Measuring the Effects of Environmental Regulations: The Critical Importance of a Spatially Disaggregated Analysis

Abstract

We examine the effects of the 1990 Clean Air Act Amendments (CAAA) on ambient concentrations of PM10 in the United States between 1990 and 2005. Consistent with prior literature, we find that non-attainment designation has no effect on the average monitor in non-attainment counties, after controlling for weather, socioeconomic characteristics at the county level and lagged concentrations. In sharp contrast, if we allow for heterogeneous treatment by type of monitor and county, we do find that the 1990 CAAA produced substantial effects. Our estimation results suggest that non-attainment counties with single monitors experienced a drop in concentrations of 10.5% relative to attainment counties. In non-attainment counties with multiple monitors, the overall effect of the regulation is an increase of ambient PM10 concentrations by 1.9%. The dirtiest monitors in these counties, however, experienced drops in PM10 of 6.1%, which suggest that regulators focus their attention on the dirtiest monitors.

Full text

Page 1

eScholarship provides open access, scholarly publishing
services to the University of California and delivers a dynamic
research platform to scholars worldwide.
Department of Agricultural and Resource
Economics, UCB
UC Berkeley
Title:
Measuring the Effects of Environmental Regulations: The Critical Importance of a Spatially
Disaggregated Analysis
Author:
Auffhammer, Maximilian, University of California, Berkeley
Bento, Antonio M., Cornell University
Lowe, Scott E., Boise State University
Publication Date:
08-29-2007
Publication Info:
Department of Agricultural and Resource Economics, UCB, UC Berkeley
Permalink:
http://escholarship.org/uc/item/8d12x8pp
Keywords:
Air Pollution, Clean Air Act, Spatial Modeling
Abstract:
We examine the effects of the 1990 Clean Air Act Amendments (CAAA) on ambient concentrations
of PM10 in the United States between 1990 and 2005. Consistent with prior literature, we find
that non-attainment designation has no effect on the average monitor in non-attainment counties,
after controlling for weather, socioeconomic characteristics at the county level and lagged
concentrations. In sharp contrast, if we allow for heterogeneous treatment by type of monitor
and county, we do find that the 1990 CAAA produced substantial effects. Our estimation results
suggest that non-attainment counties with single monitors experienced a drop in concentrations
of 10.5% relative to attainment counties. In non-attainment counties with multiple monitors, the
overall effect of the regulation is an increase of ambient PM10 concentrations by 1.9%. The dirtiest
monitors in these counties, however, experienced drops in PM10 of 6.1%, which suggest that
regulators focus their attention on the dirtiest monitors.

Page 2

I
DEPARTMENT OF AGRICULTURAL AND RESOURCE ECONOMICS AND POLICYJ
DIVISION OF AGRICULTURE AND NATURAL RESOURCES
~NIVERSITY OF CALIFORNIA AT B E R K E L E ~
.---
~RKING
PAPER NO. 1047
~
MEASURING THE EFFECTS OF ENVIRONMENTAL REGULATIONS:
THE CRITICAL IMPORTANCE OF ASPATIALLY DISAGGREGATED ANALYSIS
by
Maximilian Auffhammer, Antonio M. Bento, and Scott E. Lowe
Copyright © 2007 by the authors. All rights reserved. Readers may make verbatim copies of this document for
noncommercial purposes by any means, provided that this copyright notice appears on all such copies.
California Agricultural Experiment Station
Giannini Foundation of Agricultural Economics
August 29, 2007

