unit 7 essay

EFFECTIVENESS OF STATE POLLUTION PREVENTION PROGRAMS AND POLICIES

DONNA RAMIREZ HARRINGTON∗

States are using regulatory-, information-, and management-based policies to encourage the adoption of pollution prevention (P2) and reduce pollution. Using a sample of facilities of S&P 500 firms which report to the Toxic Releases Inventory from 1991 to 2001, this study employs dynamic panel data models to examine the effectiveness of state legislations and policies in increasing P2 and reducing toxic releases. I find that toxic waste legislations are effective in reducing toxic releases and in promoting P2, but the effect of policy instruments differ. Facilities in states with reporting requirement and mandatory planning adopt more P2 even in states that do not emphasize toxic waste reduction. The effectiveness of reporting is stronger among facilities with good environmental performance, while the potency of mandatory planning is greater among facilities with past P2 experience. In contrast, numerical goals reduce toxic pollution levels only among those which have been subjected to high levels of enforcement action. These suggest that reporting requirement and mandatory planning may be promoting the P2 practices which can improve public image and which benefit from enhanced technical know-how, but they are not causing meaningful pollution reductions, implying that the existing policies must be complemented by other approaches to achieve higher reductions in toxic pollution levels. (JEL Q55, O38, H23)

I. INTRODUCTION

Environmental policies in the United States have evolved from command and control to market-based incentives, and more recently to voluntary programs and information disclosure mechanisms. The evolution largely stems from the growing recognition that firms respond to environmental policies not only to improve com- pliance, but also to lower production costs, improve product quality, and enhance mar- ket competitiveness. Thus, prescriptive technol- ogy and emission standards are being replaced with a mix of policies that promote techniques such as pollution prevention (P2) technologies

∗The author is grateful for the comments and suggestions on an earlier draft of the paper provided by Keith Brouhle, Tom Lyon, participants of the 2009 AEA-ASSA Meetings, Robert Mohr and participants at the University of New Hampshire Department of Economics seminar series, Marc Law, and participants at the University of Vermont Depart- ment of Economics seminar series. All remaining errors are my own. Ramirez Harrington: Assistant Professor, Department of

Economics, University of Vermont, 233 Old Mill Build- ing, 94 University Place, Burlington, VT 05405. Phone +1 (802) 656-0964, Fax +1 (802) 656-8405, E-mail [email protected]

which allow firms to go beyond compliance and address strategic objectives.

In the United States, the National Pollu- tion Prevention Act of 1990 mandates that “pollution be prevented or reduced at source whenever feasible.” Beginning in 1991, the U.S. Environmental Protection Agency (EPA) began collecting annual data on 43 types of P2 activities undertaken by facilities and in doing so, it expanded the purpose of the Toxic

ABBREVIATIONS

CAA: Clean Air Act CWA: Clean Water Act EPCRA: Emergency Planning and Community Right-

to-Know Act IDEA: Integrated Data for Enforcement Analysis NPRI: National Pollutant Releases Inventory P2: Pollution Prevention PBT: Persistent Bioaccumulative Toxic RCRA: Resource Conservation and Recovery Act TLV: Threshold Limit Values TQEM: Total Quality Environmental Management TRI: Toxic Releases Inventory TSCA: Toxic Substances Control Act USEPA: U.S. Environmental Protection Agency

255 Contemporary Economic Policy (ISSN 1465-7287) Vol. 31, No. 2, April 2013, 255–278 Online Early publication February 27, 2012

doi:10.1111/j.1465-7287.2011.00312.x © 2012 Western Economic Association International

256 CONTEMPORARY ECONOMIC POLICY

Releases Inventory (TRI) under the Emergency Planning and Community Right-to-Know Act (EPCRA) by providing the public detailed infor- mation about toxic chemical releases as well as waste management activities to promote informed decision-making by industries, gov- ernment, and the public. In 1992, the National Advisory Council for Environmental Policy and Technology recommended to the U.S. EPA that P2 programs should emphasize diffusion of P2 technologies through preferential increase in the use of nonregulatory drivers, creation of incen- tives, and provision of rewards, training, and information (U.S. EPA 1992). In addition, 36 states have also legislated P2 programs, each mandating a mix of regulatory-, information-, and management-based policies. How effective these policy instruments are in promoting P2 and reducing toxic releases is the central issue addressed in this study.

The need for multiple policy instruments to address environmental goals arises because profit-maximizing firms tend to adopt fewer than optimal levels of environmental technologies and emit more than optimal pollution levels. In this study, I focus on three of the reasons for the suboptimal choices, which are the targets of the policy instruments I examine. First is exter- nalities: firms do not internalize the damages of their pollution to the rest of society (Jaffe and Stavins 2005). Second is incomplete informa- tion: the reputational benefits from environmen- tal technology adoption and pollution reduction cannot be fully realized because stakeholders that could influence firms’ choices are not fully aware of the environmental performance and activities of firms (Tietenberg 1998). Lastly, the costs of environmental technology adoption and abatement may be too high (King and Lennox 2002; Lennox and King 2004). Thus, a mix of policy instruments that compel firms to reduce their pollution levels and externalities, provide environmental-related information to the pub- lic, and lower costs of technology adoption and abatement are necessary to promote envi- ronmental technology adoption and encourage lower pollution.

Empirical comparison of the relative effec- tiveness of various policy instruments is limited. Foulon, Lanoie, and Laplante (2002) compare the impact of regulation and public disclosure of information on abatement activity while Fron- del, Horbach, and Rennings (2007) compare how standards, taxes, accounting, audits, and reports promote adoption of end-of-pipe versus

cleaner production technologies. Arimura, Hibiki, and Johnstone (2007) compare how environ- mental taxes, accounting, and standards pro- mote environmental R&D. In another study, Arimura, Hibiki, and Katayama (2008) compare the effectiveness of management-oriented ISO 14001 and publication of environmental reports on reducing natural resource use, solid waste, and wastewater effluent levels. None of these studies have analyzed the adoption of P2 activi- ties or level of toxic emissions by U.S. facilities nor have any of them analyzed how facility characteristics influence how they respond to different types of policy instruments. None of these previous studies have employed a dynamic framework that takes into account the history of innovation or history of pollution.

The objectives of this study are (1) to investi- gate whether state-level P2 legislations increase the adoption of P2 activities and reduce toxic emissions; (2) to examine the extent to which three policy instruments namely, numerical goal, reporting requirement, and mandatory P2 plan- ning, contribute to the achievement of these goals; and (3) to determine whether the potency of these policies is influenced by facility char- acteristics. To address the above objectives, I employ dynamic estimation strategies that rec- ognize the history of P2 adoption and history of pollution, while controlling for facility-specific unobservables. I use U.S. EPA TRI data on toxic releases and P2 activities of a sample of manu- facturing facilities belonging to S&P 500 parent companies from 1991 to 2001.

The results show that adoption of P2 activ- ities is higher among facilities that are located in states that have P2 legislations that empha- size toxic waste reduction. When such states additionally mandate P2 planning or require reporting of achievements and progress, facil- ities adopt more P2 activities. Furthermore, the potency of planning is greater among facilities that have adopted P2 activities in the past, while the effectiveness of reporting is greater among facilities that have demonstrated good environ- mental performance in the past, but potentially counterproductive for the extremely dirty ones. However, adoption of P2 activities is not signif- icantly influenced by the existence of numerical targets for pollution reduction. The findings also show that the toxic waste discharges are significantly lower among facilities in states where the P2 programs emphasize toxic wastes. When states prescribe pollution reduction tar- gets, toxic releases are not necessarily lower for

RAMIREZ HARRINGTON: STATE P2 POLICIES 257

all facilities, except among highly noncompli- ant ones in those states. These results imply that while traditional regulatory instruments such as emission reduction targets can contribute to the ultimate policy goal of toxic release reduction, at least among the extremely noncompliant ones, contemporary approaches such as mandatory planning and information disclosure do not.

Section II provides policy background and generates the hypotheses. Section III discusses the empirical issues and the estimation pro- cedures used. Section IV describes the sam- ple, data, and variable construction. Section V presents and discusses the results, and Section VI summarizes and concludes the paper.

II. BACKGROUND, FRAMEWORK, AND HYPOTHESES

A. State P2 Legislation and Policies

Since 1988, there are 36 states that have leg- islated P2 programs and common among them is their emphasis on source reduction activities over other means of pollution reduction such as waste disposal and treatment. Further, all of them, except for four, provide technical assis- tance by establishing an information clearing- house, on-site consultation, education, training, outreach, research assistance, guidance manu- als, waste audits, or referral services to experts through the state’s environmental agency, a pro- gram within the state department, or through the state university. However, they vary in terms of the nature of waste that they seek to reduce. Fourteen of these legislated programs empha- size the reduction of toxic and hazardous wastes, consistent with the U.S. EPA TRI, while others are subsumed under general waste minimization programs or solid waste reduction programs.

Twelve of the 36 states that currently have P2 legislation also have regulatory policies in the form of numerical goals for pollution lev- els. The targets are expressed in percentage reduction terms, relative to emission levels on a baseline year, which is usually 1987, the first year of TRI reporting. The numerical target is usually a single pollution reduction goal that needs to be achieved by a given year except for four states where it consists of multiple targets that increase in stringency through time. Despite these pollution reduction mandates, none of the legislations have indicated the specific penal- ties for noncompliance with the targets. Eigh- teen states have an information-based policy

that mandates reporting of action plans, targets, progress reports, pollution levels, or any combi- nation of these. Of the 18 states with a reporting requirement, 8 clearly require reporting of envi- ronmental performance measures. While these reports are to be submitted to the state environ- mental agency, it is not clear to what extent, if at all, the information is made publicly available. Further, while some states collect toxic informa- tion and maintain their own toxic releases inven- tory databases under the TRI Exchange Network Program, the state P2 programs do not clearly indicate whether or how the reports will fit in with the existing state-level toxic release inven- tory. Finally, 14 of the 36 states have adopted management-based regulations which require each facility to devise a P2 plan that will help it identify problems, set targets, and develop solu- tions to its pollution problems through source reduction methods. The development of facility- level plans is accompanied by reporting of progress and achievements. In the 14 states that mandate planning, the management policy is stated clearly as a feature of the state legis- lation. The policy instruments and the year of implementation are summarized in Table 1.

B. Framework and Hypotheses

In this study, my interest is to analyze how polluting entities or facilities respond to legislation and policy instruments aimed at promoting the adoption of environmental tech- nologies and/or abating pollution. Profit maxi- mizing facilities choose the level of technology adoption and the level of pollution depend- ing on the private benefits and private costs of these decisions. A facility will choose to adopt environmental technologies up to a point where its private marginal costs of environmen- tal technology adoption are equal to the pri- vate marginal benefits from technology adop- tion. In choosing the level of pollution, which is a by-product of production activities, a facil- ity will choose to release pollution up to a point where its private marginal benefits, in the form of higher output and revenues, are equal to the private marginal costs of pollu- tion. The technology adoption decision and the choice of pollution level are related to each other since the benefits from technology adop- tion are the costs from pollution that are avoided if one chooses to undertake pollution-abatement activities. These include lower compliance costs associated with violations of environmental reg- ulations, enhanced market recognition among

258 CONTEMPORARY ECONOMIC POLICY

TABLE 1 State P2 Program Legislation Dates and Policy Instruments

State

Number of Facilities in the

Sample Year of P2 Legislation

Regulatory Policy: Numerical Goal

Information-Based Policy: Reporting

Requirement

Management-Based Regulation: Mandatory

Planning

AK 0 1990 AL 22 AR 28 1993 AZ 12 1991 � (1993) � � CA 57 1989 � (1993) � � (2007) CO 7 1992 CT 19 1991 DE 6 1990 � (1992) FL 23 1991 GA 45 1990 � � (1993) HI 1 IA 23 1989 � (1994) ID 2 IL 70 1989 � (1992) IN 61 1990 KS 20 KY 37 1988 � (1997) LA 23 1992 � MA 20 1989 � (1997) � (1991) � (1994) MD 16 ME 11 1990 � (1994) � (2000) � (2000) MI 69 1994 MN 19 1990 � (1992) � (1991) MO 27 1990 � (1998) MS 21 1990 � � � MT 2 1995 NC 49 ND 2 NE 14 1992 NH 6 1996 NJ 28 1991 � (1996) � � NM 3 NV 0 NY 44 1990 � (2000) � � OH 129 1992 � OK 15 1994 OR 18 1989 � � PA 63 RI 1 SC 25 SD 2 1992 � TN 41 1991 � � TX 85 1991 � � UT 2 VA 27 1994 VT 4 1990 � (1992) � (192) WA 20 1988 � (1995) � � WI 30 1989 WV 12 1998 WY 0

Source: National Pollution Prevention Roundtable, July 15, 2008.

