How land use patterns keep driving cheap: Geographic support for transportation taxes

Abstract

Chinese

Economists tend to favour price-based approaches, such as gasoline and carbon taxes, to address the negative impacts of car travel, while urban planners tend to emphasise land use planning such as compact development. In this paper, we argue that the two approaches are synergistic. We use precinct-level data from two California referenda to show that land use planning makes pricing more feasible: voters in dense, transit-oriented neighbourhoods are more willing to support a carbon price and increased gasoline taxes. Political ideology is a more important determinant of voting patterns, but in a closely divided election, land use patterns, public transportation, and other aspects of the built environment can determine the success of a referendum on driving taxes. Our results also imply that the voluminous research on land use and transportation underestimates the long-run impacts of compact development on driving, through ignoring the ways in which urban form shapes the politics of taxation.

Keywords

built environment environment/sustainability land use driving taxes transport

Introduction

Urban planners often prescribe compact, walkable development as one solution to reduce the negative impacts of car travel. As higher density, mixed-use neighbourhoods with connected street networks and frequent transit are associated with less car ownership and use (Ewing and Cervero, 2010; Stevens, 2017), land use regulations that encourage or require such compact development patterns can help reduce greenhouse gas emissions and other forms of pollution (Ewing et al., 2008). Many urban planners and designers writing about urban sprawl do not even mention the role of prices and taxes, preferring to focus on physical design and regulatory solutions (e.g. Talen, 2015).

Policies that promote higher densities may be as much about removing existing distortionary regulations that limit heights or mandate parking as introducing new ones. Indeed, scholars across disciplines have highlighted the role of exclusionary zoning and other land use restrictions that hinder infill development in existing urban neighbourhoods (Glaeser and Ward, 2009; Levine, 2005).

Economists, however, tend to emphasise price-based rather than regulatory approaches. To the extent that carbon emissions, local air pollution, or congestion are excessive, the appropriate solution is to raise prices to marginal social cost through a carbon tax, increased gasoline taxes, or congestion pricing (Brueckner et al., 2001; Glaeser and Kahn, 2004). In turn, higher prices for car travel would soften consumer demand for lower-density, car-oriented housing.

Several studies point to the synergies between pricing and the built environment. For example, pricing may have greater impacts on travel behaviour in mixed-use and higher-density neighbourhoods with good public transportation where people have a wider range of realistic alternatives to the private car (Boarnet, 2010; Guo et al., 2011).

A further potential synergy between pricing and regulation has attracted less attention and is the subject of this paper: how land use patterns affect the feasibility of taxing car travel. Urban economists tend to sidestep the question of where taxes come from, but there is no benevolent social planner who sets taxes on gasoline, carbon and congestion at the socially efficient level. Rather, such taxes are determined by elected officials, who in turn are subject to a variety of political pressures from their constituents, political parties, and campaign contributors, or by voters directly through referenda and ballot initiatives.

The political economy of carbon taxes and cap-and-trade has attracted considerable attention, particularly at the US federal level. For example, Holland et al. (2015) seek to explain why cap-and-trade or Pigouvian taxes are shunned in favour of less efficient alternatives such as subsidies for biofuels. They find the answer lies in the distribution of gains and losses: both cap-and-trade and biofuel subsidies spread their costs fairly evenly throughout the population, but biofuel subsidies also concentrate gains in agriculturally oriented Congressional districts – creating a strong constituency behind that policy. Other research shows that Congressional representatives are more likely to support cap-and-trade legislation if their districts have lower carbon footprints, presumably because their constituents would bear lower costs (Cragg et al., 2013). But there is little equivalent to this work in the transportation domain, where local and state gasoline taxes are typically more substantial than a carbon price. The few exceptions (e.g. Holian and Kahn, 2015) are discussed below.

In this paper, we respond to the call of Boarnet (2010: 588) who bemoans the ‘caricatured nature of current policy debate’ that pits pricing and regulatory approaches against each other rather than recognising their complementarities. We take a step towards reconciling the two approaches, and suggest that the willingness of a community to tax car travel depends on its land use patterns and transportation systems. Specifically, we hypothesise that elected officials and voters will be more willing to increase gasoline taxes in compact, walkable neighbourhoods that are well served by transit. When land use regulation promotes less car-oriented development, Pigouvian taxes on car travel become more feasible.

The land use/pricing connection

Transportation costs, particularly for gasoline, have a short-run impact on vehicle travel and emissions through influencing household decisions on whether, where, and how to travel. While elasticities are often modest in the short run, they are larger in the long run (Graham and Glaister, 2002; Small and Van Dender, 2007). This makes sense as households have more margins of adjustment over longer time frames, such as buying a car with greater fuel efficiency or moving closer to work.