Page 3

Measuring the Effects of Environmental
Regulations: The critical importance of a spatially
disaggregated analysis∗
Maximilian Auffhammer
University of California, Berkeley
Antonio M. Bento
Cornell University
Scott E. Lowe
Boise State University
Date of this Draft: August 29, 2007
Abstract
We examine the effects of the 1990 Clean Air Act Amendments (CAAA) on ambient concen-
trations of PM10in the United States between 1990 and 2005. Consistent with prior literature,
we find that non-attainment designation has no effect on the average monitor in non-attainment
counties, after controlling for weather, socioeconomic characteristics at the county level and
lagged concentrations. In sharp contrast, if we allow for heterogeneous treatment by type of
monitor and county, we do find that the 1990 CAAA produced substantial effects. Our esti-
mation results suggest that non-attainment counties with single monitors experienced a drop in
concentrations of 10.5% relative to attainment counties. In non-attainment counties with mul-
tiple monitors, the overall effect of the regulation is an increase of ambient PM10concentrations
by 1.9%. The dirtiest monitors in these counties, however, experienced drops in PM10of 6.1%,
which suggest that regulators focus their attention on the dirtiest monitors.
Keywords: Air Pollution, Clean Air Act, Spatial Modeling
JEL Codes: Q53, Q58
∗We thank Calanit Saenger and Ravissa Suchato for excellent research assistance. We thank Peter Berck, Michael
Greenstone, Wolfram Schlenker and seminar participants at UC Berkeley and the 2006 ASSA meetings for valuable
comments. All errors are ours. For correspondence, contact Maximilian Auffhammer; Phone: (510) 643-5472; Email:
auffhammer@berkeley.edu or Antonio Bento, Phone: (607) 255-0626; E-mail: amb396@cornell.edu

Page 4

1.Introduction
Three empirical regularities characterize the changes in the spatial distribution of particulate matter
less than 10 microns in diameter (PM10) in the United States between 1990 and 2005: First, average
county level ambient concentrations of PM10dropped by about 25%; second, these drops were far
from uniform. Of the non-attainment counties in 1990, the average reduction in PM10in counties
with single monitors was 27% while in counties with multiple monitors this reduction was only
17%. Third, there was substantial spatial heterogeneity in reductions of PM10in non-attainment
counties with multiple monitors; the “dirtiest” monitors in these counties experienced drops that
were 9% greater than the average of the remaining monitors.
This naturally raises the following two questions: First, what is the effect of the 1990 Clean
Air Act Amendments (CAAAs) - measured by county non-attainment designations - on ambient
concentrations of PM10? Second, what is the level of spatial aggregation - county versus monitor
level - needed for the effects of the regulation to be properly captured? This paper attempts to shed
light on these questions by combining monitor level data on annual average PM10concentrations
from the EPA’s Air Quality System (AQS) between 1990 and 2005 with data from the Federal Code
of Regulations on county PM10attainment status. We ask whether county non-attainment status is
responsible for the drops in PM10experienced in single and multiple monitor counties. In the case of
multiple monitor counties, we are interested both in the overall mean change in PM10concentrations
and also in the spatial distribution of these changes, in particular, the changes in PM10at the dirtiest
monitors. Conversations with air quality regulators suggest that special attention is paid to the
dirtiest monitors in non-attainment counties, which may lead to a heterogeneous treatment effect
across monitors within non-attainment counties.
Over the years, researchers have made considerable strides in measuring the effects of federal
environmental regulations on the changes of ambient concentration of several criteria pollutants.
Henderson (1996) investigated the effects of ground level Ozone regulations in the United States
for the period of 1977-1987 on air quality and the migration of polluting facilities, using ambient
concentrations of Ozone measured at the monitor level. Along the same lines, Chay and Greenstone
(2003) and Chay and Greenstone (2005) examined the effects of total suspended particulates (TSPs)
on infant health and capitalization of air quality into property values induced by the 1970 Clean
1

Page 5

Air Act Amendments. More recently, Greenstone (2004) examined the effects of the 1970 and
1990 CAAA on Sulfur Dioxide concentration, using comprehensive county-level data files on SO2
concentrations. A key finding in this literature is that county non-attainment status designation -
the centerpiece of the CAAAs - is responsible for only modest (and often not significant) reductions
of Ozone and TSPs during the 80s.1
Our concern is that, because of the lack of a spatially-disaggregated analysis that can
capture the heterogeneity of ambient concentrations within non-attainment counties with multiple
monitors and the failure to differentiate non-attainment counties with single monitors from non-
attainment counties with multiple monitors, these studies may have potentially “averaged out”
the true effects of environmental regulations. These studies typically use ambient concentrations
measured at monitoring stations and are conducted at the county level. The problem is that, by
regressing an average of the ambient concentration of the monitors located in a county on a non-
attainment dummy, these studies could potentially underestimate the true effects of the regulation
if, the ’dirtier’ parts of non-attainment counties reduce ambient concentrations by substantially
larger amounts that the ‘cleaner’ ones. In other words, by aggregating the analysis to county
level, it looks almost as if non-attainment counties behaved identically to attainment counties,
therefore implying that environmental regulations are responsible for only minor reductions in
ambient concentrations of criteria pollutants.
Similarly, by not separating the effects of the regulations across single versus multiple mon-
itor counties, prior studies may have failed to identify the effects of the regulations. This could
simply happen if, for example, the effects experienced in single monitor counties are partially offset
by those experienced in multiple monitor counties.
This paper differs from prior literature in three distinct ways. First, we look at the impact of
federal air quality regulation on particulate matter less than 10 microns in diameter, which is often
considered to be the ”pollutant of the 90s”. Further, like Henderson (1996), we conduct our analysis
at the monitor level, yet we allow the regulation to have a differential effect on concentrations
measured at the monitoring site depending on the type of county (single vs. multiple monitors)
and type of monitor (highest versus non-highest). Finally using previously unavailable weather
1Henderson (1996) finds that county Non-Attainment status led to an additional 8% improvement in Ozone levels;
Chay and Greenstone (2005) report that 1970-1980 TSP declined 12% more in non-attainment counties than in
attainment counties
2

