Do Carrots Work? Examining the Effectiveness of EPA's Compliance Assistance Program

Do Carrots Work? Examiningthe Effectiveness of EPA’sCompliance AssistanceProgram

Sarah Stafford

Abstract

The role of compliance assistance in the U.S. Environmental Protection Agency’s overallenforcement strategy has been quite variable over the past decade and a half, increasingin prominence under the Bush administration and now slated for significantly reducedfunding under the Obama administration. While theoretical models and anecdotalevidence suggest that compliance assistance should play some role in a comprehen-sive enforcement strategy, to date there has been relatively little empirical evidence onthe actual effectiveness of existing compliance assistance programs. To help informthe debate over the appropriate use of compliance assistance, this paper uses dataon hazardous waste generators nationwide to assess the effect of federal complianceassistance programs in improving compliance with hazardous waste regulations. Thepaper also conducts a direct empirical analysis of the relationship between traditionalenforcement tools and compliance assistance. The results show that federal compli-ance assistance efforts do increase compliance, but the evidence does not suggest anyconsistent relationship between traditional enforcement and compliance assistance.Also, while states do not appear to substitute federal compliance assistance for tradi-tional enforcement, state compliance assistance programs do appear to decrease thelikelihood of inspections among the smallest hazardous waste generators. C© 2012 bythe Association for Public Policy Analysis and Management.

INTRODUCTION

Between 1998 and 2008, the U.S. Environmental Protection Agency’s (EPA) enforce-ment budget fell from over $590 million to less than $560 million in real dollars, andstaffing in its Office of Enforcement and Compliance Assurance fell from around3,900 full-time equivalent employees to 3,400.1 During this same time, complianceassistance began to play a more significant role in EPA’s overall enforcement andcompliance strategy, particularly once the Bush administration came into power.For example, in EPA’s 1997 Strategic Plan, compliance assistance was mentionedonly in passing.2 By 2000, compliance assistance was more prominently featuredin EPA’s Strategic Plan, but was clearly presented as a supplement to a strong

1 See Gray and Shimshack (2011, Figure 1).2 U.S. EPA (1997, p. 57).

Journal of Policy Analysis and Management, Vol. 00, No. 0, 1–23 (2012)C© 2012 by the Association for Public Policy Analysis and ManagementPublished by Wiley Periodicals, Inc. View this article online at wileyonlinelibrary.com/journal/pamSupporting Information is available in the online issue at wileyonlinelibrary.com.DOI:10.1002/pam.21633

2 / Effectiveness of EPA’s Compliance Assistance Program

traditional enforcement regime.3 However, in 2003, EPA’s Strategic Plan identifiedcompliance assistance as the first method through which EPA intended to increasecompliance, followed by compliance incentives and finally, enforcement.4 Thus un-der the Bush administration, compliance assistance increasingly came to be seenas a viable substitute for more traditional enforcement measures. In contrast, Pres-ident Obama’s 2012 proposed budget for the EPA eliminated all separate fundingfor compliance assistance and transferred those funds to traditional compliancemonitoring and civil and criminal enforcement.5

Interestingly, neither the increased role of compliance assistance in the Bush EPAnor the proposed decreased role of compliance assistance in the Obama EPA appearsto be based on any evaluation of the effectiveness of compliance assistance programseither in general or relative to traditional enforcement methods. While theoreticalmodels and anecdotal evidence suggest that compliance assistance should play somerole in a comprehensive enforcement and compliance strategy, to date there has beenrelatively little empirical evidence on the actual effectiveness of existing complianceassistance programs. The purpose of this paper is to provide some evidence onthis issue, specifically to determine whether compliance assistance does increaseenvironmental compliance using hazardous waste regulations as a case study. Inaddition, the study examines the relationship between compliance assistance andtraditional deterrence in increasing overall compliance.

The next section provides a description of EPA’s approach to compliance as-sistance. The third section presents a theoretical framework for the analysis anddiscusses the related literature. The fourth section explains the empirical approachand describes the data used in the analysis, and the fifth section presents the resultsof the analysis. Finally, the sixth section discusses the policy implications of theseresults, and the seventh section summarizes the results and suggests avenues forfuture research.

EPA’S APPROACH TO COMPLIANCE ASSISTANCE

EPA uses the term compliance assistance to describe activities and technical assis-tance efforts that help regulated entities understand and meet their environmentalcompliance obligations or voluntarily adopt environmentally beneficial practicesthat go beyond compliance. Compliance assistance includes facility visits and coun-seling, online and hardcopy resources such as fact sheets and guides, and trainingprograms and workshops. Compliance assistance is related to, but distinct from,compliance incentives. Compliance incentives are voluntary programs that offerfirms direct rewards to encourage increased compliance, such as providing regula-tory relief or flexibility in how certain firms comply with a regulation or decreasingpenalties for facilities that self-police to firms to encourage increased compliance.6

EPA began offering formal compliance assistance after its enforcement functionswere reorganized in 1994 to create a single Office of Enforcement and ComplianceAssurance.7 One of EPA’s first actions was to open four industry-specific compli-ance assistance centers targeted to small businesses (U.S. EPA, 1997). In 1999,

3 “While EPA will maintain a strong presence in enforcement, we will also bring a mix of innovativecompliance tools . . . to bear on environmental programs,” U.S. EPA (2000, p. 6).4 U.S. EPA (2003, pp. 111–113). The 2006 Strategic Plan contained similar language to the 2003 StrategicPlan, U.S. EPA (2006, pp. 128–133).5 U.S. EPA (2011, p. 80).6 Potoski and Prakash (2004) provide a more detailed discussion of regulatory relief programs. Stafford(2007) discusses EPA’s self-policing program.7 U.S. EPA (1996, pp. 5–19).

Journal of Policy Analysis and Management DOI: 10.1002/pamPublished on behalf of the Association for Public Policy Analysis and Management

Effectiveness of EPA’s Compliance Assistance Program / 3

EPA further refined its compliance assistance strategy by adopting a wholesaler ap-proach to compliance assistance.8 Under this approach, EPA focuses on developingcompliance assistance tools and materials and working with states, localities, andprivate providers to deliver additional assistance to the regulated community. How-ever, EPA continues to provide targeted compliance assistance directly to regulatedentities.

Today EPA provides compliance assistance materials and funds sector-specificcompliance assistance centers for regulated entities across the United States. Forexample, the ECAR center is a “‘one-stop shop’ for all automotive dismantling andrecycling operations and provides comprehensive and up-to-date environmentalcompliance assistance.”9 In addition, EPA’s 10 regional offices provide additionalcompliance assistance tools and workshops for regulated entities in each region andconduct on-site compliance assistance visits. For example, in 2010 EPA’s Region 4office held a series of compliance assistance workshops for prisons and correctionsfacilities to ensure that staff of such facilities were aware of the environmentalregulations to which such facilities were likely to be subject, what facilities neededto do to be in compliance, and places to go for additional help in complying with theregulations.10 In 2008, Region 7 developed an information guide that it translatedinto five languages to educate ethnic grocery stores about the regulations governingimported pesticide products and the selling of unregistered pesticide products thatmight be hazardous to human health.11

Many state environmental protection agencies and departments of environmentalquality also provide compliance assistance including on-site visits, compliance as-sistance hotlines, and workshops. For example, Kentucky’s Department of Environ-mental Protection has a Division of Compliance Assistance that offers a ComplianceAssistance Workbook for regulated entities to take inventory of their environmentalresponsibilities and to identify areas where assistance is needed. Staff members areavailable to answer questions via e-mail or phone and will provide on-site assis-tance and training where necessary.12 In addition to federal and state complianceassistance programs, regulated entities can seek compliance assistance from theprivate sector including trade associations, nonprofit organizations, and for-profitconsultants. As mentioned above, as part of its wholesaler approach, EPA providestraining to private sector providers to make sure that they are informed of the latestdevelopments and changes in environmental regulations.13 Of course, private sec-tor compliance assistance is paid for by regulated entities, not through governmentfunding.