Transportation costs also influence household travel through housing and land markets. The principle follows from canonical models of urban form, such as the Alonso–Mills–Muth model, which place transportation costs at the heart of the analysis. A well-known result is that at the city boundary, agricultural land rents are equal to urban land rents, which in turn are a function of transportation costs. The lower the transportation costs, the larger the radius of the city for a given population size, and thus the lower the density. Numerous studies support this result empirically, showing that higher gasoline taxes are associated with more compact and denser cities (McGibany, 2004; Molloy and Shan, 2013; Tanguay and Gingras, 2012). In turn, compactness and density are associated with less vehicle travel (Stevens, 2017).

In principle, however, there is a two-way relationship between urban form and transportation prices if urban form affects the willingness of voters and elected officials to support increased driving taxes – whether directly through increased gasoline taxes or road-user charges, or indirectly through carbon pricing. The more that a voter drives, the greater the costs they will incur from increased taxes – particularly if alternatives to driving are slower or more expensive. Thus, voters in places with lower per-capita vehicle travel and higher accessibility by public transportation, walking, and cycling would be expected to support higher taxes on car use, either directly through referenda or indirectly through voting for elected officials who promise to raise such taxes.

The relationship between urban form and political support for driving taxes is complicated by several factors. First, fuel efficiency affects the costs that a household incurs from a carbon or gasoline tax, albeit in a way that reinforces the effects of urban form: denser urban areas tend to have more fuel-efficient cars as well as less vehicle travel (Cook et al., 2015). Second, not only the costs but also the benefits of increased driving taxes through reducing congestion and air pollution externalities will vary spatially. Third, political support for driving taxes will depend on the proposed use of revenue. A tax that funds highway expansion might attract more support in car-oriented, low-density neighbourhoods, while one that funds public transportation might gain more backing in transit-oriented, high-density places. Similarly, road pricing might attract more political support were the revenue to be distributed to the cities through which the tolled freeways pass (King et al., 2007).

In short, a reasonable assumption is that the costs of driving taxes fall disproportionately on the lower-density neighbourhoods where people drive the most, while it is harder to generalise about the benefits. Empirically, the willingness of voters and elected officials to increase the cost of driving appears to be greatest in places where vehicle travel is limited. Anecdotal evidence comes from road-user pricing, which, despite the prescriptions of economists since the days of Vickrey (1963), has only been implemented at scale by four major cities – Singapore, London, Stockholm, and Oslo. Not surprisingly, all four have unusually low rates of car ownership and/or driving.

More systematic empirical studies are scarce, but two studies are notable. In Stockholm’s congestion pricing referendum, ideology partly determined voting choices, but support for the tolls was greatest among voters who would benefit the most from reduced congestion and face the lower monetary costs (Hårsman and Quigley, 2010). Meanwhile, Holian and Kahn (2015) study California’s Proposition 23, which would have effectively abolished the state’s cap-and-trade programme. They find that support for the voter initiative varied with political ideology but also with urban form, with suburban voters facing higher costs from a carbon price. Electoral precincts with higher residential density and that are closer to the city centre saw more votes in favour of carbon pricing (i.e. against Proposition 23). However, the authors were unable to distinguish between the effect of energy costs (larger suburban homes require more energy for heating and cooling) and transportation costs.

In contrast, other studies centre the role of ideology and the related concept of partisanship in explaining transportation policy preferences for taxes and expenditures. Manville and Cummins (2015) compile survey data on support for public transit funding, and find that attitudes on broader policy issues such as the environment are more predictive than whether an individual rides transit themselves. The survey findings of Nixon and Agrawal (2019) offer a similar conclusion: partisan identification is the dominant factor affecting support for a gasoline tax increase. Using data on local referenda, Congressional roll call votes, and public opinion surveys, Nall (2018) finds that density has little explanatory power once political partisanship, income, and race are accounted for. Support for highways is broad-based and bipartisan and varies little with urban form, while Democratic voters disproportionately favour pedestrian and public transit improvements.

Klein et al. (2022), meanwhile, take the partisan divide as a starting point and use survey data to explore the pathways that connect partisanship with policy preferences, including self-interest and transportation-related values. Independent of partisanship, they find a strong role for self-interest (proxied by travel behaviour) in affecting views of whether mode shift to transit, walking, and cycling should be a transportation policy goal.

Taken together, these studies indicate that ideology, or at least partisanship, is a dominant influence on support for gasoline and carbon taxes. But is there a role for urban form and public transportation service as well? Most of the studies that consider urban form (e.g. Klein et al., 2022; Nall, 2018) focus on broader aspects of transportation policy such as mode shift goals and highway and public transit spending, sometimes supported by local sales taxes. We now turn our focus to the relationship between urban form and policies such as carbon or gasoline taxes that would increase the cost of driving. We use the term ‘urban form’ in its broad sense, to encompass public transit service and transportation infrastructure as well as the physical layout of land uses. This is in line with the broader literature, which considers access to public transit as one aspect of the built environment (e.g. Ewing and Cervero, 2010).