Page 6

data, we are able to control for weather impacts at the monitor level, allowing for within county
heterogeneity of rainfall and temperature.
We address these issues by combining annual average concentrations of PM10at the mon-
itor level between 1990 and 2005 with county attainment designations for PM10. Additional data
were collected to account for other determinants of changes in PM10, including climate and eco-
nomic activity. We further control for monitor and year fixed effects to remove any unobservable
confounding factors constant by monitor or year.
We use these data to estimate two sets of models. The first is a model that replicates existing
studies of the effects of environmental regulations on other criteria pollutants (e.g., Greenstone,
2004) measured at the county level. The second is a more disaggregated model where we distinguish
single from multiple monitor counties. In addition for multiple monitor counties we allow for the
possibility of heterogeneous impacts of the regulation based on the concentration at the ’dirtiest’
monitor.
The rest of the paper is organized as follows. Section 2 provides a brief overview of PM10
regulations; section 3 describes the data sources and provides summary statistics on the trends in
monitoring and PM10concentrations between 1990 and 2005. Section 4 presents the econometric
models and section 5 the results. Section 6 concludes.
2.Basic Aspects of PM10Regulation
2.1Brief Historical Facts About PM10Regulation
Particulate Matter is a term used for a class of solid and liquid air pollutants. Total suspended
particulates (TSPs) include particles less than 100 microns in diameter. The 1971 Clean Air Act au-
thorized the Environmental Protection Agency to enforce a National Ambient Air Quality Standard
(NAAQS) for TSPs. The standards are phrased as primary and secondary standards. “Primary
standards set limits to protect public health, including the health of “sensitive” populations such
as asthmatics, children, and the elderly. Secondary standards set limits to protect public welfare,
including protection against decreased visibility, damage to animals, crops, vegetation, and build-
ings.” (see United States Environmental Protection Agency (2005) for further discussion). Each
3

Page 7

standard is defined in terms of an annual benchmark average as well as 24 hour benchmarks. From
April 30th1971 until July 1st 1987 the primary annual standard for TSPs was 260 µg/m3for the
24-hour average and 75 µg/m3for the annual average. The secondary standard for TSPs was
150 µg/m3for the 24-hour average and 60 µg/m3for the annual average (National Archives and
Records Administration, 1987).
If a county exceeded the primary annual standard for one year or the primary 24-hour
standard for more than a single day per year it was considered to be in violation of the standard. By
provisions in the Clean Air Act, the EPA can move to designate a county “non-attainment”. After a
lengthy review process, a non-attainment county was required to submit, in a state implementation
plan (SIP), the strategy that it intends to use to become in attainment with the NAAQS. If the
deficiency remains uncorrected, or if the EPA “finds that any requirement of an approved plan
(or approved part of a plan) is not being implemented”, the county is given 18 months to correct
the deficiency. If the deficiency continues to be uncorrected the EPA Administrator may impose
sanctions on the county in violation, including the withholding of federal highway funds, and the
imposition of technological “emission offset requirements” on new or modified sources of emissions
within the county (National Archives and Records Administration, 2005). In the first stage of the
sanction process, only one of the sanctions is applied at the discretion of the EPA Administrator;
if the county continues to be in violation 6 months after the first sanction, then both are applied.
These sanctions are enforced not at the state level, but at the political subdivisions that “are
principally responsible for such deficiency” (National Archives and Records Administration, 1987).
In 1987, the U.S. Environmental Protection Agency refined their particulate policy to regu-
late particulates less than 10 micrometers in diameter (PM10). The new primary standard required
the three year geometric average of PM10concentration for each monitor in a county to be less than
50 µg/m3. It further required via a secondary standard that the 24 hour average concentrations
at a monitor do not exceed 150 µg/m3. This change was implemented because a growing body
of scientific evidence indicated that the greatest health concern from particulate matter stemmed
from PM10, which can penetrate into sensitive regions of the respiratory tract.2
2For a concise analysis of the health effects from exposure to PM10, see Hall, Winer, Kleinman, Lurmann, Brajer
and Colome (1992). For an analysis of the impact of air pollution on infant health, see Currie and Neidell (2005),
and Chay and Greenstone (2003).
4

