ECONOMICS OF ENVIRONMENTAL AND PUBLIC HEALTH POLICIES
by
SHAN ZHANG
A DISSERTATION
Presented to the Department of Economics
and the Division of Graduate Studies of the University of Oregon
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
June 2023
DISSERTATION APPROVAL PAGE
Student: Shan Zhang
Title: Economics of Recycling and Public Health Policies
This dissertation has been accepted and approved in partial fulfillment of the
requirements for the Doctor of Philosophy degree in the Department of Economics
by:
Eric Zou Chair
Trudy Ann Cameron Core Member
Grant McDermott Core Member
Kory Russel Institutional Representative
and
Krista Chronister Vice Provost for Graduate Studies
Original approval signatures are on file with the University of Oregon Division of
Graduate Studies.
Degree awarded June 2023
ii
© 2023 Shan Zhang
All rights reserved.
iii
DISSERTATION ABSTRACT
Shan Zhang
Doctor of Philosophy
Department of Economics
June 2023
Title: Economics of Recycling and Public Health Policies
This dissertation is comprised of three papers that investigate the effects
of recycling policy, pollution, and public health policy. The first paper examines
the impact of China’s waste import ban on U.S. polltion emissions at the national
and state level. The second paper studies the distributional effect of China’s waste
import policy on waste transfers across local communities in California. The third
paper investigates people’s willingness to pay for public health policies to protect
the health of a community during the COVID-19 pandemic. Chapter 1 provides a
comprehensive overview of each dissertation chapter.
Chapter 2 analyzes the effect of China’s Green Sword policy (waste import
ban) on U.S. emissions at the national and state level. Using the synthetic control
method, the study finds that many states experienced significant increases in
methane emissions after the policy took effect, with the total U.S. methane
emissions from the waste industry increasing by 10%. The study also finds a
positive correlation between the waste trade each state had with China before
the ban and the increase in emissions after the policy, suggesting that the states
that relied more on trading recyclable wastes with China were more affected by the
policy.
iv
Chapter 3 examines the effects of the Green Sword policy on the relocation
of solid waste pollution across local communities in California. Using detailed waste
transfer data from California, the study finds that Black communities received
more waste transfers before the policy, but after the policy, relatively more waste
pollution relocated to lower-income White communities. The study identifies land
costs as the primary explanation for this distributional effect.
Chapter 4 (co-authored with Trudy Ann Cameron) utilizes a choice-
experiment survey of U.S. residents to determine people’s willingness to pay for
public policies to reduce illnesses and premature deaths in their communities. The
study estimates people’s ex-ante willingness to pay to avoid the actual monthly
totals of COVID-19 cases and deaths from March 2020 to April 2021 by county
and month. The estimated aggregate willingness to pay across the U.S. adult
population during this period is about 3 trillion dollars.
v
CURRICULUM VITAE
NAME OF AUTHOR: Shan Zhang
GRADUATE AND UNDERGRADUATE SCHOOLS ATTENDED:
University of Oregon, Eugene, OR, USA
University of Florida, Gainesville, FL, USA
Dalian Nationalities University, Dalian, Liaoning, China
DEGREES AWARDED:
Doctor of Philosophy, Economics, 2023, University of Oregon
Master of Business, International Business, 2018, University of Florida
Bachelor of Science, Economics, 2015, Dalian Nationalities University
AREAS OF SPECIAL INTEREST:
Environmental Economics
Health Economics
PROFESSIONAL EXPERIENCE:
Economic Consultant, World Bank, 2023
Research Economist, Resources for the Future, 2022
Research Assistant to Dr. Scott Kostyshak, University of Florida, 2017
GRANTS, AWARDS AND HONORS:
Graduate Teaching Award, University of Oregon, 2022
Gary E. Smith Professional Development Award, University of Oregon, 2022
Kleinsorge Summer Research Award, University of Oregon, 2022
Sloan Summer School Diversity Fellowship, UC Berkeley, 2021
Kleinsorge Summer Research Award, University of Oregon, 2021
Qualification Exam Scholarship Award, University of Oregon, 2019
vi
ACKNOWLEDGEMENTS
I wish to extend my deepest appreciation to my committee chair—Eric Zou,
and Trudy Cameron, who have consistently offered their unwavering support and
invaluable guidance during my journey through graduate school. I am also indebted
to Grant McDermott and Kory Russel, whose insightful feedback has significantly
contributed to different sections of this dissertation. I must also express my
heartfelt gratitude to my family and friends for their limitless support. Above
all, my husband, Ryan Buttacavoli, deserves my special thanks for his relentless
encouragement and support.
vii
For my parents, who inspired and taught me the value of education, and my future
children, to whom I hope to impart the same knowledge.
viii
TABLE OF CONTENTS
Chapter Page
I. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . 1
II. THE IMPACT OF CHINA’S WASTE IMPORT BAN ON
POLLUTION EMISSIONS IN THE UNITED STATES . . . . . . . . 5
2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2. Background and Data . . . . . . . . . . . . . . . . . . . . . 10
2.2.1. Background . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2. Data . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.3. Summary Statistics . . . . . . . . . . . . . . . . . . . 16
2.3. The impact of China’s GS policy on Emissions in the U.S. . . . . . 17
2.3.1. Raw Trends . . . . . . . . . . . . . . . . . . . . . . 17
2.3.2. State-level Emissions . . . . . . . . . . . . . . . . . . 19
2.3.3. Emission Changes and Waste Export Exposure . . . . . . . 23
2.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.5. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.6. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
III. THE EFFECTS OF CHINA’S WASTE IMPORT BAN
ON LOCAL COMMUNITIES: A CASE STUDY OF
CALIFORNIA . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2. Background and Data . . . . . . . . . . . . . . . . . . . . . 55
3.2.1. Background . . . . . . . . . . . . . . . . . . . . . . 55
3.2.2. Data . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.3. Summary Statistics . . . . . . . . . . . . . . . . . . . 60
3.3. Determinants of Pollution Relocation: Evidence from California . . . 61
ix
Chapter Page
3.3.1. Raw Patterns . . . . . . . . . . . . . . . . . . . . . . 61
3.3.2. Distributional Effects . . . . . . . . . . . . . . . . . . 63
3.3.3. Mechanism . . . . . . . . . . . . . . . . . . . . . . . 67
3.4. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.5. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.6. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
IV. WILLINGNESS TO BEAR THE COSTS OF
PREVENTATIVE PUBLIC HEALTH MEASURES . . . . . . . . . . 83
4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.1. The Original 2003 Survey . . . . . . . . . . . . . . . . 91
4.2.2. County-level Sociodemographic and Contextual Heterogeneity 94
4.3. Estimating Specification . . . . . . . . . . . . . . . . . . . . 95
4.4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.4.1. Identifying Dimensions of Heterogeneity: LASSO Estimation . 97
4.5. Benefit Transfer: 2020-21 WTP to Avoid COVID-19
Illnesses and Deaths in Each Month . . . . . . . . . . . . . . . 98
4.5.1. Preferences for a Representative Individual in
Each County, for Each County-Month of the
2020-21 Pandemic . . . . . . . . . . . . . . . . . . . . 107
4.5.2. Parametric Bootstrap Estimates of Predicted
WTP in Each County-month . . . . . . . . . . . . . . . 107
4.5.3. Scaling to Monthly National Total WTP
Amounts to Avoid COVID-19 Cases and Deaths . . . . . . 113
4.5.4. Systematic Heterogeneity in Predicted WTP to
Reduce COVID-19 Cases and Deaths . . . . . . . . . . . 116
4.6. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . 117
x
Chapter Page
V. DISSERTATION CONCLUSION . . . . . . . . . . . . . . . . . 120
APPENDICES
A. CHAPTER 2: APPENDIX . . . . . . . . . . . . . . . . . . . 122
A.1. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
A.2. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
B. CHAPTER 3: APPENDIX . . . . . . . . . . . . . . . . . . . 135
B.1. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
B.2. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
C. CHAPTER 4: APPENDIX . . . . . . . . . . . . . . . . . . . 149
C.1. Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
C.2. Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
REFERENCES CITED . . . . . . . . . . . . . . . . . . . . . . . . 159
xi
LIST OF FIGURES
Figure Page
1. U.S. Recyclable Waste Exports to China and the Rest Of
the World (ROW) . . . . . . . . . . . . . . . . . . . . . . . . 34
2. Composition of U.S. Recyclable Waste Exports to China . . . . . . . 35
3. Plastic Waste Exports from the U.S. and Other OECD
Countries to China . . . . . . . . . . . . . . . . . . . . . . . . 36
4. U.S. Waste Industry GHG Emissions . . . . . . . . . . . . . . . . 37
5. U.S. Greenhouse Gas Emissions (log.MMT) by Industry . . . . . . . . 38
6. State-level Synthetic Control Results . . . . . . . . . . . . . . . . 39
7. State-level Synthetic Control Results: Four States . . . . . . . . . . 40
8. State-level Synthetic Control Results: Placebo Tests . . . . . . . . . 41
9. State-level Synthetic Control Results—Estimates of the
Percentage and Net Change in GHG Emissions from the
Waste Industry . . . . . . . . . . . . . . . . . . . . . . . . . 42
10. Pairwise Correlations: State-level Emission Changes and
Waste Trade Exposures . . . . . . . . . . . . . . . . . . . . . . 43
11. Pairwise Correlations: State-level Emission Changes and
Paper/paperboard Exports . . . . . . . . . . . . . . . . . . . . 44
12. U.S. Waste Exports and Methane Emissions from the Waste Sector . . . 45
13. CalRecycle: Average Net Increase of Disposal Flow after
China’s GS Policy Los Angeles and Covanta Landfill . . . . . . . . . 74
14. CalRecycle: Average Net Increase of Disposal Flow after
China’s GS Policy San Francisco and Potrero Hills Landfill . . . . . . 75
15. CalRecycle: Average Net Increase of Disposal Flow by
Racial Composition . . . . . . . . . . . . . . . . . . . . . . . . 76
16. CalRecycle: Average Net Increase of Disposal Flow by
Registered Voters . . . . . . . . . . . . . . . . . . . . . . . . . 77
xii
Figure Page
17. Correlations of Disposal Flow and Destination Community Characteristics 78
18. CalRecycle: Effect of Destination Community
Characteristics on Waste Flows Before and After China’s
GS Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
19. Potential Mechanism: Disposal Flow Map by Political Deviation . . . . 80
20. Survey Respondents across the U.S. . . . . . . . . . . . . . . . . . 92
21. Aggregate WTP to Reduce the Risk of COVID-19 among
all U.S. Counties in 2020-21 . . . . . . . . . . . . . . . . . . . . 115
22. Spatial Distribution of WTP to Reduce the Risk of COVID-
19 among all U.S. Counties in December 2021. . . . . . . . . . . . . 116
A1. Other Countries’ Recyclable Waste Exports to China and
the Rest Of the World (ROW) . . . . . . . . . . . . . . . . . . . 122
A2. Composition of U.S. Recyclable Waste Exports to the Rest
of the World (ROW) . . . . . . . . . . . . . . . . . . . . . . . 123
A3. GHGRP: Spatial Distribution of Waste Facilities . . . . . . . . . . . 124
A4. Synthetic Control: Waste Industries and Other Industries . . . . . . . 125
A5. Synthetic Control: Exports of Other Control Industries . . . . . . . . 126
A6. Pairwise Correlations: Heterogeneous Effects of GS Policy
on State-level Estimations and Trade Exposures . . . . . . . . . . . 127
A7. Pairwise Correlations: Heterogeneous Effects of GS policy
on State-level Estimations and Paper Exports . . . . . . . . . . . . 128
B1. Data Comparison: EPA GHGRP v.s. CalRecycle RDRS . . . . . . . . 135
B2. CalRecycle: Recycling and Disposal Reporting System
(RDRS) Facility Locations in California by Rural and
Urban Areas . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
B3. Disposal Flow Map by Median Income . . . . . . . . . . . . . . . 137
B4. Disposal Flow Map by Environmental Vulnerability . . . . . . . . . . 138
B5. CalRecycle: Recycling and Disposal Reporting System
(RDRS) Facility Locations in California by Racial Composition . . . . 139
B6. Racial Composition Variation within Los Angeles County . . . . . . . 140
xiii
Figure Page
B7. Racial Composition Variation within Santa Clara County . . . . . . . 141
B8. Voting Variation within Los Angeles County . . . . . . . . . . . . . 142
B9. Voting Variation within Santa Clara County . . . . . . . . . . . . . 143
B10. Economies of Scale of Communities Where Destination
Facilities are Located . . . . . . . . . . . . . . . . . . . . . . . 144
B11. Economies of Scale of communities Where Destination
Facilities are Located (cont.) . . . . . . . . . . . . . . . . . . . . 145
C1. One Example of a Choice Set in the Original 2003 Survey . . . . . . . 149
xiv
LIST OF TABLES
Table Page
1. Data Sources Summary: State-level Analysis . . . . . . . . . . . . . 46
2. Summary Statistics: Recyclable Waste Exports by the U.S.
and Other Countries . . . . . . . . . . . . . . . . . . . . . . . 47
3. Models to Explain the Changes in Methane Emissions as a
Function of the Changes in Recyclable Waste Exports . . . . . . . . . 48
4. Data Sources Summary: Community-level Analysis . . . . . . . . . . 81
5. Potential Mechanisms: Model Estimates . . . . . . . . . . . . . . . 82
6. Descriptive Statistics, Public Health Policy Design Variables,
Choice Experiments Posed within the 2003 Estimating Sample . . . . . 94
7. WTP Estimation Model Results . . . . . . . . . . . . . . . . . . 99
8. Descriptive Statistics, 2020-21 COVID-19 New Cases and
Deaths (in hundreds), County-level. . . . . . . . . . . . . . . . . . 106
9. Descriptive Statistics, 2003 Estimating Sample vs 2020
Simulation Sample, County-level Heterogeneity(for candidate
interaction terms considered in Lasso model, where not all
interactions are retained.) . . . . . . . . . . . . . . . . . . . . . 108
10. County Representative Individual’s (monthly) WTP to
Reduce COVID-19 Cases and Death through 2020-21 . . . . . . . . . 113
A1. Summary Statistics: U.S. Recyclable Waste Exports (by
Type of Materials) . . . . . . . . . . . . . . . . . . . . . . . . 129
A2. Summary Statistics: U.S. GHG Emissions (MMT) by Industry . . . . . 130
A3. EPA: Waste Sector - Greenhouse Gas Emissions Reported to
the GHGRP Summary Statistics . . . . . . . . . . . . . . . . . . 131
A4. Summary Statistics: U.S. GHG Emissions by Industry . . . . . . . . . 132
A5. Synthetic Control Results: Estimates at State Level . . . . . . . . . . 133
xv
Table Page
A6. Synthetic Control Results: Emission Increases across States
(million metric tons of CO2 eq.) . . . . . . . . . . . . . . . . . . 134
B1. CalRecycle: Recycling and Disposal Reporting System
(RDRS) Disposal Flows within California, Summary
Statistics (Thousands of Tons) . . . . . . . . . . . . . . . . . . . 146
B2. Summary Statistics for Community Characteristics around
each Destination Facility . . . . . . . . . . . . . . . . . . . . . . 147
B3. Altered Distributional Effects: Fixed-effects Model for Waste
Flows from Origin Jurisdiction to Receiving Facilities and
their Local Communities, before/after China’s GS Policy . . . . . . . . 148
C1. Sources of county-level data for estimation and simulation . . . . . . . 150
C2. Homogeneous Preferences versus Model with Three Latent Classes . . . 154
xvi
CHAPTER I
INTRODUCTION
The goal of my dissertation is to shed light on important policy questions
related to recycling and their impact on emissions and pollution transfer, as well
as people’s willingness to pay for public health measures. This study examines
heterogeneity in responses to a set of environmental and health policies, utilizing
various methods including machine learning, spatial analysis, and causal inference
analysis. Chapters 2 and 3 are the author’s original work, while Chapter 4 is a
collaborative effort with Trudy Ann Cameron.
Chapter 2 of this study examines the impact of China’s waste import ban
on U.S. pollution at both the national and state levels. To estimate the policy’s
effects on emissions from the waste industry in specific states, the study uses the
U.S. EPA Greenhouse Gas Reporting Program (GHGRP), which records annual
methane emissions at the facility level for all industries. Emissions from other
industries that emit methane, but are not affected by the waste import ban, are
used as controls in the identification strategy. For instance, emissions from the
waste industry in California are used as the treatment, while emissions from other
non-waste industries, such as oil and gas, mining, and refining industries, from
California and all other states are used as the control group. The synthetic control
method is applied to all states in the U.S. resulting in a map showing the various
changes in emissions across states. The study finds that larger states, in particular,
experience a greater increase in emissions following China’s waste ban. To validate
state-specific estimates, the study examines exposure, which refers to the amount of
waste trade a state has had with China historically. The results indicate that states
1
more exposed to China’s policy shock tend to experience larger pollution increases,
on average, after the shock.
In addition to the synthetic control method, the study employs a second
method, the Bartik instrument. In this part of analysis, waste export data from
sources like USA Trade Online and UN Comtrade are utilized. The method
instruments for the current shocks in waste exports by using historical state-level
export shares multiplied by the national shock component. The study estimates
that for every metric ton of waste exported, domestic methane emissions from the
waste industry decrease by 0.89 metric tons. Overall, after the implementation
of China’s waste ban, there was a reduction of 12 million metric tons in U.S.
waste exports to China. Consequently, the U.S. methane emissions is estimated
to experienced an increase of 11 million metric tons.
In Chapter 3, I move beyond state-level analysis to examine the
distributional effects of China’s waste import ban on local communities. I
investigate which communities were previously receiving more waste treatment
and how the policy impacted them. With the ban in place, more recyclable waste
remains in the U.S., but which communities are now bearing greater pollution costs
due to the policy? Specifically, I examine which facilities are handling more waste
treatment and their proximity to White and Black communities. To achieve this,
I collect detailed demographic and socioeconomic data at the census block level,
as well as California waste transfer records. These data allow me to construct a
fixed-effects model and identify which communities are most affected by the policy.
Before the policy, I replicate the standard result that minority communities were
more likely to receive waste transfers from outside, which is consistent with much
of the existing Environmental Justice literature. However, after the policy, I find
2
that White communities experience a relatively greater increase in waste flows
compared to Black communities, which is a novel finding. The racial gap in waste
transfers narrowed after China’s policy shock. To understand what is driving this
narrowing of the gap, I propose three mechanisms: land costs, transportation costs,
and political costs. I proxy for land costs and transportation costs using population
density and distance, and use the degree of polarization in political views between
a community and its county as a proxy for the political costs of the destination
communities. I find that, prior to the waste import ban, transportation costs and
political costs were the primary drivers of waste transfers across communities.
However, after the policy shock, low land costs of a community became a significant
determinant of the communities to which the waste was relocated to.
In Chapter 4, my co-author Trudy Cameron and I utilize a 2003 general-
population choice-experiment survey of U.S. residents to assess people’s willingness
to pay for public policies that aim to reduce illnesses and avoid premature deaths
in their communities. We refine earlier models by excluding all respondent-specific
individual characteristics and incorporating county-level data on various contextual
variables from 2003. We then adapt our re-estimated model to the 2020-21 COVID-
19 pandemic, using updated contextual variables such as household incomes and
unemployment rates. We compute the model’s implied values of ex-ante willingness
to pay (WTP) to prevent the actual monthly numbers of cases and deaths from
March 2020 to April 2021, by county and month, across the contiguous U.S.
Our estimated total WTP by the U.S. adult population during this period is
approximately 3 trillion dollars, although these estimates are conservative as the
original choice scenarios did not involve infectious illnesses or pandemics. Our
analysis indicates that people from counties with a higher proportion of working
3
adults and Black residents tend to have a greater WTP for public health policies
that mitigate the risks of illness and death. Individuals from counties with higher
incomes and better health access also tend to exhibit higher WTP. If preferences
for public health programs remain relatively stable over time, our findings may
be useful for predicting contemporary WTP for public health measures, both
retrospectively for the current pandemic and prospectively for future pandemics.
4
CHAPTER II
THE IMPACT OF CHINA’S WASTE IMPORT BAN ON POLLUTION
EMISSIONS IN THE UNITED STATES
2.1 Introduction
As of 2016, half of the world’s scrap intended for recycling was traded
internationally, with more than one billion metric tons of waste transferred from
developed to developing countries over the prior two decades.1 China has been the
largest recipient of this waste, accounting for 45% of the world’s total since 1992,
and 70% when including Hong Kong, which returned to Chinese sovereignty in
1997. This pattern has enabled China’s factories to access inexpensive materials
and has thus supported its emerging economy, but recycling has resulted in
increased pollution from energy-intensive processes and contaminated imported
wastes (Kellenberg, 2012; Kellenberg, 2015; Higashida & Managi, 2014; Gregson &
Crang, 2015; Lee, Wei, & Xu, 2020).
In 2017, China enacted its “Green Sword” (GS) policy, which prohibited
the import of most plastics, papers, and other materials for recycling processors,
effectively shutting down the world’s largest market for recyclable waste. According
to the Institute of Scrap Recycling Industries, as of October 2019, U.S. scrap
exports of plastic to mainland China had decreased by 89% since early 2017, while
mixed paper exports had fallen by 96%. The Chinese government implemented
this policy to address local environmental concerns such as air and water pollution,
as well as to safeguard public health. Studies have shown that since the policy’s
implementation, Chinese cities connected to waste importation and reprocessing
1Statistics according to the UN Comtrade data
5
have significantly improved their air quality (J. Li & Takeuchi, 2021; Unfried &
Wang, 2022).
China’s GS policy has significantly affected the recycling industry in the
United States, formerly the leading exporter of recyclable waste to China.2 After
China implemented the waste ban, several lower-income countries in Southeast
Asia, despite having a lack of the necessary infrastructure to handle recyclables
correctly, have attempted to take on some of the load. However, these countries
were quickly overwhelmed by the volume of waste and most have since enforced
their own limitations on paper or plastic imports.3 Thus far, no single nation has
completely replaced China as the world’s primary market for recyclable waste.
Consequently, the United States had to confront its own waste issues, uncovering
fundamental flaws in its domestic recycling procedures. The escalating costs of
transporting recyclable materials have rendered the practice unprofitable, causing
an increase in the quantities of plastics and paper waste that end up in landfills and
incinerators, polluting the environment.
This paper quantitatively studies the impact of China’s waste import ban
on the quantity and distribution of methane emissions from the waste industry,
particularly landfill facilities, in the United States. I provide empirical evidence
necessary to answer the following questions: (1) How has the GS policy affected
domestic methane emissions at the national level in the U.S.? (2) How do changes
in methane emissions relate to recyclable waste exports at the state level in the
U.S.?
2In 2016, China imported about 17 billion tons of recyclable waste from the U.S., which made
up 72.9% of the total U.S. waste exported.
3For example, in May 2018, Indonesia required 100 percent inspection of scrap paper and
plastic imports. In March 2019, India announced a ban on scrap plastic imports.
6
The paper uses the EPA Greenhouse Gas Reporting Program to study
the causal effect of China’s GS policy on U.S. domestic methane emissions. The
synthetic control method is used to estimate the effect of the waste ban on methane
emissions from the waste industry in each state of the U.S. The study finds that
11 states, particularly larger states, have experienced a statistically significant
increase in methane emissions from their waste industry after the waste ban, with
California seeing a 9% increase in methane emissions. Aggregating the changes in
methane emissions from all states, the study finds that the overall U.S. methane
emissions from the waste industry increased by almost 10 million metric tons of
CO2 equivalent (mmt CO2 eq.), accounting for 10% of total U.S. methane emissions
from the waste industry in 2016.4 The study also finds that the more waste a state
exported before China’s waste ban, the greater the impact of the waste ban on the
state’s methane emissions.
In the second part of my analysis, I delve deeper into the connection
between U.S. emissions and exports of recyclable waste. If states with a history
of high waste exports experienced a greater increase in methane emissions after
the GS policy, then exports of recyclable waste must have been reducing emissions
successfully in those states prior to the policy’s implementation. For this analysis, I
use data on recyclable waste exports by state and year from U.S.A. Trade Online,
as well as methane emissions data from the waste industry by state and year from
the U.S. EPA Greenhouse Gas Inventory. To estimate the causal relationship
between exports and emissions, I must account for the fact that U.S. waste
exports are endogenous to U.S. economic activities and emissions. To address
this issue, I employ a Bartik shift-share instrumental variable. This instrument
4This number is calculated by aggregating the changes in emissions of all states whose
synthetic control estimates can be judged to be statistically significant.
7
starts with the initial-year shares of recyclable waste exports by state and shifts
these over time using annual aggregated recyclable waste exports from the U.S.
to China. The underlying assumption is that the initial-year shares of recyclable
waste exports by state are not related to future waste exports from the U.S. to
China. As a result, recyclable waste exports weighted by the initial-year state -
level shares are exogenous to future emissions and economic activities. Using this
method, I discover that prior to China’s GS policy, for each additional metric ton
of recyclable waste exported, U.S. domestic emissions were lower by 0.83 metric
tons of CO2 eq. This finding supports the hypothesis that U.S. recyclable waste
exports directly reduced domestic emissions from the waste industry. After China’s
GS policy was enacted, U.S. exports of recyclable waste to China declined by 12
million metric tons, accompanied by a corresponding increase of approximately 11
million metric tons of CO2 eq. in methane emissions from the U.S. waste industry.
This figure is comparable to the result obtained using the synthetic control method,
thus cross-validating my findings from the two different approaches.
My research makes several significant contributions. Firstly, mine is the
first study to investigate quantitatively the effects of China’s groundbreaking GS
policy on U.S. environmental quality at the national and state levels. The recycling
sector is an essential but under-researched area in environmental economics, and
China’s GS policy, as a significant shock in this context, has received relatively
little attention. Previously, most research focused on the efficiency of recycling
programs in developed countries and their impact on social welfare, revealing low
efficiency and low social welfare (Aadland & Caplan, 2006; Bohm, Folz, Kinnaman,
& Podolsky, 2010; Kinnaman, 2014; Kinnaman, Shinkuma, & Yamamoto, 2014).
In contrast, my research demonstrates that under China’s exogenous policy shock,
8
recycling in the U.S. not only has low efficiency but also negatively and unevenly
affects the domestic environment.
