What do residential lotteries show us about transportation choices?

Abstract

Chinese

Credibly identifying how the built environment shapes behaviour is empirically challenging, because people select residential locations based on differing constraints and preferences for site amenities. Our study overcomes these research barriers by leveraging San Francisco’s affordable housing lotteries, which randomly allow specific households to move to specific residences. Using administrative data, we demonstrate that lottery-winning households’ baseline preferences are uncorrelated with their allotted residential features such as public transportation accessibility, parking availability and bicycle infrastructure – meaning that neighbourhood attributes and a building’s parking supply are effectively assigned at random. Surveying the households, we find that these attributes significantly affect transportation mode choices. Most notably, we show that essentially random variation in on-site parking availability greatly changes households’ car ownership decisions and driving frequency, with substitution away from public transport. In contrast, we find that parking availability does not affect employment or job mobility. Overall, the evidence from our study robustly supports that local features of the built environment are important determinants of transportation behaviour.

Keywords

affordable housing car ownership public transportation

Introduction

A person’s residential neighbourhood shapes their health, employment and transportation habits – indeed, almost every aspect of their lifestyle, identity and opportunities. In turn, the choices people make based on residential location also affect others, through externalities such as pollution, road congestion and traffic collisions. Thus, urban planners and policymakers increasingly face calls to promote walkability, raise allowable building heights and densities, and reduce the amount of space dedicated to automobile parking. In principle, policies that provide more flexibility for developers will promote a mixture of local amenities and infrastructure that better matches the preferences of the community and allows more households to move to their preferred locations, thereby reducing the implicit regulatory tax imposed by many zoning regulations (Glaeser and Gyourko, 2018; Levine, 2005). In practice, the efficacy of these land-use policies in reducing transportation-related externalities depends heavily on how the built environment ultimately affects people’s behaviour.

A voluminous international literature in urban planning and economics considers how neighbourhood attributes such as public transportation access, residential density and walkability relate to automobile ownership, vehicle miles travelled and emissions (e.g. Ewing and Cervero, 2010; Giuliano and Narayan, 2003; Salon et al., 2012; Stevens, 2017; Zegras, 2010). To a lesser extent, researchers have also investigated how the accessibility of job opportunities correlates with employment and household income (Marinescu and Rathelot, 2018; Sanchez, 1999).

A significant challenge for understanding how location-based amenities such as public transportation affect residents’ travel behaviour and employment opportunities, however, is that people choose where to live, and they do so based in part on local factors such as the availability of parking and public transportation. This self-selection into residential (and workplace) locations means that the vast majority of inferences from the transportation–land use literature are susceptible to selection bias (Van Wee, 2009).¹ Fundamentally, the empirical concern is that residential location is a decision made by the residents, rather than an assignment based on some external process. For instance, individuals who do not own cars or otherwise prefer to commute via public transport are more likely to try to live nearby to major rail or bus lines, biasing upward any observed correlation between public transport access and utilisation (Glaeser et al., 2008). Individuals who prefer owning cars and driving, on the other hand, will care more about the provision of parking. Thus, most estimates in the literature that relate infrastructure features to travel behaviour lack a straightforward causal interpretation (Salon et al., 2012; Sampson et al., 2002).

A large body of research examines these self-selection challenges from theoretical, empirical and methodological perspectives. This literature includes a special issue in the Journal of Transport and Land Use (Cao, 2014) and some excellent reviews to which we refer the reader for more detail (Bhat and Guo, 2007; Cao and Chatman, 2016; Cao et al., 2009; Mokhtarian and Cao, 2008; Van Wee, 2009). One key finding is that the direction of self-selection bias is difficult to predict, as it depends on the extent to which neighbourhood characteristics match residents’ travel preferences (Cao and Chatman, 2016; Manaugh and El-Geneidy, 2015). A related line of argument suggests that residential sorting is an important channel through which development exerts an impact on travel, especially if it helps residents find housing in neighbourhoods that are consonant with their preferences (Chatman, 2014; Levine, 1998; Naess, 2014).

Often, however, a more specific causal interpretation is desired, particularly when seeking to understand the impacts of non-marginal changes to the built environment, and how policy changes will affect travel decisions in existing neighbourhoods where few people will re-sort (i.e. move) in the short term. Studies often attempt to correct for residential selection bias using statistical controls, propensity score matching, instrumental variables, joint models of residential location and mode choice, panel data, or related methods (see Cao et al., 2009; Mokhtarian and Cao, 2008, for reviews). These bias-correcting techniques considerably change quantitative results, including reducing the estimated impacts of land-use characteristics like urban density on vehicle travel by 50 per cent or more (Stevens, 2017).

All these methodological approaches, however, are only partial solutions. Longitudinal studies of movers, for example, can better control for within-household characteristics, but face the challenge that movers may be moving in order to better align their travel preferences with neighbourhood characteristics. Joint and structural models, meanwhile, require strong assumptions, especially considering that the selection bias involves both observable and unobservable factors (Pinjari et al., 2007). Because joint models estimate the choice of residential neighbourhood, they are also ill-equipped to assess the impacts of building-specific attributes such as parking provision. Indeed, while residential parking provision might be expected to have a major impact on travel behaviour (Manville and Shoup, 2005; Shoup, 2005), empirical studies are few in number and typically cannot consider biases from residential self-selection (e.g Guo, 2013; Weinberger, 2012). In short, the empirical challenge is that residential ‘self-selection leads to non-random heterogeneity in choices and behaviour’ (Van Wee, 2009: 280).