RAMIREZ HARRINGTON: STATE P2 POLICIES 259

consumers and the supply chain, and improved reputation with local communities (Arora and Cason 1996, 1999; Brunnermeier and Cohen 2003; Gray and Shadbegian 1998; Hall 2001; Khanna, Harrington, and Deltas 2009). It is through all these benefits and costs that the pol- icy instruments influence the rate of adoption of environmental technologies and the level of pollution chosen by facilities. I explore below the impact of legislated programs and policies on the adoption of environmental technologies, specifically P2 practices, and on toxic emission levels by explaining how each program or policy potentially affects the benefits and costs of P2 adoption and the benefits and costs of pollution. I then generate the hypotheses.

Legislated programs may promote environ- mental technology adoption and reduce pollution levels in two ways. First, information-sharing and technical assistance provided through the legislated programs enhance technical know- how which can lower the costs of searching, discovering, learning, acquiring, and adapting existing knowledge, enabling a facility to adopt environmental technologies and lower pollu- tion levels (Cohen and Levinthal 1989). In the absence of technical assistance, facilities may find such costs to be too high, causing abate- ment technologies to be underexploited (King and Lennox 2002; Lennox and King 2004). Sec- ond, the legislation may provide a perception of increased visibility of the state regulatory agency which monitors and enforces regula- tions. To enjoy reduced regulatory scrutiny and avoid future regulatory actions, facilities adopt environmental technologies and reduce their pollution levels to signal environmental stew- ardship (Brouhle, Griffiths, and Wolverton 2009; Brunnermeier and Cohen 2003; Gray and Shadbegian 1998; Khanna and Damon 1999; Khanna, Harrington, and Deltas 2009; Vidovic and Khanna 2007). Thus,

Hypothesis 1a: Legislated P2 programs for toxic waste reduction will lead to more P2 activities.

Hypothesis 1b: Legislated P2 programs for toxic waste reduction will lead to lower toxic releases.

The degree to which legislated programs can achieve their policy goals depends on the mix of policy instruments that are putting them into practice (Freeman et al. 1992). I analyze the impacts of three specific policies: a regulatory policy in the form of a numerical goal for pollution reduction, a mandatory information

disclosure policy in the form of a reporting requirement, and a management-based regula- tion in the form of mandatory P2 planning. Regulatory policies such as mandated maximum allowable pollution standards are direct controls to limit pollution and compliance with such a policy is promoted by imposing a penalty for exceeding the standard. The lower the allow- able level of pollution, ceteris paribus, the more motivated a facility will be to under- take abatement activities to reduce violations and liabilities associated with noncompliance. Additionally, the higher the expected penalty for exceeding the standard, ceteris paribus, the greater will be the incentive to abate pollution in order to meet the target. Thus, emission reduc- tion targets have the ability to enhance the bene- fits from environmental technology adoption and increase the costs of pollution that are associated with the anticipated penalties and lawsuits due to violations of environmental regulations (Hart and Ahuja 1996).

Empirical evidence on the effectiveness of a mandatory target is mixed. While Lanoie, Thomas, and Fearnley (1998) find that more stringent facility-specific emission limits lead to lower emissions, neither Jaffe and Stavins (1995) or Stafford (2003) find state-level manda- tory building codes to improve energy effi- ciency or mandatory state programs to improve compliance status of polluting entities, respec- tively. In contrast, the existing literature has consistently found penalties and enforcement actions on noncompliant facilities to be effec- tive not only in reducing violations but also in lowering emissions (see survey by Gray and Shimshack 2011). Further, state enforce- ment action has been shown to have significant and distinct effects on environmental compli- ance from federal enforcement actions (Earnhart 2004a, 2004b). Thus, state-level numerical goals may pose a state-level regulatory threat that will lead to better environmental outcomes:

Hypothesis 2a: Emission reduction targets will lead to more P2 activities.

Hypothesis 2b: Emission reduction targets will lead to lower toxic releases.

Information-based policies that provide reg- ulated entities a means to publicly disclose environmental-related information aimed at improving their reputation with consumers, investors, regulators, and the general public can increase benefits from pollution abatement

260 CONTEMPORARY ECONOMIC POLICY

activities and lower costs associated with high pollution levels. Disclosing information about pollution abatement activities and pollution lev- els allows a facility to respond to the demands of these stakeholders by providing them various signals: a signal of product quality improvement to its consumers, a signal of lower liabilities, fewer lawsuits, and better overall financial per- formance to its stockholders, a signal of com- pliance to enforcement agencies, and a signal of environmental stewardship to communities (see survey by Tietenberg 1998).

The empirical evidence on the efficacy of information disclosure programs in promoting better environmental performance and behavior depends on the nature of information disclosure system, the stakeholder channel through which behavior is affected, and the nature of environ- mental issue addressed. The information that is required to be reported to the U.S. TRI has been effective not only in reducing emissions directly (Khanna and Damon 1999) but also in shap- ing future regulatory action (inspection activity) and influencing subsequent stock market returns which then affect pollution levels (Decker 2005; Khanna, Quimio, and Bojilova 1998; Hamilton 1995). The U.S. EPA (2003) also documents how the TRI has been used by local communi- ties to lobby for more stringent regulations and by enforcement agencies for prioritizing facili- ties for enforcement and monitoring and for cre- ating inspection targeting lists. Bae, Wilcoxen, and Popp (2009) further show that the TRI is more useful in reducing releases and health risks when states undertake further data pro- cessing efforts to refine and analyze the data before disseminating it to the public. In Canada, mandatory reporting to the National Pollutant Releases Inventory (NPRI) has been shown to reduce chemical pollution levels among those who perceive a stronger threat of regulation and to a much less extent among those who perceive significant consumer pressure (Antweiler and Harrison 2003; Harrison and Antweiler 2003). A stronger role of information via the con- sumer channel is observed when there is a direct consumer information provision program as shown by Bennear and Olmstead (2008) who find that water utilities reduce their violations of Safe Drinking Water Act standards if they are mandated to report water contaminant levels to consumers. In the climate change arena where the specific form of regulation is forthcoming but still uncertain, voluntary information disclo- sure programs are growing but existing studies

demonstrate evidence of greenwashing behav- ior among participants in a number of voluntary programs such as the U.S. DOE 1605, U.S. DOE/EPA Climate Challenge, and the Cana- dian Voluntary Climate and Challenge Registry (Brouhle and Harrington 2010; Kim and Lyon 2011; Welch, Mazur, and Bretschneider 2000). These are in contrast to the electricity indus- try’s mandatory disclosure of fuel mix infor- mation which as Delmas, Montes-Sancho, and Shimshack (2009) find, lead to lower usage of fossil fuels and higher usage of clean fuels. As the reporting requirement in state P2 programs are mandated mostly for pollutants that are reg- ulated under an existing environmental statute, one could expect the following:

Hypothesis 3a: Reporting requirements will increase adoption of P2 activities.

Hypothesis 3b: Reporting requirements will reduce toxic releases.

Management-based regulations require the adoption of environmental management prac- tices for the fulfillment of environmental goals. It may lower the costs of environmental tech- nology adoption and pollution abatement for facilities because environmental management practices enable them to systematically review and identify pollution-reducing opportunities, undertake audits and benchmarking, and mon- itor environmental performance on very spe- cific segments of their operations (Coglianese and Nash 2006). Further, a management sys- tem can facilitate communication among differ- ent units and can promote better information flow (DeCanio, Dibble, and Amir-Atefi 2000). Such a “management innovation” has the poten- tial to enhance technological opportunities for facilities and promote adoption of environmental product and process innovations that can reduce pollution (Frondel, Horbach, and Rennings 2007).

A few studies show that voluntary adop- tion of environmental management systems promotes better environmental behavior and performance. Khanna, Harrington, and Deltas (2009) show that Total Quality Environmen- tal Management (TQEM) promotes P2 adop- tion, while Anton, Deltas, and Khanna (2004) demonstrate that a more comprehensive environ- mental management system reduces onsite and offsite releases. Arimura, Hibiki, and Katayama (2008) further find that adoption of management systems that comply with ISO 14001 standards

RAMIREZ HARRINGTON: STATE P2 POLICIES 261

can improve environmental performance even after controlling for the influence of environ- mental regulations. With regard to mandated management systems, Bennear (2007) finds that mandatory planning can promote P2 adoption and lower toxic releases, but she did not account for the presence of other policy instruments (as in Arimura, Hibiki, and Katayama 2008) which may complement or substitute for planning. Thus,

Hypothesis 4a: Mandatory P2 planning will increase adoption of P2 practices.

Hypothesis 4b: Mandatory P2 planning will lower toxic emissions.

In analyzing how a facility responds to each policy instrument, I recognize that the response may depend on facility characteristics such as facility-specific knowledge, past envi- ronmental performance, and exposure to past enforcement actions. Facility-specific knowl- edge acquired through learning is important because past experience promotes acquisition and exploration of new knowledge, enables a facility to exploit other external sources of information, and hastens the build-up of knowledge stock that enhances a facility’s abil- ity to assimilate new information (Cohen and Levinthal 1989; Lennox and King 2004; Mans- field 1968). Thus, highly technical facilities may be more capable of improving their environ- mental performance and modifying their behav- ior in response to state policies that require a significant amount of facility-specific technical information. Past environmental performance is important because responsive regulators use past performance and behavior of the facility as a sign of latent noncompliance that would war- rant future enforcement action (Decker 2005; Maxwell and Decker 2006; Innes and Sam 2008). Facilities that face greater prospect of regulatory scrutiny may therefore respond more aggressively to state policies to avoid future regulations. Finally, because past enforcement actions have been shown to be credible signals of threat (Innes and Sam 2008; Sam, Khanna, and Innes 2009; Shimshack and Ward 2005), exposure to past enforcement activity such as inspections and penalties may make facilities more vigilant in their response to state poli- cies as well. To deflect further monitoring and enforcement, a facility that has been subjected to more regulatory action may be more motivated

to adopt P2 technologies and lower pollution. Thus,

Hypothesis 5a: The extent of adoption of P2 in response to state policies is expected to be signifi- cantly influenced by different facility characteristics.

Hypothesis 5b: The extent of toxic pollution reduction in response to state policies also depends on various facility characteristics.

III. EMPIRICAL FRAMEWORK

The main objective of this study is to estab- lish the link between the legislation of P2 program and policy instruments and two mea- sures of environmental performance of a facil- ity: adoption of environmental technologies and pollution levels using panel data models in a framework with lagged dependent variable. As measure of technology adoption, I use the count of P2 activities adopted by a facility i at time t , and denote it as P2it . As measure of pollution, I use the level of toxic releases of facility i at time t , and denote it as Rit . The main explana- tory variables of interest for both are Leg it , the P2 legislation dummy variable, and Policyit , a vector of three policy dummy variables: numeri- cal goal, reporting requirement, and P2 planning. In each model, I include lagged dependent vari- ables, P2it−1 and Rit−1, respectively, to account for the history of P2 and history of pollution. I also control for various time-varying facil- ity characteristics, including regulatory pres- sures specific to the facility, (zit ), time- or period-specific dummies (Yt ), and time-invariant characteristics (ci).

I specify the toxic emissions equation with a lagged dependent variable as follows:

Rit = α0Rit−1 + Legitβ0,R + Policyitβ1,R(1) + zitγ1,R + Ytγ2,R + ci,R + εit,R

where ci includes a full set of facility dummies, industry dummies, and state dummies, and εit denotes idiosyncratic errors. If there are time- invariant facility-specific unobservables, pooled estimation of Equation (1) that contains a lagged dependent variable will yield inconsistent esti- mates (Anderson and Hsiao 1982). To derive consistent estimates, I employ the Anderson and Hsiao two-stage approach as implemented by Jaffe and Stavins (1995). The first stage involves estimating the first differenced ver- sion of Equation (1). Because the differenced lagged dependent variable is correlated with

262 CONTEMPORARY ECONOMIC POLICY

the first differenced error terms, I instrumented for the differenced lagged dependent variable using its twice lagged level (Rit−2) in the dif- ferenced equation. The second stage derives the coefficients of the time-invariant regressors by deriving a “residual” from the differenced first stage regression and regressing (using OLS) the residuals against the time-invariant factors. The “residual” is computed for each facility by mul- tiplying each coefficient estimate from the first step by the facility mean for each variable and subtracting their sum from the facility mean of the dependent variable.

