Does rationing really backfire? A critical review of the literature on license-plate-based driving restrictions

ABSTRACT

Policies limiting the number of days vehicles are permitted to circulate based on the last digit of their license plates have experienced a resurgence in popularity, particularly in Chinese cities. This paper provides a critical review of the literature on license-plate-based driving restrictions. Of the 235 papers reviewed, most (152) only briefly mention license-plate-based restriction programmes to describe contextual conditions or provide an example of a policy that influences driving or vehicle purchases. Reviewing forty empirical papers, we find a divided literature on whether and where license-plate-based driving restrictions reduce local pollution or congestion. Some differences in findings likely relate to differences in research design or outcome measurement. Variations in policy design, enforcement, and other local conditions also play an important role. We next review findings about the multiple legal and illegal strategies households employ in response to driving restrictions. The second- car hypothesis, which posits that restriction policies backfire and lead to increased local pollution due to households purchasing second cars with different final license-plate digits, has become particularly popular. Evidence for the hypothesis, however, is mixed. Households employ a range of other behavioural responses, such as shuffling driving trips to specific days and driving in lightly policed areas, that likely attenuate the effectiveness of license-plate-based driving restrictions. As a result, researchers and policymakers should not expect to find a 20% reduction in pollution or congestion from banning a fifth of vehicles from the road. Improving policy effectiveness will likely require policymakers to address intended and unintended behavioural responses through additional mechanisms.

KEYWORDS:

• Driving restriction
• second-car hypothesis
• license plate restriction
• air quality
• Hoy No Circula

Introduction

In response to increased urban air pollution and congestion, local and national governments have enacted a variety of programmes to limit the amount, location, and type of private vehicles allowed on public roads. License-plate-based driving restrictions, first implemented in cities, such as Mexico City, Santiago, and São Paulo, in the 1980s have experienced a resurgence in popularity after helping reduce Beijing's air pollution in the lead-up to the 2008 Beijing Summer Olympics. These restrictions limit the number of days that vehicles are permitted to circulate in a city or metropolitan area based on the last digit of their license plates. For example, Beijing's policy initially restricted automobiles from circulating on alternate days depending on whether the last number of the license plate was even or odd. After the Olympics, regulators adjusted the policy to restrict vehicles from circulating one day per week (Wang, Xu, & Qin, 2014).

Academic studies examining the effectiveness of license-plate-based driving restrictions in reducing congestion or pollution have produced mixed results. Eskeland and Feyzioglu (1997) and Davis (2008) concluded that Mexico City's Hoy No Circula restriction may have even increased pollution by encouraging households to purchase a second, older, and higher polluting car with a different last license plate digit to avoid the restriction. The notion that license-plate-based restrictions “backfire” and increase pollution has become pervasive throughout the academic literature, including in studies of Chinese cities where additional public policies may make purchasing a second car difficult. In Beijing, for example, the right to purchase a new or used car requires winning a low-probability lottery.

In addition to purchasing another car that is restricted on different days, drivers have a variety of available behavioural approaches to avoid the policy, such as adjusting daily or weekly driving schedules, avoiding highly policed areas, or purchasing unrestricted low-emissions vehicles (Gu, Deakin, & Long, 2017; Guerra & Millard-Ball, 2017; Wang et al., 2014). These behavioural responses may be substantially easier and less burdensome than purchasing a second car, particularly for low- and moderate-income households. If license-plate-based car restrictions are not working, it is important for regulators to understand why and adjust policies accordingly.

This paper provides the first systematic literature review of the empirical literature on license-plate-based driving restrictions. Through this review, we emphasise how policy design, urban context, and behavioural factors likely influence whether and by how much license-plate-based restrictions might reduce driving and associated pollution. The remainder of the paper is organised as follows. Section 2 describes our general approach to selecting, organising, and summarising academic papers that discuss license-plate-based restrictions. Section 3 summarises findings about the relationship between license-plate-based restrictions and local pollution. Section 4 provides a theoretical framework for how different behavioural responses to license-plate-based restrictions affect pollution and congestion and provides evidence from the literature on the nature and scale of these responses. The final section concludes with a summary of the results and key takeaways for researchers and policymakers.

Methodology

We conducted a general keyword search of papers that refer to “license plate restrictions” or “driving restrictions” using Google Scholar and the University of Pennsylvania's online academic search system Franklin. We then excluded papers that reference different types of driving restrictions or have not undergone peer review. This filtering reduced the total number of citations to the 235 papers included in our review. Most papers (152) only briefly mention license-plate-based restriction programmes, generally to describe contextual conditions or provide an example of a general policy to influence driving or vehicle purchases. For example, Zhang, Liu, Waller, and Yin (2019) list exemptions from license-plate-based restrictions as one of several possible policies to encourage the adoption of automated vehicles.

Of the remaining 83 papers, 40 provide empirical estimates of the effects of license-plate-based restrictions on pollution or congestion. We exclude simulations from this category since the findings depend on the simulation inputs. For example, Pu, Yang, Liu, Chen, and Chen’s (2015) finding that Hangzhou's license-plate-based restrictions reduced vehicle travel and pollution by 7–10% depends on the assumption that the policy reduced driving by 20% in the restricted zone at restricted times of day. The final category of papers includes simulations and papers studying the effects of restrictions on some other outcome, such as crime, transit use, or vehicle purchases. We discuss both categories of papers in greater detail in the section on behavioural responses to license-plate-based driving restrictions.

The effect of license-plate-based driving restrictions on pollution and congestion

The 40 papers that examine the effects of license-plate-based driving restrictions on pollution and congestion produce a range of results. Twenty papers report that license-plate-based driving restrictions decreased pollution or congestion, nine find no effect or an inverse effect, and eleven present mixed results (Table 1). In addition to the wide range of findings, the papers use a variety of research approaches in places with differently designed policies to examine a variety of outcomes. We discuss these differing outcomes and designs throughout this section and the next.

Table 1. Summary findings of 40 papers examining the results of license-plate-based driving restrictions.

