Journal of Urban Economics 46, 360᎐376 Ž1999. Article ID juec.1998.2127, available online at http:rrwww.idealibrary.com on

The Silver Lining of Rust Belt Manufacturing Decline Matthew E. Kahn* Department of Economics, Columbia Uni¨ ersity, New York, New York 10027 E-mail: [email protected] Received March 18, 1998; revised December 10, 1998 Between 1969 and 1996 manufacturing employment declined by 32.9% in the Rust Belt, leading to severe dislocations in cities specialized in manufacturing. However, one silver lining of reduced manufacturing activity is improved environmental quality. This paper exploits a unique merger of air quality and county manufacturing data to quantify manufacturing’s pollution intensity by industry and then uses these estimates to judge which Rust Belt cities experienced large environmental improvements. It uses valuation estimates to assess the economic magnitude of the pollution reduction. 䊚 1999 Academic Press

1. INTRODUCTION Increased foreign competition and lower demand for such products as steel has led to sharp declines in manufacturing in the Rust Belt.1 Whereas manufacturing employment in the country as a whole grew by 1.4% between 1969 and 1996, in the Rust Belt manufacturing employment fell by 32.9%. The recovery of Rust Belt cities may be slow. Declining manufacturing opportunities lead to reduced wages and increased poverty rates ŽJacobson, Lalonde, and Sullivan w27x, Neal w31x, and Wilson w43x..2 But, a silver lining of reduced manufacturing activity is improved environmental quality. For cities endowed with highly polluting industries, local quality of life may improve significantly because of manufacturing decline. * Assistant Professor of Economics and International Affairs, Columbia University and U.S Census Research Associate. I thank the Editor and two anonymous referees for numerous useful comments. In addition, I thank Richard Arnott, Joyce Cooper, Dora Costa, Mike Cragg, Ed Glaeser, and Larry Goulder for helpful comments. The opinions and conclusions expressed in this paper are those of the author and do not represent those of the U.S. Bureau of the Census. All papers are screened to ensure that they do not disclose confidential information. All errors are mine. 1 I define the Rust Belt as Illinois, Indiana, Michigan, New Jersey, New York, Ohio, Pennsylvania, and West Virginia. 2 Neal w31x reports a 31% wage premium in the primary metals industry over retail sales. Jacobson, Lalonde, and Sullivan’s w27x findings suggest that some displaced manufacturing workers suffer wage losses of $10 per hour as they transfer to the service sector. 360 0094-1190r99 $30.00 Copyright 䊚 1999 by Academic Press All rights of reproduction in any form reserved.

RUST BELT MANUFACTURING DECLINE

361

The decline in Rust Belt manufacturing offers the opportunity to study the relationship between local air quality and manufacturing activity. Because the decline in the Rust Belt was so large and because counties differed in their manufacturing composition, this allows me to identify the impact of manufacturing activity on local air quality. I use a unique merger of air quality and county manufacturing data to examine pollution intensity of manufacturing activity by two digit SIC industry. In addition, I test for the presence of cross-county pollution spillovers by studying the impact of reduced manufacturing on pollution levels in neighboring counties. My findings have implications for the future of Rust Belt cities, a future that partially depends upon the quality of life they offer ŽGlaeser w21x.. Cross-city quality of life studies have documented that households accept lower wages and pay higher rents in areas with better quality of life ŽGyourko and Tracy w23x, Blomquist, Berger, and Hoehn w9x.. By documenting the negative correlation between local air quality levels and local manufacturing activity, this paper links industry activity to city amenity levels and quality of life. If manufacturing has a large impact on local environmental quality and people value a clean environment, then Rust Belt cities such as Pittsburgh may experience sharp improvements in quality of life.3 Local environmental improvements arising from manufacturing’s decline may partially offset the loss of high paying manufacturing jobs. As local quality of life improves, Rust Belt cities might experience an influx of highly educated workers. Such workers offer city wide externalities which may accelerate such a city’s adjustment to a negative sectoral demand shock ŽRauch w36x..4 This paper is organized as follows. Section 2 presents the empirical framework for estimating the benefits of declining manufacturing. Section 3 discusses the data sources. Section 4 presents regression estimates of the size of pollution reductions attributable to reduced manufacturing. Section 5 measures the quality of life effects caused by such pollution reductions and Section 6 concludes. 3

The November 11, 1996 issue of Fortune reports, ‘‘Once upon a time, Pittsburgh was all soot, steel, and Steelers. But the smokestacks have given way to glass towers as the city has gone from working class to classy. As a home to eight FORTUNE 500 companies it still has plenty of economic strength. It’s just that in the new Pittsburgh, only 3% of the work force earns a living producing primary metals.’’ Fortune magazine ranks Pittsburgh as the 噛9 ‘‘Best City’’ to live in. Using 1980 Census data, Gyourko and Tracy w23x rank Pittsburgh’s quality of life as 噛52 of 130 cities. 4 In the aftermath of a negative local labor market demand shift, wages and rents will fall and unemployment rates will rise. There is a continuing debate on how long it takes the regional adjustment mechanism of migrants exiting and new firms moving in to converge to the long run regional equilibrium ŽTopel w39x, Blanchard and Katz w8x, Treyz, Rickman, Hunt, and Greenwood w40x..