Second, my paper is the first to study the causal relationship between trade
volume and emissions in the context of a high-pollution industry, the recycling
industry, and a specific trade policy change. In recent years, more papers have
begun to focus on the relationship between international trade policies and
emissions (J. S. Shapiro, 2016; R. Shapiro Joseph S.and Walker, 2018). Copeland,
Shapiro, and Taylor (2021) show that nearly one-fourth to one-third of global
pollution emissions stem from industrial processes related to international trade.
With increasing exposure to trade, dirty industries in rich countries tend to relocate
their production to developing countries with more-lenient environmental policies.
Many international trade policies, such as tariffs, have tended to impose more costs
on “clean” industries than on industries with pollution (J. S. Shapiro, 2021). My
research also sheds light on the pollution haven hypothesis, which posits that
developed countries tend to relocate their pollution elsewhere through trade,
typically to developing countries. Before China’s GS policy, my research indicates
that recyclable waste exports directly reduced domestic emissions in the U.S.,
supporting this hypothesis. My findings thus differ from those in the previous
trade and environment literature, which has focused mainly on the impact of trade
policies on general industries versus polluting industries.
My research demonstrates that many states in the U.S. are impacted to
varying degrees by international environmental policy changes. Therefore, my
research findings have important implications for domestic recycling policies.
Given the rapidly changing international markets, these policies must be regularly
updated. Ignoring the international context for domestic recycling policies is no
9
longer an option. Notably, recent legislative developments, such as the RECYCLE
Act of 2021, the Recycling Infrastructure and Accessibility Act of 2022, the Plastic
Waste Reduction and Recycling Research Act, and the $350 million solid waste and
recycling grant from the Infrastructure Bill 2021 passed by the U.S. EPA, indicate
that recycling is receiving greater attention at the national level.5
The rest of this chpater is organized as follows. Section 2 provides a brief
background concerning international trade in recyclable waste and methane
emissions in the United States and outlines my data sources. Section 3 identifies
the impact of China’s GS policy on domestic emissions in the U.S. Section 4
concludes with a few caveats and suggestions for future research.
2.2 Background and Data
2.2.1 Background.
Methane Emissions from Recyclables. According to the U.S. EPA,
methane emissions from landfills rank as the third largest source of human-related
methane emissions in the U.S. Anaerobic decomposition of organic food, wood, and
paper scraps are the primary sources of methane emissions from landfills. The GS
policy affects the most-recycled products and materials in the U.S., particularly
”mixed paper and paperboard” and plastic scrap, which account for 85 percent and
14 percent of total recyclable waste exports from the U.S., respectively before 2017.
5While recycling regulations are typically enforced by states or counties, various recycling
bills are currently being proposed by legislators across different states. These bills may propose
conflicting regulations due to the diverse needs and circumstances of each state. Some bills
suggest creating grant programs to educate people on recycling and improve recycling accessibility
in communities, while others propose expanding producer responsibility for material use and
reducing existing residential recycling programs.
10
After the GS policy, most of these recyclable paper products end up in landfills,
contributing to methane emissions.6
Aside from emissions and pollution from the degradation of recyclable
materials themselves, organic food residues on recyclables can also contribute to
methane emissions from landfills. Before China’s GS policy, recyclable materials
were separated and cleaned for export, resulting in less associated food waste
contamination. However, after the GS policy, these recyclable materials are often
merely thrown into the trash without being cleaned, likely contributing to an
increase in landfill methane emissions.
I choose methane emissions as the main environmental outcome in this
paper due to a number of reasons. First, methane emissions have been the most
consistently recorded emissions at U.S. waste industry facilities by the U.S. EPA
over time.7 This makes it the most suitable choice for the empirical design of this
paper. I employ methane emissions as an indicator for overall pollution emissions
originating from the waste industry.8 Second, as a greenhouse gas, methane is
85 times more potent than carbon dioxide in trapping heat, and it stays in the
atmosphere for a shorter period of time than carbon dioxide. Thus, curbing
methane emissions will have a more immediate impact on reducing global warming
in the near future. U.S. Regulators are beginning to acknowledge the significance
of reducing methane emissions as the EPA is currently reassessing its regulations
6Royer, Ferron, Wilson, and Karl (2018) find that the degradation of plastic in landfills can
also emit methane, and the longer it remains, the more methane it emits.
7Due to inadequate oversight, methane emissions are the only emissions that the waste
industry regularly monitors. Furthermore, local pollutant monitoring around landfill facilities
is lacking when compared to other types of plants.
8It is understood that an increase in waste treatment activities at a facility results in a
proportional rise in general pollution emissions, which consequently leads to greater methane
emissions.
11
on methane pollution in the oil and gas industry under the Clean Air Act (CAA).9
At some point, regulations concerning landfills, which constitute the third-biggest
source of methane pollution, will need to be reassessed. Third, methane emissions
can serve as a proxy for other types of pollution caused by recyclable wastes, such
as soil and water pollution, as well as air pollution. These types of pollution are
generally difficult to track around landfills, but they often co-occur with wastes
that generate methane emissions. In the case of air pollution, although methane
and carbon dioxide make up 90 to 98% of total landfill gases, the remaining 2
to 10% of other gases from potentially recyclable wastes include volatile organic
compounds (VOCs), nitrogen, oxygen, ammonia, hydrogen sulfides, and surface
ozone for which even small amounts can affect the health of people living nearby
(Abernethy, O’Connor, Jones, & Jackson, 2021).10
It is noteworthy that methane emissions from landfills can be captured,
converted, and utilized as a source of energy for electricity, renewable natural gas,
and direct use. Nevertheless, according to the Landfill Methane Outreach Program
(LMOP) National Map, only around 20 percent of all landfill facilities in the United
States have implemented landfill gas (LFG) energy projects as of 2022.
China’s Green Sword Policy. China’s Green Sword Policy was implemented
in response to a significant increase in U.S. recyclable waste exports to China
within a decade of China joining the WTO in 2001. The increase was due to
several factors, including China serving as an international pollution haven, China’s
economic growth and expanding manufacturing industry, and excess capacity
9https://www.epa.gov/newsreleases/us-sharply-cut-methane-pollution-threatens-climate-and-
public-health
10Abernethy et al. (2021) shows that methane removal can reduce surface ozone, a local
pollutant, and temperature.
12
on ships returning to China from the U.S. (Kellenberg, 2012; Kellenberg, 2015;
Bransetter & Lardy, 2006; Brandt, Biesebroeck, & Zhang, 2012; Higashida &
Managi, 2014; Palma, Lindsey, Quinet, & Vickerman, 2011; Olivia, 2014) As
China’s policies concerning environmental quality, regulation, and pollution evolved
between 2010 and 2019, the government introduced the Green Fence (GF) policy
in 2013, which had minimal effects on the total quantity of recyclable wastes being
imported (Greenstone, He, Li, & Zou, 2021).11 In 2017, however the much more
stringent Green Sword (GS) policy was launched, which imposed much stricter
contamination limits and banned many types of recyclables.12 This policy caused
a significant decrease in U.S. exports of all affected recyclable materials to China,
but a significant increase in U.S. waste exports to other developing countries in
South and Southeast Asia. In response, many South and Southeast Asian countries
enacted similar stringent policies to restrict waste inflows and control illegal
processing facilities. Additionally, in 2019, the Basel Convention (to which the U.S.
is not a party) was revised to limit the cross-border transfer of plastic waste from
developed nations to developing nations, and this agreement has been signed by 192
countries. Consequently, markets for U.S. recyclable wastes are much more limited
than in the past.13
11The Green Fence policy imposed strict quality standards on imported recyclable materials,
rejecting shipments that were contaminated or mixed with non-recyclable waste.
12According to the WTO Committee on Technical Barriers to Trade Notification, China forbade
the importation of 24 kinds of solid waste, including plastic waste, vanadium slag, unsorted paper,
cotton, and textile materials.
13Due to the GS policy, mixed paper and paperboard exports dropped from 15.1 billion tons in
2016 to just 5.4 billion tons in 2019 (a 64.24% decrease). Plastic scrap exports dropped from 2.89
billion tons in 2016 to 0.18 billion tons in 2019 (a 93.8% decrease). Other exports of recyclable
wastes, such as cotton waste, man-made fibers, and textiles, decreased by 96.4%, 69.8%, and
99.5%, respectively.
13
2.2.2 Data.
Trade Data. The data I utilize for trade in recyclable materials trade comes
from U.S.A. Trade Online and covers the period from 2002 to 2020. This dataset
provides annual information on the source state, destination countries, weight, and
value of exports for commodities categorized by HS4 and HS6 codes.14 Total export
weight is calculated by aggregating vessel and air weights. I limit my analysis to
scrap commodities at both HS4 and HS6 levels that have been directly affected by
China’s waste ban. Additionally, I use trade data from U.N. Comtrade for country-
level exports of recyclable waste. To examine trade in recyclable waste between
China and other countries besides the U.S., I gather data for 11 countries that have
regularly traded in substantial quantities of recyclable waste materials.
Emissions Data. To examine the correlation between trade in recyclable
wastes and pollution emissions, I rely on data from the EPA Inventory of U.S.
Greenhouse Gas Emissions and Sinks. The data covers state-level greenhouse
gas (GHG) emissions from the waste industry annually from 2002 to 2020. It is
important to note that the dataset estimates state-level emissions prior to 2010
by using national-level emissions weighted by the proportion of waste sent to
landfills by each state. For the years 1990-2009, the methodology utilized the state
percentage of waste going to landfills reported by landfills under subpart H.H. of
the Greenhouse Gas Reporting Program (GHGRP) to calculate national methane
(CH4) emissions estimates. However, the U.S. EPA improved the accuracy of state-
level methane emission calculations after 2010 by aggregating emissions reported
14HS codes are standard industry classifications used for goods exports, and HS6 codes are more
detailed than HS4 codes with six and four digits, respectively.
14
by individual waste industry facilities. To account for this measurement change, all
estimating specifications include time fixed effects.
To examine the relationship between trade in recyclable waste and pollution
emissions in more detail, I utilized methane emissions data from individual
landfill facilities reported under the U.S. EPA Greenhouse Gas Reporting Program
(GHGRP) from 2010 to 2020. This program requires large greenhouse gas emission
sources, fuel and industrial gas suppliers, and CO2 injection sites in the United
States to report their greenhouse gas data annually. The GHGRP includes various
industries, such as power plants, petroleum and natural gas systems, minerals,
chemicals, pulp and paper, refineries, and, crucially for my analysis, the waste
industry. Approximately 8,000 facilities are required to report their emissions each
year, and the GHGRP has a high compliance rate due to the absence of a quantity-
based penalty for emissions from landfill owners, and because non-compliant owners
are issued warning notices by the U.S. EPA. Appendix Figure A3 displays the
distribution of landfill facilities in the U.S. according to the EPA GHGRP. In my
analysis of statistical identification, I also used data from various other industries in
the GHGRP dataset, not including the waste industry, as control industries.
In this paper, Table 1 presents a concise overview of all the data sources
used. The primary datasets constructed for the three main analyses are as follows:
(1) annual state-level panel data from 2010 to 2020 for U.S. GHG emissions by
industry; (2) annual state-level panel data from 2002 to 2020 for GHG emissions
from the waste industry and recyclable exports.15
15Both export and emission data are at the state level since there is no smaller geographic unit
for exports that can be easily matched to emissions. Although export data exist at the departure-
port level, it is challenging to track whether the waste arriving at each port originates from within
the state or other states, making it difficult to establish a correlation between port-level exports
and local emissions.
15
2.2.3 Summary Statistics. The summary statistics for the export
of recyclable waste by the U.S. and other major waste-exporting countries are
presented in Table 2. In panel A, columns 1 and 2 display the total value of U.S.
recyclable waste exports to China and the rest of the world between 2010 and 2020,
while columns 3 and 4 show the corresponding values for 11 other countries that
regularly trade recyclable waste with China, including Australia, Austria, Canada,
France, Germany, Portugal, New Zealand, the United Kingdom, Japan, Spain, and
Finland. Panel B shows the total weight of recyclable waste exports from the U.S.,
and from these 11 other countries, to China and to the rest of the world.
Before 2017, U.S. recyclable waste exports to China accounted for over half
of the total value of U.S. recyclable waste exports to non-U.S. countries, and over
70% of total U.S. recyclable waste exports by weight. However, following China’s
2017 waste ban, U.S. recyclable waste exports to China decreased significantly,
while exports to the rest of the world initially increased but then eventually
decreased as well.
Further information about the data can be found in Appendix Tables A.1
through A.8. Appendix Table A1 provides more detailed information about the
composition of recyclable waste exports by value and weight, showing that from
2002 to 2020, the U.S. exported mixed paper/paperboard and plastic scrap with a
total value of 31,521 million USD and 15,464 million USD, which account for 66%
and 32% of the total value of recyclable waste exports. The next most-exported
recyclables, in order, were metal, fibers, cotton, iron/steel, and wool scrap.
Details regarding the composition of greenhouse gas (GHG) emissions by
industry in the U.S. can be found in Appendix Table A2. The waste, metal, and
refinery industries primarily emit methane (CH4) as their main GHG, while the
16
power plants, minerals, chemicals, and petroleum and natural gas industries emit
more carbon dioxide (CO2) than methane. The pulp and paper industry emits
approximately equal amounts of methane and carbon dioxide, while nitrogen
dioxide (NO2) emissions are relatively low across all industries, at least in
comparison to methane and CO2 emissions.
Appendix Table A3 displays changes in the overall emissions levels for the
waste industry over the years, as well as the varying numbers of facilities from
2010 to 2020. The dataset used excludes small facilities, focusing only on large
emitters in the EPA GHGRP. While total GHG emissions from the waste industry
have increased after 2017, the total number of facilities has gradually decreased,
indicating that the average emissions per facility have risen over time. On the other
hand, other industries such as power plants, metals, pulp and paper, and refineries
have experienced a reduction in their GHG emissions between 2010 and 2019.
Appendix Table A4 documents the overall GHG emissions from various industries
in the U.S., with a focus on the nine main industries out of the 72 sectors listed in
the GHGRP. This data reveals that the waste industry’s emissions have increased
both in total and on average after China’s GS policy.
2.3 The impact of China’s GS policy on Emissions in the U.S.
2.3.1 Raw Trends.
Export Trends. Figure 1 displays the trends in exports of recyclable waste
from the U.S. to China and the rest of the world. After China’s accession to
the WTO in 2001, the value of recyclable waste exports to China increased and
remained stable between 2013 and 2016. However, when China introduced its GS
policy in 2017, the value of these exports plummeted. Meanwhile, the value of
exports to the rest of the world remained stable from 2002 to 2016, but increased
17
temporarily after the waste ban diverted China’s former recyclable waste inflows
to other countries. Nevertheless, this increase was short-lived due to similar policy
changes in those other countries. The trend in the net weight of recyclable waste
exports shows an even more significant and direct impact of China’s GS policy
on U.S. exports to China and the rest of the world after 2017. These trends
demonstrate that the waste ban has decreased recyclable waste exports from the
U.S. to China and has temporarily increased exports to the rest of the world.
However, this increase lasted for only a year, and thus many recyclable wastes that
were previously exported overseas are now being processed within the U.S.
Figure 2 illustrates the composition of total recyclable waste exports from
the U.S. Paper/paperboard and plastic scrap are the most-exported materials,
accounting for about 76% and 22% of total exports by value, respectively, and
90% and 10% of total exports by weight, respectively. Figure 3 displays the trends
in relative values of plastic scrap exports from the U.S. and six other OECD
countries—Canada, France, Germany, Japan, the Netherlands, and the United
Kingdom—over time. After China’s waste ban, exports of plastic scrap to China by
the U.S. and the six other OECD countries declined by about 99 percent compared
to their 2010 trade values. Nevertheless, the U.S. plastic manufacturing industry’s
GDP has gradually increased over time. The exports of U.S. plastic scrap to the
rest of the world were stable from 2010 to 2016. However, after China’s GS policy,
these flows increased temporarily but then decreased again. Similar patterns
emerged in the plastic scrap exports of the other six OECD countries to the rest
of the world.
Emissions Trends. Trends in total greenhouse gas (GHG) emissions from the
U.S. waste industry are displayed in Figure 4 using data from the EPA GHGRP
18
from 2010 to 2020. Prior to 2017, total GHG emissions from the waste industry
had decreased over time. However, after China’s GS policy was implemented in
2017, the trend in total GHG emissions from the waste industry reversed and
began to increase. Methane emissions account for over 80% of the waste industry’s
total GHG emissions, followed by carbon dioxide and nitrous oxide. Although
total emissions have increased since 2017, the number of waste industry facilities
has gradually decreased over the past few decades, indicating that the average
emissions per facility have also increased over time.
In Figure 5, the total GHG emissions from the waste industry and several
other U.S. manufacturing industries are shown from 2010 to 2020. The raw trends
indicate that most industries have seen decreasing GHG emissions over the years,
including power plants, metals, pulp and paper, and refineries. However, a few
industries, such as chemicals, minerals, and petroleum and natural gas, have seen
increasing GHG emissions. Comparing the emissions changes of all industries with
the timing of China’s waste ban, it appears that the changes in GHG emissions for
industries other than the waste industry are not influenced by the waste ban.
2.3.2 State-level Emissions.
Synthetic Control Method. To identify the effect of China’s GS policy
on state-level methane emissions in the U.S., I use the synthetic control method.
This strategy relies on exogenous variation in methane emissions across all other
industries in the EPA GHGRP (such as power plants, petroleum and natural
gas systems, minerals, chemicals, refineries, etc.). Appendix Table A2 shows the
average emissions from industries across states by type of GHG. This identification
strategy takes advantage of the fact that other industries in the GHGRP, which
19
also emit GHGs, were negligibly affected by China’s waste ban. To exclude the
possibility that the exports of these other industries may have changed in the year
of 2017 (when the U.S. started its “trade war” with China) and thus contaminate
the control industries for emissions, I plot the exports (by weight/kg) of control
industries in Appendix Figure A5. The figure shows that the exports of the control
industries, such as oil and gas, minerals, and chemicals, did not discernibly shift as
of 2017.16
Given that waste industries from all states in the U.S. are affected to
varying degrees by China’s GS policy, data from the waste industry in other U.S.
states (shown in Appendix Figure A4b) cannot be assumed to represent an entirely
uncontaminated control pool for the waste industry in any given U.S. state of
interest. Appendix Figure A4a plots the time trends for methane emissions from
all industries (including the waste industry, in blue) using California as an example.
As this figure suggests, simply using the other industries in the same state likewise
may not provide a suitable control pool. Thus, I use all other industries from
all states to greatly enlarge my control pool. For each U.S. state, separately, I
use these other state-level industries as my control pool for the recyclable waste
industry in the state in question. For my synthetic control approach, I fit a
separate synthetic pre-policy trend that is as close as possible to the actual pre-
policy trend for each state’s landfill methane emissions (Abadie, Diamond, &
Hainmueller, 2012). The “model training” process seeks to minimize the prediction
error over the period prior to China’s GS policy:
16Exports by industry can be obtained from the U.S.A Trade Online data by Standard
International Trade Classification (SITC) code.
20
∑J ∑50
YˆN11t = wjsYjst (2.1)
j=2 s=2
Where YˆN11t is the emissions from the waste industry for a given state (i.e.
the industry “treated” by China’s GS policy, indexed as industry j = 1) that
would have been expected in the absence of China’s GS policy, t is the year of these
emissions, j = 2, ..., J is a collection of untreated industries not affected by China’s
GS policy, and s = 1, ..., 50 are all states in the U.S.17 Yjst is observed emissions
from the untreated control industries from all states. The synthetic control is
defined as a weighted average across state-industry pairs in the “donor pool” of
untreated controls. The weights on the emissions of industry j in state s are wjs.
18
I use the trained model based on data for the pre-policy period to predict
post-policy-date synthetic emissions in the absence of the GS policy for the waste
industry in a given state. The difference between the synthetic post-policy landfill
emissions trend and the actual landfill emissions trend, τ̂1t, is the estimated causal
effect of China’s waste ban on U.S. state-level methane emissions from landfills:
τ̂1t = Y11t − YˆN11t (2.2)
I use the same process, separately, for the waste industry in each of the 50
U.S. states (excluding Washington DC) and calculate the estimated causal effects of
China’s GS policy on methane emissions for each state.
Figure 6 displays emissions from 2010 to 2020 from the waste industry for
four selected states—California, Virginia, Texas, and New York—compared to
17Alaska is excluded from this analysis.
18For example, in the synthetic control for California, the donor pool includes state-industry
pairs such as California oil and gas industry, Indiana mining manufacturing industry, etc.
21
their synthetic-control counterparts. The synthetic emissions for each state track
very closely with the trajectory of actual emissions for the pre-GS policy period.
This suggests that the synthetic trend for each state likely provides a reasonable
approximation to the amount of methane that would have been emitted in each
state from 2018 to 2020 in the absence of China’s policy. Figure 7 suggests that
China’s waste ban has had a discernible effect on methane emissions from the waste
industry in these four states, and these effects have increased over time.
Placebo Tests. To evaluate the robustness of my results and calculate effective
p-values for my estimates, I run placebo tests by applying the synthetic control
method to all state-industry pairs that were not affected by China’s GS policy
during the sample period of my study. If the placebo study shows that the marked
change estimated for California’s waste industry, for example, is unusually large
relative to the emission changes for other state-industries that were not affected
by China’s waste ban, then my analysis can be assumed to provide statistically
significant evidence that the waste ban causally increased domestic methane
emissions from the waste industry in California.
Figure 8 shows my placebo test results for the four example states. The
bright blue lines reveal causal estimates of the effects of China’s waste ban on
methane emissions in California, Virginia, Texas, and New York. The muted grey
lines are the analogous causal estimates of the GS policy for other (non-waste)
state-industry combinations that can be assumed not to be affected by the GS
policy. The plots in Figure 8 show that the synthetic control estimates for the four
states are above the 90th percentile of all placebo estimates, which proxies for a
test of the statistical significance of the causal estimates for these four states.
22
I then apply the same placebo-test strategy to all other states in the U.S.
Appendix Table A5 shows the causal estimates of China’s GS policy on state-level
methane emissions and the implied p-values calculated from the placebo tests for
each state. Figure 9 shows the causal estimates of the GS policy on GHG emissions
from the waste industry by state. States such as Nevada, Montana, Virginia, and
New York show statistically significant percentage increases in GHG emissions
after China’s policy. Figure 9 also shows the net change in state-level emissions
from landfills for each state in the U.S. after the GS policy. Larger states, such as
California, New York, Texas, and Virginia have seen the largest absolute increases
in methane emissions from landfills after the waste ban.
2.3.3 Emission Changes and Waste Export Exposure.
To further explore potential factors that may correlate with the
heterogeneous effects of China’s GS policy on the U.S., I plot emission changes
for each state against both total (historical) recyclable waste exports, and the
percentage of paper as a share of waste exports. Autor, Dorn, and Hanson (2013)
find that the regions (community zones) which have a higher trade exposure with
China in manufacturing industries experienced larger decreases in manufacturing
employment in the U.S. from 1995 to 2010. In the present paper, I explore whether
a state with higher trade exposure with China, specifically in terms of recyclable
wastes, has lower domestic emissions before the GS policy but increased emissions
after the GS policy is implemented. Given that mixed paper/paperboard account
for more than 80% of exported recyclable wastes and that these materials generate
methane in landfills, I also explore whether states with a higher percentage of
waste paper exports experience greater increases in methane emissions after the
GS policy. Historical recyclable waste exports and the percentage of paper exports
23
are calculated using annual trade data from U.S.A. Trade Online. Figure 10 shows
that the increase in emissions due to the waste ban is positively correlated with
historical recyclable waste exports. The more of its recyclable wastes a state
exported overseas before the GS policy, the greater the increase in emissions it
experienced after the GS policy. Given that waste paper/paperboard is the largest
contributor to methane emissions from landfills among all recyclable wastes, I also
try to link the heterogeneous effects by states with the percentage of recyclable
paper/paperboard waste a state exported. However, I find the relatively weak
correlation shown in Figure 11. The GS policy had an effect on how much a state
exported, and not how much of the export was paper/paperboard.
The next step is to verify statistically the apparent finding that the greater
a state’s exposure to trade in recyclable wastes with China before the GS policy,
the greater the increase in emissions the state experiences after the GS policy. I
plot the raw trends of waste exports and domestic emissions from 2002 to 2019.
Figure 12 shows that waste exports are negatively correlated with domestic
emissions from the waste industry. Before China’s waste ban, the more recyclable
wastes the U.S. exported, the less methane emitted domestically in the U.S. waste
industry. After China’s waste ban, waste exports decreased dramatically and
domestically methane emissions increased. To identify the general causal effect of
recyclable waste exports on domestic emissions from the waste industry, I start
with a simple OLS regression in levels as follows:
Methaneit = α0 + αt+ β0Exportit + νi + ηit+ µt + eit (2.3)
24
Methaneit is the level of emissions in state i in year t. Exportit is the level of
recyclable exports from state i to China in year t. I then take the first difference
of emissions and exports:
∆Methaneit =Methanei,t −Methanei,t−1
(2.4)
∆Exportit =Exporti,t − Exporti,t−1
After replacing equation (2.4) with equation (2.3), and taking the difference
(subtracting the values in the previous year t − 1 from the values in year t), I get
the first difference model:
M︸ ethanei,t −︷︷Methanei,t−︸1 = α︸ 0 ︷−︷α︸0+ ︸α(t−︷(︷t− 1)︸)+β0 ︸(Exporti,t −︷︷Exporti,t−1︸)
∆Methane 0it α ∆Exportit
+ ︸νi ︷−︷ ν︸i +µ︸ t −︷︷µt−︸1+ ︸ϵi,t −︷︷ei,t−︸1
0 for each i ∆µt ∆ϵit
(2.5)
∆Methaneit = a+ β∆Exportit + ut + eit (2.6)
where ∆Methaneit is the change in methane emissions in the waste industry
for state i in year t, compared to the previous year. ∆Exportit is the change in
recyclable waste exports to China from state i in year t, compared to the previous
year. In differencing, any constant term in the model in levels drops out. The
α in equation (2.6) is therefore the linear time trend in the levels data. Year
fixed effects ut control every time pattern other than the linear time trend. State
fixed effect νi drops out for each state i after taking the first difference. I choose
the first difference model in equation (2.6) over the fixed effect model in levels
for the following reasons: (1) although the first difference model works just as
25
a fixed effects model, it relies on no serial correlation in the differenced errors
(a weaker assumption); and (2) the first difference model controls for variations
in differences across states, which is a stronger control.19 There are still several
concerns regarding first difference OLS identification in this context:
1. On the one hand, GHG emissions from most industries are monitored by the
U.S. EPA. For the waste industry, it is difficult for landfills to get permits
because they need to meet many environmental requirements. Given that
the U.S. has relatively stringent environmental regulations on local pollution
(such as the Clean Air and Clean Water Acts), it can be harder for recyclers
to find facilities to process recyclable materials. As a result, emissions from
the waste industry may be inversely related to the amount of recyclables
being exported to China, reflecting the stringent domestic environmental
regulation in the U.S.
