The Impact of The Pico Y Placa Policy On Fine Particulate Matter In Lima, Peru

Lima has been ranked among the top most polluted cities in the Americas. Vehicular emissions are the dominant source of pollution in the city. In order to reduce congestion and pollution levels during the XVIII Pan- and Parapan-American Games, Lima government ocials enacted the pico y placa policy to restrict the number of vehicles on certain heavily tracked roads in the city at rush hours between Monday to Thursday based on the last digit of their license plates. This policy was retained after the Games. In this paper we evaluate the impact of this policy on ne particulate matter concentration levels (PM 2.5 ) at a background site in the city using a difference-in-difference approach. We nd that the policy resulted in increases on PM 2.5 levels on Monday-Thursday compared to Friday-Sunday levels after the policy was enacted, compared to previous years. However, such an increase was not signicant. These results suggest the need for additional policies to reduce pollution due to trac in Lima. It also suggests the need to track the response to this policy over time to evaluate its ecacy.


Introduction
Lima is one of the most highly polluted cities in the Americas (WHO (World Health Organization), 2016). Silva et al., (2017) reported that between 2010 and 2015, the average annual concentrations of particulate matter with diameter < 10 µm (PM 10 ) and < 2.5 µm (PM 2.5 ) were 84 and 26 µg/m 3 , respectively. They found that PM 2.5 concentrations at six of the ten regulatory monitors in Lima exceeded the World Health Organization (WHO) daily guideline on 77% of the days between 2014 and 2015. PM 2.5 concentrations in Lima show seasonal trends, with the highest levels observed in the summer (Romero et al., 2020b).
Epidemiologic evidence has found that air pollution was responsible for 2,300 premature deaths due to cardiorespiratory disease in Lima adults every year between 2001-2011 (Gonzales and Steenland, 2014). Tapia et al., (2020) found that between 2000 and 2016, an interquartile range increase in PM 2.5 was associated with a 4% increase in respiratory emergency room (ER) visits. There is thus an urgent need to reduce pollution levels in Lima. Carbon (BC) is evidence of this (Underhill et al., 2015). The government has attempted to mitigate vehicular pollution by phasing out lead from gasoline, reducing the sulfur content in diesel and reducing the permissible age of vehicles. However, despite these efforts, the number of vehicles in the city is rapidly increasing. Between 2000 and 2014, emissions from registered vehicles in the Lima Metropolitan Area (LMA) increased by almost 65%. In 2019, Lima had a eet of more than 2.7 million motor vehicles, that has been projected to grow at 7% per year (C. Posada, 2018). The average age of Lima's vehicular eet exceeds 15 years for private vehicles and 22 years for public transport vehicles (BBVA, 2010), and thus the emissions from an average vehicle is much higher than a typical vehicle in the developed world.
It is exceedingly important to develop effective transportation policies to reduce vehicular emissions.
To this end, research has been undertaken to understand the spatially disaggregated impact of vehicular emissions (Romero et al., 2020a, c). Such analyses can inspire the development and implementation of targeted policies to reduce emissions. Other crucial research has examined the impact of tra c regulations on certain main avenues in Lima on air pollution (Morales-Ancajima et al., 2019). More work is needed to evaluate existing transportation policies on air pollution, so that they can be ne-tuned.
In 2019, Lima hosted the XVIII Pan-and Parapan-American Games (Games) (July 26-September 1). A major concern before and during this event were the tra c and high levels of pollution that could potentially affect the athletes. Therefore, the Peruvian government, in collaboration with regional authorities, adopted a variety of tra c control actions just before and during the Games. Most of these policies were discontinued after the end of the Games except for the pico y placa policy, which continued until mid-2020, when it was temporarily halted due to the COVID-19 emergency in the city. In this article, we evaluate the impact of the pico y placa policy on PM 2.5 concentrations in Lima using a difference-indifference methodology.
Data And Methods