362

MATTHEW E. KAHN

2. AN EMPIRICAL FRAMEWORK FOR ESTIMATING THE BENEFITS OF DECLINING MANUFACTURING This paper estimates pollution production functions to study the marginal pollution impact of different manufacturing industries. In the pollution production regressions, a data point is a county within a given year. For example, I fit the ambient air pollution levels in Cook County Illinois ŽChicago. for 1982 as a function of economic activity within the county. Defining SICŽ h. as total economic activity in an SIC two digit industry named ‘‘h’’ Žsuch as primary metals., Eq. Ž1. presents a model to explain pollution levels in county j at time t. log Ž pollution jt . s

Ý Bh )SICŽ h . jt q ␺ ) X jt q ⑀ jt

Ž 1.

h

X is a vector of county and state level attributes that affect local pollution levels and ⑀ is the error term.5 All estimates of Eq. Ž1. include state fixed effects to control for such factors as climate, regulatory environment and electric utility activity and calendar year dummies.6 Estimates of B in Eq. Ž1. indicate each industry’s marginal impact on pollution. If an industry has a small impact on local pollution levels, then its decline cannot lead to local environmental improvements. Each industry’s pollution impact can be estimated because counties differ in their manufacturing composition.7 The Rust Belt has featured greater concentrations of manufacturing activity and has experienced greater declines in manufacturing activity. This variation is useful for tracing out the pollution consequences of manufacturing. For example, the primary metals industry ŽSIC 33. has declined sharply. Pennsylvania experienced the destruction of 71% of its primary metals jobs between 1967 and 1987. This decline has been attributed to high labor costs, increased foreign competition, the growth of southern minimills and lower demand ŽCrandall w15x, Barnett and Crandall w2x, Beeson w5x, and Beeson and Giarratani w6x.. I estimate Eq. Ž1. using OLS. A county’s total manufacturing level at a point in time is assumed to not be a function of pollution levels within that county. A recent environmental regulation literature has explored whether regulatory enforcement, by raising the relative costs of doing business in a county, might encourage new manufacturing plants to locate in less regu5

An implicit assumption in estimating Eq. Ž1. is that all manufacturing within a given two digit SIC code is fungible Ži.e., has the same impact on county air pollution.. 6 Freedman and Jaggi w20x provide detailed case studies of emission activity at 100 major electric utility plants. 7 Manufacturing agglomerations are discussed in detail in Henderson w25x, Ellison and Glaeser w19x, and Dumais, Ellison, and Glaeser w18x.

RUST BELT MANUFACTURING DECLINE

363

lated areas.8 This unintended consequence of regulation is not likely to be quantitatively important in the Rust Belt because most plants were in operation before 1970 and thus, are not subject to the most stringent environmental regulations required of new plants ŽPortney w34x.. Local government officials also can ease regulatory constraints when they threaten local employment prospects ŽDeily and Gray w15, 16x.. I have estimated Eq. Ž1. instrumenting for county manufacturing using a county’s manufacturing activity five years earlier in a given two digit SIC industry and find no evidence warranting concern about this variable’s endogeneity. In estimating Eq. Ž1., I treat each county as a ‘‘bubble’’ such that local air quality is solely a function of activity within its borders. Neighboring counties’ economic activity can certainly reduce quality of life in bordering counties. To study the role of cross-boundary pollution spillovers, I use the Contiguous County File ŽICPSR tape 噛09835. to create for each county a measure of total manufacturing activity in adjacent counties. This variable is called ‘‘Border’’ in Eq. Ž2.. log Ž pollution jt . s

Ý Bh )SICŽ h . jt q ␺ ) X jt q ␪ )Borderjt q ⑀ jt Ž 2. h

In estimating Eq. Ž2., I test whether, controlling for state fixed effects and for a county’s own level of manufacturing activity in a given industry, there is evidence of cross-boundary spillovers. Regression estimates of Eqs. Ž1. and Ž2. are used to determine which counties enjoy the greatest environmental gains from manufacturing decline and which neighbors of highly polluting counties experience improved quality of life. 3. DATA Estimation of Eqs. Ž1. and Ž2. requires information on a county’s ambient air quality and proxies for economic activity within the county. This paper focuses on pollutants associated with manufacturing activity. One environmental indicator is ambient total suspended particulates. Total suspended particulates ŽTSP. has been one of the six ambient 8 For an overview of regulation’s effect on manufacturing see Gray w22x. Berman and Bui w7x use plant level regulatory data and find no evidence that environmental regulation reduces firm labor demand. Controlling for nonmanufacturing employment growth, county manufacturing growth from 1982᎐1988 was 14% lower in counties that were not in attainment with the Clean Air Act’s particulate standard in 1977 relative to counties that did not monitor air quality ŽKahn w28x.. Henderson w26x finds that chemical plants were less likely to locate in ozone nonattainment areas.

