I PO LC- RE'SEARCH WORKING PAPER 1447 Environmental Inspect'ions ~ npcin n threat of inspections reduce-: and Emissions of the Pulp Pollucion emissions. .-Moevr .nse .o isri and Paper IndustryMoevrinecosdue plants to report ~their emissions levels more.:- The Case of Quebec feunlt euaos Benoit Laplante Paul Rilstone The World Bank - . Environment, Infrastructure, and Agriculture Division Apoicy'ResearchIDepartment I - 1 April1995~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ [R ticY RIESARKCH WORKIN; PAPER 1447 Summary findings Since the carly 1970s, industrial countries have enacted regulation, little empirical work has been done abIotit the (or amended) many environmental laws and regulations effect of current monitoring strategies on pollution to control and improve air and water quality. D)eveloping emissions. countries are increasingly enacting sinlilar legislation. lIut I.aplante and Rilstone supply an empirical frameworL imposing a ceiling on a plant's emissions does not for measuring the impact of environmental inspectiznsn guarantee reduecd emissions or an improved on plant emissions. They apply it to pulp and paper environment. Ensuring the attainment of the regulation's plants in Quebec for which reliable data were available. objectives requires monitoring the behavior of the The results suggest that both inspections and the threat regulated facility and enforcing environmental stand; s. of inspections reduce pollution emissions. They also Most of the literature in environmental economics is show that a plant's decision whether to report its theoretical and simply assumes that polluters comply emissions levels to the regulator is not random. with regulations. Although monitoring and enforcement Inspections improve the frequency of reporting. problems are clearly a pitfall of environmental This paper-a product of the Environment, Infrastructuie, and Agriculture Division, Policy Research Department - is part of a larger effort in the department to investigate the impact of regulation on environmental performance. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Elizabeth Schaper, room NI0-037, extension 33457 (27 pages). April 1995. The Polir, smcareb Working Papr Sorics dissemnates the findings of dwork progress to encourage tbe exchange of bdsas aiout developrnwnt issues.own objecstiu of the series is to get the findings out quickly, cever if the presentations are less than fully polished 7-hc papers carry the namces of the author and should be used and coted accordingly. Thc findings, interpretations. and conclusions are tbe autbors' oum arnd should not becubie = to Wrd Bank, its Excutiue Board of Directors. or any of its rncrnbo coun.7ies. Produccd by the Policy Rcscarch Dissemination Center ENVIRONMENTAL INSPECTIONS AND EMISSIONS OF THE PULP AND PAPER INDUSTRY: THE CASE OF QUEBEC By Benoit Laplante The World Bank Policy Research Department, PRDEI 18 1 8 H Street, N.W. Washington, D.C. 20433 and Paul Rilstone Department of Economics Universite Laval Ste-Foy, Quebec Canada, GIK 7P4 We are very grateful to Wesley Magat, Paul Lanoie and Chuck Howe for valuable comments and suggestions. Participants at the Third Canadian Conference on Environmental and Natural Resources Economics (Ottawa, October 1993) and the Fourth Annual Conference of the European Association of Environmental and Resource Economists (Fontainebleau, June 1993) have also provided helpful comments. Finally, we thank Martine Bossi and Chantal Dallaire for their research assistance. Executive Summary Sincc the beginning of the 1970s, govenmments of developed countries have enacted (or amended) a large number of environrmental laws and regulations directed mainly al controlling and improving air and water quality. Governments of developing countries are also increasingly enacting similar legislation. However, imposing a ceiling on a plant's emissions does not necessarily imply that emissions will fall and that environmental quality will improve. Indeed, for the objectives of the regulation to be attained, the behaviour of the regulated facility has to be monitored, and enviromnental standards have to be enforced. Monitoring and enforcement issues have attracted relatively little research effort. Indeed, most of tthe literature in environmental economics simply makes the (implicit or explicit) assumption that polluters comply with the regulation. Moreover, the existing literature on these issues is for the most part theoretical. Hence, although it has long been recognized that monitoring and enforcement problems are an important pitfall of envirorunental regulation, little empirical work has been done on the impact of current monitoring strategies on pollution emissions. The purpose of this paper is twofold. First, we extend the work of Magat and Viscusi (1990) to produce a methodology and an empirical framework for measuring the impact of environmental inspections on plants' emissions. Second, we apply this methodology to pulp and paper plants in Quebec from which reliable data were available. Our results suggest that both inspections and the threat of inspections have a strong negative impact on pollution emissions. We also show evidence that the decision for a plant to report or not to report its level of emissions to the regulator is not random and that inspections improve the frequency of reporting. Until recently, there have been no data which would allow us to draw any inferences about the responses of plants to inspections in developing countries. The current paper should therefore be considered the initial round of a larger work program. It develops the relevant methodology and applies it to a situation where good measures are available. In the future, PRDEI will pursue similar work in several developing countries. 2 1) INTRODUCIION Since the beginning of the 1970s, governments of developed countries have enacted (or amended) a arge number of environmental laws and regulations directed mainly at controlling and improving air and water quality. However, imposing a ceiling on a plant's emissions does not necessarily imply that emissions will fall and that environmental quality will improve.' For the objectives of the regulation to be attained, the behaviour of the regulated community has to be monitored, and environmental standards have to be enforced. However, while a large amount of resources is devoted to designing environmental regulations, defining and negotiating environmental standards with the regulated industries, it has been acknowledged, both in Canada and the United States, that the resources devoted to monitoring and enforcement are insufficient. What is missing is a commitment of resources to checking up on whether those covered by the law and regulations are doing (or not doing) what is required of (or forbidden to) them. (Russell, 1990, p.243) This lack of resources has forced the regulator to rely on a system by which a polluter (I) is presumed to comply with the environmental standard if it is using the appropriate emissions control technology (initial compliance) and (2) has to report at regular interval its cmissions of the regulated pollutants (self-monitoring). Audits of plants and on-site inspections are rare events. For example, Russell (1983) notes that measurement of the discharges of large sources of air pollution occurs on average only once every eight and a half months. Wasserman (I 984) notes that experted inspection frequencies for minor sources are bi-annual. With respect to hazardous waste disposal, less than 10% of the regulated facilities were reached at all in 1986 (General Accounting Office, 1987). In Quebec (Canada), while 59 pulp and paper plants were in operation during the period 1985-1990, there has been a total of only 54 on-site inspections in the industry. The present system does not put pressure on agency policy makers to make the large investments in monitoring and personnel that are required to make the tedious and unending work of credible enforcement a bureaucratic reality. (Ackerman arid Stewart, 1985, p. 1333) The same holds true when effluent charges or tradeable permits are introduced. 