In explaining the count of P2 using its lagged count and other factors, I assume that the count of P2 has a Poisson distribution with the mean specified as

E(P2it |Xit , ci) = ci,P 2 exp[P2i,t−1ρ + Legitβ0,P 2

+ Policyit , β1,P 2 + zitγ1,P 2 + Ytγ2,P 2]. However, the presence of a lagged dependent variable makes within-estimation yield inconsis- tent coefficient estimates due to the correlation between the differenced lagged regressor and the error term (Cameron and Trivedi 1998). Thus, rather than employing a fixed effects model, I adopt Wooldridge’s (2005) model that assumes a specific distribution for the facility-specific effect but which allows the use of a random effects Poisson model with gamma distribution. I follow Wooldridge’s (2005) recommendation for ci to have a specific conditional distribution as follows:

ci = ai exp[α0 + f (P2i0)α01 + ziα2], where f (P2i0) is a function of the initial value of P2 on the first year of reporting, P2i,0, the vector, z i consists of time-invariant facility- specific characteristics, and ai is assumed to be independent of (P2i0, z i), and follows a gamma distribution. Wooldridge (2005) shows that after conditioning the ai out of the density (Wooldridge 2005, 50–51, equations 32–34), if only one lag is included and if the explanatory variables are exogenous, one can estimate the following model using a random effects Poisson model with gamma heterogeneity:

E(P2it |P2it−1, P2i0, Legit , zit , ci)(2) = ai exp

[ g(P2i,t−1)ρ + f (P2i0)α01

+ Legitβ0,P 2 + 3∑

k=1 Politkβ1,P 2

+ zitγ1,P 2 + Ytγ2,P 2 + α0 + ziα2 ]

where the full set of regressors consists of all time- and facility-specific variables and those that determine the facility-specific effects, ci , shown above. To satisfy these assumptions, I assume gamma heterogeneity in my random effects model,1 I use one lag of the depen- dent variable, and I include the initial P2 count, industry dummies, state dummies, and other time-invariant characteristics to define ci .2

Regarding the exogeneity assumption, par- ticularly of the legislation and policy vari- ables, which can be of concern for estimating both Equations (1) and (2) using the approaches described above, there is reason to suspect that the nature of facilities which determines the count of their P2 activities and their level of emissions could also influence whether the state they are located in legislates a P2 pro- gram and implements a particular policy. To check for this possibility, I employ a probit model to explain the legislation of the P2 pro- gram and the adoption of each of the policies using various state-level measures of environ- mental performance as well as political pres- sure and community pressure variables. These state-level explanatory variables are correlated with facility-specific measures of environmen- tal behavior and performance which I use as dependent variables in Equations (1) and (2). If the state-level explanatory variables are signif- icant in the probit models, there is reason to suspect that state legislation/policies are deter- mined simultaneously with the decision to adopt P2 and the level of toxic releases. The results

1. I compared the log likelihood and BIC of the Poisson models with normal versus gamma heterogeneity and find that the gamma heterogeneity models have higher log- likelihoods and lower BIC than the model that assumes normally distributed heterogeneity.

2. Further, as this is basically now a random effects model, I also checked if the other regressors that include measures of actual past regulatory action or proxy for forth- coming regulations are correlated with the initial count of P2. I find that the correlation between initial count and various measures of regulatory pressures (lagged releases, lagged inspections, and lagged penalties) range only from 0.10 to 0.29. Further, while these measures may be asso- ciated with total level of pollution, pollution can be abated using either P2 techniques or end-of-pipe abatement meth- ods, or both. Hence, these variables which are all measured at time t − 1 can be considered uncorrelated with initial P2 count in 1991.

RAMIREZ HARRINGTON: STATE P2 POLICIES 263

(not shown) indicate that there is no evidence that state P2 legislation and policies are adopted in response to factors that may also affect P2 adoption and toxic releases.3

As a second check, I use prelegislation or prepolicy variables in place of the actual legis- lation/policy variables in Equations (1) and (2) as in Bennear (2007). These prelegislation/pre- policy variables take on a value of 1 for states which eventually pass the legislation/adopt the policy on the year immediately preceding the year of legislation or policy implementation, and 0 otherwise. Statistical significance of these vari- ables would indicate that there are differences in the P2 counts and toxic releases of facili- ties between states which eventually adopt the legislation/policies on the year before the leg- islation/policies were adopted and those facil- ities in states which never did. This implies that there may be other time-varying factors that can explain differences in P2 counts and emission levels and that I may be wrongfully attributing the response to the legislation/policy. In Table A1, Models 1-A and 1-B are random effects P2 models, while Models 2-A and 2-B are Anderson-Hsiao models, all of which use the prelegislation/prepolicy variables. All models show that none of the prelegislation/prepolicy variables are significant, implying that the leg- islation and policy can be treated as exogenous in Equations (1) and (2). Thus the toxic emis- sions equations are estimated using Anderson- Hsiao models, while count of P2 practices are explained using random effects Poisson mod- els with initial P2 as one of the time-invariant variables that define facility-specific effects.

IV. SAMPLE DESCRIPTION, VARIABLE CONSTRUCTION, AND DATA SOURCES

A. Sample Description

The sample consists of facilities whose parent companies are among the S&P 500 firms report- ing to the TRI every year from 1991 to 2001.4

The TRI mandates the reporting of toxic releases and P2 activities by facilities which belong to

3. The full description and results of the probit model and the variables used are not included in the paper for the interest of brevity. They are available in the online version of the article.

4. 1991 is the first year of P2 reporting, and many states passed their P2 legislations between 1992 and 1998. Further, even though a number of states passed their P2 legislations in 1991 or earlier, many of the actual policies did not take into effect until the 1992 to 2001 time frame.

specific industries, have at least ten full-time employees, and which manufacture, process, or use any EPCRA Section 313 chemical in quan- tities greater than the established threshold.5 The original 11-year sample is an unbalanced panel with 34,121 observations, consisting of 5,094 unique facilities. I effectively start my analysis in 1992 and use 1991 P2 data as a facility- specific effect, a key explanatory variable in my dynamic random effects Poisson model. This drops the 1991 reporters, or 5,101 observations. Because many facilities do not report every year and reporting years are not always adjacent, an additional 3,146 observations do not have 1-year lagged data. For those which have lagged data but which do not report every year, their attrition in any year cannot be clearly attributed to hav- ing zero or below threshold toxic releases, nor can below-threshold levels be attributed to the use of P2. Thus, I only include the 1,261 S&P facilities who report every year from 1991 to 2001, for a total of 12,610 observations.6 Rela- tive to the complete unbalanced set, the facilities in this smaller balanced panel adopt approxi- mately 39% more P2 activities and emit 70% more toxic releases. They also have been sub- jected to 19% more penalties and 39% more inspections. Thus, I emphasize that the conclu- sions derived from this analysis apply only to the facilities with similar characteristics: belong- ing to large parent companies (S&P 500 firms) which report every year from 1991 to 2001 and whose toxic emissions are always at or above the threshold.

B. Dependent Variables

I use two variables to measure the response to the state legislation and policies. The first is the count of all new P2 activities or prac- tices adopted by a facility each year for all the

5. The USEPA mandates reporting to the TRI for facilities that have at least 25,000 pounds of chemicals that are manufactured or processed or 10,000 pounds otherwise used, except for certain persistent bioaccumulative toxic (PBT) chemicals. The threshold for the PBT chemicals is 100 pounds or less depending on the chemical beginning with the 2000 reporting year.

6. Even if we can attribute the reduction in releases to the adoption of P2 activities, the data on P2 adoption and releases that result from previous adoption will not be observed for the year that a facility has below threshold releases because the facility will not be required to report to the TRI. But it is very likely that facilities that have below threshold (but still positive) releases may still be adopting P2 activities. Thus, any TRI-related data will not be available for these facilities in those years.

264 CONTEMPORARY ECONOMIC POLICY

chemicals it uses, processes, and manufactures.7

The U.S. EPA classifies 43 different types of practices a facility may adopt into eight cate- gories: good operating practices, inventory con- trol, process and equipment modifications, spill and leak prevention, cleaning and decreasing, surface preparation and finishing, raw material modifications, and product modifications (U.S. EPA 2007). The first three consist of activi- ties that involve adjustments in operating and production systems that are part of day-to-day activities which are highly customized to a facility’s operations and include maintenance scheduling, recordkeeping, changes in produc- tion schedule, adoption of recirculation within a process, modification of equipment, layout or piping, and use of a different product catalyst. The next three deal with preventing visible air and water residues in work areas that result from production or cleaning activities which can be accomplished through installation of alarms and automatic shut-off valves, installation of vapor recovery systems, implementation of inspection or monitoring program for potential spill or leak sources. Such activities impact day-to-day expo- sure to water and air residues that may cre- ate hazards for employees. The last two prac- tices include the more technologically sophisti- cated P2 practices: substitution of raw materials and changes in product specification which are becoming the basis for product differentiation through product labeling and marketing cam- paigns. Each facility is allowed to report up to four practices that it adopts for each of its chemical. The dependent variable Total P 2 is therefore the sum of all the practices from all these categories adopted for all chemicals every year by each facility from 1992 to 2001. To ensure that the change in P2 adoption over time is not due to differences in the chemicals that were required to be reported, chemicals which have been added or deleted (due to changes in the reporting requirements by the EPA) over the sample period were dropped. The average Total P 2 count is 1.64 practices, with a minimum of

7. I verified if facilities do indeed report new P2 activities by examining the annually reported P2 counts for each chemical by each facility belonging to S&P 500 firms which report to TRI and compared it with their reports for the preceding year. I calculated the number of facilities for which the reported P2 counts were nondecreasing for all chemicals. This was the case for only 5.68% of all facilities and represents only 0.67% of the chemical-facility pairs (these facilities have a lower than average chemical count). Therefore, even if there was any misinterpretation of the survey question, it impacted a tiny fraction of the data.

zero and a maximum of 84. The distribution is also highly skewed, with 68% of observations having adopted zero P2 count, and 19% having between 1 and 10 counts of P2. The states with the highest mean adoption of Total P 2 among its facilities are MT (7.2) and MN (6.2), while the lowest mean adoption is observed in UT (0.05) and ND (0.45).

The other dependent variable is a measure of toxic pollution level. The TRI requires facilities to report the quantities of onsite toxic releases to air, water, land, and underground injection, as well as offsite disposals, transfers, and treat- ment, on a chemical-specific basis. The level of a facility’s toxic releases, denoted as Toxic Releases, is an aggregation of the emissions across all media and all chemicals in each time period, released onsite and offsite. To ensure that the change in toxic releases over time is not due to differences in the chemicals that were required to be reported, I dropped from the aggregation the releases for those chemicals which have been added or deleted over the sam- ple period. The level of toxic releases ranges from 0 to 3,402 million pounds, with a median of 141,245 pounds. As the Toxic Releases vari- able is highly skewed, I use the natural log of this variable (plus one).

C. Explanatory Variables

The main explanatory variables of interest are the P2 legislation decisions and adoption of pol- icy instruments at the state level. The dates of legislation and the specific details on the fea- tures regarding policy instruments are obtained and manually coded from the NPPR.8 The P2 Legislation dummy variable takes a value of 1 if the facility is located in a state with a P2 leg- islation beginning at time t , and 0 before that year. A facility located in a state that has never passed a P2 legislation or has adopted it after 2001 will have a value of 0 for this variable for the entire sample time frame, while a facil- ity located in a state that has passed it before 1991 will have a value of 1 for this variable for the sample time frame. A state is considered to have a P2 program if it legislated an act that is named as such or as long as the state legislation clearly gives highest priority of waste reduc- tion to P2. Because Total P2 and Toxic Releases are collected only for specific toxic substances

8. National Pollution Prevention Roundtable, http:// www.p2.org/inforesources/nppr_leg.html, downloaded on July 15, 2008.

RAMIREZ HARRINGTON: STATE P2 POLICIES 265

identified under EPCRA Section 313, I further differentiate states by including a Toxic dummy variable, to indicate whether the state P2 pro- gram is focused on toxic wastes.