364

MATTHEW E. KAHN

pollutants regulated under the Clean Air Act ŽPortney w34x.. Particulates are a function of manufacturing activity, coal fired electric utility plants, diesel buses, forest fires, and fugitive dust. The Environmental Protection Agency is the source of my air pollution data.9 The EPA chooses monitoring locations to identify which areas are not in attainment of the Clean Air Act standards so that it can impose more stringent regulation to bring these areas into compliance. EPA monitoring intensity varies across states.10 Using data from the EPA’s Aerometric Information Retrieval System ŽAIRS. data base, I construct each county’s annual ambient particulate level measured in micrograms per cubic meter. Most monitoring stations measure air quality every six days. The particulate data set includes data on counties in 1977, 1072 counties in 1982 and 903 in 1987. Average employment in counties in which the EPA monitored ambient particulates was roughly 10 times employment in counties that the EPA did not monitor, suggesting that the EPA is concentrating its efforts in more populated areas featuring higher levels of economic activity. To measure county manufacturing activity, I use micro plant level data on manufacturing activity from the Census Bureau’s Longitudinal Research Database ŽLRD., which is a panel data set of economic variables collected from manufacturing establishments in the Census of Manufacturers and Annual Survey of Manufacturers programs. The LRD file contains establishment level identifying information on the factors of production and the products produced ŽLRD Technical Documentation Manual w30x.. For each plant in the United States, the LRD data includes geographical and industry identifiers for each plant and each plant’s total value shipped, total employment in 1967, 1972, 1977, 1982, and 1987. For each plant in a given two digit SIC category, I add up total value shipped by county and deflate this using the Bartelsman and Gray w4x price deflator for value of shipments. This manipulation yields real total value shipped for each countyrindustry in each year. I scale this by county land area to create the density of total value shipped for each countyrindustry. This variable’s units are $10 million dollars per square mile measured in 1987 dollars. Total value shipped is a better proxy for measuring production over time than using a countyrindustry’s total employment. Over time, a plant may change its capital to labor ratio substituting towards capital. Thus, changes in total employment would overestimate the decline in plant 9 Crandall w13x reports evidence of a downward trend in particulates in the 1970s Žp. 18.. I thank Michael Greenstone for kindly providing me with the 1977 particulate data. Starting in 1987, the EPA switched to monitoring a subset of smaller particulates PM-10. 10 In California in 1981, there was one particulate monitoring station for every 161,000 people while in Ohio there was one particulate monitoring station for every 32,000 people. There is at least one particulate monitoring station in 35% of all counties. In 1981, Los Angeles county had 10 stations while Allegheny county ŽPittsburgh . had 20 stations.

365

RUST BELT MANUFACTURING DECLINE TABLE 1 County Level Summary Statistics

County level variables

Mean

Standard deviation

Particulates Total value shipped per square mile SIC 28 Žchemicals. Total value shipped per square mile SIC 29 Žrubber and petroleum. Total value shipped per square mile SIC 30 Žrubber and plastics. Total value shipped per square mile SIC 32 Žstone, clay, and glass. Total value shipped per square mile SIC 33 Žprimary metals. Total value shipped per square mile SIC 34 Žfabricated metals. Total value shipped per square mile SIC 35 Žindustrial metals. LogŽnonmanufacturing employment per square mile in 1000s.

56.00 0.046 0.022 0.012 0.0094 0.024 0.028 0.030 y5.56

19.99 0.198 0.142 0.034 0.028 0.095 0.099 0.089 1.74

Note: The county is the unit of analysis. Manufacturing’s units are tens of millions of 1987 dollars total value shipped per square mile of county land area. The data set’s time period covers 1977, 1982, and 1987. There are 2879 observations overall. Counties which did not measure particulates are not included in the data set.

production.11 The summary statistics for the pollution and manufacturing activity variables are reported in Table 1. 4. FINDINGS Which Manufacturing Industries Create Pollution? Particulate levels are lower in counties with less manufacturing activity. In 1982, the average particulate level for counties in the lowest 10% of the manufacturing activity distribution was 50.0 while it was 56.0 for counties in the top 10% of the manufacturing distribution. Although manufacturing encompasses a wide range of activities, the LRD data allow me to disaggregate manufacturing by two digit SIC industry to study which industries have the largest pollution impacts. The empirical work focuses on the seven manufacturing industries which have the largest impact on local particulate levels. Table 2 reports seven OLS regressions of the county level pollution production function presented Eq. Ž1..12 Each specification includes state fixed effects whose coefficients are suppressed. In each regression in Table 11 Regression estimates of Eq. Ž1. where county manufacturing activity is proxied for using industry total employment rather than total value shipped are available on request. These regressions yield very similar results to those presented below. In 1977, the plant level correlation between total employment and total value shipped for SIC 33 plants was 0.92 and it was 0.89 in 1987. 12 The regressions are weighted by the number of monitoring stations within a county in a given year.

SIC 29 0.088 Ž0.081 . 0.0038 Ž0.088 . 4.17 Ž0.012 . 2879 0.39 1.3

1.3

3.3

y0.246 Ž0.015 . y0.213 Ž0.016 .

SIC 28 0.168 Ž0.029 . y0.0022 Ž0.039 . 4.16 Ž0.012 . 2879 0.40 3.3

y0.0245 Ž0.0147 . y0.217 Ž0.016 .

State fixed effects included

1.9

SIC 30 1.88 Ž0.234 . y1.34 Ž0.330 . 4.15 Ž0.012 . 2879 0.41 6.4

y0.242 Ž0.015 . y0.219 Ž0.016 .

State fixed effects included

4.8

SIC 32 1.304 Ž0.185 . 0.447 Ž0.392 . 4.14 Ž0.012 . 2879 0.42 3.6

y0.235 Ž0.014 . y0.211 Ž0.015 .

State fixed effects included

3.3

SIC 33 0.903 Ž0.102 . y0.553 Ž0.112 . 4.14 Ž0.012 . 2879 0.44 8.5

y0.231 Ž0.014 . y0.200 Ž0.015 .