At both the federal and the state level, many compliance assistance programsare focused on small entities. For example, recognizing that small communitiesoften have very small staffs, Arizona’s Department of Environmental Quality has aSmall Communities Environmental Assistance Project to help communities with lessthan 10,000 residents comply with state environmental regulations.14 Virginia has aSmall Business Environmental Compliance Assistance Loan Fund that makes andguarantees loans to small businesses for the purchase and environmental pollution

8 U.S. EPA, (1999, p.4).9 See http://www.ecarcenter.org/, last accessed December 9, 2011.10 Region 4’s compliance assistance Web site, http://www.epa.gov/region4/ead/news/workshop.html, lastaccessed December 9, 2011.11 See http://www.epa.gov/compliance/resources/reports/endofyear/eoy2008/2008-sp-assistance.html,last accessed December 9, 2011.12 See http://dca.ky.gov/complianceassistance/Pages/default.aspx, last accessed December 9, 2011.13 See http://www.epa.gov/compliance/assistance/provider.html, last accessed December 9, 2011.14 Information on the Small Communities Environmental Assistance Programs is available athttp://www.azdeq.gov/function/compliance/smallcomm.html, last accessed December 9, 2011.



control equipment.15 At the federal level, EPA has partnered with the Small BusinessAdministration to provide environmental compliance guides by industry.16

Although the role of compliance assistance in EPA’s overall enforcement strategymay have changed over the years, EPA consistently acknowledges that “[c]omplianceassistance is most effective when used in an integrated strategy combining compli-ance monitoring (inspections), compliance incentives and auditing (self-disclosurepolicies) and enforcement.”17 The goal of compliance assistance is to help regu-lated entities understand their obligations under various environmental statutesand reduce the information costs associated with understanding environmentalregulations. However, EPA stated early in the development of its compliance as-sistance program that compliance assistance does not absolve regulated entities ofthe responsibility to learn and comply with regulations, and it cannot be successfulwithout an accompanying threat of formal enforcement.18

FRAMEWORK FOR THE ANALYSIS AND RELATED LITERATURE

Most of the theoretical economic literature on compliance with environmental regu-lations assumes that regulated entities make compliance decisions by comparing thecost of compliance relative to the expected cost of noncompliance. In these models,which are based on Becker’s (1968) paper on the economics of crime, the decision tocomply or not is a rational one based on the objective of profit maximization. For ex-ample, Russell, Harrington, and Vaughan (1986) provide one of the first comprehen-sive applications of Becker’s model to environmental regulation. Over the past twodecades, a wide range of rational polluter models has been developed that allows forvarious complexities such as imperfect information, self-reporting, principal–agentrelationships, and dynamic settings. The optimal enforcement regime in these mod-els may include a combination of civil and criminal penalties, targeted enforcement,or escalating penalties for repeated noncompliance, but ultimately the enforcementregimes are designed to deter violations by increasing the expected cost of noncom-pliance as efficiently as possible.19

While rational polluter models focus on the relative cost of compliance as thekey factor underlying a regulated entity’s compliance decisions, other authors haverecognized that there can be alternative motivations for compliance or reasons fornoncompliance. For example, a regulated entity may want to comply with regu-lations out of an inherent sense of duty to obey rules, because of peer or othersocial pressure, or to avoiding damage to the entity’s reputation or social license.20

However, if a regulated entity’s managers do not understand the regulatory require-ments, do not have complete information about their facility’s operations, have poorinternal environmental management systems, or simply do not have the ability tocomply, good intentioned regulated entities may still violate environmental regu-lations.21 In such cases, deterrence-based measures would be generally ineffectiveat increasing compliance, while programs that help regulated entities and theiremployees understand the regulations and how to comply should be effective.

15 See http://www.deq.state.va.us/osba/finance.html, last accessed December 9, 2011.16 See http://www.sba.gov/category/navigation-structure/starting-managing-business/managing-business/business-guides-industry, last accessed December 9, 2011.17 EPA compliance assistance Web site, http://www.epa.gov/compliance/assistance/index.html, last ac-cessed December 9, 2011.18 U.S. EPA (1998, p. 5).19 See Cohen (1999) and Heyes (2000) for surveys of this literature.20 See, for example, May (2005a).21 See, for example, Spence (2001) and Environmental Law Institute (2003).



The primary question this paper seeks to answer is whether compliance assistanceis effective at increasing compliance in practice. While one could build a rationalpolluter model where compliance assistance increases overall compliance by low-ering the cost of compliance relative to noncompliance, a finding that complianceassistance has a significant effect on compliance also provides support for thosemodels where reasons other than cost minimization drive noncompliance. In addi-tion, although some theoretical models focus on a particular motive underlying thecompliance decision, as Etienne (2011) discusses, in practice the compliance deci-sion is likely to depend on a number of different objectives and factors and will differacross facilities. For example, some regulated entities may want to both minimizecompliance costs and generally comply with the law. Other entities may care onlyabout complying with the law, but may have differing abilities to comply with thelaw based on their size and the technical capability of their staff. Thus, examiningthe effectiveness of compliance assistance compared to deterrence-based tools inincreasing compliance across a wide range of entities should provide some insightinto the relative importance of different motivations and factors in the compliancedecision for different types of entities.

In addition, as noted in the preceding section, EPA’s formal statement aboutthe role of compliance assistance in its overall enforcement strategy seems to rec-ognize a complementary relationship between traditional enforcement and othermore innovative tools such as compliance assistance and compliance incentives.Such a complementary relationship would be consistent with a number of theoret-ical models of compliance. For example in Scholz’s (1984) game-theoretic model,the optimal solution to noncompliance is to offer both cooperative compliance in-centives as carrots and also have a deterrence-based enforcement regime that worksas a stick when necessary. Similarly, the responsive regulation model of Ayers andBrathwaite (1992) calls for a mix of ex ante actions such as compliance assistanceand incentives and ex post actions such as compliance inspections and penalties.Thus, the second question that this paper seeks to answer is whether compliance as-sistance and traditional enforcement do in fact have a complementary relationshipor whether the effectiveness of compliance assistance is independent of the level oftraditional enforcement a regulated entity faces.

In literature, there are a number of empirical studies that have examined theeffectiveness of deterrence-based tools on environmental compliance in a wide va-riety of situations and for a wide range of regulated entities. A recent survey of thisliterature by Gray and Shimshack (2011) concludes that taken together these stud-ies demonstrate that traditional monitoring and enforcement significantly enhanceenvironmental compliance and reduce pollution. In contrast, to date there has beenalmost no formal analysis of the effectiveness of compliance assistance programsin increasing environmental compliance, although qualitative analyses and anecdo-tal evidence suggests they can be effective. To my knowledge, the only empiricalanalysis that has investigated the effect of compliance assistance on environmentalcompliance behavior is Stafford (2006). Stafford analyzes compliance with haz-ardous waste regulations using a wide variety of explanatory variables to measurethe level of complexity and costs associated with facility compliance and the pres-ence of various compliance programs including state compliance assistance. Shefinds that one form of compliance assistance provided by states, one-stop permit-ting assistance, does have a positive effect on the likelihood of compliance. However,the analysis only uses rough measures of state compliance assistance programs andthus cannot be used to assess the effectiveness of compliance assistance programsmore generally. As discussed in more detail in the following section, this paperlooks at regional compliance assistance programs rather than state programs andprovides a more thorough examination of the role of compliance assistance relative



to deterrence-based tools. Thus, it should provide additional valuable insight intothe role of compliance assistance in EPA’s overall enforcement strategy.

This paper will also add to the small but growing empirical literature that explic-itly addresses the different motivations behind compliance and attempts to betterunderstand the relative effectiveness of different approaches in a variety of settings.For example, Burby and Patterson (1993) examine enforcement in North Carolina’ssedimentation control program and find that in addition to traditional deterrencemechanisms, the clarity of regulations also has a significant positive effect on com-pliance, particularly in cases where the regulatory standard is performance-based.Brehm and Hamilton (1996) investigate compliance with the Toxics Release Inven-tory (TRI) reporting requirements by facilities in Minnesota and find that noncom-pliance is better explained by variables associated with the likelihood that a firm isignorant of those reporting requirements than by variables associated with deliber-ate evasion. May (2005b) examines the behavior of boatyards who are generally sub-ject to regulation under the Clean Water Act and marina operators who are largelyunregulated but are encouraged to comply with voluntary standards. May finds thatdeterrence and a sense of duty to comply motivate both types of facilities to com-ply, although deterrence is a stronger motivation for facilities subject to mandatoryregulation compared to facilities covered by voluntary standards. Finally, as notedabove, Stafford (2006) examines compliance with hazardous waste regulations us-ing a wide variety of explanatory variables to measure the level of complexity andcosts associated with facility compliance as well as deterrence-based enforcementand state compliance assistance and incentive programs. Her results suggest thatboth the deterrence and complexity models have some explanatory power and thatignoring either type of noncompliance would create a problem.

EMPIRICAL APPROACH AND DATA USED IN THE ANALYSIS

To determine the effectiveness of compliance assistance econometrically, there mustbe variation in the level of compliance assistance provided to regulated entities.Much of the federal compliance assistance is developed and delivered by EPA’s 10regional offices. These offices also have significant discretion over the form and levelof compliance assistance offered to regulated entities. Thus, there is regional vari-ation in the level of compliance assistance. In addition, states may offer additionalcompliance assistance above what is offered by the federal EPA. Finally, compli-ance assistance efforts are often targeted to particular types of regulated entities. Infact, according to EPA, the majority of compliance assistance efforts are directedat small- and medium-sized businesses and a number of industry sectors receivemore focused compliance assistance (Metzenbaum, 2007). Thus, using a nationalsample of a wide range of entities should provide a reasonable variation in the levelof compliance assistance at entities in the analysis.

