Read the PDF and write a 1.5 page single space report.
Unformatted Attachment Preview
Journal of Public Economics 101 (2013) 105–114
Contents lists available at SciVerse ScienceDirect
Journal of Public Economics
journal homepage: www.elsevier.com/locate/jpube
Do local energy prices and regulation affect the geographic concentration
Matthew E. Kahn a, b,⁎, Erin T. Mansur b, c
UCLA Institute of the Environment, La Kretz Hall, Suite 300, 619 Charles E. Young Drive East, Box 951496, Los Angeles, CA 90095, United States
National Bureau of Economic Research, United States
Dartmouth College, Department of Economics, 6106 Rockefeller Hall, Hanover, NH 03755, United States
a r t i c l e
i n f o
Received 9 September 2011
Received in revised form 18 January 2013
Accepted 11 March 2013
Available online 16 March 2013
a b s t r a c t
Manufacturing industries differ with respect to their energy intensity, labor-to-capital ratio and their pollution intensity. Across the United States, there is signiﬁcant variation in electricity prices and labor and environmental regulation. This paper examines whether the basic logic of comparative advantage can explain the
geographical clustering of U.S. manufacturing. We document that energy-intensive industries concentrate in
low electricity price counties and labor-intensive industries avoid pro-union counties. We ﬁnd mixed evidence that pollution-intensive industries locate in counties featuring relatively lax Clean Air Act regulation.
© 2013 Elsevier B.V. All rights reserved.
Between 1998 and 2009, aggregate U.S. manufacturing jobs declined by 35 percent while the total production of this industry
grew by 21 percent. 1 This loss of manufacturing jobs has important
implications for the quality of life of the middle class. Manufacturing
offers less educated workers employment in relatively well paying
jobs (Neal, 1995). Despite public concerns about the international
outsourcing of jobs, over eleven million people continue to work in
the U.S. manufacturing sector. 2 The ability of local areas to attract
and retain such manufacturing jobs continues to play an important
☆ We thank Severin Borenstein, Joseph Cullen, Lucas Davis, Meredith Fowlie, Jun Ishii,
Enrico Moretti, Nina Pavcnik, Frank Wolak, Catherine Wolfram, and the seminar participants at the 2009 UCEI Summer Camp, UBC Environmental Economics and Climate
Change Workshop 2010, the 2012 UC Berkeley Power Conference, ClaremontMcKenna College, Amherst College, the University of Alberta, the University of
Michigan, and Yale University for their useful comments. We thank Wayne Gray for
sharing data with us and Koichiro Ito and William Bishop for assisting with Fig. 1. We
thank the two anonymous reviewers for their several useful comments.
⁎ Corresponding author.
E-mail addresses: [email protected] (M.E. Kahn), [email protected]
The U.S. Bureau of Labor Statistics reports employment by sector. From 1998 to 2009,
manufacturing employment fell from 17.6 million to 11.5 million (http://data.bls.gov/
timeseries/CES3000000001?data_tool=XGtable). The United Nations Statistics division
reports gross value added by kind of economic activity at constant (2005) US dollars. From
1998 to 2009, manufacturing value went from $1348 billion to $1626 billion (http://data.
In March, 2011, 11.67 million people worked in manufacturing (NAICS 31–33)
0047-2727/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
role in determining the vibrancy of their local economy (Greenstone
et al., 2010).
Ongoing research examines the role that government regulations
and local factor prices play in attracting or deﬂecting manufacturing
employment. During a time when unemployment rates differ greatly
across states, there remains an open question concerning the role
that regulation plays in determining the geography of productive activity. A leading example of this research is Holmes’ (1998) study
that exploited sharp changes in labor regulation at adjacent state
boundaries. He posited that counties that are located in Right-toWork states have a more “pro-business” environment than their
nearby neighboring county located in a pro-union state. He used this
border-pairs approach to establish that between 1952 and 1988
there has been an increasing concentration of manufacturing activity
on the Right-to-Work side of the border. A recent Wall Street Journal
piece claimed that, between the years 2000 and 2008, 4.8 million
Americans moved from union states to Right-to-Work states. 3
In this paper, we build on Holmes’ core research methodology
along three dimensions. First, we focus on the modern period from
1998 to 2009. During this time period, the manufacturing sector experienced signiﬁcant job destruction as intense international competition has taken place (Davis et al., 2006; Bernard et al., 2006). This
time period covers the start of the recent deep downturn in the national economy and the earlier 2000 to 2001 recession. Past research
has documented that industrial concentration is affected by energy prices
Arthur B. Laffer and Stephen Moore. “Boeing and the Union Berlin Wall.” http://online.