Each of the policy instruments is constructed as a dummy variable, where 1 indicates presence of a policy beginning at time t , and 0 before that year. The specific years of implementation are obtained from the NPPR and state environmen- tal bureaus. If the legislations do not specify a year of implementation for a particular policy, it is assumed to be in effect on the same year as the legislation. The Numerical Goal variable pertains to whether target(s) exist for emission reductions. The year of adoption of a numerical goal is the year that the first target is expected to be met. I alternatively investigate models that use the actual percentage reduction in pollu- tion levels, which I denote as Target. Reporting Requirement refers to the mandatory submission of details on the action plans, targets, progress reports, or pollution levels. In this study, a state with any type of reporting, regardless of the required content, is considered to have a Report- ing Requirement. Finally, Mandatory Planning variable refers to the presence of mandatory P2 planning policy. States are considered to have a management policy if it is clearly stated as a mandatory feature of the state legislation in the NPPR. The number of facilities in the sample that are subjected to each policy in each state is indicated in Table 1.

The role of history in technology adoption and history of pollution are captured by the inclusion of the lagged dependent variables as explanatory variables. Lagged P2 is constructed as the count of new practices adopted in the previous year and is included to account for past experience in technology adoption. Lagged Toxic Releases is the volume of all toxic releases of a facility in the previous year and is included to account for time-persistent facility character- istics such as nature of operations and technical constraints that influence pollution levels. I use the natural logs of these variables (plus one).

Facility-specific regulatory pressures and market pressure variables are also included to control for influences of regulatory action (some at the federal level) and other external stake- holders. The first set includes measures of regu- latory threat. Because past enforcement actions are found to be credible sources of threat (Innes and Sam 2008; Sam, Khanna, and Innes 2009; Shimshack and Ward 2005), polluters who face a prospect of greater regulatory scrutiny because

they have been subjected to high enforcement activity in the past may adopt more P2 and reduce toxic releases more to signal environ- mental stewardship. I use the number of Lagged Inspections and the number of Lagged Penalties. The Lagged Inspections variable is the number of times a facility was inspected by state and federal environmental agencies to monitor com- pliance with mandatory regulations in the past year. Lagged Penalties refer to the number of times a facility has been cited for and has been penalized for noncompliance with federal envi- ronmental statutes, such as the Clean Air Act (CAA), the Clean Water Act (CWA), Toxic Sub- stances Control Act (TSCA), and the Resource Conservation and Recovery Act (RCRA) in the past year. Both are from the EPA Integrated Data for Enforcement Analysis (IDEA). I use both to explain Total P 2 and Toxic Releases. Both are highly skewed so I use the natural log of these variables (plus one).

To further explain Total P 2, I include Lagged Toxic Releases as a proxy for the threat of anticipated compliance costs with existing or forthcoming regulations. The reporting of P2 practices under the TRI is mandated specifically for toxic chemicals and many of these chemi- cals are regulated under different environmental statutes.9 Further, the total volume of releases captures the extent of pollution of the facility, making it more visible to the regulator (Arora and Cason 1996; Innes and Sam 2008; Khanna and Damon 1999). Thus, the lagged level of releases may proxy for the degree of poten- tial liability associated with violations of these regulations, which can increase P2 adoption.

I use two additional measure of regula- tory threat to explain Toxic Releases: Lagged Toxicity-Weighted Releases and Nonattainment. Because the risk impacts of the chemicals reported to the TRI vary, Lagged Toxicity Weighted Releases may be able to capture extent of liabilities related to health risks as in Bae, Wilcoxen, and Popp (2009). This variable is a weighted sum of toxic releases which is obtained by multiplying the volume of releases for each chemical with its corresponding toxicity weight from the Threshold Limit Values (TLV) Index and summing up over all chemicals for each facility in each year. TLV is determined by the American Conference of Governmental

9. A matrix that relates the TRI chemicals to different environmental statutes is available from the USEPA web- site: http://www.epa.gov/tri/trichemicals/reg_requirements/ 94regmat.pdf.

266 CONTEMPORARY ECONOMIC POLICY

Industrial Hygienists and is obtained from the World Bank.10 The nonattainment status of all counties in the United States according to the 1977 CAA Amendments is a designation on every county as being in attainment or nonat- tainment with national air quality standards for each of six criteria air pollutants: CO, SO2, TSP, O3, NO, and PM. Facilities in areas which are in nonattainment may face a more stringent regu- latory environment (List, Mchone, and Millimet 2004). For each county, I derived the sum of the number of pollutants for which it is in nonat- tainment of environmental standards to create the Nonattainment variable. The data is obtained from the U.S. EPA Greenbook.11 I use the nat- ural log of these variables (plus one).

The second set of explanatory variables include measures of influences from the mar- ket and local citizens because polluting entities may also be subjected to pressures from its con- sumers, supply chain, and its local community to improve environmental behavior and perfor- mance. On one hand, polluting entities can bene- fit from higher consumer sales if they can project a signal that the production process reflects environmentally responsible manufacturing, say through adoption of environmental technologies or lower emission levels. Several studies have shown that those that are in closer contact with consumers are more likely to participate in vol- untary environmental programs and adopt more comprehensive environmental management sys- tems (Anton, Deltas, and Khanna 2004; Arora and Cason 1996; Khanna and Damon 1999; Vidovic and Khanna 2007). On the other hand, intermediate good producers are exposed to both suppliers and clients in their supply chain and may be required by their suppliers and clients to comply with certain environmental standards (Hall 2001). They may even be expected by their clients to adopt specific environmental manage- ment systems to conform to ISO 14001 stan- dards (Arimura, Darnall, and Katayama 2010). Thus, there is no a priori expectation on whether Final Good dummy will have a positive or a negative effect on Total P2 or Toxic Releases. The Final Good dummy variable is based from

10. The TLV is expressed in milligrams per cubic meter and is based on “time-weighted average concentrations in air that cannot be exceeded without adverse effects for workers in a normal 8-hr work day and a 40-hr work week”. The index is available from http://www.worldbank.org/nipr/data/ toxint/.

11. Can be found at http://www.epa.gov/oar/oaqps/ greenbk/anay.html.

the facility’s four-digit SIC code as constructed in Harrington, Khanna, and Deltas (2008). It is equal to 1 if the facility produces a final good, 0 otherwise.

Because local communities can exert pres- sures on polluting entities directly through cit- izen suits or indirectly by lobbying for more stringent regulations (Earnhart 2004b; USEPA 2003) it is important to control for local pres- sures that capture willingness to pay for environ- mental quality. I use median household income, denoted as Income obtained from the Bureau of Census. Many studies have shown that sev- eral measures for economic attributes are corre- lated with various environmental outcomes such as pollution, exposure to risk, plant location decisions, and participation in voluntary envi- ronmental programs (Arora and Cason 1999; Becker 2004; Brooks and Sethi 1997; Earnhart 2004b; Hamilton 1995; Wolverton 2009). Thus, facilities in areas with higher willingness to pay for environmental quality, measured by Income, are expected to have higher Total P 2 and have lower Toxic Releases.

In addition to the regulatory and market variables, several experience-related variables are included to help explain Total P 2. Lagged Cumulative P2 is defined as the count of all new P2 activities adopted since 1991 up to the imme- diately preceding year and captures the stock of knowledge from past P2 adoption. Initial P2 is the count of P2 adopted in 1991, the first year that P2 reporting is mandated in the TRI, and it captures both the adoption on the first year of reporting and the facility-specific effect. Both variables are obtained from the TRI. The adop- tion of P2 may also be influenced by activities of other facilities in the same corporate fam- ily because experience of these related facilities may generate positive spillovers that can also enhance a facility’s capacity to develop tech- nologies that are in line with the parent company culture, needs, and priorities (Jaffe 1986). Thus, I include the average adoption of P2 by facil- ities belonging to the same parent company, which I denote as Spillover P2. It is obtained from the TRI. I use the natural log of the two experience variables and the spillover variable (plus one).

To control for the scope over which Total P 2 and Toxic Releases are reported while avoiding the potential problem of endogeneity of contemporaneous or 1-year lagged number of chemicals, the Number of Chemicals of a facility, averaged over all years is included as

RAMIREZ HARRINGTON: STATE P2 POLICIES 267

an explanatory variable.12 This data is obtained from the TRI. The value of this variable ranges from 0.67 to 79.5, with a median of 3.47. Finally, because environmental quality improve- ments may be costly and technically challeng- ing for facilities, I use R&D Intensity, the ratio of R&D expenditures to net sales as measure of parent company technical capac- ity as in Anton, Deltas, and Khanna (2004) and Khanna, Harrington, and Deltas (2009), because facilities that belong to more innova- tive companies are better able to use, assimi- late, and exploit existing information to develop new technologies (Cohen and Levinthal 1989). This data is from Research Insight. Due to the skewness of the distribution of the data for both variables, I use the natural logs (plus one) of Number of Chemicals and RD Inten- sity. State, year, and industry dummies are also constructed. The descriptive statistics are in Table 2.

V. RESULTS AND DISCUSSION

The discussion of results is organized as fol- lows. I first summarize the robust results with regard to the other factors that influence Total P2 and Toxic Releases. I then proceed with the summary of results with respect to the hypothe- ses regarding the legislation and policy vari- ables. The detailed discussion of the findings on how Total P2 (Models I to V in Table 4) and Toxic Releases (Models VI to IX in Table 5) respond to each policy variable follows in sep- arate subsections.

Overall, the empirical results show very robust results with regard to the effect of differ- ent regulatory, market and community pressure variables on Total P2 and Toxic Releases. For the case of Total P2, Models I to V show that Total P2 is higher among facilities that have higher Lagged Inspections, higher Lagged Toxic Releases, those which are intermediate good

12. The number of chemicals reported consists of chem- icals used, manufactured, processed, and disposed of, thus, the opportunities for reporting P2 activities is greater for facilities which emit more chemicals. This indicates that contemporaneous number of chemicals may be endogenous with the P2 practices adopted. Lagged chemicals may also pose some endogeneity issue because having fewer chem- icals in the previous year implies fewer opportunities to report P2 in the following year. The number of chemicals in 1991 could be used, but it is highly correlated (0.56) with 1991 P2 count, which is a key explanatory variable in this dynamic Poisson model. Hence, I use average number of chemicals between 1991 and 2001.

producers, those located in areas with higher Income, those which release a higher Number of Chemicals, and those which have greater expe- rience in P2 activities in the past. I also find the experience-related variables to have different impacts. One year Lagged P2 and Initial P2 are always positive and significant, indicating that past experience provides facilities the know-how in identifying and implementing new source reduction practices, possibly because of cost- reducing effects or momentum effects (Lagged P2 ). However, Lagged Cumulative P2 is neg- ative and significant which suggests diminish- ing opportunities for further adoption of new P2 activities. Thus, despite short-term learning effects (captured by Lagged P2 ), there may be some degree of exhaustion of new opportunities to further reduce releases of chemicals at source. Nonetheless, the combined impact of both vari- ables is positive and statistically significant (at 5% level), lending support to the inclusion of past experience as key explanatory variables. In the Toxic Releases models, Models VI to IX show Toxic Releases to be lower among facil- ities that have higher Lagged Penalties, have higher Lagged Toxicity-Weighted Releases, have lower Number of Chemicals, and which are Final good producers. I also find strong path dependence in Toxic Releases suggesting the importance of persistent facility characteristics that may constrain large reductions in pollution levels every year. One interesting result to con- trast between Total P2 and Toxic Releases is the negative and significant sign of the coeffi- cient of the Final good dummy in both the Total P2 and Toxic Releases equations, which sug- gests that it is the pressure from the supply chain that is motivating technology adoption, but it is the pressure from consumers that is motivating pollution reduction. These results can be inter- preted as a support for the view that consumers are more concerned with observable measures such as pollution levels but not necessarily with the nature of technologies adopted to achieve these ends.

I now summarize the findings with regard to the hypotheses. I find some evidence to support Hypothesis 1a and 1b. While Total P 2 is only slightly higher in states with P2 legislation that emphasize toxic waste reduction, Toxic Releases are significantly lower in states with a state P2 legislation that are geared toward reducing toxic wastes. I do not find evidence to support Hypotheses 2a and 2b: Total P2 is not significantly higher and Toxic Releases

268 CONTEMPORARY ECONOMIC POLICY

TABLE 2 Summary of Variables and Descriptive Statistics: Mean and Standard Deviation (in Parentheses).

