key: cord-1051658-ruv67f5v
authors: Chang, Hung-Hao; Meyerhoefer, Chad D.; Yang, Feng-An
title: COVID-19 prevention, air pollution and transportation patterns in the absence of a lockdown
date: 2021-08-10
journal: J Environ Manage
DOI: 10.1016/j.jenvman.2021.113522
sha: 6d605dc1314df6614d9daa96a8333e24797e813e
doc_id: 1051658
cord_uid: ruv67f5v

Recent studies demonstrate that air quality improved during the coronavirus pandemic due to the imposition of social lockdowns. We investigate the impact of COVID-19 on air pollution in the two largest cities in Taiwan, which were not subject to economic or mobility restrictions. Using a difference-in-differences approach and real-time data on air quality and transportation, we estimate that anthropogenic air pollution from local sources increased during working days and decreased during non-working days during the COVID-19 pandemic. This led to a 3–7 percent increase in CO, O(3), SO(2), PM(10) and PM(2.5). We demonstrate that the increase in air pollution resulted from a shift in preferred mode of travel away from public transportation and towards personal motor vehicles during working days. In particular, metro and shared bicycle usage decreased between 8 and 18 percent, on average, while automobile and scooter use increased between 11 and 21 percent during working days. Similar COVID-19 prevention behaviors in regions or countries emerging from lockdowns could likewise result in an increase in air pollution. Taking action to reduce the transmissibility of COVID-19 on metro cars, trains and buses, could help policymakers limit the substitution of personal motor vehicles for public transit, and mitigate increases in air pollution when lifting mobility restrictions.

In order to reduce the spread of the SARS-CoV-2 virus that causes COVID-19, many regions and countries have implemented significant restrictions on business operations and the mobility of consumers (Cheng et al., 2020a) . For example, stay-athome orders and social distancing reduce travel and provide barriers to employment and the acquisition of consumer goods. The closure of restaurants, retail establishments, and non-essential businesses likewise limit movement and reduce economic output. There is significant interest in determining how these unprecedented restrictions and the associated behavioral responses affected air quality.

Wuhan and mobility restrictions put in place throughout China thereafter. For example, Xu et al. (2020) , Wang and Su (2020) , Shi and Brasseur (2020) , Cole et al. (2020) , Fan et al. (2020) , and Almond et al. (2020) all find that lockdowns and restrictions on economic activities in the city of Wuhan and other regions of China significantly reduced air pollution. Shi and Brasseur (2020) calculates that surface PM2.5 and NO2 levels decreased by 35% and 60%, respectively, in northern China during the pandemic, while Wang et al. (2021) finds reductions in six ambient air pollutants in the Beijing-Tianjin-Hebei region and Yangtze River Delta. 1 Most studies also find that O3 increased by as much as a factor of 2 due to reductions in nitrogen oxides. Despite the offsetting effects of these pollutants, both He et al. (2020) and estimate that the improvement in China's air quality could lead to as many as 36,000 fewer premature deaths per month.

Investigations of the association between COVID-19 restrictions and air pollution in Europe (Baldasano, 2020; Menut et al., 2020; Bauwens et al., 2020; Collivignarelli et al., 2020) , Egypt (Mostafa et al., 2021) , South Korea (Bauwens et al., 2020) and India (Mahato et al. 2020) , are generally consistent with those from China. For example, Menut et al. (2020) estimates reductions in NO2 pollution across locations in Western Europe of 30% -50%, and reductions in PM10 and PM2.5 of 5% -15%. Mahato et al. (2020) finds a 53% drop in NO2 and reductions of 60% and 39% in PM10 and PM2.5, respectively, during the lockdown in Delhi, India. In addition, Mahato et al. (2020) finds concomitant increases in ground-level ozone in Delhi, as does Menut et al. (2020) in Western Europe.