2 Enforcement can take various forms: orders, fines, loss of market value or reputation, etc. See Dewees (1990), Muoghalu et al. (1990) and Laplante and Lanoie (1994) for more details. 3 Monitoring and eniborcement issues lhave attracted relatively little research CeTort.? Moreover, most ol this elTort has beeni theorctical.) E,xcept for l)eily and Gray (1991) and Magat and Viscusi ( 1990). we can only nlOtC the mere absence of* empirical analysis.5 Magat amd Viscusi (1990. hencelborth MV) have estimated the impact ol inspections on the sel/-reported discharges of biological oxygen demand (BOD) by the pulp and paper industry in Ihe United States. Since the pulp and paper industry is the largest discharger of BOI). it has been the focus of a considerable amount of regulatory ellTrt. ThIlis explains that there is, lor this industry, an extensive data base on BOD dischargc measurements per plant (the EPA Permnit Compliance System. also known as the 13CS data base) and on on-site sampling inspections by the regulator.6 Sampling inspections are cnnsidered to be the regulator's ultimate device to assess compliance with the standard and give credibility to the sell-reportinig procedure. MV have found that each inspection reduces the mean value of rcported discharges of BOD by approximately 20%. They also found that inspections have a pcnnanent effect on discharges. The purpose of this study is to measure the impact of inspections on the self-reported emissions levels of plants in the pulp and paper industry in Quebec. Our analysis differs from MV on a number of important accounts. Firstl MV measured the impact of inspections on the absolute level of emissions as well as on the status of compliance of the plants, i.e. whether plants comply or do not comply witlh the standard. However, to the extent that environrmental quality is the ultimate concem, the interest is not necessarily whetheT inspections induce compliance, but instead whether inspections effect the level of emissions exceeding the standard. Indeed, if inspections do not induce a plant to comply with the standard, they may nonetheless induce the plant to reduce the amount of emissions by which it exceeds the standard. Hence, a plant's compliance status may we note, along with Cropper and Oates (1992), that most of the literature in environmental economics simply makes the (implicit or explicit) assumption that polluters comply with the regulation. 4 Among others, see Beavis and Dobbs (1987). lIarrington (1988), Lee (1984), Linder and McBride (1984), Russell et al. (1986), and Tietenberg (1992)). 4 Fisheries have attracted a certain number of empirical analysis (see, among others, Sutinen and Andersen (1985). Anderson and Lee (1986). and Furlong (1991)). Deily and Gray (1991) examines the EPA's enforcement activities "for evidence that enforcement was responsive to the possible economic disruption from plant closings" (p. 260). Deily and Gray claim that their paper is "thefirsl empirical study of the EPA's enforcement activity at the plant level" (p. 260). 6 A "sampling inspection" is an inspection where the regulator samples the plant's effluents and measures the BOD content of the samples. Jther types of monitoring activities are also performed. See Magat and Viscusi (1990, p. 338) for more details. 4 remaini uncllaiiged as tlic result of inspcctions. and yet environmenmal quality improve. MV wrotc: 'Unrbrtunately, it is not possible to conistruct a reliable nieasure of' the amounit of' pollution in excess ol' the perrmitted an-ount since data pertaining to the level specified in thc pennit are niot available from thie Pt'S data base" (p. 345). In our data set, we do have access to the standard per plant. I lencc, wc are able to test lor the impact ol inspectionis on the level ol' emissioins r elhifie to the standard. Second., thCe Imost obvious question whicih arises in the context of' the current analysis concerns the possible endogencity of' inispectionis. Indeed, whlile past inspections are given, the regulator's current decisioni to inspect a plant may itself be e'fected by the plant's eniissions level. 1'herelore, one miglht reasonably expect that in the current period, it is the perceived probability or threat ol'an inspection (rather than an inspection per se) which is the variable ol' interest. In other words, both inspections and the probability of an inspection may have an effect on emissions. MV have rejected the hypothesis that current inspections arc exogenous and perform their estimatioris usinig only lagged inspection variables. Interviews with employces of the Quebec Ministry of the Ensi ronment strongly suggest that in any given period. the plants chosen to be inspected are not randomly picked, and in fact, that the probability of an inspection may be inversely related to the number of previous visits. This reflects the Ministry's desire to visit as many plants as possible. From a statistical perspective, this amounts to sampling witlout replacement. Our interviews also indicate that changes in production capacity may trigger an inspection.7 Consequently. we estimate an "ins;pections equation" in which inspections are a function (among others) of a variable indicating the number of inspections which have been conducted at the plant prior to the period of reporting as well as capacity. In our sample of analysis, we also reject the exogeneity of current inspections. We then re-estimate our basic model by instrumental variables using expected inspections as instruments. Our results strongly suggest that the threat of an inspection as well as actual inspections have an impact on pollution emissions. Thi-d. though the EPA Permit Compliance System lists 194 sources with BOD discharges, only 77 of these sources submitted discharge monitoring reports to iae EPA. If the missing information is not governed by a random process, this obviously raises the possibility of a selection bias. MV are aware of this problem and inform the reader that "[our] results need to be interpreted as estimates of the response to EPA inspections of fim-s whose discharge levels are regularly reported to the EPA's national data base" (p. 342). We also face the same issue in Quebec. Indeed, as mentioned above, there were 59 plants in operation over the period 1985-1990. In principle, as required by the regulation, each of these plants must submit a monthly discharge report to the Ministry of the Environment. However, only 46 of the 59 plants filed reports on a regular basis during the sample period. E In order to allowv for sample selection problems, we compute a simple 7 In such cases, the purpose of the inspection is to verify whether the change in capacity affects compliance with the standard and/or environmental quality. For some of these 46 plants. a few observation points were missing. These were smoothed over using forecasts from 12th-order univariate autoregressions. 