M.E. Kahn, E.T. Mansur / Journal of Public Economics 101 (2013) 105–114
(Carlton, 1983), environmental regulation (Becker and Henderson, 2000;
Greenstone, 2002; Walker, 2012), and labor regulation and general state
level pro-business policies (Holmes, 1998; Chirinko and Wilson, 2008).
Second, we use the border-pair methodology to study the relative
importance of these three key determinants of the geographic concentration of manufacturing jobs in one uniﬁed framework. Third, we examine the heterogeneity of industries’ response to these policies.
We estimate a reduced form econometric model of equilibrium
employment variation across counties that allow us to study how energy regulation, labor regulation and environmental regulation are
associated with the spatial distribution of employment while holding
constant the other policies of interest. Our identiﬁcation strategy exploits within border-pair variation in energy prices and regulation to
tease out the role that each of these factors play in inﬂuencing the
geographical patterns of manufacturing employment. As we discuss
below, county border pairs share many common attributes including
local labor market conditions, spatial amenities, and proximity to
markets. We compare our estimates of policy effects in regression results with different levels of geographic controls to see how robust
our results are across different speciﬁcations.
This paper studies where different industries cluster across different types of counties as a function of county regulation status. In
the case of manufacturing, we disaggregate manufacturing into 21
three-digit NAICS industries. These industries differ along three dimensions; the industry’s energy consumption per unit of output, the
industry’s labor-to-capital ratio, and the industry’s pollution intensity.
We model each county as embodying three key bundled attributes; its
utility’s average industrial electricity price, its state’s labor regulation,
and the county’s Clean Air Act regulatory status.
The basic logic of comparative advantage yields several testable
hypotheses. In a similar spirit as Ellison and Glaeser (1999), we test
for the role of geographical “natural advantages” by studying the sorting
patterns of diverse industries. Energy-intensive industries should avoid
high electricity price counties.4 Labor-intensive manufacturing should
avoid pro-union counties. Pollution-intensive industries should avoid
counties that face strict Clean Air Act regulation. We use a countyindustry level panel data set covering the years 1998 to 2009 to test
all three of these claims.
The paper also examines the relationship between energy prices and
employment for speciﬁc industries. We recognize that manufacturing is
just one sector of the economy and thus we examine how other major
non-manufacturing industries are affected by energy, labor and environmental regulation. For 21 manufacturing industries and 15 major
non-manufacturing industries, we estimate this relationship. We ﬁnd
that energy prices are not an important correlate of geographical concentration for most non-manufacturing industries. However, employment in expanding industries such as Credit Intermediation (NAICS
522), Professional, Scientiﬁc and Technical Services (NAICS 541), and
Management of Companies and Enterprises (NAICS 551) is responsive
to electricity prices with implied elasticities of approximately −.15.
In comparison, the most electricity-intensive manufacturing industry,
primary metals, has an elasticity of −1.17.
2. Empirical framework
Our empirical work will focus on examining the correlates of the
geographic clusters of employment and establishments by industry
starting in 1998. Building on Holmes’ (1998) approach, we rely heavily on estimating statistical models that include border-pair ﬁxed
effects. A border pair will consist of two adjacent counties.
Energy-intensive industries will also attempt to avoid high oil, coal, and natural gas
prices, as well. However, our identiﬁcation strategy examines differences between
neighboring counties and while there are regional differences in coal and natural gas,
these differences are likely to be small between neighboring counties.