Research design

Causal pathways

We examine the impact of urban form on support for two California ballot measures – Proposition (Prop) 23 in 2010 and Proposition 6 in 2018 – that would have repealed carbon pricing (Prop 23) and gasoline tax increases (Prop 6) that had been enacted by the state legislature. For simplicity, we refer to both as ‘driving taxes’ in the remainder of the paper. We pay particular attention to separating out the impact of urban form from that of ideology, given the confounding influence of liberal voters tending to cluster in larger, denser cities (Nall, 2018; Rodden, 2019). Specifically, we model the impact of urban form variables on support for each of the two propositions, after controlling for vote share in the Governor’s race and on a separate ballot measure that would have increased non-transportation spending or made it more difficult to raise taxes. Our multi-pronged identification strategy, discussed in more detail below, relies on controlling for ideology, as observed through voting on different contests in the same election.

Figure 1 shows the causal pathways that are considered (solid lines) and omitted (dashed lines) from the study. Urban form influences ideology through contextual effects; for example, social interactions with neighbours, housing tenure, and collective consumption change people’s attitudes and political identification (Martin and Webster, 2020; Pattie and Johnston, 2000; for broader discussions, see Weaver, 2014; Williamson, 2008). Such contextual effects are likely to be weak (e.g. Walks, 2006), but to the extent they exist, they will bias our estimates downwards because we do not capture the upper path in Figure 1 which runs through ideology. For example, people moving to a denser neighbourhood may become less conservative (Martin and Webster, 2020), perhaps because they are influenced by the political beliefs of their neighbours or due to social pressure to conform (Brown, 2023). This means that our results can be interpreted as a lower bound. A stronger effect of urban form on ideology comes through residential sorting (i.e. self-selection): people move to neighbourhoods that match their political leanings and preferences for collective versus individual service provision (Nall, 2018). We address the confounding impacts of self-selection through careful attention to controlling for ideology, as discussed below.

Figure 1.

Causal path diagram.

In the long run, ideology also affects urban form (e.g. Kahn, 2011). We are unable to capture this effect, which also means that our estimates of the impacts of urban form on driving taxes may be biased downwards since we control for or match on ideology. However, this effect of ideology on urban form is most likely to be pronounced in the long run; much of urban California was built up at a time when ideological alignments and spatial patterns of political preferences were very different from those of today. While future work might be able to untangle these multiple potential pathways, our objective in this paper is more modest – to explore the existence of a direct connection, independent of ideology, between urban form and support for increased driving taxes.

Another causal path shown in Figure 1 is through the proposed uses of revenue. However, we assume that these are independent of urban form for the two measures we study here. Opposition to both taxation measures focussed on the cost of gasoline and energy, while support emphasised clean energy (in the case of Proposition 23) and road maintenance (in the case of Proposition 6).

Dependent variables: Two California ballot measures

Proposition 23, rejected by the state’s voters in November 2010 by a margin of 62–38%, would effectively have abolished the state’s cap-and-trade programme and much of its other climate policy apparatus (Biber, 2013). At the time of the election, cap-and-trade had not yet taken effect, but California Air Resources Board projections at the time indicated that it would increase gasoline prices by 6% or $0.18 per gallon (Burtraw et al., 2012).¹ Transportation was one of the most salient aspects of the Proposition 23 campaign, and oil companies provided almost all of the campaign funding on the ‘Yes’ side (Biber, 2013). While the spatial variation in Proposition 23 support has been previously studied by Holian and Kahn (2015), among others, here we provide richer measures of urban form and more systematic controls for political ideology.

Proposition 6, which was on the ballot in November 2018, was also rejected by the state’s voters, this time by a margin of 57% to 43%. It would have effectively repealed the $0.12 per gallon gasoline tax increase that had been approved by the state legislature a year earlier, along with other fees and taxes on transportation fuels and vehicles that were part of Senate Bill 1. The revenue is primarily dedicated to maintenance of local streets and roads, with smaller amounts earmarked for public transportation and bicycle and pedestrian facilities. In disclosure, Senate Bill 1 also funds university transportation research, including some of the authors’.

Treatment variables: Measures of urban form

Our treatment variables consist of five dimensions of urban form: residential density, job density, the working population/jobs equilibrium index (which we call the population–jobs index for short in the remainder of this paper), transit frequency, and street connectivity. A large literature shows that higher values for all five variables are associated with less vehicle travel (Barrington-Leigh and Millard-Ball, 2020; Ewing and Cervero, 2010; Salon et al., 2012; Stevens, 2017), likely because of shorter travel distances that enable walking and cycling, the economies of scale of public transportation, and higher parking costs in dense neighbourhoods. We take the natural log of residential density, job density, and transit frequency (the other two variables are not as heavily right-skewed and no transformation is necessary). We standardise all five variables, and allow for non-linear relationships through including a squared term. The Appendix defines each variable and provides summary statistics.