2. Omitted variables (e.g. economic development) may increase both recyclable
exports and domestic emissions. Thus there is a potential problem of
endogeneity for the variable that measures the change of recyclable waste
exports.
3. Instead of the observed demand policy shock from China (the waste ban),
technological development—such as an increasing ability to reprocess
recyclables cleanly and safely—could also decrease the supply of recyclable
wastes to be exported.
To identify the causal effects of changes in U.S. recyclable waste exports on
U.S. domestic methane emissions from the waste industry, I employ an instrumental
19The fixed effects model builds on the assumption of no serial correlation prior to demeaning.
26
variable that accounts for the potential endogeneity of U.S. recyclable exports. I
use a Bartik shift-share instrument from the literature on international trade and
labor economics, adapted to an environmental context (Bartik, 1991; Wong, 2021).
The main IV is defined as follows: ∑{ }
IV Bartik
Eijt0
it = ∆Exportucjt (2.7)E
j jt0
where ∆Exportucjt is the change in exports from the U.S. (u) to China (c)
E
for recyclable waste of type j, in year t compared to the previous year. ijt0 is
Ejt0
state i ’s share of exports to China for recyclable waste j in the initial year t0.
The product of the initial share term and the current change in exports is then
summed across all recyclable wastes j. In other words, the shift-share instrument
is a data-regenerating process that shifts the initial export share of each state
to the trajectory of the change in total recyclable waste exports from the U.S.
to China over time. The initial year I use is 2004, the earliest year for which
complete data is available for recyclable waste exports from the U.S. to China
for recyclable materials affected by China’s policy. Given that the construction
of IV Bartikit excludes state i’s current-period recyclable waste exports to China, the
initial distribution of export shares of state i for waste j is exogenous (or at least
predetermined), relative to the subsequent changes in methane emissions for state i.
One concern for my upcoming estimation is that changes in recyclable waste
exports from the U.S. may be correlated with U.S. technological improvements and
thus the supply of recyclable materials. In that way, U.S. recyclable exports may
be jointly endogenous with domestic emissions. Naive OLS estimates may overstate
the true impact on domestic methane emissions of restrictions on recyclable waste
exports from the U.S. to China. I thus employ a second alternative instrument that
27
accounts for this potential endogenei∑ty a{s follows: }
IV Bartik
Eijt0
it,others = ∆Exportocjt (2.8)E
j jt0
Instead of using the change in exports from the U.S. to China, I exclude the
U.S. and instead construct the IV using the contemporaneous changes in export
of recyclable wastes to China from 11 other developed countries. Specifically, I
instrument for the measured change in U.S. exports of recyclable wastes with
a non-U.S. analog ∆Exportocjt, where the o subscript, for “others” replaces
the u subscript for “U.S.” This variable is constructed using data for changes
in recyclable exports (at the commodity level) from the 11 other high-income
countries to China. These 11 countries are all OECD countries which have engaged
in extensive trade in recyclable wastes with China during the past few decades.20
This instrument is valid because: (1) the instrument constructed by other OECD
countries is highly correlated with the U.S. waste exports (relevance); (2) this
instrument affects the U.S. methane emissions only through the U.S. waste exports
(exclusion restriction); and (3) this instrument does not have a direct effect on the
U.S. methane emissions (exogeneity).
After constructing the main and alternative Bartik-type instruments, I fit
models of the following form:
First stage: ∆Êxportit = α + β∆IV Bartikit + µt + vit (2.9)
Where IV Bartikit is either the U.S. or the other-country version of the
instrument. Then the fitted value for the first stage is employed in the second
stage:
20The 11 selected countries are: Australia, Austria, Canada, France, Germany, Portugal, New
Zealand, United Kingdom, Japan, Spain, and Finland.
28
Second stage: ∆Methaneit = α + β∆Êxportit + µt + eit (2.10)
In the second-stage equation, ∆Methaneit is the annual change in methane
emissions from the waste industry in state i.21 The key coefficient of interest is
β, the average annual change in methane emissions across U.S. states caused by a
one-unit change in U.S. recyclable waste exports to China. I then use this average
estimate to calculate the cumulative aggregate impact of China’s GS policy on U.S.
methane emissions from the waste industry for 2016 through 2019. The calculation
is as follows:
∑2019 [ ∑I ]
∆Mêthanetotal = β ∆Exportit (2.11)
t=2016 state=i
I begin my analysis by estimating the naive OLS specification in equation
(2.1). The coefficient β is interpreted as the change in methane emissions from
the waste industry for a 1 metric ton change in recyclable waste exports. For the
simple OLS specification, model 1 in Table 3 suggests that for every 1 metric ton
reduction in recyclable waste exports, methane emissions from the waste industry
in the U.S. increase by 0.49 metric tons per year. To address concerns about
potential reverse causality and/or endogeneity, I then estimate 2SLS equations
(4) and (5) using my basic Bartik shift-share instrument. Model 2 in Table 3
suggests that for every one metric ton reduction in recyclable waste exports,
methane emissions increase by 0.722 metric tons of CO2 equivalent per year. This
shows that the estimation bias from reverse causality tends to attenuate the point
estimate of interest toward zero. The logic here is that the point estimate shows
21I use a first-differenced estimator to address the problem of omitted variables that may bias
the estimate.
29
that the more recyclable waste was exported from the country, the less methane
pollution was emitted domestically. However, the reverse causality is that the
higher are domestic methane emissions, the more new recyclable waste is likely
to be exported due to the stringent environmental regulations of the U.S., such
as emissions caps. Thus, without accounting for this reverse causality, the OLS
specification tends to underestimate the true negative effect of waste exports on
domestic emissions. However, there is still a possibility that these may have been
nontrivial supply shocks in the recyclable waste industry instead of just China’s
GS policy, which could also bias these estimates. Thus I estimate equations (2.3)
and (2.4) again, but this time I use my alternative Bartik shift-share instrument
constructed with the recyclable waste exports from a set of 11 non-U.S. countries
to China. Model 3 in Table 3 suggests that for every 1 metric ton reduction in
U.S. exports of recyclable wastes to China, methane emissions increased by 0.893
metric tons of CO2 equivalents per year. This even-larger estimate reinforces my
finding that the decrease in U.S. recyclable waste exports to China has increased
the U.S. domestic methane emissions in general. This could be due to U.S. supply-
side shocks such as technology improvement. Technology that processes wastes
more efficiently could decrease both exports of waste and methane emissions. Thus,
without accounting for such supply-side shocks, the effect of “exposure” to waste
exports to China and domestic emissions could be underestimated.
There are still several possible threats to identification in my strategy. One
is that supply-side shocks like technology improvement may be correlated across
high-income countries. In this event, my IV estimates may be contaminated by
correlations between changes in waste exports and unobserved components of
export supply, making the impact of waste export exposure on domestic emissions
30
appear smaller than it truly is. After estimating β—the average effect across all
U.S. states annually, I then calculate the implied cumulative effect of (a) the total
reduction in recyclable waste exports due to China’s waste ban on (b) overall U.S.
national increases in methane emissions from 2016 to 2019.22 Over this time period,
the total weight of recyclable waste exports decreased from nearly 18,000,000
metric tons to just 5,500,000 metric tons. Applying equation (2.6), the U.S. total
methane emissions from the waste industry increased by approximately 8-11 million
metric tons of CO2 equivalents., which (for comparison) is about 9.68% of total
methane emissions from the U.S. waste industry in 2016, or about 5.23% of total
methane emissions from the U.S. petroleum and natural gas system in 2016.23
2.4 Conclusion
This paper examines the effects of China’s Green Sword Policy (a
comprehensive waste import ban) on environmental outcomes in the U.S. The
waste ban, designed to reduce China’s waste imports from the U.S., has resulted in
far-reaching consequences for the U.S. environment at the national and state levels.
Following the implementation of China’s waste import ban, policy-makers have
been concerned about its effects on the U.S. recycling industry, but few researchers
have studied these effects in a quantitatively rigorous fashion. My paper is the first
empirical analysis of the impact of China’s GS policy on the U.S. I use methane
emissions as an available and consistent measure of pollution outcomes from wastes,
and find that the U.S. has seen a significant increase in landfill-related methane
emissions after the GS policy, especially in larger states such as California, New
York, Virginia, and Texas. Furthermore, the observed heterogeneous changes in
22I exclude 2020 exports and emissions because of the COVID disruption in international trade
and domestic emissions.
23U.S. EPA: 1990-2020 National-level U.S. Greenhouse Gas Inventory Fast Facts
31
state-level methane emissions have been positively correlated with the amount
of waste previously exported to China by each state. This positive correlation
suggests a potential causal relationship between waste exports and domestic
emissions. In my analysis of U.S. waste export and emissions, I find that recyclable
waste exports are inversely related to domestic methane emissions in the U.S. This
result is consistent with the “pollution haven” hypothesis, namely that developed
countries tend to relocate their wastes and associated negative externalities to
developing countries in order to reduce domestic pollution levels.
Other questions for further study include what kind of spillover effects the
GS policy might have had on international policies regarding pollution. Despite
there being many new destination countries for U.S. recyclable wastes in Southeast
Asia, Africa, and the Middle East, only some countries reacted to the GS policy by
adjusting their domestic regulations to control the altered flow of waste imports,
while others did not have a prompt change in policy to control the inflow of wastes.
Another point of investigation would be waste transfers across states: my analysis
finds that among 11 states that have experienced statistically significant increases,
some are smaller states such as Nevada, North Dakota, Alabama, Kentucky, and
New Hampshire. It is difficult to connect their significant increases in methane
emissions with their own waste generation. It is worth noting that these relatively
small states might have experienced significant methane emission increases because
they are often neighbors of larger states. There is the possibility that larger states
have transferred their newly surplus amounts of recyclable wastes to neighboring
states for processing and disposal. Additionally, within the states that have seen
significant increase in emissions from wastes, it would be worth exploring which
local communities are bearing the burden of the increased amounts of waste due to
32
China’s waste ban and what are the demographics of these communities. Finally,
a caveat concerning my research is that the methane emission data I use is from
a self reporting database. Although the EPA has emphasized the high reporting
compliance rate, it would be more accurate to test the effect of the policy using
some type of ground-monitored emissions data or satellite data to verify the
significance of the policy effect. Another potential caveat lies in the identification
strategy I employed. I assume that industries not experiencing emission changes
during the implementation of China’s GS policy were unaffected by this policy.
However, this assumption might overlook potential general equilibrium effects
that could influence the outcomes. The GS policy’s breadth and impact could
potentially alter the overall demand and supply dynamics for numerous goods,
thereby indirectly affecting units in the control group.
33
2.5 Figures
Figure 1. U.S. Recyclable Waste Exports to China and the Rest Of the World
(ROW). This figure shows that U.S. recyclable waste exports to China dropped
dramatically after 2017 when China’s Green Sword (GS) policy was announced and
implemented. Meanwhile, U.S. exports in recyclable wastes to the rest of the world
increased temporarily but decreased after 2018.
34
Figure 2. Composition of U.S. Recyclable Waste Exports to China. This figure
shows the composition of recyclable wastes that were exported in the past two
decades. The listed waste materials are all wastes that are directly affected by
China’s GS policy. The left panel shows the types and percentages of waste
materials exported by value ($ USD). The right panel shows the types and
percentages of waste materials exported by weight (kg). Paper and paperboard
and plastic scraps are the most exported recyclable wastes by value and weight,
followed by metal, iron/steel, fibre, cotton, etc.
35
Figure 3. Plastic Waste Exports from the U.S. and Other OECD Countries to
China. This figure shows recyclable waste exports (taking plastic scrap as an
example) for the U.S. (red line), as well as for a set of six other OECD countries
(grey lines) to China and to the rest of the world from 2010 to 2020. All of the
export values are normalized by the 2010 export values for each country. The blue
line represents the U.S. manufacturing GDP—NAICS 31-33 (the industry code)
that includes the plastics industry. Although the plastic-manufacturing GDP of the
U.S. increased gradually over time, the U.S. plastic scrap exports to China dropped
by almost 99 percent, especially after China’s GS policy. Similar patterns are found
in other OECD countries. After China’s GS policy, plastic scrap exports by the
U.S. and by the set of six other OECD countries increased temporarily to the rest
of the world but then decreased.
36
Figure 4. U.S. Waste Industry GHG Emissions. Total emissions from the waste
industry based on the aggregated reporting records from facilities for each year
between 2010 and 2020. In the waste industry, total emissions are from methane
(CH4), carbon dioxide (CO2), and Nitrous oxide (N2O). However, the amount
of CO2 and N2O are too small compared to CH4. Thus, more than 80% of total
emissions from the waste industry are CH4. Although the number of facilities has
decreased gradually over the years, the total emissions and methane emissions of
facilities have increased since 2017.
37
Figure 5. U.S. Greenhouse Gas Emissions (log.MMT) by Industry. This figure
shows the GHG emissions from the eight main industries in the EPA GHGRP
data. Waste, power plants, and petroleum and natural gas are the industries that
have the highest emissions in the U.S. on average from 2010 to 2020. The waste
industry (in red) has seen a decrease in methane emissions from 2010 to 2017 and
an increase in methane emissions afterwards, both in total and on average. Changes
in GHG emissions of other industries are exogenous to China’s GS policy.
38
Actual Methame Emissions (solid) vs. Synthetic Methane Emissions (dashed)
Figure 6. State-level Synthetic Control Results. This plot shows the synthetic
control results from four selected states. The solid grey lines are the actual
methane emissions from the waste industry in four states. The blue dashed lines
are the synthetic methane emissions estimated by a function of other state-
industries (controls) with different weights. The differences between the actual
methane emissions and synthetic emissions are the causal effects of China’s GS
policy on the U.S. domestic methane emissions from the waste industry at state
level. California, Virginia, Texas, and New York have all seen an increase in
methane emissions after China’s GS policy. Texas’s methane emissions from the
waste industry dropped in 2020. This may be caused by a variety of reasons due to
the 2020 Covid-19 pandemic.
39
Causal Estimates (Differences between Actual and Synthetic Emissions)
Figure 7. State-level Synthetic Control Results: Four States. The causal estimates
are calculated by subtracting the synthetic emissions from the actual emissions.
Since the synthetic emissions are predicted in the absence of China’s GS policy,
the difference between actual and synthetic emissions is the causal effect of the GS
policy on emissions from the waste industry for each state. All of the four example
states have seen increasingly positive effects on methane emissions from the waste
industry.
40
Waste Industry (Blue) vs. Other Industries (Grey)
Figure 8. State-level Synthetic Control Results: Placebo Tests. This figure shows
results from placebo tests associated with the synthetic control method for the
four states. The blue lines are the causal effects of China’s GS policy on the waste
industry of each state. The grey lines are the causal effects of China’s GS policy on
other industries of different states. The p-value is calculated by the distribution of
the post/pre-GS policy ratios of the MSPE for treatment industry of a state and all
other control state-industry pairs.
41
42
Figure 9. State-level Synthetic Control Results—Estimates of the Percentage and Net Change in GHG Emissions from
the Waste Industry. Red-colored states represent state-level estimates that can be considered statistically significant at
the 10% level. Grey states represent state-level estimates that are statistically insignificant.
Figure 10. Pairwise Correlations: State-level Emission Changes and Waste Trade
Exposures. Methane emission increases for states are calculated by multiplying
the state’s methane emissions in 2016 by the percentage increases in emissions
estimated by the synthetic control method. The dots (observations) are the net
methane emission increases by state. The log of total recyclable waste exports
(before China’s GS policy) is used as a measurement of the “export exposure” of a
given state to China. The red line is the fitted line for a regression that regress the
net change in methane emission on the log of total recyclable exports at the state
level.
43
Figure 11. Pairwise Correlations: State-level Emission Changes and
Paper/paperboard Exports. Methane emission increases for states are calculated
by multiplying the methane emissions in 2016 with the percentage increases in
emissions estimated by the synthetic control method. The dots (observations)
are the net methane emission increases by state. The red line is the fitted line for
regression that regress the net change in methane emissions on paper/paperboard
exports at the state level.
44
Figure 12. U.S. Waste Exports and Methane Emissions from the Waste Sector.
The average emissions of the waste industry in the United States are derived from
the aggregated reports submitted by states from 2010 to 2020. Meanwhile, the
average waste exports from the industry are determined based on state-level waste
export data. The plots demonstrate an inverse correlation between waste exports
and domestic emissions from the waste industry in the U.S.
45
2.6 Tables
Spatial Unit Years available Frequency
UN Comtrade Data country level 2002-2020 yearly
U.S.A Trade Online Data state level 2002-2020 yearly
EPA GHG Inventory Data state level 2002-2020 yearly
EPA GHG Reporting Program Data facility level 2010-2020 yearly
Table 1. Data Sources Summary: State-level Analysis. This table summarises all of
the data sources that are used in this paper. Export and emission data are tracked
by state and year.
46
I. U.S. II. Other countries
Exports to Exports to Exports to Exports to
China rest of world China rest of world
(1) (2) (3) (4)
Panel A. Total value ($ U.S. billion) over all U.S. states
2010 3.54 2.85 3.30 4.05
2011 4.29 2.95 3.16 4.59
2012 3.96 2.41 3.76 4.89
2013 3.54 2.41 3.28 4.28
2014 3.57 2.51 3.13 4.52
2015 3.33 2.31 2.82 3.69
2016 3.16 2.29 2.62 3.55
2017 2.59 2.75 2.33 4.64
2018 1.53 3.49 1.36 5.14
2019 1.01 3.09 0.68 4.64
2020 0.86 2.78 0.33 4.33
Panel B. Total weight (billion kg) over all U.S. states
2010 17.30 6.59 13.74 14.01
2011 20.03 6.58 14.11 15.58
2012 19.88 6.15 15.33 16.09
2013 18.39 6.05 14.03 12.85
2014 18.65 6.56 13.64 16.49
2015 19.04 6.50 14.24 9.39
2016 17.99 6.69 14.26 10.67
2017 14.31 8.08 11.19 6.29
2018 7.95 12.95 6.81 9.66
2019 5.59 12.21 4.23 11.85
2020 4.71 10.38 2.19 10.78
Table 2. Summary Statistics: Recyclable Waste Exports by
the U.S. and Other Countries. “Other countries” refers to 11
selected OECD countries—Australia, Austria, Canada, France,
Germany, Portugal, New Zealand, United Kingdom, Japan,
Spain, and Finland. They all have regular trade with China in
recyclable wastes.
47
Dependent Variable: Change in Methane Emissions
2SLS 2SLS
Naive
Bartik shift-share Bartik shift-share IV
OLS
IV Other countries
(1) (2) (3)
2003-2019 first differences
Change in Exports -0.492*** -0.722*** -0.893***
(0.122) (0.114) (0.124)
2SLS first stage estimates: Change in Exports regressed on IV
IV Bartik 1.11*** 9.55***
(0.038) ((0.465)
First stage F-statistics 50 34
Year FE ✓ ✓ ✓
Observations 897 897 897
Table 3. Models to Explain the Changes in Methane Emissions as a Function of the
Changes in Recyclable Waste Exports. Each column reports a separate regression.
*p < 0.1, **p < 0.05, ***p < 0.01. The first-differenced model is like the fixed
effect model but with a less restrictive assumption. The intercept in this first-
differenced model captures all unobserved factors that may affect the emissions but
are constant over time. It also captures the linear time trend. The year-fixed effects
capture every time pattern other than the linear time trend.
48
CHAPTER III
THE EFFECTS OF CHINA’S WASTE IMPORT BAN ON LOCAL
COMMUNITIES: A CASE STUDY OF CALIFORNIA
3.1 Introduction
Each year, the United States produces over 200 million tons of solid waste,
with a substantial portion being transported beyond the borders of the originating
county or state.1 This practice is expected to continue as waste generation rates
have nearly doubled in recent years.2 However, the extra jurisdictional transfer
of waste can result in unequal exposure to waste facilities and shipments among
various demographic groups. Low-income and minority communities are often
disproportionately impacted, as these groups are more likely to live near landfills
and other polluting facilities (GAO, 1983; Chavis & Lee, 1987). In fact, people of
color are more likely to inhale contaminated air or consume water with harmful
contaminants due to their proximity to these facilities.3 This exposure can lead
to various adverse health outcomes, including respiratory diseases and cancers,
and it can also harm ecosystems, agriculture, and fisheries that are relied upon
by the affected communities. To reduce this inequality, it is important to address
these disparities and take action to minimize the impact of waste shipments on
vulnerable communities.
1The blue whale is the largest animal on Earth. 200 million tons of waste generated annually
equals 1 million blue whales.
2In 2017, 70% of hazardous waste processed in commercial facilities in Michigan originated
outside of that state, according to data from the Environmental Protection Agency. Massachusetts
is a significant exporter of materials such as wood, brick, and asphalt, for which disposal is
prohibited in their own landfills. These materials are thus transported to other states like Maine,
where they account for a significant proportion of the receiving state’s landfill.
3For example, the population of Uniontown, the second-largest city in Perry County, Alabama,
is over 90% Black, but the city permitted a large landfill to open in 2005, which eventually
accepted waste from 33 states, including toxic coal ash from one of the worst environmental
disasters in the country.
49
China’s “Green Sword” policy, which was announced in 2017, prohibited
imports into China of most recyclable waste from around the world, including the
United States. The impact of this policy on local waste transfers in the U.S. has
been significant. Prior to the policy, the U.S. was the largest exporter of solid waste
to China. However, after the policy was enacted, U.S. solid waste exports to China
decreased by almost 90%. As a result, many waste products that were previously
exported by the U.S. are now being processed domestically, putting greater pressure
on the country’s land capacity for waste management. This has led to an increased
need to transfer waste from densely populated areas to other locations, and haulers
and municipalities are facing new decisions about where to send waste and whether
to expand existing landfills or to open new landfill facilities.
In this paper, I describe a case study using the west coast U.S. state of
California to investigate the distributional impact of China’s waste ban on waste
transfers at the local community level. California is selected as a case study
because it is the largest state that exported recyclable waste and has comprehensive
data at the facility level, allowing a more rigorous analysis. Prior to the GS policy,
waste from California tended to be relocated to pollution havens like China, but
after the policy, new local or original pollution havens emerged as destinations for
excess recyclable waste. To study the distributional effects of the GS policy on
local communities, I utilize the CalRecycle Recycling and Disposal Reporting System
(RDRS), which contains detailed waste transfer records from origin jurisdictions to
destination facilities in California from 2002 to 2020. This dataset includes over 400
origin jurisdictions and nearly 150 destination facilities, with almost 800,000 tons of
waste transferred across local communities in California over the past 20 years.
50
To examine the change in patterns on waste relocation, I use China’s
waste ban as a natural experiment. I collecte local characteristics of destination
communities, such as racial composition, median income, economies of scale, and
political vote shares, and compare how these characteristics affecte waste transfers
across communities before and after the GS policy. The results indicate that before
the waste ban, communities with higher minority population shares, higher median
income, greater economies of scale, and higher Republican registration shares tend
to receive more waste transfers from other jurisdictions. However, after the waste
ban, communities with higher White population shares, lower median income, fewer
economies of scale, and lower Republican registration shares experience a greater
relative increase in waste inflows. These findings suggest that, perhaps counter-
intuitively, racial disparities in arriving waste transfers appears to be narrowing
after the exogenous GS policy shock.
I then utilize a simple theoretical model to investigate the mechanisms
behind the narrowing of racial disparities related to waste transfers after China’s
policy shock. I propose several cost metrics to explain the phenomenon. In my
model, the amount of waste pollution received by the destination community is
negatively associated with land costs, transportation costs, and political costs for
the destination community. To proxy for the costs of land and transportation, I
use population density and distance between origin and destination. For political
costs, I use the absolute difference between the Republican voting registration
share of the destination community and the Republican voting registration share
of the county where the destination facility is located. Communities with lower
political costs are likely to have less political voice and thus lower resistance to
waste pollution inflows.
51
To investigate which mechanisms have led to an increase in the relocation of
waste to lower-income White communities, I use a simple OLS model with policy
dummy interactions. The results show that before the GS policy, destination
communities located within a shorter distance and with lower political costs
receive more waste pollution transfers. However, after the exogenous GS policy
shock, communities with lower land costs appear to experience relatively greater
increases in waste inflows, and political costs associated with these flows seem to
have become lower. As a result, rural White communities with lower land costs
appear to have been relatively more likely to experience increased waste transfers
after China’s GS policy.
My paper makes several important contributions to the literature. First,
I contribute to the growing literature on racial inequities in pollution exposure,
specifically with respect to waste transfer. Previous studies have examined
residential proximity to industrial facilities and found that racial and ethnic
minority groups and/or lower socioeconomic status groups tend to live closer
to these facilities compared to other groups, and this trend persists over time.
Examples of such studies include research by Abel and White (2011) for Seattle
from 1990 to 2007, research by Hipp and Lakon (2010) for southern California
from 1990 to 2000, and research by Pais, Crowder, and Downey (2013) using
a national cohort sample from 1990 to 2007. In contrast to these studies, my
research examines how waste is relocated to different landfill facilities, taking
into account whether these facilities are situated in areas with higher minority
populations. Specifically, I investigate whether patterns in waste transfers,
conditioned on the location of the destination facilities, is systematically related to
the socioeconomic and demographic characteristics of the communities where these
52
destination facilities are located. To address challenges associated with drawing
causal inferences from this literature, such as the ecological fallacy that can affect
analysis using aggregate data, I define destination communities as those within a
3km radius of each destination facilities, which corresponds to only a few blocks of
neighborhoods.4
Second, my paper describes a novel examination of the impact of China’s
GS policy as a natural experiment for assessing racial disparity in waste transfers.