Pico y Placa
Pico y placa policy, established by Ordinance N2164 by the Metropolitan Municipality of Lima, restricted the number of vehicles on the road in Lima from Monday to Thursday from 6:30 am to 10:00 am and from 5:00 pm to 9:00 pm based on their license plate number. Speci cally, vehicles with a license plate ending in an even digit or zero, could not circulate on certain roads (displayed in Fig. 1) between Monday and Wednesday during those hours, while those with license plates ending with an odd digit could not be on the roads between Tuesday and Thursdays during the same hours (https://www.gob.pe/institucion/munilima/normas-legales/285775-2164, Metropolitan Municipality of Lima (In Spanish)). No restrictions were applied on holidays and non-working days.
The pilot period of this policy was between July 22 -August 5, 2019, with a short suspension between July 29 -July 30, 2019 because of national holidays. The second period of this policy, with additional roads, started on August 5, 2019. Because of its success of reducing tra c in main arteries, both phases were extended inde nitely, and was temporarily suspended only recently, in March 16, 2020, due to the COVID-19 emergency in Lima (https://rpp.pe/lima/actualidad/coronavirus-en-peru-covid-19, Metropolitan Municipality of Lima temporarily suspends the Pico y Placa policy (In Spanish)).

Other Tra c intervention Policies enacted during the Games
The Lima 2019 Route Plan, which restricted travel on certain roads in Lima to enable athletes to reach their venues in time, was put in place between July 19 -September 5, 2019. It was executed by the Peruvian National Police (PNP) in coordination with Lima 2019 Organizing Committee (COPAL) (https://www.lima2019.pe/en/road-plan, Lima 2019 Route Plan). This Plan comprised of three different measures (i) permanent lanes with 24 hours of restrictions were designated (ii) temporal lanes which branch off from the permanent lanes and had temporary restrictions four hours before and two hours after each competition and (iii) preferential lanes that enabled o cial Lima 2019 vehicles, public, and private transportation to circulate at the same time. Table 1 details the characteristics of all these interventions: the dates of enforcement and the times during the day that the policy came into action. Other than the pico y placa program and the Lima Route plan, the other interventions were short-term or temporary controls. Figure    August 9 -August 10, 2019 [1] https://elcomercio.pe/lima/transporte/pico-placa-sera-aplicado-dias-29-30-julio-noticia-nndc-659586noticia/?ref=ecr [2] https://www.gob.pe/institucion/mtc/noticias/45580-mtc-restringira-circulacion-de-vehiculos-pesadosen-la-carretera-central-durante-el-feriado-largo-por-estas-patrias [3] https://elcomercio.pe/lima/transporte/pico-placa-sera-aplicado-dias-29-30-julio-noticia-nndc-659586noticia/?ref=ecr [4] https://gestion.pe/peru/panamericanos-2019-plan-desvio-inmediaciones-estadio-nacional-nndc- Lima has a limited number of reference air quality monitors. Only one station monitored PM 2.5 data both before and after the pico y placa policy came into force, at the US Embassy (Latitude: -12.099398, Longitude: -76.96888).
The location of this monitoring site vis a vis the streets impacted by the tra c policies can be seen in Fig.  1. The US monitor is located away from localized sources. It thus is well placed to capture overall changes in PM 2.5 levels in Lima from the pico y placa intervention. As can be seen there are a few missing measurements in early 2016, and in late 2017 ( Figure S1 in Supplementary Information).
We plot the average hourly, daily, monthly PM 2.5 levels for the months July to December (while the pico y placa policy was in place) for 2019, and for years prior to 2019, to visualize the impact of the policy on temporal patterns of PM 2.5 levels.