State fixed effects included

3.1

SIC 34 0.360 Ž0.074 . y0.043 Ž0.091 . 4.14 Ž0.012 . 2879 0.42 3.6

y0.238 Ž0.014 . y0.210 Ž0.016 .

State fixed effects included

3.9

SIC 35 0.360 Ž0.046 . 0.080 Ž0.085 . 4.15 Ž0.012 . 2879 0.41 3.2

y0.248 Ž0.015 . y0.220 Ž0.016 .

State fixed effects included

Note: Standard errors in parentheses. Each column of this table reports a separate regression where the dependent variable is the log of a county’s ambient particulate level measured in micrograms per cubic meter. The omitted category is the 1977 year dummy.

Observations R2 Estimated % increase in particulates caused by a one standard deviation increase in this industry’s activity for counties outside of the Rust Belt Estimated % increase in particulates caused by a one standard deviation increase in this industry’s activity for counties in the Rust Belt

Constant

Total value shippedU rust belt dummy

Two digit industry included in the regression Total value shipped for industry

1987 year dummy

Independent variables 1982 year dummy

State fixed effects included

logŽpollution jt . s ␺ j q B m )SIC Ž m . jt q ⑀ jt

TABLE 2 Particulate Pollution Regressions Using One Industry to Proxy for County Manufacturing

366 MATTHEW E. KAHN

RUST BELT MANUFACTURING DECLINE

367

2, I include a different indicator of a county’s manufacturing activity. For example, the left most column of Table 2 includes a county’s manufacturing density in SIC 28. I interact the total manufacturing activity variable with a Rust Belt dummy variable to study differential pollution effects in the Rust Belt and outside of the Rust Belt. I concentrate on the seven two digit SIC manufacturing industries which have the largest impact on local particulate levels. With the exception, of SIC 29 Žrubber and petroleum., I find that all the other manufacturing indicators have a positive and statistically significant impact on particulates. The year dummies indicate that, holding manufacturing activity constant, particulates levels fell a little over 20% between 1977 and 1982 and were flat between 1982 and 1987. This reduction is likely due to the phase out of older buses, declining use of coal as a household fuel, and reduced electric utility emissions. Table 2’s last two rows show the impact of an extra standard deviation of an industry’s activity on particulates in and outside of the Rust Belt. In the Rust Belt, an extra standard deviation of SIC 28 Žchemicals. activity raises particulate levels by 3.3% which is approximately the same impact for SIC 33 Žprimary metals. and SIC 34 Žfabricated metals. and SIC 35 Žindustrial machinery.. SIC 32 Žstone, clay, and glass. has the largest Rust Belt effect of 4.8%. While statistically significant, these effects are small and suggest that only in those counties which experience large reductions in these manufacturing sectors will we observe large environmental improvements. A surprising result is that the Rust Belt interaction terms for SIC 30, and SIC 33 are negative and statistically significant, indicating that a marginal increase in manufacturing in these industries has a larger impact outside of the Rust Belt than in the Rust Belt. I had expected to find that the marginal pollution generated per unit of Rust Belt manufacturing would be higher because its capital stock is older. It is possible that because these plants are located in densely populated areas, they have faced more regulatory pressure to invest in emissions reductions. To further study manufacturing’s role in pollution production, Table 3 presents additional regression estimates of Eq. Ž1. which include all seven two digit SIC manufacturing activity levels in the same specification. The left column of Table 3 shows that controlling for state fixed effects and year fixed effects, SIC 32, 33, and 35 have a statistically significant impact on particulate levels. The results indicate that an extra standard deviation of SIC 33 increases particulates by 3.1%. The right column of Table 3 presents the same specification of Eq. Ž1. but adds to the specification the log of nonmanufacturing employment within the county. This variable controls for other economic activity in the county unrelated to manufacturing. Including this variable in the regression affects the coefficient estimates on the manufacturing proxies. SIC 32 and 33 still have a positive and statistically significant impact on particulates but the coefficient on

368

MATTHEW E. KAHN TABLE 3 Particulate Pollution Regression logŽpollution jt . s ␺ j q Ý Bm )SICŽ m. jt q ⑀ jt

Independent variables 1982 year dummy 1987 year dummy logŽnonmanufacturing employment density. Total value shipped in SIC 28 Total value shipped in SIC 29 Total value shipped in SIC 30 Total value shipped in SIC 32 Total value shipped in SIC 33 Total value shipped in SIC 34 Total value shipped in SIC 35 Constant Observations R2

State fixed effects included

State fixed effects included

y0.232 Ž0.014. y0.206 Ž0.015.

y0.236 Ž0.014. y0.215 Ž0.015. 0.055 Ž0.0054. 0.034 Ž0.026. y0.036 Ž0.029. y0.264 Ž0.1113. 0.299 Ž0.127. 0.272 Ž0.045. y0.083 Ž0.059. y0.047 Ž0.053. 4.42 Ž0.029. 2879 0.48

0.014 Ž0.027. y0.0012 Žy0.029. 0.228 Ž0.170. 0.610 Ž0.144. 0.311 Ž0.050. y0.044 Ž0.060. 0.142 Ž0.048. 4.12 Ž0.011. 2879 0.44

Note: Standard errors in parentheses. Each column of this table reports a separate regression where the dependent variable is a county’s ambient particulate level measured in micrograms per cubic meter. The omitted category is the 1977 year dummy.