Variables Without State P2 Program

With State P2 Program All Observations

Total P2 Equation

Toxic Releases Equation

Dependent Variables Total P2 1.705 1.621 1.637 X

(5.446) (4.425) (4.638) Toxic Releases 1,700,153 2,572,414 2,405,571 X

(6,957,141) (36,200,000) (32,700,000) Explanatory Variables

P2 Legislation 0.000 1.000 0.809 X X (0.000) 0.000 (0.393)

Numerical Goal 0.000 0.191 0.155 X X (0.000) (0.393) (0.362)

Reporting Requirement 0.000 0.572 0.462 X X (0.000) (0.495) (0.499)

Mandatory Planning 0.000 0.344 0.278 X X (0.000) (0.475) (0.448)

Toxic dummy 0.000 0.37164 0.3005 X X (0.000) (0.4832) (0.4585)

Target 0.5182 0.9359 0.8560 X X (5.0650) (6.2803) (6.0687)

Initial (1991) P2 2.347 2.637 2.581 X (5.783) (5.999) (5.959)

Lagged P2 1.863 1.766 1.785 X (5.786) (4.665) (4.899)

Cumulative P2 10.276 11.504 11.269 X (29.636) (27.603) (28.006)

Spillover P2 0.882 0.929 0.920 X (1.037) (1.127) (1.111)

Lagged Inspections 5.004 2.980 3.367 X X (11.017) (8.449) (9.031)

Lagged Penalties 0.070 0.087 0.083 X X (0.359) (0.507) (0.482)

Lagged Toxic Releases 1.673044 2.616549 2.436078 X X (6.717) (36.300) (32.800)

Toxicity-Weighted Releases 5,750,387 5,069,449 5,199,697 X (64,700,000) (51,300,000) (54,100,000)

Nonattainment 0.4556 0.6797 0.6368 X (0.7666) (0.9934) (0.9582)

Final Good dummy 0.274 0.358 0.342 X X (0.446) (0.479) (0.474)

Income 34134.62 37475.91 36836.80 X X (8408.175) (9177.562) (9130.221)

R&D Intensity 0.026 0.030 0.030 X X (0.020) (0.028) (0.027)

Number of Chemicals 4.970 5.541 5.432 X X (4.981) (6.477) (6.222)

Number of observations 2,412 10,198 12,610

are not significantly lower among all facilities in states which have Numerical Goals. I find evidence to support 3a and 4a but not 3b and 4b: Total P2 is higher among facilities in states with Reporting Requirement and Mandatory Planning, even among those in states that do not emphasize toxic waste reduction, but Toxic

Releases are not significantly affected by either policy. Finally, I find evidence for Hypothesis 5a and 5b: some facility characteristics do influence the adoption of Total P2 and reduction of Total Releases in response to policy instruments. The hypotheses and findings are summarized in Table 3.

RAMIREZ HARRINGTON: STATE P2 POLICIES 269

TABLE 3 Summary of Hypotheses and Evidence

Hypotheses Evidence to Support

Hypotheses

Hypothesis 1a: Legislated P2 programs for toxic waste reduction will lead to more P2 activities. � Hypothesis 1b: Legislated P2 programs for toxic waste reduction will lead to lower toxic releases. � Hypothesis 2a: Emission reduction targets will lead to more P2 activities. × Hypothesis 2b: Emission reduction targets will lead to lower toxic releases. × Hypothesis 3a: Reporting requirements will increase adoption of P2 activities. � Hypothesis 3b: Reporting requirements will reduce toxic releases. × Hypothesis 4a: Mandatory P2 planning will increase adoption of P2 practices. � Hypothesis 4b: Mandatory P2 planning will lower toxic emissions. × Hypothesis 5a: The extent of adoption of P2 in response to state policies is expected to be

significantly influenced by different facility characteristics �

Hypothesis 5b: The extent of toxic pollution reduction in response to state policies also depends on various facility characteristics.

�

A. Effect of P2 Policies on the Adoption of P2 Practices

The results of the models that explain how policy instruments affect Total P2 are in Table 4. All models (except for Model V) are estimated using random effects Poisson and include the lagged dependent variable, state dummies, year dummies, initial (1991) P2 counts, industry dummies, and time-varying covariates to con- trol for past experience in P2, fixed differences between states, national trends that affect all facilities, facility-specific characteristics, fixed differences between industries, and time-varying characteristics affecting P2 at each time period, respectively. In Table 4, Model I is the base model and includes the P2 Legislation dummy and Toxic dummy. Model II is Model I with all policy dummy variables included. Model III uses Target instead of Numerical Goal dummy. Model IV-A and IV-B include interaction terms. Model V has the same specification as Model II but is estimated using Poisson with all facility dummies in place of the Initial P2 and is pre- sented for comparison purposes. It yields similar coefficients.

All models in Table 4 show some limited evi- dence for Hypothesis 1a, no support for 2a, but broad support for Hypothesis 3a and 4a. The nonsignificant coefficient of P2 Legislation in all models indicates that any P2 Legislation is not sufficient to yield higher adoption rates for Total P2. However, the Toxic dummy variable is significant at 10% which suggests that when the effect of P2 Legislation is assessed for facilities in states that emphasize toxic pollution reduc- tion, the effect on Total P2 variable is positive and significant indicating that P2 Legislation is

effective in promoting P2 activities only in these states. This is not surprising because the data on P2 activities used in this study are obtained from the TRI and are the ones adopted specif- ically for toxic chemicals. Hence, P2 activities adopted by facilities in states whose legislation and technical assistance emphasize reduction of other types of wastes would not be reflected in this dataset, which may explain the lack of sig- nificance of the P2 Legislation variable by itself, but a significant effect when the focus of the legislation is accounted for.

The coefficients of Reporting Requirement and Mandatory Planning are positive and sig- nificant, consistent with Bennear (2007) and Khanna, Harrington, and Deltas (2009), while that of Numerical Goal is not significant. These findings suggest that the P2 adoption deci- sion of facilities positively responds to policies which may improve their reputation with exter- nal stakeholders (Reporting Requirement) and which allow them to reduce their costs of tech- nology adoption (through Mandatory Planning) but not to those that can pose threat of regula- tory action. The signs of the policy coefficients are similar whether included singly or jointly, though the magnitudes and levels of significance are higher if the policy instrument variables are included in the model one at a time (results not shown). This is especially true for Report- ing Requirement which has a coefficient of 0.62 and significant at 1% when included as the only policy instrument in the model. Nonetheless, Reporting Requirement and Mandatory Plan- ning are individually significant even when other policy variables are included in Model II, indicating that they have distinct effects and

270 CONTEMPORARY ECONOMIC POLICY

TABLE 4 Determinants of Total P2, Poisson Random Effects (RE) and Fixed Effects (FE)

Poisson RE Poisson FEa

Variables I II III IV-Ab IV-Bb V

P2 Legislation 0.03249 0.03715 0.03933 0.03708 0.07579 0.0347∗

(0.0515) (0.0518) (0.0516) (0.0518) (0.0523) (0.0521) Toxic dummy 0.23036∗ 0.23023∗ 0.22833∗ 0.19268 0.23158∗ 0.2055

(0.1351) (0.1351) (0.1351) (0.1361) (0.1350) (0.1353) Numerical Goal −0.02541 −0.02731 −0.04252 −0.0169

(0.0397) (0.0397) (0.0397) (0.0398) Reporting Requirement 0.40922∗ 0.4010∗ 0.4599∗∗ 0.4325∗∗ 0.5127∗∗

(0.2192) (0.2194) (0.2204) (0.2183) (0.2186) Mandatory Planning 0.2122** 0.2113** 0.2111** −0.0625 0.1921∗∗

(0.0937) (0.0935) (0.0937) (0.1065) (0.0938) Target −0.00161

(0.0015) Reporting Requirement ×

Lagged Toxic Releases −0.1110∗∗ (0.0450)

Mandatory Planning × Cumulative P2

0.0972∗∗∗

(0.0182) Lagged P2 0.8410∗∗∗ 0.8408∗∗∗ 0.8411∗∗∗ 0.8400∗∗∗ 0.8427∗∗∗ 0.9079∗∗∗

(0.0170) (0.0170) (0.0170) (0.0170) (0.0170) (0.0176) Cumulative P2 −0.5203∗∗∗ −0.5219∗∗∗ −0.5227∗∗∗ −0.5227∗∗∗ −0.5567∗∗∗ −0.731∗∗∗

(0.0259) (0.0259) (0.0259) (0.0259) (0.0266) (0.0257) 1991 P2 0.7083∗∗∗ 0.7101∗∗∗ 0.7106∗∗∗ 0.7097∗∗∗ 0.7171∗∗∗

(0.0532) (0.0532) (0.0532) (0.0532) (0.0533) Spillover P2 0.2772∗∗∗ 0.2794∗∗∗ 0.2812∗∗∗ 0.2823∗∗∗ 0.2824∗∗∗ 0.276∗∗∗

(0.0338) (0.0339) (0.0338) (0.0339) (0.0339) (0.0349) Lagged Inspections 0.0422∗∗∗ 0.0416∗∗∗ 0.0411∗∗∗ 0.0411∗∗∗ 0.0444∗∗∗ 0.0426∗∗∗

(0.0110) (0.0110) (0.0110) (0.0110) (0.0110) (0.0111) Lagged Penalties 0.0160 0.0180 0.0170 0.0208 0.0166 0.0167

(0.0297) (0.0297) (0.0297) (0.0297) (0.0296) (0.0298) Lagged Toxic Releases 0.0919∗∗∗ 0.0919∗∗∗ 0.0918∗∗∗ 0.1657∗∗∗ 0.0927∗∗∗ 0.0887∗∗∗

(0.0219) (0.0219) (0.0219) (0.0371) (0.0220) (0.0229) Final Good dummy −0.2941∗∗∗ −0.2946∗∗∗ −0.2946∗∗∗ −0.2976∗∗∗ −0.2949∗∗∗ −2.1567

(0.1049) (0.1050) (0.1050) (0.1051) (0.1054) (2.0794) Income 0.000014∗∗∗ 0.000015∗∗∗ 0.000015∗∗∗ 0.000015∗∗∗ 0.000013∗∗∗ 0.000018∗∗∗

(0.0000047) (0.0000047) (0.0000047) (0.0000047) (0.0000047) (0.00000637) RD Intensity 0.5469 0.3778 0.4376 0.4293 0.5270 0.6041

(0.8093) (0.8133) (0.8150) (0.8137) (0.8136) (0.8943) Number of Chemicals 0.5997∗∗∗ 0.6013∗∗∗ 0.6013∗∗∗ 0.5943∗∗∗ 0.5953∗∗∗ 3.9990∗∗∗

(0.0825) (0.0825) (0.0825) (0.0824) (0.0829) (1.2629) Constant −2.9408∗∗∗ −2.9434∗∗∗ −2.9448∗∗∗ −2.9523∗∗∗ −2.8627∗∗∗ −6.375∗∗∗

(0.6503) (0.6507) (0.6509) (0.6512) (0.6531) (1.8794) Facility dummies — — — — — 5351.45∗∗∗

SIC dummies 43.80∗∗∗ 43.89∗∗∗ 43.84∗∗∗ 43.73∗∗∗ 43.06∗∗∗ 143.98∗∗∗

State dummies 68.64∗∗ 75.84∗∗∗ 75.54∗∗∗ 75.66∗∗∗ 76.25∗∗∗ 770.70∗∗∗

Year dummies 68.88∗∗∗ 65.81∗∗∗ 65.70∗∗∗ 66.01∗∗∗ 69.66∗∗∗ 163.39∗∗∗

LR (a) 3913.46∗∗∗ 3915.72∗∗∗ 3919.11∗∗∗ 3915.11∗∗∗ 3944.32∗∗∗ — No. of observations 12,610 12,610 12,610 12,610 12,610 8,810 No. of groups 1,261 1,261 1,261 1,261 1,261 881

aThere are 8810 for Models III because those facilities which have zero P2 counts in all the years get dropped from a Poisson model with facility dummies.

bModels with other interaction terms between Reporting Requirement and Final Good dummy, Numerical Goal and Lagged Inspections or Lagged Penalties and Reporting Requirement and Lagged Inspections or Lagged Penalties are not shown for the interest of brevity but in all these models, the interaction terms are not statistically significant while the rest of the coefficients variables are similar to those in Model II. A complete set of these results are available upon request.

Standard errors in parentheses: ∗significant at 10%, ∗∗significant at 5%, ∗∗∗significant at 1%.