As in nearly all areas of social science research, randomised experiments are the gold standard by which to identify causal effects. In principle, researchers could randomly assign households to different types of neighbourhoods and then observe their behaviour, but this is rarely practical or ethical (Cao et al., 2009). Randomisation has been successfully employed to analyse how federally subsidised housing vouchers via the Moving to Opportunity programme affect economic opportunities, crime and public health outcomes (Chetty et al., 2016; Feins and Shroder, 2005; Katz et al., 2001; Kling et al., 2007; Leventhal and Brooks-Gunn, 2003; Ludwig et al., 2001; Sanbonmatsu et al., 2011). Similar lotteries for housing assistance or public housing have also been used to analyse labour market and health outcomes in Canada, Ethiopia, India and the Netherlands (Adair et al., 2016; Barnhardt et al., 2017; Bowen et al., 2018; Franklin, 2019; Van Dijk, 2019). All of these studies, however, primarily evaluate the effects of randomly moving households away from particular residential locations – such as out of government-provided housing projects – rather than the effects of assigning people to live in specific residential locations.

In this article, we leverage the housing lottery programmes in San Francisco to overcome the aforementioned research limitations and provide causal interpretations of the impacts of specific neighbourhood characteristics and parking provision on households’ transportation behaviour and economic outcomes. In San Francisco, nearly all new housing developments with 10 or more residential units must offer a government-specified share of ‘inclusionary’ units at below-market-rate (BMR) prices, either directly on site, directly off site, or indirectly off site by paying a fee. As might be expected, demand for new BMR units substantially exceeds the available supply – one recent lottery for 95 rental units attracted 6580 household applicants (Badger, 2018). Because of the very low odds of winning, eligible households generally apply indiscriminately to many different housing lotteries. Those that are fortunate to eventually win a BMR unit are thus effectively assigned to live in specific buildings and neighbourhoods. In essence, San Francisco’s housing lotteries provide as-good-as-random assignment of people into homes.

Conceptually, our approach is most similar to Lin et al. (2017) and Manville (2017), who study travel behaviour among public housing residents who have limited choices of where to live. In both of these studies, however, the as-good-as-random allocation was assumed rather than being a primary characteristic of the setting – and this key assumption could not be directly tested.

Our research design compares transportation behaviour and economic outcomes across households that won different BMR lotteries, which thereby provides effectively random variation in their residential building characteristics like on-site parking availability and neighbourhood-level characteristics such as bicycle infrastructure and accessibility of public transportation. In doing so, we provide the first evidence to our knowledge about transportation behaviour and economic outcomes for a population that is in effect randomly assigned to live in particular places. To validate our empirical strategy, we assess whether households are selective in the types of BMR housing projects for which they apply, finding that lottery participation decisions are indeed as-good-as-random. We then present findings from a survey that we conducted of about 2700 of these households currently residing in BMR units, asking them questions about their transportation choices and employment.

The responses to our survey confirm the importance of accessibility by walking, bicycling and public transport in shaping household transportation choices. Even in a city such as San Francisco, where walkable neighbourhoods are the norm and public transport quality is quite high by US standards, accessibility substantially impacts people’s travel and commuting decisions. On-site residential parking has even larger effects: increased parking causes more car ownership and more driving while reducing public transport use, regardless of a neighbourhood’s public transport accessibility. Moreover, additional parking does not improve employment or labour market mobility among households in our sample. In summary, the evidence from our study robustly supports that urban residents’ transportation behaviour – but not their employment – is affected by local features of the built environment, and particularly so by parking.

Setting: San Francisco affordable housing programmes

San Francisco is often ranked as one of the least affordable cities in the United States (e.g. NAHB, 2019). In response, the city has developed and implemented a range of programmes to increase the availability of affordable housing and to provide down payment assistance for qualified home purchases. Most of these programmes are administered by the Mayor’s Office of Housing and Community Development (MOHCD). We focus on the Inclusionary Housing programme, under which a government-specified portion of units in most new residential developments must be made available at below-market-rate prices (or rent) to households whose income is below specified thresholds. Given San Francisco’s high incomes, a two-person household generally can qualify while earning up to US$118,200, equivalent to 120% of city median income.

For private developers, the BMR percentage requirements and thresholds have varied over time since the Inclusionary Housing programme was established in 2002, but as of December 2019, 12 to 20 per cent of on-site housing units (depending on the project’s size) must be set aside for low- or middle-income households. Alternatively, developers can directly provide off-site affordable housing or pay a fee that is used to supply off-site affordable housing. BMR housing projects are also developed using a mix of public and private funds by the city’s Office of Community Investment and Infrastructure. Some projects cater to specific groups such as the elderly, people who are homeless and people with disabilities.

Although specific eligibility requirements and funding sources vary between and within these programmes, a lottery mechanism is used to allocate all BMR units that we study in this article. First, would-be residents apply to each lottery; no fee is required at this stage. Second, applicants are randomly assigned a rank. Then, eligibility is verified for those receiving a sufficiently high rank. Finally, units are offered to eligible applicants in their lottery rank order within certain priority groups. Applicants from higher-priority groups – such as tenants displaced by no-fault eviction or fire – are more likely to win BMR units, but about two-thirds of successful applicants are in the lower-priority group of residents who live or work in San Francisco.

Projects also vary by the amount of parking that is provided. In the early years of the BMR programme, projects had a one-to-one ratio of parking spaces to units, and the cost of parking was bundled in with the rent or sales price. In line with subsequent zoning reforms, however, more recent projects have unbundled parking from the rent or sales price – that is, residents are free to decline a parking space, but accepting it entails an added cost. At the same time, parking ratios of less than one space per unit or even zero parking have become more common. For projects that have a parking ratio of less than one space per unit, spaces are offered in lottery rank order within each priority group.² For example, in a project with 10 BMR units and a parking ratio of 0.5:1, the first five lottery winners would be guaranteed an offer of parking, but remaining lottery winners would only be offered parking if higher-ranked applicants declined to take (and normally pay for) a parking space. Developers are required to offer parking spaces to BMR units at the same ratio as they provide for market-rate units.