To test for a complementary relationship between compliance assistance and de-terrence, there also needs to be variation in the level of deterrence at regulatedentities. The level of deterrence depends on both the likelihood that an entity issubject to a compliance inspection or evaluation and the severity of sanctionsassociated with violations. Deterrence is often broken out into two different cat-egories: general deterrence and specific deterrence. General deterrence refers toany general measure of enforcement effort, that is, enforcement affecting the en-vironmental performance of all regulated entities, not just the particular entity atwhich the action takes place. In the United States, the level of general deterrencecan vary across states because states implement most environmental programs andhave a significant level of discretion over how to enforce those regulations. In thisstudy, I use state-level measures of both inspections and penalty levels to capture



general deterrence. Both of these measures do vary significantly across states. Spe-cific deterrence—that is, the level of deterrence a particular entity expects to face orhas faced in the past—varies across entities within a state because many regulatoryagencies appear to engage in targeting—that is, focusing enforcement resources onthose entities most likely to violate or those which have the potential to cause sig-nificant environmental harm. In this study, I use the level of inspections an entityhas received in the past to measure specific deterrence.

To take advantage of the differences in both enforcement efforts and complianceassistance efforts across states and regions, this paper examines a national universeof regulated entities. However, because EPA regulates media programs separatelyand each program employs its own enforcement regime, it would be difficult todevelop a reasonable measure of enforcement or deterrence for a given entity if Iconsidered all regulated entities in the United States. In addition, each programcollects different information about the entities that it regulates and it would bequite challenging to collect a consistent set of explanatory variables for all regulatedentities. Thus, I chose to focus on entities subject to hazardous waste regulations. Ichose the hazardous waste program over other media programs because it is a largefederal program that covers a wide variety of facilities and thus there is substantialvariation across facility types in both enforcement effort and compliance assistance.In addition, the hazardous waste program is enforced primarily through complianceinspections so that it is easy to measure deterrence levels in this program. Finally,this sector has been the focus of a number of other compliance analyses, and I caneasily compare the results to those of other studies.

The universe for the analysis includes over 350,000 regulated hazardous gener-ators in the continental United States that were identified using EPA’s RCRAInfodatabase.22 Hazardous waste generators can be grouped into three basic categories:Large Quantity Generators (LQGs), Small Quantity Generators (SQGs), and Con-ditionally Exempt Generators (CEGs). LQGs generate over 1,000 kg of hazardouswaste a month, while SQGs generate between 100 and 1,000 kg a month and CEGsgenerate less than 100 kg a month. LQGs are subject to more stringent regulationsthan SQGs, and CEGs are subject to less-stringent regulations than SQGs. In addi-tion, LQGs generally face higher levels of enforcement than SQGs and CEGs, as willbe shown in more detail below.

One might also expect the effect of compliance assistance to vary across generatortypes, as most compliance assistance efforts are directed more toward small entitiesthan toward large ones. Since smaller entities are less likely to find it cost effectiveto employee full-time compliance personnel than large entities and are less likely tohave undergone environmental audits (Evans, Liu, & Stafford, 2011), complianceassistance may make more of a difference at smaller entities. While small entitiesare not necessarily SQGs or CEGs since the small entity designation is based onnumber of employees and the generator categorization is based on the quantityof hazardous waste generated, most small businesses do fall into the SQG andCEG categories.23 Because compliance assistance may have a varying affect acrossfacility type, I decided to conduct separate analyses for each of the generator groupsin addition to analyzing all facilities together.

22 I do not include facilities that are subject to hazardous waste regulations, but do not generate haz-ardous waste such as hazardous waste transporters.23 EPA does not provide any formal estimates of the distribution of small businesses across the variousgenerator categories. However, a survey on Waste and Hazardous Materials (conducted by the NationalFederation of Independent Business Research in 2007) found that more than half of the surveyed smallbusinesses were conditionally exempt generators, 38 percent were SQGs, and just 2 percent were LQGs(Dennis, 2007).



Table 1. Regional compliance assistance measures per 100,000 regulated entities.

Entities Entities ToolsRegion States reached visited developed

1 Connecticut, Maine, Massachusetts,New Hampshire, Rhode Island,Vermont

33,189 26 1

2 New Jersey, New York 131,254 10 243 Delaware, Maryland, Pennsylvania,

Virginia, West Virginia, Districtof Columbia

6,270 25 0

4 Alabama, Florida, Georgia,Kentucky, Mississippi, NorthCarolina, South Carolina,Tennessee

6,615 31 7

5 Illinois, Indiana, Michigan,Minnesota, Ohio, Wisconsin

25,920 6 4

6 Arkansas, Louisiana, New Mexico,Oklahoma, Texas

2,355 10 0

7 Iowa, Kansas, Missouri, Nebraska 12,867 183 48 Colorado, Montana, North Dakota,

South Dakota, Utah, Wyoming6,718 1,466 1

9 Arizona, California, Hawaii, Nevada 6,388 256 110 Alaska, Idaho, Oregon, Washington 76,571 37 6

Compliance assistance may immediately affect a facility’s compliance with envi-ronmental regulation. However, it may also take time for a facility to absorb theassistance provided and use it to increase environmental performance. Thus forthis analysis, I use lagged compliance assistance efforts. In addition, I use laggeddeterrence measures rather than contemporaneous deterrence measures because fa-cilities are likely to base their estimation of current deterrence on recently observedlevels of deterrence. The use of lagged measures also minimizes any concerns aboutthe potential endogeneity of such measures.

EPA began collecting data on the type and level of compliance assistance providedby its regional offices in 2005. According to EPA staff, the first year of its compli-ance assistance data collection suffered from some inconsistencies across regions inreporting data, so I have chosen to use the data from 2006 to conduct the analysis.24

Table 1 presents three different compliance assistance measures provided by EPA:entities reached by compliance assistance, compliance assistance visits, and com-pliance assistance tools developed, all normalized by the total number of regulatedentities in the region. Note that there is reasonable variation in all three variablesacross the regions. The number of entities assisted includes entities that receiveon-site visits, entities that participate in workshops, as well as entities that receivecompliance assistance materials. Compliance assistance tools include complianceworkshops, web-based assistance and training programs, guidance documents, etc.Since the activities measured by these two variables could be very different acrossregions, these two variables are likely to be more noisy than the number of compli-ance visits.

Given the need to lag the compliance assistance data, I use 2007 compliance dataas the dependent variable in the analysis. Violations of RCRA regulations can range

24 K. Koslow, Acting Director of the Compliance Assistance and Sector Program Division of EPA’s Officeof Compliance (phone conversation, September 4, 2009).



from minor paperwork violations to major violations that pose an immediate threatto human health and the environment. However, the RCRAInfo database containsvery limited information on the nature of violations found at facilities, making itvery difficult to develop a continuous measure of the level of noncompliance at agiven facility. Thus, I model compliance as a binary variable: Facilities with anydetected violations are considered to be in noncompliance and facilities with nodetected violations are considered to be in compliance.

Because the dependent variable is binary, I use a probit model to analyze thefacilities’ compliance behavior. However, data on compliance are only availableif a facility is inspected, and thus any valid empirical method must control forthis censoring. In addition, it is likely that compliance and inspections are jointlydetermined; that is, a regulator’s decision to inspect a particular facility dependsin part on the likelihood that the facility will be noncompliant and the facility’sdecision to comply depends in part on the likelihood of inspection. To control forboth the censoring and the probability that the compliance and inspection equationmay have correlated errors, I use a censored bivariate probit model.

As discussed in more detail in Greene (1992), the censored bivariate probit usesmaximum likelihood to estimate a probit model with sample selection where theselection equation and the underlying equation of interest may have correlatederrors. More specifically, both the probability of a violation and the probability ofan inspection are modeled as latent variables, V∗

i = x1iβ1 + ε1i and I∗i = x2iβ2 + ε2i ,

respectively. Let Vi and Ii be the observable binary variables associated with thesetwo latent variables. Since regulators target facilities that are likely to be in violationfor inspection, the error terms ε1i and ε2i should be positively correlated. Therefore,the likelihood of observing a detected violation (Vi = Ii = 1) can be written as

LVi=1,Ii=1 =∑

Vi=1,Ii=1

log �2[x1, β1, x2, β2, ρ] (1)

where �2 is the bivariate normal cumulative distribution function and ρ is thecovariance between ε1i and ε2i. Similarly, the likelihood of inspecting a nonviolator(Vi = 0, Ii = 1) is

LVi=0,Ii=1 =∑

Vi=0,Ii=1

log �2[−x1, β1, x2, β2,−ρ]. (2)

Finally, if a facility is not inspected, whether the facility is in violation is unknown.Thus, the maximum likelihood function can be expressed as

L = LVi=0,Ii=1 + LVi=1,Ii=1 + LIi=0 =∑

Vi=1,Ii=1

log �2[x1β1, x2, β2, ρ]

∑

Vi=0,Ii=1

log �2[−x1β1, x2, β2,−ρ]

∑

Ii=1

log(1 − �[x2β2])

. (3)

To identify the model, there must be at least one explanatory variable that affectswhether the facility will be inspected, but does not affect whether the facility violatesthe regulations. I address the identification strategy later in this section.

