Environ Resource EconDOI 10.1007/s10640-011-9465-y

Does Disclosure Reduce Pollution? Evidence fromIndia’s Green Rating Project

Nicholas Powers · Allen Blackman · Thomas P. Lyon ·Urvashi Narain

Accepted: 15 March 2011© Springer Science+Business Media B.V. 2011

Abstract Public disclosure programs that collect and disseminate information about firms’environmental performance are increasingly popular in both developed and developing coun-tries. Yet little is known about whether they actually improve environmental performance,particularly in the latter setting. We use detailed plant-level survey data to evaluate the impactof India’s Green Rating Project (GRP) on the environmental performance of the country’slargest pulp and paper plants. We find that the GRP drove significant reductions in pollutionloadings among dirty plants but not among cleaner ones. This result comports with statisticaland anecdotal evaluations of similar disclosure programs. We also find that plants located inwealthier communities were more responsive to GRP ratings, as were single-plant firms.

Keywords Public disclosure · Pollution control · India · Pulp and paper

JEL Classification Q53 · Q56 · Q58

This research was undertaken while Nicholas Powers was a doctoral student at the University of Michigan.

N. Powers (B)The Brattle Group, 1850 M St. NW, Washington, DC 20010, USAe-mail: Nichlolas.Powers@brattle.com

A. BlackmanResources for the Future, Washington, DC, USA

A. BlackmanEnvironment for Development Center for Central America, Centro,Agronómico Tropical de Investigación y Enseñanza (CATIE), Turrialba, Costa Rica

T. P. LyonUniversity of Michigan, Ann Arbor, MI, USA

U. NarainThe World Bank, Washington, DC, USA

123

N. Powers et al.

1 Introduction

Programs that collect and disseminate information about firms’ environmental performancehave been characterized as the “third wave” in environmental regulation, after command-and-control and market-based approaches (Tietenberg 1998). Two types of national publicdisclosure programs have emerged over the past two decades (Dasgupta et al. 2007).So-called pollutant release transfer registries simply report emissions data without usingthem to rate or otherwise characterize environmental performance. More than 20 countrieshave set up such registries.1 Like the seminal US Toxic Release Inventory, most focus ontoxic pollutants not covered by conventional regulations.

The second type of national public disclosure program is called a performance evaluationand ratings program (PERP). Such programs both report emissions data and use them torate plants’ environmental performance. These programs are confined to developing coun-tries and focus mostly on conventional pollutants. Examples include Indonesia’s Programfor Pollution Control, Evaluation, and Rating (PROPER), which was the first such programto appear and is the best known; India’s Green Rating Project; the Philippines’ EcoWatchprogram; China’s GreenWatch program; and Vietnam’s Black and Green Books initiative.These programs have been touted as a means of circumventing perhaps the most dauntingobstacle to pollution control in developing countries: weak environmental regulatory insti-tutions. Public disclosure does not necessarily require an effective enforcement capability oreven a well-defined set of environmental regulations. Furthermore, the costs of the adminis-trative activities it does require—data collection and dissemination—are declining thanks toinformation technologies (Dasgupta et al. 2007).

Notwithstanding the promise and growing popularity of public disclosure, we know littleabout whether it actually improves environmental performance, in either industrialized ordeveloping countries. As discussed in the next section, most evaluations to date have beenanecdotal, and only a few rigorous analyses have appeared.

To help fill this gap, this paper evaluates the impact of India’s Green Rating Project (GRP)on the environmental performance of the country’s largest pulp and paper plants. To ourknowledge, it is the first rigorous analysis of the GRP’s environmental impact and only thesecond such evaluation of a developing country public disclosure program.2 To identify theeffect of the GRP, we control for other factors that drive cross-sectional and intertemporalvariation in pollutant discharges using exceptionally detailed plant-level data, including bothprimary survey and secondary census data.

Our analysis suggests that the GRP drove significant reductions in pollution loadingsamong dirty plants but not among cleaner ones. This result comports with statistical andanecdotal evaluations of similar disclosure programs (Dasgupta et al. 2007; García et al.2007, 2009). We also find that plants in wealthier communities were more responsive toGRP ratings, as were single-plant firms.

The remainder of the paper is organized as follows. Section 2 reviews the literatureon public disclosure and presents a conceptual framework for our empirical analysis. Sec-tion 3 provides background information on the GRP and the Indian pulp and paper industry.

1 Countries that have at least initiated a web-accessible pollution release transfer registry include Austria,Australia, Canada, Chile, the Czech Republic, Denmark, England, France, Germany, Hungary, Italy, Japan,Mexico, the Netherlands, Norway, Scotland, South Korea, Spain, and Sweden (Dasgupta et al. 2007; Kerretand Gray 2007).2 To the best of our knowledge, the only other developing country program that has been rigorously evaluatedis PROPER; see García et al. (2007, 2009) for details.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

Section 4 describes the empirical framework and the data used in the regression analysis.Section 5 presents our econometric results, and Section 6 concludes.