Publication	Location	Main research approach	Main findings
Studies finding license-plate-based restrictions reduced pollution or congestion
Viard and Fu (2015)	Beijing	Before and after (odd-even and one day per week) examination of daily air quality index and hourly television viewership (regression discontinuity design)	Reduced pollution and increased television viewership (assumption related to less time spent commuting)
Liu, Yan, and Dong (2016)	Beijing	Before and after (odd-even and one day per week) examination of daily air quality index and hospitalisations (regression discontinuity design)	Reduced pollution and hospital admissions
Lu (2016)	Beijing	Before and after examination of reducing policy hours and later increasing penalty for violations on air quality index (regression discontinuity design)	Strengthening the policy associated with improved air quality. Weakening the policy associated with reduced air quality
Gu et al. (2017)	Beijing	Compares household behaviour on restricted days to household behaviour on unrestricted days using travel diaries	Decreased car trips on restricted days
Liu, Huang, Guo, and Guo (2017)	Beijing	Compares daily hospital admission for respiratory issues on high-restriction and low-restriction days using #4 "death" plate (roughly 14% instead of 21% banned)	Reduced hospital admissions on days with more cars banned
Zhang et al. (2017)	Beijing	Examine unique subway card data to identify users that only use the subway on one consistent day per week, assume these switched from driving due to policy, and estimate pollution reduction from removing trip	A small shift in trips from private cars to metro (0.12%) with a corresponding decrease in CO2 emissions
Zhong et al. (2017)	Beijing	Compares daily/hourly air quality index, traffic congestion index, and daily ambulance call data using “death” plate to compare high- and low-restriction days	Lower pollution, congestion, and hospital admissions on days with more vehicles banned
Yang et al. (2018)	Beijing	Compares daily public transit use, taxi use, and congestion using “death” plate to compare high and low restriction days	Reduced congestion and increased taxi and transit use on days with more restricted vehicles
Chen, Qin, et al. (2020)	Beijing	Examines shift in traffic congestion index using “death” plate design by geography and compares to shifts in travel speeds	Finds that traffic congestion index increases by 0.5–0.7 on days with fewer cars restricted
Liu and Kong (2021)	Beijing	Compares differences in changes in air pollution index on weekdays and weekends in Beijing relative to Tianjin using a differences-in-differences approach	Finds a 3%-to-5% reduction in pollution from the license-plate policy with another 8%-to-12% reduction attributable to other policies.
Rajabov et al. (2020)	Beijing	Compares multiple particulates using “death” plate to compare high- and low-restriction days	Finds reductions in AQI, PM2.5, PM10, CO, and NO2, but no difference in O3
Yang et al. (2020)	Beijing	Examines shift in vehicle density using “death” plate design by geography and compares to shifts in travel speeds	Statistically significant increase in vehicle density and decrease in travel speeds on days when fewer cars are restricted. Effects most pronounced in central locations
Zhang, Chen, et al. (2020)	Beijing	Compares household mode choice at restricted and unrestricted hours on weekdays and weekends using travel diaries	Residents travelling at restricted hours less likely to use a car relative to differences between the same time periods on unrestricted weekends
Zhang, Shan, Wang, Pang, and Guo (2020)	Beijing, Tianjin	Before and after examination of daily air quality index (regression discontinuity design)	Modest air quality improvement in both cities
Rao et al. (2017)	Delhi	Intercept survey asking respondents about effects of restriction on current and recalled travel behaviour	Increased speeds, shorter travel times, and shift from cars to transit
Mishra et al. (2019)	Delhi	Compares PM1.0 and PM2.5 on six days during temporary odd-even restriction to three days in weeks preceding restriction on three major arterials	Particulate matter decreased by an average of 5% on arterials during restricted hours
Han, Liu, and Lu (2020)	Jinan	Compares daily CO, PM10, and mortality rates before and after temporary restrictions	Reduction in CO (46.6%), PM10 (33.1%), and morality (0.313 per million people per percent of PM10)
Liu et al. (2018)	Langfang	Before and after (every other day vs. one day per week) comparison of non-compliance, travel speed, and traffic volumes by space and time of day (regression discontinuity design)	Traffic speeds increased with more frequent restriction
Carrillo et al. (2016)	Quito	Before and after examination of hourly CO in restricted zones and times relative to unrestricted times and zones	Improved air quality at restricted times and locations
Barahona et al. (2020)	Santiago	Compares the share of high-polluting vehicles in the fleet in Santiago relative to other municipalities over time (regression discontinuity design)	Santiago has significantly fewer older cars without catalytic converters and these emit less pollution on average
Studies with negative or null findings
Wang et al. (2014)	Beijing	Compares residents’ mode choice on days when household vehicle is restricted using travel diaries	No significant change in probability of travelling by car instead of transit, bike, or foot
Xiao, Chang, Guo, Chen, and Tian (2019)	Beijing	Before and after examination of daily air quality	No evidence of improved air quality
Bonilla (2019)	Bogota	Before and after examination of two phases (peak hours only and 6am-8pm restrictions) on hourly CO (regression discontinuity design)	No evidence of improved air quality
Chen, Zheng, et al. (2020)	Chinese cities (11)	Before and after comparison of annual PM10 and car registrations in eleven cities with restrictions compared to 100 without restrictions	Air quality worsened and car registrations increased in cities with restrictions
Ye (2017)	Lanzhou	Before and after examination of hourly air quality index (regression discontinuity design)	No evidence of improved air quality. Some evidence of worsened air quality
Eskeland and Feyzioglu (1997)	Mexico City	Before and after comparison of quarterly gasoline consumption	Increased gasoline consumption
Davis (2008)	Mexico City	Before and after examination of hourly, daily, and weekly air quality (regression discontinuity design)	No evidence of improved air quality. Some evidence of worsened air quality
Davis (2017)	Mexico City	Before and after examination of policy change to include Saturday restrictions on air quality (regression discontinuity design)	No evidence of improved air quality
Guerra and Millard-Ball (2017)	Mexico City	Compares the daily vehicle travel of households with a single banned car to that of similar households with a single car that is likely exempt because of vehicle age using travel diaries	No evidence of reduced driving from households with a banned car
Studies with mixed findings
Sun et al. (2014)	Beijing	Compares peak travel speeds and daily PM10 using “death” plate to compare high and low restriction days	Traffic speeds increased on days with more restricted vehicles, but no air quality improvement
Ma and He (2016)	Beijing	Before and after examination of hourly air quality index	Air quality improved in short run but worsened in the long run
Xie et al. (2017)	Beijing	Before and after examination of daily air quality index	Air quality improved in short run but not in the long run
Zhang et al. (2017)	Bogota	Before and after examination of hourly air quality measures (regression discontinuity design)	Decrease in NO but increases in NO2, NOx, and O3
Chen, Zhang, and Chen (2021)	Chinese cities (15)	Compares change in PM2.5 over time relative to 158 control cities using a differences-in-differences approach with propensity score matching	No evidence of reduced pollution across the full sample but variation in results across regions and city types with an 11% reduction in air pollution outside of provincial capitals
Chowdhury et al. (2017)	Delhi	Compares satellite-based PM2.5 estimates during a 15-day odd-even restriction to time periods immediately before and after the restriction	Reduces PM2.5 by 4–6%, which is within the 10% uncertainty range of the satellite-based estimates and therefore a null finding
Mohan et al. (2017)	Delhi	Compares PM2.5, traffic speeds, and congestion measures immediately before, during and after implementation of two-week odd-even scheme	Car volumes decrease by 7–19% but there is an increase in exempt auto rickshaws, two-wheelers, and buses. Speeds up around 5% on average and no measurable change in pollution
Joshi et al. (2016)	Delhi	Compares PM2.5 measures immediately before and after implementation of odd-even scheme	Pollution decreased immediately after policy implementation with a subsequent increase, likely due to forest fires
Huang et al. (2017)	Lanzhou	Before and after examination of hourly air quality measures (regression discontinuity design)	Air quality improved in short run but not in the long run
Gallego et al. (2013)	Mexico City	Before and after examination of daily air quality measures (regression discontinuity design)	Air quality improved in short run but worsened within a year
de Grange and Troncoso (2011)	Santiago	Before and after examination of long-term policy and short-term emergency measures on hourly vehicle flows and daily transit ridership	Long-term policy of excluding cars with catalytic converters not associated with reduction in driving. Short-term emergency measures extending the ban to all vehicles associated with reduced driving and increased metro use