Methods and data

Housing lottery applications

To validate our assumption that housing assignments are as good as random, we use a data set of all 107,310 applications to 59 BMR housing lotteries held between July 2015 and June 2018, which we call our applicant sample. Because the applicant sample is only available for a three-year period, it excludes many housing projects that are included in our primary survey sample.³ However, the applicant sample also encompasses 14 projects which are not present in our survey sample; these projects are currently managed by a non-profit housing organisation or by another city agency, precluding survey distribution to these units. In addition to lottery rank and priority group status, the applicant sample provides basic demographic information from lottery applications such as income, gender and race.

Household survey design

Our primary data survey sample consists of all BMR units for which we have occupancy and parking data, and comprises 2654 units in 197 projects that were occupied as of April 2019. Almost all (2605) of these units were built under the Inclusionary Housing programme. We obtained data on project-level characteristics directly from MOHCD and supplemented these data using land-use permit approval records to fill in missing data such as parking ratios. As shown in Figure 1 and Table 1, our survey sample provides meaningful variation in households’ building-level and neighbourhood characteristics. BMR units are distributed throughout the city, giving a range of walking environments and proximity to public transport, as well as substantial variation in on-site parking availability.⁴ Units are roughly evenly split between rentals (53% of units) and for-sale units; although more projects are ownership (65%), these tend to be smaller in scale.

Figure 1.

Locations of surveyed below-market-rate residential projects in San Francisco.

Table 1.

Summary attributes of surveyed below-market-rate (BMR) projects.

	Mean	SD	Range
Number of residential projects	197
Year completed	2008	6	1992–2018
Number of on-site BMR units	13.5	22.3	1–170
Total on-site residential units	86	105	10–540
Parking ratio (spaces per unit)	0.77	0.43	0.00–2.42
Distance to nearest rail stop (metres)	611	593	38–3203
AllTransit performance score	9.8	0.3	7.5–10.0
Walk Score	93	12	16–100
Bike Score	85	16	22–100

Notes: Table 1 shows statistics for San Francisco BMR projects. Walk Score and Bike Score are measured on a 0–100 scale, and are obtained from walkscore.com. The public transport score refers to the AllTransit Performance Score calculated by the Center for Neighborhood Technology. It considers frequency, connectivity and access to jobs, and is measured on a 0–10 scale.

Our survey asked all BMR residents in our survey sample about their frequency of travel by mode; their car ownership; their employment status; the location of their workplace or school (if any); and their interactions with neighbours. These survey questions are not intended to calculate vehicle miles travelled or other common metrics of transportation usage, which would require a substantially more complex survey instrument, impacting response rates and increasing recall bias. Similarly, our employment questions allow us to create coarse measures of labour market mobility.⁵

The survey questionnaire was posted with a reply-paid envelope in June 2019 to 2654 primary occupants of inclusionary housing units in the MOHCD database. Supplemental Appendix Figure A2 shows the paper survey instrument. In addition, 1693 of these (same) occupants received an email version of the survey with a personalised link to an online survey platform. Questions were provided in English, Spanish, Filipino and Chinese. As an incentive for participating, respondents were eligible to win one of 10 randomly awarded US$100 Visa gift cards. After merging cases where we received both an online and a postal response, we obtained 779 completed surveys, a response rate of 29.4%. We attribute this high rate to our efforts to keep the survey very short (one side of an A9-size card) and simple, the minimally intrusive questions, the pecuniary incentives, and the twin modes of distribution.

Transportation accessibility measures

Our analyses consider how four primary measures of transportation accessibility affect household behaviour. We quantify private automobile accessibility using each building’s ratio of parking spaces per residential unit. We use the Center for Neighborhood Technology’s AllTransit performance score to measure public transport frequency and quality, and we use the WalkScore company’s Walk Score and Bike Score metrics to measure accessibility by walking and cycling, respectively. Walk Score and Bike Score were accessed via the API at www.walkscore.com, and are based on accessibility of retail, services and other destinations, as well as neighbourhood design factors such as block length and bike lane provision (Hall and Ram, 2018; Manaugh and El-Geneidy, 2011; Osama et al., 2020).

Whereas many analyses of how land use relates to transportation behaviour focus on the ‘D’ variables such as density, land-use diversity (mix), and distance to public transport (see Ewing and Cervero, 2010, for a meta-analysis), our accessibility variables arguably provide a better measure of household transportation choices in our setting (Handy, 2018). For one, each of our measures is specific to a particular mode of transportation. In contrast, factors like urban density and street connectivity can affect household decisions through multiple channels, such as by making frequent public transportation service feasible and by providing more direct travel paths for walking to local destinations (Barrington-Leigh and Millard-Ball, 2020). Accessibility is also the more proximate influence – households do not make travel decisions directly in response to density, land-use mix or connectivity, but in response to how these factors affect accessibility. Finally, public transport accessibility can be changed more directly through policy, for example by changing service frequencies or routes. We therefore focus on accessibility-oriented metrics rather than (e.g.) sheer density, because these proxies more comprehensively reflect the variation in households’ abilities to access destinations using particular travel modes.

As would be true for nearly any measure of transportation accessibility, our explanatory variables are correlated through spatial variation. For instance, a location that has a relatively higher Walk Score is also likely to have a relatively higher public transport score, and buildings with good public transport accessibility tend to have less parking.⁶ To address this collinearity, our preferred regression specifications include the parking ratio and only one of the other accessibility measures, an approach that captures the meaningful spatial variation in accessibility while providing regression estimates that can readily be interpreted.

Results

Demonstrating as-good-as-random housing assignment

We begin our empirical analysis by demonstrating that assignment of lottery-winning households to housing units is as-good-as-random, which facilitates causal inference. To do so, we examine the patterns of lottery participation and repeat-entrant behaviour among households in our applicant sample. While each lottery is itself random by design, households might possibly choose to selectively enter only certain lotteries, for example by forgoing the chance to rent or buy in a building without parking or in one that is distant from a public transport stop.

