Energy Research & Social Science 70 (2020) 101773

Contents lists available at ScienceDirect

Energy Research & Social Science

journal homepage: www.elsevier.com/locate/erss

Original research article

A pollution paradox? The political economy of environmental inspection T

and air pollution in China
Lan Zheng, Ming Na
⁎

School of Economics, Hefei University of Technology, Hefei 230601, China

ARTICLE INFO ABSTRACT

Keywords: To scrutinize the implementation of environment policy by local governments, the central government dis­
Environmental inspection patched inspection teams to all the provinces across China. Do the central inspections effectively encourage local
Air pollution politicians to implement environment policy and improve air quality? In this paper, we assess the impact of the
Government regulation central inspection on urban air pollution, using daily pollution and weather data for 130 Chinese cities over
China
2016–2017 and employing both difference-in-difference and regression discontinuity design methods. Our main
finding is that, compared to uninspected provinces, the inspected provinces experience statistically significant
improvement in air quality. Particularly, the significant negative effect of inspection on air pollution is mainly
reflected in PM2.5, the pollutant about which the public is most concerned and main assessment indicator for
controlling air pollution by the central government. However, the improvement in air quality turns out to be
short-term effect, followed by higher air pollution after the inspection. We also conduct heterogeneous analysis
and find that the later rounds of inspection, key control areas, and regions with higher level of economic de­
velopment are more sensitive to the inspection with sizable reduction effects on air pollution, compared to the
first round of inspection, non-key control areas and less economic development regions. In addition, we perform
a series of robustness checks to confirm that our findings are stable and reliable.

1. Introduction negative impact on people’s outdoor travel frequency [9], social in­
teractions [10], productivity [11], and even affects stock yields [12,13].
China’s extraordinary economic development in the past three As the largest developing country in the world, the deterioration of
decades has been accompanied by heavy environmental pollution, air pollution in China has not only attracted massive attention for its
especially air pollution1 [1–3]. Air pollution, represented by haze pol­ influence on global environmental change, but also contributed to the
lution, has severe adverse influences on people’s health and lifespan increasing public awareness of the detrimental effects of air pollution
[4]. According to a World Bank report [5], air pollution is the fourth [14]. Against this background, the Chinese government has made
leading fatal health risk, which resulted in 4.8 million premature deaths controlling air pollution an imperative issue on the national agenda,
worldwide in 2013. The problem is particularly pressing for developing followed by a series of policies and programs which are initiated to
countries, more than 90% of air pollution –related deaths occurring in reduce the emission of environmental pollutants [3]. For example, the
low and middle-income countries [6]. Hsu et al. [7] report that more Air Pollution Control Action Plan was published in 2013; the En­
than 50% of the Chinese population is exposed to unsafe air, and one of vironmental Protection Law (revised), considered as the strictest en­
five deaths can be attributed to air pollution. Chen et al. [8] estimate vironmental law, was promulgated in 2014; and the Air Pollution
that the life expectancy of residents in cities north of China’s River Huai Prevention and Control Law was implemented in 2015[15]. In 2018,
is about 5.5 years less than that of residents in other areas of the the State Council issued the “Three-Year Action Plan to Win the Blue-
country because of the practice of burning coal for winter heating. Sky Defense War2”[16].
Further, because of reduced visibility and dust, air pollution also has a Despite of these fairly strict regulations issued by the central

⁎
Corresponding author.
E-mail address: [email protected] (M. Na).
1
China state television (CCTV) has reported that fine airborne particulates in Beijing rose above 900 µg per cubic meter in January 2013, compared with World
Health Organization guidelines of no more than 25.
2
The Blue-Sky Defense Three-Year Action Plan is a pollution prevention action plan deployed by the Chinese government to continuously improve air quality and
retain more blue sky for the masses.

https://doi.org/10.1016/j.erss.2020.101773
Received 30 December 2019; Received in revised form 29 August 2020; Accepted 2 September 2020
Available online 25 September 2020
2214-6296/ © 2020 Published by Elsevier Ltd.
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

government, the implementation of environmental law remains diffi­ pollutant emissions and demonstrate to the inspection teams their de­
cult. The impediment arises from that local governments, considered as terminations to fulfill environmental targets. This leads to several
important players in monitoring and enforcement of national environ­ questions. What is the impact of CCEPI’s inspection on air pollution? Is
mental policies, have not faced sufficient incentives to prioritize en­ it effective? To what extent is it effective? Is this a “gust of wind” or a
vironmental issues. In China’s cadre evaluation system, GDP growth long-term mechanism? Given the concerns about air pollution in China,
performance is traditionally given far more weight in career promotion it is necessary to explore the implementation effect of CCEPI’s inspec­
evaluations. This political incentive system pushes local governments to tion.6
prioritize economic growth and revenue extraction even at the expense In this paper, we present the empirical evaluation of the impact of
of broader environmental interests, a situation called “pollution for CCEPI’s inspection on air pollution in Chinese cities. Since every in­
promotion” [17–21]. To pursue their own economic interests, local spection round covers a few provinces according to a sequence de­
officials have incentives to form coalitions with other economic entities, termined by the central government, it makes the uninspected pro­
including state-owned firms, to implement lax pollution abatement vinces an ideal control group. That is, when some provinces are under
policies or levy only nominal fines [22–24]. Moreover, regular rotations inspection, the other provinces become their control group; this allows
of local politicians create inherent tensions between economic effi­ the application of the difference-in-difference (DID) method of analysis.
ciency and environmental sustainability [25,26]. In this context, they Further, there exists a discontinuity in the inspection of provinces, so
tend to weight more on short-term economic growth than long-term we also use regression discontinuity design (RDD) method. In sum, to
interests of environmental protection. control for unobserved confounding factors that may affect daily air
To tackle problems of weak environmental law enforcement, the quality, we employ both DID and panel RDD methods to identify the
Central government has undertaken a number of measures to signal its effect of inspection on air pollution during the inspection period.
strong commitment on pollution reduction as a policy priority [27]. In We use the daily air pollution and weather data for 130 cities from
2005, the Chinese central government specified for the first time the 30 provinces (including four municipalities) in China over a two-year
inclusion of environmental protection in the performance assessment period (2016–2017). The major finding is that provinces inspected by
system, which stresses accountability and is crucial in cadre promotion CCEPI experience substantial improvement in air quality compared to
decisions [28]. Under the One Vote Veto system, there is no chance of uninspected provinces, with the value of the air quality index (AQI)
promotion if the binding environmental targets are not fulfilled falling by 3.7–10.9% during the inspection period. Further, we conduct
[29,30]. That means, the promotion of local leaders has been linked to an analysis on the individual pollutants in AQI, including PM2.5,
the environmental performance, measured with hard pollution reduc­ Nitrogen Dioxide (NO2), Ozone (O3), PM10, Sulfur Dioxide (SO2), and
tion targets. Moreover, the Chinese central government established an Carbon Monoxide (CO). The estimation results show a significant and
environmental inspection mechanism as early as July 2015, by dis­ negative effect of CCEPI’s inspection on particulate matter (PM) pol­
patching central inspection teams to investigate and correct environ­ lution (including that by PM10 and PM2.57) and SO2, while having re­
mental violations across the country [16]. It is aimed to create strong latively weak impacts on other pollutants. One possible explanation is
pressure for local governments to examine their environmental pro­ that PM is easily perceived by the public that people are more sensitive
tection duties, through the inspection led by the Party Central Com­ to haze than to other aspects of air pollution8. The other possible ex­
mittee and the State Council. In July 2016, the first round of inspection planation is that the central government uses PM as the main assess­
conducted by the Central Commission for Environmental Protection ment indicators for air pollution control9. To investigate whether the
Inspection (CCEPI) was launched by sending eight inspection teams to impact of inspection on air pollution is sustainable, we run tests using
eight different provinces3. In the subsequent rounds, CCEPI has strived different time horizons ranging from 10 to 60 days (in 10-day units)
to cover all provinces, including districts and cities, by regularly dis­ after the inspection. The results show there exists deterioration in air
patching inspection teams for one-month periods to examine how local quality 30 days after the inspection, with the AQI rising by 4.8%. We
politicians carry out their environmental duties4. also get similar results in the estimation of the six pollutants. Thus, the
However, influenced by the traditional obsession with GDP growth improvement in air quality turns out to be temporary, which is followed
performance and bureau leadership, local governments have to balance by higher levels of air pollution after the inspection. In sum, the CCEPI’s
the goals of environmental protection and regional economic growth. inspection achieves its prioritized target of fighting air pollution during
Unlike economic growth, air quality can be improved in the short term short-term periods, but the results have not been sustained. Moreover,
by temporarily shutting down polluted factories or enforcing vehicle we perform a series of robustness checks to confirm that our findings
restriction rule. In other words, even if local politicians devote more are stable and reliable.
resources towards economic growth at the expense of air quality during Our findings have important economic implications. The World
their terms in office, they still could take anti-pollution measures to Bank [5] reports that air pollution not only causes more than 7 million
improve air quality during important time periods. For example, Shi premature deaths each year, but also results in lost labor income for
et al. [31] find that local government takes temporary measures for working-age men and women. In terms of economic loss, China is the
controlling air pollution to create “political blue sky” during the “Two worst affected country, with an estimated 10% loss in GDP owing to air
Sessions”5 periods. Alkon and Wang [32] show that Chinese govern­
ment intentionally reins in air pollution during a two-week period in
preparation for a military parade in Beijing to ensure clear, blue skies 6
The World Bank report [5] states that at the country level, the highest po­
on the day of the parade. Similarly, we might expect that under CCEPI’s pulation-weighted mean concentration of PM2.5 estimated in 2013 was in
intense scrutiny, local bureaucrats have strong incentives to reduce Mauritania (70μg/m3), followed by that in China (55μg/m3) and Saudi Arabia
(54μg/m3). Li et al. [33] estimate that the annual average concentrations of the
six pollutants in more than 100 Chinese cities exceeded the CAAQS (Chinese
3
The eight provinces covered in the first round were Inner Mongolia, Ambient Air Quality Standards) Grade II standard during 2014–2016.
7
Heilongjiang, Jiangsu, Jiangxi, Henan, Guangxi, Yunnan, and Ningxia. We PM10 is the concentration of particulate matter with diameter of 10 µm and
provide more detailed information in section 2. smaller, and PM2.5 is the concentration of particulate matter with diameter 2.5
4
The schedule and province list for each round of inspection are determined µm or less.
8
by the central government. Usually, the official outlets release the news on the See https://www.epa.gov/pm-pollution for more information about the
day or a few days before the start of new round of inspection. severe problems caused by PM2.5.
5 9
“Two Sessions” refers to the annual sessions of National People’s Congress According to the Notice of “Measures for the Implementation of the Air
and the National Committee of the Chinese People’s Political Consultative Pollution Prevention and Control Action Plan (Trial)” issued by the General
Conference. Office of the State Council in April 2014.

