus access, tenure and type of housing, or onstreet parking availability; nor did it specically test the walking-distance thresholds included here. The study area could also play a role. Subway access in New York City is highly correlated with more generalized transit accessibility. The comparatively weak inuence of rail access found in the present study is all the more remarkable given that New Jersey is so well served by rail and the share of rail commuting is so high. Although rail service undoubtedly attracts auto users in a way that buses do not, in some contexts it may also siphon off bus riders, walkers, and bikers. To test this hypothesis in the case of the commute to work, I estimated some additional commute mode regressions using binomial logit, like those presented in Table 5. Controlling for other factors, rail station distance was highly positively correlated with rail commuting, but negatively correlated with buses, walking and biking, ferry

Chatman: Does TOD Need the T?

29

use, and working at home.4 The apparent substitution between rail and other non-auto modes helps to explain why auto use varies relatively little as a function of distance to rail. Some rail stations are located far from job and shopping clusters, and regional-level accessibility and distance to downtown are often shown to be more highly associated with travel patterns than are neighborhood characteristics (see Boarnet, 2011; Ewing & Cervero, 2010; Handy, 1993). Thus, some housing developments near rail might lead to unintended increases in auto use. This implies a continuing need for an explicit accounting of scale in specifying measures of the built environment to account for local, subregional, and regional measures (Chatman, 2008; Zhang & Kukadia, 2005). The relationships among travel patterns, rail access, parking availability, and built environment measures are more complex than represented here. It is possible, for example, that rail investments could have played some role in either a market or political sense in increasing population density (cf. Bailey et al., 2008), increasing the number of grocery stores, and decreasing the amount of parking provided. But these results suggest rail plays at most an indirect role, and likely not a strong one, since the direct measure of rail is insignicant in all of the controlled models.

Policy Implications
Current sustainability policies are often quite focused on investing in rail and developing housing near rail stations. For example, California Senate Bill 375, a widely observed and admired attempt to incorporate climate planning within regional transportation and land use planning, gives special consideration to transit priority projects: dense housing development within a half mile of a major transit station or high-quality transit corridor (Cal. Govt. Code 21155.1). Such a focus primarily on TODs to reduce greenhouse gases could miss the boat. These results suggest that a better strategy in many urban areas would be to incentivize housing developments of smaller rental units with lower on- and off-street parking availability, in locations with better bus service and higher subregional employment density. Rail station areas may be among the most likely to be targeted for housing development proposals because developers are aware that public opposition is often lower near rail stations and because policymakers and urban planners believe that rail access will mitigate trafc impacts. But such a policy will not serve long-

term sustainability interests if, in fact, rail investments and rail-proximate housing make little difference in auto use in and of themselves. The focus on rail is particularly problematic in cases where developments near rail stations are simply transit adjacent, with high amounts of parking, low density, and large units being offered for sale. Denser housing developments coupled with good management of automobile parking could reduce auto use in many contexts, and there could be a substantial market for it. Previous research has suggested the need to reduce parking requirements to take account of the fact that demand for parking is lower in places with transit service (e.g., Rowe, Bae, & Shen, 2011). But parking requirements likely themselves affect travel by oversupplying parking (Cutter & Franco, 2012); in other words, parking demand may be lower in places with rail service partly because parking is scarce. Public agencies are heavily involved both in regulating minimum amounts of off-street parking and in providing and regulating on-street parking. Developers could be allowed to provide less off-street parking, while on-street parking could be priced, managed, and permitted in order to mitigate spillover effects (Shoup, 2005). Future population growth in the United States may well be concentrated in cities, and on-street parking may become scarce while private off-street parking will become very expensive to construct. If so, existing policies regarding on- and off-street parking could signicantly constrain densication and inll development. It is fortunate if access to rail is not a primary factor in reducing auto use, not only because rail infrastructure is expensive, but also because the fraction of available land near rail stations is limited. That said, ubiquitous higher housing density and scarce on- and off-street parking could cause greater local auto congestion if not carefully managed. In fact, positive regional and global effects may result from those negative local impacts, if they quash more driving. However, negative local impacts induce cities to frown on dense development and neighbors to protest it. How can urban planners bring about a more widespread relaxation of parking regulations, height limits, oor-to-area ratio standards, and general plans that restrict the form and location of development and redevelopment? That is the planning puzzle that deserves our focused attention. The pursuit of rail-oriented development may be a distraction from it.
Acknowledgments
A number of people contributed to the research study upon which this article draws. Project manager Stephanie DiPetrillo designed the physical layout of the paper survey, identied much of the new TOD portion of