Page 8

2.2 Sources of PM10Pollution
Particulate matter enters the atmosphere in one of two ways: primary particulate matter is emitted
directly into the atmosphere as a solid or liquid; secondary particulate matter is formed in the
atmosphere by reactions between precursor gases such as organic gases, nitrogen oxides (NOx), and
sulfur oxides (SOx). In general, the contribution of the secondary PM10precursor gases to total
ambient PM10is substantially larger than the contribution of primary particulate matter.
In California, for example, the California Air Resources Board estimates that in the year
2000, there were approximately 2,400 tons of primary PM10 emitted on a daily basis. Of these
2,400 tons, 6% was emitted by stationary industrial sources, 5% was emitted directly from mobile
sources, 15% was generated from paved roads, and the remaining 74% was produced by area-wide
sources. The area-wide sources include residential fuel combustion (7%), farming operations (9%),
construction and demolition (9%), unpaved road dust (27%), fugitive windblown dust (12%), and
burning and waste disposal (10%).
In addition to the primary PM10emissions, 10,847 tons of secondary PM10precursor gases
were emitted into the atmosphere on a daily basis in California in the year 2000. These precursor
gases include 3,591 tons of NOx, 333 tons of SOx, and 6,923 tons of organic gases (California
Air Resources Board, 2001). The actual contribution of the secondary PM10 precursor gases to
ambient PM10concentration levels depends on the ambient concentrations of the precursor gases
themselves, as well as the atmospheric chemistry of the region, including the relative humidity,
temperature, wind speed and direction (Foresman, Kleeman, Kear and Niemeier, 2003). In this
case one may find two areas with similar secondary PM10precursor gas releases that have different
secondary PM10 ambient concentrations, depending on their location-specific characteristics. In
the case of the South Coast Air Basin, the PM10reduction efficiency calculations, which allow one
to estimate the primary and secondary emissions required to produce a single unit increase in the
ambient concentration of PM10, indicate that NOxemissions in 1990 contributed to over half of
the total ambient PM10concentration (see Foresman et al. (2003) for further discussion).
5

Page 9

3.Overview of the Trends in PM10Concentrations
and Regulations
To implement the analysis, we compiled the most detailed data available on concentrations, non-
attainment status and other relevant determinants of concentrations, including climate and eco-
nomic activity. This section describes the data sources and presents summary statistics on national
trends in PM10, the distribution of monitors and mean concentrations of between 1990 and 2005.
3.1 PM10Concentrations and Attainment Status Data
The concentrations data were obtained from the Air Quality Standards (AQS) database, maintained
by the EPA. For each PM10 monitor operating between the 1990 and 2005 period, these data
include a number of monitor characteristics including the location of the monitor, and hourly
readings during operation. For estimation purposes, we calculated the annual monitor averages
across hourly readings from the set of monitors that operated for at least 75% of the year, which
is how the EPA designates a monitor as active.
The annual county attainment status designations were determined from the annual Code
of Federal Regulations (CFR). For each of the criteria pollutants, the CFR reports the county
attainment status in one of the following categories: “does not meet primary standards”, “does not
meet secondary standards”, “cannot be classified” and “better than national standards”. For some
criteria pollutants, CFR indicates that only part of a county did not meet the primary standards.
Based on this information, we assigned a county to be non-attainment if the whole county or parts
of it failed to meet the “primary” or “secondary” standards.
3.2Additional Data: Attainment Status with other criteria pollutants, Climate
and Economic Activity
We supplement the data on PM10concentrations and attainment status with additional relevant
data, reflecting the need to capture other determinants of the change in PM10. Since attainment
status is not only assigned for PM10, but for five other criteria pollutants, it is important to separate
the impact of policy induced reductions in precursor emissions to the pollutant of interest. We
therefore control for yearly county non-attainment status for TSP, Ozone, SOxand NOxcollected
6