Comparing the geographic concentration of employment within a
border pair controls for many relevant cost factors. Manufacturing
ﬁrms face several tradeoffs in choosing where to locate, how much
to produce, and which inputs to use. To reduce their cost of production, they would like to locate in areas featuring cheap land, low
quality-adjusted wages, lax regulatory requirements and cheap energy. They would also like to be close to ﬁnal consumers and input suppliers in order to conserve on transportation costs. Within a border
pair, we posit that local wages are roughly constant as are location speciﬁc amenities and proximity to input suppliers and ﬁnal consumers.
Our unit of analysis will be a county/industry/year. First we study
the geographic concentration of 21 manufacturing industries using
the U.S. County Business Patterns (CBP) data over the years 1998 to
2009. 5 The CBP reports for each county and year the employment
count, establishment count and establishment count by employment
size. This last set of variables is important because the CBP suppresses
the actual employment count and reports a “0” for many observations
(Isserman and Westervelt, 2006). 6
Throughout this paper, we assume that each industry differs with
respect to its production process (and hence in their ﬁrms’ response
to electricity prices and regulation) but any two ﬁrms within the
same industry have the same production function. In general, energy
inputs and the ﬁrm’s environmental control technology may be either
substitutes or complements with labor in a given industry (Berman
and Bui, 2001). Our paper studies the effects of regulations on overall
employment, combining both these substitution effects as well as
Our main econometric model is presented in Eq. (1). Estimates of
Eq. (1) generate new ﬁnding about the equilibrium statistical relationship between regulation, electricity prices and manufacturing location choices between 1998 and 2009. The unit of analysis is by
county i, county-pair j, industry k, and year t. County i is located in
utility u and state s. In most of the speciﬁcations we report below,
we will focus on counties that are located in metropolitan areas. 7
empijuskt ¼ β 1 P elec
ut þ β 2 P ut ⋅ElecIndexkt þ β 3 Right s ⋅LabCapRatiokt
þβ 4 Nonattainit þ β 5 Nonattainit ⋅PollIndexk þ β 6 NoMonitori
þβ 7 NoMonitori ⋅PollIndexk þ θ1 ElecIndexkt þ θ2 Right s þ θ3 LabCapRatiokt
þθ4 PollIndexk þ f ðPollit Þ þ δZi þ α j þ γkt þ π st þ εijuskt :
In this regression, the dependent variable will be a measure of
county/industry/year employment. The ﬁrst term on the right side of
Eq. (1) presents the log of the average electricity prices that the industry faces in a speciﬁc county. The second term allows this price effect
to vary with the industry’s electricity-intensity index. In the regressions, the electricity-intensity index is normalized to range from 0 to
1 for ease in interpreting the results. 8 Third is an interaction term
between whether state s has Right-to-Work laws (Right) and the
County Business Patterns (http://www.census.gov/econ/cbp/download/index.htm).
We use 1998 as our start date because this was the ﬁrst year in which NAICS rather than
SIC codes where used. All data use the 2002 NAICS deﬁnitions.
The CBP suppress employment counts to protect ﬁrms’ privacy in certain cases. In
35 percent of our observations, employment equals zero despite there being a positive
count of establishments in that county, industry and year. To address this issue, we impute
the employment data using the establishment count data when suppression occurs. The
CBP provides the counts of establishments by ﬁrm size category. We take the midpoint
of employment for each of these categories and use the county/industry/year establishment count data across the employment size categories (1–4, 5–9, 10–19, 20–49, 50–99,
100–249, 250–499, 500–999, 1000–1499, 1500–2499, 2500–4999 and 5000+) to impute
the employment count for observations that are suppressed. We top code the 5000+ employment observations at 6000.
MSA counties account for most of the population (78% of the 1995 US population),
manufacturing establishments (78% in sample), and manufacturing workforce (74% in
The NBER productivity data report electricity intensity in electricity usage (in
kWh) per dollar value of shipments. We normalize this measure to range from zero
to one to simplify the interpretation of the price coefﬁcients.
M.E. Kahn, E.T. Mansur / Journal of Public Economics 101 (2013) 105–114
industry’s labor-to-capital ratio (LabCapRatio). Finally, we examine
the effect of environmental policy. This includes the interaction of an
indicator of nonattainment status (Nonattainment) and a continuous
index of pollution from an industry (PollIndex). We also examine the
interaction effect of an indicator of whether a county does not monitor
the pollutant of interest (NoMonitor) and the PollIndex variable.