Estimates from U.S. data are more mixed than in other countries. Brodeur et al. (2020) determine that stay-at-home orders reduced levels of PM2.5 by 25%, while Bauwens et al. (2020) report reductions of NO2 in the northeastern cities of New York, Philadelphia, and Washington D.C. of 21% -28%. In contrast, Bekbulat et al. allowing haze pollution in Beijing to persist. J o u r n a l P r e -p r o o f (2020) does not find that PM2.5 and O3 concentrations in the U.S. fell outside the normal range of statistical variability. Zangari et al. (2020) likewise reports no significant differences in PM2.5 and NO2 during the 2020 New York City shutdown relative to the same weeks in 2015 -2019.

Finally, Dang and Trinh (2020) compiles data from 178 different countries and confirms the finding from most individual country or city studies that lockdowns decreased air pollution. They also conclude that decreases in mobility following the lockdowns likely reduced air pollution. Elsaid et al. (2021) conducts a similar worldwide review, and finds that lockdowns dramatically decreased COX, NOX, SOX and PM emissions, but increased O3 due to reductions in nitrogen emissions.

While studies hypothesize reductions in industrial production, transportation and other human activities are responsible for the lower levels of ambient air pollution that occurred during COVID-19 lockdowns, they typically do not make a direct link to specific activities. Exceptions include Sahraei et al. (2021) which shows that in 12 countries with lockdowns there was a concurrent decrease in public transit usage, and Dang and Trinh (2020) , which used Google Mobility Reports to show lower mobility in locations where government policies were more stringent. Likewise, Zeng and Bao (2021) finds that much of the decrease in PM10, PM2.5 and NO2 that occurred during lockdowns was due to lower levels of human migration. Hua et al. (2021) shows that J o u r n a l P r e -p r o o f NO2 emissions in Beijing gradually increased as traffic volumes escalated following the start of the lockdown and Cicala et al. (2020) predicted a decline in CO2 in a simulation study based on decreased vehicle traffic in the U.S.

We contribute to this literature by examining changes in air pollution in Taipei and New Taipei City, the two largest cities in Taiwan, which were not subject to the strict lockdowns imposed in many other major cities. As a result, we are able to how air quality changed during the early stages of the pandemic due to residents' attempts to avoid a COVID-19 infection. Using a difference-in-differences framework and realtime data on both air pollution and transportation, we draw a direct link between higher pollution levels and changes in transportation patterns that we ascribe to COVID-19 prevention behavior.

We compiled data from several administrative sources in order to conduct our analysis. We collected data on confirmed cases of COVID-19 from the Taiwan Center of the Disease Control (CDC); several air quality measures from the Environmental Protection Agency (EPA); metro usage from the Taipei Rapid Transit Corporation; shared bicycle usage from the Taipei City Government, and motor vehicle traffic data J o u r n a l P r e -p r o o f from the Ministry of Transportation and Communication. 2 Our study period is January 1 -March 31, 2017 -2020.

The first confirmed case of COVID-19 in Taiwan was identified in Taipei City on January 22, 2020. We compiled the number of the confirmed cases of COVID-19 registered in both Taipei and New Taipei City during each day between January 22, 2020 and March 31, 2020. Confirmed cases are those that have been validated by medical testing, whereas unconfirmed cases are suspected by physicians to be COVID-19 based on patient-reported symptoms. We use the former because confirmed cases are listed on the CDC website and reported to the public during CDC press conferences. From these data we created two variables to measure COVID-19 cases. The first is a binary indicator for the period when confirmed COVID-19 cases existed in Taiwan (=1 if January 22 -March 31 of 2020; =0 otherwise), while the second is a continuous measure of the cumulative number of the confirmed cases in each day of the sample period in each of the two cities.

For the analysis of all outcome variables we defined the treatment and control groups in 2020 and 2017 -2019, respectively. In each year, the post-treatment period corresponds to January 22 -March 31 and the pre-treatment period is January 1 -January 21.

We obtained measures of air quality from 19 air quality monitoring stations located across Taipei and New Taipei City (see Figure 1 China as control variables. We merged the COVID-19 variables into the air quality dataset using city and date. In total, the air quality data from Taipei and New Taipei City consist of 5,741 station-day observations across both the treatment and control periods, of which 1,190 occurred in the 2020 post-treatment period (i.e., the period corresponding to the COVID-19 outbreak).