Each of these five variables provides an incomplete picture of urban form, and the theoretical relevance of some of them has been questioned in previous research. For example, the population–jobs index, which captures the balance between the working age population and jobs, addresses only one aspect of the ultimate goal of increasing accessibility, and accessibility itself may be more important for providing households with more residential and transportation choices, rather than reducing car use and congestion (Levine, 1998). Our interest here is less in the specific contribution of each individual measure, and more about whether we can trace a link between any (or all) aspects of urban form and political support for driving taxes.

Data sources

The raw vote counts, precinct boundaries, and geographic conversion files are taken from the California Statewide Database (available at https://statewidedatabase.org). Four of our urban form measures are from the EPA Smart Location Database (available at https://www.epa.gov/smartgrowth/smart-location-mapping#SLD). Street connectivity is calculated based on the method in Barrington-Leigh and Millard-Ball (2019); we multiply their measure of street disconnectivity (SNDi) by −1 in order to align it with our other measures of urban form.

The scale of our analysis is the voter precinct.² The four urban form measures from US EPA are reported by census block group, which we rescale to precincts based on population weights.³ Voting data and street connectivity are already aggregated to the precinct level. While an individual-level analysis would offer considerable advantages in avoiding ecological inference challenges, such data are not available. Moreover, the pitfalls of ecological inference are more limited in our analysis since voters in the same precinct experience similar urban form regardless of individual demographic characteristics. (A similar ecological approach is employed by Hårsman and Quigley (2010) in their precinct-level study of the congestion pricing referendum in Stockholm.)

Regression specifications

The primary consideration in our regression specifications is to separate out the influence of urban form on support for driving tax increases from the influence of political ideology. Our first two specifications respectively include linear and cubic controls for ideology, as measured through both the Governor’s race in that year and a voter initiative on a non-transportation spending measure: Proposition 1 in 2018, which would have issued $4 billion in bonds for housing and veterans’ programmes, and Proposition 26 in 2010, which would have introduced a two-thirds supermajority requirement for most future voter-approved taxes. A “yes” vote on Proposition 1 and a “no” vote on Proposition 26 indicate a more fiscally liberal ideology, and capture the overall willingness of voters to support tax increases. The gubernatorial vote share provides a multidimensional measure of ideology.

Our third specification uses a propensity score matching algorithm. Matching methods aim to achieve covariate balance and reduce the importance of parametric assumptions, but are typically used in the context of a binary treatment. Because our treatment (urban form) is continuous, we use the covariate balancing generalised propensity score method proposed by Fong et al. (2018). We use the CBPS R package to match on two different treatments: residential density and job density. In essence, we weight each observation to construct a sample that minimises the weighted correlation between the treatment (residential density or job density) and other covariates. The major advantage of matching is to increase robustness to nonlinear relationships and other forms of model misspecification. While matching cannot control for unobserved confounders, our primary concern is to control for ideology, which we do observe in the form of voting on the gubernatorial race and a non-transportation spending measure.

Our fourth specification allows the urban form of neighbouring precincts to influence voting decisions through a spatial lag. For example, a voter’s travel costs and thus willingness to support tax increases could be affected by the mix of uses and densities in nearby blocks, not just the characteristics of their home precinct.

We also conduct a series of robustness tests that add additional covariates: voting on two additional races (California Secretary of State and US Senate), and/or demographics (gender, race as measured by the percentage of white people, median age, percentage of the population that are children, and median income). We report these specifications in the Appendix, but our results are not qualitatively affected, likely because the impact of these variables on voting behaviour is already captured by ideology.

We do not wish to control for car ownership and other transportation outcomes because those decisions are partly a result of ideology and partly a result of urban form (in addition to other factors such as income and job location). The ideological aspects of car ownership are already addressed through controlling for ideology. To the extent that car ownership is also a product of urban form, controlling for car ownership would bias our estimates of the impacts of urban form.

As shown below, our regression specifications provide broadly similar results, indicating that our findings are robust to different functional forms and non-parametric relationships. However, in all four specifications, our measures of ideology are limited to those expressed through the Governor’s race and the non-driving tax-related propositions, and may not capture other dimensions of voter preferences. In short, we cannot rule out biases caused by unobservable aspects of ideology.