Table 2 lists the variables used in the analysis and presents summary statisticsfor each of the three categories of generators: LQGs, SQGs, and CEGs. All of the



Table 2. Summary statistics for the variables in analysis, by generator type: Variable mean(standard deviation).

All generators LQGs SQGs CEGs

Number of observations 350,572 10,342 172,069 168,161Dependent variablesInspected07 0.038 0.383 0.036 0.020

(0.192) (0.486) (0.186) (0.139)Violated07 0.014 0.183 0.012 0.005

(0.117) (0.387) (0.110) (0.069)Deterrence measuresState Inspections06 0.022 0.025 0.022 0.023

(0.018) (0.020) (0.017) (0.020)Change in State Inspections06–07 0.031 0.033 0.0011 0.061

(0.553) (0.546) (0.533) (0.572)State Average Penalty06 3.604 3.512 3.853 3.354

(5.615) (5.474) (5.902) (5.280)Facility Inspections06 0.051 0.672 0.041 0.023

(0.393) (1.776) (0.262) (0.183)Facility Inspection History02–06 0.242 3.081 0.172 0.138

(1.398) (6.582) (0.756) (0.562)Compliance assistance measuresRegional CA Visits06 0.001 0.001 0.001 0.001

(0.002) (0.002) (0.002) (0.003)Historic State CA Program 0.852 0.843 0.888 0.817

(0.355) (0.364) (0.316) (0.387)Deterrence and compliance assistance interactionsRegional CA Visits06 0.007 0.006 0.006 0.0073

× State Average Penalty06 (0.064) (0.059) (0.065) (0.068)State CA Program 3.256 2.926 3.195 3.583

× State Average Penalty06 (5.683) (5.331) (5.552) (5.996)Facility characteristicsWaste Generated07 −0.874 3.9620 0.122 −2.191

(1.488) (2.066) (0.294) (0.281)Waste Managed07 −18.324 −15.344 −18.411 −18.418

(1.514) (8.001) (0.458) (0.257)Transporter 0.016 0.045 0.012 0.018

(0.124) (0.208) (0.107) (0.132)Multimedia 0.202 0.864 0.203 0.161

(0.401) (0.342) (0.402) (0.367)Facility Violations06 0.104 1.683 0.083 0.028

(2.264) (9.750) (1.858) (1.070)Facility Violation History02–06 0.485 7.142 0.382 0.181

(6.185) (24.872) (5.364) (3.071)State characteristicsState Violations06 0.043 0.048 0.044 0.042

(0.040) (0.041) (0.042) (0.036)State Environmental Group 1.126 1.138 1.075 1.178Revenues06 (2.386) (2.496) (1.630) (2.960)State TRI Releases06 0.394 0.43 0 0.277 0.513

(0.981) (0.824) (0.609) (1.248)State Government Ideology06 0.510 0.493 0.500 0.523

(0.221) (0.242) (0.220) (0.220)State Household Income07 51.421 50.407 52.912 49.958

(6.950) (7.349) (6.458) (7.087)State Percent College 34.861 33.950 35.441 34.324Degree05 (4.687) (4.817) (3.743) (5.418)



facility-level variables are extracted from EPA’s RCRAInfo database that includesdata on each facility’s location, regulatory status, compliance history, enforcementhistory, and whether the facility is regulated by another media program. Note thatthere is significant variation in the means of the facility-level variables across thethree groups.

The two dependent variables are Inspected07 and Violated07. Inspected07 is equal to1 if any state or regional official inspects the facility in 2007, regardless of the pur-pose of the inspection. As shown, LQGs face a 30 percent chance of being inspectedin a given year while SQGs and CEGs face a much smaller likelihood. Violated07 isequal to 1 if there is any detected hazardous waste violation at the facility in 2007,regardless of the type or severity of the violation. Of course, violations cannot be de-tected if the facility is not inspected, so the higher percentage of LQGs that violatedis due, at least in part, to the higher probability of being detected. The empiricalmethod described above will allow me to determine whether LQGs are more likelyto violate even after controlling for the different probability of being inspected.

The first set of explanatory variables captures the level of deterrence at eachfacility in the analysis. State Inspections06 measures the total number of RCRAinspections conducted in the state in 2006 normalized by the total number of RCRAregulated facilities (not just generators) and is one proxy for the level of generaldeterrence each facility faces. While generators are unlikely to know the valuesof this variable explicitly, recent enforcement levels are likely to provide the bestestimate a facility has of current enforcement efforts and thus this variable is likelyto be highly correlated with the level of deterrence a facility experiences. Change inState Inspections06-07 measures the change in the total number of RCRA inspectionsconducted in the state between 2006 and 2007. To identify the model, I do excludeChange in State Inspections06-07 from equation 1. From a theoretical standpoint, Ibelieve this exclusion is valid as an increase in the total number of inspections in astate will increase the probability regulators will inspect a facility. However, sucha change is unlikely to be common knowledge, and thus it should not affect thefacility’s violation decision.25 State Average Penalty06 measures the average penaltyper violation in the state in 2006 in thousand dollars and is a second measure ofgeneral deterrence at facilities in the analysis. A facility’s past inspection historyaffects the level of specific deterrence it faces. Two variables measure facility-levelinspections, Facility Inspections06, which indicates the number of inspections at thefacility in 2006, and Facility Inspection History02-06, which is a count of the numberof inspections between 2002 and 2006. Both of these variables do vary significantlyacross generator types.

In the analysis, federal compliance assistance is measured by the variable RegionalCA Visits06, which counts the number of entities that receive a compliance assistancevisit in the region, normalized by the total number of regulated entities in the re-gion. As shown in Table 1, the number of entities visited in the sample ranges froma high of 1,465 per 100,000 entities in Region 8 to a low of 6 per 100,000 entities inRegion 5.26 While this is the only measure of regional compliance assistance usedin the final regression reported in this paper, I did consider the two alternate mea-sures of regional compliance assistance presented in Table 1 as well. However, asdiscussed above, these other two variables (Entities Reached and Tools Developed)are likely to be more noisy than the number of compliance visits. In addition, when

25 From an econometric standpoint, when this variable is included in Equation 1, the coefficient onChange in State Inspections06-07 is not significant.26 The high number of visits in Region 8 is due to over 1,500 visits to auto service centers. However, theresults are robust even if these visits are excluded from the Region 8 count.



included in the analysis, neither of the coefficients for these two variables was sta-tistically significant.

The analysis also includes a variable, Historic State CA Program, that measureswhether the state has historically had a compliance assistance program. Ideally, Iwould like a variable that measures whether the state had a compliance assistanceprogram in 2006, but unfortunately data on compliance assistance programs during2006 are not available.27 Thus, I used historic data on state compliance assistanceprograms from a survey conducted by the Environmental Council of the States(ECOS) in 2000.28 I also considered a series of dummies capturing the differentaspects of states’ historic compliance assistance programs, such as whether the stateenvironmental agency had a separate compliance assistance division, provided on-site consultations to assist facilities with compliance, or had a compliance assistancehotline. However, due to the binary nature of these variables and high correlationsacross them, including them in the analysis caused conversion problems in theregression. Note that both the regional and state compliance assistance variablesare also interacted with the general deterrence proxy, State Average Penalty06.

In addition to the deterrence and compliance assistance measures, I include anumber of facility- and state-level characteristics in the analysis. Waste Generated07is the natural log of the tons of waste generated by the facility in 2007, and WasteManaged07 measures the log of the tons of waste managed at the facility in 2007.29

Transporter indicates whether the facility is also licensed to transport hazardouswaste, and Multimedia indicates whether the facility is regulated under an environ-mental program other than the hazardous waste program. Both of these variablesmeasure the level of regulatory complexity a facility would face, and while there isnot a lot of variation across types with respect to the Transporter variable, LQGs aresignificantly more likely to be regulated under other media programs than SQGs orCEGs. Facility Violations06 indicates the number of violations detected at the facilityin 2006, while Facility Violation History02-06 is a count of the number of detectedviolations at the facility between 2002 and 2006. Here too, there are significantdifferences across facility types.

