Research
950 Main Street • Worcester, MA 01610
Tel: 508-793-7711 •

Active Learning and Research
Economics professor Wayne Gray and undergraduate research assistant Melanie Lajoie examined the impact of environmental regulations on the pulp and paper industry.

Printable Format

Technology change, emissions reductions, and productivity

A paper presented at the Allied Social Science Association meetings in New Orleans (AERE session) on January 6, 2001 by Wayne B. Gray (Clark University and NBER) and Ronald J. Shadbegian (University of Massachusetts at Dartmouth, Ph.D. Clark University)

Introduction
Air pollution in the paper industry
Determinants of air pollution emissions
Data description
Results
Conclusions
References
Tables
Footnotes
Acknowledgements

Introduction

The past 30 years have seen significant improvements in US environmental quality, driven in large part by reductions in industrial emissions. This paper provides indirect evidence on the reasons for emissions reductions over time, by looking at the experience of plants in the pulp and paper industry. We focus on the differences in air pollution emissions across plants, explaining differences in emissions using differences in air pollution abatement expenditures, local regulatory stringency, the mix of end-of-pipe and change-in-process air pollution abatement investments, and the productivity level of the plant.

Much of the early empirical research on the impact of environmental regulation concentrated on the relationship between productivity and reported pollution abatement costs. Denison (1979) used growth accounting to calculate the expected impact of regulation on productivity. Gray (1986,1987) compared all manufacturing industries and found that high-abatement-cost industries had a bigger productivity slowdown in the 1970s.Barbera and McConnell (1986,1990) first looked at time-series variation in a few selected industries and found some impacts of abatement cost on productivity, and then looked at indirect effects on productivity as regulation changed the use of other inputs (especially energy). Gray and Shadbegian (1995) found that plants with higher abatement costs had lower productivity levels, though Berman and Bui (2001) found little impact of abatement costs on productivity for oil refineries.[1] Becker and Henderson (1999) and Greenstone (1999) use plant-level production data combined with air pollution attainment status, finding that plants in non-attainment areas had significantly higher costs.Gollop and Roberts (1983) use information about regulatory stringency at individual electric utilities to show a significant impact of regulation on productivity. Becker (2000) also finds evidence that reported abatement costs may not reflect all of the information about differences in regulatory stringency contained in the county non-attainment status data.

There has also been some examination of the benefits of regulation, relating air quality to emissions from manufacturing facilities.Shadbegian, et. al. (2000) find that differences in abatement costs across paper mills are related to the benefits of pollution reductions at the mill. Kahn (1999) quantifies an improvement in air quality associated with plant closings in certain high-polluting industries. Henderson (1996) finds a significant impact of county attainment status on peak air quality measures in non-attainment areas.

We use confidential annual plant-level Census data from the Longitudinal Research Database (LRD) to develop a comprehensive database of 68 U.S. paper mills. We have used this data in earlier papers, studying the impact of environmental regulation on plant-level productivity and investment in Gray and Shadbegian (1995,1998). Using the LRD, we can identify each plant's production, investment, productivity, age, and production technology. We have plant-level air pollution abatement capital investments from the Pollution Abatement Costs and Expenditures (PACE) survey, which allows us to create an air pollution abatement capital stock, and also to distinguish between end-of-pipe and change-in-process investments. To the Census data we add plant-level air pollution emissions information from various EPA datasets in the 1980s, the dependent variable for our analyses. We also add characteristics of the plant's production technology, taken from the Lockwood Directory, and measures of local regulatory stringency faced by the plant.

In our simplest cross-plant analysis, we find that aggregate emissions (a weighted average of particulates and sulfur dioxide) are significantly lower in plants with a larger air pollution abatement capital stock. We also find that end-of-line abatement capital seems to be more clearly connected with lower emissions than is change-in-process abatement capital. A 10 percent increase in abatement capital stock appears to reduce annual emissions by about 2.5 percent. Translating these impacts into dollars suggests a reasonable return: one dollar of abatement capital stock providing an annual return of about 30 cents in pollution reduction benefits.

Local regulatory stringency appears to play an important role in determining a plant's emission level. Plants in non-attainment counties have an average of 23 percent lower emissions than those in attainment counties, all else equal. A plant's productivity is also associated with emissions, with 10 percent higher productivity being associated with 2.5 percent lower emissions. A plant with older or less productive equipment stock, as proxied by the age or speed of its paper machines, tends to have higher emissions.

Our analysis also yielded several unexpected results, which may influence the interpretation of the results. Chief among these are the positive coefficients on air pollution abatement operating costs (higher operating costs being associated with higher pollution emissions) and the impact of adding fixed-effects to the model (shifting the sign of air pollution abatement capital from significant negative to significant positive). Perhaps these have to do with reverse causality: higher emissions, for whatever reason, creating a need for greater pollution abatement expenditures. In any event, the results related to local regulatory stringency and plant-specific productivity are more robust, with stricter local regulation and higher productivity associated with lower emissions in all specifications.

When we examine emissions of particulates and sulfur dioxide separately, we find some differences between the two. For example, in our data cases of county non-attainment are primarily due to excessive concentrations of particulates (not sulfur dioxide). Therefore it is not surprising that particulate emissions are more affected by non-attainment . We also find that abatement capital is more associated with reducing sulfur dioxide emissions than particulates: this may be connected to differences in the abatement techniques used for the two pollutants.

Section 2 provides some information about the generation and regulation of air pollution in the paper industry. Section 3 presents a brief model of the determinants of pollution emissions, in conjunction with a plant's productivity and technology. Section 4 describes the data used in the analysis. Section 5 provides the results, and section 6 concludes the paper.

Air Pollution in the Paper Industry

Pulp and paper mills are major sources of both air and water pollution. The key distinction for production technology among paper mills is whether or not the plant begins the papermaking process with a pulping stage or not. Pulping plants begin the process with trees, separating out the wood fibers by a variety of chemical and mechanical methods. non-pulping mills can begin with purchased pulp, or with recycled paper to provide the fibers for the paper. During the papermaking stage, a combination of fiber and water is set on a wire mesh and passed through several sets of steam-heated dryers to dry into paper.