Methods
In order to evaluate the impact of the many tra c policies on PM 2.5 levels in the city, in an ideal experiment, we would compare PM 2.5 concentrations in the absence of these policies, with concentrations after the policies were enacted. Unfortunately, such a counterfactual scenario does not exist in reality. Thus, in this study a Difference in Difference approach (DiD) is applied in order to estimate the effects of such interventions during our study period, by exploiting the fact that the pico y placa policy was in force on Monday-Thursday, but not Friday-Sunday.
Brie y, we estimate the difference in the difference in PM 2.5 levels after the restriction for Monday-Thursday (treatment group) compared to PM 2.5 concentrations on Friday-Sunday (control group) after the pico y placa intervention (July-December, 2019) compared with levels in previous years.
We restrict our analysis to July 31-December 31 (the months in 2019 for which the pico y placa was in force) for all years for which we have data to account for the seasonal changes in PM 2.5 concentrations.
Our strategy is encapsulated in regression Eq. 1.
Y t = β 0 + β 1 (P 1 ) x Y 2019 + β 2 (P 1 ) + β 3 (P 2 ) + β 4 .x t + u t ………………... (1) Y t denotes average hourly PM 2.5 from the US Embassy monitor: on the pico y placa policy intervention for July 31-December 31 across multiple years. P 1 is a dummy variable that is 1 for data between Monday and Thursday and 0 otherwise, indicating the treatment group for which the pico y placa policy was in place.
Y 2019 is a dummy variable indicating the intervention has occurred. It is 1 for 2019 and 0 otherwise, indicating the intervention took place. P 2 is a dummy variable corresponding to the other policy interventions, to separate out the impacts of the different interventions from that of the pico y placa policy. From Table S1, several policy interventions were enacted before and during the Games. The Lima 2019 Route Plan is the longest running intervention, with others in force for only a short time concurrently. We assign P 2 = 1, when the Lima 2019 Route Plan is enforced: July 19 -September 5, 2019, and 0 otherwise.
x t is a vector of covariates that includes a cubic polynomial to account for meteorological variables: temperature, humidity, wind speeds, as well as interactions of these variables with day-of-year and day-ofweek. It also includes xed effects corresponding to year, month of year, hour of day, and interactions between hour of day and P 1 to account for temporal trends in PM 2.5 concentrations.
The error is u t . We clustered errors by month.
In We control for other short-term policy interventions in 2019 using P 2 . In order to further avoid the confounding impacts of other policies: P 2 , we repeat the regression only considering September 6 -December 31, 2019, and October -December 31 (well after the Pan-and Parapan-American games).
The identi cation assumption of the DiD model is that unobserved factors that could affect PM 2.5 values are not correlated with the treatment, conditional on the covariations (x t ) and that these unobserved factors do not affect PM 2.5 concentrations in a manner that is nonlinear in time and not captured by the current independent variables in the model. Therefore, we also present supplementary analyses, where we limit the years in the treatment group to 2018 to ensure that other time-varying factors do not affect our results.

Data availability
The data that support the ndings of this study are available from the corresponding author upon reasonable request. Figure 2 displays the average hourly, daily, monthly PM 2.5 levels for the months July to December (while the pico y placa policy was in place) for 2019, and for years prior to 2019, to explore the impact of the policy on temporal patterns of PM 2.5 levels.