30

Journal of the American Planning Association, Winter 2013, Vol. 79, No. 1

the survey sample, managed the eld survey of on-street parking, and helped write a report on the project from which some of the Study design section was extracted. Marc Weiner coordinated the household survey, with the able assistance of Orin Puniello. The mailings and data entry were carried out by ABT/SRBI, under the direction of Chintan Turakhia and David Ciemnecki. Marc and Chintan also advised on the questionnaire and sampling design. Dan Tischler coded verbatim occupational responses into standard occupational classications. Nick Klein carried out the construction of most spatial measures in GIS, with initial work by Nicholas Tulach and Kyeongsu Kim. The grocery store counts were done by Matt Brill. The parking audits, and management of parking data, were carried out by Nick Klein, Lewis Thorwaldson, Katie Thielman, Milan Patel, Rodney Stiles, Liz Thompson, Charu Kukreja, Andrew Besold, Aaron Sugiura, Michael Parenti, and Graydon Newman. Thanks to Mike Manville, Robert Noland, Robert Cervero, and three anonymous reviewers for their very helpful comments on previous drafts.

Notes
1. Housing age was reported by survey respondents and supplemented with information about the year of development for known multifamily projects. Almost 20% of respondents reported that they did not know the age of the unit they were living in or did not answer the question; only 6% of those were in multifamily units known to be new. The remaining units are assumed to be at least eight years old. 2. Alternative methods such as structural equations, nested logit, or two-stage least squares could be used to control for the potential endogeneity of residential location, public transit, population density, parking, or other dependent variables (e.g., Bailey et al., 2008; Cervero & Murakami, 2010; Deka, 2002; Salon, 2009). Such efforts require plausibly exogenous instruments and historical data, which are not present in this dataset, but could be the subject of future research. 3. Multicollinearity generally did not present problems in these data, with the exception of the variable for on-street parking and, in the models restricted to near-station households, the subregional built environment variables. For example, for the 14 models presented here, the variance ination factor on distance to rail averaged 1.99 with a range of 1.72 to 2.29. When independent variables of interest were statistically insignicant in the presence of variance ination, I removed other collinear variables to see if signicance occurred once variance ination was reduced. Statistical signicance was generally unaffected, except for the spatial variables; as a result the set of spatial variables varies slightly for each of the model sets, except that Models 4 and 5 in each set are kept consistent with Model 2. 4. The carpooling model does a poor job of explaining the likelihood of carpooling; distance to rail is not signicant, nor are many of the other built environment variables. I ran other variants of this modal categorization but results were very similar. Detailed results are available upon request.

References
Arrington, G. B., & Cervero, R. (2008). Effects of TOD on housing, parking, and travel: TCRP Report 128. Washington, DC: Transportation Research Board. Bailey, L., Mokhtarian, P. L., & Little, A. (2008). The broader connection between public transportation, energy conservation and greenhouse gas reduction. Fairfax, VA: ICF International. Belzer, D., & Autler, G. (2002). Transit-oriented development: Moving from rhetoric to reality. Washington DC: Brookings Institution Center on Urban and Metropolitan Policy.