Air pollution is generated primarily during the pulping process. Most pulping mills incorporate large boilers, either power boilers to generate energy for the pulping process and steam for the papermaking dryers, or recovery boilers to recycle chemicals in some pulping techniques. Pulping mills have a convenient supply of fuel, the remaining parts of the trees after the wood fiber is extracted, and cogeneration plants are common, with boilers generating high-pressure steam to run electric power turbines. The resulting low-pressure steam used in the papermaking dryers. Non-pulping mills are more likely to purchase their energy inputs or use small boilers to provide needed energy and steam. The papermaking process can produce some air pollution, as some paper is chemically treated to produce smoother surfaces, but this is definitely less serious than the air pollution created by pulping.

This study covers air pollution emissions during the 19791990 period. During this period, air pollution in the U.S. was regulated by the 1977 amendments to the 1970 Clean Air Act. Each year, U.S. counties are designated as 'attainment' (meeting ambient air quality standards) or 'non-attainment ' (violating ambient air quality standards) for each of several criteria pollutants. Plants located in non-attainment counties face substantially more stringent regulation than those in attainment counties, with limitations on new plant openings and modifications of existing plants. These regulatory pressures can come from both state environmental agencies and the federal EPA, which oversees activity in non-attainment areas more closely. In addition to county-level differences based on attainment status, there can be differences across states in their regulatory stringency, determined by such factors as the state's budget constraint and political support for environmental issues within the state.

During the 1980s, regulatory attention for air pollution from paper mills was concentrated on sulfur dioxide and particulate emissions. Regulators also focused their attention on the emissions of volatile organic compounds (VOC), contributors to ozone pollution, but paper mills are not generally major sources of VOCs. Thus the relevant attainment status designations are for particulates and sulfur dioxide, and emissions of those pollutants are likely to face more regulatory pressures, and therefore be lower, in non-attainment counties.

An important feature of the regulatory process is the grand-fathering of existing plants. For the most part, existing manufacturing plants were not subject to regulations as stringent as those applied to new facilities. This is at least partly justified by the extreme difficulty in retrofitting existing facilities, designed before pollution control was a major priority. For example, reducing air pollution emissions can involve capturing vapors from the production process and burning them in a recovery boiler; in one older mill this required installing hundreds of yards of extra piping because the recovery boiler was located in a distant building. Thus older production facilities are likely to have higher emissions for two reasons: they face less regulatory pressure, and pollution abatement is more difficult for them. As facilities are updated and rebuilt with newer equipment, some of these drawbacks of being older may be reduced, as the new equipment is likely to incorporate some pollution-reducing characteristics. Updating an existing facility can also bring the plant under closer scrutiny by regulators, as some facility renovations lead the facility to be treated as if it were a 'new' source.

Determinants of Air Pollution Emissions

The determinants of the level of air pollution emissions from a given paper mill can be separated into two groups:

(1) EMIT = f( EMIT*() , ABATE() ),

factors influencing the amount of 'uncontrolled' emissions, EMIT*(), generated by the production process in the absence of any special efforts by the plant to abate pollution, and factors influencing the fraction of that pollution that is abated, ABATE(), before it reaches the environment. The first category includes the plant's size, capital intensity, production technology, and age. The second category includes the plant's air pollution control equipment, air pollution abatement operating expenses, regulatory pressures facing the plant, and the overall efficiency of the plant's management in achieving its abatement goals.

Consider the first set of factors, those affecting uncontrolled emissions, in more detail:

(2) EMIT* = EMIT*(SHIP, PRODCAP, PULP, AGE).

The pollution generating process is assumed to have constant returns, relative to the plant's output level, or shipments (SHIP), so that doubling a plant's output would be expected to double its pollution. Given that both the EMIT and SHIP variables are measured in log form, this would translate into a coefficient of unity on SHIP. There may be some economies of scale in air pollution abatement, which in our categorization of factors would cause the plant's output to appear also in the determinants of ABATE. In this case, the coefficient on SHIP would be less than one.

Plants with more production capital (PRODCAP), on average, are likely to generate more air pollution. This occurs because most air pollution arises from burning fossil fuels for energy. Capital-intensive plants tend to require more energy to operate than labor-intensive ones, with larger power boilers and more air pollution. As noted earlier, the key element of paper mill technology that affects air pollution emissions is whether or not the plant incorporates a pulping process (PULP), with pulping mills having substantially higher emissions.

The age of the plant is also likely to affect emissions, with newer plants (or plants that have undergone substantial revisions) being designed to reduce emissions. Given current regulatory constraints, there was by the 1980s no way for a U.S. paper plant to buy 'dirty' papermaking equipment, or to design a dirty upgrade to a pulp facility. This variable also spans the two categories to some degree: we choose to consider age as a factor affecting uncontrolled emissions EMIT* before the plant does its abatement, but age could instead be treated as reflecting difficulties on the ABATE side of the equation, though the prediction remains the same greater age should be associated with greater emissions.

Now consider the determinants of air pollution abatement:

(3) ABATE = ABATE(AIRCAP, AIRPAOC, NONATTAIN, VOTE, TFP).

A key element affecting the ability of the firm to reduce air pollution is the plant's level of air pollution abatement capital (AIRCAP). During the study period 19791990, most of the abatement of air pollution was done with large capital equipment: scrubbers and precipitators connected to smokestacks. This represents so-called 'end-of-line' pollution abatement (AIRCAPEOL). An alternative type of capital investment to reduce pollution is 'change-in-process ' abatement (AIRCAPCIP), where the production process is redesigned to some degree to reduce pollution. change-in-process investment is more difficult to quantify, since it requires allocating the overall investment between the goals of production and abatement, so it may be subject to more errors of measurement (tending to bias its coefficient towards zero in the empirical analysis). However, CIP investment may also reflect a greater willingness on the part of the firm to be creative in its abatement procedures, and thus may be associated with greater abatement success.

In addition to capital expenditures for air pollution abatement, operating costs may also play an important role. Without proper maintenance and operation, pollution control equipment may fail to operate as designed. There may also be areas where labor (e.g. more workers to check for and fix process failures) may be substitutable for capital in air pollution abatement. Thus we would expect AIRPAOC, as well as AIRCAP, to be positively associated with abatement efficiency, and therefore lead to lower emissions.