2
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

pollution. Conversely, the welfare benefit of reduction in pollution empirical perspective. Our study investigates how the central inspec­
exposure, even in the short run, seems to include more working hours tion affected the air pollution and conclude that inspected provinces
and lower welfare spending. experience statistically significant improvement in air quality compared
Moreover, our findings have important policy implications domes­ to uninspected provinces, indicating that CCEPI’s field inspections are
tically and internationally. The impact of CCEPI’s inspection on fighting effective in encouraging local governments to take measures to fight air
air pollution is not sustainable and local governments tend to cope with pollution.
the central government’s inspection in the way of “surface manage­ Third, our study contributes to a growing body of research on the
ment”. Beyond the case of China, these results shed lights on plausible relationship between local officials’ behavior and environment gov­
challenges in the environmental governance system of central inspec­ ernance. Chen [42] argues that the current Chinese environmental
tion, reveal the tensions between centralized environmental targets and protection supervision system belongs to campaign-style governance
decentralized implementation of environmental issues, and suggest model so that the central inspection can’t ensure the sustainability of
more effort is needed to encourage local politicians to pursue broader environment governance. Local officials are motivated to devote their
environmental interests. efforts to short-term environmental goals. Shi et al. [31] find that local
Between 2016 and 2017, CCEPI completed four rounds of inspec­ politicians strategically create a temporary “blue sky” during the “Two
tions. We find that the impacts of inspection on air pollution vary over Sessions” period. Our findings also suggest that local bureaucrats take
the four rounds. Importantly, the number of rounds also appears sub­ temporary measures for controlling air pollution to “coping” with the
stantive and statistically significant for explaining the reduction in air central inspection. However, the relapse of environmental violation and
pollution. Compared to the first round, later rounds of inspection lead even the deterioration of air pollution after the inspection indicate that
to a remarkable mitigation of air pollution in inspected provinces. This the campaign-style governance only has a very short-term effect on air
is consistent with our expectation that the provinces inspected later are pollution control. Additionally, our study is very close to Wu and Hu
likely to learn from the experience of the provinces inspected earlier, [16] on the questions, the impact of China’s central environmental
and, thus, take necessary measures to curb pollutant emissions when protection inspection on air pollution. Yet, our analysis differs in that
they are inspected. As 19 provinces are classified as key control areas by we use city-level data with DID method, while Wu and Hu [16] use
the Ministry of Environmental Protection (MEP), we divided the sample provincial data with the RDD method. Our findings support their con­
into two subgroups and estimate the effects on key control areas and clusion that central inspection has negative effect on the AQI, but the
other areas. The empirical evidence confirms that key control areas are effect is only short-term and unsustainable.
more sensitive to inspections than others during the inspection period. The remainder of this paper is organized as follows. We provide
We also estimate whether the level of economic development influences background information in Section 2. In Section 3, we present data and
the impact of inspection on air pollution. Taking three economic re­ establish the model specification. Section 4 reports the main findings. In
gions having different levels of economic development, we find that the Section 5, we conduct further analyses on heterogeneous effects. Ro­
level of development is associated with the effect of inspection on air bustness check is reported in Section 6. Section 7 concludes the paper.
pollution. The higher the level of development is, the larger the effect of
inspection on air pollution will be. 2. Background
This paper contributes to the literature in several ways. First, it
relates to the fields of environmental politics. Some studies demonstrate As early as July 2015, the Chinese central government deliberated
how air pollution is associated with the citizen’s evaluation of the and adopted the “Environmental Protection Inspection Scheme (for
government (e.g., Alkon and Wang, 2018 [32]; Kim et al. 2019 [34]). Trial Implementation),” clearly establishing the mechanism of en­
Other studies debate the relevance of decentralization and environ­ vironmental protection inspection. Under the leadership of the MEP,
mental governance. Van Der Kamp et al. [35] show that decentraliza­ CCEPI teams were set up on January 4, 2016, with members from the
tion undermines the implementation of centrally environmental trans­ MEP, Central Discipline Inspection Commission, and Central
parency regulations. Zhang et al. [36] study whether China’s central Organization Department. The CCEPI is one of the major institutional
super-vision and local environmental decentralization management arrangements of the Communist Party of China Central Committee and
could be combined and the results support the central vertical reg­ the State Council to make local governments take responsibility for
ulatory reform. Alkon and Wong [26] focus on China’s power sector to promoting ecological progress and protecting the environment.
explain the tensions between decentralized delegation and centralized In the first two months of 2016, the CCEPI pilot was launched in
environmental goals. Our study suggests that with the increasing public Hebei Province. Five months later, the first round began with eight
dissatisfaction with air pollution, local governments fail to play their inspection teams being dispatched to 8 regions, which included Inner
role in environmental implementation due to their own parochial Mongolia, Heilongjiang, Jiangsu, Jiangxi, Henan, Guangxi, Yunnan,
economic interest. To make local officials take their environmental and Ningxia. The teams carried out supervision and inspection over a
protection responsibility, the Chinese central government established one-month period. Since then, 30 provinces of the Chinese mainland
the central inspection mechanism. Our findings contribute to the small have been inspected in the four rounds separately in July and
but growing literature on how central inspection affects the im­ November 2016, and in April and August 2017.
plementation of environmental policies by local officials. Before the inspection, schedules and other relevant details were
Second, it relates to the research on environmental governance. In published in newspapers and posted on MEP’s website. On the first day
literature, the existing studies explore the impact of policy instruments of inspection, the CCEPI team usually holds a mobilization meeting in
of environmental governance from perceptive of bureaucratic cadre the inspected province, stressing the importance of ecological devel­
mandates (Wang [27]), binding environmental targets (Kostka and Mol opment and environmental protection in promoting a “comprehensive
[37]), participatory practices (Xie [38]), public participation (Zheng four-pronged strategy,” as well as the main purposes of environmental
et al. [39], Li et al. [15]), and non-government monitoring of local inspection. In response, local governments are required to cooperate
governments (Anderson et al. [40]). However, as a new instrument of with the central inspection team in one coordinated effort against on­
environmental governance, the research on central government’s in­ going pollution problems.
spection is rather limited. Lo [41] notes that some assessments of the During the inspection, the CCEPI team collects information on en­
central government’s inspection can only be based on local government vironmental protection by making phone calls; receiving mails; lis­
self-assessment reports which gives local governments the opportunity tening to briefings or consulting reports; holding private talks; ex­
to provide false reports. Yet, the implementation effect of the central amining and reading relevant documents; and through local visits and
government’s environmental inspection has not been examined from an field trips, surveys, and questionnaires. In addition, it also includes