5 binary clhoice model ol' reporting and then augmcnt our basic model with a correction term suggested by I-leckniian ( 1 979). Our results suggest that inspections havc an impact not only on the levels of 'emissions, but also on reporting frequencies.9 Hence, the bencfits of inspections are not simply that they reduce pollution emissions; they also providc the regulator with morm information by inducing more frequunt reporting. F'inall. we estimate the impact of inspections not only on the reported discharges of BOD, but also on reported discharges of total suspended solids (TSS). It should be noted that the technology used to abate BO3D differs from the one used to reduce TSS. It is found that inspections do not have the same effect on the emissions of these two pollutants. Tlhis suggests that a monitoring strategy cannot be developed irrespective of the pollutant (and therefore the abatement technology) thal is the object of the regulation. The rest of the paper proceeds as follows. In Section 11, we present and describe our data set. In Section III, models and results are presented. We conclude in Section IV. 11) THE INDUSTRY AND THE DATA SET"0 (A) The industry The pulp arid paper industry is an important economic agent in the province of Quebec. In 1989, more than 31 000 individuals were employed by the industry which paid more than one billion dollars in wages and salaries (Quebec, 1990). In that same year, it was estimated that the industry's capital made up 25% of the capital of the entire manufacturing industry in the province. Newsprint represents by far the most important output with 56% of total production (L'Association des Industries Forestieres du Quebec, 1991). The province of Quebec is the largest producer of newsprint in Canada with 45% of Canadian production and one of the largest in the world with 14% cf world production in 1989. Most of its output (73%) is exported to the United States; this represents 20% of Quebec's total exports (Quebec, 1990). If the industry is a major contributor to Quebec's economic activity, it is also one of the most important sources of conventional pollutants.'" The BOD load produced by the industry is MV do address non-reporting although not with a formal model. They test whether or not there is a statistically significant difference in the fiequency of reporting before and after an inspection, for the 77 plants of their sample. They find that inspections increase the frequency of reporting of those plants. '° For a more detailed discussion of the industry and the regulation, see Nemetz (1986) and Sinclair (1991). " These include BOD and TSS. Conventional pollutants do not include toxic emissions such as dioxins and furans. 6 estimated to represent more than 60% of the total BOD load produced by the manufacturing industry in Qudbec. TI'his represenls thc cquiivalent of the BOD produced by approximately 15 million individuals. Hence, one may expect that a reduction in the production ol conventionial pollutants by the pulp and paper industry would have a significant impact on water quality in the province. This presumably explains that so much attention has been dcvoted to thc emissions control activities of the industry. In Canada, jurisdiction ovcr water pollution control (and more gcnerally over pollution control) is shared bv the federal and provincial goveLnments. The basis of the ovcrlap rclies on the Constitution Act of 1867.12 Insofar as water pollution is concerned, the fcderal government has played an important role through its Fisheries Act'3 under which it has introduced the Pulp and Paper Effluent Reguiations'" in 1971. Similarly, the government of Quebec, pursuant to its Environmental Quality Act' 5, has introduced the RNglement sur les fabriques de pdtes el papiers (I . As of May 1992, new federal and provincial regulations were introduced for the pulp and paper industry whereby new emissions standards for TSS, BOD, toxicity, dioxins and furans have been defined. The standards contained in the provincial regulation are at least as stringent as those contained in the federal regulation.'7 However, for the period covered by our sample of data (1985- 1990), only the Quebec regulation contained standards for BOD and TSS (and not on toxicity). Hence, only the latter is relevant for the current study. These standards are uniform and apply to every plant in the industry. They are set in kilograms per ton of production. It is therefore important to understand that the total amount of BOD and TSS that a plant can emit in any given 12 The invol vement of the federal government in matters of environmental protection is made possible through its jurisdiction over fisheries, harbours, criminal law, and its residual power to legislate for the peace, order and good government of Canada. The appropriate roles and responsibilities of federal and provincial go-ernments are the subject of an everlasting debate (see, for example, Kennett (1991)). 13 Revised Statutes of Canada, 1970, c. F-14. 14 C.R.C. 1978. c. 830. 15 L.R.Q., c. Q-2. 16 R.R.Q., 1981, c. Q-2, r. 12. 7 These regulations were preceded by the adoption of an administrative agreement which makes the Quebec government the primary agent in dealing with the industry on environmental issues. In particular, the Quebec government is solely respoiisible for collecting data on pollution emissions. The federal government will have ongoing access to the information thus compiled and is therefore able to oversee the plants' compliance with the federal regulation. A plant that is not complying with the provincial regulation (and therefore not complying with the federal regulation) may face enforcement actions from both levels of government. 7 period is u lunction ole its output productioni during that period: dhC greater its production. the greater is the allowable discharge. A plant's compliance witlh the regulation is assessed by coimparing the allowable discharge willt the total load reported bv tile plant." (13) 'Ilic data set According to the Rfg/lem7eint sur lesj,/&briques de p61 es el papiers, plants are required to submit monthly reports to the Qudbcc Ministry ol the Environment concerning the plants' discharges ol' TSS and BOD during the month. Measures have to be taken at times and intervals specified by the regulation. Se(l-monilo ring is the most important source of information used by the regulator to assess a plant's compliance with the standards. All the data used in this study have been provided by the Quebec Ministry of the Environment; most of them are issued from the Department's annual publication Bilans annuels de coqformitd environmentale - secleur des pd/es et papiers. These documents are based on the monthly reports of all mills of the province and contain the mill's monthly discharges of BOD and T'SS. The reports also indicate the allowable discharges of each individual plant for each individual month.'t9 As mentioned above, a large number of observations are missing from the monthly reports filed by the plants of the industry. A natural and important question arises as to whether these are missing in a random or systematic manner. In the former case, estimation can proceed in a fairly straightforward manner with the missing observations smoothed over in an appropriate way. On the other hand, if there is a systematic pattern to the non-reporting, this can lead to a selection bias in the usual least squares estimates. After an examination of the data, we decided to divide the missing observations into two categories. In a number of cases, some of the plants had neglected to report their emission levels on a few occasions in what seemed to be an unsystematic way. These observations were treated as randomly missing and were replaced by forecasts from 12-th order univariate autoregressions. This left us with a data set including information on 46 of the 59 plants. ,x in the United States, the regulation set a limit per pound of pulp and paper nroduced. Then, the total amount of BOD that a plant can discharge on any given day is obtained by multiplying the limit by the total number of pounds of pulp and paper the plant produces on that day. It appears difficult to compare the Quebec emissions standard to their American counterpart since they were defined very differently. In particular, in Qudbec, allowable discharges were defined for each and every stage of production, from wood washing (whether it be logs or wood chips) to the mnaking of the final product. They also varied according to the production process. However, interviews with the Qudbec Ministry of the Environment suggest that the allowable discharges per ton of output in Quebec and the United States were practically the same. 9 The reports also indicate the monthly production of each plant. However, this information is confidential. Moreover, given the complexity with which allowable discharges are calculated, it is not possible to find out what was the output production in any given period from knowing what was the allowable discharges for that same period. 