State Violations06 measures the total number of RCRA violations detected in thestate in 2006 normalized by the total number of RCRA regulated facilities in thestate. As with most of the state-level variables, there are some minor differences inmean and standard deviations across the generator types, but these differences aremuch less significant than the differences in facility-level variables. State Environ-mental Group Revenues06 measures the total revenues (in thousands) collected bynonprofit environmental groups in the state in 2006 normalized by the number ofregulated entities, and is included to control for nongovernmental deterrence thatcould affect a facility’s compliance decision.30 State TRI Releases06 measures the

27 It is possible to determine whether a state currently has a compliance assistance program, but giventhat compliance assistance programs can be offered by many different divisions of a state environmentalagency, data on when existing compliance programs were created and whether states that do not currentlyhave programs ever did are not readily available.28 Fifteen states did not respond to the ECOS survey used to gather these data (see Shewmake, 2004).For this analysis, I assumed that nonrespondents did not have any compliance assistance programs.29 Because quantity of waste generated and managed both have a skewed distribution (e.g., the meanquantity generated is over 2,400 tons, although about 90 percent of facilities generate less than 500 tons),the log of tons generated performs significantly better than tons. Any facility listed in RCRAInfo as anLQG that did not generate hazardous waste in 2007 was dropped from the analysis as all LQGs arerequired to report their waste-generation quantities. However, depending on state requirements, SQGsand CEGs may not have to report waste quantities. For SQGs and CEGs that did not report, I assumedgeneration equal to the maximum quantity for that type.30 I collected the data on environmental group revenues from the National Center for Charitable Statis-tics’ Guidestar Database.



total releases of toxic chemicals released in 2006 by entities subject to EPA’s TRIreporting requirements normalized by the state’s gross domestic product in 2006,and is included to control for differences in the pollution intensity of a state’s eco-nomic base.31 State Government Ideology06 is a measure of the ideology of the electedofficials in each state in 2006 using a 0 to 100 scale with higher scores indicating amore liberal ideology. The construction of this variable is described in Berry et al.(1998).32 To capture other potentially relevant state demographic characteristics,I also include State Household Income07, which measures the median householdincome in each state in 2007, and State Percent College Degree05, which measures thepercent of individuals 25 or older who have earned at least an associate degree.33

RESULTS

Table 3 presents the results of the censored bivariate probit regressions. Because theexplanatory variables include a number of state-level variables, I cluster the standarderrors by state. As discussed in the section above, I run four separate models: onethat combines all types of generators together with dummy variables for generatortypes, and one for each type of generator separately.

Focus first on the results for the deterrence variables in the violation equation.General deterrence is measured by two variables, State Inspections06 and State Av-erage Penalty06. The coefficient on State Inspections06 is negative and significant inthree of the regressions, while the coefficient on State Average Penalty06 is negativeand significant in two. Thus, general deterrence results in a lower probability of aviolation for all generators overall, as well as for CEGs and SQGs. However, LQGsare not significantly affected by these general deterrence measures. This is consis-tent with Gunningham, Thorton, and Kagan’s (2005) finding that deterrence is moreimportant to small- and medium-sized companies than it is to large corporations.Specific deterrence is captured by both Facility Inspections06 and Facility Inspec-tion History02-06. Facility Inspections06 does have the expected negative coefficient inthree of the four regressions and is significant in two, while 5 Year Inspection His-tory has a positive and significant coefficient in two of the regressions. The positivecoefficient may be due to the fact that some facilities are more likely to violate dueto their underlying characteristics and that regulators target those facilities. Thefact that neither of these variables is significant for conditionally exempt generatorsis likely to stem from the fact that CEG facilities have an extremely low probabilityof inspection—less than 2 percent of CEGs are inspected in a given year.

Next consider the regional compliance assistance variable, Regional CA Visits06.It has a negative and significant coefficient in two of the regressions—the overallregression and the SQG regression—providing evidence that regional complianceassistance efforts do decrease noncompliance. While the coefficient on RegionalCA Visits06 is neither negative nor significant in the CEG regression, note that thecoefficient on the Regional CA Visits06 × State Average Penalty06 is negative and

31 The TRI reporting program requires over 23,000 industrial and other facilities that use toxicchemicals to provide information on the release of over 600 toxic chemicals into the environ-ment. The TRI is arguably EPA’s most comprehensive source of data on pollution as it in-cludes releases to air, water, and land. Data on 2006 TRI releases by state are available athttp://www.epa.gov/tri/tridata/tri06/pdr/SectionB.pdf. Data on 2006 GDP by state were obtained fromthe Bureau of Economic Analysis.32 The 2006 measure of government ideology is available at http://www.uky.edu/∼rford/stateideology.html.33 The household income data were taken from the American Community Survey 2005 to 2007 three-year estimates, while the educational attainment data came from the 2005 American Community Survey.Both are available from the U.S. Census Bureau.



Table 3. Results of the censored bivariate probit, by generator type: Coefficient (standarderror).


Violation equationState Inspections06 −7.358** −4.685 −5.919** −13.048**

(2.603) (3.022) (2.946) (5.301)State Average Penalty06 −0.089** −0.100 −0.102** −0.034

(0.034) (0.061) (0.041) (0.043)Facility Inspections06 −0.040 −0.039* −0.16** 0.038

(0.025) (0.024) (0.051) (0.071)Facility Inspection History02–06 0.013* 0.015** 0.023 0.034

(0.007) (0.007) (0.042) (0.047)Regional CA Visits06 −40.712** −10.945 −61.507** 51.276

(16.857) (19.553) (15.731) (46.959)Historic State CA Program −0.109 0.003 −0.232 −0.161

(0.113) (0.147) (0.190) (0.227)Regional CA Visits06 1.654* 2.03* 1.488 −53.098**

× State Average Penalty06 (0.871) (1.097) (0.911) (24.618)State CA Program 0.046 0.064 0.055 0.022

× State Average Penalty06 (0.035) (0.061) (0.038) (0.047)LQG 0.398**

(0.132)SQG 0.155**

(0.071)Waste Generated07 0.068** 0.036** 0.231** 0.098**

(0.010) (0.01) (0.049) (0.047)Waste Managed07 −0.005 0.0001 0.008 0.006

(0.003) (0.005) (0.013) (0.021)Transporter 0.020 −0.005 0.0004 0.030

(0.075) (0.074) (0.134) (0.206)Multimedia 0.201** 0.071 0.244** 0.128

(0.064) (0.054) (0.103) (0.108)Facility Violations06 0.004 0.003 0.006 0.012

(0.004) (0.003) (0.010) (0.009)Facility Violation History02–06 0.006** 0.005** 0.007* 0.007*

(0.002) (0.002) (0.004) (0.004)State Violations06 4.872** 3.103** 5.435** 5.800**

(1.434) (1.172) (2.550) (2.122)State Environmental Group −0.032** −0.021** −0.039** −0.043**

Revenues06 (0.011) (0.007) (0.015) (0.016)State TRI Releases06 8.335* 5.832 10.912 4.576

(4.889) (7.414) (6.791) (4.477)State Government Ideology06 0.059 −0.027 0.124 −0.202

(0.215) (0.274) (0.273) (0.405)State Household Income07 −0.042** −0.021 −0.062** −0.025*

(0.011) (0.015) (0.013) (0.014)State Percent College 0.06** 0.035 0.083** 0.036Degree05 (0.02) (0.023) (0.026) (0.027)Constant −1.039** −0.741 −0.424 −0.675

(0.477) (0.452) (0.569) (0.877)

significant. Taken together, these results suggest that regional compliance assis-tance does have a negative effect on violations in the SQG and CEG regressions,but does not in the LQG regression—a finding that is consistent with the theory dis-cussed in earlier sections that compliance assistance may be more needed at smallerentities.



Table 3. Continued


Inspection equationState Inspections06 8.357** 11.526** 7.986** 7.68**

(2.699) (2.166) (3.222) (3.088)Change in State Inspections06–07 2312.179** 2095.429** 2679.81** 1569.617*

(1061.382) (759.678) (1320.557) (814.884)State Average Penalty06 −0.020 −0.059* −0.048 0.034

(0.030) (0.035) (0.039) (0.027)Facility Inspections06 0.017 −0.09** −0.029 0.284**

(0.042) (0.037) (0.084) (0.065)Facility Inspection History02–06 0.191** 0.135** 0.225** 0.247**

(0.016) (0.013) (0.025) (0.034)Regional CA Visits06 −17.174 5.449 −32.997 2.114

(26.422) (12.839) (40.782) (17.84)Historic State CA Program −0.299** −0.387** −0.431** −0.072

(0.128) (0.128) (0.170) (0.148)Regional CA Visits06 0.622 −1.600** 1.013 1.343

× State Average Penalty06 (0.945) (0.479) (1.270) (0.973)State CA Program 0.008 0.067** 0.034 −0.059**