RAMIREZ HARRINGTON: STATE P2 POLICIES 271

reinforce each other’s influence on P2 adop- tion, similar to findings by Arimura, Hibiki, and Katayama (2008). Additionally, these policies increase Total P2 adoption even in states that do not emphasize toxic waste reduction (and they remain significant even if Toxic dummy is dropped from all models in Table 4), imply- ing that the effectiveness of such policies is not constrained by the emphasis of the state’s P2 legislation.

To demonstrate the magnitude of the effect of Reporting Requirement, Mandatory Planning, or both, I compute for the ratio of the condi- tional mean of P2 counts between states with and without each policy, which is a function of the relevant coefficients.13 The magnitude of the coefficients of Reporting Requirement and Mandatory Planning are very consistent across the models and are approximately 0.41 and 0.21, respectively. Thus, facilities in states with reporting requirement have 1.5 times as much P2 counts as those in states which do not require reporting, while facilities in states with manda- tory planning have 1.2 times as much P2 counts as those in states without mandatory planning. For those facilities in states with both Report- ing Requirement and Mandatory Planning, their count of P2 practices is 1.86 times that of facil- ities in states with neither. (These three ratios are statistically different from one at the 1%, 5%, and 5% levels, respectively). If one takes into account that being in states whose P2 pro- grams emphasize toxic waste reduction leads to even higher Total P2, the coefficients from the models suggest that facilities in states with a toxic waste reduction focus and which impose a Reporting Requirement, Mandatory Planning, or both will respectively have 1.97, 1.61, and 2.43 times as much P2 counts as those in states with- out them. (These ratios are statistically different from one at 1%.)

I further investigated the role of Numerical Goal in Model III where I alternatively use the level of pollution reduction, Target, in place of the Numerical Goal dummy and I obtain results similar to those in Models I and II, indicating that even differences in stringencies of targets do not affect adoption of Total P2. In fact, Numerical Goal also remains insignificant when its interactions with Lagged Inspections and Lagged Penalties are included, and the

13. The ratio is computed as E(P 2it |P 2it−1,P 2i0,2it ,ci ,Legit =1,Policyit =1) E(P 2it |P 2it−1,P 2i0,2it ,ci ,Legit =1,Policyit =0) = exp[βPolicy].

interactions are also insignificant (results not shown). The results suggest that regardless of how much regulatory action or enforcement activity that a facility has been subjected to in the past, Numerical Goal is an ineffective policy tool in promoting P2. These results hold if the Toxic dummy is dropped from all models in Table 4.

I further investigate whether the response to Reporting Requirement and Mandatory Planning depend on facility characteristics in Models IV-A and IV-B by including interaction terms. In Model IV-A, the interaction between Report- ing Requirement and Lagged Toxic Releases is significant, while none of the other pol- icy variables or any other explanatory vari- ables experience a change in sign, magnitude and significance. Specifically, the sign of the interaction term is negative, which implies that facilities that respond more positively to Report- ing Requirement by adopting more P2 are those which have been able to demonstrate good envi- ronmental performance in the past (those with low Lagged Toxic Releases). This suggests that the mandatory information disclosure program promotes better environmental behavior among those which have good news to report, but not for those whose toxic releases are too high. Using the coefficients in Model IV-A, the level of toxic releases needs to be extremely high, 3,402 million pounds, (99th percentile) before reporting will have a negative impact on the P2 count.14 Hence, for most of the facilities in the study, P2 counts are higher when report- ing is mandated in the state P2 program, but extremely dirty ones are “penalized” by the pol- icy. Note further that the sample used in this analysis has relatively higher toxic releases than those in the original unbalanced sample of all TRI reporters. Thus, even among these relatively high emitters, the results show that a policy that mandates reporting will still promote P2 adoption among most facilities. These results suggest that the response to a reporting require- ment may be driven by the desire to reduce potential enforcement action in the future, which is consistent with the findings of Maxwell and

14. In models with an interaction, the ratio can be computed for facilities in states with and without reporting requirement for various values of the Lagged Releases, that is, exp[βReport + γReport−Lagged Releases(Lagged Releases ∗ Report)it ]. The ratio is significantly lower than unity (at the 11%) when the natural log of lagged releases is at least 8.13 (lagged releases is at least 3,402 million pounds).

272 CONTEMPORARY ECONOMIC POLICY

Decker (2006) and Decker (2005) who show that disclosure of information can be used to lower the probability of being inspected and that those which have reported lower toxic releases per unit of output receive fewer inspections. In an alternative model where Reporting Require- ment is interacted with Final good dummy, the interaction term is not significant but the rest of the coefficients are unchanged (results not shown).

Model IV-B further investigates how a facil- ity responds to Mandatory Planning given its characteristics. It includes an interaction term between Mandatory Planning and Lagged Cumulative P2. In this model, Mandatory Plan- ning becomes negative but insignificant (it used to be positive and significant in the model with- out the interaction term), but the interaction term is positive and significant. This implies that for facilities that have not adopted any P2 in the past (36% of the sample), mandatory planning will not result in significantly higher (or lower) count of P2 activities. But for those facilities which have adopted at least one P2 practice in the past and are in states with Mandatory Plan- ning, they adopt significantly more P2 practices than those in states without such a policy. How much mandatory planning can promote P2 is shown by the ratio of the conditional mean of P2 counts of facilities in states with planning to the conditional mean of those in states without plan- ning for various values of Lagged Cumulative P2. The ratio is significantly greater than one at the 5% level of significance for facilities with at least 30 counts of Lagged Cumulative P2 prac- tices in the past (within the 90th percentile).15

These findings lend support to Bennear (2006) who asserts that the management-based regula- tions will be successful when the regulated facil- ity enjoys low costs to pollution reduction. Since the literature shows how past experience lowers cost of adoption, the findings in this study imply that a high enough accumulation of P2-specific experience lowers a facility’s costs of exploit- ing other sources of knowledge and integrating new knowledge generated through P2 planning, which in turn lowers the costs of finding new P2 opportunities.

15. If an interaction with Lagged Cumulative P2 is included, the ratio can be computed for facilities in states with and without mandatory planning for various values of the Lagged Cumulative P2, that is, exp[βPlanning + γPlanning−Lagged Cumulative P2(Lagged Cumulative P2∗Plann- ing)it ]. The ratio is significantly higher than unity when the natural log of cumulative P2 is 3.43 or when count is at least 30 practices.

B. Effect of P2 Policies on Toxic Releases

The results of the models explaining Toxic Releases are in Table 5. All models are esti- mated using the Anderson and Hsiao (1982) two-step approach and include the lagged depen- dent variable, state dummies, year dummies, industry dummies, and time-varying covari- ates to control for the past history of pollu- tion, fixed differences between states, national trends that affect all facilities, fixed differences between industries, and the time-varying char- acteristics affecting toxic releases at each time period, respectively. Model VI is the base model that includes P2 Legislation and Toxic dummy. Model VII includes all the policy instruments, while Model VIII uses Target in place of Numer- ical Goal. Model IX-A and IX-B include inter- action terms. All models in Table 5 show that P2 legislations are effective in reducing Toxic Releases in states whose legislated programs emphasize toxic waste reduction, consistent with Hypothesis 1b. Similar to Earnhart (2009), the findings here suggest that state agencies are able to provide an additional threat to facilities that are distinct from the threat associated with fed- eral legislations.

However, I do not find evidence to support Hypotheses 2b, 3b, and 4b: none of the policy instruments I investigate are able to encourage further toxic pollution reduction. In conjunc- tion with the significant effect of the legisla- tion, these results are consistent with findings by Thornton, Gunningham, and Kagan (2005) who find that promulgation of laws and uncer- tainty surrounding their enforcement may be more salient than actual sanctions. Except for Numerical Goal, the coefficients of the policy variables are slightly larger when the policy variables are included one at a time but their coefficients remain insignificant in those mod- els (results now shown). In Model VI, I use the actual level pollution reduction Target in place of the Numerical Goal dummy, and I find that states with more stringent targets do not have lower toxic pollution either. Despite the unexpected findings, the results regarding the insignificant effect of Numerical Goal or Target is consistent with findings of Jaffe and Stavins (1995) and Stafford (2003) who find mandated state-level codes to be ineffective in induc- ing energy efficiency and compliance, respec- tively. The nonsignificant effect of Mandatory Planning however, is in contrast with the find- ings of Bennear (2006) and Arimura, Hibiki, and Katayama (2008). Reporting Requirement

RAMIREZ HARRINGTON: STATE P2 POLICIES 273

TABLE 5 Determinants of Toxic Releases, Anderson–Hsiao Estimators

Variables VI VII VIII IX-Aa IX-Ba

P2 Legislation 0.02712 0.02744 0.02757 0.02734 0.02829 (0.0265) (0.0265) (0.0265) (0.0265) (0.0265)

Toxic dummy −0.31253∗∗∗ −0.31258∗∗∗ −0.31241∗∗∗ −0.31268∗∗∗ −0.31384∗∗∗ (0.0903) (0.0903) (0.0902) (0.0903) (0.0902)

Numerical Goal −0.00266 −0.00731 0.00457 (0.0188) (0.0202) (0.0190)

Reporting Requirement 0.05883 0.05881 0.05776 0.06935 (0.1011) (0.1011) (0.1012) (0.1011)

Mandatory Planning 0.02036 0.0205 0.02061 0.02068 (0.0389) (0.0389) (0.0389) (0.0389)

Target 0.00019 (0.0003)

Numerical Goal × Lagged Inspections 0.00607 (0.0094)

Numerical Goal × Lagged Penalties −0.08169∗∗∗ (0.0298)

Lagged Inspections −0.00207 −0.00215 −0.00211 −0.00294 −0.00214 (0.0034) (0.0034) (0.0034) (0.0036) (0.0034)

Lagged Penalties −0.02332∗ −0.02351∗ −0.02363∗ −0.02364∗ −0.00632 (0.0127) (0.0127) (0.0127) (0.0127) (0.0141)

Lagged Toxic Releases 1.01843∗∗∗ 1.01841∗∗∗ 1.01793∗∗∗ 1.01910∗∗∗ 1.01629∗∗∗

(0.1285) (0.1286) (0.1285) (0.1287) (0.1283) Lagged Toxicity-Weighted Releases −0.40768∗∗∗ −0.40777∗∗∗ −0.40760∗∗∗ −0.40810∗∗∗ −0.40690∗∗∗

(0.0464) (0.0464) (0.0464) (0.0464) (0.0463) Nonattainment 0.02741∗ 0.02783∗ 0.02774∗ 0.02787∗ 0.02810∗

(0.0161) (0.0161) (0.0161) (0.0161) (0.0161) Final Good dummyb −0.17499∗∗∗ −0.17517∗∗∗ −0.17512∗∗∗ −0.17530∗∗∗ −0.17320∗∗∗

(0.0165) (0.0165) (0.0165) (0.0165) (0.0164) Income 0.00000193 0.00000211 0.00000207 0.00000214 0.00000214

(0.00000830) (0.00000830) (0.00000830) (0.00000830) (0.00000829) RD Intensity 0.2425 0.24278 0.23518 0.2445 0.22624

(0.3908) (0.3909) (0.3908) (0.3910) (0.3903) Number of Chemicalsb 0.83050∗∗∗ 0.83063∗∗∗ 0.83066∗∗∗ 0.83095∗∗∗ 0.82113∗∗∗

(0.0112) (0.0112) (0.0112) (0.0112) (0.0111) Constant −0.0027 −0.00306 −0.0031 −0.00309 −0.00317

(0.0110) (0.0110) (0.0110) (0.0110) (0.0110) SIC dummiesb 88.35∗∗∗ 88.42∗∗∗ 88.40∗∗∗ 88.44∗∗∗ 88.30∗∗∗

State dummiesb 21.62∗∗∗ 21.51∗∗∗ 21.49∗∗∗ 21.52∗∗∗ 21.76∗∗∗

Year dummies 22.51∗∗∗ 22.30∗∗∗ 22.20∗∗∗ 22.30∗∗∗ 21.93∗∗∗

No. of observationsc 11,349 11,349 11,349 11,349 11,349 No. of groups 1,261 1,261 1,261 1,261 1,261

aModels with other interaction terms between Reporting requirement or P 2 Legislation and Lagged Inspections or Lagged Penalties are not shown for the interest of brevity but in all these models, the interaction terms are not statistically significant while the rest of the coefficients variables are similar to those in Model VII. A complete set of these results are available upon request.

bThese are the variables that are estimated in the second stage of the Anderson–Hsiao two-step framework. cThere are 10,088 observations in Models VI–IX because the Anderson–Hsiao models require 2-year lagged levels as

instrument for the first differenced 1-year lagged dependent variable. Standard errors in parentheses: ∗significant at 10%, ∗∗significant at 5%, ∗∗∗significant at 1%.

is not significant either, which is contrary to findings of many studies that show mandatory information disclosure mechanisms yield signifi- cant pollution reductions. However, if viewed in conjunction with the findings of Bae, Wilcoxen,

and Popp (2009), my results may suggest that state agencies need to further process and disseminate the reported information better to make them more relevant to the public and local communities before these external stakeholders

274 CONTEMPORARY ECONOMIC POLICY

can effectively use the information to influence the environmental performance of facilities.