8 Ihis dala set was used to) estimiailte the eflbect of' inspectionis without controlling fhr samiiple selectioni issues. '['lic 13 reimiainilng piants hlid lihiled to report their emissions to sucih an extent that it was not evei possible to smooth these over witlh autoregressions. I'lhese were treated als possibly imlissinlg in a nonranidonm miainner. th1us leuding to a sample sceltion problem. 'I'is issue w;11 be discussed in mlorc detail below. In addition to iltose in thc regulat measurements carried out by eacih planit, emissionis are also mcastired during thc periodic sampling inspections coniducted by the regulator. Inspections consist of (I) the regulator and the producer each taking samples from thc mill's eflluents, (2) mcasuring their *FSS and BOD conitents, and (3) comparing thesc measures witli the applicable slandards.21'1 'lhe i Qudhcc Ministry ol thc Environment perlormed 54 sampling inspections from 1985 to 1990. 1 lowever, since 13 plants are excludcd from our initial sample of analysis, only 47 of these inspectionis are initially accounited for. Belore presenting our model. some descriptive statistics are of intercst. These appear in Table 1. Note first that tile avwrage production of both BOD and TSS is above the norm. In fact, 37.38% of the self-reported discharges of TSS are above the norm (35.75% for BOD). In MV's sample of analysis, the occurrencc of reported violations for BOD is 25.2%. Note also that the unconditional probability of inspections in any given month is 0.0148, or approximately 1.5%. In MV, this probability is approximately 4.25% so that the probability of an inspection is almost 3 times higher in MV's sample. Variables of the form PRODi (i = 1,....5) represent dummy variables for the plant's type of production. Newsprint is by far the most important good produced by these plants. hliese will be used to reflect that plants have different operations and technology. Finally, variables of the form REGi (i =.8) are dummy variables for the region in which the plant is located. A question which naturally arises with self-reporting is whether the plants accurately report their emissions levels. To some extent, this is an unresolvable problem and the results should be interpreted conditional on the fact that the reporting was conducted by the plants themselves. However, there are several reasons to expect that the reported emissions are not completely inaccurate. First, the technology used by the plants is by now well-knowrn and has been used for a relatively long period of time. Hence, knowing the precise technology used by any given plant, its actual production, and the waste water treatment facilities it is using, relatively good estimates of its pollution load can be obtained. Second. it should be noted that fraud in reporting is a serious criminal offence. Third, our discussions with various parties indicate that unionised employees are very prone to inform the regulator about a plant's wrongdoing with respect to the management of its waste. Finally, at the same time as a sampling inspection takes place, plants are also required to perform a sampling, independently of thoF? usually conducted for their monthly reports. Given the presence of an inspector, one would therefore expect the plants' measurements of BOD and TSS to 20 It is important to recognize that the purpose of an inspection is not to detennine the accuracy of previous reports. This is technically impossible to do since the TSS and BOD discharges of previous months have "disappeared" from the mill's vicinity. 9 Ie LIccullilC 101i) i[ IeaslI IIIOSC S&tiiplings. 'Thiis prvides an additional source of inforniation regarding 1he IILCliIricy ol' their reporls. We thltis conlulcted paircd dilThrencc of menscs tcsts using, as a measure ol' reportinig zccuraLrley. 1 difTl'crenee lbtweeni tlhe plants' loud measured in presencie of uw inspector andli televels indicatecl oni the mothlily repirts Ilfr iliat samei period. 21 As indicated in liblc 2. the resLultilg test statistics do nol inidicate ally syste:matic lailsilicaLtion ol rcsults. 2 111) MODE)LS ANI) REStJLTS In tihis section, we proceed in thrce steps. First, we discuss least squarcs estimates of the basic m)odel to examine the efTects of inspections without controlling cithcr ror possible cndogencity o* the inspections or possible selection biascs (section (A)). Second, we allow and test kor the possibility that currcnt inspections arc endogcnous. and then cstimatc our model by instr umental variables (section (B)). In both of thcse sections, thc estimates are calculatcd using the data lor the 46 plants whose reports were basically complete. Finally, we tcst for the possibility that the process governing non-reporting may not be random, and then modify our model as suggested by Heckman (1979). In this last section (section (C)). we also allow inspections to be endogenous. (A) Irhe bnasic model Our objective is to test for the impact of inspections on two sets of variables: (1) the absolute discharges of BOD and TSS and (2) the level of discharges of BOD and TSS relative to their respective standards. The basic model we estimate is of the same form regardless of the pollution variable of interest. Let Pi, denote the pollution variable associated with plant i in period i..23 In the absence of sample selection corrections, the equations estimated are of the following forn: 12 P,1 - a + ,320-12+0 INS, +Z:o .INSt.j+ 1 2REGi+ P 3PRODi + P 4CAP +y t+c,1 (1) i=]1..46;tl= ,..60 The first variable is the plant's lagged value of pollution. This variable is included to capture potential seasonal effects, which may be strong (especially for BOD) in Quebec with important variations of temperature between summer and winter. This variable also reflects the fact that the 21 For example, if an inspection took place in May, we would compare the plant's measure from the sample taken by the inspector with the load reported by the plant for the month of May. 22 It should be said that this is a very simple measure of reporting accuracy that would not be an accurate measure under a number of scenarios. 