Page 10

from the CFR.3
In addition to regulation, there are other physical factors influencing ambient concentrations
of PM10. Temperature and rainfall affect the formation of secondary PM10as well the presence
of primary particulates. Since microclimates vary greatly within states and large counties, we do
not use county averages, but use rainfall and temperature at the monitor. We control for February
and July rainfall and temperatures, which have been shown to be highly correlated with particulate
concentrations, since they proxy for how cold/wet each winter was and how warm/dry each summer
was at the monitor level. We use the PRISM Group (2007) dataset, which provides monthly data
based on all US weather stations extrapolated to a set of 4km grids covering the continental United
States between 1990 and 2005, allowing us to construct weather observations at the pollution
monitor location.
Finally, emissions of particulate matter are strongly correlated with economic activity.
While GDP is not available at the county level, the Bureau of Economic Analysis (BEA) releases
annual estimates of personal income at the county level. This indicator has been widely used in the
Environmental Kuznets Curve literature at the state level (e.g., Millimet, List and Stengos, 2003).
We include the county level real personal income for each year and county in our sample.
3.3National Trends in Monitoring and concentrations
3.3.1Distribution of monitors by Counties
Table 1 presents annual summary information on counts of monitors by county for monitors that
were active at least 75% of the year. The second column reports the number of counties with
active monitors. As a result of the 1990 CAAA, both the number of operating monitors and the
geographical coverage of PM10readings increased substantially between 1990 and 2005. In 1991
only 2 counties had active monitors; by the end of our sample 95 counties had active monitor
readings.
Columns (3)-(10) present the count of counties as well as the mean of monitors by attainment
status in ‘single’ and ‘multiple’ monitor counties. In the year 2000, for example, a total of 51
counties had active monitors at least 75% of the year. Of these, 28 were single-monitor attainment
3In 1997 the EPA began to regulate fine particulates. Non-attainment designations for fine particulates were first
assigned in 2005. We further do not control for lead non-attainment status.
7

Page 11

counties, 16 single-monitor non-attainment counties, 4 multiple monitor attainment counties, and
3 multiple monitor non-attainment counties. The number of monitor years from multiple monitor
non-attainment counties is 42% larger than the number of monitors in multiple monitor attainment
counties.
The spatial distribution of monitors active during our sample period across the United States
is quite varied. In 2005, Maricopa (AZ), Dona Ana (NM) and Allegheny (PA) were the counties
with the largest number of monitors. As of 2005, the population in counties monitored were 153
million people, which is about 52% of the total US population. The overall spatial distribution
of monitors reflects the EPA’s concerns of measuring concentrations of in areas that are heavily
populated.4
3.3.2 Trends in PM10between 1990 and 2005
Table 2 shows the mean annual concentrations between 1991 and 2004 by attainment status and
“single” versus “multiple” monitor counties based on all active monitors. The following trends stand
out: First, by 1991 the mean annual concentrations in non-attainment counties were already below
the national air quality primary standard (NAAQS) for PM10. In single-monitor non-attainment
counties concentrations were 38.19 µg/m3. In multiple-monitor non-attainment counties, concen-
trations were 35.30 µg/m3. More surprisingly, the average of the monitors with the highest mean
annual concentration in non-attainment counties was below the NAAQS.
Second, as figure 2 shows, the relative reductions in PM10concentrations between 1991 and
2004 were remarkable, regardless of initial attainment status. On average, single monitor counties
experienced a larger drop in PM10. Single monitor attainment counties experienced a reduction
in of about 29.5%, and single monitor non-attainment counties experienced a reduction of 35.12%.
The drops in multiple-monitor counties were less dramatic: 17.26% and 19.33% for attainment and
non-attainment counties, respectively.
Third, a comparison of the last two columns of table 2 reveal that between 1990 and 2005,
the reductions in the annual mean concentrations of PM10for the highest monitor were 39.46%,
which is very close to the drop experienced by single monitor non-attainment counties.This
provides some evidence in support of the hypothesis that regulators targeted the ’dirtiest’ areas in
4A map of monitor location can be found at the EPA’s website: http://www.epa.gov/airtrends/pm.html. A map
of monitors active for more than 75% of the year which are used in this paper is available upon request.
8