Regulatory pressures are also likely to contribute to the extent of pollution abatement at a plant. It could be argued that regulatory pressure should be treated as a factor affecting the plant's desire to abate air pollution, operating through the plant's choice of values for AIRCAP and AIRPAOC rather than independently entering the ABATE equation. Given our imperfect measures of the plant's actual abatement efforts, as captured by AIRCAP and AIRPAOC, some component of mis-measured abatement activity is likely to remain, and could be captured by measures of regulatory stringency. Alternatively, if there is some degree of ordinary inefficiency in the allocation of resources to pollution abatement, facing a high degree of regulatory scrutiny is likely to focus the plant's attention on abatement issues, reducing this inefficiency and increasing abatement. As noted earlier, the main indicator of the air pollution regulatory stringency faced by a plant is the attainment status of the county in which the plant is located, NONATTAIN. Another variable, VOTE, measuring differences across states in their political support for environmental regulations is also included, to capture possible state-level factors.

Finally, we consider the possibility that plants may have different overall efficiency, possibly due to managers with different abilities. Some may be especially good in motivating their workers, gaining extra output (or abatement) from the same amount of resources. Here we use the total factor productivity of the plant, TFP, to indicate overall efficiency, either of the plant manager or of the entire plant workforce. We would expect this to increase abatement efficiency, but it need not do so. It is possible, for example, that some managers concentrate on regulatory issues, while others concentrate on production: in this case the two types of efficiency would be substitutes rather than complements. Having high productivity achieved by speeding up and cutting corners in the production process could provide another reason for higher productivity to be associated with a greater probability of emissions, through more frequent accidental releases of emissions (although such a temporary increase in emissions is unlikely to be captured in our emissions measures).

The resulting equation for estimation includes both emission and abatement components:

(4) EMIT = f(SHIP, PRODCAP, PULP, AGE, AIRCAP, PAOC, NONATTAIN, VOTE, TFP)

Since we have data on productive efficiency available for each plant in each time period, we can focus on the unexplained portion of emissions and see if that is related to the unexplained portion of efficiency. This is done through a seemingly unrelated regression (SUR) model, where equation (4), excluding TFP, is estimated simultaneously with a TFP equation:

(5) TFP = f(PAOC, NONATTAIN, VOTE, PULP, AGE).

Here productive efficiency is explained by pollution abatement spending (using total pollution abatement operating costs for all media, not just air pollution), regulatory pressures (which may limit the plant's ability to adjust its production process), technology, and age. If more efficient producers are also more efficient at air pollution abatement, we would expect to see a negative correlation between the estimated residuals from equations (4) and (5) generated as part of the SUR analysis.

Data Description

Our research was carried out at the Census Bureau's Boston Research Data Center, where we can access confidential Census databases developed by the Census's Center for Economic Studies. The principal source for our sample of plants is the Longitudinal Research Database (LRD). The LRD contains annual information on a large sample of individual manufacturing plants from the Census of Manufactures and Annual Survey of Manufacturers over time (for a more detailed description of the LRD data, see McGuckin and Pascoe (1988)). The LRD includes data on each plant's real inventory-adjusted shipments (SHIP), labor, materials, and investment spending. From the LRD data we can calculate a productivity index, TFP, for each plant.[2] Using LRD data we also calculate a measure of the plant's total capital stock, TOTCAP, based on a standard perpetual inventory calculation.

We combine the LRD data with another plant-level Census data source: the Pollution Abatement Costs and Expenditures (PACE) survey, conducted annually by the Census Bureau. The PACE questionnaire is sent to a subset of firms in the Annual Survey of Manufactures, over-sampling high-pollution plants such as paper mills. We require the plants in our sample to have PACE data in all available years from 19791990, shrinking our dataset from the 116 plant sample used in much of our prior research to a sub-sample of 68 plants with complete pollution abatement investment data from 19791990. [3]

Using this PACE data, we calculate the stock of total pollution abatement capital (POLCAP) and air pollution abatement capital (AIRCAP) in place at each plant over time.[4] The 'productive' capital stock of the plant, PRODCAP, is calculated as the difference between TOTCAP and POLCAP. The PACE survey also distinguishes between end-of-line investment and change-in-process investment spending: these data are used to separate the overall air pollution abatement capital into AIRCAPEOL and AIRCAPCIP. Finally, the PACE survey contains information on air pollution abatement operating costs (AIRPAOC), which is also tested to see whether it is related to the plant's air pollution emissions. We also use PAOCRAT, total pollution abatement operating costs as a fraction of the plant's peak capacity (an average of the highest two years of shipments) in the analysis, as an explanatory variable possibly affecting the plant's productivity (PAOCRAT was found to be significantly related to TFP levels in Gray and Shadbegian, 1995).

We combine the LRD and PACE data with two other plant-level information sources: the Lockwood Directory and various EPA air pollution datasets. The Lockwood Directory is an annual listing of pulp and paper mills, from which we extracted several pieces of information about the plants' production technology. The Lockwood Directory includes information on the production technology being used at each mill, which we use to construct a PULP dummy (indicating that the plant begins its papermaking process with raw wood). We also gather information on the age and operating speed of each paper machine in operation at a mill and use a weighted average of these values in our analysis, PMAGE and PMSPEED.[5]

In addition to the plant-level data, we use two measures of the local regulatory stringency faced by the plant. The stringency of each state's pollution abatement effort is proxied by VOTE, the League of Conservation Voters' pro-environment voting score for the state's Congressional delegation during each congressional session. In prior research, VOTE was found to be significantly related to manufacturing plant location decisions (Gray, 1997) and investment decisions (Gray and Shadbegian, 1998).

An alternative measure of regulatory stringency specific to air pollution regulation is NONATTAIN, a dummy variable indicating whether the plant was located in a county that failed to attain the ambient air quality standards for particulates or sulfur dioxide. The attainment status of each county is published annually in the Federal Register, and there is some variation in attainment status over time.[6] For the paper mills in our sample, non-attainment status is nearly always due to excessive particulates; sulfur dioxide non-attainment is much less common.

The key dependent variable for our analyses, the plant's air pollution emissions, comes from various EPA regulatory datasets which span the 1980s. We combine information from several years of the Compliance Data System (CDS) with 1980 and 1985 data from the National Emissions Data System (NEDS). For the post1985 period the Aerometric Information Retrieval System (AIRS) contains air emissions data. We should note that these datasets do not provide a continuously varying measure of emissions. They vary no more often than annually, and often several years go by between changes in the emissions values. This reflects two factors. First, much of the early emissions data we use, especially in the CDS, was intended for categorizing plants as major or minor sources (above or below 100 tons per year of a given pollutant), rather than for assembling a complete history of emissions, so less attention was paid to regularly collecting updates. Second, most plants are reporting a 'calculated' emissions number, based on the capacity of the production process times the expected emissions per unit of capacity times the design efficiency of the installed pollution control equipment. This number is less likely to change over time than would a continuously-measured stream of data on actual emissions of each pollutant.