× State Average Penalty06 (0.029) (0.033) (0.037) (0.028)LQG 0.754**

(0.078)SQG 0.184**

(0.07)Waste Generated07 0.074** 0.025** 0.228** 0.158**

(0.008) (0.007) (0.019) (0.015)Waste Managed07 0.002 0.013** 0.002 0.030*

(0.003) (0.003) (0.009) (0.015)Transporter 0.196** 0.177* 0.192** 0.227**

(0.069) (0.101) (0.09) (0.113)Multimedia 0.311** 0.092* 0.292** 0.279**

(0.041) (0.048) (0.045) (0.055)Facility Violations06 0.016** 0.015** 0.022** 0.005

(0.003) (0.003) (0.005) (0.008)Facility Violation History02–06 −0.003** −0.004** −0.004* −0.001

(0.001) (0.001) (0.002) (0.003)State Violations06 −2.142* −0.613 −3.184** −1.566

(1.218) (0.726) (1.535) (1.045)State Environmental Group −0.018 −0.005 −0.025 −0.009Revenues06 (0.012) (0.008) (0.019) (0.009)State TRI Releases06 4.554 2.579 8.131 −1.772

(3.967) (2.969) (5.134) (3.079)State Government Ideology06 −0.133 0.001 −0.142 −0.143

(0.238) (0.184) (0.311) (0.253)State Household Income07 −0.022 0.004 −0.039 0.006

(0.019) (0.009) (0.027) (0.013)State Percent College 0.012 −0.008 0.025 −0.022Degree05 (0.025) (0.014) (0.032) (0.02)Constant −1.141** −0.485 −0.387 −0.858*

(0.454) (0.296) (0.728) (0.506)

**Significant at the 5% level; *significant at the 10% level.



Figure 1. Predicted Probability of Violation for a Representative Facility UnderVarying Levels of State Penalties and Regional Compliance Assistance.

Interpreting the relationship between regional compliance assistance and generaldeterrence (as proxied by State Average Penalty06) can be a little difficult to do fromTable 3 directly. While the coefficient on the Regional CA Visits06 × State AveragePenalty06 interaction is positive and significant in the regression for all generatorsand LQGs, it is negative and significant in the CEG regression. To help the readerbetter understand how regional compliance assistance and state penalties togetheraffect facility violations, Figure 1 presents the predicted probability of violation for arepresentative facility under varying levels of state penalties and regional complianceassistance.34

In all four regressions as the level of penalties increases, the probability of aviolation decreases for each underlying level of regional compliance assistance. Thatis, within each grouping of bars on each of the four charts, the predicted probabilityof a violation decreases. However, the effect of regional compliance assistance giventhe underlying level of deterrence is not consistent. First focus on the All Generatorsand SQGs charts on the left side of the figure. For each level of state penalties (i.e.,similarly shaded bars), increasing the level of regional compliance assistance doesdecrease the probability of violation. In addition, in the SQG regression, it is clearthat the effect of increasing the level of regional compliance assistance depends onthe underlying level of state penalties. Note that the decrease in violation is larger

34 The representative facility has the mean values for all continuous explanatory variables and the medianvalues for discrete explanatory variables. The representative facility varies across the four regressions, asdo the coefficients on each variable.



at the 25th percentile of state penalties (the most lightly shaded bars) than it is forthe 75th percentile of state penalties (the most darkly shaded bars). This is due tothe positive coefficient on the interaction between regional compliance assistanceand state penalties and shows that the benefit of increasing regional complianceassistance decreases as the level of penalties increases. This pattern is present in theAll Generators regression as well, although the effect is not as pronounced and thusnot as visually apparent. In the LQGs chart, the underlying probabilities conform tothe same pattern although there is very little perceptible change in the probability ofviolation as regional compliance assistance increases. That is, for LQGs as well thereare small decreases in the likelihood of a violation as regional compliance assistanceincreases, but the effect is smaller at higher levels of state penalties. This impliesthat compliance assistance and general deterrence do not enhance each other, buttend to crowd each other out.

In contrast, note that in the Conditionally Exempt Generators chart in the lowerright-hand corner of the figure, the predicted probability of a violation increases asregional compliance assistance increases for the 25th percentile of state penalties,but decreases for the 50th and 75th percentiles of state penalties. This pattern is dueto the positive (but not significant) coefficient on the regional compliance variableitself combined with the negative (and significant) coefficient on the interactionbetween regional compliance assistance and state penalties. Thus for the represen-tative CEG, a moderately high level of penalties is necessary for regional complianceassistance to be effective at decreasing the likelihood of a violation. Given the dif-ference in inspection rates at CEGs compared to other facilities, it could be the casethat the complementary relationship suggested by the results of this regression onlyapply at very low levels of inspections and once inspections reach higher levels, thisrelationship no longer holds.

Moving on to the results for the Historic State CA Program variable, note thatin contrast to the results in Stafford (2006), the existence of some sort of a statecompliance assistance program does not have a significant effect on compliance:While the coefficients on Historic State CA Program are negative in three of the fourregressions, they are never significant. This result does not necessarily imply thatall state compliance assistance programs are ineffective, but instead could be dueto the fact that there is significant variation across state programs. Also note thatthe coefficient on the interaction between Historic State CA Program × State AveragePenalty06 is also never significant.

The remaining variables in the violation equation are generally consistent acrossthe four regressions and furthermore are generally consistent with expectations.LQG and SQGs are both more likely to violate than CEGs, which is consistent withLQGs and SQGs being subject to more stringent regulations than CEGs. Withineach category of generators, facilities that generate more waste have a higher prob-ability of violation suggesting that, holding the stringency of regulations constant,larger facilities are also more likely to violate. In two of the regressions, facilitiesthat are subject to regulation for other environmental media programs (e.g., theClean Air and Clean Water Act regulation) are also more likely to violate, suggestingthat the complexity of a facility’s environmental management requirements makecompliance more difficult. Facilities with poorer compliance records in the past arealso more likely to violate suggesting that there may be facility-specific factors notcaptured by the variables in the regression that make some facilities more likelyto violate. This is consistent with Gunningham, Kagan, and Thorton’s (2003) find-ing that managerial attitudes or a facility’s environmental management style has asignificant effect on a facility’s overall environmental performance.

Focusing on the state-level variables, facilities in states with successful environ-mental groups are less likely to violate, suggesting that active environmental groupspositively influence compliance outcomes. This result is consistent with Scholz and



Wang’s (2006) finding that local environmental organizations have a positive effecton regulated entities’ compliance with the National Pollutant Discharge EliminationSystem (NPDES) program.35 Finally, note that in three of the four regressions, thecoefficient on State Household Income07 is negative and significant, while in twoof the four regressions the coefficient on State Percent College Degrees06 is positiveand significant. The former result is consistent with Scholz and Wang’s findingthat local per capita income is inversely related to the likelihood of a violation atNPDES-regulated facilities while the latter results contrast with their finding thatthe local percentage of college graduates is also inversely related to the likelihoodof a violation.

Turning now to the results for the inspection equation, most of the coefficientson the explanatory variables are also consistent with expectations and the findingsof previous studies on hazardous waste compliance. There are, however, a coupleof interesting new results. First, note that the coefficient on Regional CA Visits06is never significant, indicating that in general regional compliance assistance doesnot decrease the probability of an inspection. Thus, regional compliance assistancedoes not appear to be substituting for more traditional enforcement. However, thisfinding does not hold for state compliance assistance efforts. The negative andsignificant coefficient on Historic State CA Program in three of the four regressionssuggests that regulators may see state compliance assistance as a substitute forinspections; that is, if a state has a compliance assistance program, there is lessneed for traditional enforcement efforts such as inspections. Alternatively, if statecompliance programs were complementary to traditional enforcement efforts, onewould expect to see a positive relationship between state programs and inspectionprobability.

One might expect compliance assistance and inspections to be substitutes for eachother when a state’s enforcement budget is constrained. In such a case, there is aclear trade-off between having a compliance assistance program and conductingmore facility inspections. Budgetary pressures would not necessarily result in asimilar trade-off between regional compliance assistance and state inspection effortsas the regional compliance assistance efforts are funded by the federal EPA not stateagencies, which is consistent with the insignificant result on Regional CA Visits06.36

The insignificant coefficient on Historic State CA Program in the CEG inspectionequation may be due to the fact that the inspection rate at CEGs is very low to beginwith—only 2 percent of CEGs facilities are inspected on an annual basis. The factthat state compliance assistance efforts appear to be substitutes for state inspectionsat LQGs and SQGs is troubling given that this study finds no evidence that statecompliance assistance efforts have a significant effect on facility compliance.

Interestingly, other than State Violations06 none of the state control variables havesignificant coefficients in any of the four regressions. Recall that most inspectionsare conducted by state officials, so it is somewhat surprising that differences inthe likelihood of inspection are not partially explained by these factors. These re-sults provide an interesting contrast to the results from Scholz and Wang’s (2006)analysis of NPDES enforcement. Scholz and Wang find that local environmentalinstitutions, per capita income, and state government ideology all have a positiveeffect on the probability of inspection in the NPDES program. However, for thefirst two variables, Scholz and Wang use watershed-level data, not state-level data.It may be the case that differences in inspection probability do depend on localdemographic characteristics, but not on state-level characteristics.