Bento, A. M., Cropper, M. L., Mobarak, A. M., & Vinha, K. (2005). The impact of urban spatial structure on travel demand in the United States. Review of Economics and Statistics, 87(3), 466478. Boarnet, M. G. (2011). A broader context for land use and travel behavior, and a research agenda. Journal of the American Planning Association, 77(3), 197213. California Senate Bill 375. Cal. Govt. Code 21155.1 Retrieved from http://www.leginfo.ca.gov/cgi-bin/displaycode?section=prc&group= 2100122000&le=2115521155.3 Calthorpe, P. (1993). The next American metropolis: Ecology, community, and the American dream. New York, NY: Princeton Architectural Press. Cervero, R., Ferrell, C., & Murphy, S. (2002). Transit-oriented development and joint development in the United States: A literature review. Washington, DC: Transit Cooperative Research Program. Cervero, R., & Murakami, J. (2010). Effects of built environments on vehicle miles traveled: Evidence from 370 US urbanized areas. Environment and Planning A, 42(2), 400418. Chatman, D., & DiPetrillo, S. (2010). Eliminating barriers to transitoriented development. New Jersey Department of Transportation, Bureau of Research. Retrieved from http://www.nj.gov/transportation/refdata/ research/reports/FHWA-NJ-2010-002.pdf Chatman, D. G. (2008). Deconstructing development density: Quality, quantity and price effects on household non-work travel. Transportation Research Part A: Policy and Practice, 42(7), 10081030. Cutter, W. B., & Franco, S. F. (2012). Do parking requirements signicantly increase the area dedicated to parking? A test of the effect of parking requirements values in Los Angeles County. Transportation Research A, 46(6), 901925. Deka, D. (2002). Transit availability and automobile ownership: Some policy implications. Journal of Planning Education and Research, 21(3), 285300. Dill, J. (2003). Transit use and proximity to rail: Results from large employment sites in the San Francisco, California, Bay Area. Transportation Research Record, 1835(1), 1924. Dillman, D. A. (1978). Mail and telephone surveys: The total design method. New York, NY: Wiley. Dillman, D. A., Dillman, J. J., & Makela, C. J. (1984). The importance of adhering to details of the Total Design Method (TDM) for mail surveys. New Directions for Program Evaluation, 1984(21), 4964. Ewing, R., & Cervero, R. (2010). Travel and the built environment: A meta-analysis. Journal of the American Planning Association, 76(3), 265294. Federal Highway Administration. (2009). National Household Travel Survey [Data les]. Retrieved from http://nhts.ornl.gov/download.shtml Guo, Z. (2013). Does residential parking supply affect household car ownership? The case of New York City. Journal of Transport Geography, 26(January), 1828. Guo, Z., & Xu, P. (2012). Duet of the commons: The impact of street cleaning on car usage in the New York City area. Journal of Planning Education and Research, 33(January), 3448. Handy, S. L. (1993). Regional versus local accessibility: Implications for nonwork travel. Transportation Research Record, 1400, 101107. Loo, B. P. Y., Chen, C., & Chan, E. T. H. (2010). Rail-based transitoriented development: Lessons from New York City and Hong Kong. Landscape and Urban Planning, 97(3), 202212. Lund, H. M., Cervero, R., & Willson, R. W. (2004). Travel characteristics of transit-oriented development in California. San Francisco, CA: Bay Area Rapid Transit District.

Chatman: Does TOD Need the T?

31

Pushkarev, B., & Zupan, J. M. (1977). Public transportation and land use policy. Bloomington: Indiana University Press. Rowe, D. H., Bae, C. H. C., & Shen, Q. (2011). Evaluating the impact of transit service on parking demand and requirements. Transportation Research Record, 2245, 5662. Salon, D. (2009). Neighborhoods, cars, and commuting in New York City: A discrete choice approach. Transportation Research Part a-Policy and Practice, 43(2), 180196. Shoup, D. C. (2005). The high cost of free parking. Chicago, IL: Planners Press. U.S. Census Bureau. (2008). LEHD Origin-Destination Employment Statistics (LODES) [Data le]. Retrieved from http://lehd.ces.census.gov/data/ Vance, C., & Hedel, R. (2007). The impact of urban form on automobile travel: Disentangling causation from correlation. Transportation, 34(5), 575588.

Weinberger, R. (2012). Death by a thousand curb-cuts: Evidence on the effect of minimum parking requirements on the choice to drive. Transport Policy, 20, 93102. Weinberger, R., Seaman, M. , & Johnson, C. (2009). Residential off-street parking impacts on car ownership, vehicle miles traveled, and related carbon emissions. Transportation Research Record, 2118, 2430. Zegras, C. (2010). The built environment and motor vehicle ownership and use: Evidence from Santiago de Chile. Urban Studies, 47(8), 17931817. Zhang, M. (2010). Can transit-oriented development reduce peak-hour congestion? Transportation Research Record, 2174,148155. Zhang, M., & Kukadia, N. (2005). Metrics of urban form and the modiable areal unit problem. Transportation Research Record, 1902, 7179.

The MIT Press

Planning Ideas That Matter

Livability, Territoriality, Governance, and Reective Practice edited by Bishwapriya Sanyal, Lawrence J. Vale, and Christina D. Rosan This volume of benchmark essays provides critical perspectives on some of the most important planning conversations in North America over the past ninety years. John Friedmann, University of British Columbia
408 pp., 34 illus., $27 paper

City Cycling
edited by John Pucher and Ralph Buehler This impressive book thoroughly documents the individual, community, and national benets of getting more people on bikes and proposes specic measures for making cycling safe and feasible for everyone. Andy Clarke, President, League of American Bicyclists
368 pp., 62 illus., $27.95 paper

now in paper

Topologies
The Urban Utopia in France, 1960-1970 Larry Busbea Superb the rst book-length study of a time and place when technologically innovative design proposals ourished on architects drawing boards but languished in the corridors of power. Martin Filler, The New York Review of Books
2400 pp., 137 illus., $19.95 paper

The MIT Press mitpress.mit.edu