I investigate the lack of significance of the policy instruments to determine whether there are specific types of facilities that would reduce their toxic emissions in response to the poli- cies by constructing several interaction terms between Lagged Inspections or Lagged Penal- ties and Numerical Goal, Reporting Require- ment, Mandatory Planning, or P2 Legislation dummies and included each interaction term one by one in the model. I show in Models IX-A and IX-B two of these models which include interaction terms between Numerical Goal and Lagged Inspections (Models IX-A) and Lagged Penalties (IX-B). I find that among all interaction terms investigated, only the inter- action between Numerical Goal variable with Lagged Penalties is negative and significant (in Model IX-B) while the rest of the coefficients remain unchanged, providing some evidence for Hypothesis 5b. While the coefficient of Lagged Penalties is slightly lower and loses significance in Model IX-B, the elasticity of Toxic Releases with respect to Lagged Penalties is now stronger (but still very modest), at −0.09, compared to elasticity in Models VI-VIII, at only −0.02. The findings also show that state pollution reduc- tion targets are effective in reducing emissions only among highly noncompliant facilities, that is, those which have already been subjected to more enforcement action in the past. Specif- ically, toxic emissions are significantly lower among those facilities located in states with a Numerical Goal if they have been subjected to at least eight penalties (75th percentile) in the previous year.16 This is consistent with Earnhart (2009) who find that the potency of deterrence instruments is greater among polluting entities that are facing more regulatory scrutiny.

VI. SUMMARY AND CONCLUSIONS

Regulations for pollution reduction have increasingly emphasized P2 over end-of-pipe abatement through legislation of P2 programs offering a combination of numerical goals, reporting requirements, and management poli- cies. The findings of this study show a significant effect of the toxic waste reduction legislation in reducing toxic emissions, and to a less extent in

16. This is the number of penalties that would make the marginal effect of Numerical Goal dummy statistically negative in Model IX-B.

promoting P2 adoption, which indicates that the potential for enforcement action and provision of technical assistance through the state legis- lation may be providing additional regulatory threat while also reducing the costs of under- taking pollution abatement. However, the policy instruments play very different roles in further influencing P2 adoption and the level of toxic releases. Specifically, the results show that facil- ities in states with reporting requirement and mandatory planning adopt significantly more P2 practices, even in states that do not emphasize toxic waste reduction, while numerical goals prescribed under the state P2 programs do not yield significantly higher P2 counts. On the other hand, numerical goals yield significantly lower toxic pollution levels but only among those which have been subjected to greater enforcement action in the past. Thus, while policy instruments enhance the effect of toxic waste legislation in increasing P2 adoption, they are largely ineffectual in further reducing pol- lution levels. Overall, these results are consis- tent with the survey findings of Koehler (2007) who finds significant demonstration of efforts (through participation in voluntary environmen- tal programs) in response to various measures of regulatory threat and other external pressures, but no meaningful or long-term environmental performance improvements.

Several insights can be gleaned from these findings. First, the significance of reporting requirement and mandatory planning for P2 adoption suggest some consistency with the growing emphasis on management- and information-based instruments to promote P2 activities. It also lends some support to the find- ings of Lent and Wells (1994) who find that the trend and pattern of firm investment in environ- mental management shows a shift in investment priorities from remediation and regulatory com- pliance toward investments that will reduce pol- lution and increase competitive advantage. As management- and information-based policies are more flexible and allow firms to respond to mar- ket factors, it is not very surprising that they are the types of approaches that may promote strate- gic and beyond-compliance types of activities such as P2.

Second, that reporting increases P2 adoption but does not reduce toxic releases shows a very limited role for information disclosure as a pol- icy tool, at least at the state level, contrary to numerous positive findings about the TRI. The disclosure of environmental-related plans

RAMIREZ HARRINGTON: STATE P2 POLICIES 275

and actions to the state regulatory agency seem to create incentives to adopt P2 practices that would enable facilities to sufficiently signal envi- ronmental stewardship through their technology adoption behavior possibly to deflect regulatory action and other pressures, but without making any meaningful pollution reduction. By improv- ing the quality of information that is made pub- licly available, say through better data processing and dissemination as Bae, Wilcoxen, and Popp (2009) find is necessary for TRI data, the state agencies may be able to create incentives for facilities to undertake P2 activities that do not only improve reputation through their demonstra- tion of efforts but also activities that can actually reduce toxic releases. Nonetheless, the findings emphasize the power of good news: how infor- mation disclosure promotes P2 among those who have demonstrated good behavior in the past and that information disclosure may be potentially detrimental to extremely dirty facilities.

Third, the significant impact of a manage- ment policy in promoting P2 adoption con- tributes to both economic literature and policy regarding the role that environmental manage- ment systems play in promoting adoption of environmentally responsible practices especially in the context of the EPA’s specific emphasis on management based tools to promote P2. A key contribution of this study is the observed com- plementarity between planning and past expe- rience, implying that P2 planning as a policy tool cannot entirely operate in a vacuum; it is a more potent instrument among facilities that have accumulated enough P2-specific expertise in the past. Thus, despite the recent popularity of employing management-based regulations, it would be foolish to abandon other policy tools that can promote even modest adoption of P2 practices to allow facilities to build experience and know-how that will allow them take advan- tage of the benefits of P2 planning. The find- ings also reveal that not only are management based regulations effective in promoting P2, the impact of management systems is distinct from

and reinforces the effect of other policies such as reporting requirement, which is consistent with findings of Foulon, Lanoie, and Laplante (2002) and Arimura, Hibiki, and Katayama (2008). However, while the findings suggest that P2 planning is able to lower the cost of P2 adop- tion, they also imply that P2 planning is not able to facilitate adoption of other abatement techniques that lead to significant reduction in toxic releases. Reduction of toxic releases may be more responsive to other abatement activi- ties (say, end-of-pipe techniques) that are not promoted by P2 planning. Because P2 planning may be effective in promoting specific means of pollution reduction, but not in achieving the ultimate goal of lowering pollution levels, it may need to be complemented by other policy instruments.

Finally, any discussion of the role of pol- icy instruments in inducing environmentally responsible behavior is incomplete without pro- viding the dynamic context to the analysis. Thus, despite the finding that they only pro- mote investment in environmental technologies without bringing about consequential environ- mental improvements and that the coefficients of the policy instruments are small compared to the very large and statistically strong coef- ficients of history of P2 adoption and history toxic releases, the impact of the policy instru- ments would be larger if we consider how the effect of past behavior perpetuates further out into the future. Thus, even with modest short- term impacts on P2 and toxic releases, this study shows that reporting requirement and manda- tory planning can promote P2 in the long term because the effect of accumulated knowledge and expertise is strong. For the case of toxic releases, strong path dependence in the level of releases suggests it may still be worthwhile to pursue numerical targets, even though it is only effective among highly noncompliant facilities because this policy significantly reduces levels of releases, which can then reduce future growth rate of pollution among violators.

276 CONTEMPORARY ECONOMIC POLICY

APPENDIX

TABLE A1 Robustness Check Using Prelegislation and Prepolicy Variables

Dependent Variable: Total P2 Poisson Random Effects

Dependent Variable: Toxic Releasesa Anderson–Hsiao Two-Step Model

Variables 1-A 1-B 2-A 2-B

Pre-P2 Legislation −0.00167 −0.00365 0.0272 0.02716 (0.0611) (0.0611) (0.0187) (0.0188)

Pre-Numerical Goal 0.04835 −0.0253 (0.0467) (0.0337)

Pre-Reporting Requirement 0.07751 0.01006 (0.3053) (0.0141)

Pre-Mandatory Planning −0.16798 0.00606 (0.1038) (0.0742)

Lagged P2 0.84059∗∗∗ 0.83996∗∗∗

(0.0170) (0.0170) Cumulative P2 −0.52117∗∗∗ −0.51990∗∗∗

(0.0259) (0.0259) 1991 P2 0.70909∗∗∗ 0.70843∗∗∗

(0.0532) (0.0532) Spillover P2 0.28803∗∗∗ 0.29033∗∗∗

(0.0332) (0.0333) Lagged Inspections 0.04097∗∗∗ 0.04016∗∗∗ −0.00214 −0.0022

(0.0110) (0.0110) (0.0034) (0.0034) Lagged Penalties 0.01604 0.01753 −0.02433∗ −0.02434∗

(0.0297) (0.0297) (0.0127) (0.0128) Lagged Toxic Releases 0.09665∗∗∗ 0.09748∗∗∗ 1.01852∗∗∗ 1.02020∗∗∗

(0.0218) (0.0218) (0.1286) (0.1288) Lagged Toxicity-Weighted Releases −0.40747∗∗∗ −0.40812∗∗∗

(0.0464) (0.0465) Nonattainment 0.02472 0.02521

(0.0161) (0.0161) Final Good dummyb −0.29342∗∗∗ −0.29331∗∗∗ −0.17511∗∗∗ −0.17531∗∗∗

(0.1049) (0.1049) (0.0165) (0.0165) Income 0.000014∗∗∗ 0.000014∗∗∗ 0.00000216 0.00000224

(0.00000467) (0.00000467) (0.00000820) (0.00000831) RD Intensity 0.59635 0.52452 0.24588 0.24503

(0.8092) (0.8100) (0.3910) (0.3913) Number of Chemicalsb 0.59583∗∗∗ 0.59531∗∗∗ 0.83104∗∗∗ 0.83130∗∗∗

(0.0825) (0.0825) (0.0112) (0.0112) Constant −2.94637∗∗∗ −2.94605∗∗∗ −0.00301 −0.00314

(0.6505) (0.6502) (0.0110) (0.0110) SIC dummiesb 43.62∗∗∗ 43.64∗∗∗ 88.42∗∗∗ 88.43∗∗∗

State dummiesb 66.68∗∗∗ 66.72∗∗∗ 18.65∗∗∗ 18.69∗∗∗

Year dummies 74.79∗∗∗ 73.16∗∗∗ 23.61∗∗∗ 23.46∗∗∗

LR (a) 3915.45∗∗∗ 3910.73∗∗∗ — — No. of observations 12,610 12,610 11,349 11,349 No. of groups 1,261 1,261 1,261 1,261

aThere are 10,088 observations in Models 2-A and 2-B because the Anderson–Hsiao models require 2-year lagged levels as instrument for the 1-year lagged differenced dependent variable.

bThese are the variables that are estimated in the second stage of the Anderson–Hsiao two-step framework. Standard errors in parentheses. +Significant at 15%, ∗significant at 10%, ∗∗significant at 5%, ∗∗∗significant at 1%.

RAMIREZ HARRINGTON: STATE P2 POLICIES 277

REFERENCES

Anderson, T. W., and C. Hsiao. “Formulation and Estima- tion of Dynamic Models Using Panel Data.” Journal of Econometrics, 18(1), 1982, 47–82.

Anton, W. R. Q., G. Deltas, and M. Khanna. “Incentives for Environmental Self-Regulation and Implications for Environmental Performance.” Journal of Environmen- tal Economics and Management, 48(1), 2004, 632–54.

Antweiler, W., and K. Harrison. “Toxic Release Invento- ries and Green Consumerism: Empirical Evidence from Canada.” Canadian Journal of Economics, 36(2), 2003, 495–520.

Arimura, T. H., A. Hibiki, and N. Johnstone. “An Empiri- cal Study of Environmental R&D: What Encourages Facilities to be Environmentally Innovative,” in Envi- ronmental Policy and Corporate Behaviour, edited by N. Johnstone. Cheltenham, UK: Edward Elgar, 2007.

Arimura, T. H., A. Hibiki, and A. H. Katayama. “Is a Vol- untary Approach an Effective Environmental Policy Instrument? A Case for Environmental Management Systems.” Journal of Environmental Economics Man- agement, 55(3), 2008, 281–95.