For the twenty papers with positive findings, the most common research design is to examine measures of air quality from air quality monitoring stations before and after policy implementation while controlling for other temporal predictors of pollution, such as the weather and time of day. Regression discontinuity designs, with the year or month of policy implementation identified as the discontinuity, are particularly common. Several papers include additional control groups. For example, Carrillo, Malik, and Yoo (2016) use a differences-in-differences approach instead of a discontinuity design and conclude that Quito's license-plate-based restrictions reduced carbon emissions by 9–11% during peak traffic hours around monitoring stations within the restriction zone compared to monitoring stations outside of the zone. Liu and Kong (2021) use a differences-in-differences approach comparing Beijing to Tianjin to try to net out the effect of the license-plate-based restrictions from other policies to reduce local pollution. The authors conclude that this approach leads to a smaller estimated reduction in pollution than Viard and Fu’s (2015) estimates. While both studies find a 15–20% reduction in Beijing's local pollution, Liu and Kong (2021) only attribute the license-plate-based restriction with a 3–5% reduction in pollution.

The next most common research approach relies on a location-specific aversion to license plates ending in the number 4, which is associated with the Chinese word for death. On days that restrict the numbers 4 and 9 from circulating, Yang, Lu, Liu, and Guo (2018) report that just 14% of Beijing's cars are banned compared to 20–22% on other days. Days with fewer restricted vehicles are associated with a 3–20% increase in ambulance calls for heart and fever-related symptoms, 22% higher congestion, and 12% higher daily NO2 concentrations (Zhong, Cao, & Wang, 2017). The “death” plate method has recently also been used as an exogenous instrument to tease out causal relationships between congestion and travel speed (Yang, Purevjav, & Li, 2020) and congestion and pollution (Chen, Qin, Tan-Soo, Xu, & Yang, 2020). Other general approaches include comparing household-level travel behaviour on banned days (Gu et al., 2017), at banned times of day (Zhang, Chen, Du, & Wang, 2020), and before and after the policy (Rao, Madhu, & Gupta, 2017).

The nine papers that produce null or inverse findings use similar methodological approaches. The majority employ a before-and-after approach, generally a regression discontinuity design. Several also include additional controls using a differences-in-differences approach instead of a discontinuity design. For example, Chen, Zheng, Yin, and Liu (2020) find that pollution, car registrations, and pollution per car increased rather than decreased in eleven Chinese cities with license-plate-based driving restrictions compared to 100 cities without restrictions. Two papers use household-level survey data. Guerra and Millard-Ball (2017) find no substantial drop in vehicle travel for households with just one banned car compared to households with a similarly aged car that is eligible to be exempted from the policy after passing a tailpipe emissions check. Wang et al. (2014) use a similar approach to Gu et al. (2017) but do not find a statistically significant change in the probability of travelling by car on days when Beijing's residents are banned from driving.

The eleven papers with mixed results also generally use regression discontinuity designs or similar approaches to examine air quality before and after policy implementation. The most common mixed finding is that air quality improved immediately after the policy intervention but not in the long run (Gallego, Montero, & Salas, 2013; Huang, Fu, & Qi, 2017; Ma & He, 2016; Xie, Tou, & Zhang, 2017). However, the timing of restriction policies often coincides with other, sometimes substantial, temporal changes that are also likely associated with pollution and congestion. Claims about long-run influences of driving restrictions are not well-suited to regression discontinuity designs. The further the pollution data move from the days, weeks, and months immediately before or after policy implementation, the less likely that differences in pollution have anything to do with driving restrictions.