A reasonable hypothesis is that households are not selective, given the low probability of winning any lottery. As shown in Figure 2, BMR projects attract up to 6575 applicants, while the average lottery in our applicant sample offers only 27 units (median = 11 units; maximum = 170 units). With the exception of a handful of projects that cater to the elderly or other specific populations, the odds of winning a rental unit lottery are extremely small – the average success rate of these applications is only 1.2%. Lotteries for ownership units attract a much smaller pool of applicants, likely because of the need to obtain a mortgage down payment, but ownership lotteries still have an average success rate of only 12%. Across all lotteries in our applicant sample, the average success rate is 1.5%.

Figure 2.

Number of entrants across BMR residential lotteries by application outcomes.

We confirm our hypothesis that households are not selective using regression analysis. Specifically, we estimate whether a household is more (or less) likely to participate in a particular housing lottery depending on how the characteristics of that lottery differ from those of the first lottery that the same household entered. For each household in the applicant sample, we defined the set of relevant lotteries as those with a lottery date between that of the first and last lotteries in our data set that the household entered. We then estimate whether a household’s decision to skip or enter each of these lotteries is explained by the characteristics of the associated project. We measure project characteristics (such as the parking ratio) in terms of their absolute differences from that of the first lottery in our data set that the household played, which we take to be the baseline preferences of the household. All explanatory terms are first standardised using a z-transformation (mean = 0, SD = 1). We control for household-specific fixed effects, as some households are more attentive in general to lottery availability and participate more often overall. We also control for lottery-specific fixed effects, as some lotteries are relatively better advertised, have less restrictive eligibility criteria, or otherwise attract entry from a broader section of the population.

We find no evidence that households skip lotteries based on project or neighbourhood characteristics such as parking and public transport accessibility. Table 2 presents linear probability models for lottery skipping using different subsamples of lottery-applying households. The first column includes all households that we observe playing at least two lotteries (as there cannot be skipping by households that played only a single lottery). If households were selectively participating based on their baseline preferences, then we should see that an absolute change in project characteristics – relative to those of the first lottery entered by that same household – would be associated with a larger propensity to skip a particular lottery. In contrast, the regression estimates indicate no evidence of lottery selectivity. For instance, we find that a one standard deviation difference in the parking ratio of the building, equivalent to 0.43 parking spaces per unit, is associated with a tiny 0.5 percentage point decrease in the probability of the household participating in the lottery. We find similarly small and almost always statistically insignificant relationships for the other explanatory terms. The one significant estimate, for Walk Score, is quantitatively small and thus has little practical import, further indicating a lack of selectivity in lottery participation.

Table 2.

Identification tests for lottery skipping: Regression estimates.

	Dependent variable: I{skipped lottery}
	(1)	(2)	(3)
Abs(Δ) in std. parking ratio	0.005 (0.005)	0.003 (0.007)	0.010 (0.015)
Abs(Δ) in std. public transport score	−0.001 (0.005)	−0.007 (0.016)	0.011 (0.022)
Abs(Δ) in std. Walk Score	0.014** (0.005)	0.021 (0.014)	0.034 (0.019)
Abs(Δ) in std. Bike Score	−0.004 (0.005)	0.003 (0.011)	−0.004 (0.015)
Sample	All applicants	Winner occupants	Survey respondents
Average skip rate	0.779	0.759	0.743
Household fixed effects	Yes	Yes	Yes
Lottery fixed effects	Yes	Yes	Yes
Number of households	18,574	481	159
Number of lotteries	59	59	59
Observations	290,085	10,165	3441
R ²	0.397	0.345	0.340

Note: Standard errors two-way clustered by household and lottery. An observation is either a household application to a lottery (107,310 in total) or a household skipping of a lottery (182,775). Each column presents estimates from a separate linear probability regression model. *p < 0.05. **p < 0.01. ***p < 0.001.

The same null patterns continue to hold as we restrict the estimation samples in Column (2) to use only households that (eventually) won a lottery, that is, those that we surveyed, or even further restrict in those Column (3) to our survey respondents.⁷ On the whole, the evidence in Table 2 clearly supports that households – quite understandably – are not selective in their participation in these low-odds housing lotteries. Nor do we detect any bias from differential patterns of survey response – for example, if households in buildings with better public transport access or lower parking ratios are more or less likely to respond to the survey (Supplemental Appendix Table A1).

Survey analysis for transportation

Having demonstrated as-good-as-random assignment of people into homes, the remainder of our analysis focuses on the household survey that we fielded. We begin by examining the relationship between household car ownership and a building’s parking provision and neighbourhood transportation accessibility. Figure 3 demonstrates a clear and substantive trend: the more parking in a building, the more likely a resident household is to own a car. In buildings with no on-site parking, only 38% of households own a car. In buildings with at least one parking space per unit, more than 81% of households own automobiles. Moreover, for buildings with intermediate amounts of parking, the pattern in Figure 3 shows monotonically increasing car ownership rates.

Figure 3.

Survey responses for car ownership by residential parking ratio.

A similar relationship between parking provision and car ownership is shown by the regression models in Table 3. In Column (1), a minimal univariate linear specification indicates that a one standard deviation increase in a building’s parking ratio – about 0.43 additional spaces per unit – causes a household to be 14 percentage points more likely to own a car.⁸ As discussed above, parking ratios are correlated with the other neighbourhood-level factors such as public transport accessibility and walkability. However, Columns (2) to (4) show very similar estimates (12 percentage points) using specifications that also include regressors for accessibility by public transport, walking and bicycling, along with survey respondent-level controls. That the estimate remains unchanged when adding the control terms is further indirect evidence of the as-good-as-random assumption. Public transport accessibility emerges as a somewhat smaller influence on car ownership, and is insignificant in Column (4), likely because of the strong collinearity noted above between public transport, walking and bicycling accessibility. On the whole, car ownership appears to be strongly influenced by features of the local built environment.⁹

Table 3.