The Taipei metro system includes 119 stations that serve all sections of Taipei and New Taipei City. We excluded 11 stations from our analysis because they opened on January 31, 2020, right after the beginning to our treatment period. For each of the remaining 108 stations, we collected data on the number of people who departed from and exited the station on every day during our treatment and control periods. In total, our sample of metro ridership contains 38,772 station-day observations, of which 7,452 occurred in the 2020 post-treatment period.

From the same metro database we created several indicator variables to measure characteristics of the stations and their surroundings. These include variables that indicate whether the station is a terminal station, connected to other metro lines, on an airport route or at an airport, connected to a high speed rail station, and whether a school or a traditional night market is near the station.

In 2011 the Taipei city government created a bike sharing system called U-bike to J o u r n a l P r e -p r o o f provide transportation to and from public transportation. The program has grown to 966 stations that serve most of Taipei and New Taipei City. U-bike stations are located at metro stations and bus stops as well as strategic locations near places of business and high density residential areas. The initial cost of bike rental is only five New Taiwan dollars, making the use of U-bikes very popular among metro and bus users. 3

In May, 2020 the turnover rate of U-bikes was 5.6 times per bike per day (Cheng et al., 2020b) . We excluded 10 stations that were opened in 2020. For the remaining 956 stations, we collected data on the number of departures and exits from each U-bike station (i.e. U-bike rentals) in each day of our treatment and control periods as well as information on the characteristics of U-bike stations. The latter includes the year the station opened and the number of bike docks at each station. There are 222,007 rentalday observations in the U-bike data, 48,540 of which occurred in the 2020 posttreatment period.

There are 35 traffic monitoring stations along the main bridges that connect Taipei and New Taipei City. Since many people live in New Taipei City and work in Taipei, the monitors effectively capture commuting patterns for all types of motor vehicles.

We collected information on the number of cars, vans/trucks/buses and scooters passing through each traffic monitor during every hour of the treatment and control periods. 4 In total, our motor vehicle samples contains 309,400 station-hour observations, of which 60,460 are in the 2020 post-treatment period.

We collected additional data on three district-level characteristics that may explain variation in the outcome variables: the geographic area of each district, monthly population, the wind speed in each district, and the daily rainfall in each district. 5 For inclusion in our traffic analysis we also collected information on the daily gasoline price, and for the air pollution models we collected data on the amount of coal used in power generation in metric tons per day. 6 Finally, we include in all of our models a control for the number of daily inbound visitors in Taiwan. 7

We use a difference-in-differences model (DiD) to identify the causal effect of confirmed COVID-19 cases on air quality, motor vehicle usage, metro usage, and Ubike rentals using panel data (Wooldridge, 2010) . The pre-treatment period is January 1 -January 21 and the post-treatment period is January 22 -March 31. The control group is composed of observations in 2017-2019 and the treatment group is observations in 2020. 8 Because the air quality, public transportation and shared bicycle outcome variables are right-skewed, we implemented the following log-linear specification:

(1)

( 1 ) = 1 + 1 ⋅ + 1

where 1 is the outcome variable for station in city during time . Station is either air monitoring station, metro station or U-bike station, and time is day. is either our discrete or continuous measure of confirmed COVID-19 cases in city at time . 1 is a vector of the explanatory variables associated with the outcome variable, , , and are fixed effects for city, month and year, and 1 is the random error term. 9 , 10

In the case of motor vehicle traffic, all of the traffic monitoring stations are located on main bridges that connect Taipei and New Taipei City, which means that there is no city-level variation in the dependent variable. In addition, the outcome variable has some zero values, so we estimated the following DiD model:

(2) 2 = 2 + 2 ⋅ + 2 ′ 2 + + + 2 .

In equation (2) the time index denotes the one hour period used to collect traffic data, and is either our discrete or continuous measure of confirmed COVID-19 in both cities combined at time .