Several studies find improvements along one dimension but not along another. For example, Sun, Zheng, and Wang (2014) employ the “death” plate design and find increased traffic speeds but no change in emissions on days with more restricted vehicles. Types of pollutants may also matter. Zhang, Lin Lawell, and Umanskaya (2017) use a regression discontinuity design on pollutants in Bogota and find the policy associated with decreases in NO but increases in NO2, NOX, and O3. Rajabov, Liu, and Rajabov (2020) find Beijing's policy associated with statistically significant decreases in most pollutants and an overall air quality index but not in O3. Total pollution likely also matters. For example, Chowdhury et al. (2017) find Delhi's policy reduces PM2.5 by 4–6%, but that this is within the 10% uncertainty of the satellite-based estimates and therefore a null finding. However, Mohan, Tiwari, Goel, and Lahkar (2017) report that cars contribute just 5% of Delhi's total PM2.5 and that these reductions are offset by increases in unrestricted auto rickshaws, motorised two-wheelers and taxis. A recent study of Covid-19 restrictions across 325 Chinese cities finds that lockdowns reduce overall pollution by 12% relative to the cities without lockdowns, but that differences vary substantially by pollutant (Wang, Liu, & Zheng, 2021). For example, PM2.5 and PM10 decreased by 13–15%, but SO2, NO2, and CO only decreased by 3–4%.

Other researchers find different relationships across different types of policies. For example, de Grange and Troncoso (2011) find that Santiago's long-term policy, which excludes cars with catalytic converters, had no impact on the use of private cars, but that short-term emergency bans that extended the policy to all vehicles reduced car use by 5.5% and increased Metro use by 3%. Liu, Li, Wang, and Shang (2018) find that a tightening of Langfang's restrictions increased overall travel speeds, while Davis (2017) finds that extending the restriction policy to weekends as well as weekdays did not reduce local pollutants in Mexico City. Several recent studies examine the effects of a two-week temporary odd-even restriction in Delhi by comparing several days of direct measurements of traffic and pollution during the restriction compared to measurements before and after the restriction (Chowdhury et al., 2017; Joshi et al., 2016; Mishra, Pandey, Pandey, & Kumar, 2019; Mohan et al., 2017). These studies produce mixed findings, but generally rely on only a few data points.

A note on research design and findings

We do not exclude any peer-reviewed papers that examine license-plate-based restrictions based on research design. Although there is a range in research quality, we do not see a systematic difference in positive, negative, or null findings across studies with strong or weak designs. We do, however, generally find that studies from Beijing conclude that the policy reduced pollution and congestion, whereas studies from Mexico City do not. Studies from other geographies produce mixed results or do not have a large enough sample of papers to draw conclusions. In addition to differences in outcome measures and research design, variations in policy design, enforcement, and other local conditions likely influence the efficacy of license-plate-based driving restrictions. We discuss how and why policy differences likely influence the effects of license-plate-based driving restrictions throughout the remainder of this review.

Behavioural responses to license-plate-based driving restrictions

Figure 1 presents a diagrammatic summary of the interrelated ways that vehicle purchase decisions and travel behaviour on restricted and unrestricted days are likely to influence the effects of license-plate-based driving restrictions on pollution and congestion. For example, households that have a single restricted car may respond to the policy by switching modes, driving during unrestricted hours, violating the policy, or shuffling trips to other days. Some households might respond by choosing not to purchase a car or, depending on local context, purchasing a low-polluting exempt vehicle. This section summarises academic findings about how households respond to driving restrictions through vehicle purchases and travel behaviour on restricted and unrestricted days. We emphasise how these decisions relate to overall pollution and congestion. We also identify and highlight gaps in the existing literature and opportunities for future research.

Figure 1. License-plate-based restrictions’ relationships to vehicle purchase decisions, travel behaviour, congestion, and pollution.

Vehicle purchases

In addition to reducing the number of days residents drive their cars, license-plate-based driving restrictions likely influence vehicle purchase decisions.

The second-car hypothesis

Purchasing a second car with a different license plate is the most referenced behavioural response to driving restrictions. Although 145 papers reviewed mention this second-car hypothesis as a reason that license-plate-based restrictions fail or might fail to reduce pollution, just nine present empirical evidence. The results are mixed. Due to the popularity of the second-car hypothesis, Appendix A provides additional insight into its origins and diffusion, as well as the strength of the initial evidence.

Five studies compare shifts in vehicle sales or registrations before and after the policy implementation. Although all but one of these studies find modest increases in registrations, these studies also rely on time-series data with few observations immediately before and after policy implementation. As with Davis’ (2008) examination of Mexico City, Bonilla (2019) observes a modest increase in used vehicle sales after the tightening of Bogota's license plate restriction and concludes that the study provides mild evidence that the new policy increased used vehicle sales. Chen et al. (2020) provide additional spatial controls and find modest, statistically significant increases in car registrations in eleven Chinese cities with restrictions relative to other Chinese cities without restrictions. Guerra and Millard-Ball (2017) include controls for other Mexican metropolitan areas and conclude that vehicle registrations in metropolitan Mexico City increased no more quickly than in the rest of Mexico after Hoy No Circula's initial implementation.

Five papers, including Eskeland and Feyzioglu (1997), rely on household-level data and generally find a weak relationship between the policy and households’ second car purchases. For example, Guerra and Millard-Ball (2017) find that households with a restricted vehicle are no more likely to own a second car than similar households with a similarly aged but unrestricted vehicle. Gu et al. (2017) look at the travel behaviour of households with one restricted and one unrestricted car compared to households with two unrestricted cars. Although the data sample is small, it appears that households with one restricted and one unrestricted car increase the share of trips by car as one household member picks up and drops off others.