23 In some specifications, P1, is the absolute discharges while in others, it is the discharges in excess of the norm. 10 installation of emissions control equipment typically requires a long time. To this extent, the lagged pollution variable could also be interpreted as a proxy for the production technology. I-ence, we would expect that the (12-month) lagged value of pollution to be a good explanatory variable for current pollution.24 The second group of variable reflects the efTect of current inspections and indicate whether the plant was inspected in period t. The third group of variables indicate whether the plant was inspected in period t-j. An empirical question concems the appropriate number of lag lengths to include in the analysis. When we included four lags in the model, the corresponding coefficient estimates were generally negative, of the same magnitude and statistically significant. However, as a referee pointed out, to test whether the efflects of inspections are persistent, it is preferable to include also less recent inspections. With twelve lagged inspections, the estimates were still generally negative and of the same size, but the individual coefficients had small t-ratios. To circumvent this problem, we then conducted Wald tests to see whether we could reject the hypothesis that the coefficients were equal. Since we were unable to reject this hypothesis for each of the models, we have imposed this constraint on the coefficients of lagged inspections. The resulting point estimates are substantially sharper and, in fact, yield considerable evidence that the effects of inspections are persistent, if not permanent. REG and PROD are 8 X I and 5 X I vectors of dummy variables reflecting the plant's location and type of output.25 The CAP variable indicates plant i's daily productive capacity at time t. It should be noted that plants periodically change their productive capacities, and this is in fact the case in the sample period. Plants with higher capacities should produce higher levels of pollution. However, it is important to remember that allowable discharges are also a function of output and consequently, higher levels of pollution do not necessarily imply that a plant is more likely to be out of compliance. The final variable allows for a time trend in pollution emissions. Using quarterly data, MV have instead used a set of quarterly dummy variables and report that there was no interesting pattern in the results. With monthly data, a similar procedure leads to an important loss in the degrees of freedom and so we used a simple linear time trend. Moreover, a time trend has a straightforward interpretation, namely the overall trend in pollution emissions in the absence of inspections. MV reports having regressed absolute level of discharges against a linear time trend and found no significant relationship. As shown below, this is not so in our case. The results from these estimates are presented in Table 3. There are four sets of results corresponding to the four measures of pollution emissions.26 First note that the coefficient on the twelve-month lagged dependent variable is, as expected, positive and has a strong effect on absolute discharges, especially so for BOD. MV obtained a similar result for BOD. Second, the coefficients on current and past inspections are always negative, although not always statistically significant. This is especially the case when discharges are measured relative to the norm. This 24 We have also experimented with other lag lengths. It had little effect on the overall results. 25 For identification, REG9 and PROD6 are left out of the estimated models. 26 These equations were estimated separately. We also computed seemingly unrelated regression. The results were very similar. 11 suggests that the means by which BOD and TSS emissions are reduced also have an impact on the norm.27 MV lound that each inspection reduces the mean value of absolute BOD discharges by approximately 20%. Our results indicate that lagged inspections reduce absolute discharges of BOD by approximately 7%. Significant coefficients on regions (especially on REG, and REG2) indicate that there might be important regional differences in the nature of the relationship that exists between the regulator and the regulatees and/or the monitoring and enforcement procedure across regions. As expected, other things being equal, plants with larger capacity should have higher levels of absolute discharges, but need not be out of compliance. The statistically significant negative coeflicient on time indicates that once the impact of inspections is accounted for, there is a trend for both pollution discharges and discharges relative to the norm to fall over time. This is in contrast to the results reported by MV. (B) Endogenous inspections The most obvious question which arises in the context of this study concems the possible endogeneity of inspections and the consequent impact on the least squares estimates. If inspections are endogenous and corrclated with the same variables which determine current pollution levels, then the least squares estimates will be biased in general. To put this another way, it may not be contemporaneous inspections which have an effect on effluent levels so much as the probability of an inspection. To control for this (and to identify the resulting parameters), it is necessary to model the inspections using some variables which do not enter the basic model. Interviews with employees of the Quebec Ministry of the Environment indicate that inspections are motivated by two considerations. First, plant size seems to be a factor: smaller plants are less likely to be inspected than larger plants. Moreover, plants which make changes to their productive capacities are more likely to be inspected. Second, there seems to be an effort to visit as many plants as possible. In other words, the plants to be inspected in any given period do not appear to be chosen randomly. An obvious implication of this "sampling without replacement" strategy is that a plant knows that, all things being equal, the probability of an inspection is inversely related to the number of previous visits. It therefore appears appropriate to estimate an "inspections equation" where inspections are a function of variables in the basic pollution equation as well as a variable indicating the number of inspections which have been conducted at the plant prior to the current period: cumi1 = XINSiT (2) T ' I(2 Since inspections are a qualitative variable, a simple way to model inspections is as the following: INSi, = 1[S Xi, >itj i = 1,2....,46; t = 1,2,...60 (3) 27 This would be the case if output were to fall as a result of inspections. Unfortunately, we are unable to substantiate this possibility since we did not have access to plant's production data. 12 whcrc I I-] is the usual indicator function, Xi, contains the variables determining inspections, and Tjj, is a variable which could capture, for example, some unobserved tolerance level above which an inspection is conducted. For simplicity, we assumed that -qit are identically and independently distributed normal random variables so that equation (3) is simply a probit model. Table 4 provides the results of this probit regression of inspections on a constant, the number of past inspections, capacity and a time trend.28 As far as inspections are concerned, it is interesting to note that they are not clumped together at the beginning of the period, but rather seem first to decline and then jump at the end of the period.29 As a result of this, we made the inspections equation quadratic in the time trend variables. The results confirm what one could expect: the probability of an inspection is a decreasing iunction of past inspections and an increasing function of capacity.30 Also, all things being equal, the probability of being inspected appcars to be increasing over time. This can be interpreted as a proxy for additional resources being committed over time to monitoring activities. Given this, it is sensible to consider testing for the exogeneity of current inspections. In fact, for three of the four Wald tests (see Table 5), exogeneity of current inspections is strongly rejected so that the least squares estimates of the parameters in equation (1) are most certainly biased. This being the case, it is instructive to consider the effeets of reestimating the model using the fitted values from the inspections equation (3) and the other right hand side variables of equation (1) (apart from current inspections) as instmrnents. The results appear in Table 