Arimura, T. H., N. Darnall, and H. Katayama. “Is ISO 14001 a Gateway to More Advanced Voluntary Action? The Case of Green Supply Chain Management.” Journal of Environmental Economics and Management, 61(2), 2011, 170–82.

Arora, S., and T. Cason. “Why Do Firms Volunteer to Exceed Environmental Regulations: Understanding Participation in EPA’s 33/50 Program.” Land Eco- nomics, 72(4), 1996, 413–32.

. “Do Community Characteristics Influence Environ- mental Outcomes? Evidence from the Toxics Release Inventory.” Southern Economic Journal, 65(4), 1999, 691–716.

Bae, H., P. Wilcoxen, and D. Popp. “Information Disclo- sure Policy: Do State Data Processing Efforts Help More Than the Information Disclosure Itself?” Jour- nal of Policy Analysis and Management, 29(1), 2009, 163–82.

Becker, R. A.“Pollution Abatement Expenditure by U.S. Manufacturing Plants: Do Community Characteris- tics Matter?” B.E. Journal of Economic Analysis and Policy, 3(2), 2004. Accessed January 5, 2012. http:// www.bepress.com/bejeap/contributions/vol3/iss2/art6.

Bennear, L. S. “Evaluating Management-Based Regulations: A Valuable Tool in the Regulatory Toolbox?” in Leveraging the Private Sector: Management Strategies for Improving Environmental Performance, edited by C. Coglianese and J. Nash. RFF Press, 2006.

. “Are Management-Based Regulations Effective? Evidence from State Pollution Prevention Programs.” Journal of Policy Analysis and Management, 26(2), 2007, 327–48.

Bennear, L. S., and S. Olmstead. “The Impacts of the ‘Right to Know’: Information Disclosure and the Violation of Drinking Water Standards.” Journal of Environmental Economics and Management, 56(2), 2008, 117–30.

Brooks, N., and R. Sethi. “The Distribution of Pollu- tion: Community Characteristics and Exposure to Air Toxics.” Journal of Environmental Economics and Management, 32(2), 1997, 232–50.

Brouhle, K., and D. R. Harrington. “GHG Registries: Par- ticipation and Performance under the Canadian Volun- tary Climate Challenge Program.” Environmental and Resource Economics, 47(4), 2010, 521–48.

Brouhle, K., C. Griffiths, and A. Wolverton. “Evaluating the Role of EPA Policy Levers: An Examination of a Voluntary Program and Regulatory Threat in the Metal-Finishing Industry.” Journal of Environmental Economics and Management, 57(2), 2009, 166–81.

Brunnermeier, S., and M. Cohen. “Determinants of Environmental Innovation in US Manufacturing Indus- tries.”Journal of Environmental Economics and Man- agement, 45(2), 2003, 278–93.

Cameron, A. C., and P. K. Trivedi. Regression Analysis of Count Data. Cambridge: Cambridge University Press, 1998.

Coglianese, C., and J. Nash. “Management-Based Strategies: An Emerging Approach for Environmental Protec- tion,” in Leveraging the Private Sector: Management Strategies for Improving Environmental Performance, edited by C. Coglianese and J. Nash. Washington, DC: RFF Press, 2006.

Cohen, W., and D. Levinthal. “Innovation and Learning: The Two Faces of R&D.” The Economic Journal, 99(397), 1989, 569–96.

DeCanio, S., C. Dibble, and C. K. Amir-Atefi. “The Impor- tance of Organizational Structure for the Adoption of Innovations.” Management Science, 46(10), 2000, 1285–99.

Decker, C. S. “Do Regulators Respond to Voluntary Pollu- tion Control Efforts? A Count Data Analysis.” Con- temporary Economic Policy, 23(2), 2005, 180–94.

Delmas, M., M. J. Montes-Sancho, and J. Shimshack. “Information Disclosure Policies: Evidence from the Electricity Industry.” Economic Inquiry, 48(2), 2009, 483–98.

Earnhart, D. “Panel Data Analysis of Regulatory Factors Shaping Environmental Performance.” Review of Eco- nomic and Statistics, 86(1), 2004a, 391–401.

. “The Effects of Community Characteristics on Polluter Compliance Levels.” Land Economics, 80(3), 2004b, 408–32.

. “The Influence of Facility Characteristics and Per- mit Conditions on the Effectiveness of Environmental Regulatory Deterrence.” Journal of Regulatory Eco- nomics, 36(3), 2009, 247–73.

Foulon, J., P. Lanoie, and B. Laplante. “Incentives for Pol- lution Control: Regulation or Information?” Journal of Environmental Economics and Management, 44(1), 2002, 169–87.

Freeman, H., T. Harten, J. Springer, P. Randall, M. A. Curran, and K. Stone. Industrial Pollution Preven- tion: A Critical Review. Pollution Prevention Research Branch, Risk reduction Engineering Laboratory, US Environmental Protection Agency, Cincinnati, Ohio, 1992.

Frondel, M., J. Horbach, and K. Rennings. “End-of-Pipe or Cleaner Production? An Empirical Comparison of Environmental Innovation Decisions across OECD Countries,” in Environmental Policy and Corporate Behaviour, edited by N. Johnstone. Cheltenham, UK: Edward Elgar, 2007.

Gray, W., and R. Shadbegian. “Environmental Regulation, Investment Timing, and Technology Choice.” The Journal of Industrial Economics, 46(2), 1998, 235–56.

Gray, W., and J. Shimshack. “The Effectiveness of Environ- mental Monitoring and Enforcement: A Review of the Empirical Evidence.” Review of Environmental Eco- nomics and Policy, 5(1), 2011, 3–24.

Hall, J. “Environmental Supply Chain Innovation.” Greener Management International, 35, 2001, 105–19.

Hamilton, J. T. “Testing for Environmental Racism: Preju- dice, Profits, Political Power?” Journal Policy Analysis and Management, 14(1), 1995, 107–32.

Harrington, D. R., M. Khanna, and G. Deltas. “Striving to Be Green: The Adoption of Total Quality Environmen- tal Management.” Applied Economics, 40(23), 2008, 2995–3007.

Harrison, K., and W. Antweiler. “Incentives for Pollu- tion Abatement: Regulation, Regulatory Threats, and

278 CONTEMPORARY ECONOMIC POLICY

Non-Governmental Pressures.” Journal of Policy Anal- ysis and Management, 22(3), 2003, 361–82.

Hart, S. L., and G. Ahuja. “Does It Pay to Be Green? An Empirical Examination of the Relationship between Emission Reduction and Firm Performance.” Business Strategy and the Environment, 5(1), 1996, 30–37.

Innes, R., and A. Sam. “Voluntary Pollution Reductions and the Enforcement of Environmental Law: An Empirical Study of the 33/50 Program.” Journal of Law and Economics, 51(2), 2008, 271–96.

Jaffe, A. B. “Technological Opportunity and Spillovers of R&D: Evidence from Firms’ Patents, Profits, and Market Value.” American Economic Review, 76(5), 1986, 984–1001.

Jaffe, A. B., and R. N. Stavins. “Dynamic Incentives of Environmental Regulation: The Effects of Alternative Policy Instruments on Technology Diffusion.” Journal of Environmental Economics and Management, 29(3), 1995, S43–63.

. “A Tale of Two Market Failures: Technology and Environmental Policy.” Ecological Economics, 54(203), 2005, 164–74.

Khanna, M., and L. Damon. “EPA’s Voluntary 33/50 Pro- gram: Impact on Toxic Releases and Economic Perfor- mance of Firms.” Journal of Environmental Economics and Management, 37(1), 1999, 1–25.

Khanna, M., D. R. Harrington, and G. Deltas. “Adoption of Pollution Prevention Techniques: The Role of Manage- ment Systems, Demand-Side Factors and Complemen- tary Assets.” Environmental and Resource Economics, 44(1), 2009, 85–106.

Khanna, M., W. Quimio, and D. Bojilova. “Toxics Release Information: A Policy Tool for Environmental Protec- tion.” Journal of Environmental Economics and Man- agement, 36(3), 1998, 243–66.

Kim, E. H., and T. P. Lyon. “Strategic Environmental Dis- closure: Evidence from the DOE’s Voluntary Green- house Gas Registry.” Journal of Environmental Eco- nomics and Management, 61(3), 2011, 311–26.

King, A., and M. Lennox. “Exploring the Locus of Prof- itable Pollution Reduction.” Management Science, 48(2), 2002, 289–99.

Koehler, D. “The Effectiveness of Voluntary Environmental Programs—A Policy at a Crossroads?” The Policy Studies Journal, 35(4), 2007, 689–722.

Lanoie, P., M. Thomas, and J. Fearnley. “Firms Responses to Effluent Regulations: Pulp and Paper in Ontario, 1985–1989.” Journal of Regulatory Economics, 13(2), 1998, 103–20.

Lennox, M., and A. King. “Prospects for Developing Absorptive Capacity through Internal Information Pro- vision.” Strategic Management Journal, 25(4), 2004, 331–45.

Lent, T., and R. P. Wells. “Corporate Environmental Man- agement Survey Shows Shift from Compliance to Strategy,” in Environmental TQM. 2nd ed., edited by J. T. Willig. New York: McGraw Hill, 1994.

List, J. A., W. A. Mchone, and D. L. Millimet. “Effects of Environmental Regulation on Foreign and Domestic Plant Births: Is There a Home Field Advantage?” Journal of Urban Economics, 56(2), 2004, 303–26.

Mansfield, E. The Economics of Technical Change. New York: Norton, 1968.

Maxwell, J. W., and Decker, C. S. “Voluntary Envi- ronmental Investment and Responsive Regulation.”

Environmental and Resource Economics, 33(4), 2006, 425–39.

Sam, A., M. Khanna, and R. Innes. “Voluntary Pollution Reduction Programs, Environmental Management, and Environmental Performance: An Empirical Study.” Land Economics, 85(4), 2009, 692–711.

Shimshack, J. P., and M. Ward. “Regulator Reputation, Enforcement, and Environmental Compliance.” Jour- nal of Environmental Economics and Management, 50(3), 2005, 519–40.

Stafford, S. L. “Assessing the Effectiveness of State Regu- lation and Enforcement of Hazardous Waste,′′ Journal of Regulatory Economics, 23(1), 2003, 27–41.

Thornton, D., N. A. Gunningham, R. A. Kagan. “General Deterrence and Corporate Environmental Behavior.” Law and Policy, 25(2), 2005, 262–88.

Tietenberg, T. “Disclosure Strategies for Pollution Control.” Environmental and Resource Economics, 11(3–4), 1998, 587–602.

U.S. Environmental Protection Agency. Improving Tech- nology Diffusion for Environmental Protection: Report and Recommendations of the Technology Innovation and Economics Committee. Washington, DC, 1992.

. How Are the Toxics Release Inventory Data Used? Government, Business, Academic and Citizen Uses. Washington, DC, 2003.

. Toxic Chemical Releases Inventory Reporting Forms and Instructions. Washington, DC, 2007.

Vidovic, M., and N. Khanna. “Can Voluntary Pollution Prevention Programs Fulfill Their Promises? Further Evidence from the EPA’s 33/50 Program.” Journal of Environmental Economics and Management, 53(2), 2007, 180–95.

Welch, E., A. Mazur, and S. Bretschneider. “Voluntary Behavior by Electric Utilities: Levels of Adoption and Contribution of the Climate Challenge Program to the Reduction of Carbon Dioxide.” Journal of Policy Anal- ysis and Management, 19(3), 2000, 407–25.

Wolverton, A. “Effects of Socio-Economic and Input- Related Factors on Polluting Plants’ Location Deci- sions.” The B.E. Journal of Economic Analysis & Policy, 9(1), 2009. Accessed January 5, 2012. http:// www.bepress.com/bejeap/vol9/iss1/art14.

Wooldridge, J. M. “Simple Solutions to the Initial Con- ditions Problem in Dynamic Nonlinear Panel Data Models with Unobserved Heterogeneity.” Journal of Applied Econometrics, 20(1), 2005, 39–54.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Appendix S1. Probit Models for State P2 Legislation/ Policy Adoption. Table S1. Descriptive Statistics of Variables for Probit Models Table S2. Probit Model Results: P2 Legislation and Policy Adoption

Copyright of Contemporary Economic Policy is the property of Wiley-Blackwell and its content may not be

copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.