SIC 32 falls by half. SIC 33s coefficient remains unchanged. Perhaps due to multicollinearity, the coefficients for SIC 30 and SIC 35 are now negative. The coefficient on nonmanufacturing employment density indicates that a 1% increase in this variable raises particulate levels by 5.5%.13

13

In results that are available on request, I have estimated the regressions in Table 3 using an instrumental variables approach. I use a county’s manufacturing level five years before as an instrument for current manufacturing activity. My estimates of Eq. Ž1. using two stage least squares yield very similar results as compared to OLS estimates using the same sample.

RUST BELT MANUFACTURING DECLINE

369

E¨ idence of Cross Boundary Spillo¨ ers The results presented in Tables 2 and 3 focus solely on explaining county pollution as a function of activity within that county’s borders. Using the LRD data to create total economic activity in adjacent counties, I estimate Eq. Ž2. to quantify cross-boundary spillovers. If such spillovers are large, then counties adjacent to heavy manufacturing counties will experience a windfall from reduced neighbor activity.14 Spillovers are also important to quantify because much environmental regulation is carried out at the state level. In the absence of Coasian bargaining, a state regulator may not be concerned if pollution is ‘‘blowing down wind’’ to a neighboring state. Estimation results for Eq. Ž2. are presented in Table 4. Table 4 presents seven regressions and mirrors Table 2 with the exception that the Rust Belt dummy variable interaction has been dropped. I include in the regression specification the sum of the bordering counties’ total value shipped for that two digit SIC Žmeasured in $10 billion 1987 dollars.. The left column of Table 4 regresses the log of particulates on state fixed effects, year fixed effects, a county’s density of SIC 28 and the sum of SIC 28 manufacturing activity in adjacent counties. Across the seven specifications, I find evidence that particulates are an increasing function of own county’s SIC manufacturing and I also find evidence of cross-boundary spillovers. For the primary metals industry ŽSIC 33., an extra standard deviation of local manufacturing increases pollution by 3.5% while an extra standard deviation of neighbor primary metals activity increases pollution by 1.1%. SIC 32 features the largest spillover impact of 4.1%. Measuring The Quality of Life Impact of Declining Manufacturing The previous section used the merger of ambient pollution data with county manufacturing data to quantify the impact of different manufacturing industries on own county and neighboring county pollution levels. While certain industries such as SIC 33 and SIC 32 consistently have a positive effect on pollution, the size of the coefficients is such that only counties which experienced significant reductions in manufacturing would experience sharp pollution reductions. The Rust Belt has experienced large activity reductions in primary metals ŽSIC 33.. The primary metals industry is highly centralized in the Rust Belt. In 1967, 6.9% of the nation’s manufacturing jobs were in SIC 33 industry and 67.4% of these jobs were in New York, New Jersey, Pennsyl14

Between 1977 and 1987, Westmoreland county in Pennsylvania experienced a particulate level drop from 144.6 to 50.7. This county has only a small level of manufacturing activity but is adjacent to a high manufacturing county ŽAllegheny. whose output has fallen sharply between 1977 and 1987.

0.7 1.9

2.8 1.7

% increase in county particulates caused by a standard deviation of county production in this SIC % increase in county particulates caused by a standard deviation of adjacent county production in this SIC 3.2

2.15

0.633 Ž0.228 . 0.595 Ž0.108 . 4.15 Ž0.012 . 2879 0.41

y0.239 Ž0.015 . y0.221 Ž0.016 . SIC 30

State fixed effects included

4.1

3.3

1.25 Ž0.153 . 1.28 Ž0.178 . 4.15 Ž0.012 . 2879 0.43

y0.225 Ž0.014 . y0.211 Ž0.015 . SIC 32

State fixed effects included

1.1

3.5

0.371 Ž0.049 . 0.093 Ž0.049 . 4.14 Ž0.012 . 2879 0.43

y0.229 Ž0.014 . y0.198 Ž0.015 . SIC 33

State fixed effects included

2.7

2.9

0.297 Ž0.040 . 0.222 Ž0.045 . 4.15 Ž0.012 . 2879 0.42

y0.233 Ž0.015 . y0.207 Ž0.015 . SIC 34

State fixed effects included

3.5

3.0

0.335 Ž0.040 . 0.251 Ž0.046 . 4.14 Ž0.012 . 2879 0.42

y0.252 Ž0.015 . y0.231 Ž0.016 . SIC 35

State fixed effects included

Note: Standard errors in parentheses. Each column of this table reports a separate regression where the dependent variable is a county’s ambient particulate level measured in micrograms per cubic meter. The units for total value shipped for all adjacent counties is in 10 billions of 1987 dollars. The omitted category is the 1977 year dummy.

Observations R2

0.051 Ž0.032 . 0.127 Ž0.051 . 4.15 Ž0.012 . 2879 0.39

y0.243 Ž0.015 . y0.211 Ž0.016 . SIC 29

0.142 Ž0.019 . 0.082 Ž0.036 . 4.15 Ž0.012 . 2879 0.40

y0.243 Ž0.0147 . y0.217 Ž0.016 . SIC 28

State fixed effects included

Own County’s total value shipped Total value shipped for all adjacent counties Constant

Two digit industry included in the regression

1987 year dummy

Independent variables 1982 year dummy

State fixed effects included

logŽpollution jt . s ␾ l q B m )SIC Ž m . jt q ␺ m )Border Ž m . jt q ⑀ jt

TABLE 4 Cross-County Pollution Spillovers

370 MATTHEW E. KAHN

371

RUST BELT MANUFACTURING DECLINE TABLE 5 Employment Trends in SIC 33 ŽPrimary Metals.