3
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

participatory methods such as setting up a hotline, receive mails to Table 1

allow citizens to voice complaints or report violations. After the in­ Inspection Schedule of CCEPI.
spection, the team thoroughly sorts and systematically summarizes the Round of inspection Start date End date Province
problems found during the inspection before delivering a formal report
to the MEP and the local governments. In the light of the report, the 1 7/14/2016 8/14/2016 Inner Mongolia
1 7/19/2016 8/19/2016 Heilongjiang
Discipline Inspection Group and the CPC Committee of the Departments
1 7/15/2016 8/15/2016 Jiangsu
will seriously investigate and punish the discipline breaches. In the four 1 7/14/2016 8/14/2016 Jiangxi
rounds of inspection, more than 135,000 complaints have been re­ 1 7/16/2016 8/16/2016 Henan
ceived, 18,199 government officials held accountable, and fines of RMB 1 7/14/2016 8/14/2016 Guangxi
1.43 billion imposed. 1 7/15/2016 8/15/2016 Yunnan
1 7/12/2016 8/12/2016 Ningxia
2 11/29/2016 12/29/2016 Beijing
3. Data, summary statistics, and model specification 2 11/28/2016 12/28/2016 Shanghai
2 11/26/2016 12/26/2016 Hubei
2 11/28/2016 12/28/2016 Guangdong
3.1. Data description
2 11/24/2016 12/24/2016 Chongqing
2 11/28/2016 12/28/2016 Shaanxi
To align with the World Health Organization’s standards, the MEP 2 11/30/2016 12/30/2016 Gansu
released the latest revision of Chinese Ambient Air Quality Standards 3 4/28/2017 5/28/2017 Tianjin
(CAAQS) in 2012. For the first time, the AQI, the authoritative index as 3 4/28/2017 5/28/2017 Shanxi
3 4/25/2017 5/25/2017 Liaoning
an indicator for overall air pollution, measured emissions of six in­ 3 4/27/2017 5/27/2017 Anhui
dividual pollutants—PM2.5, NO2 (nitrogen dioxide), O3 (ozone), PM10, 3 4/24/2017 5/24/2017 Fujian
SO2 (sulfur dioxide), and CO (carbon monoxide). Particularly, the first 3 4/24/2017 5/24/2017 Hunan
three pollutants are newly incorporated into the air pollution index 3 4/26/2017 5/26/2017 Guizhou
4 8/7/2017 9/7/2017 Sichuan
[43]. Since 2013, AQI data on the concentration of pollutants has been
4 8/11/2017 9/11/2017 Jilin
released to the public on a daily basis for real-time scrutiny. The AQI 4 8/11/2017 9/11/2017 Zhejiang
data used in this paper is collected from the MEP website (http:// 4 8/10/2017 9/10/2017 Shandong
datacenter.mep.gov.cn/). We explore how city-level daily air pollution 4 8/10/2017 9/10/2017 Hainan
is associated with the implementation of central environmental pro­ 4 8/15/2017 9/15/2017 Tibet
4 8/8/2017 9/8/2017 Qinghai
tection inspection. 4 8/11/2017 9/11/2017 Xinjiang
The city-level AQI, ranging from 0 to 500, is divided into six cate­
gories. An AQI below 50 is classified as level 1 and indicates excellent Note: This table presents the start and end dates of provincial visits by CCEPI
air quality; an AQI between 51 and 100 is level 2 and implies good air during 2016 and 2017.
quality; an AQI between 101 and 150 is level 3, indicating slight to light
pollution. An AQI between 151 and 200 is level 4, which denotes measures and control variables over the sample period. The average
moderate pollution, and an AQI between 201 and 300 is level 5, in­ level of air quality, which is captured by AQI, is about 80 for the sample
dicating heavy air pollution. Finally, an AQI above 300 is level 6 pol­ cities in the two-year period studied. As per the air quality categories,
lution, and suggests severe air quality. A higher AQI indicates higher this corresponds to level 2 and implies good air condition, suggesting
pollutant concentrations and more harmful health effects. the air quality is acceptable and only some hypersensitive individuals
Research shows that meteorological conditions, including air pres­ should reduce their outdoor activities11. However, we can discern a
sure, wind, temperature, and precipitation, all play important roles in huge difference in air quality across cities and time; AQI levels range
determining daily air quality variation [44]. For example, PM is af­ from the maximum of 500 to a minimum level of 12. The descriptive
fected by precipitation and wind speed, and ozone is most likely to statistics of the six pollutants are also reported in Panel A. It is worth
reach unhealthy levels on hot sunny days. Following previous studies mentioning that the average density for PM2.5 is around 45 μg/m3
[45,46], we choose temperature, humidity, wind, precipitation, and (micrograms per cubic meter of air), which is higher than the annual
sunshine duration as the control variables. Daily weather data is ob­ average level 35 μg/m3specified by the CAAQS.12 The average density
tained from the Data Share Center of the China Meteorological Ad­ for PM10 is 83 μg/m3, which exceeds the desirable annual mean level of
ministration. The detailed definition of the variables can be found in 70 μg/m3. The higher average levels of particulate matters indicate that
Appendix A. air pollution in China is still serious and continues to be a challenge.
In this study, we examine four rounds of inspection by CCEPI. However, the median levels of PM2.5 and PM10 are around 34 μg/m3and
Details about the inspections are manually collected from the MEP 63 μg/m3, respectively.
website. Given that the inspections are spread over a two-year period, In panel B, we divide the sample into two subgroups: the first one
we select January 1, 2016 to December 31, 2017 as the sample period; covers the inspection period, whereas the other is the non-inspection
it covers 30 provinces and four municipalities of China. 10Table 1 gives period. Then, we conduct mean difference tests between these two
detailed information of CCEPI’s schedule for each province and muni­ subgroups for air quality. The result shows that compared with the non-
cipality, starting from 2016. After comparing the data from different inspection period, the average AQI is 6 points lower during the in­
sources, we remove outliers and exclude the missing observations for spection period; thus, the inspection leads to an improvement in air
key variables. This leads to a final sample of 80,532 daily observations quality. It is supported by statistically significant difference between
on 130 cities.

3.2. Summary statistics 11

According to the MEP, an AQI below 100 is considered as “blue sky.”
12
In China Ambient Air Quality Standard(CAAQS), 24-hour and annual
Table 2 reports the descriptive statistics for variables used in the average levels for different pollutants are specified. Particularly, there are two
study. In Panel A, we document the detailed statistics of air quality standards in light of different functional areas. The primary standard is for
nature reserves, scenic spots and other areas requiring special protection; the
secondary standard is for residential areas, commercial transportation mixed
10
The province of Hebei is excluded from the study since it is considered as areas, cultural areas, industrial areas, and rural areas. In the paper, we refer to
the single pilot province for the inspection in January 2016. the secondary standard.

4
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 2 3.3. Model specifications

Summary Statistics.
Panel A: Main variables We use daily air pollution and weather data for 130 cities over a
two-year period to assess the effect of CCEPI inspection on air pollution.
N Mean S.D. Min Max P25 Median P75 To control for unobserved confounding factors that may affect daily air
quality, we employ both the DID and the panel RDD methods. Due to
Inspection 80,532 0.049 0.216 0 1 0 0 0
AQI 80,532 79.592 50.852 12 500 49 67 94 different inspection schedule, the provinces not undergoing inspection
CO 80,532 1.007 0.583 0 11.585 0.616 0.9 1.2 can be taken as the control group. This constitutes period-difference
NO2 80,532 30.570 17.591 2 168 18 27 40 between the inspection and non-inspection periods, and province-dif­
O3 80,532 88.896 41.336 3 300 58 82 114 ference between the inspected and uninspected provinces. We thus
PM10 80,532 82.925 96.292 0 8811 41 63 100
adopt the time-varying DID specification based on the following re­
PM2.5 80,532 44.428 42.044 1 1793 20 34 54
SO2 80,532 20.367 22.807 1 470 8 14 23 gression set-up:
Temperature 80,532 14.103 12.260 −38.8 42.3 6.9 16.8 23.5
Sunshine 80,532 5.996 4.183 0 14.6 1.7 6.9 9.3 log (APit )= 0 + 1 Inspectionit + Zit + i + µt + it (1)
Humidity 80,532 65.952 19.386 4 100 53 69 81
Precipitation 80,532 230.660 765.860 0 3270 0 0 2.8
where APit is the measure of air pollution in city i on day t; the measure
Wind 80,532 2.233 1.241 0 15.4 1.4 2 2.8
includes AQI and its six constituent pollutants. We use the logarithm
Panel B: Mean difference test form to interpret the regression coefficient as a percentage change.
Inspectionit is a dummy variable equal to 1 when city i is under CCEPI’s
Non-Inspection (1) N During inspection (2) Diff (1) - (2) t-value inspection on day t, and zero otherwise. The coefficient 1measures the
Mean N Mean impact of inspection on air pollution. Thus, a negative 1indicates that
AQI 76,569 79.880 3963 74.038 5.842*** 7.054
inspection improves air quality, whereas a positive 1suggests that in­
spection leads to a deterioration in air quality. Zit is a vector of control
CO 76,569 1.012 3963 0.899 0.113*** 11.899 variables related to weather factors for city i on day t. As environmental
NO2 76,569 30.779 3963 26.530 4.249*** 14.847 pollution is partially a function of the economy, we also control for the
O3 76,569 88.357 3963 99.323 −10.967*** −16.312
economic variable. i is a city-specific dummy variable to control for
PM10 76,569 83.610 3963 69.672 13.938*** 8.890
PM2.5 76,569 44.926 3963 34.801 10.126*** 14.803 time-invariant, unobserved city characteristics. To address any poten­
SO2 76,569 20.697 3963 13.986 6.711*** 18.098 tial effect of seasonality on air pollution, we also include year, month-
Note: This table reports the descriptive statistics for the variables used in the
by-year, and week-by-month dummies in the regression. Additionally,
study. Panel A presents the detailed statistics of air quality measures and con­ air pollution is notably affected by both production and consumption
trol variables over the sample period. Panel B reports summary statistics for key activities, although the magnitudes of the effects are still debatable
variables during the inspection period and non-inspection period. Mean dif­ [46]. We include dummies of day-by-week and public holiday in the
ferences test between inspection group and non-inspection group are conducted estimation to remove these impacts. µt is a vector of time-variant
and the results are reported in Panel B. ***, **, and * represent statistical control variables including year, month, week, day dummies, and a
significance at the 1%, 5%, and 10% levels, respectively. dummy for public holiday. it is the error term and we cluster standard
errors at province level to control for within-province correlation of air
the subsamples of inspected and uninspected provinces. We also find pollution.
significant difference between the inspected and uninspected provinces Although DID can identify the average impact, its estimation may be
in the average levels of all pollutants. influenced by other factors; thus, we use panel RDD method to better
Fig. 1 plots the monthly average trend for air quality measures (AQI solve the potential endogeneity problem.13 Importantly, as Wu and Hu
and the six pollutants) during the sample period. The pattern appears to [16] point out that RDD has been widely used in air pollution literature
have markedly seasonal characteristics, with the mean levels for AQI to investigate whether air quality is changed before and after an event
and pollutants higher in winter than summer for all pollutants except [49–51]. Thus, we follow the literature to analyze whether there is a
O3. This suggests that we need to consider seasonal adjustments in the “jump” in air quality owing to the sudden shock of inspection. The
regressions. model is designed as:

log(APit)= 0 + 1 Inspection it + 1 f(x) + Z it + i + µt + it (2)

where x is a running variable of time trend capturing the number of

days away from inspection, that is, its value is negative, 0, and, positive
before, during, and after the inspection period, respectively. Without
loss of generality, we consider symmetric time trend before, during, and
after the inspection period. f(x) is a flexible polynomial function of x
controlling for unobserved trends that may affect daily air quality.
Following the literature14, we focus on the estimates of the time trend
up to the third order. The definition of other variables is the same as
shown in Eq. (1).