Prior research has demonstrated that minority communities in the United States
are disproportionately exposed to hazardous waste and pollution, and the affected
groups are less able to relocate to avoid such exposure (S. H. Banzhaf & Walsh,
1994; Baden & Coursey, 2002; S. H. Banzhaf & Walsh, 2008; Depro, Timmins,
& O’Neil, 2015; S. Banzhaf, Ma, & Timmins, 2019). Moreover, remediation of
contaminated sites may primarily benefit home-owning households and wealthier
households who subsequently move into the area and bring about gentrification
(Cameron & McConnaha, 2006; Depro, Timmins, & O’Neil, 2011; S. H. Banzhaf &
Walsh, 2013). In this study, I explore how China’s GS policy affects racial disparity
in waste transfers based on existing environmental disparities in the United States.
I first confirm that, prior to the policy’s implementation, minority communities
experienced greater waste pollution, in line with prior research. However, my
research highlights a new finding that, following the GS policy, lower-income White
communities have experienced relatively more waste pollution. This shift in racial
disparity suggests that one impact of the GS policy has been to narrow racial
disparities in waste transfers. My analysis of the underlying mechanisms suggests
4To guarantee that landfill facilities are in close proximity to residential areas, I have selected
a 3km radius, considering that most landfills in the United States span a land area of several
hundred acres.
53
that, in the wake of this policy change, land costs (rather than transportation or
political costs) played a more significant role in determining the destinations of
waste flows.
Third, my paper addresses the issue of “pollution displacement.” Previous
research has shown that pollution is displaced from the global North to South
(Copeland, Shapiro, & Taylor, 1994; Cherniwchan, 2017). Additionally, Tanaka,
Teshima, and Verhoogen (2021) examines how a 2009 tightening of U.S. air-
quality standards for lead pollution led to relocation of battery recycling facilities
to Mexico and concomitant changes in infant health in Mexico. Pollution can
also relocate from highly polluted regions to less polluted regions within the
same country (Henderson, 1996; Becker & Henderson, 2000; Greenstone, 2002;
J. S. Shapiro & Walker, 2021). This relocation can be either unintentional or
strategic. For example, Ho (2023) shows how NIMBY regulations, which restrict
waste transfers between states, can unintentionally induce waste relocation among
local communities within a state. Meanwhile, Morehouse and Rubin (2021)
demonstrate how decision-makers strategically sited coal-fired power plants at
the border of their counties to comply with the Clean Air Act, thereby exporting
emissions to neighboring counties via prevailing winds. Most of the pollution
displacement studies focus on endogenous environmental regulations within the
U.S. (Hernandez-Cortes & Meng, 2020; J. S. Shapiro & Walker, 2021). Currie,
Voorheis, and Walker (2023) find that a significant proportion of the decrease in
Black-White disparities in pollution exposure can be attributed to differential
effects of the Clean Air Act (CAA) in communities with different relative
populations of Black and White individuals. In contrast, my study emphasizes the
exogenous impact of a foreign policy shock on local pollution relocation in the U.S.
54
and finds that land costs became the most critical determinant of the narrowing
racial disparity regarding waste transfers after the policy.
The rest of this chapter is structured as follows: Section 2 provides a brief
overview of California’s waste transfer system and China’s policy on restricting
waste imports from the United States, along with a description of the data sources
used. The impact of China’s GS policy on pollution relocation in California is
analyzed in Section 3. Section 4 investigates the potential mechanisms behind
changes in the pattern of waste transfers between Black and White communities.
Finally, Section 5 concludes with some limitations and recommendations for future
research.
3.2 Background and Data
3.2.1 Background.
Local communities and waste transfers. Historically, small, local dumps
were the primary places for disposing solid waste before the practice evolved to
transporting waste to larger, regional facilities. However, the 1990s brought about
more rigorous government regulations aimed at safeguarding public health and
the environment. These new measures prompted significant improvements in
waste management technologies and necessitated the closure of many landfills
that fell short of the prescribed standards. In the wake of this, contemporary,
large-scale landfill facilities were established, leading to a considerable rise in the
transportation of solid waste across counties and states. Between 1989 and 1999,
the movement of waste between states soared by 300%, escalating from 10 million
tons to 30 million tons (Repa, 2005). Trucks are still the main vehicles used for
transporting waste between facilities, though some landfills have begun using
trains to haul trash from urban centers, with the aim of reducing truck traffic and
55
its associated emissions. Nowadays, waste continues to be transported between
communities. For example, New York City dispatches its waste to locations in
upstate New York, New Jersey, and even Virginia, while Los Angeles sends its
waste to northern California, Oregon, and Nevada. Presently, Uniontown, Alabama,
is a recipient of waste, including hazardous coal ash, from as many as 33 different
states.
The international transportation of waste is influenced by a variety of
factors, including environmental regulations, labor costs, and trade imbalances.
One example of this has been the trade in recyclable waste between the United
States and China. In the ten years following China’s accession to the WTO in
2001, U.S. exports of recyclable waste to China increased from 5 billion tons to
20 billion tons, representing an increase from 27.2% to 59.8% of total recyclable
waste exports from the U.S. This increase can be attributed to several factors.
First, China was initially viewed as a “pollution haven” where recycling processes
that generated pollution could still be used because of the lack of stringent
environmental regulations in the country. Second, China experienced significant
economic development and demand for raw materials, particularly in manufacturing
industries, and as a result, the import of recyclable waste from developed countries
tended to increase alongside expanding domestic industrial activity and economic
growth. Lastly, because China frequently had a trade surplus with the United
States, ships returning to China from the U.S. often had excess capacity and could
therefore transport recyclable waste at relatively low marginal costs.
In contrast, the transportation of waste within a nation or among local
communities can be attributed to factors such as disposal fees, land expenses,
transport costs, biased landfill location decisions, and uneven enforcement of
56
environmental regulations. In certain uncommon instances, evidence suggests
that communities with limited political influence have been indirectly selected
as sites for industries or sources of pollution, including waste disposal. These
communities are more susceptible to environmental pollution due to a lack of
political representation, limited resources for challenging polluting industries, and
biased zoning and regulatory practices. For instance, an indigenous community
near the Standing Rock Sioux reservation, with arguably less voting power and
political clout, has been contesting the pipeline’s rerouting from a predominantly
White community due to concerns over water pollution and the disturbance of
sacred lands.5
This paper reports upon an investigation of the impact of China’s recyclable
waste import ban on waste shipments at the local level within the U.S. state of
California. Specifically, I analyze a dramatic policy change affecting in international
waste shipments and its effects on intranational waste shipments. China had been
the largest importer from the U.S. until 2016, when revisions to its environmental
regulations and policies led to the implementation of stricter inspection and
contamination limits under the Green Fence and then Green Sword policies.
China’s ban on imports of many types of recyclables has resulted in a shift from
international waste relocation to domestic waste relocation, which has differentially
affected local communities. This research examines the distributional consequences
of an international trade-related policy on waste shipments received by local
communities in California, which is home to roughly 10% of the population in the
U.S.
5Standing Rock Sioux and Dakota Access Pipeline: https://americanindian.si.edu/nk360/plains-
treaties/dapl
57
Environmental Equity. Over the past four decades, a significant body of
research has thoroughly documented the prevalence of environmental injustice,
highlighting that pollution and environmental degradation disproportionately
affect minority communities. This field of research emerged in the late 1970s when
a middle-class Black neighborhood in Houston, Texas, discovered a solid waste
facility was planned for their area. In 1987, a national study, spearheaded by
United Church of Christ minister Benjamin Chavis, validated that race was the
primary determinant of whether a person would live near a hazardous waste site,
even after adjusting for geographical location and income. The most pronounced
disparities in hazardous environmental exposure occur in communities that are
both economically disadvantaged and belong to minority groups (Bullard, 1983;
GAO, 1983; Chavis & Lee, 1987).
The issue of unequal pollution exposure in minority communities is a
significant concern, as it can result in disparities in health, education, and other
outcomes. Research by Hoek et al. (2013) demonstrated a negative association
between air pollution and health outcomes, while accounting for race and income.
Various studies have utilized quasi-experimental designs to establish a robust causal
link between pollution exposure and health consequences for both children and
adults (Chay & Greenstone, 2003; Currie & Neidell, 2007; Currie & Walker, 2011;
Currie, Davis, Greenstone, & Walker, 2015; Schlenker & Walker, 2016; Currie,
Greenstone, & Moretti, 2011; Persico, Figlio, & Roth, 2016).
Furthermore, long-lasting effects of pollution exposure on health, education,
and economic outcomes have been observed, including human capital accumulation,
labor market performance, family structure, and welfare reliance (Black, Devereux,
& Salvanes, 2007; Currie & Moretti, 2007; Oreopoulos, Stabile, Walld, & Roos,
58
2008; Sanders, 2012; Figlio, Guryan, Karbownik, & Roth, 2014). Studies by Aizer,
Currie, Simon, and Vivier (2018) and Persico et al. (2016) found that lead exposure
and proximity to Superfund sites can influence test scores and other educational
outcomes. By reducing such exposure, the gap in educational outcomes between
disadvantaged children and their peers can be significantly narrowed.
Addressing environmental inequality requires understanding the underlying
mechanisms contributing to such disparities. S. Banzhaf et al. (2019) highlights
several key factors: 1) discriminatory siting of harmful facilities, 2) low-income
households prioritizing basic needs over environmental quality, leading to
higher pollution exposure, 3) Coasean bargaining, where firms and households
negotiate pollution compensation, but unclear environmental property rights leave
communities vulnerable, and 4) the government’s role in distributing pollution
through legislation, monitoring, enforcement, and regulatory decisions, which
can be influenced by various factors and interest groups, potentially resulting in
inequitable exposure to environmental hazards.
Building on the previous environmental inequality mechanisms, I suggest
three potential factors influencing local waste transfers. First, land cost and
population density could directly impact waste disposal tipping fees, leading to
low-density rural areas, often predominantly White in California, receiving excess
waste. Second, transportation costs can influence waste haulers’ decisions, with
areas farther from cities possibly receiving less waste due to distance. Third, local
political influence can affect waste transfers through the permitting process, as
landfills in the United States, whether private or public, need permits from local or
state government agencies to open or expand. This process, involving compliance
with environmental regulations, zoning requirements, and public health and safety
59
standards, could result in unequal waste exposure. My paper aims to examine
which mechanism dominates general waste transfer patterns and investigates if the
dominant mechanism changes after an exogenous policy shock from China.
3.2.2 Data.
California Disposal Flow Data. To investigate the impact of China’s GS
policy on pollution relocation in California, I analyze data from CalRecycle’s
Recycling and Disposal Reporting System (RDRS) from 2002 to 2021. The RDRS
provides facility-level information on disposal flows, including waste flows (in
disposal tons), for each origin jurisdiction and destination facility. These data
are collected quarterly and cover more than 450 origin jurisdictions and over 250
destination disposal facilities over the 2002 to 2021 time period.
Socioeconomic and Demographic Data. To capture the characteristics
of destination communities, I collected data from various sources. To measure
racial composition, I utilize census-block level data from the U.S. Census. Median
income data is drawn from the U.S. 5-year ACS at the census-block-group level.
For economies of scale, I rely on data from the Waste Business Journal, which
includes geographic coordinates for all recycling-related facilities across the United
States, such as landfills, composters, recycling centers, and transfer stations. Lastly,
I obtain data on political ideologies from California’s Statewide Database (SWDB)
for Elections, specifically, presidential election data at the precinct level. The
summary of all data sources utilized in this paper is displayed in Table 4.
3.2.3 Summary Statistics. Appendix Table B1 provides summary
statistics for CalRecycle RDRS disposal flows from 2002 to 2020. The dataset
covers of approximately 464 origin jurisdictions and 263 disposal facilities, on
60
average, over time. Columns 2 and 4 of Appendix Table B1 show that the average
disposal quantities shipped from the origin jurisdictions have increased since 2013,
as have the disposal quantities received by the destination facilities. In contrast,
columns 1 and 3 indicate that the numbers of origin jurisdictions and destination
facilities have decreased over time. Appendix Table B2 presents summary statistics
for community characteristics around each destination facility in the CalRecycle
RDRS data. These community characteristics are calculated for different buffers
around each destination facility to ensure the model’s robustness.
3.3 Determinants of Pollution Relocation: Evidence from California
My state-level analysis described in chapter 2 of this dissertation constitute
a first step towards understanding the causal effect of China’s waste ban on
aggregate emission levels in the U.S. Among my state-level analyses in last chapter,
I find that California’s methane emissions from the waste industry increased by
an average of 9.4 percent per year following China’s GS policy.6 To gain a more
detailed understanding of the disaggregated local environmental effects of the GS
policy, I use facility-level data on disposal flows within California to explore some
distributional effects of China’s GS policy and propose some potential mechanisms
to explain these effects.7
3.3.1 Raw Patterns.
Waste Flows in California. To depict visually the distributional effects of
China’s GS policy on pollution relocation, I plot the spatial distribution of all
6The number is calculated using U.S. EPA GHGRP data.
7I compare the EPA GHGRP data with the CalRecycle RDRS data and find that the EPA
data (in methane emissions) are highly correlated with CalRecycle disposal flow data (in tons).
See Figure B1 in Appendix.
61
destination facilities in the CalRecycle RDRS dataset on the map of California.
Appendix Figure B2 shows that most facilities in the CalRecycle data are located
near urban areas (highlighted in yellow) in California, and fewer facilities are
located in more-remote areas or agricultural regions. I then plot disposal flows on
a California map using the coordinates of each origin jurisdiction and destination
facility. To illustrate, I pick a specific city source (Los Angeles) and a destination
facility (Covanta Stanisalaus Landfill) and show general patterns of pollution
relocation in the state of California for this origin and this destination. Figure 13
shows four things: (1) the different destinations for waste pollution shipped
outside the source city; (2) the different origins for waste pollution shipped into
the destination community; (3) the size of the increase in waste shipments sent by
each origin jurisdiction after China’s GS policy; and (4) how much of an increase
in waste shipments was received by each of the destination facilities after China’s
GS policy. The maps in Figure 13 show that most shipments of waste pollution
are transported to destination facilities either in remote rural areas or in suburbs
immediately outside urban areas (yellow areas) of California.8 To explore further
the characteristics of the destination communities where receiving facilities are
located, I then plot an analogous spatial map based on the racial composition and
political affiliation for California.9 Figure 15 shows that for waste being shipped
from Los Angeles, for example, most changes in pollution relocation have involved
increased shipments to remote and light-shaded areas, with higher proportions
of White residents, or to darker-shaded areas where larger shares of minority
populations reside. Figure 16 shows that pollution relocation has also increased
8Figure 14 shows waste transfers from San Francisco and and into Potrero Hill Landfill.
9I use census-tract data for the racial composition map and precinct-level data for the political
affiliation map.
62
shipments of waste going to more-remote Republican-leaning districts. Appendix
Figure B3 and Appendix Figure B4 shows changes in disposal flows based on
median income and pollution vulnerabilities of the destination communities in
California after China’s waste ban.10 To address the concern that disposal facilities
are more likely to be sited in minority communities, I plot all landfill facilities in
the CalRecycle dataset in relation to the racial composition and political affiliation
of the communities. Appendix Figure B4 shows that although many facilities are
situated close to minority communities or Republican-leaning districts (darker green
or red areas), some facilities are located in White communities or Democratic-
leaning districts (lighter green or blue areas).
Altered Distributional Effects of Pollution Relocation. The maps
that summarize disposal flows motivate my research questions concerning the
distributional effects of China’s GS policy: (1) How has China’s GS policy, as a
specific international trade policy shock, affected existing patterns of pollution
relocation? (2) Are there environmental justice concerns with regard to changes
in waste pollution relocation?11
3.3.2 Distributional Effects.
Correlation. To identify factors that potentially determine disposal flows
and the relocation of waste pollution, I start with some simple scatter plots.
I select six characteristics of destination communities that may correlate with
10Pollution vulnerability is reported by the Office of Environmental Health Hazard Assessment
(OEHHA) CalEnvironScreen4.0. CalEnviroScreen is a screening methodology that can be used to
help identify California communities that are disproportionately burdened by multiple sources of
pollution.
11Pollution relocation refers to activities that transfer negative environmental externalities to
places outside of the local community.
63
patterns of waste pollution relocation that create a net increase in disposal inflows.
These characteristics include racial compositions (White and Black shares of the
population), median income, economies of scale, and political affiliation. I define
“economies of scale” for a community by counting the number of other related
facilities/industries that are within a 15 km buffer around each destination facility12
Figure 17 shows that the White share of the population of the destination
community is inversely related to net increases in disposal inflows. However, the
Black share of the population is directly related to waste pollution inflows. These
relationships suggest that the higher the share of Whites in the population of the
destination community, the lesser the increase in disposal inflows.
Among other factors, the distances between origin jurisdiction and
destination communities(facility) are inversely related to the net increase in waste
inflows to destination communities. Nearby communities tend to receive greater
net increases in waste inflows. The median income of the destination community is
also negatively correlated with the net increase in waste inflows, which shows that
the lower the median income for a community, the greater the increase in waste
inflows it experiences after China’s GS policy. Economies of scale for destination
communities are negatively correlated with the waste inflows. The fewer similar
facilities are located near the destination facility, the greater the increase in waste
inflows to such communities. Finally, the Republican voting registration share of
the destination community is positively correlated with net increases in waste flows.
12Racial composition measure is defined by the average population share by race within a 3-km
buffer of the destination facility(see more details in Appendix Figure B6 and Figure B7). Political
affiliation is defined by average registration share by party within a 3-km buffer of the destination
facility (see more details in Figure B8 and Figure B9). The economies of scale measure is defined
by the number of waste facilities that are within a 15-km buffer of the destination landfill facility
for a disposal shipment. Waste facilities can be composting sites (CO), landfills (LF), recycling
centers (MR and MW), and transfer stations (TS) (see more details in Appendix Figure B10 and
Appendix Figure B11).
64
This shows that the greater the Republican share in the destination community,
the greater the increase in waste inflows experienced by the destination community
after China’s waste ban.
Fix-effects model. Knowing that California has seen a significant increase
in methane emissions and waste pollution after China’s GS policy, I investigate
the distributional effects of China’s policy on waste flows for local communities
(at the census-block level) in the state of California. I apply a fixed-effects model
which includes the distances between the origins and destinations for inter-regional
waste flows. For each destination community, the model also includes racial
composition, median income, economies of scale in the waste industry, and the
voting registration share of residents.13 The model specification is as follows:
Yijt =α + β[1log(Distij) + β2log(R]j) + β[3log(Xjt) ] [ ]
+ β4 log(Distij)× 1(post) + β5 log(Rj)× 1(post) + β6 log(Xjt)× 1(post)
+ ζo+ θd + µod + ηt + ϵijt
(3.1)
The dependent variable Yijt is the tons of the waste transported from
jurisdiction i to destination (facility) community j in year-quarter t.14 Distij is
the distance between origin jurisdiction i and destination community j. Rj is the
racial composition of destination community j. 1(post) is an indicator variable,
which takes the value of 1 for the year of China’s waste ban and beyond. Xjt is
a set of socioeconomic factors such as median income, economies of scale in the
13The voting registration data used here at the precinct level is from Statewide Database
(SWDB) election data.
14I define the destination community as the areas that are within a 3km buffer of the
destination facilities (S. Banzhaf et al. (2019)).
65
waste industry, and political affiliation for destination j.15 To reveal any altered
distributional effects caused by China’s GS policy, I interact the characteristics
of the destination communities with the policy indicator variable. In Appendix
Table B3, I present results that are adjusted for different types of fixed effects,
including origin-county fixed effects (ζo), destination-county fixed effects (θd), year-
quarter fixed effects (ηt), and an error term (ϵt).
Figure 18 shows my estimates of the altered distributional effects caused
by China’s GS policy. Before China’s waste ban, the estimates and their 90% and
95% CIs show that the sizes of waste flows are negatively correlated with distances
between origin and destination communities. The greater the proportion of Black
residents in the destination community, the more waste pollution the community
receives from other places. The greater the proportion of White residents in
the destination community, the less waste pollution the community receives.
These coefficient estimates on the racial composition variables for the destination
communities confirm that well-documented racial disparities in pollution exposure
also exist with regard these patterns of waste pollution relocation. The median
income and economies of scale for destination communities are both positively
correlated with the amount of waste the destination communities receive.16 The
higher the percentage of registered Republican voters in a destination community,
the more waste pollution is transported to that community on average, prior to
China’s waste ban.
15Economies of scale for community j are measured by counting how many waste-related
facilities are within a 5km buffer of the destination facility.
16Comparing to all destination communities with a waste facility, communities with more
economies of scale may provide more job opportunities to the community, which could account,
in part, for a higher median income.
66
I then use the interaction terms in the model to compare pollution
relocation patterns before and after China’s GS policy. After China’s waste ban,
the positive coefficient on Black share of the population shows that the Black
communities continue to receive more waste pollution from elsewhere. However,
communities with higher White population shares now are more affected—
communities with a higher White share of their population have seen a greater
increase in their incoming shipments of wastes. Communities that are more remote,
have fewer economies of scale, and lower Republican shares receive more waste
pollution from elsewhere after China’s GS policy takes effect, compared to before
that event.
3.3.3 Mechanism.
Theory model. In this section, I present a stylized model to illustrate the
potential determinants of the observed altered waste shipments within California.
The amount of waste being transported to other locations depends on the amount
of waste generated by each origin jurisdiction and the cost of transporting this
waste from that origin to each destination. The more waste the origin jurisdiction
generates, the more waste can potentially be transported to other places. The lower
the cost of transferring the waste, the more waste is likely to be transported to
other locations. This relationship can be summarized as:
TranspWasteijt = f(TotalWasteit, Costijt) (3.2)
+ −
TranspWasteijt is the amount of waste transported from origin jurisdiction i to
destination facility j at time t. TotalWasteit is the amount of waste generated in
jurisdiction i at time t. Costijt is the monetary costs and non-monetary costs of
67
transporting this waste along with its negative externalities from origin jurisdiction
i to destination facility j at time t.
My empirical analysis, in the previous section, suggests several factors that
affect inter-regional pollution flows. Before China’s waste ban, waste pollution
tended to be transferred to communities with a higher percentage of Black
residents. This is still the case after the GS policy. However, after China’s GS
policy, waste transfers shifted, to some extent, to lower-income White communities,
and to less remote communities. The GS policy does not seem to have exacerbated
the usual environmental disparity across communities with regard to waste
pollution relocation. Instead, it has tended to narrow this relative disparity across
communities. Black communities have continued to receive, in absolute terms,
more waste shipments after the policy shock. However, White communities have
experienced a greater increase in waste pollution, relative to the shipments they
received before the GS policy.
There are three potential mechanisms to explain this altered distributional
effect. The reason some White communities are receiving more waste after China’s
GS policy may be due to some of their characteristics. These increasingly used
destinations are in California communities that tend to have (1) lower land costs;
(2) lower transportation costs; (3) lower political costs. I assume the cost of waste
relocation, in the case of recyclable waste transfers, depends on land values (LCjt),
transportation costs (TCijt), and political costs (PCijt) incurred in destination
communities where the receiving facilities are located. I also assume that the
amount of waste (TotalWastei) generated by an origin jurisdiction is relatively
constant over time.
68
Costijt = f(LCj, TCij, PCij) (3.3)
+ + +
Three metrics. First, land costs tend to be lower in places where
population densities are lower. Waste pollution tends to be transferred to places
with lower land costs, where tipping rates (i.e., disposal fees) are lower. Based on
this logic, I use population density (people/acre) as a proxy for land values.
LCjt = f(Populationjt) (3.4)
+
Second, waste pollution also tends to be transferred to closer locations to minimize
transportation costs. I use distance (in kilometers) between the origin and
destination as a proxy for transportation costs.
TCjt = f(Distanceijt) (3.5)
+
Third, waste shipments to some communities (in this case, voting precincts)
might be motivated by political cost. I define political cost as the deviation of the
destination community’s vote share from that of its county.17 If the community’s
vote registration share by party is very different to the vote registration share by
party of its county, then this community has a lower political cost.18 The voting
registration share of the community is the percentage of those who registered as
Republican among all registered voters in the population of the voting precinct.
17Vote share is defined as the Democrats/Republican registrations among all all registered
voters.
18The voting data is from Statewide Data Base (SWDB). SWDB collects the Statement of Vote
and the Statement of Registration along with various geography files from each of the 58 counties
for every statewide election. The Statement of Vote is a precinct-level dataset and precincts in
California change frequently between elections. The goal of the SWDB is to make election data
available that can be compared over time, on the same unit of analysis—a precinct, a census block
or a census tract.
69
Pjt = f(︸V otesjt ︷−︷V otesc︸t) (3.6)
−
V otesjt is the Republican registration share of the community where destination
facility j is located, calculated at time t. V otesct is the Republican registration
share of county c to which destination community j belongs at time t.19 The
difference between the community and county registration shares by party reflects
the political cost of transporting waste and its externalities to community j. The
greater the difference between the community and its county in terms of “political
voting registration shares”, the lower the political cost for waste inflows to that
community. For example, a very Republican community in an overall Democratic
county may have a high vote discrepancy and, thus, a lower resistance to increased
waste shipments for various reasons: (1) such a community may have less political
influence within the county; (2) its residents may also have a very different
philosophy, relative to the county as a whole, concerning environmental issues; or
(3) it may be harder to change the minds of such voters in such communities about
their political affiliations. Consequently, waste haulers may hear fewer complaints
from such communities if they increase their shipments to such places. For all three
of these reasons, these communities may put up the least resistance to increased
waste relocation after China’s GS policy shock. Figure 19 shows the spatial
distribution of voting registration discrepancies across communities (precincts) in
California. The lighter the color, the more the destination community deviates from
its county in political ideology, regardless of the dominant party in that county.
I estimate an ad hoc regression specification in order to examine which of
these potential mechanisms best accounts for my finding that waste pollution has
19Here the vote shares are registrations votes.
70
relocated relatively more to poorer, more remote, and White communities after
China’s GS policy shock. The regression is as follows:
[ ]
Yijt =α + β1Cij + β2 Cij × 1(post) + µod + ηt + ϵodt (3.7)
Cij are the three cost metrics: land cost, transportation cost, and political cost—
approximated by population density, distance between origin and destination, and
discrepancy (absolute difference) of political vote shares between community and
county. Population density is from the 2010 census data at the census-block level.