Survey responses for car ownership: Regression estimates.

	Dependent variable: I{own any cars}
	(1)	(2)	(3)	(4)
Std. parking ratio	0.143***(0.023)	0.121***(0.026)	0.121***(0.025)	0.118***(0.026)
Std. public transport score		−0.048***(0.012)	−0.037*(0.016)	−0.021(0.025)
Std. Walk Score				−0.030(0.029)
Std. Bike Score				0.014(0.032)
Average car ownership	0.668	0.668	0.668	0.668
Controls	No	No	Yes	Yes
Observations	758	758	758	758
R ²	0.075	0.086	0.137	0.138

Note: Standard errors clustered by residential project. Controls are residency duration, residency type (rent or own), household income, household size, survey recipient gender and survey recipient race. Each column presents estimates from a separate linear probability regression model. *p < 0.05. **p < 0.01. ***p < 0.001.

In addition to impacting car ownership, parking ratios and transportation accessibility also affect household transportation mode decisions. Figure 4 shows the raw correlations between project- and neighbourhood-level transportation availability characteristics (columns) and surveyed households’ travel behaviour (rows). As expected, the frequency of driving (bottom row) increases with the building’s parking ratio and decreases with neighbourhood public transport, walking and cycling accessibility. The frequency of bicycling, walking and public transport use (the top three rows) show the opposite relationship to that for driving. Across the board, these correlations strongly support the conclusion that households choose between driving and other modes of travel based on the quality and availability of modes of transportation.

Figure 4.

Survey responses for household transportation utilisation: Correlation matrix.

The raw correlations provide compelling evidence that transportation choices depend on features of the local built environment. To more formally estimate the importance of these transportation availability measures in shaping households’ choices, we present multivariate regression analysis in Table 4. In Panel [A], the dependent variables are a respondent’s self-reported frequency of travel by single-occupant vehicle, public transportation, walking and bicycle, respectively. The survey asked how often the respondent travels by each mode, on a 1 to 4 ordinal scale where a value of 4 is ‘daily’, a value of 3 is ‘2–3 times a week’, a value of 2 is ‘2–4 times a month’ and a value of 1 is ‘less often’. As expected, increasing accessibility by public transport, walking or bicycling increases the frequency of use of the corresponding mode, even after controlling for respondents’ household characteristics, as well as for the building’s parking ratio. Nearly all of the estimates are statistically significant at the 5% level (the p-value for public transport score in Column (2) is 0.055), and most magnitudes are nontrivial. A one standard deviation increase in the building’s Walk Score, for instance, causes about a 24 percentage point increase in the likelihood that a household’s walking behaviour falls into a more frequent category.

Table 4.

Survey responses for transportation utilisation: Regression estimates.

	Transportation mode
	Private car (1)	Public transport (2)	Walking (3)	Bicycling (4)	Car if choice is car or public transport (5)
Panel [A] Frequency of use of transportation modes for all trips
Std. parking ratio	0.202***(0.059)	−0.242***(0.051)	−0.178**(0.060)	−0.063(0.037)
Std. public transport score	−0.152**(0.049)	0.097(0.051)
Std. Walk Score			0.235***(0.061)
Std. Bike Score				0.090***(0.023)
Panel [B] Primary mode of transportation for commute trips
Std. parking ratio	0.073*(0.030)	−0.069*(0.034)	0.013(0.029)	−0.004(0.013)	0.115**(0.041)
Std. public transport score	−0.086***(0.016)	0.001(0.016)			−0.017(0.019)
Std. Walk Score			0.059***(0.015)
Std. Bike Score				0.029***(0.007)
Controls	Yes	Yes	Yes	Yes	Yes
Panel [A] dep. var. avg.	2.195	2.627	2.875	1.351	–
Panel [A] observations	766	766	766	766	–
Panel [A] R²	0.082	0.082	0.110	0.075
Panel [B] dep. var. avg.	0.384	0.314	0.211	0.068	0.550
Panel [B] observations	544	544	544	544	380
Panel [B] R²	0.102	0.053	0.065	0.049	0.098

Note: Standard errors clustered by residential project. Controls are residency duration, residency type (rent or own), household income, household size, survey recipient gender and survey recipient race. For Panel [A], each dependent variable frequency measure is treated as a continuous outcome formed from a four-point scale where a value of 4 is daily, a value of 3 is 2–3 times a week, a value of 2 is 2–4 times a month and a value of 1 is less often. For Panel [B], dependent variables are binary indicators for primary commuting mode and each regression includes only respondents who are either employed or a student. Each column presents estimates from a separate linear regression model. *p < 0.05. **p < 0.01. ***p < 0.001.

In the case of public transport use, a building’s parking ratio also has an effect – and one that is more than twice as large as that of public transport accessibility. More on-site parking reduces public transport use while increasing the frequency of driving by a similar amount. The parking ratio also has a smaller but still statistically significant negative impact on the frequency of walking. Our estimates show no detectable impact of on-site parking on bicycling, although bicycling frequency is low for this population – 83% of respondents report bicycling ‘less often’ than 2–4 times a month.

The analysis above refers to all trips made by survey respondents. Similar patterns are shown in Panel [B] of Table 4 for commute trips to work or school. Increased residential parking leads to a higher probability of commuting by private car (driving alone or carpooling) and a lower probability of commuting by public transport. Greater public transport accessibility has the opposite effects, although these results are not always as statistically significant.