State

1967

1972

1977

1982

1987

Nation 1281 1143 1114 854.1 701.1 Illinois 108.6 98 89 60.7 43.2 Indiana 111 103 102 86.2 60.3 New Jersey 37.6 31 21 21.6 18.3 New York 72.9 58 53 38.2 22 Ohio 169.6 142 134.2 97.6 65.6 Pennsylvania 233.1 186 172 118.7 67.3 Six State’s share of 57.2% 54.1% 51.2% 49.4% 39.5% national employment in SIC 33

1967 to 1987 % change y45.3 y60.2 y45.7 y51.3 y69.8 y61.4 y71.1

Note: Employment expressed in 1000s. Data Source: Printed Records of the Census of Manufacturers Geographic Area Series.

vania, Ohio, Indiana, Illinois, Michigan, and Wisconsin. Between 1967 and 1987, employment in primary metals plants ŽSIC 33. fell by 45.3% from 1.28 to 0.701 million jobs. At the same time, the Rust Belt experienced a 62% decline in employment in primary metals plants with the Rust Belt’s share of employment in this industry falling from 57.2% to 39.5% Žsee Table 5.. No other Rust Belt industry experienced such sharp contractions. The environmental benefits of reduced manufacturing is a function of how much pollution declines and of household willingness to pay for a reduction in pollution. Since pollution is a local public bad, the aggregate benefits are also a function of the size of the population exposed to the pollution. To estimate the extent of the decline in pollution brought about by reduced manufacturing activity, I present estimates of actual particulate declines versus predicted particulate declines for five Rust Belt cities. Table 6’s left column reports each area’s actual decline in ambient particulates between 1977 and 1987. The right column uses the regression model presented in the right column of Table 6 to predict how the change in a county’s economic activity between 1977 and 1987 affects the county’s particulate levels. The right column of Table 6 reports the change in a county’s particulate levels attributable to changes in manufacturing activity and nonmanufacturing employment. Manufacturing’s decline triggered a reduction of Pittsburgh pollution of 28.7 units. This represents roughly half of the total reduction in Pittsburgh’s pollution level reduction of 60.7. Changes in manufacturing activity in Chicago reduced its pollution levels by 10.5 units and Youngstown experienced a reduction of 9.7 units.

372

MATTHEW E. KAHN TABLE 6 Pollution Reductions in Some Rust Belt Counties

Geographical area

1977᎐1987 actual decline in particulate levels

1977᎐1987 predicted decline in particulate levels given the change in economic activity

Pittsburgh, PA Gary, IN Chicago, IL Detroit, MI Youngstown, OH

y60.7 y14.8 y16.4 y29.5 y57.2

y28.7 y7.7 10.5 y6.6 y9.7

Note: This table’s left column reports the area’s decline in ambient particulates between 1977 and 1987. The right column uses the regression model presented in the right column of Table 3 to predict the change in an area’s pollution level which is attributable to the decline in economic activity. Particulate are measured in micrograms per cubic meter.

How much would each Pittsburgh household be willing to pay for a 28.7 unit reduction in particulates? Epidemiologists have studied the impact of particulate exposure on morbidity and mortality. Based on the meta-analysis by Dockery and Pope w17x, Pittsburgh’s particulate decline would reduce mortality rates by 1.5%.15 Given that the United States’ mortality rate was 8.8 per thousand in 1975, the pollution decline would save roughly 50 statistical lives per year in a county with a population of 500,000. The compensating differentials literature’s estimates of particulate valuation offer a second method for valuing improved air quality. In accord with basic compensating differentials theory, areas with higher pollution levels feature lower home prices. Hedonic price estimates are informative about consumer valuation but do not uniquely identify consumer willingness to pay to avoid pollution ŽPalmquist w32x.. Both within city and across city, hedonic housing price studies have documented particulate capitalization ŽSmith and Huang w38x, Blomquist, Berger, and Hoehn w9x, Clark and Nieves w12x, Gyourko and Tracy w23x.. Evaluating over 80 city level hedonic studies which estimated the implicit capitalization of particulates into home prices, Smith and Huang’s w38x meta-analysis reports a mean home

15 Particulate exposure also increase morbidity. Ransom and Pope w35x study daily Utah hospital admissions from 1985᎐1991 for breathing problems caused by small particulate matter caused by the local steel mill. They have a ‘‘natural experiment’’ because this steel mill is the only major producer of particulates in the area and because for one year the mill shut down due to a labor dispute. When the steel mill was open, the area averaged 12.6 violations of the 24 hour particulate standard while when the mill was closed the particulate standard was never violated.