Vote data at the precinct level is from the 2016 presidential election. I examine
which of the three potential mechanisms appears to dominate as an explanation for
the altered distributional patterns in waste pollution relocation.
Results. Table 5 shows that before China’s waste ban, i.e., 1(post) = 0, waste
pollution tended to be relocated to remote places with low land values. Waste also
tended to be transported to places with relatively low political costs. However,
after the waste ban, more waste pollution has been relocated to destinations that
are further away, with lower land costs but higher political costs. Furthermore,
the effect of land costs on the altered distributional patterns is more statistically
significant (at 5% level) than the effect of political costs (at 10% level). Although
land costs and political costs both seem to influence waste pollution relocation,
these estimates suggest that land costs may be more important than political costs
as a determinant of the decisions about where to transport excess amounts of waste
pollution in the event of an exogenous policy shock.
71
3.4 Conclusion
This paper investigates the far-reaching environmental consequences of
China’s waste import ban on local communities in the United States, particularly
in California. The study examines changes in the burden of waste pollution on
various communities before and after the policy is implementation. By utilizing
detailed demographic, socioeconomic, and waste transfer data, the research also
contributes to the Environmental Justice literature by shedding light on changes in
pollution disparities between Black and White communities as a result of the policy
shock.
My findings reveal that China’s waste import ban has led to a counter-
intuitive decrease in the pollution gap between Black and White communities.
Minority communities have a history of being exposed more to waste pollution.
However, after the waste ban, lower-income White communities experienced a
relatively greater increase in waste inflows. This unexpected outcome prompts
an investigation into the mechanisms driving this change. I propose three
explanations—land costs, transportation costs, and political costs—to determine
which factors are responsible for the narrowing racial gap. My analysis shows that
before the waste import ban, transportation and political costs were the main
determinants of waste transfers across communities. However, after the policy
shock, low land costs for a community have become a significant determinant for
waste relocation. This research highlights the broader environmental implications of
China’s waste import policy for the US, beyond its immediate trade impacts, and
emphasizes the importance of studying the domestic recycling industry and related
policies in environmental economics.
72
This study still has several limitations, however. First, the waste data used
in the local pollution relocation analysis cannot accurately track the amount or
composition exclusively of recyclable wastes that are transferred locally, since
disposal data incorporates both regular and recyclable wastes. Second, this paper
has examined the effects of China’s waste ban on emissions and pollution relocation
related to landfill facilities only; data specifically for recycling facilities are not yet
publicly available due to privacy issues. Finally, the GS policy might not be the
only factor that has caused the observed narrowing of the racial disparity for waste-
transfer destinations; other unobserved factors such as local policies or ideological
changes in environmental equity might simultaneously have begun to have an effect
on California’s waste transfers around 2017.
While this study focuses on California, future research could potentially
leverage satellite data to directly detect pollution caused by waste and evaluate
the long-term effects of the policy on US communities. Additionally, further
investigation into the distributional effects of the policy on waste facility siting
and capacity expansions in minority or low-income communities is warranted, if we
wish, to understand more completely how the waste ban has altered the landscape
of waste transfers across local communities. This paper is the first to use empirical
data from multiple sources and spatial scales, as well as a variety of econometric
methods, to assess the policy’s effects on the US environment at national, state,
and local levels, providing valuable insights into its equity implications.
73
3.5 Figures
Figure 13. CalRecycle: Average Net Increase of Disposal Flow after China’s GS
Policy Los Angeles and Covanta Landfill. These maps show the disposal flows from
source cities (Los Angleles) and disposal flows to destination facilities (Covanata
Stanislaus, Inc.), as examples. They show (1) where the disposal goes from Los
Angeles and (2) where disposals originate for the Covanata Stanislaus, Inc. The
color of the arrows shows the increase in amount of disposal flows after China’s GS
policy. From the source city, most of the disposal has gone to rural or suburban
areas outside the urban areas (yellow areas). Disposal that was transferred to closer
rural areas increased more (represented by the darker color of curves with arrows)
after China’s GS policy.
74
Figure 14. CalRecycle: Average Net Increase of Disposal Flow after China’s GS
Policy San Francisco and Potrero Hills Landfill. These maps show the disposal
flows from source cities (San Francisco) and disposal flows to destination facilities
(Potrero Hills Landfill), as examples. They show (1) where the disposal goes from
San Francisco and (2) where disposals originate for the Potrero Hills Landfill. The
color of the arrows shows the increase in amount of disposal flows after China’s GS
policy. From the source city, most of the disposal has gone to rural or suburban
areas outside the urban areas (yellow areas). Disposal that was transferred to closer
rural areas increased more (represented by the darker color of curves with arrows)
after China’s GS policy.
75
Figure 15. CalRecycle: Average Net Increase of Disposal Flow by Racial
Composition. This map shows the net increase in disposal flow after China’s GS
policy based on racial composition. The geographic unit for racial composition is
the census tract. The color of the arrows shows the increase in amount of disposal
flows after China’s GS policy. The map shows that most of the destinations for
disposal transfer are in “minority” areas (represented by darker blue/green).
However, the “White” areas (represented by lighter green) have seen a greater
increase (darker red arrows) in waste transfers after China’s GS policy.
76
Figure 16. CalRecycle: Average Net Increase of Disposal Flow by Registered
Voters. This map shows the net increase in disposal flows after China’s GS
policy based on voting registrations. The geographic unit for vote shares is the
voting precinct. The map shows that more destinations of waste transfers are in
Republican precincts (red) in California. The color of the arrows shows the increase
in amount of disposal flows after China’s GS policy.
77
Figure 17. Correlations of Disposal Flow and Destination Community
Characteristics. These plots show potential factors determining the community-
level waste pollution relocation in California after China’s GS policy. The
percentage of the White population, the distance between origin and destination,
median income, and the economies of scale of destination communities are
negatively correlated with the net increase in waste inflows. The percentage of
the Black population and the percentage of Republican voting registrations are
positively correlated with the net increase in waste pollution inflows.
78
Figure 18. CalRecycle: Effect of Destination Community Characteristics on Waste
Flows Before and After China’s GS Policy. This plot shows the results of the
fixed-effects model. Before the GS policy, the White share of the population in a
community is negatively correlated with the amount of waste transported into the
community. The Black share of the population is positively correlated with the
amount of waste transported into the community. However, this pattern changed
after the GS policy. The White communities have seen a greater increase in waste
inflows than Black communities. The estimate (after the GS policy) that does not
intersect with the CIs of estimates (before the GS policy) shows that the change is
statistically significant.
79
Figure 19. Potential Mechanism: Disposal Flow Map by Political Deviation.
Political deviation is one of the three potential mechanisms to explain waste
transfers across communities. Blue/green indicates the political deviation of a
community from its county in terms of Republican voting registration shares
(voting registration is the only data that is available at the precinct level in
Statewide Election Database.). White spaces indicate where no data was available.
Political deviation is calculated by the absolute difference between a community’s
Republican voting registration share and its county’s Republican voting registration
share. The higher political deviation a community has, the lower the political cost
such a community has for waste pollution inflows.
80
3.6 Tables
Spatial Unit Years available Frequency
CalRecycle Disposal Flow Data jurisdiction by facility level 2002-2020 quarterly
U.S. Census Data census block level 2000-2020 decennial
ACS 5-year Data census block group level 2002-2017 5-year
Waste Business Journal facility level 1992-2020 yearly
Statewide Database Election Data precinct level 2000-2020 4-year
Table 4. Data Sources Summary: Community-level Analysis. This table summarizes all of the
data sources that are used in this paper. Disposal flow data is aggregated to origin jurisdiction,
destination facility, and year. For the census data, since the geographic units are small and
data frequency is low, I use 2010 census-block level data for racial composition, 2013 ACS 5-
year data for median income, and 2016 precinct-level election data for vote share.
81
Dep.Variable: Disposal shipment (tons) (1) (2) (3) (4)
Transportation costs -0.326∗∗∗ -0.476∗∗∗
(0.113) (0.112)
Transportation costs×1(post) 0.031 0.0196
(0.049) (0.063)
Land costs 0.019 -0.063
(0.052) (0.060)
Land costs×1(post) -0.017 -0.057∗∗
(0.020) (0.024)
Political costs 0.028 -0.011
(0.041) (0.032)
Political costs×1(post) -0.107∗ 0.101∗
(0.062) (0.057)
County FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Quarter FE ✓ ✓ ✓ ✓
R2 0.642 0.638 0.654 0.664
Observations 293,238 291,016 210,767 209,647
Two-way clustered standard errors at the county-year level in all models. *p < 0.1, ** p < 0.05,
*** p < 0.01.
Table 5. Potential Mechanisms: Model Estimates. Transportation costs are
approximated by the distances between origin jurisdictions and destination
facilities. Land costs are approximated by the population density of the
communities where the destination facilities are located. Political costs are defined
as the discrepancy (absolute difference) between the community Republican
voting registration share (at the precinct level) and the county Republican voting
registration share. For example, community A has 30 percent of Republican
voters, and the county it resides in has 45 percent of Republican voters. The
political cost of community A as a destination community for waste shipment is
|30− 45| = | − 5| = 5, which is a “high” political cost. Because these communities,
which have similar political ideologies as the county, are more likely to be resistant
to the increased waste relocation due to the GS policy.
82
CHAPTER IV
WILLINGNESS TO BEAR THE COSTS OF PREVENTATIVE PUBLIC
HEALTH MEASURES
This chapter is co-authored with Trudy Ann Cameron. I had an essential
role in developing the initial idea that led to this project. I also wrote a significant
amount of code that generates the results in this paper. I wrote and edited many
sections of the paper. I have also presented this paper at multiple seminars and
conferences.
4.1 Introduction
Many policies and regulations are intended to protect human life and health.
In the context of the recent global pandemic, given the externalities associated with
infectious disease, public health policies have been essential. To analyze the benefits
and costs of public health measures, policymakers must take into account the level
of (and heterogeneity in) people’s willingness to bear the costs of appropriate public
health measures.
It is challenging to monetize the social benefit from costly policies to
protect human life and health. Economists typically use a measure called the
Value of a Statistical Life (VSL) to quantify society’s willingness to bear the costs
of small reduction in mortality risks for a large number of people. VSL can be
interpreted as a marginal rate of substitution between individual private mortality
risk and money. Mathematically, VSL is the marginal utility of a small reduction
in mortality risk divided by the marginal utility of a small change in income. In
2006, for example, the U.S. Environmental Protection Agency (EPA) estimated
that people in the U.S. are willing to pay about $7,000,000 for one “statistical” life.
83
This number means, for example, people are willing to pay about $70, on average,
to reduce the probability of death by 1/100,000 for 100,000 people.1
For the COVID-19 pandemic, Echazu and Nocetti (2020) calculate society’s
overall willingness to pay for morbidity and mortality risk reductions. They
estimate that the aggregate social WTP for a sizeable reduction in infection risk
during a pandemic may be on the order of $3T to $7T. This dramatic estimate
for WTP (for all statistical lives “lost”) for risk reduction during an infectious
pandemic likely reflects the fact that people are willing to pay not just for a
reduction in their own risk of illness and death, but also to permit reductions in
the stringency of pandemic restrictions. Cameron (2010) points out that VSL, as
a “one-size-fits-all” measure, can hinder our ability to understand distributional
effects of risk-reducing policies or interventions. A single VSL—where the majority
of estimates of the VSL are derived from labor-market studies where the risk in
question is sudden death in an industrial workplace accident—may also fail to
reflect the particular features of COVID-19 as a specific health threat. Likewise,
the populations for which wage-risk VSLs are typically estimated (prime-aged
white male workers in hazardous occupations) may be a poor approximation to
the characteristics of the populations most seriously affected by COVID-19.
The research described in this paper constitutes an exercise in “benefits
function transfer” (Smith, Van Houtven, & Pattanayak, 2002), where the “study
sample” is an existing survey-based choice experiment fielded to more that 1400
respondents in a representative probability sample of households in counties
across the U.S. in 2003 (Bosworth, Cameron, & DeShazo, 2009). The goal in that
1EPA’s estimates of the value of a mortality risk reduction were reviewed in a white paper
called ”Valuing Mortality Risk Reduction in Environmental Policy” included 33 studies between
1988 to 2009. See line 694 in this white paper.
84
original study was to determine the social benefits from public health policies to
reduce illness and deaths from different types of health threats in the respondent’s
community. For the current benefits transfer task, the “policy samples” consist
of the populations of all counties across the U.S. during the 2020-21 COVID-19
pandemic.
Benefits transfer has been widely used to in environmental economics to
supply information for benefit-cost analyses to support policy decisions when a new
study is not affordable or when no time is available to conduct a thorough new
study (Richardson, Loomis, Kroeger, & Casey, 2015). Benefits function transfer
exercises can involve study and policy samples at different points in time where
conditions may be different. For example, Price, Dupont, and Adamowicz (2017)
evaluate the temporal stability of willingness-to-pay values from two identical
stated preference surveys undertaken in 2004 and 2012. The surveys were designed
to capture the trade-offs between (a) risk reductions for two health endpoints
related to tap water, and (b) monetary costs. Across these two time periods, their
study found no significant differences in real-valued WTP, or in the structure of
heterogeneous preferences.2
In the broader environmental benefits literature, it is also a common practice
to estimate a benefits function for one country, and then to attempt to transfer this
benefits function to another country. These efforts can be challenging, however,
because there are often cultural differences between countries (especially between
developed and developing countries) that can call into question whether the
preferences estimated in one country should be expected to hold in another country
2Benefit function transfers maybe be derived from just one study, or they may combine the
results for several related studies to “triangulate” the conditions for which a new benefits estimate
is needed.
85
(Brander, Beukering, & Cesar, 2007; Lindhjelm & Navrud, 2008; Ready & Navrud,
2006). In this paper, fortunately, we seek to transfer a benefits function only
between two different time periods in the U.S. This requires only that we assume
that U.S. preferences over public health policies and net incomes be relatively
stable across time, after controlling for changes over time in the variables that
systematically affect these preferences. It also requires the assumption that cross-
sectional differences among U.S. counties in 2003 have similar effects on public
health policy preferences as do changes over time in the characteristics of these U.S.
counties.
Instead of using a single one-size-fits-all VSL, our research estimates people’s
WTP for public health policies that reduce both illnesses and deaths, in light of
both the relevant cost and the expected duration of such policies. Furthermore,
rather than focusing on private WTP to reduce an individual’s personal mortality
risk, we emphasize a specifically public program, where people are asked their WTP
for reduction in the risk of illness and deaths in their broader community. In our
current analysis, we interpret counties as communities. Although counties are
not the smallest geographic regions we might use, they are the most appropriate
administrative units in the context of the original survey. During a public health
crisis like the COVID-19 pandemic, publicly available data on cases and deaths are
also commonly reported at the county level.
Assessing people’s willingness to pay for community-level public health
policies is essential for public health policymakers for four reasons. First, people
from the same community often have more in common than do people from
different communities, in terms of sociodemographics, ethnicities, economic status,
and health characteristics. To the extent this is true, community characteristics
86
may systematically affect individuals’ preferences for public health policies. Second,
during an infectious pandemic like COVID-19, people’s behaviors and actions are
intimately related to the health and well-being of others who live in the same
community. Third, pandemic policies have often been tailored to conditions in
specific counties as authorities attempt to allocate public health resources more
efficiently. Fourth, many communities struggle with specific types of health risks
systematically. For example, Lincoln, Abdou, and Lloyd (2014) find that Black
communities tend to suffer more from obesity and depression than do White
communities. Yancy (2020) finds that, during COVID-19, infection rates within
Black-dominated communities have sometimes been three times higher than that
in White-dominated communities. Even more strikingly, the COVID-19 death
rate for Black communities has been as much as six times higher than for White
communities. With more-refined knowledge about their population’s willingness to
bear the costs of community-level health policies, county-level decision makers can
implement public health measures with greater confidence that their strategies will
deliver positive net social benefits for their constituents.
In this research, we re-analyze some high-quality stated-preference choice-
experiment survey data from an original 2003 study reported in Bosworth et al.
(2009) that reveals people’s preferences for randomized public policies that benefit
community-level health.3 To permit out-of-sample forecasting, our re-analysis
substitutes county-level explanatory variables for the individual-specific variables
that were largely relied-upon to explain respondents’ choices in the original study.
We collect new data on county-level policy contexts with the requirement that our
3The 2003 survey was one of four surveys funded by research grants from the U.S. EPA and
the National Science Foundation, and was fielded using Knowledge Networks, the leading research-
quality representative consumer panel available in the U.S. at the time.
87
measures for all these county-level variables be available for both (a) the 2003
context and (b) the more-recent context of the 2020-21 pandemic. We need to
control for differences, both across counties and between 2003 and 2020-21, in each
county’s mix of socio-demographic characteristics, incomes, political affiliations,
health status, and access to medical care. If people’s basic preferences for policies
to reduce risks to public health have remained sufficiently stable between 2003 and
2020-2021, after controlling for shifts in all of these explanatory variables, lessons
from our 2003 survey can illuminate people’s likely policy preferences today. While
we cannot identify a premium for infectious diseases, it will be helpful at least to
understand what people would be willing to give up simply to avert illnesses and
premature deaths at the scale of the recent pandemic.
We first estimate a latent class model and discern three classes of
preferences. Within each class, people’s preferences are driven by different
combinations of policy attributes and community characteristics. There is evidence
of considerable heterogeneity. Next, we use LASSO methods to help select the
most important observable determinants of heterogeneity in preferences for public
health policies using our 2003 data. Then, based on the updated community-level
characteristics in counties across the U.S. in 2020-21, we use the fitted model to
predict overall WTP for policies to reduce monthly generic cases and deaths on a
scale commensurate with county-level casualties from the COVID-19 pandemic. For
example, we find that people from Black-dominated counties have a higher WTP
for public health policies than those from White-dominated counties. Residents of
counties that have populations which are younger or more highly educated have
lower WTP for public health interventions to reduce illnesses and deaths on a scale
88
such as the recent pandemic risks, compared to those who live in counties with
older and less-educated populations.
Stated preference methods, such as those employed for this paper, are
used frequently to quantify preferences in health economics, health technology
assessment, risk-benefit analysis, and health services research (Mühlbacher &
Johnson, 2016). A few contemporary survey-based discrete choice experiments have
sought to understand public perceptions of COVID-19 pandemic interventions and
to identify preference classes across individuals. Rees-Jones, Attoma, Piolatto, and
Salvadori (2020) conduct a survey of 2,516 Americans concerning their preferences
for both short- and long-term expansion to governmental-provided healthcare and
unemployment insurance programs. That study finds that preferences for such
programs are positively affected by the county’s COVID-19 deaths, unemployment
caused by COVID-19, and how respondents perceive the consequences of COVID-
19. Chorus, Sandorf, and Mouter (2020) use survey-based choice experiments
to infer people’s preferences from the trade-offs they are willing to make among
policy effects, including health-related effects, impacts on the economy, education,
and personal income. They find that “the average citizen, to avoid one fatality
directly or indirectly related to COVID-19, is willing to accept a lasting lag in the
educational performance of 18 children, or a lasting (> 3 years) and substantial
(> 15%) reduction in net income of 77 households.”
In an earlier, pre-COVID context, Cook, Zhao, Chen, and Finkelstein (2018)
use a survey in Singapore regarding the trade-offs between risks of infectious
diseases and the inconvenience of government interventions to prevent outbreaks
of infectious disease. They find that respondents prefer more-intense interventions
and prefer scenarios with fewer deaths and lower taxes. L. Li, Long, and Rad
89
(2020) use a survey-based choice experiment in three U.S. states and empirically
quantify “willingness to stay home.” They find broad support for statewide mask
mandates. Their estimate of WTP to reduce new cases is large, and demographic
and socioeconomic factors are the main drivers of the heterogeneity in individuals’
willingness to stay home. Reed, Gonzalez, and Johnson (2020) also use a survey-
based choice experiment in the U.S. to quantify Americans’ acceptance of COVID-
19 infection risks from lifting public health restrictions earlier and to reduce
economic impact of the pandemic.4
Other recent papers focus on factors that affect people’s responses to
COVID-19. Cattapan, Ackerverney, Dobrowolsky, Findlay, and Mandrona (2020)
find that the need for community engagement is pressing in a pandemic crisis.
Engagement is essential to ensure that policy-making is built on equity, access,
and inclusion. Adeel et al. (2020) find that the sub-national policies of U.S. states
and Canadian provinces are more important than the national-level policies in each
country.
Some studies focus on the benefit-cost analysis of restrictive public health
policies during COVID-19. For example, Viscusi (2020) applies a standard Value of
a Statistical Life (VSL) to monetize COVID-19 deaths for the first half of 2020
and produces a U.S. mortality cost estimate of $1.4 trillion. Miles, Stedman,
and Heald (2020) conduct a benefit-cost analysis of U.K. public health policies
during COVID and find that the costs of continuing severe restrictions are large
compared to benefits. Dorantes, Kaushal, and Muchow (2020) use county-level data
on COVID-19 mortality and infections, as well as the county-level information on
4They find four classes of people among all respondents: “risk-minimizers”, “waiters”,
“recovery-supporters”, and “openers”. Political affiliation, race, household income, and
employment status were all associated with class membership.
90
the adoption of non-pharmaceutical interventions (NPI) and find that NPIs slowed
infection rates in counties where the healthcare system might otherwise have been
overwhelmed by the pandemic. They also suggest that political ideology is a factor
that may have limited the effectiveness of those measures in Republican-dominated
counties.
4.2 Data
4.2.1 The Original 2003 Survey. Our survey from 2003 was
originally employed in an analysis that takes advantage of the characteristics of
individual survey respondents to explain their policy preferences in that 2003
context. The original analysis is described in Bosworth et al. (2009). The 2003
survey produced 1,466 completed responses, and was designed specifically to
elicit individuals’ willingness to pay for publicly provided health policies.5 Each
respondent faces a choice between either of two different health policies and the
status quo. For example, Policy A might be described as reducing air pollutants
that cause heart disease; and Policy B might reduce pesticides in foods that cause
adult leukemia. The status quo “Neither Policy” option would involve no change
in community health risks, but also no cost to the respondents’ household. Each
policy is also described in terms of a set of attributes that includes cases and
premature deaths prevented in this community, duration of the policy, and the
cost of the policy. The randomized illness labels include respiratory disease, cancer,
leukemia, colon/bladder cancer, asthma, lung cancer, heart disease, heart attack,
and stroke. See Appendix Figure C1 for one instance of the randomized choice sets
used in the survey.
5See Johnston et al. (2017) for an inventory of current best practices in SP research.
91
The original survey was fielded in June of 2003 and was distributed to
members of a premium nationally representative consumer panel (Knowledge
Networks) that produced a representative sample of respondents from counties
throughout the conterminous U.S. The essentially national scope of the survey
captured extensive geographic variation in sociodemographics, voting patterns,
health status, and access to medical care. Figure 20 maps the geographic
distribution across counties of our 1,466 respondents.
Figure 20. The map highlights every U.S. county containing one or more online
survey respondents.
The main policy attributes described in each policy-choice task include
monthly cost, policy duration, the size of the affected population, illnesses avoided
and premature deaths averted. Our basic model allows for “status quo” effects,
i.e., a discrete mass of utility, positive or negative, associated with the “Neither
Policy” option, regardless of the specified attributes of either of the the two public
health policies under consideration. Importantly, each policy choice was followed
by a “self-interest” question about the degree to which the respondent or their
family would personally benefit from that particular public health policy. Briefly,
the relevant policy attributes for the present study were:
92
– Affected population in thousands: Across respondents, but not within
a respondent’s version of the survey, the original survey varies the size of the
population affected by the policy. While it would have been ideal to describe
this population as that of the respondent’s own county, the anonymity of the
survey prevented the tailoring of policy options specifically (in advance) to
match each prospective respondents’ county of residence. We asserted, about
each pair of policies, that these two policies will be implemented for the “X
thousand people living around you.” We randomized X (among 1, 2, 3, 4, 5,
6, 7, 8, 9, 10, 15, 20 (2-3% in each case), 30 (4%), and 50, 100, 500, and 1,000
(8-15% each).
– Policy duration: Each prospective policy to reduce public health risks was
described as a commitment to pay the cost of the policy for a specified time
period.
– Total illnesses avoided and deaths averted: Over the specified time
horizon, each policy is described as being expected to result in a specific
number of cases avoided and a specific number of premature deaths averted.
Preliminary models revealed that WTP for these public health policies is
not simply linear in these policy attributes. Instead, people appear to derive
diminishing marginal utility from additional avoided illnesses or averted
premature deaths. Likewise, preferences are nonlinear in the policy’s duration
and in the size of the affected population.6
– Status quo (or conversely, “Any policy”) effects: Respondents are
allowed to choose “Neither policy” in every choice set, if they do not like
either of the offered policies. Best practices in choice modeling include making
an allowance for a status-quo effect. Equivalently, we use an indicator that
equals one for “any policy (regardless of its effectiveness or duration)” and
zero for the “Neither policy” alternative.
– Monthly cost to your household: Each prospective policy was associated
with a specified private household cost, expressed both per month and
annually, with a reminder of the duration of the commitment.
Given that we need a model that can be transferred to the 2020-21 COVID-
19 context, we must forgo the use of any of the available individual-specific
variables that were collected by the 2003 survey for those respondents. In place
of these individual-specific variables, we recruit new county-level variables that are
6For the models in this paper, we employ logarithmic or shifted logarithmic transformations for
these variables, since these functions seem best to explain people’s choices.
93
both available and consistently measured both close to the time of the original 2003
survey and likewise close to the time of the recent pandemic.
Most of our 1,466 respondents made five policy choices each. For our
estimating sample, then, the 14,466 non-status-quo policies described in our choice
experiments have randomized levels of each attribute, with the attribute levels
in each case designed to span a wide range of possible policy choice scenarios.
Fortunately, the original design spans the potentially relevant ranges of attributes
for the 2020-21 pandemic. The arbitrary randomized distribution of the program
design attributes used in the 2003 survey is summarized in Table 6.
mean sd
Pop. affected/county pop. 2.706 8.341
Duration of policy (months) 167.9 116.6
Baseline illnesses 1004.7 2334.5
Number of illnesses avoided 606.9 13854.