6. With the exception of BOD emissions relative to the norm, the coefficient estimates on current and lagged inspections from the IV estimation are all negative and strongly significant. Apart from being substantially more significant, note that the magnitude of the coefficient on current inspections is much larger when estimated with instrumental variables. This is attributable to the fact that with IV estimation, current inspections (a discrete indicator variable) are effectively replaced by the conditional probability of an inspection, which has a smooth distribution and takes values on a much shorter interval. The strongly negative coefficient estimates on lagged inspections indicate a persistent, if not permanent, effect from inspections. The results now indicate that past inspections reduce absolute BOD discharges by approximately 28% (compared to 20% obtained by MV). Since an altemative interpretation of the IV estimates is that inspections in our basic model (equation(l)) are replaced by expected inspections, it appears that the threat of an inspection may have most effect on pollution emissions. This is not to say that actual inspections have no impact on 28 We also ran regressions using the region and product indicators, but these did not improve the fit of the model. 29 The number of inspections for each year in the data set is the following: 15 (1985); 9 (1986); 6 (1987); 8 (1988); 3 (1989); 13 (1990). 30 Estimates were also obtained using other variables such as previous pollution levels. Results were not improved. 13 a plant's pollutioni control behaviour. But it does indicate that this beiaviour is also a function of the probability of beinig inispected. If the inspection strategy is dr..termined by sampling wvitlhout replacement. then one may suggest that lagged inspections might have the opposite sign since once the rcgulator has come by once, the plant may (correctly) guess that it will not come back for a large number of periods.3' While this is possible, it may also be the case that inspeclions prompt changes in the planl's behavior Ihal ar1c of a permanent nature. One can think of numerous reasons including changes in equipment, emiployce lunctions and simple changes in the employer and employees awareness of the regulations. The sign of thcsc coeflicients is therefore an empirical matter. We find them to be significantly negative.32 The other coeflicient estimates are very similar to those when least squares were used. (C) Missing data As mentioned above, the exclusion of missing observations can result in a selection bias if the filing of a report is in fact not a *andom event, leading to inconsistent parameter estimates. As a first step in allowing for sample selection issues, we estimate a "reporting" equation to predict the probability that a plant reports its emissions levels. Since we do have some information on the plants even if they do not report, we are able to compute a simple binary choice model of reporting as a function of cumulated inspections, capacity, as well as a time trend (which is again specified as quadratic). In other words we calculate the coefficients from a model written as the following: REP,l = 1[5 Xi, > jħij, i = l,2,...59; t= 1,2,...60 (4) Note that for these estimates the entire data on all 59 plants was used. This was estimated using a probit model, that is, assumning that the gi,u are nornally distributed. The results for this regression are summarised in Table 7. Note that cumulated inspections have a strong positive effect on reporting. This result is important in itself as it indicates an important secondary function of inspections in the reporting/monitoring process. It also seems clear that larger plants, having more resources at their disposal, are more likely to file their reports. There does not seem to be any significant trend in report filing that is not captured by the CUMi variable.33 Overall it seems clear that be act of reporting is not random, although it is not necessarily clear that this is due any strategic planning on the part of the plants. Having estimated the parameters in this equation -we went back to the subsample of 46 plants and augmented the basic model with a correction term as suggested by Heckman (1979). The equation of interest becomes the following: 3 1 This point was raised by one of the referees. 32 Note moreover that this effect is controlled for when equation (1) is estimated by instrumental variables in which case current inspections is effectively replaced by the probability of an inspections given, amongst other things, cumulated past inspections. 33 In fact, when REPi was regressed only on CUMi, the corresponding coefficient was strongly significant. 14 12 Pil=aX ++ Pi.,-12+ -EjlNS1.j+P2REGi+P3PRODi+ 4CAPI,+y t+sXij, +e-i (5) j-O i = 1,2,...,46; t 1,2,...,60 where Xi, = + (6 Xi,) / (D Xi,). + and (D are the standard normal density and cumulative distribution and e denotes the probit estimate of 8. In this context a serves as an estimate of selection bias. Under the null hypothesis that the data are missing in a random manner, ca should equal zero. This equation was also estimated using instrumental variables.34 With respect to the effects of inspections, in all four cases the instrumental variable estimates of the coefficients on inspections are all significantly negative, except for BOD discharges relative to the norm as shown in Table 8.35 Moreover, with the inclusion of a sample selection correction, it is interesting to note that the sign on the time trend is negative and statistically significant in 3 cases out of 4. This may be evidence that, apart from inspection inducements, there is no effort on the part of plants to reduce their emission levels. IV) CONCLUSION Securing compliance with environmental standards is a difficult task. Current monitoring practices and enforcement initiatives (or the lack thereof) have been increasingly criticised. Regulators are therefore experimenting new approaches. Because of limited resources and the resulting need to establish priorities, each EPA program at agency headquarters in Washington, D.C. has developed compliance monitoring plans and enforcement response policies. These stategies generally direct the most intensive efforts to those segments of the regulated community most likely to be in non-compliance. (Silverman, 1990) Similarly in Canada, 34 Consistent estimates of the standard errors in this case were obtained using the method developed by White (1980). Once again, we constructed Wald tests for endogeneity of current inspections. Here, in all four cases, the test statistics were large enough to reject the exogeneity of inspections. 35 Overall the inclusion of a sample selection correction led to much precise estimates as evidenced by the t-statistics. 15 tJpon evaluating the results of the National Inspection Plan at the conclusion of the 1990-91 year, EInvironment Canada found that all regulations did not require the same levcl of compliance verification. and decided on a target-oriented approach. (Canada, 1992) Ilowever, for such an approach to be effective, one must have a clear understanding of plants' pollution control behaviour. Regulators must be able to observe characteristics of plants and industries and from these characteristics, predict whose "most likely to be in non-compliance". In particular, one needs to know how current monitoring practices affect pollution behaviour and in the light of this knowledge, re-allocate, if necessary, monitoring resources more efficiently. We have shown evidence in this paper that both inspections and the threat of inspections have an impact of emissions. We have also shown evidence that the decision to self-report level of emissions is not random and that inspections improve the frequency of reporting. Once this effect is taken into consideration, the impact of inspections on emissions is even larger. These results have direct implication on the allocation of scarce monitoring resources. In particular, credibly increasing the probability of inspections can induce a significant change in plants' pollution behaviour. The quality of our environment crucially depends on the credibility of the monitoring activities and enforcement actions practised by regulators. Along with Cropper and Oates (1992), we do believe this to be an area of research "where economic analysis may make some quite useful contributions" (p. 697). 