RUST BELT MANUFACTURING DECLINE

373

price capitalization of $100 per particulate unit and a median of $22 Ž$1982 dollars..16 In a recent paper, Chay and Greenstone w11x find a much larger impact of particulates on home prices. Unlike cross-sectional hedonic estimates, their methodology uses a county level panel data set of home prices and particulates. They regress the change in home prices on the change in county particulates and instrument for county pollution levels using a county’s environmental regulatory status as an instrumental variable. They find that a one unit reduction in particulates results in about a 1% increase in home values.17 Thus, based on the mean hedonic home price capitalization estimated by Smith and Huang w38x the Pittsburgh home’s price would appreciate by $2,800. Using Chay and Greenstone’s estimates and assuming the average home price is $60,000, the Pittsburgh air quality improvement would lead home prices to rise by roughly $15,000. While there may be a silver lining of reduced polluting activity in the Rust Belt, the costs of achieving this amenity gain may be regressive.18 Air quality has improved because manufacturing plants have reduced activity. Air quality gains will tend to increase home values but reduced economic opportunities will lead to lower county home prices. Home owners who work in the manufacturing industry might simultaneously lose their job and experience a decline in their home’s price.19 The likely winners from reduced manufacturing activity would be footloose service firms and retired renters seeking local higher quality of life. As manufacturing activity declines other environmental margins are likely to improve. I have focused solely on manufacturing’s impact on ambient particulates. There are also other environmental margins which are affected by primary metals manufacturing. For example, Toxic Release Inventory data ŽU.S. EPA w42x. indicate that the primary metals industry is the second largest producer of production related waste ŽSIC 28 chemicals is the largest.. Further research could link manufacturing to other environ16 Within city hedonic studies have generated larger particulate valuation estimates than research using cross-city data. In their cross-city study, Blomquist, Berger, and Hoehn w9x report a yearly full price of $0.36 per unit of particulates. 17 Chay and Greenstone w11x are able to replicate the smaller cross-sectional particulate capitalization estimates found by researchers using these techniques. 18 The incidence of local land improvement depends on cross-city migration costs and on how the amenity improvement was achieved ŽHarrison and Rubinfield w24x, Ackerman w1x, Pines and Weiss w33x.. 19 This paper has not explored within county pollution variation. This would require detailed information on the location of air quality monitoring stations and the demographics of households living near those stations. Brooks and Sethi w10x have documented using Toxic Release Inventory data at the zip code level that, even controlling for income measures, blacks are exposed to more pollution than other demographic groups.

374

MATTHEW E. KAHN

mental attributes. 20 Ozone is an important pollutant that is not analyzed in this paper.21 5. CONCLUSION Based on a spatial merger of county level ambient air quality data and micro manufacturing data, this paper has linked local pollution levels to local manufacturing activity levels in two digit SIC industries. I found that Rust Belt counties that had a high concentration of primary metals activity experienced significant improvements in environmental quality as measured by particulates. I also found evidence of significant cross-county pollution spillovers. Reduced manufacturing activity in one county lowers pollution levels in adjacent counties. My results suggest that a full costrbenefit snalysis of the regional transition from manufacturing to services must account both for the economic dislocations experienced by cities and for the improvements in quality of life. The silver lining of Rust Belt manufacturing decline is that residents of the densely populated cities of the Rust Belt will be exposed to lower levels of pollution. Because the environmental cost of living in these cities has fallen over time, such cities may increasingly attract high skill workers and firms. Rust Belt cities, however, have an older capital infrastructure and colder climate than other regions, so whether reduced pollution levels will be enough to attract more high skilled workers than in the past remains a topic for future research.

REFERENCES 1. S. R. Ackerman, On the distribution of public program benefits between landlords and tenants, Journal of En¨ ironmental Economics and Management, 5, 167᎐184 Ž1977.. 2. D. Barnett and R. Crandall, ‘‘Up from the Ashes: The Rise of the Steel Minimill in the United States,’’ Brookings Institution Ž1986.. 3. D. Barnett and L. Schorsih, ‘‘Steel: Upheaval in a Basic Industry,’’ Ballinger Ž1983.. 4. E. Bartelsman and W. Gray, The nber manufacturing productivity database, technical working paper 噛205, NBER Ž1996.. 5. P. Beeson, Sources of the decline in manufacturing in large metropolitan areas, Journal of Urban Economics, 28, 71᎐86 Ž1990.. 20

One environmental margin that I have explored is ambient sulfur dioxide. Coal fired electric power plants are the major contributor to ambient sulfur dioxide. In results that are available on request, I find that primary metals plants do not have a positive impact on this pollutant. I find strong evidence of a temporal decline in sulfur dioxide per unit of electric utility coal consumption. 21 Henderson w26x documents which industries have contributed to local ozone problems. These include: plastics ŽSIC 282,307., industrial organic chemicals ŽSIC 286., steel ŽSIC 331. and petroleum refining ŽSIC 291..