13
Lee and Lemieux[47] proposed that regression discontinuity could avoid
the endogeneity in parameter estimation and faithfully reflect the correlations
between variables. For more information about the regression discontinuity
design method, please see Angrist and Pischke [48].
Fig. 1. Monthly Average of AQI and Pollutant Levels. (For comparison, we 14
Regarding the discussion of high-order polynomials of the assignment
multiply the mean value of CO by 100). variable, please see Gelman and Imbens [52] and Gelman and Zelizer [53].

5
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 3 estimate the PM2.5-related premature deaths in China and the results
Regression results with DID method (Dependent variable: log(AQI)). show that 0.03 million premature deaths were attributed to PM2.5 long-
Variable (1) (2) (3) (4) term exposure per μg/m3. Conditional on our regression result that
PM2.5 falls by 8.6% during the inspection period, that is, 3.8 μg/m3,
Inspection −0.069** −0.010 −0.037* −0.037** 114,000 premature deaths are reduced by the inspection in China15.
(0.032) (0.021) (0.019) (0.017)
Similarly, we observe the pronounced reduction of 5% in PM10
Temperature −0.002 0.019***
(0.003) (0.005)
concentration and 4.5% in O3, with statistically significance at the 10%
Sunshine −0.006* 0.008*** level. Among the remaining pollutants, we also find that they are ne­
(0.0035) (0.003) gatively associated with the inspection, although none of them are
Humidity −0.009*** −0.002** statistically significant at the conventional level. Our results suggest
(0.001) (0.001)
that in spite of the overall improvement in air quality during the in­
Precipitation −0.000 0.001
(0.000) (0.000) spection period, local politicians are more strongly motivated to control
Wind −0.070*** −0.063*** the emission of PMs on special occasions. On one hand, the public is
(0.013) (0.012) most concerned about PMs as they are main causes of haze days, which
Log(GDP) 0.100** 0.052
reduce visibility and pose serious health problems including irritating
(0.042) (0.078)
City FE N Y N Y
airways, and exacerbating respiratory diseases such as asthma. On the
Year FE N Y N Y other hand, the central government considers the abatement in PMs as
Month FE N Y N Y the key indicator of the effectiveness of air pollution control measures.
Week FE N Y N Y From the regression results reported in Table 3, we conclude that
Day FE N Y N Y
CCEPI’s inspection encourages local officials to improve air quality.
Holiday FE N Y N Y
N 80,532 80,532 80,532 80,532 However, is this improvement sustainable? To examine this question,
Adjusted R2 0.001 0.396 0.123 0.455 we include some dummy variables in the regressions to represent dif­
ferent horizons after the inspection. Specifically, using a 10-day period
Note: This table reports the impact of CCEPI’s inspection on air pollution by as a unit, we denote the 10-day horizon after the inspection with After1,
using the DID method. The dependent variable is the logarithm of AQI; the the 20-day horizon with After2, and so on, such that the 60-day horizon
control variables are a set of weather factors including temperature, humidity,
is denoted by After6. The estimation results are reported in Table 5.
wind, precipitation and sunshine hours. In addition, we also control for the
Consistent with previous results, air pollution falls by 3.5–4.3% during
economy by including an economic variable, measured as the logarithm of
provincial GDP. The variable inspection is a dummy variable equal to 1 for the the inspection period when we include control variables, and city-
sample during the inspection period, and 0 otherwise. The fixed effect dummies specific and time-varying fixed effects in the regression. After the in­
of city, year, month, week, day, and holiday are also included in the estimation. spection, as shown in column (4), air pollution rises by 1.2–6%. No­
Robust standard errors clustered by province are reported in parentheses. ***, tably, the coefficients on the variables After3 through After6 are sta­
**, and * represent statistical significance at the 1%, 5%, and 10% levels, re­ tistically significant. The immediate deterioration in air quality
spectively. demonstrates, to some extent, that local politicians are likely to take
temporary measures to improve air quality during special periods, such
4. Main results as inspections, but this improvement is followed by higher emissions of
pollutants. Our results confirm the findings of Shi et al. [31] that the
To examine the effect of CCEPI’s inspection on air pollution, we use “blue sky” created by local politicians is neither persistent nor sus­
DID and panel RDD methods to run regressions. We assume that the air tainable. When we ignore cross-time variations and holiday fixed ef­
quality will improve substantially during the inspection period; so, the fects, as shown in columns (1) and (2), we may reach the opposite
regression yields a negative coefficient on the dummy variable of in­ conclusion—the impacts of inspection on air pollution are persistent,
spection. with negative coefficients on dummy variables After2 through After6.
Again, the different regression results prove that air pollution is affected
4.1. DID estimation by seasonal characteristics.
We also run the regression on the six atmospheric pollutants using
We first estimate model (1) with DID method, and Table 3 reports different time horizons after the inspection. The findings are presented
the regression results. The dependent variable is the logarithm of AQI. in Table 6. As shown in Table 5, cities experience statistically insig­
In column (1), we can see that air pollution falls by 6.9% during the nificant increases in all air pollutants, except SO2 and NO2, after the
inspection period. However, when we add city and time-variant fixed inspection visit. It is worth noting that the highest fall —of 5.4— during
effects in column (2), the effect of inspection on air pollution becomes the inspection period is in PM2.5 levels; however, PM2.5 also registers
significantly smaller and statistically insignificant. In columns (3) and the highest rebound among atmospheric pollutants, as shown by the
(4), we report regression results by including the control variables re­ variable After6. The overall result suggests that inspections manage to
lated to weather and economy. Compared with the non-inspection curb air pollution only temporarily, and regular post-inspection mon­
period, air pollution decreases by 3.7% during the inspection period, itoring is needed to improve environmental protection.
with statistically significance at the 5% level in column (4).
Because AQI is calculated by measuring the concentrations of six 4.2. RDD estimation
atmospheric pollutants, it is necessary to run the regression to access
the impact of the CCEPI’s inspection on each pollutant. The findings are As mentioned above, DID estimation cannot solve endogeneity error
presented in Table 4, with the logarithm of each pollutant as the de­ because of omitted variables. Therefore, we use Eq. (2) with the panel
pendent variable. Among the six pollutants, PM2.5 has the largest fall of RDD method to estimate the impact of CCEPI’s inspection on air pol­
8.6% during the inspection period, with statistically significance at the lution. Given the inspection period of one month, we use symmetric
1% level. This finding has important health implications. Studies have window of one month before, during, and after the inspection with
found that particulate matter is an important risk factor for cardio­ polynomial regressions. Before the estimation, a visual representation
pulmonary disease and mortality. Xie et al. [54] estimate that the of the RDD method is plotted. From Fig. 2, we see the apparent
mortality rate caused by PM2.5 pollution in 2015 accounted for 1.75‰,
or approximately 2.62 million people and 31.14% of all deaths in
China. Li et al. [55] use local concentration–response function to 15
The mean level of PM2.5 is 44.4 μg/m3 for the sample period.

6
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 4
Regression results on six pollutants with DID method.
Variable log(PM10) log(PM2.5) log(CO) log(NO2) log(O3) log(SO2)

Inspection −0.050* −0.086*** −0.012 −0.032 −0.045* −0.056

(0.029) (0.029) (0.030) (0.025) (0.027) (0.045)
Temperature 0.021*** 0.019*** 0.005** 0.008*** 0.016*** −0.021***
(0.002) (0.003) (0.002) (0.002) (0.003) (0.003)
Sunshine −0.001 −0.001 0.000 0.003*** 0.033*** 0.006***
(0.002) (0.002) (0.001) (0.001) (0.002) (0.002)
Humidity −0.006*** 0.000 0.004*** −0.001** −0.004*** −0.008***
(0.001) (0.001) (0.001) (0.000) (0.001) (0.001)
Precipitation 0.000* 0.000 −0.000*** −0.000*** 0.000*** −0.00***
(0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Wind −0.084*** −0.116*** −0.078*** −0.163*** 0.008* −0.109***
(0.009) (0.008) (0.005) (0.008) (0.005) (0.008)
Log(GDP) −0.009 0.020 −0.010 −0.002 0.140** −0.074
(0.073) (0.114) (0.049) (0.033) (0.057) (0.069)
City FE Y Y Y Y Y Y
Year FE Y Y Y Y Y Y
Month FE Y Y Y Y Y Y
Week FE Y Y Y Y Y Y
Day FE Y Y Y Y Y Y
Holiday FE Y Y Y Y Y Y
N 80,531 80,532 80,526 80,532 80,532 80,532
Adjusted R2 0.531 0.511 0.546 0.673 0.532 0.635

Note: This table reports the impact of CCEPI’s inspection on six different atmospheric pollutants by using the DID method. The dependent variable is the logarithm of
pollutants, such as PM10, PM2.5, CO, NO2, O3, and SO2. Their regression results appear in columns (1) through (6). The variable inspection is a dummy variable equal
to 1 for the sample during the inspection period, and 0 otherwise. The fixed effect dummies of city, year, month, week, day, and holiday are also included in the
estimation. Robust standard errors clustered by province are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels,
respectively.