In estimating these policy-relevant variables, we try to control
for potentially confounding factors. There are several variables that
we would estimate in a traditional difference-in-differences model,
including the direct effects of ElecIndex, Right, LabCapRatio, and
PollIndex: θ1 − θ4. However, all of these are perfectly collinear with
the various ﬁxed effects that we estimate. For example, the direct
effect of Right-to-Work states cannot be separately identiﬁed given
the inclusion of state-year ﬁxed effects. We do control for a ﬂexible
function of pollution concentration levels, pollit.9 The Z vector has county variables: a county’s population in 1970, its distance to the nearest
metropolitan area’s Central Business District (CBD), the county’s land
area, and the log of the 1990 housing values.10 In the core speciﬁcations
we control for a county-pair ﬁxed effect, industry-year ﬁxed effects and
state-year ﬁxed effects. We rely heavily on these border-pair ﬁxed
effects to soak up spatial variation in local labor market conditions,
climate amenities, and proximity to intermediate input providers and
ﬁnal customers. Past studies such as Dumais et al. (2002) have emphasized the importance of labor pooling as an explanation for why ﬁrms in
the same industry locate close together. The industry-year ﬁxed effects
control for any macro level changes in demand due to shifting national
consumption trends or world trade. 11 The state–year ﬁxed effects control for local labor market conditions such as local wage trends and any
state policy that affects a ﬁrm’s propensity to locate within a state. For
example, some states such as Missouri have low taxes while others
such as California do not.12
We use several different dependent variables. We begin by examining the number of manufacturing employees. We also present
results that focus on an industry’s percentage of total county employment. In another speciﬁcation, we report results for the natural log of
employment, which is estimated only for observations with positive
employment. As discussed below, 14 percent of our observations
have no establishments and thus no employees.
For each manufacturing industry, we can measure the electricity
intensity and the labor–capital ratio. These data are from NBER Productivity Data Base and cover 1997 to 2009. 13 Below, we will also
Counties are more likely to be assigned to nonattainment status if their ambient air
pollution levels in the recent past have been higher. If booming counties have high regulation levels, then a researcher could conclude that regulation raises employment
levels when in fact reverse causality is generating this relationship. To sidestep this
problem, we include a ﬂexible function of the county’s ambient pollution level.
Adjacent counties are unlikely to be “twins.” The classic monocentric model of urban economics predicts that counties closer to a major Central Business District will
feature higher population densities and higher land prices than more suburban
counties. We have also estimated speciﬁcations that included other county attributes
such as a dummy indicating whether the county is the metropolitan area’s center
county and another dummy that indicates whether the county is adjacent to an Ocean
or a Great Lake. The results are robust to controlling for these variables and are available on request. In Appendix Table A1, we present formal tests of whether our explanatory variables included in the Z vector are “balanced.” We ﬁnd that these covariates
vary by treatment for high electricity prices, labor regulation, and environmental regulation. In a regression reported in Table 5, we include linear trends for each covariate
to test whether our results are robust.
Linn (2009) documents that linkages between manufacturing industries amplify
the effect of macro energy price shocks. Given that energy-intensive industries are important input suppliers to other industries, there could be industry–year effects driven
by such linkages. Including the industry–year ﬁxed effects helps to address this issue.
For more on the macroeconomics impacts of energy price changes see Killian (2008).
Recent empirical work has documented that minimum wage differences across
states do not inﬂuence the locational choices of low skill jobs (Dube et al., 2010).
See http://www.nber.org/data/nbprod2005.html. We thank Wayne Gray for providing us with data that extends the sample through 2009.
present results for non-manufacturing industries but we cannot
measure their electricity, labor, or pollution intensity. As such, our
main results focus on manufacturing where we can test for the role
of geographic regulations in attracting employment activity.
The interaction terms presented in Eq. (1) allow us to test three hypotheses. The ﬁrst hypothesis is that energy-intensive industries cluster
on the low electricity price side of the border. The second hypothesis is
that labor-intensive industries cluster on the Right-to-Work Side of the
border. The third hypothesis is that high emission industries cluster in
the low environmental re …
Purchase answer to see full