In equation (1), 100 times the parameter 1 is a semi-elasticity that measures the effect of COVID-19 on the outcome variable in percentage terms. In equation (2), 2 measures the effect of COVID-19 on 100s of cars, vans/trucks/buses or scooters per hour travelling between Taipei and New Taipei City. Both 1 and 2 are identified by comparing differences in outcome variable before and after the advent of COVID-19 (or across the pre-and post-COVID-19 level of infections) within the same calendar month in the control period (2017-2019) and the treatment period (2020) . 11 We computed the standard errors in all models using the two-way-cluster-robust variance approach proposed by Cameron and Miller (2015) to cluster on both station and day.

In panel A of Table 1 

We present semi-elasticity estimates in Table 2 from the full DiD model of air quality, as specified in equation (1), and full estimation results for the CO equation in Appendix We performed several specification and robustness tests of the DiD air quality J o u r n a l P r e -p r o o f models, which we describe more fully in Appendix A, Section 1. First, we tested whether the assumption of parallel pre-treatment trends in the air quality outcomes is met using the approach developed by Mora and Reggio (2012; 2015) . We found that models for of CO, O3, SO2, NO2, and PM2.5, but not PM10, are consistent with the null hypotheses of common parallel linear trends. We then conducted a falsification test by using the 2020 COVID-19 discrete indicator and case count variables as placebo treatments in 2019, 2018 and 2017, while retaining the remaining two years as the control group. All models passed the falsification test except some formulations of PM10 model. Finally, we increased the flexibility of our DiD specification to allow for differential linear pre-treatment trends across the two cities and estimated models with station fixed effects. Semi-elasticity estimates for all outcomes are robust across these different specifications. Overall, the sensitivity analyses suggest a small bias in the semi-elasticity estimates from the PM10 models.

Because motor vehicle traffic is a major source of pollution in Taipei and New

Taipei City, we investigated whether the change in pollution levels due to COVID-19 could have resulted from a shift in transportation patterns. indicate that increases and decreases in motor vehicle traffic both increase in magnitude with the number of confirmed cases in each city when the sample is split between working and non-working days.

While transportation by personal motor vehicle increases air pollution, the use of the metro system and U-bikes reduces pollution by decreasing the combustion of fossil fuels associated with vehicle travel. We present the results for the use of metro and U-bikes together because the latter are often used jointly with the metro in order J o u r n a l P r e -p r o o f to travel the first and last miles of a journey. Table 4 contains semi-elasticity estimates of COVID-19 on metro departures and exits and on U-bike rentals. 14 Using the full sample, we find that metro departures and exits decreased 17.7% during the COVID-19 period and 0.2% per additional confirmed case of COVID-19 (Table 4 , panel A).

When we split the sample into working and non-working days, we find that metro use decreased more during the latter than the former. In particular, metro departures and exits went down 25.2% due to COVID-19 during non-working days, but only decreased by 11.2% during working days.

U-bike rentals also decreased due to behavioral change among residents during the COVID-19 period, but to a lesser extent than the reduction in metro use. The overall decrease in U-bike rentals was 8.1% during the COVID-19 period, or 0.1% per confirmed case of COVID-19 (Table 4 , panel B). Similar to metro use, U-bike rentals decreased more during non-working days than working days. In particular, rentals decreased 15.6% due to COVID-19 during non-working days, but 3.1% during working days.

As a robustness check, we re-estimated all of our transportation models with station-level fixed effects using the full sample in each dataset. The estimates, 14 Appendix Tables A7 and A8 contain descriptive statistics for the control variables in the metro and U-bike datasets, respectively, and full estimation results for the metro and U-bike models are reported in Tables A9 and A10, respectively. reported in appendix Table A11 , are similar to the main DiD estimates.