Page 12

multiple monitor non-attainment counties.
Fourth, most of the additional effort to reduce in non-attainment counties took place be-
tween 1990 and 1997. In fact, Table 2 reveals that in non-attainment counties concentrations
actually leveled off or slightly rose after 1997, especially in multiple monitor counties.
4.Econometric Model
In this section, we describe the econometric strategy adopted to measure the effects of PM10
attainment status on changes in concentrations. Let Dj,tbe an indicator variable that equals one
when the entirety or part of county j is designated non-attainment in year t and 0 if it is in
attainment. Let Yj
i,tdenote the PM10concentrations of monitor i in county j in year t. Consistent
with the literature, our basic econometric model is equation 1 below:
Yj
i,t+1− Yj
Yj
it
it
= αDj,t+ Xj,tβ + Pi,tϕ + θt+ δi+ ηi,t
(1)
where α is the parameter of interest and measures the difference in the percent change in
PM10concentrations between non-attainment and attainment counties. Formally, α represents the
average treatment effect of attainment status in non-attainment counties, and is given by:
α = E

Yj
i,t+1− Yj
Yj
it
it
|Dj,t= 1

− E

Yj
i,t+1− Yj
Yj
it
it
|Dj,t= 0


Xj,tis a vector of controls, which vary over time at the county level. These include non-
attainment status of monitors in county j for other criteria pollutants (i.e TSP, NOx, SOx and
Ozone) in the same year that Dj,tis measured and a county-level measure of income. pi,tis a vector
of controls, which vary at the monitor level. In this paper we include rainfall and temperature at the
monitor level, as described in the data section. θtis a year fixed effect that is common to monitors
located in attainment and non-attainment counties, δiis a monitor fixed effect that controls for
monitor specific unobservables that are invariant over time, and ηi,tis the idiosyncratic unobserved
component of the percent change in PM10concentrations, which is assumed to be stationary ergodic.
Further, we follow Greenstone (2004) and include one lag of PM10concentrations in order
9

Page 13

to remove the unwanted correlation between Dj,tand ηi,t. In this paper we present models with a
one-period linear lag.5
The model described by equation (1) is appropriate to measure the average effect of at-
tainment status on the average percent change of PM10county concentrations. However, it does
not allow us to disentangle the potential differential impact of the non-attainment status across
single versus multiple monitors’ counties and across ‘dirtier’ versus ‘cleaner’ monitors’ within non-
attainment counties. In order to address these concerns, we estimate an augmented specification
of equation (1).
Yj
i,t+1− Yj
Yj
it
it
= αDj,t+γ1Dj,tMj,t+γ2(1 − Dj,t)Mj,t+λDj,tMj,tHi,t+Xj,tβ+Pi,tϕ+θt+δi+ηi,t (2)
where Mj,ttakes the value 1 if a non-attainment county has multiple monitors, zero oth-
erwise.
Hi,t is a dummy equal to one, if a monitor has the highest concentration during the
non-attainment designation year out of all active monitors in the county. It therefore only equals
one if a monitor is the highest monitor in a multiple monitor county during the designation year.
Equation (2) allows for the effects of the regulation to have differential impacts. In contrast
to equation(1), α isolates the effect of non-attainment status on non-attainment counties with
single monitors. γ1captures the additional effect of the regulation on a monitor located in non-
attainment counties with multiple monitors (in these counties the effect of the regulation is given by
(α+γ1)). Since dirtier counties are likely to have more monitors, we would expect γ1to be positive.
γ2captures the additional effect of the regulation on a monitor located in an attainment county
with multiple monitors relative to that experienced by a single monitor county. Finally, λ isolates
the difference in percent annual reductions of ambient concentrations of the highest monitor in a
multiple-monitor non-attainment county with respect to the other (“non-highest”) monitors in the
same county. The effect of the regulation on the percent change in concentrations at the highest
monitor in a multiple monitor non-attainment county is given by (α+γ1+λ). If indeed regulators
targeted the dirtier areas of non-attainment counties, λ should be negative.
λ is an estimate of the average treatment effect for the dirtiest monitor in the multiple
monitor counties. The concentrations at these dirtiest monitors will vary, and one would expect
5However, since the true lag structure is unknown, we experimented with different lag lengths using the SIC and
found that only the first lag was significant. We therefore include a single lag of ambient concentrations.
10