RUST BELT MANUFACTURING DECLINE

375

6. P. Beeson and F. Giarratani, Spatial aspects of capacity change by U.S. integrated steel producers, Journal of Regional Science, 38, 425᎐444 Ž1998.. 7. E. Berman and L. Bui, Environmental regulation and labor demand: Evidence from the South Coast Air Basin, NBER working paper 噛6299 Ž1997.. 8. O. J. Blanchard and L. F. Katz, Regional evolutions, Brookings Papers on Economic Acti¨ ity, 0Ž1., 1᎐75 Ž1992.. 9. G. Blomquist, M. Berger, and J. Hoehn, New estimates of quality of life in urban areas, American Economic Re¨ iew, 78, 89᎐107 Ž1988.. 10. N. Brooks and R. Sethi, The distribution of pollution: community characteristics and exposure to air toxics, Journal of En¨ ironmental Economics and Management, 32, 233᎐250 Ž1997.. 11. K. Chay and M. Greenstone, Does air quality matter? Evidence from the housing market, mimeo Ž1998.. 12. D. Clark and L. Nieves, An interregional hedonic analysis of noxious facility impacts on local wages and property values, Journal of En¨ ironmental Economics and Management, 27, 235᎐253 Ž1994.. 13. R. Crandall, ‘‘Controlling Industrial Pollution: The Economics and Politics of Clean Air,’’ Brookings Institution, Washington, D.C. Ž1983.. 14. R. Crandall, ‘‘Manufacturing on the Move,’’ Brookings Institution, Washington, D.C. Ž1993.. 15. M. Deily and W. Gray, Enforcement of pollution regulation in a declining industry, Journal of En¨ ironmental Economics and Management, 21, 260᎐274 Ž1991.. 16. M. Deily and W. Gray, Compliance and enforcement: air pollution regulation in the U.S. steel industry, Journal of En¨ ironmental Economics and Management, 31, 96᎐111 Ž1996.. 17. D. Dockery and A. Pope, Epidemiology of acute health effects: summary of time-series studies, in ‘‘Particles in Our Air: Concentrations and Health Effects,’’ R. Wilson and J. Spengler ŽEds.., Harvard University Press Ž1996., pp. 123᎐148. 18. G. Dumais, G. Ellison, and E. Glaeser, Geographic concentration as a dynamic process, NBER Working Paper 噛 6270 Ž1997.. 19. G. Ellison and E. Glaeser, Geographic concentration in U.S. Manufacturing industries: a dartboard approach, Journal of Political Economy, 105, 889᎐927 Ž1997.. 20. M. Freedman and B. Jaggi, ‘‘Air and Water Pollution Regulation: Accomplishments and Economic Consequences,’’ Quorum Books Ž1993.. 21. E. Glaeser, Are cities dying?, Journal of Economic Perspecti¨ es, 12Ž2., 139᎐160 Ž1998.. 22. W. Gray, Manufacturing plant location: does state pollution regulation matter?, NBER Working Paper 噛5880 Ž1997.. 23. J. Gyourko and J. Tracy, The structure of local public finance and the quality of life, Journal of Political Economy, 99, 774᎐806 Ž1991.. 24. D. Harrison and D. Rubinfield, The distribution of benefits from improvements in urban air quality, Journal of En¨ ironmental Economics and Management, 5, 313᎐332 Ž1978.. 25. V. Henderson, Efficiency of resource usage and city size, Journal of Urban Economics, 19, 47᎐70 Ž1986.. 26. V. Henderson, The effect of air quality regulation, American Economic Re¨ iew, 86, 789᎐813 Ž1996.. 27. L. Jacobson, R. Lalonde, and D. Sullivan, Earnings losses of displaced workers, American Economic Re¨ iew, 83, 685᎐705 Ž1993.. 28. M. Kahn, Particulate pollution trends in the united states, Regional Science and Urban Economics, 27, 87᎐107 Ž1997.. 29. J. Kohlhase, The impact of toxic waste sites on housing values, Journal of Urban Economics, 30, 1126 Ž1991..

376

MATTHEW E. KAHN

30. Longitudinal Research Database, Technical Documentation Manual, US Department of Commerce, Bureau of Census Ž1992. 31. D. Neal, Industry specific human capital: evidence from displaced workers, Journal of Labor Economics, 13, 653᎐677 Ž1995.. 32. R. Palmquist, Hedonic methods, in ‘‘Measuring the Demand for Environmental Quality. Contributions to Economic Analysis,’’ no. 198, J. B. Braden and C. D. Kolstad ŽEds.., Elsevier, New York, Ž1991.. 33. D. Pines and Y. Weiss, Land improvement projects and land values, Journal of Urban Economics, 3, 1᎐13 Ž1976.. 34. P. Portney, Air Pollution policy, in ‘‘Public Policies for Environmental Protection,’’ P. Portney ŽEd.., Resources for the Future, Washington D. C. Ž1990.. 35. M. Ransom and C. Pope, External health costs of a steel mill, Contemporary Economic Policy, 13, 86᎐97 Ž1995.. 36. J. Rauch, Productivity gains from geographic concentration of human capital: evidence from the cities, Journal of Urban Economics 34, 380᎐400 Ž1993.. 37. C. Russell and W. Vaughan, ‘‘Steel Production; Processes, Products, and Residuals,’’ Resources for the Future, Washington, D.C. Ž1976.. 38. V. K. Smith and J. Huang, Can hedonic models value air quality? A meta-analysis, Journal of Political Economy, 103, 209᎐227 Ž1995.. 39. R. Topel, Local labor markets, Journal of Political Economy, 94 Žsupplement., S111᎐143, Ž1986.. 40. G. Treyz, D. Rickman, G. Hunt, and M. Greenwood, The dynamics of us internal migration, Re¨ iew of Economics and Statistics, 75, 209᎐214 Ž1993.. 41. United States Environmental Protection Agency, National Air Quality and Emissions Trends Report 1988, EPA-450r4-90-002 ŽMarch 1990.. 42. United States Environmental Protection Agency, 1994 Toxic Release Inventory Public Data Release, EPA-749-R-96-02 ŽJune 1996.. 43. W. Wilson, ‘‘The Truly Disadvantaged,’’ University of Chicago Press Ž1987..