Results
It can be seen that PM 2.5 concentrations peaked in September, 2019 compared to previous years, and then decreased for the months of October -December 2019 (lowest monthly levels compared with others months before and during 2019) as shown in Fig. 2. No signi cant change was identi ed in July 2019 compared with previous years. Average hourly variation shows increases in the diurnal peaks likely corresponding to the morning rush-hour tra c in Lima at around 9:00 am and 11 pm in 2019 of PM 2.5 , in comparison to previous years. Such increases in peak levels of PM 2.5 appear to be particularly pronounced on Wednesdays and Thursdays (during the pico y placa policy) in 2019, although we observe such increases on the weekend as well when the pico y placa policy was not in force.
The peaks of PM 2.5 measured, however do not appear to have been temporally displaced after the intervention occurred, indicating that the intervention has likely not resulted in changes in tra c patterns. There is a notable decrease in the residual for months of November and December, 2019.
Such an exercise in plotting the residuals provides some information on if there are other factors affecting PM 2.5 conditional on factoring in the various daily, weekly, monthly and seasonal trends of PM 2.5 , which as mentioned previously is a key assumption in the DiD approach. The similarity of the residual before the impacts of the interventions indicates that this assumption holds in our analysis. Table 2 presents the DiD estimation results. We see that the pico y placa policy is associated with a small increase in PM 2.5 concentration at the US embassy, of ~ 3 µg/m 3 (Analysis 1 in Table 2) which is not signi cant. When we controlled for other policy interventions, this did not change.   Frésard (1998) found the policy had a disproportionate impact on low-income families that could not afford clean cars. In addition, a large number of families bought a second car, which in many cases, was more polluting than the rst car. This research found that the effect of this restriction on tra c capacities was limited (5% reduction in tra c) Mexico city introduced, 'Hoy No Circula' in 1989 Davis (2008) found that the Hoy No Circula policy had no impact on air pollution in Mexico city. This study found evidence that this policy led to the total increase in vehicular sales, instead of nudging residents to using non motorized transport Bogota, Colombia in 1998 Cantillo and Ortúzar (2014) report that such a policy resulted in a decrease in trips by cars in the short term. However, in the long term, trips by cars returned to original levels.
Similarly PM 10 levels decreased in the short-term, but also rebounded as vehicle trips went up. Quito, Ecuador in May 2020 A signi cant 9-11% reduction in CO concentration during peak hours was observed after the implementation of the pico y placa program in the period Jan 2008 -Dec 2012 (Carrillo et al., 2016). The research reports that it is likely the policy worked here because many residents were too poor to be able to purchase additional cars.
Beijing and Tianjin during the 2008 Olympic Games Cai and Xie (2011) report that the policy was able to signi cantly alleviate pollution during the Olympic Games and was an effective temporary solution to reduce pollution.
New Delhi, India during 2016 (Chowdhury et al., 2017) found no clear reduction in PM 2.5 during the period of the odd-even policy enactment. They said this was likely because of unfavorable meteorological conditions, and the dominance of other sources of pollution during this period.
This result did not change when 1. We restricted the days considered to September 6 -December 31 when the other interventions had ended (Analysis 2). Here we can see that the difference between the concentrations increases to ~ 6 µg/m 3 but is still not signi cant 2. From Fig. 3, it appears that there could be other unusual events occurring in September, possibly related to the Pan-and Parapan-American games that our model had not accounted for. We therefore, also rerun our analysis considering days in October -December (Analysis 3 concentrations. More work is required to continue to track the impact of the policy on vehicular volume in Lima. 2. From Fig. 2, it appears that PM 2.5 concentrations are higher on Monday and Wednesday, but we also see that the peaks on the weekends also increased post July, 2019 compared to concentrations in previous years. This could mean that pico y placa has changed overall tra c ow patterns over every day of the week. Alternatively, 2019 could have also been an outlier year for Lima, with more visitors in the city due to the Games. This possibility warrants the need for continued monitoring to keep tracking the impact of the pico y placa policy. 3. From Fig. 3, it appears that an unusual event is occurring between July-September, 2019. It is unclear what causes this peak. When we restricted our experiment to only considering the months of October-December, the increase in PM 2.5 concentrations on Monday-Thursday compared to Friday-Sunday dwindled. This could mean that residents of Lima had adjusted to the policy, tra c patterns had reduced on Monday-Thursday, resulting in a much smaller difference between levels of PM 2.5 .
This suggests that the response to the policy is still adapting and needs to be tracked over time. Alternatively, another event occurred in Lima that caused increased levels of PM 2.5 concentrations on Monday-Thursday compared to Friday-Sunday. For example, as mentioned earlier, there could have been an in ux of visitors in the city. Our model does not capture such an event. Again, this result suggests the need to continue tracking the effects of this policy. 4. The impacts of the policy are localized/affect certain parts of the city. In this experiment we have only used measurements from one monitor. There is a need to expand monitoring in Lima to evaluate the overall impact of the pico y placa policy in different parts of the city.
The pico y placa has been enacted in other locations. Past research in other cities has shown that such a policy only has limited short-term effects on tra c and like our research, has shown to have limited impact on air pollution.
Despite the fact that most studies (including ours) have shown that the pico y placa policy does not typically result in lowering air pollution levels in the long term, many governments have adopted it as a strategy to reduce pollution. It is important to track the impact of such policies to make a case for additional regulation to decrease pollution.