The emissions data is provided separately for the major criteria air pollutants. Our analysis focuses on particulates and sulfur dioxide, since they are most commonly reported and were the major focus of air pollution regulation for this industry during the time period. We measure the emissions of each pollutant, PT and SO2, in tons per year and then aggregate them together into EMIT, weighting them using a measure of the relative health damages for the two pollutants (based on Shadbegian, et. al (2000), one ton of PT emissions = 2.45 tons of SO2 emissions).[7]

Results

Table 1 presents summary statistics and variable descriptions for all the variables included in the analysis. Examining the dependent variables, we see that the greater weight placed on emissions of particulates (2.45:1) is more than offset by the greater mean emissions of sulfur dioxide (789:3355), so SO2 emissions account for about 60% of the total value of EMIT. Each of the emission variables shows substantial variability across observations, with the mean being exceeded by the standard deviation in each case.

The air pollution abatement capital stock at a typical plant is substantial, approximately 8% of the productive capital stock at the plant: $4.46 million of $54.6 million. The average value of the total pollution abatement capital stock is $8.5 million, so air pollution abatement requires over half of the plants' abatement capital. Of this, about two-thirds represents end-of-line spending and one-third is change-in-process spending. Pollution abatement operating costs represent 1.64% of total costs, with the majority of those costs going to abating water pollution: about twice as large as air pollution costs see Shadbegian, et. al. (2000).

Of the plants in our sample, slightly less than half are located in counties that are in non-attainment for either particulates or sulfur dioxide as noted earlier, nearly all of these cases of non-attainment refer to particular concentrations rather than to sulfur dioxide concentrations. Slightly over half of our sample are pulping mills, beginning the papermaking process with raw fiber from trees. Many of the plants are quite old, with an average plant's paper machines dating back to 1940 (60 years before the year 2000).

We also include a measure of the cross-sectional nature of the variables, %CS, among the descriptive statistics. This shows that nearly all of the variation in the dependent and independent variables used in our analysis can be explained by plant dummies. Since many of our variables are likely to be measured with some error (which may be providing some of the within-plant variation), this may complicate the estimation process making it difficult to identify the determinants of changes over time, and forcing us to rely on primarily cross-sectional analyses.

Turning to the empirical results in Table 2, we see that plants which incorporate a pulping process generate significantly more air pollution, as do more capital-intensive plants. We also find that emissions are roughly proportional to the plant's output. Older plants show a slight tendency for greater air emissions. There is also a sizable decline in emission levels over the decade.

Looking at the abatement-related variables, plants with larger stocks of air pollution abatement capital (all else equal) tend to have lower emissions. When we distinguish between end-of-line and change-in-process capital, it is end-of-line that appears to be more related to emissions reduction, though both types of abatement capital have negative coefficients. Being located in a non-attainment county also tends to be associated with lower emissions, even though we are controlling for the air abatement capital in place. The VOTE variable shows a slight negative relationship to emissions, but the effect is insignificant.

Consider the coefficient on air pollution abatement capital stock, about .247. This indicates that a plant with a 10 percent larger air pollution abatement capital stock, all else equal, would have about 2.5 percent lower emissions. Is this a worthwhile investment? At the mean level of AIRCAP in our data, a 10 percent increase would cost about $450,000. This would achieve a reduction in annual aggregate air pollution emissions of about 1300 tons. We can compare these benefits and costs, using information from Shadbegian, et. al. (2000) which includes a calculation of the health benefits per ton of reduced emissions which translates into about $1000 benefits per ton in these terms.[8] Thus a $450,000 investment yields annual benefits of $130,000, about a 30 percent annual rate of return on the investment. This benefit-cost ratio is noticeably less than that found in Shadbegian, et. al. (2000), where the benefits from air pollution abatement were calculated at about 20 times the costs, but part of that difference is explained by the earlier paper concentrating on abatement operating costs rather than capital stocks (so both benefits and costs are in terms of annual dollars).

Table 3 adds air pollution abatement operating costs to the analysis. The total air pollution abatement capital stock, end-of-line air abatement capital, and non-attainment variables maintain or strengthen the negative relationships they had with emissions, as seen earlier. However, operating costs are significantly positively related to emissions, an unexpected result. Higher emissions could lead to higher paperwork costs, but ordinarily one would expect that spending money for pollution abatement should reduce pollution, whether the spending is for capital expenses or operating costs. In the following tables we omit the abatement operating cost measures; including them tends to strengthen the negative coefficient on the air pollution abatement capital stock, but the puzzling positive coefficient on AIRPAOC remains consistent in all specifications.

Table 4 shifts the analysis to plant productivity measures. Plants with higher productivity have significantly lower emissions. The PMSPEED variable shows a similar impact: plants with faster paper machines tend to have significantly lower emissions recalling that paper machine speed is not causally related to emissions, but serves as a proxy for the technology incorporated in the plant's operations overall. Note that once we control for a plant's productivity, the PMAGE coefficient changes sign from the earlier tables, though it is not statistically significant. The differences between change-in-process and end-of-line capital become much smaller after controlling for the plant's productivity level.

Table 5 presents the results from a fixed-effect analysis, applied to a variety of the earlier models. The local regulatory stringency measures, NONATTAIN and VOTE, are significantly stronger, and the productivity and scale effects remain strong. However, the air abatement capital results switch signs in the fixed-effect models: within a plant's data, years with a high air abatement capital stock are associated with more, not less, emissions. This result is consistent when we test a wide variety of similar models. Note that the productivity effects, while still significant, are now substantially smaller than they were in the earlier regressions. This indicates that part of the explanation for the large TFP effects in Table 4 could be related to cross-sectional differences in efficiency (or other correlated factors) across plants.