Barahona, Gallego, and Montero (2020) report that although more households own cars in Santiago than the rest of Chile, higher incomes and household structure explain this difference, not the car-restriction policy. Moncada et al. (2018) employ a similar methodology to look at changes in households’ probability of purchasing a second vehicle in Bogota and Villavicencio, Colombia, using household travel surveys from 1996 and 2011 in Bogota and 2008 and 2012 in Villavicencio. Including controls for working-age adults, children, household motorcycles, and inconsistent linear income categories, the authors find a 20% increase in the probability of purchasing an additional vehicle for residents of Bogota relative to residents of Villavicencio. Despite comparing a capital city of 7.5 million residents to a city of 500,000 at different time periods, the authors attribute the entirety of this 20% increase to Bogota's 1998 restriction policy and the policy's tightening a decade later.

In addition to the empirical analyses, Gu et al. (2017), Guerra and Millard-Ball (2017), and Barahona et al. (2020) also describe the small share of households that potentially avoid the policy through additional car purchases. In Beijing, only 2.2% of households own more than one car (Gu et al., 2017). In Mexico City, only 1.2% of households own at least two cars that are both likely to be subject to the driving restriction (Guerra & Millard-Ball, 2017). In Santiago, less than 5% of households own more than one car, and many own newer, exempt vehicles (Barahona et al., 2020). Although Moncada et al. (2018) do not present statistics on their dependent variable, just 3.3% of households reported having two or more cars or pickups on Bogota's 2011 household travel survey (Secretaría Distrital de Movilidad de Bogotá, 2013).

Other vehicle purchase decisions

Household members might also respond by choosing not to own a car or purchasing specific types of vehicles that are exempt from the driving restrictions. If a substantial number of households choose not to purchase cars or shed them, this would likely reduce metropolitan congestion and pollution. Using a contingent valuation approach, Blackman, Alpízar, Carlsson, and Planter (2018) estimate that Hoy No Circula reduces cars’ average annual value by about $130 per year. If cars are worth less, then residents are less likely to buy them. Based on this reduction in value, Eskeland and Feyzioglu (1997) estimate that about 8% of car-owning households would sell their cars in response to Hoy No Circula. Given its potential importance and car ownership rates in places like Mexico City and Beijing, there is surprisingly little research into the role that driving restrictions might play in reducing the probability of buying a car.

Vehicle exemption policies likely also play an important role. In Mexico City, for example, the policy changed to exempt cars with catalytic converters in 1996 and has since undergone several substantial changes. By 2017, most cars on the road in metropolitan Mexico City were exempt from general travel restrictions (INEGI, 2017). Comparing the age of the vehicle fleet across municipalities, Barahona et al. (2020) conclude that Santiago's vehicle exemptions reduced the share of older and high-polluting vehicles relative to other similarly wealthy municipalities. Chinese cities that exempt electric vehicles from driving restrictions have also seen a substantial increase in electric vehicle sales (Diao, Sun, Yuan, Li, & Zheng, 2016; Lu, Yao, Jin, & Pan, 2020; Rao, 2020; Wang, Tang, & Pan, 2017). Given the importance of vehicle exemptions on purchases, there is a need for additional research into how exemptions shape travel behaviour. For example, exemptions likely reduce average pollution per mile travelled but also encourage additional travel in newer vehicles with lower operating costs. There may also be important spatial and socioeconomic impacts. Only relatively wealthy households are likely to view purchasing a new, low-emission vehicle as a possible response to driving restrictions.

Travel on restricted days

For households without exempt cars, there are multiple potential responses to car travel on restricted travel days.

Mode choice

Perhaps the most studied response is switching modes. As with general findings about congestion and pollution, the evidence is mixed. In Mexico City, neither Davis (2008) nor Guerra and Millard-Ball (2017) find evidence of increased transit or taxi use. Others observe statistically significant increases in transit and taxi use in Delhi (Mohan et al., 2017) and Beijing (Gu et al., 2017; Yang et al., 2018). Zhang, Long, and Chen (2019) find that transit use increased after policy implementation in some but not all six Chinese cities and attribute these differences to differences in policy implementation. Cheng, Huang, Qu, Zhang, and Li (2020) conclude that taxi use increased most in wealthier areas with worse transit in Xi’an. Other general findings include increased home values near transit (Xu, Zhang, & Zheng, 2015), a small increase in shared electric bike use (Campbell, Cherry, Ryerson, & Yang, 2016), an increase in bike share (de Buen Kalman, 2021), and an increase in bicycle use (Gu et al., 2017).

Shuffling trips

In addition to influencing mode choice, license-plate restrictions might encourage drivers to avoid taking trips, shuffle trips to unrestricted hours, or shuffle trips to unrestricted days. Avoiding taking trips is relatively understudied, but Gu et al. (2017) find no evidence of reduced trip-making in Beijing. Shuffling trips outside of restricted hours appears to occur in some cities but may be a niche approach. In most cities with license-plate restrictions, only a small fraction of trips occurs outside of restricted hours. During Santiago's emergency restrictions, de Grange and Troncoso (2011) identify a statistically significant 3.5% increase in traffic during the hours prior to the ban. In Bogota, Zhang et al. (2017) observe an increase in NO2 during both restricted and unrestricted hours but a decrease in NO only during restricted hours. By contrast, Carrillo et al. (2016) conclude that Quito's decreased pollution is not offset by increases in pollution at unrestricted hours or unrestricted geographies. Similarly, neither Gu et al. (2017) nor Guerra and Millard-Ball (2017) find a statistically significant shift in driving outside of restricted hours using household travel data. Since most non-exempt cars are only used two to three weekdays per week, Guerra and Millard-Ball (2017) conclude that shuffling car trips from a restricted day to an unrestricted day is a particularly easy behavioural response to driving restrictions.

Cheating

Finally, many households may respond to travel bans by cheating. Although some authors assume near-universal policy compliance (Davis, 2008, 2017), most studies suggest that there is substantial non-compliance with driving restrictions (Guerra & Millard-Ball, 2017; Liu et al., 2018, 2020; Mohan et al., 2017; Wang et al., 2014). For example, Wang et al. (2014) find that nearly half of regulated car owners violated the restriction rules in metropolitan Beijing and that violations are more likely to occur during peak hours, on social trips, and on trips outside of the city centre. In an exception, Viard and Fu (2015) report high rates of compliance based on license plate data from a centrally located parking garage in Beijing. Central locations, however, are where the policies are most likely to be enforced (Guerra & Millard-Ball, 2017; Wang et al., 2014).