16 TABLE 1 DESCRIPTIVE STATISTICS OF SAMPLE (Monthly data 1985:1 - 1990:12 for 46 plants) Variable Mean Standard deviation Total Effluent Production 47.309 49.5464 Total Suspended Solids Emissions (TSS) 5.5386 6.1210 Standards 5.2679 4.0883 Biological Oxygen Demand Emissions (BOD) 19.2401 28.4372 Standards 18.4768 26.7975 Inspections 0.0148 0.1207 Violation of TSS Standard 0.3738 0.4839 Violation of BOD Standard 0.3575 0.4793 PROD 1 (1= Kraft Pulp) 0.1957 0.3968 PROD2 (1 = Newsprint) 0.4130 0.4925 PROD3 (1= Recycled Pulp) 0.0652 0.2469 PROD4 (1 = Office Paper) 0.0217 0.1459 PROD5 (1 = Chemical Pulp) 0.1522 0.3592 PROD6 (1 = Other) 0.1522 0.3592 REGI (1= located in region 1) 0.1087 0.3113 REG2 0.1304 0.3368 REG3 0.1522 0.3592 REG4 0.2174 0.4125 REG5 0.0652 0.2469 REG6 0.1087 0.3113 REG7 0.1087 0.3113 REG8 0.0652 0.2469 REG9 0.0435 0.2040 Capacity of Production 15.8922 12.0868 17 TABLE 2 PAIRED DIFFERENCE OF MEANS TESTS BOD TSS Mean measurements with regulator present 19.1593 8.2632 Mean self-reported measurements, regulator absent 19.0697 6.6543 Difference 0.0896 1.6089 t-difference 0.10230952 0.2144231 18 TABLE 3 EMISSIONS EQUATIONS ORDINARY LEAST SQUARES' (Sainplc size - 2716) Independent Absolute dicharges Discharges relative to norm Variables BOD TSS BOD TSS CONSTANT 0.8783 3.6740 0.1063 2.1150 (0.6566) (8.8756) (0.0347) (4.4517) Pi.t-12 0.8144 0.4228 0.1511 0.4196 (80.812) (33.796) (8.0051) (31.199) INS, -4.6976 -0.5796 -2.5810 -0.7632 (-2.9283) (-1.1704) (-0.7014) (-1.3377) INS; -1.3115 -0.6413 -1.0186 -0.4082 (-2.5466) (-4.0472) (-0.8648) (-2.2364) PROD, 0.9211 1.9236 -2.0380 1.1488 (1.0119) (6.7883) (-0.9758) (3.5403) PRO')2 1.3253 1.3577 1.9298 0.6156 (1.4742) (4.8576) (0.9373) (1.9234) PROD3 8.5664 3.6086 19.270 2.0956 (6.3952) (9.0274) (6.7732) (4.6436) PROD4 0.3520 0.3841 0.5537 -0.0703 (0.2582) (09127) (0.1771) (-0.1450) PROD5 -0.1381 0.0784 0.9640 0.1465 (-0.1869) (0.3439) (0.5688) (0.5579) REG, -4.6449 -5.4594 -8.4068 -3.0172 (-3.2435) (-12.392) (-2.6303) (-6.0522) REG2 -3.8455 -3.5180 -6.3141 -1.5032 (-3.0151) (-8.9461) (-2.1723) (-3.3394) REG3 -0.4137 -2.5488 -1.7323 -1.1595 (-0.3340) (-6.6878) (-0.6118) (-2.6440) REG4 -0.1182 -3.3124 3.3458 -1.6605 (-0.0989) (-8.9735) (1.2239) (-3.9180) REGs -1.2703 -3.3482 -1.3965 -1.8530 (-0.8952) (-7.6601) (-0.4308) (-3.6891) REG6 -0.5655 -2.6949 1.0118 -1.4283 (-O.4162) (-6.4467) (0.3258) (-2.9672) REG7 -1.7044 -3.8094 -1.6109 -2.4143 (-1.3197) (-9.6178) (-0.5489) (-5.3086) REGs 1.0611 -3.2194 2.7014 -1.5990 (0.7671) (-7.5468) (0.8522) (-3.2504) CAP 5.9976 4.0212 3.0602 -0.5524 (7.5076) (17.105) (1.9065) (-2.2230) TIME -1.7881 -1.7016 -3.8478 -1.7560 (-2.8668) (-8.7511) (-2.6881) (-7.8269) R2 0.891 0.720 0.115 0.370 The dependent variable is the appropriate pollution variable divided by 1000. 19 TABLE 4 INSPECTIONS EQUATION (Sample size = 2716) Independent variaibles Coefflcient t-stats CONSTANT -2.5442 -10.586 ci'M, -0.1956 -1.912 CAP1, 0.5955 3.525 lIME -0.7887 -0.844 l'IME2 1.2067 1.400 IJ(i-lI .IK11.11 I(X)I) -1ST STMIiTIICS: 17.345 20 TABLE 5 WALD SPECIFICATION TEST FOR EXOGENEITY OF CURRENT INSPECTIONS (Sample size = 2716) Variables Value of Wakl's statistic BOD 49.32 1 rss 22.398 BOD-NORM 0.6235 TSS-NORM 27.620 21 TABLE 6 EMISSIONS EQUATIONS INSTRUMENTAL VARIAP4J,E ESTMATION' (samplc siwj = 'b) Independent Absolute dicharges Discharges relative to norm Variables BOD TSS BOD TSS CONSTANT 6.5776 4.9203 1.6032 3.7160 (1.6989) (5.4476) (0.4309) (3.2848) PU,12 0.8198 0.4116 0.1524 0.3946 (32.854) (17.579) (7.7901) (13.816) INS, -193.40 -40.402 -51.927 -52.318 (-2.8843) (-2.5961) (-0.8032) (-2.6467) INS. X -5.3703 - 1.5054 -2.0836 - 1.5240 (-2.7960) (-3.3786) (-1.1268) (-2.7116) PROF), 3.2633 2.4485 -1.4241 1.8301 (1.3620) (4.3735) (-0.6188) (2.6148) PROD2 2.6024 1.6674 2.2683 1.0202 (1.1488) (3.1582) (1.0446) (1.5480) PROD3 6.3182 3.3143 18.725 1.7864 (1.8573) (4A550) (6.1936) (1.9610) PROD4 -0.0397 0.3026 0.4529 -0.1917 (-0.0118) (0.3908) (0.1401) (-0.1973) PROD, -1.2867 -0.1679 0.6605 -0.1719 (-0.6885) (-0.3904) (0.3681) (-0.3186) REG, -10.294 -6.7833 -9.91 33 -4.7418 (-2.5331) (-7.0605) (-2.5791) (-3.9634) REG2 -6.5861 -4.1565 -7.0414 -2.3073 (-1.9986) (-5.4348) (-2.2365) (-2.4234) REG3 -5.4460 -3.6407 -3.0557 -2.5434 ( 1.5380) (-4.4394) (-0.8993) (-2.4809) REG4 -5.3100 -4.4456 1.9752 -3.1003 (-1.5268) (-5.4882) (0.5907) (-3.0648) REG5 -6.0546 -4A013 -2.6602 -3.1937 (-1.5554) (-4.8760) (-0.7126) (-2.8294) REG6 -7.0729 4.0936 -0.7028 -3.2133 (-1.7369) (-4.3426) (-0.1796) (-2.7205) REG7 -8.4692 -5.2892 -3.3973 -4.3181 (-2.1223) (-5.6912) (-0.8879) (-3.7032) REG8 -3.2430 -4.1279 1.5830 -2.7202 (-0.8669) (-4.7956) (0.4416) (-2.5326) CAP 6.9972 4.3558 3.3644 -0.2397 (3.4841) (9.6463) (1.9739) (-0.4686) TIME -0.2092 -1.3085 -3.3238 -1.2794 (-0.1234) (-3.3634) (-2.0401) (-2.6396) 0.555 0.379 0.077 0.068 The dependent variable is the appropriate pollution variable divided by 1000. 22 TABLE 7 NONRANDOM REPORTING EQUATION (Sampic size = 3496) Independent variables Coefficient t-stats CONSTANT 0.4720 4.959 CUMPt 0.3859 6.443 CAPi, 1 .5473 13.159 TIME 0.6967 1.689 TIME2 -0.8375 -2.127 LOG-LIKELIHOOD TEST STATISTIC FOR ZERO SLOPE COEFFICIENTS: 301.76 23 TABLE 8 EMISSIONS EQUATIONS INSTRUMENTAI, VARIABLE ESTIMATION] (Sample size - 2716) Independent Absolute dileIarges Vischargeie relative to norm Variables DOD TSS WOD TSS CONS1'ANT 5.4537 4.4666 1.7141 3.3552 (0.8149) (2.9382) (0.5576) (1.8645) Pu-12 0.8191 0.4095 0.1524 0.3922 (23.058) (8.8315) (1.2871) (7.4'nI) INS, -210.21 -47.558 -50.218 -58.061 (-1.8784) (-1.9156) (-0.8589) (-'.9095) INS.. -5.5509 -1.5839 -2.0651 -1.5872 (-2.4181) (-2.9996) (-1.5643) (-2.5080) PROD, 3.6189 2.6050 -.4602 1.9546 (1.4040) (4.2949) (-0.4645) (2.6977) PROD2 3.0048 1.8396 2.2286 1.1570 (1.2620) (3.2905) (1.2262) (I.6884) PROD3 6.1961 3.2663 18.742 1.7517 (1.9968) (4.3360) (4.5815) (1.9357) PROD4 -0.1722 0.2466 0.4663 -0.2378 (-0.4125) (2.0029) (1.5022) (-1.6863) PROD5 -1.3209 -0.1836 0.6641 -0.1846 (-1.0410) (-0.6257) (0.8492) (-0.5232) REG, -10.558 -6.9048 -9.8891 -4.8390 (-1.6845) (-4.8202) (-2.9635) (-2.9306) REG2 -6.6477 -4.1878 -7.0363 -2.3298 (-1.1857) (-3.2960) (-2.6874) (-1.5842) 2EG3 -5.9189 -3.8413 -3.0088 -2.7011 (-0.9838) (-2.8705) (-1.0849) (-1.7143) REG4 -5.7143 -4.6209 2.0158 -3.2379 (-0.9576) (-3.4598) (0.6825) (-2.0671) REG5 -6.1659 -4.4537 -2.6493 -3.2339 (-1.0384) (-3.3582) (-0.8748) (-2.0807) KEG6 -7.7194 -4.3666 -0.6382 -3.4294 (-1.1948) (-3.0655) (-0.2054) (-2.0291) REG7 -8.8416 -5.4508 -3.3605 -4.4467 (-1.4084) (-3.8940) (-1.0930) (-2.7076 REG2 -3.2857 -4.1539 1.5888 -2.7367 (-0.5512) (-3.1591) (0.5303) (-1.7704) CAP 8.1369 4.8395 3.2522 0.1314 (2.6350) (6.0190) (2.4415) (0.1652) TIME 0.4036 -1.2298 -3.3438 -12188 (0.2233) (-2.8422) (-3.2572) (-2.3644) RANDOM 4.3839 1.8109 -0.4348 1A387 (0.8985) (1.4787) (-0.0664) (0.9963) 0.519 0.321 0.079 0.058 The dependent varable is the appropriate pollution vaiable divided by 1000. 