35 However, Scholz and Wang’s analysis is conducted at the watershed level, not the state level.36 Of course, a severely resource constrained state might substitute regional compliance assistance fortraditional enforcement efforts and shift resources out of compliance and enforcement all together.



To provide a sense of the magnitude of the effects of the explanatory variables,Table 4 presents the change in the predicted probability (in percentage points) thata representative facility violates in 2007 for various changes in the explanatoryvariables.37 The representative facility varies across the four regressions, as do thecoefficients on each variable. Thus in the regression that includes all generators,the predicted baseline violation probability is 9.7 percent, while it is 32.3 percentfor LQGs, 11.7 percent for SQGs, and 4.2 percent for CEGs. To make the changesin explanatory variables roughly comparable across variables, I increased each con-tinuous variable by one standard deviation to see how such an increase changes thepredicted probability of violation. For binary variables, I examine what happens ifthe variable takes a value opposite to the value for the representative facility. Theabsolute value of the change in the explanatory variable is provided in the table inbrackets.

While this table does provide some general understanding of the size of the rel-ative effect each variable has on the violation probability, the estimates need tobe interpreted with care. For example, increasing the number of state inspectionsper RCRA-regulated facility by one standard deviation results in about a 2 percent-age point decrease in violation probability at the representative facility in the AllGenerators regression. Because the standard deviation of State Inspections06 for allfacilities is 0.02, this change requires additional two inspections of every 100 RCRA-regulated facilities in the state. All RCRA-regulated facilities in the state will alsohave decreased probability of violation, although the magnitude of each facility’s de-crease will depend on the facility’s characteristics and thus could be higher or lowerthan the representative facility. In comparison, for the entire universe of facilities,increasing the number of regional compliance assistance visits per EPA-regulatedentity by one standard deviation (two additional visits per 1,000 facilities) results ina 1.6 percentage point decrease in violation probability at the representative facility.All RCRA-regulated facilities in the region will then have decreased probability ofviolation, with the magnitude depending on each facility’s characteristics. While thechange in state inspections at first might seem to have a slightly larger per facilityeffect, such a change affects a smaller number of facilities (state-wide rather thanregion-wide effects) and may also require a larger number of inspections. Compar-ing results across regressions, note that roughly the same change in state inspectionsper RCRA-regulated facility causes a 2.1 percentage point decrease in the probabilityof violation at the representative facility in the All Generators regression, but onlya 1.9 percentage point decrease at the representative facility in the SQG regression.However, an increase in Regional CA visits of two per 1,000 EPA-regulated facilitiesresults in a 1.6 percentage point decrease in the probability of violation at the rep-resentative facility in the All Generators regression compared to a 2.3 percentagepoint decrease at the representative facility in the SQG regression.

These estimates show that the changes induced by a one standard deviation in-crease in state inspections and compliance assistance visits are of roughly the samemagnitude, and smaller than the change in the probability of violation effected bya one standard deviation increase in State Average Penalty06 for all but the CEGregression. Comparing the deterrence variables to facility-level variables, note thata one standard deviation increase in any of the deterrence variables is more thanoffset by a facility being an LQG. Finally, note that many of the other state variableshave effects on the probability of violations that are comparable to the effects of

37 Just as in Figure 1, the representative facility has the mean values for all continuous explanatoryvariables and the median values for discrete explanatory variables.



Tab

le4

.Mar

gin

alef

fect

sfo

ra

rep

rese

nta

tive

faci

lity

(ab

solu

tech

ange

inex

pla

nat

ory

vari

able

pre

sen

ted

inb

rack

ets)

.

All

LQ

GS

QG

CE

G

Pre

dic

ted

pro

bab

ilit

yof

insp

ecti

on9.

7%32

.3%

11.7

%4.

2%C

han

gein

pre

dic

ted

pro

bab

ilit

yin

per

cen

tage

poi

nts

from

chan

gein

Sta

teIn

spec

tion

s 06

−2.1

2%

[0.0

2]−3

.28%

[0.0

2]−1

.87

%[0

.02]

−1.8

5%

[0.0

2]S

tate

Ave

rage

Pen

alty

06−6

.08

%[5

.62]

−16.

58%

[5.4

7]−8

.05

%[5

.92]

−1.3

9%[5

.28]

Fac

ilit

yIn

spec

tion

s 06

−0.2

7%[0

.39]

−2.4

3%

[1.7

8]−0

.80

%[0

.26]

0.06

%[0

.18]

Fac

ilit

yIn

spec

tion

His

tory

02-0

60

.32

%[1

.40]

3.5

8%

[6.5

8]0.

34%

[0.7

6]0.

18%

[0.5

6]R

egio

nal

CA

Vis

its 0

6−1

.61

%[0

.002

]−0

.83%

[0.0

02]

−2.3

4%

[0.0

02]

1.48

%[0

.003

]H

isto

ric

Sta

teC

AP

rogr

am−1

.73%

[−1.

00]

−0.1

1%[−

1.00

]−3

.95%

[−1.

00]

−1.2

7%[−

1.00

]R

egio

nal

CA

Vis

its 0

6×

Sta

teA

vg.P

enal

ty06

1.9

3%

[0.0

6]4

.43

%[0

.06]

2.02

%[0

.07]

−4.2

4%

[0.0

7]S

tate

CA

Pro

gram

×S

tate

Ave

rage

Pen

alty

065.

25%

[5.6

3]12

.90%

[5.3

3]7.

09%

[5.5

5]1.

35%

[6.0

0]L

QG

8.6

7%

[1.0

0]S

QG

2.9

3%

[1.0

0]W

aste

Gen

erat

ed07

1.8

4%

[1.4

9]2

.69

%[2

.07]

1.3

9%

[0.2

9]0

.25

%[0

.28]

Was

teM

anag

ed07

−0.1

2%[1

.51]

0.04

%[8

.00]

0.07

%[0

.46]

0.01

%[0

.26]

Tra

nsp

orte

r0.

34%

[0.1

2]−0

.18%

[0.2

1]0.

01%

[0.1

1]0.

28%

[0.1

3]M

ult

imed

ia3

.91

%[1

.00]

−2.5

0%[−

1.00

]5

.51

%[1

.00]

1.29

%[1

.00]

Fac

ilit

yV

iola

tion

s 06

0.16

%[2

.26]

1.17

%[9

.75]

0.21

%[1

.86]

0.11

%[1

.07]

Fac

ilit

yV

iola

tion

His

tory

02-0

60

.61

%[6

.19]

4.6

3%

[24.

87]

0.8

1%

[5.3

6]0

.20

%[3

.07]

Sta

teV

iola

tion

s 06

3.7

4%

[0.0

4]4

.71

%[0

.04]

5.1

7%

[0.0

4]2

.28

%[0

.04]

Sta

teE

nvi

ron

men

talG

rou

pR

even

ues

06−1

.25

%[2

.39]

−1.8

4%

[2.5

0]−1

.19

%[1

.63]

−1.0

4%

[2.9

6]S

tate

TR

IR

elea

ses 0

61

.48

%[0

.98]

1.75

%[0

.82]

1.36

%[0

.61]

0.54

%[1

.25]

Sta

teG

over

nm

ent

Ideo

logy

060.

22%

[0.2

2]−0

.23%

[0.2

4]0.

55%

[0.2

2]−0

.39%

[0.2

2]S

tate

Hou

seh

old

Inco

me 0

7−4

.07

%[6

.95]

−5.3

1%[7

.35]

−6.0

9%

[6.4

6]−1

.37

%[7

.09]

Sta

teP

erce

nt

Col

lege

Deg

ree 0

55

.70

%[4

.69]

6.35

%[4

.82]

7.3

0%

[3.7

4]2.

09%

[5.4

2]

Sig

nif

ican

tef

fect

sin

bol

d.



the state enforcement variables. Thus, much of the variation in the likelihood of aviolation is due to factors beyond the state regulators’ control.

POLICY IMPLICATIONS

This study provides evidence that federal compliance assistance is effective at in-creasing compliance among some types of hazardous waste generators, particularlythose facilities that generate relatively small quantities of hazardous waste. How-ever, the study does not find any increase in compliance associated with federalcompliance assistance at LQGs. These results suggest that federal compliance as-sistance is not an effective tool for increasing compliance at large generators, eitherbecause assistance is not needed or perhaps because compliance assistance as it iscurrently provided is not useful for such facilities. Thus, any suggestion that thecompliance rates of larger hazardous waste generators can be increased through in-creased compliance assistance should be reviewed carefully. Without a significantchange in the type of compliance assistance offered, such efforts are unlikely to beeffective.