The increase in motor vehicle use and decrease in metro and U-bike use that we identify is consistent with the increase in air pollution levels during working days in the COVID-19 period. In addition, the reduction in motor vehicle use during nonworking days is consistent with the decrease in NO2 and PM10 we find during nonworking days. However, it is possible that the changes in industrial pollution could have also affected air pollution. Because Taiwan did not impose a social lockdown or implement restrictions on business activities, we suspect that differences in industrial pollution across the pre-and post-COVID-19 period were minor. Nonetheless, industrial output could have been affected by shifts in demand during the pandemic. Taiwan We investigated the sensitivity of our air pollution results to changes in manufacturing by re-estimating our DiD models while controlling for manufacturing sales in each city. Due to limitations on data availability, we could only access J o u r n a l P r e -p r o o f manufacturing sales data in 2020 and 2019. Therefore, we use observations in 2019 as the control group in these models. 

Cars, vans, trucks, buses and scooters are a major source of air pollution in many cities around the world. Collectively, they emit all of the primary pollutants we track in our study, and are major contributors to the formation of secondary pollutants (Union of Concerned Scientists, 2018; Zhang and Batterman, 2013; Platt et al., 2014) .

In Taipei and New Taipei City, scooters were the most popular method commuting to work prior to the pandemic, accounting for approximately 30% of trips. Scooter engines emit significant amounts of carbon monoxide, nitrogen oxides and particulate matter (more per mile than cars), and they are regulated less stringently than other types of motor vehicles (Platt et al., 2014; Carpenter, 2014) . Because the use of .

As vaccines are deployed and other regions or countries emerge from lockdowns, they should expect individuals to exhibit similar preferences for personal vehicle use that we observe in Taiwan. This could lead to congestion and higher levels of pollution. Initiatives to improve the safety of public transportation and associated public information campaigns could help to limit the avoidance of such cleaner forms of transportation. In addition, research suggests that individuals de-prioritize environmental protection following periods of high unemployment (Kenny, 2019) and

there is evidence that rollbacks of environmental regulations intended to lessen the economic effects of COVID-19 could lead to longer term worsening of deadly air pollution and result in additional deaths (Persico and Johnson, 2021; Gardiner, 2020) .

Our findings suggest that scaling back air pollution regulations due to a perceived tradeoff between environmental protection and economic growth could compound the deterioration in air quality.

As the pace of worldwide vaccination effects increases, and COVID-19 infection rates begin to decline, governments will face mounting pressure to lift restrictions on economics activities intended to reduce the spread of COVID-19. For example, China has begun to lift COVID-19 restrictions in some provinces, while most countries in Western Europe are also lifting restrictions (IMF, 2021) . However, if individuals, J o u r n a l P r e -p r o o f particularly those who are unvaccinated, do not feel safe traveling among the general population, they may engage in prevention behaviors similar to those we identify in Taiwan (Brackett, 2020) . Fear of public transportation could also result in preferences for work-at-home arrangements, which could increase demand for residential energy (Hinson, 2020) . The resulting increase in air pollution from personal motor vehicle use and greater time spent at home could compound a rise in air pollution from industrial sources as business seek to meet the pent-up demand for consumer goods resulting from months of limited access to retail outlets.

In order to reduce avoidance of public transportation, policymakers can consider several actions. Scientific reviews indicate that physical distancing of 1m or more, facemask use, and eye protection all decrease the transmissibility of COVID-19 (Chu et al., 2020; Brooks and Butler, 2021) . Mask mandates on public transportation are among the easiest policies to implement and enforce because they involve a low cost to the public, and compliance is observable. However, experts agree that physical distancing is the most effective public health measure to reduce the spread of COVID-19 (IDSA, 2020). Implementing physical distancing on metro/train cars and buses reduces the capacity of the public transportation system, but it could increase overall ridership if potential users feel safer. In addition, policy makers need to evaluate other initiatives to implement physical distancing without reducing system use, such as transmission during the use of public transportation, policy makers can limit the avoidance that we identify in Taiwan during the early stages of the coronavirus pandemic. Although some measures may result in additional costs, our findings suggest the reductions in air pollution from encouraging the use of public transport could be substantial. Not only does reducing air pollution lower future disease risks of otherwise healthy individuals, it also reduces COVID-19 mortality (Persico and Johnson, 2021; Isphording and Pestel, 2020) .