24 REEERENCFS Ackerman, B.A. and R.B. Stewart (1985), "Comment: Enforcing Environmental Law", Stanford Law Review, 37, 1333-1365. Anderson, L.G. and D.R. Lee (1986), "Optimal Governing Insrument, Operation Level, and Enforcement in Natural Resource Regulation: the Case of the Fishery", American Journal ofAgricultural Economics, 68 (3), 678-90. Association de l'lndustrie Forestiere du Quebec (1991), L'Industrie des Pdies et Papiers, Quebec. Beavis. B. and 1. Dobbs (1987), "Firm Behavior under Regulatory Control of Stochastic Envirom,nental Wastes by Probabilistic Constraints", Journal of Environmental Economics and Management, 15, 112-27. Canada, (1992), The Green Plan: a Frameworkfor Discussion on the Environment, Ministry of the Environment, Ottawa, Ontario. Cropper, M. and W.E. Oates (1992), "Environmental Economics: a Survey", Journal of Economic Literature, 30, 675-740. Deily, M.E. and W.B. Gray (1991), "Enforcement of Pollution Regulations in a D)eclining Industry", lournal of Environmental Economics and Management, 21, 260-74. Dewees, D.N. (1990), "The Effect of Environmental Regulation: Mercury and Sulphui Dioxide", in M.L. Friedland (eds), Securing Compliance: Seven Case Studies, Toronto, University of Toronto Press. Furlong, W.J. (1991), "The Deterrent Effect of Regulatory Enforcement in the Fishery", Land Economics, 67, 116-29. General Accounting Office (1987), Hazardous Waste: Facility Inspections are not thorough and complete, Report RCED-88-20, Washington, D.C. Harrington, W. (I988). "Enforcement Leverage when Penaltes are Restricted", Journal of Public Economics, 37,29-53. Heckman, J. (1979), "Sample Selection Bias as a Specification Error', Econometrica, 47, 153-61. Kennett, S.A. (1991), Managing Interjurisdictional Waters in Canada: a Constitutional Analysis, Canadian Institute of Resources Law, Faculty of Law, The University of Calgary, Calgary, Alberta. 25 Laplante, B. and P. Lanoie (1994), "The Market Response to Environmental Incidents in Canada: a Theorctical and Empirical Analysis", Southern Economic Journal, 60, 657-72. Lee, D.R. (1984), "The Economics of Enforcing Pollution Taxation", Journal of Environmental E:conomics and Management, 11, 147-60. Linder, S.H. and M.E. McBride, (1984), "Enforcement Costs and Regulatory Reform: the Agency and Firm Responses", Journal of Environmental Economics and Management, 11, 327-46. Magat, W.A. and W.K. Viscusi (1990), "Effectiveness of the EPA'S Regulatory Enforcement: the Case of Industrial Effluent Standards".,Journal ofLaw and Economics, 33, 331-60. Muoghalu, M.L., Robison, H.D. and J.L. Glascock (1990), "Hazardous Waste Lawsuits, Stockholders Returns, and Deterrence", Soulhern Economic Journal, 57, 3 57-70. Nemetz, P.N. (1986), "Federal Environmental Regulation in Canada", Natural Resources Journal, 26,357-70. Quebec (1990), Bilan Annuel de Conformite Environnementale - Secteur des Pdtes et Papiers, Ministere de l'Environnement du Quebec, Direction des programmes sectoriels, Qudbec. Russell, C.S. (1983), Pollution Monitoring Survey: Summary Report, Resources for the Future, Washington, D.C. Russell, C.S. (1990), "Monitoring and Enforcement", in P.R. Portney eds., Public Policies for Environmental Protection, Resources for the Future, Washington, D.C. Russell, C.S., Harrington, W. and W.J. Vaughan (1986), Enforcing Pollution Control Laws, Resources for the Future, Washington, D.C. Silverman, S.L. (1990), "Federal Enforcement of Environmental Laws", Massachusetts Law Review, 75, 95-8. Sinclair, W.F. (1991), "Controlling Effluent Discharges from Canadian Pulp and Paper Manufacturers", Canadian Public Policy, 17, 86-105. Sutinen, J. and P. Andersen (1985), "The Economics of Fisheries Law Enforcement", Land Economics, 61, 387-97. Tietenberg (1992), Innovation in Environmental Policy: Economic and Legal Aspects of Recent Developments in Environmental Enforcement and Liability, Edward Elgar. 26 Wasserman, C. (1984), Improving the Efficiency and Effectiveness of Compliance Monitoring and Enforcement of Environmental Policies, United States: A National Review, prepared for the Organization for Economic Cooperation and Development, Paris. White, H. (1980), "A Heteroskedasticity Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity", Economeirica, 48, 817-38. 27 Policy Research Working Paper Series Contact Title Author Date for paper WPS1425 On the Intersecloral Migraiion of Donald Larson February 1995 J Jacobson Agricultural Labor Yair Mundlak 33710 WPS1426 Russian Unemployment: Its Simon Commander February 1995 V. Reid Magnitude. Characteristics, and Ruslan Yemtsov 35195 Regional Dimensions WPS1427 Corporate Governance and Equity Stijn Claessens February 1995 F. Hatab Prices: Evidence Irom the Czech 35835 and Slovak Republics WPS1428 Short-Term Supply Response to a Bruno Boccara February 1995 M. Pfeiffenberger Devaluation: A Model's Implications Fabien Nsengiyumva 34963 for Primary Commodity-Exporting Developing Countries WPS1429 The World Trade Organization's Bernard M. Hoekman March 1995 F. Hatab Agreement on Govemment Petros C. Mavroidis 38535 Procurement: Expanding Disciplines, Declining Membership? WPS 1430 Intergovemmental Fiscal Relations Richard M. Bird March 1995 G. Coward and Poverty Alleviation in Viet Nam Jennie I. Litvack 80494 M. Govinda Rao WPS1431 The Industrial Pollution Projection Hemamala Hettige March 1995 A. Williams System Paul Martin 37176 Manjula Singh David Wheeler WPS1432 Using Financial Futures in Trading Ignacio Mas March 1995 K. Binkley and Risk Management Jesus SaA-Requejo 81143 WPS1433 Enterprise Restructuring in Eastern Bemard Hoekman March 1995 F. Hatab Europe: How Much? How Fast? Gerhard Pohl 35835 Where? Preliminary Evidence from Trade Data WPS1434 Government Credit Policy and Charles W. Calomirs March 1995 P. Infante Industrial Performance (Japanese Charles P. Himmelberg 37642 Machine Tool Producers, 1963-91) WPS1435 The Political Economy of Formal Arup Banerii March 1995 D. Evans Sector Pay and Employment in J. Edgardo Campos 37496 Developing Countries Richard H. Sabot WPS1436 An Empirical Model of Sunk Costs Mark J. Roberts March 1995 J. Ngaine and the Decision to Export James R. Tybout 37959 Policy Research Working Paper Series Contact Title Author Date for paper WPS1437 Travel Mode Substitution in Sao Joffre Swait March 1995 C. Jones Paulo: Estimates and Implications Gunnar S. Eskeland 37699 for Air Pollution Control WPS1438 Trade Refomn. Efficiency, and Growth Ejaz Ghani March 1995 A. Nokhostin Carl Jayarajah 34150 '^'PS1439 Nontariff Barriers Africa Faces: What Azita Amjadi March 1995 S. Upscomb Did the Uruguay Round Accomplish, Alexander Yeats 33718 and What Remains to Be Done? WPS1440 Poverty and Social Transfers in Christiaan Grootaert March 1995 N. Sachdeva Poland 82717 WPS1441 The Significance of Credits and Douglas Galbi March 1995 N. Castillo Subsidies in Russian Agricultural 33490 Reform WPS1442 Energy Price Increases in Einar Hope March 1995 C. Jones Developing Countries: Case Studies Balbir Singh 37699 of Colombia. Ghana, Indonesia, Malaysia, Turkey, and Zimbabwe WPS1443 Policy-Based Finance, Financial Dimitri Vittas April 1995 P. Infante Regulation, and Financial Sector Akihiko Kawaura 37642 Development in Japan WPS1444 Roads, Lands, Markets, and Kenneth M. Chomitz Apri 1995 E. Schaper Deforestation: A Spatial Model David A. Gray 33457 of Land Use in Belize WPS1445 Human Capital and Industry Wage Chris N. Sakellarou April 1995 I. Conachy Structure in Guatemala 33669 WPS1446 Review of Integrated Approaches Donna J. Lee April 1995 C. Spooner to River Basin Planning, Ariel Dinar 32116 Development, and Management WPS1447 Environmental Inspections and Benoit Laplante April 1995 E Schaper Emissions of the Pulp and Paper Paul Rilstone 33457 Industry: The Case of Quebec WPS1448 Environmental Regulation and Susrnita Dasgupta April 1995 E. Schaper Development: A Cross-Country Ashoka Mody 33457 Empirical Analysis Subhendu Roy David Wheeler