The impact of parking and transportation accessibility on commute mode choice appears to be more muted than for non-work trips. This might be because commute trips are relatively more constrained, for example by workplace parking options or public transport proximity, whereas non-commuting trips entail more choice of potential destinations for (say) shopping or recreation. Another constraint relates to long distances that may preclude walking or bicycling. For this reason, Column (5) of Table 4 restricts the sample to only commutes made either by public transport or private car. These estimates show even more clearly that commuters use either private car or public transport based on the building’s on-site parking availability. Also, note that the outcome measures for all trips and commute trips are not directly comparable. Panel [A] considers ordinal frequencies of use of all modes for all purposes, whereas Panel [B] uses binary outcomes for respondents’ primary mode of commuting.

Survey analysis for employment

Finally, we evaluate employment outcomes and focus on two key transportation factors that the literature suggests may affect labour market opportunities, particularly for low-income workers. Access to public transportation and access to private vehicles have both been found to improve employment outcomes, although the evidence is mixed (Blumenberg and Ong, 2001; Blumenberg and Pierce, 2014; Grengs, 2010; Sanchez, 1999; Sanchez et al., 2004).

For our surveyed households, who are essentially randomly assigned to a residential location, Table 5 suggests that neither public transport accessibility nor parking ratios have any impact on the probability of a respondent being employed full time (Column (1)). There is a similar null relationship with other labour market outcomes in Columns (2) to (4).¹⁰ One possibility is that these estimates are only indicative of the strong economy and minimal unemployment in San Francisco at the time of our survey in 2019. An alternative explanation is that small changes in car ownership and public transport access have little relevance for employment prospects after residential self-selection is fully accounted for via as-good-as randomisation. For example, residing in a low-accessibility neighbourhood might be correlated more generally with being unemployed because of some third factor such as discrimination in both housing and employment markets.

Table 5.

Survey responses for employment status: Regression estimates.

	Dependent variable
	Working full time (1)	Working part time (2)	Looking for work (3)	Student (4)
Std. parking ratio	−0.014(0.017)	0.008(0.014)	0.004(0.007)	0.003(0.012)
Std. publictransport score	0.001(0.010)	0.0004(0.008)	0.0002(0.003)	−0.004(0.008)
Dep. var. average	0.865	0.095	0.023	0.026
Controls	Yes	Yes	Yes	Yes
Observations	660	660	660	660
R ²	0.075	0.028	0.029	0.046

Note: Standard errors clustered by residential project. Controls are residency duration, residency type (rent or own), household income, household size, survey recipient gender and survey recipient race. Each of these regressions includes only respondents who are employed, looking for work, or a student. Each column presents estimates from a separate linear probability regression model. *p < 0.05. **p < 0.01. ***p < 0.001.

Conclusions

In this article, we use San Francisco’s residential housing lotteries to study how local parking and transportation accessibility affect household behaviour. Because the odds of winning any specific lottery are low and there are no monetary costs of entry, households are understandably quite unselective about which lotteries they enter. As we demonstrate, those who are fortunate to win any lottery are thus as-good-as-randomly assigned into living in particular residential locations. Our primary contribution is to present findings with a straightforward causal interpretation, in contrast to nearly all research on the implications of transportation accessibility for travel and employment outcomes, which is susceptible to selection bias from residential sorting. We also demonstrate the importance to travel choices of on-site parking provision, which has been ignored in most transportation–land use studies because of data limitations and because this selection bias is likely to be even more acute. Finally, we demonstrate the potential to use accessibility metrics to measure land use characteristics, in place of density and other ‘D’ variables which are one step removed in the causal chain.

The generalisability of our findings is qualified because our evidence is limited to a single city, and to households that are eligible for affordable housing programmes. In San Francisco, however, these programmes target a wide income range: households earning up to 120% of the median income (US$118,200 for a two-person household in 2019) are often eligible for these housing lotteries.

We find that neighbourhood-level accessibility has statistically significant and quantitatively meaningful impacts on household decisions about car ownership and travel. In this way, we confirm the findings of the larger literature on the land use–transportation connection (e.g. Ewing and Cervero, 2010; Stevens, 2017), and show that earlier findings are not simply a product of selection bias.

Specifically, greater public transport accessibility reduces the propensity to own and drive a car, while increasing the propensity to use public transport. Greater walk and bicycle accessibility also increases the propensity to use those modes. These findings are not surprising, but confirm that the land use–transportation relationships commonly shown in this literature are not simply a product of self-selection and other biases. Even within San Francisco, public transport accessibility substantially affects car ownership and travel behaviour – increasing public transport accessibility from the level of an outer suburban neighbourhood (the fifth percentile) to the citywide median would increase the share of those commuting by public transport rather than by car by 6.5 percentage points. San Francisco is a more walkable, bikeable and public transport-accessible city compared with most locations, suggesting that even more substantial household responses to increased bus frequency, for example, might be expected in places where public transport service is minimal at present.

We also document a more novel relationship between the residential parking provided in a building and transportation outcomes. Given that households who wish to own a car likely have numerous external parking options – to park on-street, park in a public garage or rent a space in a nearby building – one might surmise that neighbourhood-level rather than building-level parking supply would most affect transportation outcomes. However, we show that a building’s parking ratio not only influences car ownership, vehicle travel and public transport use, but has a stronger effect than public transport accessibility. Buildings with at least one parking space per unit (as required by zoning codes in most US cities, and in San Francisco until circa 2010) have more than twice the car ownership rate of buildings that have no parking. If parking is provided on-site for free or at a reduced price (typically, US$100 per month), then households appear to take advantage of this amenity. In contrast, households without access to on-site parking are more likely to forgo car ownership altogether.

One natural concern about reducing required parking ratios is that this might limit employment opportunities, particularly for lower-income households such as those we study in this article. Given that many jobs are inaccessible by public transport, access to a car can theoretically improve employment outcomes and labour market turnover. However, we find no evidence that this trade-off exists.