Travel on unrestricted days

Most unrestricted travel behaviour relates directly to restricted travel behaviour. For example, if someone cheats or shuffles trips to times immediately before the travel ban, driving behaviour on unrestricted days would likely be unaffected. If travel is shifted to unrestricted days, by contrast, driving would tend to increase on those days. Shifts in travel behaviour, however, could also have larger network effects. For example, if traffic decreases due to the policy, the resultant decrease in congestion would likely attract new drivers due to latent demand. Researchers have observed that increased driving tends to quickly fill and congest new highway investments and road expansions (Cervero & Hansen, 2002; Downs, 1992, 2004; Duranton & Turner, 2011).

Conclusion

In this paper, we review the empirical literature examining whether license-plate-based driving restrictions reduce pollution or congestion. Half of the studies reviewed find that the restrictions reduce pollution or congestion. A quarter find no effect or an inverse effect. A quarter present mixed results, such as short-term reductions with no long-term effect. In short, the academic literature does not indicate that this type of rationing backfires, at least not universally. While differences in findings likely relate to differences in research approach and outcome measurement, policy design and behavioural responses also play a role.

Only studies from Mexico City, the origin of the hypothesis that license-plate-based restrictions fail because households purchase additional cars, consistently find that restriction policies do not reduce pollution or driving. While the hypothesis has disseminated widely and offers a compelling and digestible narrative about why license-plate-based restrictions do not work, this assertion rests on shaky empirical grounds. As shown in Appendix A, the seminal papers fail to show a clear causal relationship between Hoy No Circula and increased vehicle purchases. Moreover, metropolitan residents of Mexico City have been able to purchase fully or partially exempt vehicles for the past 25 years. Two-thirds of private cars on the road in metropolitan Mexico are exempt from the policy outside of environmental emergency days (INEGI, 2017).

As a result, adjusting a specific policy to address second-car purchases will likely not do much to change the policy's ability to reduce local congestion and pollution. There are many other simpler and less expensive ways to avoid or adjust to the policy than purchasing a second car. Many of these adjustments, such as shuffling car trips to different days or hours, likely reduce policy effectiveness. Better understanding these behavioural responses to car bans is critical to understanding whether and why restriction policies work.

Based on a review of these behavioural responses, we conclude with four main takeaways. First, researchers and policymakers should not expect to find anything close to a 20% reduction in pollution or congestion from banning a fifth of vehicles from the road. In the most studied cities, like Beijing and Mexico City, many vehicles are exempt from the policy. Behavioural responses, such as cheating or shuffling car trips to unrestricted hours and unrestricted days, will also reduce policy effectiveness. In addition, private cars only represent a share of the total traffic on the street, and traffic only produces a share of total local pollution. These shares, moreover, vary by pollutant, city, and neighbourhoods within cities.

Second, based on the complex and interrelated behavioural responses to license-plate-based driving restrictions, researchers and policymakers should focus evaluation on the correct ecological unit, the household. Most existing studies, which rely on aggregate pollution and congestion data, are poorly suited to understanding how driving restrictions work or fail to work. There is also a notable lack of qualitative research, such as interviews and focus groups, to provide insight into the multiple ways that residents respond to and potentially avoid driving restrictions. While additional study into the second-car hypothesis may be warranted, researchers should focus on other behavioural responses. Whether and to what extent households choose not to purchase a car is particularly understudied and potentially important. The overall effects of vehicle exemptions on pollution and congestion are also poorly understood.

Third, license-plate-based restrictions have a wide variety of implementation and enforcement strategies. As such, findings from one context may translate poorly to another. Mexico City and Santiago's policies, for example, have long exempted newer vehicles with lower emissions. Beijing's policy recently began to exempt electric vehicles. These exemptions may influence air pollution by changing the composition of the vehicle fleet slowly over time. Existing research designs, which generally focus on the immediate moments before and after policy implementation, would not capture these effects. Similarly, Beijing's car lottery, enforcement policies, and recent exemptions for electric vehicles make it a substantially different context from Delhi and the Latin American cities where other studies into license-plate-based restrictions have been conducted. In order to better understand the limitations and potential benefits of driving restrictions, policymakers and researchers should focus on specific behavioural responses to specific policies.

Fourth and finally, policymakers interested in increasing the effectiveness of license-plate-based restrictions should address the full suite of potential behavioural responses described in Section 4. While vehicle exemptions for low-polluting vehicles may help modernise the vehicle fleet and reduce pollution (Barahona et al., 2020), they likely increase total driving and congestion. Encouraging desired behavioural responses while discouraging undesired responses is critical to increasing effectiveness. Improving enforcement likely plays an important role. The existing literature suggests that residents frequently drive on restricted days, particularly outside of central parts of large metropolitan areas. Extending the hours of restrictions can help discourage residents from shuffling car trips right before or after the restrictions start. By contrast, little can be done to discourage shuffling car trips from one day to another. Instead, policymakers could work to encourage shifts to transit and other modes by improving their quality through new investments or subsidies. Other policies to increase the costs of automobility more directly, such as congestion charging and reducing parking supply, would also complement license-plate-based driving restrictions. Policies that facilitate driving, such as building additional road capacity or requiring parking in new commercial developments, will tend to reduce the effectiveness of driving restrictions.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the USDOT Tier 1 University Transportation Center “Cooperative Mobility for Competitive Megaregions (CM2)” under USDOT Award No. 69A3551747135.

Appendix A.