Table 5 discontinuity around the beginning of the inspection period (day = 0),
Regression results with DID method for different horizons after the inspection which increases our confidence in the estimation achieved with the
period (Dependent variable: log(AQI)). RDD method.
Variable (1) (2) (3) (4) Table 7 shows estimate results of our main model (2), with log(AQI)
as the dependent variable. The sample is restricted to values one month
Inspection −0.043* −0.035 −0.035* −0.040* before, during, and after the inspection period. As seen in the table, the
(0.024) (0.034) (0.020) (0.023)
coefficients on inspection are stable and statistically significant at the
After1 0.001 0.001 0.008 0.012
(0.027) (0.026) (0.020) (0.020) 1% level when control variables are included. Specifically, during the
After2 −0.070*** −0.067*** 0.005 0.017 inspection period, air pollution decreases by 9.7–10.9%.
(0.023) (0.022) (0.021) (0.025) We continue to regress the logarithm of each pollutant on dummy
After3 −0.030 −0.030 0.044 0.048* variable inspection and control variables, with the time trend terms,
After4 (0.028) (0.027) (0.027) (0.028)
After5 −0.031 −0.030 0.044 0.048*
and city-specific and time-variant fixed effects included. Table 8 shows
After6 (0.028) (0.027) (0.027) (0.028) the relevant results. As expected, pollutant emissions are negatively
−0.077*** −0.057** 0.050* 0.060** related to the inspection across all columns, and are statistically sig­
(0.030) (0.027) (0.028) (0.026) nificant at the 10% level. Similar to previous results, PM2.5 has the
−0.068** −0.038* 0.039 0.038*
largest decrease—of 10.9%—among all pollutants; this confirms that
(0.024) (0.022) (0.024) (0.023)
Control Variables Y Y Y Y the inspection objective of fighting air pollution is fulfilled to some
City FE N Y N Y degree during the inspection period.
Year FE N N Y Y In Tables 7 and 8, we focus on the global polynomial regressions.
Month FE N N Y Y We also report estimated results from local linear, and quadratic
Week FE N N Y Y
polynomial regression in Table 9. Reassuringly, both approaches gen­
Day FE N N Y Y
Holiday FE N N Y Y erate very similar results.
N 80,532 80,532 80,532 80,532
Adjusted R2 0.124 0.351 0.199 0.455
5. Further analysis
Note: This table reports regression results using the DID method, by including
different horizons after the inspection period. The dependent variable is the In this section, we test for heterogeneous effects. Specifically, we
logarithm of AQI; the control variables are a set of weather factors, including test whether the effect of inspection on air pollution is greater among
temperature, humidity, wind, precipitation, and sunshine hours. In addition, we provinces in later rounds of inspection, in key control areas strictly
also control for the economy by including an economic variable, measured as regulated by the MEP, or in regions with high level of economic de­
the logarithm of provincial GDP. The variable inspection is dummy variable velopment.
equal to 1 for the sample during the inspection period, and 0 otherwise. The
independent variables After1, After2, After3, After4, After5, and After6 refer to
horizons of ten days, twenty days, thirty days, forty days, fifty days, and sixty 5.1. Does the order of rounds matter?
days after the inspection period, respectively. The fixed effect dummies of city,
year, month, week, day, and holiday are also included in the estimation. Robust As mentioned earlier, there are four rounds of inspections in our
standard errors clustered by province are reported in parentheses. ***, **, and * sample. We assume that the first round of inspection might have a weak
represent statistical significance at the 1%, 5%, and 10% levels, respectively. effect on air pollution because local politicians have little experience in
taking preemptive actions to improve air quality at short notice. As

7
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 6
Results on six pollutants using the DID method for different horizons after the inspection period.
Variable log(PM10) log(PM2.5) log(CO) log(NO2) log(O3) log(SO2)

Inspection −0.048* −0.054** −0.033* −0.033 −0.031 −0.006

(0.027) (0.025) (0.018) (0.020) (0.026) (0.027)
After1 −0.002 −0.033 0.022 0.001 −0.014 −0.064*
(0.029) (0.023) (0.035) (0.020) (0.047) (0.049)
After2 −0.008 −0.043 0.003 0.001 −0.015 0.004
(0.052) (0.070) (0.030) (0.021) (0.040) (0.054)
After3 0.012 −0.001 0.006 −0.032* −0.001 −0.021
(0.063) (0.076) (0.033) (0.019) (0.041) (0.054)
After4 0.001 0.027 0.025 −0.035 0.060** −0.048
(0.062) (0.081) (0.029) (0.034) (0.041) (0.044)
After5 0.034 0.044 0.050* 0.007 0.055* −0.031
(0.046) (0.065) (0.027) (0.032) (0.030) (0.045)
After6 0.056 0.068 0.005 −0.016 0.046* −0.014
(0.044) (0.057) (0.024) (0.031) (0.042) (0.048)
Control Variables Y Y Y Y Y Y
City FE Y Y Y Y Y Y
Year FE Y Y Y Y Y Y
Month FE Y Y Y Y Y Y
Week FE Y Y Y Y Y Y
Day FE Y Y Y Y Y Y
Holiday FE Y Y Y Y Y Y
N 80,531 80,532 80,526 80,532 80,532 80,532
Adjusted R2 0.531 0.511 0.546 0.673 0.530 0.635

Note: This table reports the impact of CCEPI’s inspection on six different atmospheric pollutants using the DID method by including different horizons after the
inspection period. The dependent variable is the logarithm of pollutants, such as PM10, PM2.5, CO, NO2, O3, and SO2, and their regression results appear in columns
(1) through (6). The variable inspection is a dummy variable equal to 1 for the sample during the inspection period, and 0 otherwise. The independent variables
After1, After2, After3, After4, After5, and After6 refer to time horizons of ten days, twenty days, thirty days, forty days, fifty days, and sixty days after the inspection
period, respectively. Control variables are a set of weather factors including temperature, humidity, wind, precipitation, and sunshine hours. The economic variable,
measured as the logarithm of provincial GDP is also controlled. The fixed effect dummies of city, year, month, week, day, and holiday are also included in the
estimation. Robust standard errors clustered by province are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels,
respectively.

Fig. 2. Mean levels of AQI 30 days before and after inspection.

inspections proceed, the provinces to be inspected may realize the pronounced improvement in air quality.
importance of the inspection and take urgent measures in a timely
manner to reduce air pollution. Therefore, we expect larger reductions 5.2. Key control areas versus other areas
in pollution during later rounds of inspection. Specifically, we divide
the whole sample into four groups to check the progressive effects of In 2013, the MEP released the “Notice on Special Emission Limits on
inspection on air pollution across the four rounds. The estimation re­ Air Pollutants,” in which 19 provinces were classified as key control
sults are presented in panel A of Table 10. The indicator variable in­ areas [13]16. We want to estimate whether these key control areas show
spection has an insignificant but positive coefficient in the first round. a greater reduction in air pollution than other areas during the in­
Conversely, during the follow-up rounds of inspection, the reduction in spection period. The regression results are reported in Panel B of
air pollution is of a much larger and significant magnitude. These
findings are consistent with our anticipation that provinces covered in
16
later rounds of inspection are more likely to learn from the experience The key control areas are Beijing, Tianjin, Hebei, Shanghai, Jiangsu,
of the provinces inspected in the first round and exhibit more Zhejiang, Guangdong, Liaoning, Shandong, Hubei, Hunan, Chongqing, Sichuan,
Fujian, Shanxi, Shaanxi, Gansu, Ningxia, and Xinjiang.

8
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 7
Regression results with RDD method.
Variable (1) (2) (3) (4) (5) (6)

Inspection −0.106** −0.097** −0.104** −0.107*** −0.102*** −0.109***

(0.049) (0.036) (0.039) (0.036) (0.029) (0.030)
Temperature 0.038*** 0.038*** 2.687***
(0.004) (0.004) (0.313)
Sunshine 0.013*** 0.013*** 0.358
(0.002) (0.002) (0.253)
Humidity −0.003** −0.003** −0.219
(0.001) (0.001) (0.139)
Precipitation −0.000 −0.000 −0.001
(0.000) (0.000) (0.001)
Wind −0.052*** −0.052*** −0.052***
(0.008) (0.008) (0.008)
Log(GDP) 0.311 0.311 0.264
(0.252) (0.254) (0.238)
Time trend 1st order 2nd order 3rd order 1st order 2nd order 3rd order
City FE Y Y Y Y Y Y
Year FE Y Y Y Y Y Y
Month FE Y Y Y Y Y Y
Week FE Y Y Y Y Y Y
Day FE Y Y Y Y Y Y
Holiday FE Y Y Y Y Y Y
N 10,414 10,414 10,414 10,414 10,414 10,414
Adjusted R2 0.533 0.533 0.535 0.616 0.616 0.618

Note: This table reports the impact of CCEPI’s inspection on air pollution by using the RDD method. The dependent variable is logarithm of AQI; the control variables
are a set of weather factors including temperature, humidity, wind, precipitation, and sunshine hours. The economic variable, measured as the logarithm of
provincial GDP is also controlled. The variable inspection is a dummy variable equal to 1 for the sample during the inspection period, and 0 otherwise. The sample is
restricted to one month before, during, and after the inspection period. Trend is in its liner term; trend2 and trend3 are the 2nd order and 3rd order of trend,
respectively. The fixed effect dummies of city, year, month, week, day, and holiday are also included in the estimation. Robust standard errors clustered by province
are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.