Table 6 considers a seemingly unrelated regression model, allowing for a correlation between the unexplained variation in emissions and in productivity across observations. Not surprisingly, given our earlier results for TFP, we find a significant negative correlation in the ordinary regression model. This result persists across the models estimated, although when we incorporate fixed-effects into the analysis the correlation becomes insignificant. The individual estimated coefficients are similar to those found earlier, with AIRCAP, NONATTAIN, PMAGE, lower output, and nonPULP associated with lower emissions, as in the ordinary regressions. As before, the results for output and local regulatory stringency retain their values in the fixed-effects model, while the abatement capital results change sign. In the productivity equation, we find that pollution abatement expenditures are associated with lower productivity. Plants in non-attainment areas also have somewhat lower productivity, even after allowing their overall pollution abatement operating costs to have a direct effect on their productivity through PAOCRAT, similar to the results found in Becker (2000).

In Tables 711 we present much of the same analysis, disaggregating emissions into particulate and sulfur dioxide. There are notable differences in the estimated coefficients across the two pollutants. In the basic models of Table 7, sulfur dioxide is substantially more sensitive to air pollution abatement capital, to end-of-line capital, and to PMAGE. Particulates, on the other hand, are more sensitive to local regulatory stringency, whether measured by VOTE or NONATTAIN. Both pollutants show the same surprising positive relationship with AIRPAOC, although including AIRPAOC in the regressions largely eliminates the differences between the two pollutants in terms of the impact of air pollution abatement capital, either in terms of the total impact or the differences in impact between end-of-line and change-in-process capital.

Both the fixed-effect models (Table 9) and the seemingly unrelated regression models (Tables 1011) show similar results to those seen earlier. Most notably, AIRCAP is again surprisingly positively related to emissions in the fixed-effects model. Also as before, the measures of local regulatory stringency are stronger for particulates than for sulfur dioxide, although both pollutants show some significant effects. Finally, the plant productivity measure is once again marginally negative, and in the seemingly unrelated regression models in Tables 1112 we again find a negative relationship between the residuals from the emissions and productivity equations.

Conclusions

We have examined the relationship between air pollution emissions at paper mills and their air pollution abatement capital stock, local regulatory stringency, and productivity. In our simplest regression analysis, we find that aggregate emissions (a weighted average of particulates and sulfur dioxide) are significantly lower in plants with a larger air pollution abatement capital stock. We also find that end-of-line abatement capital seems to be more clearly connected with lower emissions than is change-in-process abatement capital. A 10 percent increase in abatement capital stock appears to reduce annual emissions by about 2.5 percent. Translating these impacts into dollars suggests a reasonable return: one dollar of abatement capital stock providing an annual return of about 30 cents in pollution reduction benefits.

Local regulatory stringency also appears to play an important role in determining a plant's emission level. Plants in non-attainment counties have an average of 23 percent lower emissions than those in attainment counties, even controlling for the plant's abatement capital stock. A plant's productivity performance is also associated with its emissions, with 10 percent higher productivity being associated with 2.5 percent lower emissions. A plant with older or less productive equipment stock, as proxied by the age or speed of its paper machines, tends to have higher emissions.

Our analysis also yielded several unexpected results, which may influence the interpretation of the results. Chief among these are the positive coefficients on air pollution abatement operating costs (higher air pollution operating costs associated with higher air pollution emissions) and the impact of adding fixed-effects to the model (shifting the sign of air pollution abatement capital from significantly negative to significantly positive). Perhaps these have to do with reverse causality: higher emissions, for whatever reason, creating a need for greater pollution abatement expenditures. In any event, the results related to local regulatory stringency and plant-specific productivity are more robust than those for air pollution capital stock, with stricter local regulation and higher productivity associated with lower emissions in all specifications.

We plan future research in this area, particularly incorporating more recent air pollution emissions data, and adding data for water pollution emissions which are measured directly. This may help identify whether the puzzling differences between abatement operating costs and capital stocks can be explained by our reliance on emission numbers that are, for the most part in the 1980s, calculated based on design efficiencies of the pollution abatement capital equipment installed at the plant, rather than direct measurement of emissions.

References

Barbera, Anthony J. and Virginia D. McConnell, "Effects of Pollution Control on Industry Productivity: a Factor Demand Approach." Journal of Industrial Economics, December 1986, 161-72.

, "The Impact of Environmental Regulations on Industry Productivity: Direct and Indirect Effects." Journal of Environmental Economics and Management, January 1990, 5065.

Bartelsman, Eric J. and Wayne B. Gray, "The NBER Manufacturing Productivity Database." NBER Technical Working Paper 205, October 1996.

Becker, Randy A., "Air Pollution Abatement Costs and County non-attainment Status: What Does the PACE Survey Tell Us?" Presented at the International Atlantic Economic Conference, October, 2000

Becker, Randy A. and J. Vernon Henderson. "Effects of Air Quality Regulations on Polluting Industries," Journal of Political Economy, 108(2), 379421, April 2000.

Becker, Randy A. and J. Vernon Henderson. "Costs of Air Quality Regulation," NBER Working Paper 7308, August 1999.

Berman, Eli and Linda Bui, "Economic Regulation and Productivity: Evidence from Oil Refineries," Review of Economics and Statistics, forthcoming 2001.

Denison, Edward P., Accounting for Slower Economic Growth: The U.S. in the 1970s. Washington: The Brookings Institution, 1979.

Gollop, Frank M. and Mark J. Roberts, "Environmental Regulations and Productivity Growth: the Case of Fossil-Fueled Electric Power Generation." Journal of Political Economy, August 1983, 654-74.

Gray, Wayne B., "The cost of regulation: OSHA, EPA and the Productivity Slowdown." American Economic Review, December 1987, 998-1006.

, "Productivity Versus OSHA and EPA Regulations." UMI Research Press, Ann Arbor, MI, 1986.

Gray, Wayne B. and Ronald J. Shadbegian, "Environmental Regulation, Investment Timing, and Technology Choice." Journal of Industrial Economics, June 1998, 235256.

, "Pollution Abatement Costs, Regulation, and plant-level Productivity," NBER Working Paper 4994, January 1995.

Greenstone, Michael. "The Impacts of Environmental Regulations on Industrial Activity: Evidence from the 1970 and 1977 Clean Air Act Amendments and the Census of Manufactures," University of California at Berkeley, mimeograph, 1999.

Henderson, J. Vernon. "Effects of Air Quality Regulation," American Economic Review, 86(4), 789813, September 1996.

Kahn, Matthew E. "The Silver Lining of Rust Belt Manufacturing Decline," Journal of Urban Economics, 46(3), 360376, November 1999.