This study makes an important contribution to the literature as the first to link higher levels of several major air pollutants to a shift in transportation patterns from J o u r n a l P r e -p r o o f Note: The sample period is January 1 -March 31 in each year. #1 -drawn from the air quality dataset. #2 -drawn from the moto vehicle traffic dataset (see Table A2 ). #3 -drawn from the metro use dataset (see Table A4 ). #4 -drawn from the U-bike rental dataset (see Table A5 ).

J o u r n a l P r e -p r o o f J o u r n a l P r e -p r o o f 

Ambiguous pollution response to COVID-19 in China

COVID-19 lockdown effects on air quality by NO2 in the cities of Barcelona and Madrid (Spain)

Impact of coronavirus outbreak on NO2 pollution assessed using TROPOMI and OMI observations

2020. PM2.5 and ozone air pollution levels have not dropped consistently across the US following societal covid response

More people turning to cars because of fears of coronavirus infection on public transit

On the effects of COVID-19 safer-at-home policies on social distancing, car crashes and pollution

Effectiveness of mask wearing to control community spread of SARS-CoV-2

A practitioner's guideline to cluster robust Inference

Motorcycles and emissions: the surprising facts

Air pollution reduction and mortality benefit during the COVID-19 outbreak in China

Beyond adoption of shared bike: A case study on antecedents and consequences of brand attachment in Taiwan

Physical distrancing, face masks, and eye protection to prevent person-to-person transmission of SARS-CoV-2 and COVID-19: a systematic review and metaanalysis

Expected health effects of reduced air pollution from COVID-19 social distancing

The impacts of the Wuhan Covid-19 lockdown on air pollution and health: a machine learning and augmented synthetic control approach

Lockdown for CoViD-2019 in Milan: What are the effects on air quality?

Does the COVID-19 pandemic improve global air quality? New cross-national evidence on its unintended consequences

Effects of COVID-19 on the environment: An overview on air, water, wastewater, and solid waste

The impact of the control measures during the COVID-19 outbreak on air pollution in China

Pollution made COVID-19 worse. Now, lockdowns are clearing the air

COVID-19, city lockdowns, and air pollution: evidence from China

COVID-19 is changing residential electricity demand

Infectious Disease Society of America (IDSA), 2020. Physical distancing remains critical to contain COVID-19

Policy responses to COVI-19

Quantitative estimation of meteorological impact and the COVID-10 lockdown reductions on NO2 and PM2.5 over Beijing area using Generalized Additive Models (GAM)

Pandemic meets pollution: poor air quality increase deaths by COVID-19

Economic conditions and support for the prioritisation of environmental protection during the Great Recession

The effects of increased pollution on COVID-19 cases and deaths

Two-stroke scooters are a dominant source of air pollution in many cities

Public transit usage and air quality index during the COVID-19 lockdown

The response in air quality to the reduction of Chinese economic activities during the COVID-19 outbreak

Ventilation and coronavirus (COVID-19)

Comparison of air pollutants and their health effects in two developed regions in China during the COVID-19 pandemic

Response to COVID-19 in Taiwan -big data analytics, new technology, and proactive testing

A preliminary assessment of the impact of COVID-19 on environment -a case study of China

Econometric analysis of cross section and panel data

Impact of the COVID-19 event on air quality in central China

Air quality changes in New York City during the COVID-19 pandemic

The percentage magnitude of the COVID-19 effect is evaluated using the sample mean of the dependent variable in the full sample. The full list of explanatory variables in each regression is reported in Appendix Table A2. Standard errors are clustered by day and traffic monitoring station

Non-working days (NꞏT=13,500)

Non-working days (NꞏT=77,250)

* indicate significance at the 1%, 5% and 10% level, respectively

Supplementary data to this article can be found online at:Declaration of competing interest: The authors declare that they have no competing interests that could have appeared to influence the work reported in this paper.