Public transport accessibility evolves over decades and a concerted effort to improve local infrastructure requires large amounts of public funding. Parking ratios, in contrast, require only regulatory changes to zoning codes: removing minimum requirements from zoning codes and possibly replacing them with maxima instead. Such zoning reforms could also yield other benefits, including reducing housing costs and increasing land available for new housing and commercial development, as well as reducing motor vehicle trips and associated harms. Our findings suggest that the potential for private automobile trip reductions is large and does not depend on car-free households relocating to car-free buildings. Moreover, reducing space dedicated to parking appears to come without employment downsides. Where streets are relatively walkable and public transport service is frequent, parking emerges as the key factor shaping household travel behaviour – and parking is a factor that is highly amenable to low-cost policy reforms that can rapidly provide benefits.

Supplemental Material

sj-pdf-1-usj-10.1177_0042098021995139 – Supplemental material for What do residential lotteries show us about transportation choices?

Supplemental material, sj-pdf-1-usj-10.1177_0042098021995139 for What do residential lotteries show us about transportation choices? by Adam Millard-Ball, Jeremy West, Nazanin Rezaei and Garima Desai in Urban Studies

Footnotes

Acknowledgements

We are grateful for financial support from the University of California Institute of Transportation Studies. We thank the San Francisco Mayor’s Office of Housing and Community Development – particularly Maria Benjamin, Alea Gage, Charles MacNulty and Teresa Yanga – for sharing data and insights on the city’s affordable housing lottery programmes. We appreciate assistance with translations and survey logistics from Amy Bezek, Carmelle Catamura, Ingrid Chen, Brianna Delgado, Siena Elvins, Esther Lau, Ben Read, Stephanie Segur, Paul Sosbee, Harshul Srivastava and Jiaxin Zhou. This manuscript benefited from comments by Laura Podolsky and Mark Garrett at UC-ITS, and comments from editor Yingling Fan and several anonymous reviewers at Urban Studies. Any errors are our own.

Declaration of conflicting interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by a grant from the University of California Institute of Transportation Studies with funding from the Road Repair and Accountability Act of 2017 (Senate Bill 1). The contents of this report reflect the views of the author(s), who are responsible for the facts and the accuracy of the information presented. This document is disseminated under the sponsorship of the State of California in the interest of information exchange and does not necessarily reflect the official views or policies of the State of California.

ORCID iD

Adam Millard-Ball

Supplemental material

Supplemental material for this article is available online.

Notes

References

Adair

Kopp

Distasio

, et al. (2016) Housing quality in a randomized controlled trial of Housing First for homeless individuals with mental illness: Correlates and associations with outcomes. Journal of Urban Health 93(4): 682–697.

Badger

(2018) These 95 apartments promised aﬀordable rent in San Francisco. Then 6,580 people applied. The New York Times, 12 May. Available at: https://www.nytimes.com/2018/05/12/upshot/these-95-apartments-promised-affordable-rent-in-san-francisco-then-6580-people-applied.html (accessed 24 March 2020).

Barnhardt

Field

Pande

(2017) Moving to opportunity or isolation? Network eﬀects of a randomized housing lottery in urban India. American Economic Journal: Applied Economics 9(1): 1–32.

Barrington-Leigh

Millard-Ball

(2020) Global trends toward urban street-network sprawl. Proceedings of the National Academy of Sciences 117(4): 1941–1950.

Bhat

Guo

(2007) A comprehensive analysis of built environment characteristics on household residential choice and auto ownership levels. Transportation Research Part B: Methodological 41(5): 506–526.

Blumenberg

Ong

(2001) Cars, buses, and jobs: Welfare participants and employment access in Los Angeles. Transportation Research Record: Journal of the Transportation Research Board 1756(1): 22–31.

Blumenberg

Pierce

(2014) A driving factor in mobility? Transportation’s role in connecting subsidized housing and employment outcomes in the Moving to Opportunity (MTO) program. Journal of the American Planning Association 80(1): 52–66.

Bowen

Quintiliani

Bhosrekar

, et al. (2018) Changing the housing environment to reduce obesity in public housing residents: A cluster randomized trial. BMC Public Health 18(1): 883.

Cao

(2014) Residential self-selection in the relationships between the built environment and travel behavior: Introduction to the special issue. Journal of Transport and Land Use 7(3): 1–3.

10.

Cao

Chatman

(2016) How will smart growth land-use policies aﬀect travel? A theoretical discussion on the importance of residential sorting. Environment and Planning B: Planning and Design 43(1): 58–73.

11.

Cao

Mokhtarian

Handy

(2009) Examining the impacts of residential self-selection on travel behaviour: A focus on empirical ﬁndings. Transport Reviews 29(3): 359–395.

12.

Chatman

(2014) Estimating the eﬀect of land use and transportation planning on travel patterns: Three problems in controlling for residential self-selection. Journal of Transport and Land Use 7(3): 47–56.

13.

Chetty

Hendren

Katz

(2016) The eﬀects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity experiment. The American Economic Review 106(4): 855–902.

14.

Christensen

Sarmiento-Barbieri

Timmins

(2020) Housing discrimination and pollution exposures in the United States. NBER Working Paper No. 26805. Cambridge, MA: National Bureau of Economic Research.

15.

Ewing

Cervero

(2010) Travel and the built environment. Journal of the American Planning Association 76(3): 265–294.

16.

Feins

Shroder

(2005) Moving to Opportunity: The demonstration’s design and its eﬀects on mobility. Urban Studies 42(8): 1275–1299.

17.

Franklin

(2019) The demand for government housing: Evidence from lotteries for 200,000 homes in Ethiopia. Working Paper, London School of Economics.

18.