Baseline deaths 96.16 472.0
Number of deaths avoided 102.1 467.9
Policies 14466
Table 6. Descriptive Statistics, Public Health
Policy Design Variables, Choice Experiments
Posed within the 2003 Estimating Sample
4.2.2 County-level Sociodemographic and Contextual
Heterogeneity. Respondents to the original survey considered an aggregate
of 7,233 choice sets. The randomized design of the choice experiments permits
the estimation of a set of homogeneous preferences without any risk of bias from
94
omitted policy attributes. In this paper, however, we seek to identify important
dimensions of preference heterogeneity. We permit policy preferences to vary
systematically with the characteristics of the community-of-residence (county) for
each respondent. Models with adequate preference heterogeneity can potentially
allow us to predict changes in demand for public health policies, over time, in
response to changes in sociodemographics, political ideologies, and healthcare
access.7 The cross-sectional variation in the original sample can be exploited to
accomodate differences in the composition of county populations across the 17-
year interval between the original 2003 study period and the recent 2020-21 policy
period.
4.3 Estimating Specification
We specify indirect utility as linear in net income. This is a common
practice and is expedient because this functional form allows the individual’s own
household income level to drop out of the utility difference that drives the model.
This leaves only the policy cost as a dollar-denominated measure that can be used
to calculate the marginal rates of substitution that can be interpreted as marginal
willingnesses to pay for avoided illnesses and avoided premature deaths.8
Preliminary exploration of the data has revealed that people tend to
experience diminishing marginal utility from illnesses prevented and premature
deaths averted. Given that microeconomic theory does not guide the functional
form of utility beyond an expectation of diminishing marginal utility, we generalize
our additively separable shifted-logarithmic form to a more flexible translog-type
specification that is quadratic in these shifted log transformations. We include
7See data source in Appendix C1.
8This description of the model assumes a basic familiarity with utility-theoretic conditional
logit choice models.
95
the square of each logged variable and the interaction between these logged terms
to yield a translog-type specification in terms of the changes in the numbers of
illnesses and dea(ths assoc)iated with policy A.
9
V Ai = α Y
A
i (− ci ) ( )2
+ β1log (∆illnessesA +)1 + β2lo(g ∆illnessesA)+ 12
+ β3[log ∆( deathsA + 1 +)β4log ∆( deathsA + 1 ) ] (4.1)
+ β5 log ∆illnesses
A + 1 × log ∆deathsA + 1 + β6(0) + ϵA
where β6 is the lump of utility associated with the status-quo alternative, which
involves no policy. For Policy A in equation (4.1), of course, there is no status-quo
utility increment/decrement.10 Under the status quo alternative, in the absence
of the policy, there will be no cost, but also no changes in the baseline numbers
of illnesses or deaths, so that indirect utility will be determined simply by the
individual’s income and any utility associated with the status quo:
V Ni = α (Yi) + β6(1) + ϵ
N (4.2)
Thus, in a pairwise choice between just Policy A and No Policy (N), the utility-
difference will depend on the cost of the policy, the expected cases of illness
avoided, and the expected number of premature deaths averted under the chosen
9A shifted logarithmic transformation adds one to the argument of the log function, ensuring
that the function takes a value of zero when the argument is zero. An alternative to our
specification in equation (4.1 where utility is expressed in terms of reductions in illnesses and
deaths (which should be “goods”) would be to use absolute illnesses and deaths, with and without
each policy (which would imply that each attribute was a “bad”, likely to confer a negative
marginal utility).
10If the interaction term in equation (4.1 does not have a statistically significant coefficient, the
level curves of the indirect utility function would to be circular, rather than elliptical.
96
policy: ( )
V Ai − V Ni = α −c(Ai ) ( )
+ β1log (illnessesA + 1 +) β log[ de(athsA2 + 1 ) ] (4.3)2
+ β3log (∆illnessesA +)1 + [β4 lo(g ∆illnessesA)+] 1
+ β [log (∆deathsA
2
5 + 1 +)β6 log (∆deathsA + 1) ] (4.4)
+ β7 log ∆illnesses
A + 1 × log ∆deathsA + 1
+ β8(−1) + (ϵA − ϵN)
Note that if baseline levels of illness or death are to affect utility within this
particular framework, they need to be interacted with the changes in the
numbers of illnesses and deaths under each policy. To limit the complexity of
the specification, we will allow baseline illnesses to shift only the marginal utility
of reductions in the number of illnesses, and allow baseline deaths to shift only
the marginal utility of reductions in the number of deaths. We also allow the
baseline marginal utility parameters in the equation 4.3 to vary with selected
sociodemographic variables for each respondent’s county. The coefficients on
these interactions capture the extent to which these county-level variables affect
the underlying preference parameters β1, ..., β6. As is typical, we assume that the
marginal utility of net income is approximately constant.11
4.4 Results
4.4.1 Identifying Dimensions of Heterogeneity: LASSO
Estimation. Table 7 provides parameter estimates for a set of three increasingly
complex specifications. After employing our shifted log transformations, Model 1 in
11This description of the basic model assumes pairwise choices between a single policy and
the status quo. In the data, however, respondents are asked to choose between a pair of policies
and the status quo alternative. The model in equation (4.3) can readily be generalized to
accommodate three-way policy choices.
97
Table 7 is even simpler than equation (4.3), being linear and additively separable.
Model 2 is a homogeneous-preferences model that is consistent with equation (4.3),
involving some key interactions between the basic attributes. Model 3 permits
the preferences in equation (4.3) to vary systematically with the characteristics
of each respondent’s county (circa the 2003 time period). To identify the subset
of more-important sources of systematic heterogeneity in policy preferences across
counties, we force the basic attributes into the model. We then interact each of the
basic attributes with all of the available county-level data and subject just these
interaction terms to LASSO variable selection.
characteristics and select the most important interactions using LASSO model
estimation.12 We use a LASSO model with 10-fold cross-validation to yield the
variables and interactions in the model specification in section 3. We then use the
LASSO-selected variables in a conditional logit model with individual fixed effects
to produce both the parameter means and their asymptotic variance-covariance
matrix for use in deriving WTP estimates for our 2020 WTP simulation.13 Table 7
model 3 provides the preliminary results based on LASSO-selected variables and
binary choice model estimation.
4.5 Benefit Transfer: 2020-21 WTP to Avoid COVID-19 Illnesses and
Deaths in Each Month
In contrast to the wide variety of choice scenarios presented to respondents
in our 2003 study sample, we wish to use our estimated model to simulate WTP in
12Packaged LASSO algorithms for logit models appear to be limited to binary choice
specifications. We assume that the same set of preferences underlie our three-way choices as would
drive the two pairwise choices that would be consistent with these three-way choices would remain
the preferred alternatives if it was to be paired with either of the two non-chosen alternatives from
the three way choice.
13Double Lasso: Use machine learning Lasso algorithm to select the variables. Then take the
selected variables back into a maximum likelihood the conditioned logit model with individual
fixed effects.
98
Table 7. WTP Estimation Model Results
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
Preferred alternative in choice scenario
Monthly cost -0.01*** -0.01*** -
0.015***
(0.0007) (0.0009) (0.004)
... × Unemployment (v last month) 0.0069**
(0.0025)
Policy duration -0.02*** - -0.013**
0.013***
(0.002) (0.005) (0.0048)
Log(base illnesses + 1) -0.037. 0.32* 0.44**
(0.02) (0.13) (0.15)
... × County prop. Hispanic -0.77***
[ ] (0.14)
Log(base illnesses + 1) 2 -0.017* -0.037**
(0.0065) (0.011)
... × County log. median income -0.082*
(0.038)
... × County poverty rate -0.297**
(0.113)
... × County prop. obesity -0.57*
(0.238)
... × County prop. excessive-drinking -0.143*
(0.065)
... × Primary care physicians rate -
0.001***
(0.0003)
99
Table 7 (continued).
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
Log(base illnesses + 1) × Log(duration) -0.028 -0.061*
(0.021) (0.030)
... × County log. median income 0.178***
(0.043)
... × County prop. obesity -1.182**
(0.328)
... × PM2.5 0.008**
(0.003)
... × Primary care physicians rate 0.002**
(0.0008)
Log(base illnesses + 1) × (Affected -2.43* -3.67
pop/1000)−1
(1.33) (7.15)
... × County prop. Hispanic 69.49*
(10.26)
... × PM2.5 0.58*
(0.295)
Log(base deaths + 1) 0.039 -0.4 -0.23
(0.031) (0.19) (0.31)
... × County prop. aged 65-84 12.96
6.235
... × County prop. excess.-drink. 0.039 -0.4 -6.234
(0.031) (0.19) (2.167)
(Log(base deaths + 1))2 0.011. -0.0016
(0.012) (0.0056)
... × County prop. Black 0.198*
(0.081)
Log(base deaths + 1) × Log(duration) 0.06. 0.063
(0.035) (0.042)
... × Primary care physicians rate 0.001**
(0.0004)
100
Table 7 (continued).
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
... × Preventable hospitalization rate 0.003*
(0.001)
Log(base deaths + 1) × (Affected 3.65. 14.82 *
pop/1000)−1
(1.96) (9.47)
... × County log. median income -40.8*
(9.57)
... × County prop. aged 65-84 -328.7*
(122)
Log(∆ illness + 1) 0.068*** 0.039 -0.26***
(0.0097) (0.052) (0.035)
... × Unemployment (v last month) 0.353*
(0.151)
... × County prop. Asian 6.46***
(1.89)
... × County poverty Rate -4.05*
(1.85)
... × County avg. physical unhealthy 0.278**
days
(0.087)
(Log(∆ illness + 1))2 0.0073 -0.016.
(0.0038) (0.009)
... × County prop. Republican 0.053***
(0.015)
... × Unemployment (v last month) 0.029**
(0.011)
... × County prop. aged 0-17 -0.534*
(0.219)
... × County prop. Black 0.097**
(0.037)
101
Table 7 (continued).
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
... × County prop. Asian -0.304*
(0.153)
... × County log. median income 0.04*
(0.018)
... × County avg. physical unhealthy -
days 0.031***
(0.007)
... × County prop. obesity -0.339*
(0.171)
... × Primary care physicians rate -
0.0004**
(0.0001)
... × Preventable hospitalization rate 0.0006*
(0.0002)
Log(∆ illness + 1) × Log(duration) -0.0061 -0.111**
(0.0088) (0.038)
... × County log. median income 0.059*
(0.029)
... × preventable hospitalization rate -0.013*
(-
0.0005)
Log(∆ illness + 1) × (Affected 0.077 1.17*
pop/1000)−1
(0.052 ) (0.50)
... × Unemployment (v last month) 0.37*
(0.17)
... × County prop. Black -3.48***
(0.84)
Log(∆ deaths + 1) 0.20*** 0.45*** 0.51***
(0.018) (0.091) (0.11)
102
Table 7 (continued).
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
(Log(∆ deaths + 1))2 -0.0083 -0.14*
(0.007) (0.061)
... × County prop. Black 0.198*
(0.081)
... × Primary care physicians rate -0.0005*
(0.0002)
Log(∆ deaths + 1) × Log(duration) -0.034 0.23*
(0.016) (0.094 )
... × County poverty rate -1.51**
(0.51)
... × County prop. excessive-drinking -0.44*
(0.23)
Log(∆ deaths + 1) × (Affected -0.21* 1.01
pop/1000)−1
( 0.085) (0.7)
... × County prop. Asian -8.97*
(4.52)
... × County prop. obesity -14.8*
(5.85)
... × County excessive-drinking rate -6.82*
(3.17)
... × Primary Care Physicians Rate -0.008*
(0.003)
1=Status quo 0.68*** 1.13*** 2.53*
(0.071) (0.13) (1.11)
... × Unemployment (v last month) -0.88**
(0.33)
... × County prop. Black 5.62***
(1.67)
... × County prop. Asian 8.78*
(3.97)
103
Table 7 (continued).
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
... × County prop. aged 65-84 -5.4*
(2.2)
... × Primary care physicians rate -0.0051.
(0.0019)
1=Status quo × (Affected pop/1000)−1 -2.75*** 8.01.
(0.62) (4.57)
... × County prop. Republican -10.77
**
(3.85)
... × County prop. Black 36 ***
(6.82)
... × County log. median income 9.98 *
(4.41)
... × County prop. college -39.35
***
(11.81)
... × County prop. smoker 107.8***
(23.51)
... × County prop. excessive-drinking 39.6***
(19.6)
(1=Status quo × (Affected pop/1000)−1)2 2.04*** -11.78*
(0.60) (5.12)
... × Unemployment (v last month) 5.36**
(1.96)
... × County prop. Black -30.9***
(7.0)
... × County prop. aged 65-84 -81.36*
(39.52)
... × County log. median income -25.8**
(7.24)
104
Table 7 (continued).
(1) (2) (3)
Parsi- Homo- Double
monious geneous Lasso
... × County prop. Asian 75.1**
(33.5)
... × County prop. college 40.84**
(14.03)
... × County poverty rate 85.25***
(25.72)
... × County avg. physical unhealthy -7.08***
days
(1.88)
... × County prop. smoker 108.3**
(24.82)
... × County prop. excessive-drinking -55.1**
(19.65)
... × Preventable hospitalization rate 0.11**
(0.04)
Max. log-likelihood - - -9935.82
11674.45 11627.78
No. respondents 1518 1518 1466
No. choices 7492 7492 7233
No. alternatives 22476 22476 21699
2020-21 by a representative individual in each U.S. county to prevent the numbers
of COVID-19 cases and deaths recorded in each month for which data are available.
We wish to simulate a measure of the household costs that people would have
been willing bear, if a public health policy in 2020-21 could reduce new illnesses
and baseline deaths to zero. COVID-19 is infectious, so until all of the cases are
eliminated, people cannot return to a normal life. Table 8 shows the hundreds of
105
new COVID cases each month across the entire U.S., along with the thousands of
the reported deaths. The policy we wish to simulate for 2020-21 is the reduction of
these baseline cases and deaths to zero.
Month 03/2020 04/2020 05/2020 06/2020 07/2020
mean/sd mean/sd mean/sd mean/sd mean/sd
COVID-19 cases 0.58 2.74 2.24 2.62 5.97
4.91 17.28 11.72 14.05 31.05
COVID-19 deaths 0.014 0.18 0.13 0.07 0.08
0.16 1.61 0.79 0.39 0.44
Month 08/2020 9/2020 10/2020 11/2020 12/2020
mean/sd mean/sd mean/sd mean/sd mean/sd
COVID-19 cases 4.54 3.74 5.84 13.50 19.77
28.01 12.01 16.50 40.97 83.00
COVID-19 deaths 0.09 0.07 0.07 0.11 0.23
0.46 0.29 0.19 0.30 0.71
Month 1/2021 2/2021 3/2021 4/2021 5/2021
mean/sd mean/sd mean/sd mean/sd mean/sd
COVID-19 cases 19.11 7.33 5.51 6.05 2.82
85.75 25.84 19.96 20.46 8.79
COVID-19 deaths 0.29 0.22 0.12 0.08 0.05
1.42 1.06 0.57 0.36 0.21
Observations 3142 3142 3142 3142 3142
3142
Table 8. Descriptive Statistics, 2020-21 COVID-19 New Cases and Deaths
(in hundreds), County-level.
106
4.5.1 Preferences for a Representative Individual in Each
County, for Each County-Month of the 2020-21 Pandemic. In lieu of
each individual respondent’s characteristics, our estimating specification explains
the choices of individuals using only the characteristics of the county in which the
individual resides. The distribution of characteristics of the U.S. counties used in
simulating our WTP amount for 2020 are shown in Table 9.
4.5.2 Parametric Bootstrap Estimates of Predicted WTP
in Each County-month. We estimate our models in utility space, so the
calculations of WTP involve dividing other coefficients by the estimated marginal
utility of net income, where all the maximum likelihood parameters in the model
are distributed asymptotically joint normal. We used a large number of draws from
the joint distribution of the parameters to calculate the predicted distribution of
WTP to reduce to zero all COVID cases and deaths in each county-month, with
the distribution being determined by the noise in the parameters’ estimates. Given
that there were no opportunities for respondents to record a negative willingness to
pay, we interpret negative calculated point values of WTP values as zero, using a
Tobit-like interpretation. We calculate monthly average WTP to reduce to zero all
cases and deaths attributed to COVID-19 from March 2020 to February 2021 across
all counties in the U.S.
For March 2020 through February 2021, Table 10 shows for a representative
county resident across all U.S. counties, the average monthly WTP to reduce to
zero of the COVID-19 cases and deaths. These monthly estimates vary by the
changes of monthly new cases, deaths, and unemployment rates at the county
level. May, July, August, September, and October 2020 had relatively high WTP
to reduce new COVID cases and deaths compared to other months. These monthly
107
Table 9. Descriptive Statistics, 2003 Estimating Sample vs 2020 Simulation Sample,
County-level Heterogeneity(for candidate interaction terms considered in Lasso
model, where not all interactions are retained.)
Study Sample Policy Sample
2003a 2020-2021b
mean (sd) mean (sd)
County prop. aged 0-17 0.254 (0.0289) 0.22 (0.033)
County prop. aged 18-24 0.096 (0.029) 0.086 (0.033)
County prop. aged 65+ 0.129 (0.038) 0.193 (0.046)
County prop. White 0.773 (0.168) 0.835 (0.161)
County prop. Black 0.114 (0.129) 0.091 (0.146)
County prop. Asian 0.029 (0.044) 0.013 (0.026)
County prop. Hispanic 0.105 (0.137) 0.093 (0.138)
County prop. Native American 0.008 (0.026) 0.015 (0.058)
County prop. uninsured 0.160 (0.057) 0.114 (0.050)
108
Table 9 (continued).
Study Sample Policy Sample
2003a 2020-2021b
mean (sd) mean (sd)
County fractionalization (0-1) 0.383 (0.219) 0.280 (0.196)
Rep/(Dem+Rep), Pres. Election 0.511 (0.121) 0.667 (0.161)
County Med. Income 34766.67 (9392.89) 37219 (10592.8)
Hospitals per 10000 population 0.221 (0.338) 0.56 (0.876)
County prop. college degree 0.509 (0.104) 0.524 (0.107)
County overall Poverty 0.124 (0.0433) 0.144 (5.65)
County PM2.5 11.066 (2.623) 6.59 (1.47)
County prop. Fair or Poor 0.158 (0.043) 0.179 (0.047)
Health
Avg. Num. Physically Unhealthy 3.566 (0.72) 3.99 (0.6.95)
Days
109
Table 9 (continued).
Study Sample Policy Sample
2003a 2020-2021b
mean (sd) mean (sd)
Avg. Num. Mentally Unhealthy 3.475 (0.682) 4.183 (0.594)
Days
County prop. Smoker 0.203 (0.046) 0.175 (0.035)
County prop. Obesity 0.272 (0.0404) 0.33 (0.054)
County prop. Excessive Drink 0.165 (0.04) 0.175 (0.0317)
Primary Care Physicians Rate 0.906 (0.442) 0.543 (0.034)
Preventable Hospitalization Rate 70.7 (19.4) 48.67 (18.28)
∆ unempl (Jun. ’03 vs previous 0.678 (0.408)
month)
∆ unempl (Mar. ’20 vs previous 0.467 (0.934)
month)
∆ unempl (Apr. ’20 vs previous 7.663 (4.928)
month)
110
Table 9 (continued).
Study Sample Policy Sample
2003a 2020-2021b
mean (sd) mean (sd)
∆ unempl (May ’20 vs previous -2.119 (2.451)
month)
∆ unempl (Jun. ’20 vs previous -1.887 (2.227)
month)
∆ unempl (Jul. ’20 vs previous -0.594 (1.523)
month)
∆ unempl (Aug. ’20 vs previous -1.179 (1.354)
month)
∆ unempl (Sep. ’20 vs previous -0.682 (1.258)
month)
∆ unempl (Oct. ’20 vs previous -0.649 (1.092)
month)
∆ unempl (Nov. ’20 vs previous 0.0454 (1.052)
month)
111
Table 9 (continued).
Study Sample Policy Sample
2003a 2020-2021b
mean (sd) mean (sd)
∆ unempl (Dec. ’20 vs previous 0.2093 (1.024)
month)
∆ unempl (Jan. ’21 vs previous 0.4107 (1.058)
month)
∆ unempl (Feb. ’21 vs previous -0.182 (0.593)
month)
∆ unempl (Mar. ’21 vs previous -0.375 (0.570)
month)
∆ unempl (Apr. ’21 vs previous -0.5905 (0.573)
month)
∆ unempl (May. ’21 vs previous
month)
Observations 1466 respondents 3142 counties
a Descriptive statistics, across respondents, for the counties in which they reside;
b Descriptive statistics across 3142 counties or other county FIPS geographic areas.
112
Month 03/2020 04/2020 05/2020 06/2020 07/2020
med/mean/sd med/mean/sd med/mean/sd med/mean/sd med/mean/sd
0 321.12 0 6.82 384.86
WTP(dollars) 111.19 954.69 230.93 255.78 1084.63
(483.69) (5902.04) (989.29) (1014.05) (8863.67)
Month 08/2020 09/2020 10/2020 11/2020 12/2020
med/mean/sd med/mean/sd med/mean/sd med/mean/sd med/mean/sd
113.33 164.62 204.56 309.94 342.14
WTP(dollars) 377.38 453.63 532.97 636.93 693.35
(1228.37) (2068.25) (1834.24) (2278.77) (2445.96)
Month 01/2021 02/2021 03/2021 04/2021 05/2021
med/mean/sd med/mean/sd med/mean/sd med/mean/sd med/mean/sd
327.46 232.55 189.10 180.07 163.82
WTP(dollars) 679.62 500.20 432.96 433.46 366.32
(2830.37) (2038.80) (1676.48) (1691.01) (1202.98)
Observations 3142 3142 3142 3142 3142
Table 10. County Representative Individual’s (monthly) WTP to Reduce COVID-
19 Cases and Death through 2020-21
WTP amounts were larger during the first wave of the pandemic spring and
summer. The average monthly WTP increased from March 2020 to May 2021.
4.5.3 Scaling to Monthly National Total WTP Amounts to
Avoid COVID-19 Cases and Deaths. Our benefits transfer exercise predicts
monthly WTP amounts for a representative adult in each U.S. county over the
course of the pandemic from March 2020 through February 2021. It is possible to
scale these WTP amounts to a national average for all U.S. adults by weighting
113
these county averages by the population of adults aged 18 and over in each county.
We use county populations aged 18 and over, according to the 2019 5-year ACS
estimates, to build a set of weights that sum to the overall number of counties. To
get a rough estimate of the national average WTP in each month, we multiply the
WTP point estimate for each county in that month by the corresponding weight,
sum, and divide by the number of counties to yield average WTP that can be
applied for all 251 million adults in the U.S.
The aggregate WTP across the whole adult population of the U.S. adults is
then just this national average times 251 million. These totals, by month, are, for
2020: March (123 billion), April (606 billion), May (118 billion), June (125 billion),
July (402 billion), August (104 billion), September (130 billion), October (160
billion), November (221 billion), December (253 billion); and for 2021: January
(262 billion), February (192 billion), March (174 billion), Apr (179 billion), and
May (154 billion). The cumulative U.S. national WTP, for all adults over 18
through March 2020 to May 2021, is about 3 trillion dollars.
It may be tempting to compare this aggregate WTP amount to the sizes
of the various “stimulus packages” provided during the pandemic. However, the
context for the trade-offs between policy cost and reductions in cases and deaths, in
our study sample, did not include an economic shutdown or excessive job losses or
business failures. The various stimulus packages during the pandemic were intended
to compensate for the collateral economic damages caused by the pandemic, rather
than simply to reduce cases and deaths. Our 3 trillion dollar estimate of WTP for
March 2020 through Apr 2021 should probably be interpreted as people’s net WTP
to reduce cases and deaths, after the compensation represented by the various
stimulus programs for other pandemic costs.
114
Figure 21. Aggregate WTP to Reduce the Risk of COVID-19 among all U.S.
Counties in 2020-21. The upper graph displays the total U.S. willingness-to-pay
(WTP) for reducing COVID cases and fatalities over time, while the lower graph
presents the actual cases and deaths from COVID during the same period. The
aggregated WTP generally mirrors the trends in cases and deaths, although factors
like unemployment may influence WTP for community health in specific months,
such as June 2021.
115
Figure 22. Spatial Distribution of WTP to Reduce the Risk of COVID-19 among
all U.S. Counties in December 2021. Estimations of WTP across counties vary over
time based on factors such as cases and deaths, unemployment rates, and monthly
changes in unemployment rates. Additionally, they differ spatially due to each
county’s unique sociodemographic characteristics. The spatial distribution of WTPs
reveals that counties on the west and east coasts exhibit higher values than those in
the Midwest. The WTP for reducing COVID cases and fatalities in counties shows
considerable variation.
4.5.4 Systematic Heterogeneity in Predicted WTP to Reduce
COVID-19 Cases and Deaths. We also explore a latent class model that
uses only the policy attributes in the utility-difference function, but introduce
county-level covariates for each respondent to explain each person’s probability
of preference-class membership in 2003. In our latent class model, we find three
distinct classes of people driven by different features of the preventative public
health policies. We label these three preference classes as “cost-conscious,”
“comprehensive,” and “indifferent-or-altruistic.”14 Then, we explore the systematic
heterogeneity in predicted WTP to reduce COVID-19 cases and deaths in our 2020-
21 simulation. We find that the counties with a population where the proportion
of people aged below 45 is lower than the national median have a higher WTP to
14See Latent class analysis detail in Appendix Table C2.
116
reduce the risk of COVID-19, especially when entering the winter season (Nov 2020
to Feb 2021). For different ethnic and political groups, we find that for our policy
sample in 2020-21, counties with a proportion of Black residents greater than the
median have a higher WTP to reduce the risk of COVID-19 than White counties.
For political affiliation, we find that Democrat-dominated counties have a higher
WTP through March 2020 to April 2021. For the health access level, the higher
health-access counties with the primary care physicians rates and preventable
hospitalization rates above the median rates among all counties have a higher WTP
to reduce the risk of COVID-19 through public health policies. And lastly, for
income level, the WTP of higher-income counties is higher than the lower-income
counties.