Results

Figure 2 shows the distribution of support for tax increases across the state. (Recall that a ‘no’ vote equates to support for leaving the tax in place.) The patterns reflect the political geography of California, with coastal areas, particularly in the north, strongly opposed to both measures, and the interior of the state offering more support. Our urban form metrics (Figure 3) show a slightly different pattern: while the coastal metropolitan areas of San Francisco and Los Angeles stand out for their higher residential and job densities, frequent transit, and connected streets, the same is not true of more rural coastal areas, even though the entire coast tended to vote against both propositions. Moreover, the population–jobs index (which measures the balance between working-age population and jobs) is greatest in the interior Central Valley and sparsely populated rural areas, perhaps as a result of more localised labour markets.

Figure 2.

Voting outcomes from Proposition 6 (2018) and Proposition 23 (2010). Results are plotted at the precinct level.

Figure 3.

Spatial variation in urban form metrics. Results are plotted at the precinct level. Variables are log-standardised, that is, they represent z-scores of the logged variable. White areas indicate no data.

The scatter plot matrices (Figures 4 and 5) confirm this picture. Visually, there is a strong association between each proposition and ideology, as measured through voting in the gubernatorial race and the non-transportation tax related measures on the ballot (Proposition 26 in 2010 and Proposition 1 in 2018). While there is also a bivariate association between voting on our two propositions and each of our measures of urban form, there is more variation (reflected in the scatter around the 45-degree line), and the strength of the association varies across the measures.

Figure 4.

Pairwise relationships between voting on Proposition 6 (top row) and ideological and log-standardised urban form variables.

Figure 5.

Pairwise relationships between voting on Proposition 23 (top row) and ideological and log-standardised urban form variables.

We now consider the relationship between each measure of urban form and voting on the two propositions, while controlling for the effects of ideology as discussed above. Figure 6 plots the coefficients for the five models discussed in the Methods section, each of which offers a different way of controlling for the influence of ideology; the spatial lag model also captures some spatial spillovers. The Appendix provides the same results in tabular form.

Figure 6.

Coefficient plots. We show five different model specifications, discussed in the text, that control for the influence of ideology in different ways.

The first point to emphasise is the general consistency between the five models in terms of the sign and magnitude of the coefficients. In the spatial lag models, some of the effect of each treatment variable is picked up through the spatially lagged term, indicating the importance of urban form in the broader neighbourhood, not just the voter’s precinct of residence. The Akaike Information Criterion (AIC) scores (reported in Tables A-4 and A-5) favour the models with flexible controls, but AIC is a measure of fit whereas our primary interest is in separating out the influence of urban form.

Second, residential and job density have the largest consistent effects, predicting support for taxes in both the Proposition 6 and 23 referenda. Transit frequency and the population–jobs index have predictive power for Proposition 6 but not 23, while street connectivity has a small negative effect.

Third, nonlinear effects are apparent, especially in the Proposition 6 models; the squared terms for residential and job density and for transit service are positive and significant, even though in most cases the linear coefficient is negative or indistinguishable from zero. For example, higher residential density and transit frequency are only associated with opposition to Proposition 6 above about 3.5 housing units per acre and 5.7 peak-hour transit trips per square mile. Plausibly, an increase from low to moderately low densities and transit service has little impact on mode choice or travel costs and thus support for driving taxes, with the impact only felt at moderate to higher levels.

An alternative and more intuitive way of examining the impacts of urban form on voting is through the predictive plots shown in Figure 7. In each precinct, we predict Proposition 6 and 23 vote shares using our base model and the actual values of each urban form metric, but with ideology adjusted to one of two predefined levels. The blue markers and associated lowess fit show an election in which every precinct voted for the Governor and the non-transportation taxation measure (Proposition 1 in 2018 and Proposition 26 in 2010) according to the statewide means. The orange markers and lowess fit show an election in which there was a 50:50 split in voting for the gubernatorial candidate and the non-transportation taxation measure.

Figure 7.

Predictive plots. Each point represents our model’s predicted outcome for that precinct, with our ideological measures (e.g. the Governor’s race) adjusted to match the statewide means (blue) or a 50:50 election (orange).

The plots in Figure 7 are a useful complement to the coefficient plots (Figure 6) for two reasons. First, the plots in Figure 7 make any non-linearities readily apparent. Second, they account for the correlations between our measures of urban form. Residential and job density, for example, often covary: low-density residential areas rarely have much in the way of employment. Density and street connectivity are also associated with transit service, given that limited demand makes it hard to provide frequent transit to low-density neighbourhoods with culs-de-sac and other disconnected street patterns. Indeed, the land use-transportation literature often emphasises the synergistic effects of different aspects of urban form (e.g. Ewing and Cervero, 2010). Thus, rather than seeking to isolate the contribution of each measure of urban form, Figure 7 accounts for their combined impacts.