Table 8
Regression results on six pollutants by using the RDD method.
Variable log(PM10) log(PM2.5) log(CO) log(NO2) log(O3) log(SO2)

Inspection −0.085** −0.109** −0.035* −0.058** −0.056* −0.095***

(0.033) (0.042) (0.02) (0.023) (0.028) (0.024)
Time trend Y Y Y Y Y Y
Control variables Y Y Y Y Y Y
City FE Y Y Y Y Y Y
Year FE Y Y Y Y Y Y
Month FE Y Y Y Y Y Y
Week FE Y Y Y Y Y Y
Day FE Y Y Y Y Y Y
Holiday FE Y Y Y Y Y Y
N 10,414 10,414 10,413 10,414 10,414 10,414
Adjusted R2 0.658 0.649 0.745 0.773 0.688 0.710

Note: This table reports the impact of CCEPI’s inspection on six different atmospheric pollutants by using the RDD method. The dependent variable is the logarithm of
pollutants, such as PM10, PM2.5, CO, NO2, O3, and SO2. Their regression results appear in columns (1) through (6). The variable inspection is a dummy variable equal
to 1 for the sample during the inspection period, and 0 otherwise. The sample is restricted to one month before, during, and after the inspection period. Control
variables are a set of weather factors including temperature, humidity, wind, precipitation, and sunshine hours. The economic variable, measured as the logarithm of
provincial GDP is also controlled. The polynomial time trend terms up to the third order, and the fixed effect dummies of city, year, month, week, day, and holiday
are also included in the estimation. Robust standard errors clustered by province are reported in parentheses. ***, **, and * represent statistical significance at the
1%, 5%, and 10% levels, respectively.

Table 10. For the group of key control areas, the effect of inspection on implemented the coastal opening policy and have a high level of eco­
air pollution is four times more than that in the group consisting of nomic development; the Central region includes underdeveloped areas;
other areas, with significance at 1% level, suggesting that key control and the Western region consists of less developed areas.17 In light of
areas are more sensitive to the inspection. these, we run a regression to assess whether the effect of the inspection
on air pollution varies across the economic regions. The regression

5.3. Economic regions

17
Eastern region includes the provinces of Shanghai, Tianjin, Shandong,
According to the directive of the National Development and Reform Guangdong, Guangxi, Hebei, Jiangsu, Zhejiang, Hainan, Fujian, Liaoning, and
Commission, China was officially divided into three regions (Eastern, Beijing; the central region includes the provinces of Shanxi, Inner Mongolia,
Central, and Western) as early as 1986, at the Sixth Session of the Jilin, Anhui, Jiangxi, Henan, Hubei, Hunan; Western region includes the pro­
National People's Congress; this was done for the purpose of economic vinces of Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang, Sichuan, Yunnan,
policy. The Eastern region includes mostly provinces that first Guizhou and Tibet.

9
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 9
Regression results with Local Linear Polynomial.
Variable Log(AQI) log(PM10) log(PM2.5) log(CO) log(NO2) log(O3) log(SO2)

Panel A
Inspection −0.88*** −0.8** −0.11*** −0.01 −0.99*** −0.12*** −0.04
(0.02) (0.03) (0.03) (0.02) (0.03) (0.03) (0.03)
P 1 1 1 1 1 1 1
N 80,532 80,532 80,532 80,532 80,532 80,532 80,532
Panel B
Inspection −0.11*** −0.16** −0.101** −0.006 −0.08** −0.13*** −0.03
(0.03) (0.045) (0.046) (0.035) (0.04) (0.04) (0.05)
P 2 2 2 2 2 2 2
N 80,468 80,468 80,468 80,468 80,468 80,468 80,468

Note: This table reports the impact of CCEPI’s inspection on air pollution and six different atmospheric pollutants by using local linear polynomial time trend. The
dependent variable is the logarithm of pollutants, such as AQI, PM10, PM2.5, CO, NO2, O3, and SO2. Their regression results appear in columns (1) through (7). The
variable inspection is a dummy variable equal to 1 for the sample during the inspection period, and 0 otherwise. Control variables are a set of weather factors
including temperature, humidity, wind, precipitation, and sunshine hours. The economic variable, measured as the logarithm of provincial GDP is also controlled. P is
the order of the local polynomial to be used, 1 is for the linear polynomial, 2 is for the quadratic polynomial. The fixed effect dummies of city, year, month, week, day,
and holiday are also included in the estimation. Robust standard errors clustered by province are reported in parentheses. ***, **, and * represent statistical
significance at the 1%, 5%, and 10% levels, respectively.

Table 10 6.1. Different time windows

Regression results with heterogeneity analysis.
Variable (1) (2) (3) (4)
For estimation with RDD, the use of different time windows may
capture any unusual shift in air pollution levels during the inspection
Panel A 1st round 2nd round 3rd round 4th round period. To check whether our main findings are sensitive to the window
Inspection 0.042 −0.178** −0.130*** −0.070** width, we select 5-day, 10-day, 15-day, 20-day, 25-day, and 40–day
(0.030) (0.089) (0.050) (0.033)
N 4070 1816 1587 2942
windows. Table 11 reports the results, which show that the fall in air
Adjusted R2 0.624 0.542 0.515 0.617 pollution ranges from 6.4% and 10.6% during the inspection period for
Panel B Key control areas Other areas the different time windows. As the time window becomes wider, the
Inspection −0.171*** −0.040 effect of inspection on air pollution is more pronounced. These results
(0.049) (0.046)
with different-sized windows remain robust.
N 5639 4775
Adjusted R2 0.507 0.643
Panel C Eastern region Central region Western region 6.2. Placebo test
Inspection −0.222*** −0.080* −0.081***
(0.054) (0.045) (0.030)
N 4070 2941 4808
Our identification strategy relies on the assumption that different
Adjusted R2 0.608 0.667 0.654 inspection schedules make provinces not undergoing inspection be
served as the control group. To do the placebo test, we re-estimate our
Note: This table reports the effects of inspection on air pollution with hetero­ models by shifting the inspection date one year ahead. The logarithm of
geneity analysis using the RDD method. Panel A presents regression results for AQI and each pollutant are regressed as the dependent variables. As
different rounds of inspection; panel B reports results for key control areas and shown in Table 12, none of our effects is statistically significant any
other areas; and Panel C lists relevant results for three economic regions. The
more, ensuring the robustness of our results.
dependent variable is the logarithm of AQI; inspection is a dummy variable
equal to 1 for the sample during the inspection period, and 0 otherwise. The
control variables are a set of weather factors including temperature, humidity, 6.3. Category of AQI
wind, precipitation, and sunshine hours. The economic variable, measured as
the logarithm of provincial GDP is also controlled. The polynomial time trend As Wallace [56] notes, there are some doubts that officials have
terms up to the third order, and the fixed effect dummies of city, year, month, incentives to manipulate and distort statistics. One issue of concern is
week, day and holiday are also included in the estimation. Robust standard that the MEP defines “blue sky” as an AQI below 100. Thus, local
errors clustered by province are reported in parentheses. ***, **, and * re­ governments fulfill their environmental target, in a sense, if the AQI is
present statistical significance at the 1%, 5%, and 10% levels, respectively. below 100; otherwise, the officials responsible may fail the perfor­
mance assessment. Thus, some scholars believe that local politicians
results are reported in Panel C of Table 10. We find that all regions have incentives to falsify data on AQI by increasing its level above 100.
experience statistically significant reduction in air pollution during the To avoid the effects of false data, we drop some ranges of AQI around
inspection period. However, the effect of inspection is strongest in the the threshold of 100 [31]. Specifically, we remove AQI levels in the
Eastern region (having the highest level of economic development), following ranges: (95, 105), (90, 110), (85, 115), and (80, 120).
more than double of the effect in the Central and Western region. This Table 13 presents the results. The coefficients are negatively significant,
result suggests that people’s awareness of environmental protection implying that our findings are stable. In sum, the government could
may improve with the development of the economy. temporarily reduce air pollution during the inspection period.

7. Conclusion
6. Robustness check
In this paper, we investigate the effect of inspection by CCEPI on
Because the reliability of estimation results is possibly affected by
urban air pollution in China. Although China has made great progress
other conditions, it is necessary to conduct some robustness checks. In
in establishing a system of environmental laws, problems of weak en­
this section, we run regressions to test the reliability of our main results.
forcement of these laws still remain. Since July 2016, the MEP has

10
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Table 11
Robustness Check with different time windows.
Variable (1) ± 5 days (2) ± 10 days (3) ± 15 days (4) ± 20 days (5) ± 25 days (6) ± 40 days

Inspection −0.064** −0.080*** −0.091*** −0.097*** −0.106*** −0.098***

(0.029) (0.031) (0.032) (0.033) (0.031) (0.031)
N 5166 6306 7406 8408 9410 12,308
Adj. R2 0.666 0.649 0.629 0.621 0.619 0.601

Note: This table reports the effects of inspection on air pollution over different time windows using the RDD method. Columns (1) through (6) present regression
results for 5 days, 10 days, 15 days, 20 days, 25 days, and 40 days. The dependent variable is logarithm of AQI; the independent variable inspection is a dummy
variable equal to 1 for the sample during the inspection period, and 0 otherwise. We include a set of weather factors as control variables in the regression; these
include temperature, humidity, wind, precipitation, and sunshine hours. The economic variable, measured as the logarithm of provincial GDP is also controlled. The
polynomial time trend terms up to the third order, and the fixed effect dummies of city, year, month, week, day, and holiday are also included in the estimation.
Robust standard errors clustered by procince are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.