4.6 Conclusions
This paper models people’s willingness to bear the costs of public health
policies to reduce health risks to their communities. We re-purpose an existing
2003 survey of public health policy preferences, omitting the available individual-
level characteristics for the 2003 sample, and expanding the variety of county-level
characteristics employed. Almost 18 years passed between the original nationwide
survey. However, the U.S. EPA is likewise still making use of a suite of empirical
estimates of people’s willingness to trade off money for mortality risk reductions—
the so-called “value of a statistical life”—from the 1970s, 1980s, and 1990s, after
adjusting these numbers to current dollars. This suggests an implicit assumption
that people’s preferences with respect to mortality risks are highly stable over time.
We have noted several examples of stated-preference choice experiments
concerning COVID-19, conducted very early during the recent pandemic. However,
none of these contemporaneous studies has elicited such detailed data from its
survey respondents.
117
In our re-analysis of the 2003 survey data, we use a conditional logit model
with heterogeneous preferences (where variable selection is based on double
LASSO estimation) and a latent class model. In our conditional logit model
with heterogeneity in preferences, we allow for heterogeneity only with county-
level demographic characteristics and other contextual variables, rather than any
individual-specific characteristics. We first use a machine learning algorithm—
double LASSO—to winnow down all of the possible interaction terms between the
policy attributes and the county-level characteristics that are available for both
the 2003 context and the 2020 context. In our latent class model, we identify three
distinct preference classes in our sample: “cost-conscious,” “comprehensive,” and
“indifferent-or-altruistic.”
Finally, we simulate WTP amounts during the COVID-19 pandemic by
transferring our fitted model from our “study” sample in 2003 to our “policy”
sample consisting of all U.S. counties in 2020-2021. We replace the “cases
prevented” and “premature deaths prevented” attributes for the randomized public
health policies described in the original stated-preference choice experiments with
the actual county-level monthly COVID-19 cases and deaths during March 2020
through February 2021. We also update all the county-level characteristics from the
2003 era to the 2020-2021 era. We interpret predicted WTP amounts in 2020-21 as
WTP for a representative adult in every U.S. county.
Our estimated aggregate WTP across the U.S. population from March 2020
to April 2021 is about 3 trillion dollars. In April 2020, the U.S. had the highest
total WTP to reduce cases and deaths of COVID-19 because of the drastic increase
in new COVID-19 cases, deaths, and unemployment during the month. The large
aggregate WTP persisted for the rest of 2020 and started decreasing in February
118
2021 as the pandemic become more under control because of vaccinations and
stabilized unemployment numbers.
Information about the public’s willingness to bear the costs of pandemic
control will be important in the event of future pandemics. An understanding
of systematic differences in willingness to pay across counties with different
sociodemographics can potentially help county-level governments decide upon
locally appropriate and acceptable public health interventions.
119
CHAPTER V
DISSERTATION CONCLUSION
This dissertation has highlighted the significance of recycling and public
health policies through a comprehensive examination of China’s waste import
ban and its effects on the international recycling market and environmental
outcomes, and an estimation of people’s WTP for public health during a pandemic.
Despite the limited research on these policies, the study uncovers numerous policy
implications that can inform future policy decisions and interventions.
Chapter 2 of this dissertation delved into the impact of China’s waste
import ban on the U.S. environment, revealing a substantial increase in U.S.
methane emissions following the policy’s implementation. By employing a
synthetic control method, I have determined that several U.S. states, including
California, Virginia, New York, and Texas, experienced significant increases in
emissions from their waste industries. These findings demonstrate that recyclable
wastes, which were once profitable, have become an environmental burden due
to the international policy change. I have demonstrated a direct cause-and-effect
relationship between the export of waste and domestic emissions from the waste
industry. My findings show that for every metric ton of recyclable waste exported,
there is a decrease in domestic CO2 emissions by 0.89 metric tons.
Chapter 3 investigated the distributional effects of China’s waste ban on
local communities in California. I have employed a fixed effects model was utilized
to analyze waste transfers and the socioeconomic and demographic characteristics
of the involved communities. I find that, following China’s policy change, lower-
income white communities have received relatively more waste from other locations,
reducing the racial disparity associated with waste transfer. I identify land costs as
the primary mechanism for this pattern change.
120
Chapter 4 assessed the willingness of people to pay for public health policies
that benefit both themselves and others. Using survey data from 2003 and applying
a benefit transfer-function technique, we estimate that the aggregate willingness to
pay for U.S. adults to prevent COVID-19 cases and deaths during March 2020-Feb
2021 was around $3 trillion, similar to the size of COVID-19 stimulus payments
in 2020. These findings suggest that public health policies remain relevant and
valuable, with potential applications to future pandemics if people’s preferences
remain constant.
This dissertation emphasizes the crucial role of policy evaluations,
particularly in the realms of recycling and public health policies. The findings
indicate that China’s waste import ban has far-reaching environmental
consequences in the U.S. at national, state, and local levels, extending beyond its
direct trade impact. Additionally, the study highlights the continuing importance
of public health policies, as evidenced by the estimated aggregate willingness to pay
of $3 trillion to reduce COVID-19 cases and deaths, even without factoring in the
problem of a contagious cause.
121
APPENDIX A
CHAPTER 2: APPENDIX
A.1 Figures
Figure A1. Other Countries’ Recyclable Waste Exports to China and the
Rest Of the World (ROW).“Other countries” refers to 11 selected OECD
countries—Australia, Austria, Canada, France, Germany, Portugal, New Zealand,
the United Kingdom, Japan, Spain, and Finland. They all have regular trade with
China in recyclable wastes. Recyclable waste exports from other countries to China
decrease drastically by value and weight. After the GS policy, recyclable waste
exports from other countries to the rest of the world increased temporarily and
fell eventually after the GS policy. These plots show that most of the developed
countries that previously exported a substantial share of their recyclable wastes are
now dealing with these wastes on their own.
122
Figure A2. Composition of U.S. Recyclable Waste Exports to the Rest of the
World (ROW). This plot shows the composition of recyclable waste materials
exported from the U.S. to the rest of the world. Mixed paper/paperboard is still
the material that accounts for the greatest percentage of the total exports by value
and weight. Plastic scrap is the second most exported recyclable waste. Compared
to exports to China, the U.S. exported to the rest of the world lower percentages of
plastic scrap and higher percentages of metal, textile, fibers, and cotton scraps.
123
Figure A3. GHGRP: Spatial Distribution of Waste Facilities. This map shows the
locations of all landfill facilities in, the U.S., according to the EPA Greenhouse Gas
Reporting Program (GHGRP). There are more landfill facilities in states where
populations are denser.
124
Figure A4. Synthetic Control: Waste Industries and Other Industries. In plot a,
the blue line represents methane emissions for California’s waste industry; The grey
lines represent emissions from non-waste industries in California. In plot b, the blue
line again represents methane emissions from California’s waste industry, and the
grey lines represent methane emissions from waste industries in other states. These
plots show that neither “waste industries from other states” nor “other industries
within the same state” are separately the most suitable control donor for synthetic
controls.
125
Figure A5. Synthetic Control: Exports of Other Control Industries. These plots
show the net export weight by manufacturing industry. The emissions of these
manufacturing industries are used as donor groups in the synthetic control method.
The plots show no discernible changes in exports of these control manufacturing
industries after 2017, which means that the emissions from the control industries
are at least not obviously contaminated by changes in their exports.
126
Figure A6. Pairwise Correlations: Heterogeneous Effects of GS Policy on State-
level Estimations and Trade Exposures. This figure shows the correlation between
the state-level causal estimates of the GS policy and the recyclable waste export
exposure. There is a positive correlation between the percentage change in methane
emissions and recyclable waste export exposure by state.
127
Figure A7. Pairwise Correlations: Heterogeneous Effects of GS policy on State-level
Estimations and Paper Exports. This figure shows the correlation between the
causal estimates and the percentage of paper exports by each state. There is no
apparent correlation between the percentage change in methane emissions and the
percentage of paper scrap exports by state.
128
129
A.2 Tables
Total Value Total Weight Percentage Percentage
$ U.S. million million kg of total value of total weight
(1) (2) (3) (4)
Slag, dross of manufacture of iron or steel 71.9 473 0.15 0.17
Slag, ash, and residues containing metals 292.9 70.2 0.61 0.003
Plastics 15464.1 38756.4 32.4 14.23
Paper and paperboard 31521.9 232466.6 66.11 85.38
Wool and animal hair 4.8 2.3 0.05 0.00008
Wool 0.15 0.02 0.0001 0.0000008
Cotton 116.1 199.5 0.24 0.007
Fibers 150.3 256.9 0.32 0.009
Textile 51.6 51.8 0.11 0.002
Table A1. Summary Statistics: U.S. Recyclable Waste Exports (by Type of Materials). The listed waste
materials are all wastes that are directly affected by China’s GS policy. Columns (1) and (2) are the total
value and weight of recyclable waste exports from the U.S. to China from 2003 to 2020. Columns (3) and
(4) are the percentages of each waste material out of all waste materials exported.
Total Emissions Methane CO2 NO2
(1) (2) (3) (4)
Power Plants 20485.96 35.94 20481.52 76.99
Minerals 1193.32 1.22 1198.38 2.39
Waste 1118.70 1000.99 311.01 3.85
Chemicals 1053.17 1.92 991.47 64.83
Petroleum and Natural Gas Systems 985.61 88.45 896.96 0.56
Metals 815.38 793.13 1.71 0.29
Pulp and Paper 190.61 560.45 487.80 3.89
Refineries 102.77 102.07 2.87 0.26
Table A2. Summary Statistics: U.S. GHG Emissions (MMT) by Industry. Emissions
by industry are calculated by adding up the emissions from all facilities in each
industry for 2010-2020. Power plants have the largest total emissions across all
industries. The waste industry (in bold) has the highest methane emissions out of
all industries.
130
Year 2010 2011 2012 2013 2014 2015
Number of facilities 1303 1328 1342 1331 1328 1255
Total emissions (MMT.CO2e) 110.9 104.6 105.3 102.1 101.7 101
Facility emissions, mean (TMT.) 85.1 78.8 78.5 76.7 76.6 80.5
Facility emissions, sd (TMT.) 90.7 83.8 86.3 85.5 85.7 85.6
Emissions by greenhouse gas (CO2e)
Carbon dioxide (CO2) 9.9 10.7 10.8 11.1 11.1 11.4
Methane (CH4) 101.1 94.1 94.7 91.2 90.8 89.9
Nitrous oxide (N2O) 0.352 0.352 0.356 0.353 0.352 0.351
Year 2016 2017 2018 2019 2020
Number of facilities 1227 1221 1218 1204 1201
Total emissions (MMT.CO2e) 98.2 96.8 99.6 101.4 96.9
Facility emissions, mean (TMT.) 80 79.3 81.8 84.2 80.7
Facility emissions, sd (TMT.) 89 90.5 97.6 102.5 92.3
Emissions by greenhouse gas (CO2e)
Carbon dioxide (CO2) 11.7 10.6 11 10.7 10.3
Methane (CH4) 86.7 86.4 88.8 90.9 86.2
Nitrous oxide (N2O) 0.358 0.344 0.352 0.345 0.335
Sample size: 12,757
Table A3. EPA: Waste Sector - Greenhouse Gas Emissions Reported to the
GHGRP Summary Statistics. Each observation in the sample is a reporting record
from a facility from 2010 to 2020. The number of facilities has decreased gradually
over the years. However, the total and average emissions of facilities have increased
after 2017.
131
Petroleum and Pulp
Power Plants Minerals Waste Chemicals Metals Refineries
Natural Gas Systems and Paper
(1) (2) (3) (4) (5) (6) (7) (8)
Panel A. Sum over all states
2010 2295.21 100.97 110.91 104.11 65.45 90.79 43.82 27.72
2011 2136.86 100.50 104.59 83.93 94.21 82.64 16.88 17.17
2012 1995.041 104.89 105.35 80.35 96.16 78.09 16.38 6.81
2013 2006.49 108.39 102.07 85.41 93.75 77.11 15.09 6.57
2014 1997.66 113.97 101.76 89.87 96.59 77.96 15.69 6.90
2015 1874.29 112.45 101.03 93.30 98.15 69.73 15.62 6.69
2016 1771.08 108.15 98.21 99.13 76.19 69.10 14.39 6.39
2017 1696.25 111.58 96.85 100.97 79.71 69.56 13.48 6.41
2018 1710.59 113.32 99.59 104.75 90.67 72.06 13.35 6.46
2019 1577.77 112.02 101.41 106.94 97.16 69.17 12.87 6.37
2020 1424.73 107.07 96.91 104.41 97.56 59.17 13.01 5.28
Panel B. Average cross all states
2010 42.50 2.24 2.17 2.60 1.49 2.67 1.22 1.73
2011 39.57 2.28 2.05 2.09 2.00 2.36 4.97 1.07
2012 36.95 2.38 2.07 2.06 2.09 2.11 4.82 0.76
2013 37.16 2.46 2.00 2.19 2.08 2.14 4.31 0.82
2014 36.99 2.59 1.99 2.30 2.15 2.23 4.48 0.99
2015 34.71 2.56 1.98 2.39 2.18 1.99 4.34 0.96
2016 33.42 2.46 1.89 2.48 1.69 1.97 3.99 1.06
2017 32.00 2.54 1.86 2.46 1.81 2.05 3.74 1.07
2018 32.28 2.58 1.92 2.62 1.97 2.06 3.71 1.08
2019 29.77 2.55 1.95 2.74 2.16 1.98 3.58 1.27
2020 26.88 2.43 1.86 2.68 2.12 1.69 3.72 1.06
No. of Facilities 1446 364 1268 336 1225 280 143 15
Table A4. Summary Statistics: U.S. GHG Emissions by Industry. The no. of
facilities are the average numbers of facilities in each industry during 2010 to 2020.
Power plants, waste, and petroleum and natural gas are the industries that have
the most facilities in the U.S. on average from 2010 to 2020. The waste industry (in
bold) has seen a decrease in methane emissions from 2010 to 2017 and an increase
in methane emissions afterwards, both in total and on average.
132
Estimate P-value No. placebos Estimate P-value No. placebos
(1) (2) (3) (4) (5) (6)
Alabama 0.100** 0.040 24 Mississippi -0.009** 0.020 50
California 0.087* 0.052 57 South Dakota -0.063 0.500 8
Florida 0.043 0.260 49 Wyoming -0.139 0.231 39
Georgia 0.050 0.211 37 Utah -0.036 0.444 45
Hawaii 0.047 0.208 47 Maryland -0.016 0.520 57
Illinois 0.043** 0.047 42 Delaware -0.095 0.250 8
Kentucky 0.083** 0.024 40 Oklahoma -0.019 0.439 82
Louisiana 0.020 0.313 31 Connecticut -0.055 0.333 66
Missouri 0.023 0.571 6 Massachusetts -0.031 0.489 47
Montana 0.230 0.333 5 Maine -0.288 0.111 9
North Dakota 0.190* 0.100 5 Nebraska -0.084 0.258 217
New Hampshire 0.043* 0.067 29 South Carolina -0.049 0.352 105
Nevada 0.340* 0.100 9 Idaho -0.216 0.500 2
New York 0.147** 0.011 87 Pennsylvania -0.032 0.412 151
Ohio 0.060** 0.015 65 Arizona -0.078 0.288 59
Oregon 0.063 0.211 37 Michigan -0.031 0.493 73
Texas 0.083* 0.100 19 Colorado -0.089 0.222 167
Virginia 0.180* 0.919 87 Iowa -0.118 0.200 110
Washington 0.107* 0.067 15 Indiana -0.055 0.353 34
West Virginia 0.033 0.214 14 Minnesota -0.103 0.222 9
Tennessee -0.072 0.333 33 Wisconsin -0.164 0.127 110
Kansas -0.179 0.428 7 New Jersey -0.104 0.188 202
North Carolina -0.093 0.463 41
Table A5. Synthetic Control Results: Estimates at State Level. Each row (state) is
a separate synthetic control and placebo test process. The number of placebos in
each case is the number of control state-industry pairs being used in the synthetic
control process for each treatment state. Each P-value is calculated by post/pre-
Proposition 99 ratios of the MSPE for the “treatment” waste industry of a state
and all its control state-industry pairs. Post- and pre-MSPE is calculated by taking
the average of the differences between actual emissions and synthetic emissions over
the years after and before China’s GS policy. *p < 0.1, **p < 0.05, ***p < 0.01.
133
Avg. Emissions Tot. Emission Increase Avg. Emissions Tot. Emission Increase
Before GS Policy After GS Policy Before GS Policy After GS Policy
2010-2017 2018-2020 2010-2017 2018-2020
(1) (2) (3) (4)
Alabama 2.957 0.988 Mississippi 4.634 -0.0380
California 7.823 1.797 South Dakota 0.186 -0.043
Florida 8.028 1.141 Wyoming 0.146 -0.068
Georgia 4.453 0.696 Utah 0.667 -0.071
Hawaii 0.561 0.078 Maryland 1.659 -0.079
Illinois 3.534 0.557 Delaware 0.213 -0.086
Kentucky 1.899 0.485 Oklahoma 1.982 -0.112
Louisiana 2.056 0.164 Connecticut 0.926 -0.148
Missouri 1.381 0.096 Massachusetts 1.599 -0.150
Montana 0.339 0.196 Maine 0.301 -0.204
North Dakota 0.294 0.206 Nebraska 0.913 -0.232
New Hampshire 0.390 0.049 South Carolina 1.579 -0.233
Nevada 0.281 0.428 Idaho 0.349 -0.263
New York 3.132 1.370 Pennsylvania 3.701 -0.331
Ohio 5.057 0.845 Arizona 1.430 -0.372
Oregon 1.035 0.194 Michigan 4.634 -0.390
Texas 10.297 2.702 Colorado 1.382 -0.395
Virginia 3.433 1.996 Iowa 1.148 -0.406
Washington 0.972 0.353 Indiana 0.349 -0.442
West Virginia 0.731 0.067 Minnesota 1.464 -0.493
Tennessee 2.485 -0.492 Wisconsin 1.406 -0.627
Kansas 1.666 -0.638 New Jersey 2.234 -0.679
North Carolina 3.501 -1.021
Table A6. Synthetic Control Results: Emission Increases across States (million
metric tons of CO2 eq.). Average emissions before the GS policy are calculated by
taking the mean of total emissions of each state over the years 2010-2017. The total
increase in emissions after the GS policy is calculated by summing up the emission
increase in each year from 2018 to 2020. The larger states, such as California,
Texas, and New York, have seen a greater increase in methane emissions from the
waste industry after China’s GS policy.
134
APPENDIX B
CHAPTER 3: APPENDIX
B.1 Figures
Figure B1. Data Comparison: EPA GHGRP v.s. CalRecycle RDRS. To carry
the result from the state-level estimate, I compare the two data sources for
state and facility-level analysis. This plot shows that the emissions data in EPA
GHGRP and the data for tons of disposal in CalRecycle are highly correlated. This
correlation demonstrates that the result I find from the state-level analysis—i.e.
that California has seen an increase in methane emissions from the waste industry
after China’s GS policy—can be used in an analysis of facility-level distributional
effects in California. Given that California has seen an overall increase in emissions
and pollution from the waste industry due to China’s GS policy, I will estimate how
local communities have been affected differently by this policy change.
135
Figure B2. CalRecycle: Recycling and Disposal Reporting System (RDRS) Facility
Locations in California by Rural and Urban Areas. This map shows the location of
all landfill facilities in the CalRecycle RDRS data. The yellow areas are the urban
areas in California. The map shows that most landfill facilities are located in rural
regions or suburbs outside the urban areas.
136
Figure B3. Disposal Flow Map by Median Income. The median income is mapped
using 2013 ACS 5-year data at the census block group level. The colors of the
arrows show the increase in amount of disposal flows after China’s GS policy.
137
Figure B4. Disposal Flow Map by Environmental Vulnerability. Environmental
vulnerability is calculated by the Office of Environmental Health Hazard
Assessment (OEHHA). California Communities Environmental Health Screening
Tool is a screening methodology that evaluates multiple pollution sources and
stressors and measures a community’s vulnerability to pollution. The higher the
score, the more vulnerable is the community to pollution. The colors of the arrows
show the increase in amount of disposal flows after China’s GS policy.
138
Figure B5. CalRecycle: Recycling and Disposal Reporting System (RDRS) Facility
Locations in California by Racial Composition. This map shows the location
of all landfill facilities in the CalRecycle RDRS data by racial composition.
Racial composition is plotted by census block group level. The map shows that
most destination facilities are located in the darker areas where more minority
population resides. However, some facilities are still located in the lighter areas
where the white population lives.
139
Figure B6. Racial Composition Variation within Los Angeles County. This map
shows the racial composition variations across different communities (the black
circle: within 3 km buffers of the destination facilities) in Los Angeles county.
In the fixed-effects model estimation, the racial composition variable is at the
census block-level, and is time-invariant (as of 2010). Thus, when I add county
fixed effects, the variation in racial composition is the different communities within
a county. Some communities may include multiple census blocks.
140
Figure B7. Racial Composition Variation within Santa Clara County. This map
shows the racial composition variations across different communities (the black
circle: within 3 km buffers of the destination facilities) in Santa Clara county.
In the fixed-effects model estimation, the racial composition variable is at the
census block-level, and is time-invariant (as of 2010). Thus, when I add county
fixed effects, the variation in racial composition is the different communities within
a county. Some communities may include multiple census blocks.
141
Figure B8. Voting Variation within Los Angeles County. Voting Variation within
County. Similar to the racial composition, vote shares of communities (black circle:
with 3km buffers of the destination facilities) are also time-invariant (as of 2016) at
the precinct level. Thus, when I add county fixed effects, the variation for voting
registration share is the different communities within a county. Some communities
may include multiple voting precincts.
142
Figure B9. Voting Variation within Santa Clara County. Voting Variation within
County. Similar to the racial composition, vote shares of communities (black circle:
with 3km buffers of the destination facilities) are also time-invariant (as of 2016) at
the precinct level. Thus, when I add county fixed effects, the variation for voting
registration share is the different communities within a county. Some communities
may include multiple voting precincts.
143
Figure B10. Economies of Scale of Communities Where Destination Facilities are
Located. This figure shows how economies of scale are defined. The map shows
the number of facilities that are within a 15 km buffer of the destination facility
of disposal shipment. The red dot is the destination facility from CalRecycle
as a destination for disposal transfer. The blue marks are other types of related
facilities within a 15km buffer. They are composts (CO), landfills (LF), recycling
centers (MR and MW), and transfer stations (TS). The more facilities around
the destination landfill facility, the higher the economy of scale there is in the
community where the destination landfill is located.
144
Figure B11. This figure shows how economies of scale are defined. The map shows
the number of facilities that are within a 15 km buffer of the destination facility
of disposal shipment. The red dot is the destination facility from CalRecycle
as a destination for disposal transfer. The blue marks are other types of related
facilities within a 15km buffer. They are composts (CO), landfills (LF), recycling
centers (MR and MW), and transfer stations (TS). The more facilities around
the destination landfill facility, the higher the economy of scale there is in the
community where the destination landfill is located.
145
B.2 Tables
Origin Jurisdiction Destination Facility Distance
(1) (2) (3) (4) (5)
mean mean
shipments shipments
Year no.jurisdictions no.facilities mean (km)
sent received
(1000 tons) (1000 tons)
2002 434 86.6 162 232.0 93.3
(214.6) (476.8) (121.6)
2003 421 94.4 158 251.6 93.9
(271.5) (501.4) (121.0)
2004 424 96.2 152 268.3 96.1.29
(269.5) (522.9) (122.2)
2005 419 100.3 149 281.9 94.6
(283.9) (534.5) (119.5)
2006 412 99.5 148 276.9 89.8
(271.9) (517.2) (109.1)
2007 414 93.6 142 272.9 92.2
(261.8) (507.3) (110.7)
2008 417 84.2 133 264.0 89.4
(235.0) (465.9) (103.0)
2009 412 74.7 134 229.7 98.7
(211.2) (431.2) (120.4)
2010 417 72.0 131 229.4 101.7
(201.1) (414.6) (123.6)
2011 416 71.5 134 221.9 76.4
(204.2) (408.5) (90.9)
2012 414 70.3 131 222.1 71.6
(203.9) (405.2) (71.5)
2013 412 72.7 133 225.2 85.8
(214.1) (405.6) (101.6)
2014 411 75.1 130 237.5 87.1
(235.4) (427.2) (99.1)
2015 410 80.3 128 257.2 89.2
(250.8) (456.1) (104.9)
2016 420 82.9 126 276.3 90.1
(257.4) (473.1) (102.8)
2017 420 89.2 127 294.9 90.3
(275.3) (501.5) (103.4)
2018 417 94.7 128 308.5 90.3
(284.8) (519.3) (98.8)
2019 418 96.5 127 317.6 87.0
(288.9) (532.9) (90.2)
2020 419 96.2 128 314.8 87.8
(285.3) (528.8) (83.7)
Sample Size 281339
Table B1. CalRecycle: Recycling and Disposal Reporting System (RDRS)
Disposal Flows within California, Summary Statistics (Thousands of Tons). Each
observation in the sample is a waste shipment from an origin jurisdiction to a
destination facility during 2002-2020. The average distance is calculated by taking
the mean of origin-destination pairs in each year.
146
3 km Buffer 5 km Buffer 10 km Buffer
% White Population 57.12 53.67 52.37
(27.07) (25.91) (24.01)
% Black Population 2.78 3.24 4.07
(4.98) (4.83) (4.93)
% Hispanic Population 32.79 35.19 35.50
(25.65) (24.91) (22.48)
Median Income 63.156 61.137 59.921
($Thousand) (24.616) (21.974) (20.503)
5 km Buffer 10 km Buffer 15 km Buffer
Economies of Scale 1.96 4.04 6.42
(no. of facilities) (1.79) (3.95) (6.39)
# of destination facilities 264
Table B2. Summary Statistics for Community Characteristics around
each Destination Facility. # of facilities are all destination facilities
from 2002 to 2020 in the CalRecycle RDRS data. Racial composition is
from the U.S. Census decennial data 2010 at census block level. Median
income is from the U.S. Census American Community Survey (ACS)
5-year survey in 2013 at census block group level. Economies of scale
are calculated using Waste Business Journal (WBJ) data at the facility
level. See Appendix, Figure 4, for a detailed definition of economies
of scale. I choose 15km distance buffers for economies of scale since
there are normally no other facilities within a 3 km buffer of destination
landfill facilities.