LockwoodPost Pulp and Paper Directory, Miller-Freeman Publishing Company, various issues.

McGuckin, Robert H. and George A. Pascoe, "The Longitudinal Research Database: Status and Research Possibilities." Survey of Current Business, November 1988.

Shadbegian, Ronald J., Wayne B. Gray, and Jonathan Levy, "Spatial Efficiency of Pollution Expenditures", Presented at the Western Economic Association Meetings (June 1999); Harvard University's Kennedy School of Government (March 2000); and the National Bureau of Economic Research (April 2000).

U.S. Bureau of the Census, "Pollution Abatement Costs and Expenditures," U.S. Govt. Printing Office, Washington, DC, various issues.

U.S. Environmental Protection Agency, "The Benefits and Costs of the Clean Air Act, 1970 to 1990," October, 1997.

Tables

Table 1: Summary Statistics
816 observations (68 plants * 12 years)
All values are expressed in 1972 dollars, using the paper industry (SIC 2621) price deflator from Bartelsman and Gray (1996).
%CS = percent of variation that is cross-sectional (Rsquared of regression on plant dummies); '*' indicates fixed (%CS = 100)

Plant Air Pollution Emissions
Variable	Mean	Std. Dev.	%CS	Description
PT	789.06	1315.42,	.73	Particulate emissions (tons/yr)
SO2	3354.80	3927.02	.87	Sulfur Dioxide emissions (tons/yr)
EMIT	5300.11	6024.12	83	Weighted sum (PT*2.455 + SO2)

Pollution Abatement Spending
Variable	Mean	Std. Dev.	%CS	Description
POLCAP	8511.44	8720.56	.89	Total pollution abatement capital
AIRCAP	4458.62	5088.66	.84	Air pollution abatement capital
AIRCAPEOL	2905.97	3241.95	>.9	Air pollution end-of-line
AIRCAPCIP	1552.65	2541.26	.56	Air pollution change-in-process
AIRPAOC	965.10	1526.57	.76	Air pollution operating costs
PAOCRAT	1.64	1.21	.79	Total PAOC / peak plant capacity

Local Regulatory Stringency
Variable	Mean	Std. Dev.	%CS	Description
NONATTAIN	.39	.49	>.9	County non-attainment for PT or SO2
VOTE	57.86	15.69	.80	LCV pro-environment voting index

Plant Characteristics
Variable	Mean	Std. Dev.	%CS	Description
PMSPEED	1634.92	554.82	>.9	Average paper-machine speed (fpm)
PMAGE	64.19	25.34	*	Average paper-machine age (in 2000)
PULP	.56	.50	*	Plant includes pulping process
PRODCAP	54562.03	49080.74	>.9	Productive capital stock
SHIP	46490.71	28233.23	.89	Value of annual plant shipments
TFP	89.57	21.10	.34	Total factor productivity index

Variables in Logs, for Regressions
LPT	5.14	2.15	Log(PT)
LSO2	6.81	2.37	Log(SO2)
LEMIT	7.38	2.31	Log(EMIT)
LAIRCAP	7.75	1.22	Log(AIRCAP)
LAIRCAPEOL	7.32	1.24	Log(AIRCAPEOL)
LAIRCAPCIP	6.61	1.22	Log(AIRCAPCIP)
LAIRPAOC	11.38	4.47	Log(AIRPAOC)
LSHIP	10.57	.61	Log(SHIP)
LPRODCAP	10.53	.91	Log(PRODCAP)

Table 2: Aggregate Emissions--Abatement Capital and Regulatory Measures(Dep. Var. = LEMIT; 816 observations)

All regressions include dummies for missing PMAGE data.

A B C D E F

CONSTANT -6.574
(-7.05) -6.315
(-5.75) -5.948
(-6.35) -6.875
(-6.96) -6.615
(-5.78) -6.276
(-6.33)

LAIRCAP -0.247
(-2.63) -0.247
(-2.62) -0.246
(-2.68)

LAIRCAPEOL -0.219
(-2.29) -0.221
(-2.31) -0.205
(-2.14)

LAIRCAPCIP -0.066
(-0.66) -0.064
(-0.64) -0.080
(-0.82)

VOTE -0.002
(-0.54) -0.002
(-0.53)

NONATTAIN -0.230
(-2.20) -0.228
(-2.20)

LPRODCAP 0.43
(2.86) 0.439
(2.89) 0.397
(2.67) 0.46
(3.21) 0.47
(3.24) 0.430
(2.98)

LSHIP 1.074
(7.33) 1.051
(6.84) 1.054
(7.34) 1.080
(7.36) 1.056
(6.90) 1.065
(7.38)

PULP 0.866
(5.88) 0.856
(5.72) 0.881
(5.97) 0.884
(5.99) 0.873
(5.83) 0.898
(6.08)

PMAGE 0.004
(1.88) 0.004
(1.91)
0.004
(1.97) 0.004
(1.86) 0.004
(1.89) 0.004
(1.96)

yr80 0.071
(0.32) 0.073
(0.34) 0.066
(0.30) 0.071
(0.32) 0.073
(0.34) 0.065
(0.30)

yr81 0.000
(0.00) 0.004
(0.02) -0.004
(-0.02)
0.000
(0.00)
0.004
(0.02) -0.006
(-0.03)

yr82 -0.065
(-0.30) -0.048
(-0.22) -0.073
(-0.33) -0.068
(-0.31) -0.051
(-0.23) -0.077
(-0.35)

yr83 -0.265
(-1.14) -0.246
(-1.05) -0.274
(-1.18) -0.270
(-1.15) -0.250
(-1.05) -0.283
(-1.20)

yr84 -0.229
(-0.99) -0.210
(-0.89) -0.243
(-1.04) -0.235
(-1.00) -0.215
(-0.91) -0.253
(-1.08)

yr85 -0.703
(-3.21) -0.686
(3.09) -0.716
(-3.28) -0.712
(-3.18) -0.695
(-3.06) -0.730
(-3.27)

yr86 -0.723
(-3.16) -0.710
(-3.07) -0.730
(-3.18) -0.733
(-3.09) -0.719
(-3.01) -0.745
(-3.14)

yr87 -0.614
(-2.59) -0.595
(-2.48) -0.624
(-2.64) -0.621
(-2.56) -0.601
(-2.45) -0.636
(-2.62)

yr88 -0.680
(-2.91) -0.663
(-2.79) -0.690
(-2.96) -0.694
(-2.91) -0.676
(-2.80) -0.708
(-2.98)

yr89 -0.655
(-2.59) -0.637
(-2.49) -0.665
(-2.64) -0.676
(-2.62) -0.657
(-2.52) -0.689
(-2.69)

yr90 -0.681
(-2.68) -0.667
(-2.60) -0.691
(-2.73) -0.711
(-2.80) -0.696
(-2.71) -0.723
(-2.85)

R-squared 0.671 0.671 0.673 0.672 0.672 0.674

Table 3: Aggregate Emissions and Abatement Operating Costs
(Dep. Var. = LEMIT; 816 observations)
All regressions include dummies for missing PMAGE data and years.