Giuliano

Narayan

(2003) Another look at travel patterns and urban form: The US and Great Britain. Urban Studies 40(11): 2295–2312.

19.

Glaeser

Gyourko

(2018) The economic implications of housing supply. Journal of Economic Perspectives 32(1): 3–30.

20.

Glaeser

Kahn

Rappaport

(2008) Why do the poor live in cities? The role of public transportation. Journal of Urban Economics 63(1): 1–24.

21.

Graham

(2018) Identifying and estimating neighborhood eﬀects. Journal of Economic Literature 56(2): 450–500.

22.

Grengs

(2010) Job accessibility and the modal mismatch in Detroit. Journal of Transport Geography 18(1): 42–54.

23.

Guo

(2013) Home parking convenience, household car usage, and implications to residential parking policies. Transport Policy 29: 97–106.

24.

Hall

Ram

(2018) Walk score^® and its potential contribution to the study of active transport and walkability: A critical and systematic review. Transportation Research Part D: Transport and Environment 61: 310–324.

25.

Handy

(2018) Enough with the “D’s” already – Let’s get back to “A”. Transfers 1: 1–3.

26.

Katz

Kling

Liebman

(2001) Moving to Opportunity in Boston: Early results of a randomized mobility experiment. The Quarterly Journal of Economics 116(2): 607–654.

27.

Kling

Liebman

Katz

(2007) Experimental analysis of neighborhood eﬀects. Econometrica 75(1): 83–119.

28.

Leventhal

Brooks-Gunn

(2003) Moving to Opportunity: An experimental study of neighborhood eﬀects on mental health. American Journal of Public Health 93(9): 1576–1582.

29.

Levine

(1998) Rethinking accessibility and jobs–housing balance. Journal of the American Planning Association 64(2): 133–149.

30.

Levine

(2005) Zoned Out. Regulation, Markets, and Choices in Transportation and Metropolitan Land Use. Washington, DC: RFF Press.

31.

Lin

Wang

Guan

(2017) The built environment, travel attitude, and travel behavior: Residential self-selection or residential determination? Journal of Transport Geography 65: 111–122.

32.

Ludwig

Duncan

Hirschﬁeld

(2001) Urban poverty and juvenile crime: Evidence from a randomized housing-mobility experiment. The Quarterly Journal of Economics 116(2): 655–679.

33.

Manaugh

El-Geneidy

(2011) Validating walkability indices: How do diﬀerent households respond to the walkability of their neighborhood? Transportation Research Part D: Transport and Environment 16(4): 309–315.

34.

Manaugh

El-Geneidy

(2015) The importance of neighborhood type dissonance in understanding the eﬀect of the built environment on travel behavior. Journal of Transport and Land Use 8(2): 45–57.

35.

Manville

(2017) Bundled parking and vehicle ownership: Evidence from the American Housing Survey. Journal of Transport and Land Use 10(1): 27–55.

36.

Manville

Shoup

(2005) Parking, people, and cities. Journal of Urban Planning and Development 131(4): 233–245.

37.

Marinescu

Rathelot

(2018) Mismatch unemployment and the geography of job search. American Economic Journal: Macroeconomics 10(3): 42–70.

38.

Mokhtarian

Cao

(2008) Examining the impacts of residential self-selection on travel behavior: A focus on methodologies. Transportation Research Part B: Methodological 42(3): 204–228.

39.

Naess

(2014) Tempest in a teapot: The exaggerated problem of transport-related residential self-selection as a source of error in empirical studies. Journal of Transport and Land Use 7(3): 57–79.

40.

NAHB (2019) Housing opportunity index. 3rd quarter, 2019. Technical report. Washington, DC: National Association of Home Builders.

41.

Osama

Albitar

Sayed

, et al. (2020) Determining if walkability and bikeability indices reﬂect pedestrian and cyclist safety. Transportation Research Record: Journal of the Transportation Research Board 2674(9): 767–775.

42.

Pinjari

Pendyala

Bhat

, et al. (2007) Modeling residential sorting eﬀects to understand the impact of the built environment on commute mode choice. Transportation 34: 557–573.

43.

Salon

Boarnet

Handy

, et al. (2012) How do local actions aﬀect VMT? A critical review of the empirical evidence. Transportation Research Part D: Transport and Environment 17(7): 495–508.

44.

Sampson

Morenoﬀ

Gannon-Rowley

(2002) Assessing ‘neighborhood eﬀects’: Social processes and new directions in research. Annual Review of Sociology 28(1): 443–478.

45.

Sanbonmatsu

Ludwig

Katz

, et al. (2011) Moving to Opportunity for fair housing demonstration program. Final impacts evaluation. Technical report. Washington, DC: Department of Housing and Urban Development.

46.

Sanchez

(1999) The connection between public transit and employment: The cases of Portland and Atlanta. Journal of the American Planning Association 65(3): 284–296.

47.

Sanchez

Shen

Peng

Z-R

(2004) Transit mobility, jobs access and low-income labour participation in US metropolitan areas. Urban Studies 41(7): 1313–1331.

48.

Shoup

(2005) The High Cost of Free Parking. Washington, DC: Planners Press.

49.

Stevens

(2017) Does compact development make people drive less? Journal of the American Planning Association 83(1): 1–12.

50.

Van Dijk

(2019) The socio-economic consequences of housing assistance. Working Paper, University of Chicago.

51.

Van Wee

(2009) Self-selection: A key to a better understanding of location choices, travel behaviour and transport externalities? Transport Reviews 29(3): 279–292.

52.

Weinberger

(2012) Death by a thousand curb-cuts: Evidence on the eﬀect of minimum parking requirements on the choice to drive. Transport Policy 20: 93–102.

53.

Zegras

(2010) The built environment and motor vehicle ownership and use: Evidence from Santiago de Chile. Urban Studies 47(8): 1793–1817.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.28 MB