Two major conclusions are evident from the predictive plots in Figure 7. First, the ideological gap – represented by the difference between the blue and orange lines – is striking in the case of Proposition 6, but almost absent in the case of Proposition 23. Proposition 6 failed because of the clear ideological leaning of the electorate; the Democratic gubernatorial candidate, Gavin Newsom, won with 62% of the vote. Had the Governor’s race and Proposition 1 been evenly split (orange line), Proposition 6 would have passed in the majority of precincts. Proposition 23, in contrast, would have failed in almost every precinct under such a 50:50 election. This may be due to a legacy of bipartisan support for climate policy in California that remained through 2010; after all, AB 32, which Proposition 23 would have effectively repealed, was signed into law by Republican governor Arnold Schwarzenegger.

Second, the population–jobs index and street connectivity appear to be less useful predictors of support for driving taxes, at least in a consistent manner across the two propositions that we study. This is somewhat surprising given that the literature implies that these measures should have similar effects to density and transit service. In contrast, residential density, employment density, and transit frequency do emerge as useful predictors. For example, moving from the 1st percentile of residential density to the 99th percentile increases opposition to Proposition 6 from 57% to 60%, holding our ideological measures at their statewide means. For transit frequency, the effect of a shift from the 1st to the 99th percentile is an increase in opposition from 56% to 60%. However, Figure 7 indicates that the impact is not necessarily linear; the major gains in support come from moving from moderately high to very high levels of density and transit service.

Conclusion

Compact development and transportation pricing have often been considered as separate policy approaches for addressing the negative impacts of car travel. Manville (2017) laments the focus of planners and planning researchers on how compact development affects driving, calling instead for an emphasis on pricing reforms such as eliminating subsidies for driving through free parking and free roads.

However, cities have withstood more than 50 years of economic consensus about the desirability of congestion pricing. Once approved, pricing may be a rapid and efficient way to force drivers to internalise the safety, environmental, and other externalities of driving. But the day of that approval may never come.

Much has been written about how auto-oriented urban form locks in car dependency through path dependencies in land use patterns and public transportation systems. From a political economy perspective, such patterns may also lock in cheap driving, or at least make it more difficult to price roads and gasoline correctly. The literature already shows how higher gasoline taxes lead to more compact development. Here, we demonstrate that the relationship is two-way: compact development and frequent transit can also lead to voter support to increase taxes on driving. The precise contributions of residential density, employment density, frequent transit, and so on are hard to separate out empirically, but in practice, these measures of urban form and public transportation are highly correlated and mutually reinforcing.

The magnitude of the effect is modest, but could make the difference in a close election. Moreover, we might expect small effects in our study because voters were choosing whether to keep already enacted taxation increases. The taxes may also have had relatively low salience, being bundled into the price at the pump rather than a separate charge as with, for example, a congestion pricing scheme.

While our results are limited to two California referenda, it is reasonable to infer that the effect also exists where taxes are determined legislatively rather than through referenda. To some extent, council members and other elected officials respond to the preferences of their constituents.

The policy implications of our findings are indirect, but reinforce the case for creating urban neighbourhoods where people have a range of travel options, not just the private car. Our results imply that the voluminous research on land use and transportation underestimates the long-run impacts of compact development and public transportation on driving. Such studies are typically unable to capture the indirect effect of compact development on increasing political support for taxes on driving.

Such taxes are already politically salient, as in the case of mass protests against fuel taxes in places such as France and the United Kingdom. However, the political economy of driving taxes is likely to become more critical in the coming years with the rapid adoption of electric vehicles, which do not pay the driving taxes levied on gasoline. While the technology was still embryonic, there was a case for subsidies, but as electric vehicles become mainstream, taxing travel based on miles driven becomes more urgent. One rationale relates to replacing the lost revenue from gasoline taxes (Wachs, 2009). Another is to price externalities; while electric vehicles are generally preferable from a climate policy and air quality perspective, their greater weight exacerbates road safety problems (Shaffer et al., 2021), and they of course congest the roads just as much as vehicles powered by internal combustion.

New taxes on vehicle travel might attract limited support at the national scale, where car-dependent rural and suburban residents might understandably object given that they have little alternative to driving. But in the same way that congestion pricing has gained most traction in places where public transportation is frequent and reliable and walking is an option for most trips, mileage fees or other replacements for gasoline taxes might be rolled out first as pilots in urban centres. Our results point to a geographic strategy to roll out new pricing schemes: target pockets of support in cities that are more dense, walkable and transit-oriented.

Footnotes

Appendix

Table A7.