The second-car hypothesis

This Appendix provides an overview of the origins, use, and diffusion of the second-car hypothesis. The Appendix then revisits and assesses the strength of the initial evidence from two foundational papers that developed and popularised the hypothesis. The hypothesis is so popular that it has been cited in the press as definitive evidence that licence-plate-based driving restrictions will not reduce pollution in cities like Paris or anywhere else (Mathiesen, 2014). The hypothesis has also been used outside of the transportation literature, for example, to explain why public health communications about Valley Fever, a fungal respiratory disease, might backfire and increase disease transmission in central California (Matlock, Hopfer, & Ogunseitan, 2019). Thirty-seven out of the 40 empirical papers reviewed in Section 3 refer to the second-car hypothesis.

Origins

Eskeland and Feyzioglu (1997) provide the first peer-reviewed empirical study of Mexico City's license-plate-based car restriction Hoy No Circula. Creating a predictive model of aggregate gasoline consumption prior to the policy and applying the model to data after the policy's enactment, Eskeland and Feyzioglu (1997) find that gasoline consumption increased significantly more after the policy than their model would have predicted. The authors attribute this increase to the second-car hypothesis: Households responded to the policy by purchasing older, higher polluting vehicles and driving more than they would have without the policy. Two earlier working papers use similar language to discuss the likely effects of Hoy No Circula (Eskeland, 1992; Levinson & Shetty, 1992). Neither working paper provides empirical evidence to support the claim. Levinson and Shetty (1992) summarises pollution-restriction programmes in several cities and states that Hoy No Circula had “backfired … . with families having purchased second cars, usually older and dirtier, rather than face time on Mexico City's crowded public transit system” (Levinson & Shetty, 1992, p. 32). Eskeland (1992) presents a similarly worded assertion and argues that a gasoline tax would be more efficient than a license-plate-based restriction. Between 1997 and 2007, two peer-reviewed papers cite Eskeland and Feyzioglu (1997) as evidence that second-car purchases make license-plate-based driving restrictions ineffective and share the same lead author (Eskeland, Jimenez, & Liu, 1998; Eskeland & Xie, 1998).

Davis (2008), the second foundational paper, is the first to formally test and find evidence in support of the second-car hypothesis. The focus of the paper is a series of regression discontinuity designs examining whether local pollutants increased or decreased after Hoy No Circula was implemented in Mexico City. To explain why pollution did not decrease and possibly increased, the study also examines shifts in metro system use and second-hand car purchases. Presenting a regression discontinuity design on annual car sales, Davis (2008) concludes that Hoy No Circula led to a statistically significant uptick in car sales.

Diffusion and use

Starting in 2008, the number of papers citing the second-car hypothesis increases sharply (Figure A1). This increase likely corresponds to Davis (2008) but also to Beijing's implementation of a restriction policy in the lead-up to the 2008 Summer Olympics and the spread of license-plate-based restriction policies to additional Chinese cities. Of the 145 peer-reviewed studies that reference the second-car hypothesis, 19 cite Eskeland and Feyzioglu (1997), 48 cite Davis (2008), and 42 cite both. Another 36 papers cite neither, but cover similar topics, refer to the same cities, and use similar language of backfiring when referencing the second-car hypothesis. Several studies cite a 2014 Guardian article (Mathiesen, 2014) about license-plate-based restrictions as evidence of a policy backfire (Li, Wang, Chen, & Wang, 2020; Tan, Chung, Shi, & Chiu, 2017).

Figure A1. Cumulative total number of peer-reviewed publications referencing the second-car hypothesis and seminal papers from 1995 to 2020.

Referencing the second-car hypothesis

The 145 academic papers that discuss the second-car hypothesis examine a variety of geographies (Figure A2) and topics. The vast majority (141) examine some type of environmental regulation. About a third (51) study restriction policies directly and are mostly based in Latin America or China. Mexico City and Beijing's programmes have received particular academic attention.

Figure A2. Geographical distribution of papers referencing the second-car hypothesis by focus on license-plate-based restrictions. Papers with multiple study cities are counted multiple times.

The remaining two-thirds (94 papers) are from all over the world and cover a wide range of research topics, including travel credits (Wang, Yang, Zhu, & Li, 2012), car lotteries (Yang, Liu, Qin, & Liu, 2014), and subway expansion (Li, Liu, Purevjav, & Yang, 2019). Four papers do not discuss environmental policies at all. For example, Durango (2011) focuses on identifying the areas of Bogota and Medellin most impacted by traffic congestion and potential congestion mitigation strategies.

Across studies, the second-car hypothesis is frequently referenced to describe a failed environmental policy or to discuss the risk of a policy having unintended consequences. Most (94) reference the second-car hypothesis in their introduction, literature, or background sections to motivate the study under consideration. Most treat the second-car hypothesis as an established fact. For example, Parry (2012) references the second-car hypothesis to dismiss license-plate-based driving restrictions as a potential congestion-mitigation strategy and to help elevate the importance of GPS-based congestion charges.

Twenty-one additional papers use terms, such as backfiring, counterproductive, or unintended consequences, when discussing the second-car hypothesis in relation to environmental regulation. For example, Jakob (2017, p. 97) discusses Ecuador's policies to reduce greenhouse-gas emissions and uses the license plate-based driving restriction as “ … a salient example of a policy that is relevant to climate change … ” but one that is “ … unlikely to achieve its goal and might even result in exacerbating the problem it intends to address.”

Although the existing literature tends to frame the second-car hypothesis as a universal phenomenon, most papers (80) reference Mexico City's policy specifically. Even the term backfiring has a direct connection to Eskeland and Feyzioglu’s (1997) paper title. Moreover, studies that do not mention Mexico City directly often reference findings from Mexico City. Of the papers that cite neither Eskeland and Feyzioglu (1997) nor Davis (2008), twelve mention Hoy No Circula or Mexico City when discussing the second-car hypothesis. Nevertheless, the existing literature uses the second-car hypothesis to discuss a variety of environmental policies in cities and countries throughout the globe (Figure A2).