Page 14

more effort to be expended at dirtier monitors within this cohort. Therefore we will allow for a
heterogeneous treatment effect at the highest monitors. We will interact λDj,tMj,tHi,t with the
once lagged concentration at the monitor. A significant and negative coefficient would suggest that
the higher the concentrations are at the dirtiest monitor, the larger the percent reductions.
Models (1) and (2) address the potential impact of regulation on the annual average con-
centrations. From Table 1 we can clearly tell that many counties, which are not in attainment, had
annual average concentrations below the non-attainment concentration. Non-attainment, however,
can also be due to violating the hourly standard. Model (3) below explains variation in the share
of operating days monitor exceeded the hourly standard:
Vj
i,t+1−Vj
i,t= αDj,t+γ1Dj,tMj,t+γ2(1 − Dj,t)Mj,t+λDj,tMj,tHi,t+Xj,tβ+Pi,tϕ+θt+δi+ηi,t (3)
Where Vj
i,tis the number of days in year t for which monitor i in county j exceeded the
24 hour primary standard over the number of days the monitor was active. We would expect the
coefficient estimates on regulation to be negative and significant. For this measure, it is not clear a
priori at which geographic level of aggregation regulation is going to have a detectable impact on
ambient concentrations.
As further robustness checks, we will run models (1) and (2) using the level difference in
concentrations as the dependent variable instead of using percentage changes. This specification
is more sensitive to outliers, which is why we prefer the percent change specification. In addition,
we will explore the removal of California from the sample as well as an alternate definition of the
highest monitor.
5.Results
Table 3 displays the central estimation results. The entries are the parameter estimates and their
estimated robust standard errors (in parentheses). Models (1) and (2) refer to equation (1), while
models (3)-(5) refer to equation (2). We remind the reader that the dependent variable is the
percent change in PM10. Since the coefficients of interest are α, γ and λ, the table displays the
estimates of these parameters. However, at the bottom, we identify the additional set of controls
11

Page 15

used in each model. Models (1) - (5) include the weather and income variables linearly. Models (6)
- (10) are identical specifications, yet income and weather enter as second order polynomials. This
improves the fit of the models significantly. We will therefore restrict our discussion to models (6)
- (10).
The key finding from the first two specifications is that, independent of the fact whether
one includes lagged ambient PM10levels as covariates, the county non-attainment designation does
not explain a statistically significant share of the variation in the percent reductions in PM10. This
finding is consistent with Greenstone (2004), albeit for a different pollutant and time period.
Once we augment specification (1) by dummies for multiple monitor counties, we obtain
a statistically significant and sizeable estimate of the treatment effect for single monitor non-
attainment counties. The estimated coefficient can be interpreted as non-attainment designation
leading to a 10.9% decrease in ambient concentrations. For multiple monitor non-attainment coun-
ties this effect is actually a 2.1% increase in ambient concentrations on average.
The results differ drastically when we allow for differentiated impacts of the regulation on
non-attainment counties. Model (9) in table 3 highlights our key findings. Three results stand
out: First, unlike models (6)-(7), even after controlling for lagged PM10, county non-attainment
designation is associated with a decrease in PM10of about 10.5% in non-attainment counties with
single monitors.
Second, the estimate of γ - the additional effect of the regulation in non-attainment counties
with multiple monitors - is positive, suggesting that in response to the regulation, non-attainment
counties with multiple monitors reduced PM10by lower amounts that non-attainment counties with
single monitors. Indeed, this specification suggests that non-attainment status is associated with an
increase in PM10in non-attainment counties with multiple monitors of about 2%. By not allowing
for the distinction between single and multiple monitor counties, equation (1) fails to identify this
effect of the regulation.
Third, the estimate of λ - which isolates the behavior of the highest monitor in non-
attainment counties with multiple monitors - is negative, suggesting that indeed regulators appear
to have targeted dirtier areas in non-attainment counties. Compared with other monitors in non-
attainment counties with multiple monitors, the highest monitor experienced a reduction of about
8.1% more in response to the regulation. Compared to attainment counties, our model suggests
12