Robustness tests for Prop. 23

	Linear	Election	Census	Election + Census	D2B_E5MIX	D2A_JPHH	D4A	D3A
	Base model	Additional controls			Alternative dependent variables
		Added controls (SEN and SOS)	Added controls (demographics)	Added election + census controls	Different jobs-housing balance measures		Different transit measure	Different street network measure
D1A	0.012 (0.001)	0.011(0.001)	0.0098 (0.001)	0.0082 (0.001)	0.013 (0.001)	0.008 (0.002)	0.012 (0.002)	0.0047 (0.002)
D1C	0.0063 (0.002)	0.0045 (0.002)	0.0049 (0.002)	0.0037 (0.002)	0.0051 (0.001	0.013 (0.002)	0.0067 (0.002)	0.0031 (0.002)
D2A_JPHH						−0.0094 (0.002)
D2B_E5MIX					0.0028 (0.001)
D2C_WREMLX	−0.0012(0.001)	−0.000035(0.001)	0.001(0.001)	0.0013(0.001)			−0.001(0.001)	−0.001(0.001)
D4A							0.0007(0.001)
D4D	−0.0007(0.001)	−0.0021 (0.001)	0.0007(0.001)	−0.000062(0.001)	−0.0006(0.001)	−0.0009(0.001)	−0.0077 (0.001)	−0.0018 (0.001)
D3A								0.0076 (0.002)
NegSNDi	−0.0074 (0.001)	−0.0073 (0.001)	−0.0004(0.001)	−0.0005(0.001)	−0.0072 (0.001)	−0.0069 (0.001)	−0.0018 (0.001)
D1A²	−0.0006(0.001)	−0.0006(0.001)	−0.0007(0.001)	−0.001 (0.001)	−0.0007(0.001)	−0.0004(0.001)	0.000092(0)	0.0017 (0.001)
D1C²	0.0004(0)	0.0007(0)	−0.0008(0)	−0.0007(0)	0.0006(0)	−0.0004(0)		0.0009 (0)
D2A_JPHH²						0.0019 (0.001)
D2B_E5MIX²					0.0009(0)
D2C_WREMLX²	0.0005(0.001)	−0.0002(0.001)	−0.0007(0.001)	−0.0012(0.001)			0.0004(0.001)	0.0004(0.001)
D4A²							0.0026 (0.001)
D4D²	−0.0026 (0.001)	−0.0023 (0.001)	−0.003 (0.001)	−0.0026 (0.001)	−0.0026 (0.001)	−0.0024 (0.001)		−0.0031 (0.001)
D3A²								−0.0008(0.001)
NegSNDi²	0.000058(0.001)	−0.004(0)	0.00006(0.001)	−0.0005(0)	0.000063(0.001)	0.000095(0.001)	−0.000034(0.001)

These tests explore the sensitivity of the results to different specifications. Standard errors are given in parentheses. Bolding indicates p < 0.05.

Acknowledgements

We appreciate comments on earlier drafts from Chris Barrington-Leigh, Michael Manville, and Noam Rosenthal and from three anonymous reviewers.

Declaration of conflicting interests

The author(s) 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 made possible through project funding (2021-65) received by the University of California Institute of Transportation Studies Resilient and Innovative Mobility Initiative, funded by the State of California.

ORCID iDs

Adam Millard-Ball

Purva Kapshikar

Notes

References

Barrington-Leigh

Millard-Ball

(2019) A global assessment of street-network sprawl. PLOS ONE 14(11): e0223078.

Barrington-Leigh

Millard-Ball

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

Biber

(2013) Cultivating a green political landscape: Lessons for climate change policy from the defeat of California’s Proposition 23. Vanderbilt Law Review 66(2): 399–462.

Boarnet

(2010) Planning, climate change, and transportation: Thoughts on policy analysis. Transportation Research Part A Policy and Practice 44(8): 587–595.

Brown

(2023) Partisan Conversion Through Neighborhood Influence: How Voters Adopt the Partisanship of their Neighbors. Working Paper. Available at: https://jacobrbrown.com/#featured (accessed 13 September 2023).

Brueckner

Mills

Kremer

(2001) Urban sprawl: Lessons from urban economics. Brookings-Wharton Papers on Urban Affairs 2001: 65–97.

Burtraw

McLaughlin

Szambelan

(2012) California’s New Gold. A Primer on the Use of Allowance Value Created under the CO₂ Cap-and- Trade Program. Discussion Paper 12-23, May. Washington, DC: Resources for the Future. Available at: https://www.rff.org/publications/working-papers/californias-new-gold-a-primer-on-the-use-of-allowance-value-created-under-the-co2-cap-and-trade-program/ (accessed 1 August 2022).

Cook

Sanchirico

Salon

, et al. (2015) Empirical distributions of vehicle use and fuel efficiency across space: Implications of asymmetry for measuring policy incidence. Transportation Research Part A Policy and Practice 78: 187–199.

Cragg

Zhou

Gurney

, et al. (2013) Carbon geography: The political economy of congressional support for legislation intended to mitigate greenhouse gas production. Economic Inquiry 51(2): 1640–1650.

10.

Ewing

Bartholomew

Winkelman

, et al. (2008) Growing Cooler: The Evidence on Urban Development and Climate Change. Washington, DC: Urban Land Institute.