Papers that examine the effects of license-plate-based driving restrictions reference the second-car hypothesis for a wider range of reasons, such as explaining null results (Ye, 2017), describing why findings might differ across places (Gu et al., 2017; Xie et al., 2017), motivating a specific research design (Gu et al., 2017; Guerra & Millard-Ball, 2017), or explaining why short-run and long-run policy effects might differ (Ma & He, 2016; Xu, Grant-Muller, & Gao, 2017). For example, Ma and He (2016) consider second-car purchases a policy adaptation that could make pollution worse in the long run despite initial improvements. Most, however, simply present the second-car hypothesis as an established fact or typical finding in the paper introduction or literature review. Xiao et al. (2019, p. 299), for example, reference Eskeland and Feyzioglu (1997) before attributing an increase in Beijing's private vehicle fleet to “ … the fact that some consumers purchased additional cars to circumvent the driving restriction policy.”

Strength of the initial evidence

For all its prominence in the academic literature and popular press, the original papers examining the second-car hypothesis do not present strong evidence in its support. In revisiting this evidence, we do not intend to criticise the authors or their papers. Eskeland and Feyzioglu (1997) produced the seminal academic paper on license-plate-based driving restrictions, and Davis (2008) wrote the most-cited and likely best-regarded paper on license-plate-based driving restrictions. Instead, we intend to demonstrate the weaknesses of the second-car hypothesis’ foundations. While some households may indeed purchase a second car in response to restriction policies, there is little to suggest that this causes the policies to backfire.

Eskeland and Feyzioglu (1997) present the second-car hypothesis as a way to explain a statistically significant increase in fuel expenditures in the years immediately after Hoy No Circula. The authors test this explanation by developing a household-level model of car purchases. Strikingly, the authors find that although some households would likely purchase a second car to avoid the policy, “a somewhat greater number of households would want to reduce their number of cars” (Eskeland & Feyzioglu, 1997, p. 400). The authors dismiss these empirical findings, however, with an observation that aggregate used car sales increased in Mexico City after Hoy No Circula and an anecdotal assertion that “ … most observers believe that the opposite occurred” (Eskeland & Feyzioglu, 1997, p. 393). The initial finding that fuel expenditures increased due to Hoy No Circula is also relatively weak. The research design relies on the assumption that differences in fuel expenditures are entirely attributable to Hoy No Circula. The policy's implementation, however, coincides with the end of a nearly decade-long recession that had substantially reduced household income and almost certainly reduced car purchases and household driving. The only economic controls in the model of fuel expenditures are the quarterly number of international phone calls, which is used as a proxy for household prosperity. Davis (2008, p. 64) conducts a similar test of gasoline sales using a regression discontinuity design and concludes that Hoy No Circula is associated with “a small and statistically insignificant change in gasoline sales.”

Employing another series of regression discontinuity designs, Davis (2008, p. 68) reports that Hoy No Circula is associated with an increase in the number of registered vehicles that “is statistically significant at the 1 percent level.” The analysis, however, includes only 30 data points that overlap with multiple additional exogenous shocks, such as a major recession and a redesign of the policy to exempt cars with catalytic converters. Moreover, a shifting linear trend in increased vehicle sales (Davis, 2008, fig. 9, 2008, fig. 10) begins in 1986, three years prior to Hoy No Circula.

To test whether the reported results depend on the specification, we download the INEGI data on vehicle sales and re-estimate models of vehicle sales in Mexico City, other nearby states, and the entire country (Table A1). For convenience, we report the original parameter estimates (Davis, 2008, Table 10), which differ slightly from our own estimates, likely due to INEGI making minor revisions to the 2004 and 2005 data. We choose to present the model with the fifth-order polynomial time trend. The fourth- and sixth-order time trends produce consistent results. Figure A3 plots the data and predicted discontinuities for the six geographies in Table A1.

Table A1. Regression discontinuity estimates of the relationship between Hoy No Circula and annual vehicle registrations (natural log) using a fifth-order polynomial time trend.

	Parameter estimates		Standard errors
Geography	Davis (2008)	Re-Estimate	Unadjusted	Newey West^a	Bootstrapped^b
Mexico City	0.189	0.198	0.097.	0.094*	0.147
Mexico State	NA	0.220	0.225	0.233	0.986
Aguascalientes	NA	0.186	0.063**	0.050**	0.186
Puebla	NA	0.160	0.105	0.081.	0.160
Queretaro	NA	0.166	0.088.	0.078.	0.146
Country	NA	0.164	0.035***	0.076*	0.131

Notes: Significance codes “***” 0.001 “**” 0.01 “*” 0.05 “.” 0.1; ^aNewey West standard errors apply one-year lag. ^bBootstrapped standard errors estimated over 50,000 iterations.

These robustness checks indicate that increased vehicle sales in Mexico City are not related to Hoy No Circula. First, Hoy No Circula appears to have had a similarly sized effect on vehicle registrations in Mexico City as it did throughout the country and in multiple states without license-plate-based restrictions. In the case of Aguas Calientes, Queretaro, Puebla, and the country, the parameter estimates are statistically different from zero at the 10 percent level when applying Newey-West standard errors. By contrast, vehicle registrations do not appear to increase in Mexico State, much of which is subject to Hoy No Circula. Second, the confidence level of 1 reported in the paper is likely a typo and should be a confidence level of 10 (Davis, 2008, Table 10). The Newey-West correction for serial correlation, moreover, decreases rather than increases the size of the standard errors. Using bootstrapping to account for the random ways that other exogenous shocks might drive the findings, the standard errors increase substantially. Third, reducing the sample to an equal number of years before and after the discontinuity does not produce statistically significant results at the 10 percent level for Mexico City using any of the polynomial time trends reported by Davis (2008). Thus, although the study provides the first empirical evidence supporting the second-car hypothesis, the evidence is not strong. In an updated analysis that examines a shift in the restrictions to apply on Saturdays, Davis (2017) makes no mention of second-car purchases.

Figure A3. Regression discontinuity plots of the relationship between Hoy No Circula and annual vehicle registrations (natural log) using a fifth-order polynomial time trend.