Table 12
Placebo tests.
Variable log(AQI) log(PM10) log(PM2.5) log(CO) log(NO2) log(O3) log(SO2)

Inspection −0.03 −0.0080 −0.012 −0.011 0.047 −0.048 0.011

(0.04) (0.038) (0.054) (0.039) (0.031) (0.032) (0.056)
N 80,532 80,532 80,532 80,532 80,532 80,532 80,532
Adj R2 0.455 0.531 0.510 0.546 0.673 0.532 0.635

Note: This table reports the impact of CCEPI’s inspection on air pollution and six different atmospheric pollutants by using local linear polynomial time trend. The
dependent variable is the logarithm of pollutants, such as AQI, PM10, PM2.5, CO, NO2, O3, and SO2. Their regression results appear in columns (1) through (7). We set
the inspection date one year ahead so that the variable inspection is equal to 1 for the date one year before the inspection period, and 0 otherwise. Control variables
are a set of weather factors including temperature, humidity, wind, precipitation, and sunshine hours. The economic variable, measured as the logarithm of
provincial GDP is also controlled. The fixed effect dummies of city, year, month, week, day, and holiday are also included in the estimation. Robust standard errors
clustered by province are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.

is followed by even higher air pollution levels after the inspection. We

Table 13 also find that the effect of inspection on air pollution is larger in the
Robustness Check on removing some AQI ranges. later rounds of inspection, key control areas, and regions with a high
level of economic development. We perform a series of robustness
Variable (1) remove (2) remove [90, 110] (3) remove [85,
[95,105] 115]
checks to confirm that our findings are stable and reliable.
However, our study presents empirical evidence that the inspection
Inspection −0.047** −0.045** −0.039* winds up with a temporary effect to fight air pollution and leads to un-
(0.022) (0.02) (0.023) sustained compliance of environmental protection policies. Due to the
N 75,530 70,767 65,983
high costs of monitoring compliance, the overall frequency of central
Adjusted R2 0.461 0.468 0.476
inspections is not very high. One potential approach is to facilitate
Note: This table reports the effects of inspection on air pollution using the RDD bottom-up monitoring of environmental performance of local govern­
method when we remove some AQI ranges. Specifically, we drop the following ments, such as encouraging citizens, non-governmental organizations
AQI ranges: (95, 105), (90, 110) and (85, 115). The dependent variable is and media to enhance the accountability of environmental governance.
logarithm of AQI; the independent variable inspection is a dummy variable Moreover, the recurring of pollution violations and poor implementa­
equal to 1 for the sample during the inspection period, and 0 otherwise. The tion of environmental policies by local governments suggest that eco­
sample is restricted to 30 days before, during, and after the inspection. We nomic efficiency comes into tension with environmental protection.
include a set of weather factors, including temperature, humidity, wind, pre­
Local governments, when pursuing their economic interests, should
cipitation, and sunshine hours, as control variables in the regression. The
consider how to boost economic growth in a clean and economically
economic variable, measured as the logarithm of provincial GDP is also con­
trolled. The polynomial time trend terms up to the third order, and the fixed
viable way, such as promoting the rapid development of renewable
effect dummies of city, year, month, week, day, and holiday are also included in energy. According to the Chinese 13th Five-Year Plan (2016–2020), the
the estimation. Robust standard errors clustered by province are reported in non-fossil energy sources are becoming the prominent part of the
parentheses. ***, **, and * represent statistical significate at the 1%, 5%, and Chinese energy and the five sixths of carbon reduction target will be
10% levels, respectively. achieved by developing and utilizing clean energy technology, and by
improving energy efficiency of high-consuming industries. Only this
conducted four rounds of inspection by dispatching teams to various way may achieve goals of reducing air pollution and ensure the public
provinces all over China; the aim is to examine how local politicians to enjoy the basic environmental rights of clean air.
carry out their environmental duties. This offers an ideal setting for us
to examine how local politicians react to the inspection.
With the DID and RDD methods, our main finding is that provinces Declaration of Competing Interest
inspected by CCEPI experience substantial improvement in air quality
compared to uninspected provinces. The pattern of lower emissions is The authors declare that they have no known competing financial
more pronounced in the case of PM2.5, the pollutant responsible for the interests or personal relationships that could have appeared to influ­
most concern among the public owing to the severe harm it can cause. ence the work reported in this paper.
However, the improvement in air quality turns out to be temporary, and

11
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

Acknowledgements Author note

This work was supported by project of Humanities and Social Lan Zheng is an associate-professor in School of Economics, Hefei
Science of Anhui Education Department [grant numbers University of Technology. Her specialty is energy economy.
JS2017AJRW0015, JS2018AHZS0020]. Ming Na is an associate-professor in School of Economics, Hefei
University of Technology. Her specialty is energy economy.

Appendix A

We list the detailed definition of variables in Table A1.

Table A1
Variable descriptions
Variables Descriptions Source

Inspection A dummy variable that equals to 1 if the province is under inspection, 0 otherwise Hand collected
Temperature Average temperature measured as centigrade Data Share Center of China Meteorological Administration
Wind Average wind speed measured as kilometers per hour Data Share Center of China Meteorological Administration
Precipitation Cumulative level of precipitation in 24 h measured as millimeters Data Share Center of China Meteorological Administration
Humidity Average humidity measured as percentage Data Share Center of China Meteorological Administration
Sunshine Sunshine duration in 24 h measured as hours Data Share Center of China Meteorological Administration
AQI Average Quality Index Website of the MEP
PM2.5 Particulate matter with diameters 2.5 µm and smaller, measured as μg/m3 Website of the MEP
PM10 Particulate matter with diameters 10 µm and smaller, measured as μg/m3 Website of the MEP
NO2 Nitrogen dioxide, measured as μg/m3 Website of the MEP
O3 Ozone, measured as μg/m3 Website of the MEP
SO2 Sulfur dioxide, measured as μg/m3 Website of the MEP
CO Carbon monoxide, measured as μg/m3 Website of the MEP

References responsibility on financial performance: the moderating role of government reg­

ulation and organizational slack, J. Clean Prod. 166 (2017) 1323–1334 http://
www.scholarmate.com/S/VBKnyX.
[1] J. Liu, J. Diamond, China’s environment in a globalizing world, Nature 435 (7046) [16] R. Wu, P. Hu, Does the “Miracle Drug” of environmental governance really improve
(2005) 11–79 http://en.cnki.com.cn/Article_en/CJFDTOTAL-SJHJ200504016.htm. air quality? Evidence from China’s system of central environmental protection in­
[2] W. Lyu, Y. Li, D.B. Guan, H.Y. Zhao, Q. Zhang, Z. Liu, Driving forces of Chinese spections, Int. J. Environ. Res. Public Health (2019) 850.
primary air pollution emissions: an index decomposition analysis, J. Clean. Prod. [17] M. Peng, H. Zhang, R. Evans, D. Zhong, K. Yang, Actual air pollution, environmental
133 (2016) 136–144 https://www.sciencedirect.com/science/article/abs/pii/ transparency, and the perception of air pollution in China, J. Environ & Dev. 28 (1)
S0959652616303699. (2019).
[3] D. Zhao, H. Chen, X. Li, X. Ma, Air pollution and its influential factors in China's hot [18] V. Shih, C. Adolph, M. Liu, Getting ahead in the communist party: explaining the
spots, J. Clean. Prod. 185 (2018) 619–627 https://www.onacademic.com/detail/ advancement of central committee members in china, Amer. Pol. Sci. Rev. 106 (1)
journal_1000040202356310_5510.html. (2012) 166–187 https://www.onacademic.com/detail/journal_
[4] Y. Guo, H. Zeng, R. Zheng, S. Li, A.G. Barnett, S. Zhang, X. Zou, R. Huxley, W. Chen, 1000038592068710_461a.html.
G. Williams, The association between lung cancer incidence and ambient air pol­ [19] C. Xu, The fundamental institutions of China’s reforms and development, J. Eco. Lit.
lution in China: a spatiotemporal analysis, Environ. Res. 144 (2016) 60–65 https:// 49 (4) (2011) 1076–1151 https://www.aeaweb.org/articles?id=10.1257/jel.49.4.
www.sciencedirect.com/science/article/abs/pii/S0013935115301341. 1076.
[5] World Bank Report, The cost of air pollution–strengthening the economic case for [20] X. Lu, P.F. Landry, Show me the money: interjurisdiction political competition and
action (2016). https://openknowledge.worldbank.org/bitstream/handle/10986/ fiscal extraction in China, Amer. Pol. Sci. Rev. 108 (3) (2014) 706–722 https://
25013/108141.pdf?sequence=4&isAllowed=y. www.onacademic.com/detail/journal_1000040883157910_b481.html.
[6] World Health Organization, 9 out of 10 people worldwide breathe polluted air, but [21] S. Eaton, K. Genia, Authoritarian environmentalism undermined? Local leaders’
more countries are taking action (2018). http://www.who.int/news-room/ detail/ time horizons and environmental policy implementation in China, China Qua. 218
02-05-2018-9-out-of-10-people-worldwide-breathe-polluted-air-but-more-coun (2014) 359–380 https://www.onacademic.com/detail/journal_
tries-are-taking-action. 1000037485245210_87b8.html.
[7] A. Hsu, D.C. Esty, M. Levy, A. Sherbinin, 2016 Environmental Performance Index [22] B.V. Rooij, Implementation of Chinese environmental law: regular enforcement and
(EPI), Yale University, New Haven, 2016 http://www.indiaenvir­ political campaigns, Dev. Chan. 37 (1) (2006) 57–74 https://www.onacademic.
onmentportal.org.in/content/424381/2016-environmental-performance-index/. com/detail/journal_1000034503031310_9aef.html.
[8] Y. Chen, J.G. Zhe, K. Naresh, G. Shi, The promise of Beijing: evaluating the impact [23] X. Zhang, L. Ortolano, L. Zhongmei, Agency empowerment through the adminis­
of the 2008 Olympic Games on air quality, J. Environ. Econ. Manag. 66 (2013) trative litigation law: court Enforcement of pollution levies in Hubei Province,
424–443 https://www.sciencedirect.com/science/article/abs/pii/ China Qua. 202 (2010) 307–326 https://www.cambridge.org/core/journals/china-
S0095069613000685. quarterly/article/agency-empowerment-through-the-administrative-litigation-law-
[9] A.L. Sexton, Health and environmental implications of Americans? Time use re­ court-enforcement-of-pollution-levies-in-hubei-province/
sponses to external stimuli essays on air-quality alerts and daylight savings time, 842783B0DBA103AA755D5F94A9303E3F.
Int. J. Work Organ. & Emotion 1 (1) (2012) 87–102. [24] C. Lo, G.E. Fryxell, B.V. Rooij, W. Wang, P.H. Li, Explaining the enforcement gap in
[10] S.Q. Zheng, X.N. Zhang, Y. Sun, The mechanism of the effect of air pollution on China: local government support and internal agency obstacles as predictors of
outdoor activities of urban residents, J. Tsinghua Univ. 56 (1) (2016) 89–96. enforcement actions in Guangzhou, J. of Environ Mana. 111 (2012) 227–235
[11] Z.J. Graff, M. Neidell, The impact of pollution on worker productivity, Am. Econ. https://www.sciencedirect.com/science/article/pii/S0301479712003891.
Rev. 102 (7) (2012) 3652–3673 https://www.nber.org/papers/w17004. [25] S. Eaton, G. Kostka, Authoritarian environmentalism undermined? Local leaders’
[12] G.M. Lepori, Air pollution and stock returns: evidence from a natural experiment, J. time horizons and environmental policy implementation in China, The China
Empir. Finance 35 (2016) 25–42 https://www.sciencedirect.com/science/article/ Quarterly 218 (2014) 359–380.
abs/pii/S0927539815001061. [26] M. Alkon, A. Wong, Authoritarian energy transitions undermined? Environmental
[13] X. Wu, S. Chen, J. Guo, G. Cao, Effect of air pollution on the stock yield of heavy governance cycles in China’s power sector, Energy Res. Soc. Sci. 68 (10) (2020),
pollution enterprises in China's key control cities, J. Clean Prod. 170 (2018) https://www.sciencedirect.com/science/article/abs/pii/S2214629620301080.
399–406 https://kns.cnki.net/kcms/detail/detail.aspx?filename= [27] A. Wang, The search for sustainable legitimacy: environmental law and bureau­
SJES84B9585CB232CF6CF6D3EB7A3F46EBD9&dbcode=SJES. cracy in China, Soc. Sci. Electr. Publ. 37 (2) (2013) 365–440 https://escholarship.
[14] Y. Hao, Y. Deng, Z.N. Lu, H. Chen, Is environmental regulation effective in China? org/uc/item/8hc948zx.
Evidence from city-level panel data, J. Clean. Prod. (2018) 966–976. [28] J.J. Schreifels, Y. Fu, E.J. Wilson, Sulfur dioxide control in China: policy evolution
[15] D. Li, C. Cao, L. Zhang, X. Chen, S. Ren, Y. Zhao, Effects of corporate environmental during the 10th and 11th five-year plans and lessons for the future, Energy Pol. 48