11.

Ewing

Cervero

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

12.

Fong

Hazlett

Imai

(2018) Covariate balancing propensity score for a continuous treatment: Application to the efficacy of political advertisements. The Annals of Applied Statistics 12(1): 156-177.

13.

Glaeser

Kahn

(2004) Sprawl and urban growth. In: Henderson

Thisse

J-F

(eds) Handbook of Regional and Urban Economics. Vol. 4: Cities and Geography. Amsterdam: Elsevier, pp.2481–2527.

14.

Glaeser

Ward

(2009) The causes and consequences of land use regulation: Evidence from Greater Boston. Journal of Urban Economics 65(3): 265–278.

15.

Graham

Glaister

(2002) The demand for automobile fuel: A survey of elasticities. Journal of Transport Economics and Policy 36(1): 1–26.

16.

Guo

Agrawal

Dill

(2011) Are land use planning and congestion pricing mutually supportive? Journal of the American Planning Association 77(3): 232–250.

17.

Hårsman

Quigley

(2010) Political and public acceptability of congestion pricing: Ideology and self-interest. Journal of Policy Analysis and Management 29(4): 854–874.

18.

Holian

Kahn

(2015) Household demand for low carbon policies: Evidence from California. Journal of the Association of Environmental and Resource Economists 2(2): 205–234.

19.

Holland

Hughes

Knittel

, et al. (2015) Some inconvenient truths about climate change policy: The distributional impacts of transportation policies. The Review of Economics and Statistics 97(5): 1052–1069.

20.

Kahn

(2011) Do liberal cities limit new housing development? Evidence from California. Journal of Urban Economics 69(2): 223–228.

21.

King

Manville

Shoup

(2007) The political calculus of congestion pricing. Transport Policy 14(2): 111–123.

22.

Klein

Ralph

Thigpen

, et al. (2022) Political partisanship and transportation reform. Journal of the American Planning Association 88(2): 163–178.

23.

Levine

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

24.

Levine

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

25.

Manville

(2017) Travel and the Built Environment: Time for Change. Journal of the American Planning Association 83(1): 29–32.

26.

Manville

Cummins

(2015) Why do voters support public transportation? Public choices and private behavior. Transportation 42(2): 303–332.

27.

Martin

Webster

(2020) Does residential sorting explain geographic polarization? Political Science Research and Methods 8(2): 215–231.

28.

McGibany

(2004) Gasoline prices, state gasoline excise taxes, and the size of urban areas. Journal of Applied Business Research 20(1): 33–45.

29.

Molloy

Shan

(2013) The effect of gasoline prices on household location. The Review of Economics and Statistics 95(4): 1212–1221.

30.

Nall

(2018) The Road to Inequality: How the Federal Highway Program Polarized America and Undermined Cities. Cambridge, United Kingdom; New York, NY: Cambridge University Press.

31.

Nixon

Agrawal

(2019) Would Americans pay more in taxes for better transportation? Answers from seven years of national survey data. Transportation 46(3): 819–840.

32.

Pattie

Johnston

(2000) ‘People who talk together vote together’: An exploration of contextual effects in Great Britain. Annals of the American Association of Geographers 90(1): 41–66.

33.

Rodden

(2019) Why Cities Lose: The Deep Roots of the Urban–Rural Political Divide. New York, NY: Basic Books.

34.

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.

35.

Shaffer

Auffhammer

Samaras

(2021) Make electric vehicles lighter to maximize climate and safety benefits. Nature 598(7880): 254–256.

36.

Small

Van Dender

(2007) Fuel efficiency and motor vehicle travel: The declining rebound effect. The Energy Journal 28(1): 25–51.

37.

Stevens

(2017) Does compact development make people drive less? Journal of the American Planning Association 83(1): 7–18.

38.

Talen

(ed.) (2015) Retrofitting Sprawl. Addressing Seventy Years of Failed Urban Form. Athens: University of Georgia Press.

39.

Tanguay

Gingras

(2012) Gas price variations and urban sprawl: An empirical analysis of the twelve largest Canadian metropolitan areas. Environment and Planning A 44(7): 1728–1743.

40.

Vickrey

(1963) Pricing in urban and suburban transport. The American Economic Review 53(2): 452–465.

41.

Wachs

(2009) After the motor fuel tax: Reshaping transportation financing. Issues in Science and Technology 25(4): 85–88.

42.

Walks

(2006) The causes of city–suburban political polarization? A Canadian case study. Annals of the American Association of Geographers 96(2): 390–414.

43.

Weaver

(2014) Contextual influences on political behavior in cities: Toward urban electoral geography. Geography Compass 8(12): 874–891.

44.

Williamson

(2008) Sprawl, spatial location, and politics: How ideological identification tracks the built environment. American Politics Research 36(6): 903–933.