On the other hand, since the results show that compliance assistance does increasecompliance among smaller facilities, any significant defunding of compliance as-sistance efforts is likely to decrease overall compliance. If the funds currently usedfor compliance assistance were to be redirected to traditional deterrence efforts,it is unclear what the net effect might be. Given that the current rate of enforce-ment for smaller generators is quite low, if the redirected funds were to be targetedtoward smaller facilities, the net effect could be positive. However, if funds wereredirected toward larger generators, the net effect might be negative. Thus, it isnot clear how redirecting funds from compliance assistance to traditional enforce-ment would affect overall compliance with hazardous waste regulations. Any suchchange should therefore be studied carefully to determine the appropriate mix ofcompliance assistance and traditional enforcement efforts.

The results of the analysis do not find a consistent complementary relationshipbetween compliance assistance and deterrence across all facilities. However, at con-ditionally exempt generators there is a complementary relationship. This may bedue to the fact that such facilities face a relatively low baseline probability of in-spection. The results suggest that for compliance assistance to be most effective atCEGs, EPA may want to increase the baseline inspection probability. Also, any de-crease in inspection probability at these facilities is likely to significantly underminethe effectiveness of compliance assistance.

SUMMARY

The purpose of this study is to examine the effectiveness of compliance assistance aswell as to explore the relationship between the carrot of compliance assistance pro-grams and the stick of traditional deterrence efforts. I use data on hazardous wastegenerators nationwide to assess the effect of federal regional compliance assistanceprograms in improving compliance with hazardous waste regulations. The resultsof this study show that federal compliance assistance efforts administered by EPA’sregional offices do increase compliance. The findings also show that while smallergenerators do increase their compliance in response to regional compliance assis-tance, the largest hazardous waste generators do not respond to regional complianceassistance efforts. With respect to the relationship between deterrence efforts andcompliance assistance, there is no consistent relationship: For the smallest genera-tors, increased deterrence enhances the effectiveness of compliance assistance, butfor facilities as a whole the opposite is true.



The analysis finds no evidence that historic state compliance assistance programsdecrease violations, although that may be because there is no differentiation betweentypes of compliance assistance programs states offer. However, there is evidencethat compliance assistance programs at the state level are being used in place ofmore traditional enforcement, as the likelihood of an inspection is lower if a statehas historically had a compliance assistance program. Given that the probabilityof a violation is not significantly affected by historic state compliance programs,any substitution of resources from inspections to compliance assistance at the statelevel is likely to have a detrimental effect on overall compliance. Additional researchneeds to be done using more current and more detailed measures of state complianceassistance programs to better understand the full implications of state complianceassistance efforts.

One limitation of this study is that it examines a cross-section of facilities at onepoint in time and only considers the immediate effectiveness of compliance assis-tance. It may be the case the effectiveness of compliance assistance is not immediate.One could argue compliance assistance should have a persistent effect on facilitybehavior so facilities that receive compliance assistance change their behavior fora number of years, not just the year following the assistance. In addition, the effectof compliance assistance may be cumulative—it may take several years of receivingassistance for facilities to change their behavior. Further examination of the effectof compliance assistance taking a longer time frame into consideration both forcompliance assistance efforts as well for compliance behavior should help to shedadditional light on the cumulative effect of compliance assistance on environmentalcompliance.

SARAH STAFFORD is Professor of Economics and Public Policy at the College ofWilliam and Mary, 104E Morton Hall, PO Box 8795, Williamsburg, VA 23187.

References

Ayers, I., & Braithwaite, J. (1992). Responsive regulation: Transcending the deregulationdebate. Oxford: Oxford University Press.

Becker, G. S. (1968). Crime and punishment: An economic approach. Journal of PoliticalEconomy, 76, 169–172.

Berry, W. D., Ringquist, E. J., Fording, R. C., & Hanson, R. L. (1998). Measuring citizenand government ideology in the American states, 1960–93. American Journal of PoliticalScience, 42, 327–348.

Brehm, J., & Hamilton, J. T. (1996). Noncompliance in environmental reporting: Are violatorsignorant, or evasive, of the law? American Journal of Political Science, 40, 444–477.

Burby, R. J., & Patterson, R. G. (1993). Improving compliance with state environmentalregulations. Journal of Policy Analysis and Management, 12, 753–772.

Cohen, M. (1999). Monitoring and enforcement of environmental policy. In T. Tietenberg &H. Folmer (Eds.), International yearbook of environmental and resource economics, III(pp. 44–106). Cheltenhem: Edward Elgar Publishing.

Dennis, W. J. (2007). Waste and hazardous materials. NFIB National Small Business Poll, 7,1–18.

Environmental Law Institute (2003). Beyond enforcement? Enforcement, compliance assis-tance, and corporate leadership programs in five Midwest states. Washington, DC: Author.

Etienne, J. (2011). Compliance theory: A goal framing approach. Law & Policy, 33, 1467–9930.

Evans, M., Liu, L., & Stafford, S. (2011). A facility-level analysis of the long-term consequencesof environmental auditing among hazardous waste generators. Journal of Regulatory Eco-nomics, 40, 279–302.



Gray, W. B., & Shimshack, J. P. (2011). The effectiveness of environmental monitoring andenforcement: A review of the empirical evidence. Review of Environmental Economics andPolicy, 5, 3–24.

Greene, W. H. (1992). A statistical model for credit scoring. Working Paper No. EC-92-29.New York: New York University Department of Economics, Stern School of Business.

Gunningham, N., Kagan, R., & Thornton, D. (2003). Shades of green: Business, regulation,and environment. Stanford, CA: Stanford University Press.

Gunningham, N., Thornton, D., & Kagan, R. (2005). Motivating management: Corporatecompliance in environmental protection. Law & Policy, 27, 289–316.

Heyes, A. (2000). Implementing environmental regulation: Enforcement and compliance.Journal of Regulatory Economics, 17, 107–129.

May, P. J. (2005a). Compliance motivations: Perspectives of farmers, homebuilders, andmarine facilities. Law & Policy, 27, 317–347.

May, P. J. (2005b). Regulation and motivations: Examining different approaches. PublicAdministration Review, 65, 31–44.

Metzenbaum, S. (2007). Measuring compliance assistance outcomes. U.S. EPA State ofScience and Practice White Paper. Washington, DC: EPA. Retrieved August 27, 2008, fromhttp://www.epa.gov/compliance/resources/reports/compliance/research/mcao-whitepaper.pdf.

Potoski, M., & Prakash, A. (2004). The regulation dilemma: Cooperation and conflict inenvironmental governance. Public Administration Review, 64, 152–163.

Russell, C. S., Harrington, W., & Vaughan, W. J. (1986). Economic models of monitoring andenforcement: Enforcing pollution control laws. Washington, DC: Resources for the Future.

Scholz, J. T. (1984). Cooperation, edeterrence, and the ecology of regulatory enforcement.Law and Society Review, 18, 179–224.

Scholz, J. T., & Wang, C. -L. (2006). Co-optation or transformation? Local policy networksand federal regulatory enforcement. American Journal of Political Science, 50, 1540–5907.

Shewmake, T. (2004). Using compliance rates to manage. The Journal of the EnvironmentalCouncil of the States, 17–22.

Spence, D. B. (2001). The shadow of the rational polluter: Rethinking the role of rationalactor models in environmental law. California Law Review, 89, 917–998.

Stafford, S. (2006). Rational or confused polluters? Evidence from hazardous waste compli-ance. Contributions to Economic Analysis and Policy, 5, Article 21.

Stafford, S. (2007). Should you turn yourself in? The consequences of environmental self-policing. Journal of Policy Analysis and Management, 26, 305–326.

U.S. Environmental Protection Agency. (1996, July). FY 1995 enforcement and complianceassurance accomplishments report. Washington, DC: Author.

U.S. Environmental Protection Agency. (1997, September). EPA strategic plan. Washington,DC: Author.

U.S. Environmental Protection Agency. (1998, September). Role of the EPA inspector inproviding compliance assistance: Final report. Washington, DC: Author.

U.S. Environmental Protection Agency. (1999, September). Innovative approaches to enforce-ment and compliance assurance: Action plan for innovation. Washington, DC: Author.

U.S. Environmental Protection Agency. (2000, September). Strategic plan. Washington, DC:Author.

U.S. Environmental Protection Agency. (2003, September). 2003-2008 EPA strategic plan:Direction for the future. Washington, DC: Author.

U.S. Environmental Protection Agency. (2006, September). 2006-2001 EPA strategic plan:Charting our course. Washington, DC: Author.

U.S. Environmental Protection Agency. (2011, February). FY 2012 EPA budget in brief. Wash-ington, DC: Author.


Do Carrots Work? Examining the Effectiveness of EPA's Compliance Assistance Program

Documents

Transcript of Do Carrots Work? Examining the Effectiveness of EPA's Compliance Assistance Program