A B C D E F

CONSTANT -6.577
(-6.99) -6.335
(-5.81) -5.862
(-6.21) -6.854
(-6.89) -6.605
(-5.80) -6.176
(-6.20)

LAIRCAP -0.315
(-3.27) -0.315
(-3.27) -0.318
(-3.40)

LAIRCAP-EOL -0.318
(-3.37) -0.320
(-3.40) -0.307
(-3.27)

LAIRCAP-CIP -0.040
(-0.40) -0.038
(-0.39) -0.055
(-0.57)

LAIRPAOC 0.046
(3.17) 0.046
(3.16) 0.049
(3.32) 0.049
(3.37) 0.049
(3.36) 0.051
(3.50)

VOTE -0.002
(-0.51) -0.002
(-0.52)

NONATTAIN -0.263
(-2.52) -0.258
(-2.51)

LPRODCAP 0.441
(2.84) 0.448
(2.87) 0.401
(2.64) 0.487
(3.31) 0.494
(3.35) 0.446
(3.08)

LSHIP 1.074
(7.29) 1.052
(6.82) 1.051
(7.30) 1.065
(7.19) 1.042
(6.75) 1.046
(7.20)

PULP 0.790
(5.23) 0.780
(5.11) 0.802
(5.32) 0.803
(5.34) 0.793
(5.20) 0.816
(5.42)

PMAGE 0.005
(2.04) 0.005
(2.07) 0.005
(2.15) 0.005
(2.03) 0.005
(2.06) 0.005
(2.14)

R-squared 0.676 0.676 0.679 0.678 0.678 0.680

Table 4: Aggregate Emissions and Plant Productivity
(Dep. Var. = LEMIT; 816 observations)
All regressions include dummies for missing PMAGE and PMSPEED data and years.

A B C D E F

CONSTANT -7.010
(-7.76) -7.465
(-8.25) -5.869
(-5.61) -7.406
(-7.73) -8.003
(-8.22) -6.294
(-5.77)

LAIRCAP -0.249
(-2.75) -0.353
(-3.79) -0.249
(-2.81)

LAIRCAP-EOL -0.166
(-1.67) -0.224
(-2.25) -0.154
(-1.55)

LAIRCAP-CIP -0.125
(-1.21) -0.184
(-1.64) -0.137
(-1.39)

NONATTAIN -0.241
(-2.38) -0.244
(-2.43)

VOTE -0.004
(-1.10) -0.003
(-1.06)

TFP -0.025
(-8.13) -0.025
(-8.18) -0.026
(-8.36) -0.025
(-8.06) -0.025
(-8.13) -0.026
(-8.30)

LPRODCAP 0.081
(0.56) 0.123
(0.80) 0.052
(0.37) 0.106
(0.75) 0.157
(1.03) 0.074
(0.53)

LSHIP 1.714
(11.45) 1.866
(12.57) 1.659
(10.89) 1.739
(11.37) 1.900
(12.44) 1.689
(10.92)

PULP 0.868
(6.22) 0.978
(7.53) 0.863
(6.05) 0.884
(6.33) 0.996
(7.68) 0.880
(6.17)

PMSPEED -0.460
(-7.57) -0.479
(-7.66)

PMAGE 0.000
(-0.13) -0.003
(-1.49) 0.000
(-0.01) 0.000
(-0.11) -0.003
(-1.46) 0.000
(0.00)

R-squared 0.695 0.718 0.697 0.696 0.720 0.698

Table 5: Aggregate Emissions Fixed Effect Models
(Dep. Var. = LEMIT; 816 observations)
All regressions include year dummies.

A B C D E F

LAIRCAP 0.436
(4.06) 0.508
(4.81) 0.379
(3.72)

LAIRCAP-EOL 0.224
(1.62) 0.247
(1.75) 0.205
(1.48)

LAIRCAP-CIP 0.347
(4.04) 0.424
(4.77) 0.294
(3.50)

NONATTAIN -0.971
(-5.86) -0.913
(-5.63) -0.890
(-5.46) -0.849
(-5.27)

VOTE -0.010
(-2.67) -0.009
(-2.44)

TFP -0.009
(-2.32) -0.006
(-1.61) -0.008
(-2.11) -0.006
(-1.50)

LPRODCAP 0.412
(3.06) 0.331
(2.28) 0.250
(1.79) 0.458
(3.36) 0.400
(2.75) 0.297
(2.11)

LSHIP 0.997
(6.49) 1.522
(8.42) 1.156
(6.04) 0.920
(5.76) 1.367
(7.52) 1.080
(5.68)

R-squared 0.908 0.905 0.910 0.909 0.906 0.910

Table 6: Aggregate Emissions and Productivity Seemingly Unrelated Regressions Models
(Models B1 and B2 include plant fixed effects)
(816 observations)
(* = correlation significant at 5% level)
All regressions include dummies for missing PMAGE data and years.

A1 A2 B1 B2

Dep. Var.: LEMIT TFP LEMIT TFP

CONSTANT -7.713
(-6.82) 101.676
(35.72)

LAIRCAP -0.233
(-2.93) 0.390
(4.05)

PAOCRAT -2.647
(-5.23) -0.506
(-0.72)

NONATTAIN -0.196
(-1.84) -2.088
(-1.75) -0.915
(-5.87) -1.373
(-0.62)

VOTE -0.002
(-0.52) -0.011
(-2.95)

LPRODCAP 0.155
(1.23) 0.274
(2.22)

LSHIP 1.467
(8.95) 1.076
(6.70)

PULP 0.875
(6.67) 2.055
(1.45)

PMAGE 0.003
(1.47) -0.105
(-4.01)

R-squared 0.670 0.452 0.909 0.748

Cross-Equation Residual Correlation -.21* -.06

Table 7: Specific Emissions Models Abatement Capital and Regulatory Determinants
(816 observations)
All regressions include dummies for missing PMAGE data and years.