147
Dep.Variable: Disposal shipments received (tons) (1) (2) (3) (4)
Distance (log km) -0.300∗∗∗ -0.233∗∗∗ -0.216∗∗∗ -0.029
(0.078) (0.078) (0.084) (0.113)
Distance (log)×1(post) 0.070∗ 0.068 0.061 0.059
(0.041) (0.048) (0.045) (0.048)
White share (log %) -0.459∗∗∗ -0.620∗∗∗ -0.599∗∗∗ -1.197∗∗∗
(0.184) (0.027) (0.183) (0.163)
White share (log %)×1(post) 0.269∗ 0.267∗ 0.271∗ 0.496∗∗
(0.144) (0.161) (0.156) (0.195)
Black share (log %) 0.152∗∗∗ 0.178∗∗∗ 0.200∗∗∗ 0.340∗∗∗
(0.047) (0.063) (0.078) (0.091)
Black share (log %)×1(post) 0.069 0.092 0.083∗ 0.084∗∗
(0.045) (0.057) (0.049) (0.039)
Hispanic share (log %) -0.315 -0.203 -0.204 -0.635
(0.214) (0.211) (0.199) (0.111)
Hispanic share (log %)×1(post) -0.061∗∗∗ -0.065∗ -0.044 -0.072
(0.022) (0.028) (0.032) (0.085)
Median income (log $) 1.702∗∗∗ 1.806∗∗∗ 1.969∗∗∗
(0.279) (0.352) (0.351)
Median income (log $)×1(post) -0.097 -0.165∗∗ -0.151∗∗∗
(0.062) (0.069) (0.048)
Economies of scale 0.121 0.531∗∗∗
(0.161) (0.184)
Economies of scale×1(post) -0.110 -0.354∗∗∗
(0.074) (0.089)
Republican votes (log %) 1.000∗∗∗
(0.344)
Republican votes (log %)×1(post) -0.667∗∗
(0.300)
County FE ✓ ✓ ✓ ✓
Year FE ✓ ✓ ✓ ✓
Quarter FE ✓ ✓ ✓ ✓
Observations 226,128 226,128 226,188 217,212
Table B3. Altered Distributional Effects: Fixed-effects Model for Waste Flows
from Origin Jurisdiction to Receiving Facilities and their Local Communities,
before/after China’s GS Policy. Economies of scale are the numbers of related
facilities, such as transfer station, landfills, and recycling centers, that are within
5 km buffers of the destination facilities in CalRecycle RDRS data. Data for
Republican votes is from California Statewide Database (SWDB) election data at
the precinct level. It is defined as the percentage of the population that registered
as Republicans in the 2016 election year among all voting registrations.
148
APPENDIX C
CHAPTER 4: APPENDIX
C.1 Figures
Figure C1. One Example of a Choice Set in the Original 2003 Survey. Each
respondent faces a choice between either of two different health policies and the
status quo.
149
C.2 Tables
Table C1. Sources of county-level data for estimation and simulation
Variable Sources for Sources for
ca. 2003 data ca. 2020 data
(estimation) (simulation)
County population 2000 Census 2018 5-yr ACS
Population affected ”People living 1.0 (i.e. county
around you” in population as a
choice scenario, as proportion of county
a proportion of the population)
population in the
respondent’s county
Median household income 2000 Census STF3 2018 5-yr ACS
Table P53, P053001
Unemployment rate, county-level, BLS monthly for for Feb-May for 2020
current month May, June 2003
150
Table C1 (continued).
Variable Sources for Sources for
ca. 2003 data ca. 2020 data
(estimation) (simulation)
Change in unemployment rate BLS May-June 2003 Feb-Mar, Mar-Apr,
since last month, county level Apr-May, May-June
for 2020
Ethnic mix. Proportion of county 2000 Census 2018 5-yr ACS
population: pblack, pasian,
phispanic, pother
Ethnic fractionalization. For 7 calculated from 2000 calculated from 2018
racial groups: white, black, asian, Census 5-yr ACS
amerind, hawaii-pacisl, other,
multi-race
Age distribution. Proportions 2000 Census 2018 5-yr ACS
of population in each age group:
0-17, 18-24, 65 plus
151
Table C1 (continued).
Variable Sources for Sources for
ca. 2003 data ca. 2020 data
(estimation) (simulation)
Last Presidential election vote David Leip’s US David Leip’s US
shares: Democratic, Republican, Election atlas for Election atlas for
Green, Libertarian, Other 2000 2016
Hospitals per 100,000 population. CDC Open Data same
Health insurance coverage, (US Census same
county level Bureau, 2008 -
2018 Small Area
Health Insurance
Estimates (SAHIE)
using the American
Community Survey
(ACS)a )
152
Table C1 (continued).
Variable Sources for Sources for
ca. 2003 data ca. 2020 data
(estimation) (simulation)
Air quality1 Van Donkelaaret al. same
Health indicators2 Robert Wood same
Johnson Foundation
Program County
Health Rankings
and Roadmaps
1Given that COVID-19 is primarily a respiratory disease, baseline airquality may be important.
We have only one environmental variable in this research– particular matter (PM2.5). PM2.5
pollution consist of tiny particles in the air ofdiameter less than 2.5 micrometres. These particles
of dust or soot can be inhaled andhave the potential to cause long-term health problems.
2The variableswe employ in our re-estimation and simulation models include the percentage of
adultswho report smoking, obesity, and excessive drinking. For seniors in each county, we alsouse
clinical care data for “access to care”, and “quality of care” from Medicare claims.
153
Table C2. Homogeneous Preferences versus Model with Three Latent Classes
Description Three Latent Classes of Preferences
Class 1 Class 2 Class 3
Homog. Cost- Compre- Indiff./
Pref. Consc. hensive Altruist.
Latent classes of preferences (marginal utilities of policy attributes)
Monthly cost of policy -0.004*** -0.01*** -0.003* 0.1
Policy duration -0.01*** -0.02** -0.01*** -0.5
Base. cases of illness - - - 0.028
Reduction in cases 0.00008*** - 0.0001** 0.06
Base. prem. deaths - - 0.0003* 0.022
Reduction in deaths 0.000098. - 0.0004 0.33
Private benefit 0.6*** -0.53*** 2.1*** 0.74
Status quo alternative 1.54*** -0.77** 3.9*** 13.53
Class membership propensities (relative to Class 1)
154
Table C2 (continued).
Description Three Latent Classes of Preferences
Class 1 Class 2 Class 3
Homog. Cost- Compre- Indiff./
Pref. Consc. hensive Altruist.
log(County population) n/a 0 1.91***
Cnty pr. Repub. vote n/a 0 1.18* 3.49*
Proportions of county population in different age brackets
% pop. age 0-17 n/a 0
% pop. age 25-44 n/a 0 -8.31* -22.52*
% pop. age 45-64 n/a 0 9.36* 20.39*
% pop. age 65-84 n/a 0 -8.15*
Proportions of county population in different racial/ethnic groups
% pop. White n/a 0 -5.86*
155
Table C2 (continued).
Description Three Latent Classes of Preferences
Class 1 Class 2 Class 3
Homog. Cost- Compre- Indiff./
Pref. Consc. hensive Altruist.
% pop. Black n/a 0
% pop. Native Amer. n/a 0 -27.57***
% pop. Asian n/a 0 -10.89** -37.32*
% pop. multi-race n/a 0
% pop. Hispanic n/a 0
log(Med. income/100K) n/a 0 -1.69***
Hospitals per 10K pop. n/a 0
County unempl (current) n/a 0
∆ unempl (v last month) n/a 0
Health insurance coverage n/a 0 0.13.
156
Table C2 (continued).
Description Three Latent Classes of Preferences
Class 1 Class 2 Class 3
Homog. Cost- Compre- Indiff./
Pref. Consc. hensive Altruist.
% adults completing college n/a 0 8.05*
or bachelor degree
Poverty percent, all ages n/a 0 -0.089**
Average PM2.5 n/a 0
% Fair or Poor Health n/a 0
Average Number of n/a 0
Physically Unhealthy Days
Average Number of Mentally n/a 0
Unhealthy Days
% Smokers n/a 0
% Adults with Obesity n/a 0
157
Table C2 (continued).
Description Three Latent Classes of Preferences
Class 1 Class 2 Class 3
Homog. Cost- Compre- Indiff./
Pref. Consc. hensive Altruist.
% Excessive Drinking n/a 0 -0.032 0.23***
Prim. Care Physic. Rate n/a 0 -0.003
Preventable Hosp. Rate n/a 0
Observ. 1466 1466 1466
158
REFERENCES CITED
Aadland, D., & Caplan, A. J. (2006). Curbside Recycling: Waste Resource or
Waste of Resources? Journal of Policy Analysis and Management , 25 (4),
855–74.
Abadie, A., Diamond, A., & Hainmueller, J. (2012). Synthetic Control Methods for
Comparative Case Studies: Estimating the Effect of California’s Tobacco
Control Program. Journal of the American Statistical Association, 105 (490),
493-505.
Abel, T. D., & White, J. (2011). Skewed Riskscapes and Gentrified Inequities:
Environmental Exposure Disparities in Seattle, Washington. American
Journal of Public Health, 101 , 246-254.
Abernethy, S., O’Connor, F., Jones, C., & Jackson, R. (2021). Methane removal
and the proportional reductions in surface temperature and ozone.
Philosophical Transactions of the Royal Society(379).
Adeel, B. A., Catalano, M., Catalano, O., Gibson, G., Muftuoglu, E., Riggs, T.,
Sezgin, H. M., Shvetsova, O., Tahir, N., VanDusky, J., Zhao, T., & Zhirnov,
A. (2020). COVID-19 Policy Response and the Rise of the Sub-National
Governments. Canadian Public Policy , 46 , 565-584. doi:
10.3138/cpp.2020-101
Aizer, A., Currie, J., Simon, P., & Vivier, P. (2018). Do low levels of blood lead
reduce children’s future test scores? American Economic Journal: Applied
Economics , 10 (1), 307-41.
Autor, D., Dorn, D., & Hanson, G. (2013). The China Syndrome: Local Labor
Market Effects of Import Competition in the United States. America
Economic Review , 103(6), 2121-2168.
Baden, B. M., & Coursey, D. L. (2002). The Locality of Waste Sites within the
City of Chicago: a Demographic, Social, and Economic Analysis. Resource
and Energy Economics , 24 (1-2), 53-93.
Banzhaf, S., Ma, L., & Timmins, C. (2019). Environmental Justice: The
Economics of Race, Place, and Pollution. Journal of Economic Perspectives ,
33 (1), 185-208.
Banzhaf, S. H., & Walsh, R. P. (1994). Environmental Equity: the Demographics
of Dumping. Demography , 31 (2), 229-248.
159
Banzhaf, S. H., & Walsh, R. P. (2008). Do People Vote with Their Feet? An
Empirical Test of Tiebout’s Mechanism. American Economic Review , 93 (3),
843-63.
Banzhaf, S. H., & Walsh, R. P. (2013). Segregation and Tiebout Sorting: The Link
between Place-based Investments and Neighborhood Tipping. Journal of
Urban Economics , 74 , 83-98.
Bartik, T. J. (1991). Who Benefits from State and Local Economic Development
Policies? W.E. Upjohn Institute for Employment Research.
Becker, R., & Henderson, V. (2000). Effects of Air Quality Regulations on
Polluting Industries. Journal of Political Economy , 2 (108), 379-421.
Black, S. E., Devereux, P. J., & Salvanes, K. G. (2007). From the cradle to the
labor market? the effect of birth weight on adult outcomes. Quarterly
Journal of Economics , 122 (1), 409-39.
Bohm, R., Folz, D. H., Kinnaman, T. C., & Podolsky, M. J. (2010). The Costs of
Municipal Waste and Recycling Programs. Resources, Conservation and
Recycling , 54 (8), 64–71.
Bosworth, R., Cameron, T. A., & DeShazo, J. R. (2009). Demand for
environmental policies to improve health: Evaluating community-level policy
scenarios. Journal of Environmental Economics and Management , 57 (3),
293–308. doi: 10.1016/j.jeem.2008.07.009
Brander, L., Beukering, P. V., & Cesar, H. S. (2007). The recreational value of
coral reefs: A meta-analysis. Land Economics , 63 , 209-218. doi:
10.1016/j.ecolecon.2006.11.002
Brandt, L., Biesebroeck, J. V., & Zhang, Y. (2012). Creative Accounting or
Creative Destruction? Firm-level Productivity Growth in Chinese
Manufacturing. Journal of Development Economics , 2 (97), 339-351.
Bransetter, L., & Lardy, N. (2006). China’s Embrace of Globalization. NBER
Working Paper(12373).
Bullard, R. D. (1983). Solid waste sites and the black houston community.
Sociological Inquiry , 53 (2-3), 273-88.
Cameron, T. A. (2010). Euthanizing the value of a statistical life. Review of
Environmental Economics and Policy , 4 (2), 161–178. doi:
10.1093/reep/req010
Cameron, T. A., & McConnaha, I. T. (2006). Evidence of Environmental
Migration. Land Economics , 82 (2), 273-90.
160
Cattapan, A., Ackerverney, J. M., Dobrowolsky, A., Findlay, T., & Mandrona, A.
(2020). Community Engagement in a Time of Confinement. Canadian
Public Policy , 14 , 288-298. doi: IO.3138/cpp.2020-064
Chavis, B. F., & Lee, C. (1987). Toxic Wastes and Race in the United States: A
National Report on the Racial and Socio-Economic Characteristics of
Communities with Hazardous Waste Sites. New York: Commission for
Racial Justice, United Church of Christ..
Chay, K. Y., & Greenstone, M. (2003). The impact of air pollution on infant
mortality: Evidence from geographic variation in pollution shocks induced
by a recession. Quarterly Journal of Economics , 118 (3), 1121-67.
Cherniwchan, J. (2017). Trade liberalization and the environment: Evidence from
NAFTA and U.S. manufacturing. Journal of International Economics ,
C (105), 130-149.
Chorus, C., Sandorf, E. D., & Mouter, N. (2020). Diabolical dilemmas of
COVID-19: An empirical study into Dutch society’s trade-offs between
health impacts and other effects of the lockdown. PLoS ONE , 15 , 1-19. doi:
10.1371/journal.pone.0238683
Cook, A. R., Zhao, X., Chen, M. I., & Finkelstein, E. A. (2018). Public preferences
for interventions to prevent emerging infectious disease threats: A discrete
choice experiment. BMJ Open, 8 , 1-12. doi: 10.1136/bmjopen-2017-017355
Copeland, B., Shapiro, J., & Taylor, S. (1994). North-south Trade and the
Environment. The Quarterly Journal of Economics , 3 (109), 755-87.
Copeland, B., Shapiro, J., & Taylor, S. (2021). Globalization and the Environment.
NBER Working Papers(28797).
Currie, J., Davis, L., Greenstone, M., & Walker, R. (2015). Environmental health
risks and housing values: Evidence from 1,600 toxic plant openings and
closings. American Economic Review , 105 (2), 678-709.
Currie, J., Greenstone, M., & Moretti, E. (2011). Superfund cleanups and infant
health. American Economic Review , 101 (3), 435-41.
Currie, J., & Moretti, E. (2007). Biology as destiny? short-and long-run determi-
nants of intergenerational transmission of birth weight. Journal of Labor
Economics , 25 (2), 231-64.
Currie, J., & Neidell, M. (2007). Getting inside the ‘black box’ of head start
quality: What matters and what doesn’t. Economics of Education Review ,
26 (1), 83-99.
161
Currie, J., Voorheis, J., & Walker, R. (2023). What Caused Racial Disparities in
Particulate Exposure to Fall? New Evidence from the Clean Air Act and
Satellite-Based Measures of Air Quality. American Economic Review , 113 ,
71-97.
Currie, J., & Walker, R. (2011). Traffic congestion and infant health: Evidence
from e-zpass. American Economic Journal: Applied Economics , 3 (1), 65-90.
Depro, B., Timmins, C., & O’Neil, M. (2011). Hazardous Waste Cleanup,
Neighborhood Gentrification, and Environmental Justice: Evidence from
Restricted Access Census Block Data. American Economic Review , 101 (3),
620-24.
Depro, B., Timmins, C., & O’Neil, M. (2015). White Flight and Coming to the
Nuisance: Can Residential Mobility Explain Environmental Injustice?
Journal of the Association of Environmental and Resource Economists , 2 (3),
439-468.
Dorantes, C. A., Kaushal, N., & Muchow, A. N. (2020). Is the Cure Worse than
the Disease? County-Level Evidence from the COVID-19 Pandemic in the
United States. NBER Working Paper No. 27759 , 1-35. Retrieved from
https://www.nber.org/papers/w27759 doi: 10.3386/w27759
Echazu, L., & Nocetti, D. C. (2020). Willingness to pay for morbidity and
mortality risk reductions during an epidemic. Theory and preliminary
evidence from COVID-19. GENEVA Risk and Insurance Review , 45 (2),
114–133. doi: 10.1057/s10713-020-00053-0
Figlio, D., Guryan, J., Karbownik, K., & Roth, J. (2014). The effects of poor
neonatal health on children’s cognitive development. American Economic
Review , 104 (12), 3921-55.
GAO. (1983). Siting of Hazardous Waste Landfills and their Correlation with
Racial and Economic Status of Surrounding Communities. Washington, DC:
General Accounting Office (GAO).
Greenstone, M. (2002). The Impacts of Environmental Regulations on Industrial
Activity: Evidence from the 1970 and 1977 Clean Air Act Amendments and
the Census of Manufactures. Journal of Political Economy , 6 (110),
1175-1219.
Greenstone, M., He, G., Li, S., & Zou, E. (2021). China’s War on Pollution:
Evidence from the First Five Years. Review of Environmental Economics
and Policy , 2 (15), 281-299.
162
Gregson, N., & Crang, M. (2015). From waste to resource: The trade in wastes and
global recycling economies. Annual Review of Environment and Resources ,
1 (40), 151-176.
Henderson, V. J. (1996). Effects of Air Quality Regulation. American Economic
Review , 4 (84), 789-813.
Hernandez-Cortes, D., & Meng, K. C. (2020). Do Environmental Markets Cause
Environmental Injustice? Evidence from California’s Carbon Market. NBER
Working Paper(27205).
Higashida, K., & Managi, S. (2014). Determinants of Trade in Recyclable Wastes:
Evidence from Commodity-based Trade of Waste and Scrap. Environment
and Development Economics , 19 (2), 250-270.
Hipp, J. R., & Lakon, C. M. (2010). Bringing Satellite-Based Air Quality
Estimates Down to Earth. Health and Place, 16 , 674-683.
Ho, P. (2023). The Costs and Environmental Justice Concerns of NIMBY in Solid
Waste Disposal. Journal of the Association of Environmental and Resource
Economists , 3 (10), 607-654.
Hoek, G., Krishnan, R. M., Beelen, R., Peters, A., Ostro, B., Brunekreef, B., &
Kaufman, J. D. (2013). Long-term air pollution exposure and
cardio-respiratory mortality: A review. Environmental Health, 12 (43).
Johnston, R. J., Boyle, K. J., Vic Adamowicz, W., Bennett, J., Brouwer, R., Ann
Cameron, T., Michael Hanemann, W., Hanley, N., Ryan, M., Scarpa, R.,
Tourangeau, R., & Vossler, C. A. (2017). Contemporary guidance for stated
preference studies. Journal of the Association of Environmental and
Resource Economists , 4 (2), 319–405. doi: 10.1086/691697
Kellenberg, D. (2012). Trading Wastes. Journal of Environmental Economics and
Management , 1 (64), 68-87.
Kellenberg, D. (2015). The Economics of the International Trade of Waste. Annual
Review of Resource Economics , 1 (7), 109-125.
Kinnaman, T. C. (2014). Effects of Norms and Opportunity Cost of Time on
Household Recycling. Resources, Conservation and Recycling , 85 , 5–10.
Kinnaman, T. C., Shinkuma, T., & Yamamoto, M. (2014). The Socially Optimal
Recycling Rate: Evidence from Japan. Journal of Environmental Economics
and Management , 68 , 54–70.
163
Lee, J., Wei, S.-J., & Xu, J. (2020). The welfare Cost of a Current Account
Imbalance: A “Clean” Effect. Research Collection School Of Economics ,
1-57.
Li, J., & Takeuchi, K. (2021). Import Ban and Clean Air: Estimating the Effect of
China’s Waste Import Ban on the Ozone Pollution. Kobe University
Discussion Paper .
Li, L., Long, D., & Rad, M. R. (2020). Stay-at-Home Orders and the Willingness
to Stay Home During the COVID-19 Pandemic: A Stated-Preference
Discrete Choice Experiment. SSRN .
Lincoln, K. D., Abdou, C. M., & Lloyd, D. (2014). Race and socioeconomic
differences in obesity and depression among black and non-hispanic White
Americans. Journal of Health Care for the Poor and Underserved , 25 (1),
257–275. doi: 10.1353/hpu.2014.0038
Lindhjelm, H., & Navrud, S. (2008). How Reliable Are Meta-Analyses for
International Benefit Transfer? Ecological Economics , 66 , 425-435. doi:
10.1016/j.ecolecon.2007.10.005
Miles, D., Stedman, M., & Heald, A. (2020). LIVING with COVID-19:
BALANCING COSTS against BENEFITS in the FACE of the VIRUS.
National Institute Economic Review , 253 , 60-76. doi: 10.1017/nie.2020.30
Morehouse, J., & Rubin, E. (2021). Downwind and out: The Strategic Dispersion
of Power Plants and their Pollution. Working paper .
Mühlbacher, A., & Johnson, R. F. (2016). Choice Experiments to Quantify
Preferences for Health and Healthcare: State of the Practice. Applied Health
Economics and Health Policy , 14 , 253-266. doi: 10.1007/s40258-016-0232-7
Olivia, G. (2014). Determinants of European Fright Rates: The Role of Market
Power and Trade Imbalance. Transportation Research Part E: Logistics and
Transportation Review(62), 23-33.
Oreopoulos, P., Stabile, M., Walld, R., & Roos, L. L. (2008). Short-, medium-, and
long-term consequences of poor infant health, an analysis using siblings and
twins. Journal of Human Resources , 43 (1), 88-138.
Pais, J., Crowder, K., & Downey, L. (2013). Unequal Trajectories: Racial and
Class Differ- ences in Residential Exposure to Industrial Hazard. Social
Forces , 92 , 1189-1215.
Palma, A., Lindsey, R., Quinet, E., & Vickerman, R. (2011). China’s Embrace of
Globalization. A Handbook of Transport Economics .
164
Persico, C., Figlio, D., & Roth, J. (2016). Inequality before birth: The
developmental consequences of environmental toxicants. NBER Working
Paper , 22263 .
Price, J., Dupont, D., & Adamowicz, W. (2017). Willingness to accept tradeoffs
among Covid-19 cases, social-distancing restrictions, and economic impact:
A nationwide US study. Environmental and Resource Economics , 68 ,
643-662. doi: 10.1007/s10640-016-0039-x
Ready, R., & Navrud, S. (2006). International benefit transfer: Methods and
validity tests. Ecological Economics , 60 , 429-434. doi:
10.1016/j.ecolecon.2006.05.008
Reed, S., Gonzalez, J. M., & Johnson, F. R. (2020). Willingness to accept tradeoffs
among Covid-19 cases, social-distancing restrictions, and economic impact:
A nationwide US study. medRxiv , 11 , 1438-1443. doi:
10.1101/2020.06.01.20119180
Rees-Jones, A., Attoma, J., Piolatto, A., & Salvadori, L. (2020). COVID-19
Changed Tastes for Safety-Net Programs. NBER Working Paper No. 27865 ,
1-20. Retrieved from http://www.nber.org/papers/w27865.pdf
Repa, E. W. (2005). Interstate movement of municipal solid waste. bulletin report.
National Solid Waste Management Association..
Richardson, L., Loomis, J., Kroeger, T., & Casey, F. (2015). The role of benefit
transfer in ecosystem service valuation. Ecological Economics , 115 , 51-58.
doi: 10.1016/j.ecolecon.2014.02.018
Royer, S., Ferron, S., Wilson, S., & Karl, D. (2018). Production of Methane and
Ethylene from Plastic in the Environment. PLOS ONE .
Sanders, N. J. (2012). What doesn’t kill you makes you weaker: Prenatal pollution
exposure and educational outcomes. Journal of Human Resources , 47 (3),
826-50.
Schlenker, W., & Walker, R. W. (2016). Airports, air pollution, and
contemporaneous health. Review of Economic Studies , 83 (2), 768-809.
Shapiro, J. S. (2016). Trade Costs, CO2, and the Environment. American
Economic Journal: Economic Policy , 8 (4), 220-54.
Shapiro, J. S. (2021). The Environmental Bias of Trade Policy. Quarterly Journal
of Economics , 136 (2), 831-886.
165
Shapiro, J. S., & Walker, R. (2021). Where is Pollution Moving? Environmental
Markets and Environmental Justice. AEA Papers and Proceedings , 111 ,
410-14.
Shapiro, R., Joseph S.and Walker. (2018). Why is Pollution from U.S.
Manufacturing Declining? The Roles of Environmental Regulation,
Productivity, and Trade. American Economic Review , 108 (12), 3814-54.
Smith, V. K., Van Houtven, G., & Pattanayak, S. K. (2002). Benefit transfer via
preference calibration: ”Prudential Algebra” for policy. Land Economics ,
78 , 132-152. doi: 10.1007/s10640-016-0039-x
Tanaka, S., Teshima, K., & Verhoogen, E. (2021). North-South Displacement
Effects of Environmental Regulation: The Case of Battery Recycling.
American Economics Journal: Insights (Forthcoming).
Unfried, K., & Wang, F. (2022). Importing Air Pollution? Evidence from China’s
Plastic Waste Import. IZA Institute of Labor Economics(15218).
Viscusi, W. K. (2020). Pricing the global health risks of the COVID-19 pandemic.
Journal of Risk and Uncertainty , 61 , 101-128. doi:
10.1007/s11166-020-09337-2
Wong, W. (2021). The Round Trip Effect: Endogenous Transport Costs and
International Trade. American Economics Journal: Applied Economics
(Forthcoming).
Yancy, C. W. (2020). COVID-19 and African Americans. JAMA - Journal of the
American Medical Association, 323 (19), 1891–1892. doi:
10.1001/jama.2020.6548
166