12
L. Zheng and M. Na Energy Research & Social Science 70 (2020) 101773

(3) (2012). Policy (2015) 152–159.

[29] Y. Jin, H. Andersson, S. Zhang, Air pollution control policies in China: a retro­ [42] H.S. Chen, Legalization of environmental protection supervision system: posi­
spective and prospects, Int. J Environ. Res Publ. Health 13 (12) (2016) 12–19 tioning, dilemma and its Way out, Law Rev. 3 (2016) 176–187 http://www.en.cnki.
https://www.mdpi.com/1660-4601/13/12/1219. com.cn/Article_en/CJFDTotal-FXPL201703019.htm.
[30] S. Zheng, M.E. Kahn, W. Sun, D. Luo, Incentivizing China’s urban mayors to miti­ [43] B. Zhang, C. Cao, Policy: four gaps in China's new environmental law, Nature 517
gate pollution externalities: the role of the central government and public en­ (7535) (2015) 433–435 https://www.nature.com/news/policy-four-gaps-in-china-
vironmentalism, NEBER working paper (2013).https://www.sciencedirect.com/ s-new-environmental-law-1.16736.
science/article/abs/pii/S0166046213000690. [44] J. Liang, L. Langbein, Performance management, high-powered incentives, and
[31] Q. Shi, F. Guo, S. Chen, “Political blue sky” in fog and haze governance—evidence environmental policies in China, Int. Public Manag. J. (2015) 346–385.
from the local annual ‘two sessions’ in China, China Ind. Econ. 5 (2016) 40–56 [45] Y.J. Guo, Y.H. Zhang, Will air quality affect the stock market? J. Fin. Res. 2 (2016)
http://en.cnki.com.cn/Article_en/CJFDTOTAL-GGYY201605004.htm. 71–85.
[32] M. Alkon, E.H. Wang, Pollution Lowers Support for China’s Regime: Quasi- [46] S. Fu, Y. Gu, Highway toll and air pollution: evidence from Chinese cities, J.
Experimental Evidence from Beijing, J. Polit. 80 (1) (2018) 327–331 https://www. Environ. Econ. Manag. 83 (2017) 32–49 https://www.sciencedirect.com/science/
onacademic.com/detail/journal_1000040102252210_0dd4.html. article/abs/pii/S009506961630451X.
[33] R. Li, L.L. Cui, J.L. Li, A. Zhao, H.B. Fu, Y. Wu, L.W. Zhang, L.D. Kong, J.M. Chen, [47] D. Lee, T. Lemieux, Regression discontinuity designs in economics, J. Econ. Lit. 48
Spatial and temporal variation of particulate matter and gaseous pollutants in China (2010) 281–355 https://www.nber.org/papers/w14723.
during 2014–2016, Atmos. Environ. 161 (2017) 235–246 https://www. [48] J. Angrist, J.S. Pischke, Mostly harmless econometrics: an empiricist's companion,
sciencedirect.com/science/article/abs/pii/S1352231017303096. Princeton University Press, Princeton, New Jersey, 2009.
[34] S.E. Kim, S.P. Harish, R. Kennedy, X. Jin, J. Urpelainen, Environmental degradation [49] V.B. Viard, S. Fu, The effect of Beijing’s driving restrictions on pollution and eco­
and public opinion: the case of air pollution in Vietnam, J. Environ. Develop. 29 (2) nomic activity, J. Public Econ. (2015) 98–115.
(2020) 196–222 https://journals.sagepub.com/doi/abs/10.1177/ [50] X.C. Yang, F. Peng, L.L. Kang, Green Metro: the Effects of New Metro Line 3 on Air
1070496519888252?journalCode=jeda. Quality in Nanjing, J. Gansu Polit. Sci. Law Inst. 4 (2017) 82–94 http://www.en.
[35] D. Van Der Kamp, P. Lorentzen, D. Mattingly, Racing to the bottom or to the top? cnki.com.cn/Article_en/CJFDTOTAL-GSXX201704009.htm.
Decentrali-zation, revenue pressures, and governance reform in China, World Dev. [51] Q. Chen, F.K. Sun, Y.Q. Xu, Does Winter Heating Cause Smog? Evidence from a City
95 (2017) 164–176 https://www.sciencedirect.com/science/article/abs/pii/ Panel in North China, Nankai Econ. Study 4 (2017) 25–40 http://en.cnki.com.cn/
S0305750X15308020. Article_en/CJFDTotal-NKJJ201704002.htm.
[36] B. Zhang, X. Chen, H. Guo, Does central supervision enhance local environmental [52] A. Gelman, G. Imbens, Why high-order polynomials should not be used in regres­
enforcement? Quasi-experimental evidence from China, J. Public Econ. (2018) sion dis-continuity designs? NBER Working Paper (2014).
70–90. [53] A. Gelman, A. Zelizer, Evidence on the deleterious impact of sustained use of
[37] G. Kostka, A.P.J. Mol, Implementation and participation in China’s local environ­ polynomial regression on causal inference, Res. Polit. (2015) 1–7.
mental politics: challenges and innovations, J. Environ. Pol. Plann. 15 (1) (2013) [54] Z. Xie, Y. Qin, L. Zhang, R. Zhang, Death effects assessment of PM2.5 pollution in
3–16 https://www.tandfonline.com/doi/abs/10.1080/1523908X.2013.763629. China, Pol. J. Environ. Stud. 27 (4) (2018).
[38] L. Xie, Environmental governance and public participation in rural China, China Inf. [55] J. Li, H. Liu, Z. Lv, R. Zhao, F. Deng, C. Wang, A. Qin, X. Yang, Estimation of PM2.5
30 (2) (2016). mortality burden in China with new exposure estimation and local concentration-
[39] S. Zheng, G. Wan, W. Sun, Citizen complaints and urban environmental governance, response function, Environ. Poll. 243 (Part B) (2018) 1710–1718. https://www.
Manag. World 6 (2013) 72–84. sciencedirect.com/science/article/abs/pii/S0269749118332627.
[40] S.E. Anderson, M.T. Buntaine, M. Liu, B. Zhang, Non‐governmental monitoring of [56] J.L. Wallace, Juking the Stats? Authoritarian information problems in China,
local governments increases compliance with central mandates: a national‐scale British. J. Politic, Sci, 2014 https://www.cambridge.org/core/journals/british-
field experiment in China, Am. J. Political Sci. 63 (7) (2019) 626–643. journal-of-political-science/article/juking-the-stats-authoritarian-information-pro­
[41] K. Lo, How authoritarian is the environmental governance of China? Environ, Sci. blems-in-china/3A9711B9C9A38CCFD7C31AD5D90344FA.

13