A B C D E F

Dep. Var.: LPT LSO2 LPT LSO2 LPT LSO2

CONSTANT -8.760
(-7.57) -7.958
(-6.86) -6.523
(-4.72) -8.641
(-6.78) -6.076
(-5.21) -8.303
(-7.30)

LAIRCAP -0.135
(-1.27) -0.233
(-2.22) -0.137
(-1.30) -0.233
(-2.22) -0.132
(-1.42) -0.234
(-2.21)

NONATTAIN -1.038
(-7.84) 0.127
(1.06)

VOTE -0.016
(-4.02) 0.005
(1.25)

LPRODCAP 0.315
(1.63) 0.491
(3.33) 0.377
(1.92) 0.471
(3.18) 0.172
(1.02) 0.511
(3.34)

LSHIP 1.087
(5.60) 1.045
(6.14) 0.886
(4.24) 1.108
(6.22) 0.996
(5.49) 1.056
(6.22)

PULP 1.435
(8.91) 0.823
(4.75) 1.344
(8.48) 0.850
(4.86) 1.496
(9.70) 0.815
(4.73)

PMAGE -0.001
(-0.23) 0.008
(3.13) 0.000
(0.04) 0.008
(3.02) 0.000
(0.00) 0.008
(3.07)

R-squared 0.472 0.562 0.481 0.562 0.515 0.562

Table 8: Specific Emissions Models Abatement Operating Costs
(816 observations)
All regressions include dummies for missing PMAGE data and years.

A B C D E F

Dep. Var.: LPT LSO2 LPT LSO2 LPT LSO2

CONSTANT -9.069
(-7.37) -8.108
(-6.75) -8.743
(-7.38) -7.960
(-6.85) -9.018
(-7.21) -8.090
(-6.71)

LAIRCAP -0.246
(-2.21) -0.293
(-2.71)

LAIRCAP-EOL -0.102
(-0.98) -0.262
(-2.33) -0.263
(-2.51) -0.351
(-3.09)

LAIRCAP-CIP -0.074
(-0.61) 0.002
(0.02) -0.029
(-0.24) 0.025
(0.24)

LAIRPAOC 0.070
(4.25) 0.040
(2.43) 0.074
(4.40) 0.044
(2.65)

LPRODCAP 0.342
(1.79) 0.529
(3.72) 0.335
(1.70) 0.499
(3.31) 0.382
(1.99) 0.547
(3.80)

LSHIP 1.108
(5.66) 1.029
(5.98) 1.087
(5.52) 1.045
(6.14) 1.082
(5.46) 1.015
(5.87)

PULP 1.454
(9.01) 0.837
(4.83) 1.293
(7.77) 0.756
(4.33) 1.308
(7.85) 0.765
(4.38)

PMAGE -0.001
(-0.22) 0.008
(3.11) 0.000
(-0.15) 0.008
(3.23) 0.000
(-0.16) 0.008
(3.22)

Rsquared 0.473 0.563 0.486 0.565 0.487 0.567

Table 9: Specific Emissions Models Fixed Effect Models
(816 observations)
All regressions include year dummies.

A B C D E F

Dep. Var.: LPT LSO2 LPT LSO2 LPT LSO2

LAIRCAP 0.655
(5.35) 0.485
(3.89) 0.429
(3.83) 0.387
(3.13)

LAIRCAP-EOL 0.317
(2.45) 0.095
(0.58)

LAIRCAP-CIP 0.266
(2.70) 0.376
(3.69)

NONATTAIN -1.588
(-7.57) -0.697
(-3.36) -1.524
(-7.14) -0.626
(-3.03)

VOTE -0.016
(-3.49) -0.008
(-1.67) -0.015
(-3.37) -0.007
(-1.41)

TFP -0.009
(-2.00) -0.008
(-2.01) -0.004
(-0.93) -0.006
(-1.49) -0.003
(-0.79) -0.006
(-1.42)

LPRODCAP 0.347
(2.41) 0.405
(1.95) 0.219
(1.72) 0.343
(1.66) 0.238
(1.87) 0.434
(2.03)

LSHIP 1.623
(7.51) 1.299
(6.27) 1.023
(4.56) 1.021
(4.56) 0.941
(4.18) 0.947
(4.22)

Rsquared 0.855 0.854 0.871 0.857 0.871 0.858

Table 10: Specific Emissions Models Seemingly Unrelated Regression
(816 observations)
(* = correlation significant at 5% level)
All regressions include dummies for missing PMAGE data and years.

A1 A2 B1 B2

Dep. Var.: LPT TFP LSO2 TFP

CONSTANT -6.243
(-4.89) 102.705
(36.33) -10.705
(-7.90) 101.748
(35.75)

LAIRCAP -0.125
(-1.39) -0.221
(-2.32)

PAOCRAT -2.952
(-5.72) -2.753
(-5.40)

NONATTAIN -0.978
(-8.07) -2.098
(-1.71) 0.145
(1.14) -2.082
(-1.74)

VOTE -0.014
(-3.42) 0.004
(0.98)

LPRODCAP -0.027
(-0.19) 0.262
(1.74)

LSHIP 1.270
(6.86) 1.510
(7.69)

PULP 1.427
(9.63) 1.715
(1.22) 0.842
(5.39) 2.176
(1.54)

PMAGE -0.001
(-0.24) -0.112
(-4.20) 0.007
(2.49) -0.105
(-4.00)

Rsquared 0.518 0.435 0.560 0.452

Cross-Equation Residual Correlation

-.19* -.17*

Table 11: Specific Emissions Models Seemingly Unrelated Regression Fixed Effects
(816 observations)
All regressions include year dummies.

A1 A2 B1 B2

Dep. Var.: LPT TFP LSO2 TFP

LAIRCAP 0.437
(4.07) 0.397
(3.19)

PAOCRAT -0.610
(-0.84) -0.502
(-0.71)

NONATTAIN -1.590
(-9.16) -1.189
(-0.52) -0.699
(-3.47) -1.376
(-0.62)

VOTE