2 Literature Review and Conceptual Framework

This section briefly reviews the empirical literature on public disclosure programs, focusingon two questions: do they improve environmental performance, and if so, how? Then, draw-ing on the literature, it presents a heuristic graphical model of public disclosure to underpinthe econometric analysis.

2.1 Does Public Disclosure Improve Environmental Performance?

Only a few papers have evaluated PERPs like the GRP, but all have concluded that theprograms have generated environmental benefits. Of these papers, to our knowledge, onlytwo—García et al. (2007, 2009)—present rigorous statistical analyses. Using panel data onboth participants and nonparticipants, the authors find that Indonesia’s PROPER spurredsignificant emissions reductions in wastewater discharges, particularly among plants withpoor compliance records. Dasgupta et al. (2007) present a largely qualitative evaluation ofPERPs in Indonesia, the Philippines, China, and Vietnam. In each program, the authors findthat a large number of plants initially rated “noncompliant” improved to “compliant” overtime (although plants rated “flagrant violators” and “compliant” tended to remain in thesecategories). However, as the authors themselves point out, without any type of baseline orcontrol group, it is not possible to attribute observed changes in compliance to public disclo-sure. These changes may have resulted from any number of contemporaneous confounders,including ratcheting up of formal regulatory pressure and the diffusion of clean technolo-gies. Wang et al. (2004) provide a more detailed but still primarily anecdotal evaluation ofthe Chinese PERP that suggests it succeeded in improving environmental performance.

Several recent papers evaluate public disclosure initiatives other than PERPs and also findsignificant effects on environmental performance. Bennear and Olmstead (2008) presentevidence that a 1996 amendment to the US Safe Drinking Water Act requiring communitydrinking water systems to publicly report regulatory violations reduced the incidence ofsubsequent violations. Similarly, Delmas et al. (2007) show that regulations requiring USelectric utilities to mail bill inserts to consumers reporting the extent of their reliance onfossil fuels led to a significant decrease in fossil fuel use. And Foulon et al. (2002) find thata policy of publicly disclosing the identity of plants that are noncompliant or “of concern”spurred emissions reductions in a sample of pulp and paper plants in British Columbia.

Finally, a number of papers have examined the US Toxic Release Inventory (Konar andCohen 1996; Khanna and Anton 2002; Bui 2005; Koehler and Spengler 2007). Since theprogram began in 1986, total reported releases of the toxics it covers have fallen by at least45%. However, it is not clear that public disclosure has been responsible for this decline.Data on toxic releases are not available for the period preceding the program, or for plantsthat fall outside the program, and as a result, the usual means of estimating releases absentthe program are not available (Bennear and Olmstead 2008).3

3 Moreover, several papers propose alternative explanations for the observed reductions in toxic releases,including the imposition of more stringent conventional regulation (Bui 2005), plants’ practice of substitut-ing unlisted toxics for listed ones (Greenstone 2003), and simple underreporting of emissions (Koehler andSpengler 2007).

123

N. Powers et al.

2.2 How Might Public Disclosure Improve Environmental Performance?

Tietenberg (1998) lists seven hypothetical “channels” through which public disclosure mayspur improved environmental performance. To simplify the exposition, we group them intofour broad categories. Each is designated with a single lowercase letter that corresponds tothe notation used in the analytical model presented in the next subsection:

(g) output market pressures—disclosure may affect the demand for firms’ goods;(k) input market pressures—disclosure may affect the demand for firms’ securities and

their ability to hire and retain employees;(j) judicial pressures—disclosure may encourage private citizens to sue polluters; and(r) regulatory pressures—disclosure may build support for new pollution control legisla-

tion or more stringent enforcement of existing legislation.

Based on the literature discussed below, we add two more categories:

(c) community pressures—disclosure may enhance pressures from community groups andnongovernmental organizations; and

(t) managerial information—disclosure may provide new information to managers abouttheir plants’ discharges and options for reducing them.

Empirical research aimed at disentangling and identifying the effects of these channels islimited. To date, to our knowledge, such research has only demonstrated the effect of three:(k) input market pressure, (r) regulatory pressure, and (t) managerial information.

Most research in this area has focused on channel (k), input market pressure, specificallystock market pressure. For our purposes, the most relevant paper is Gupta and Goldar (2005),which tests whether GRP ratings affect the stock prices of Indian companies in the pulp andpaper, chlor-alkali, and automobile sectors, three of the four sectors that have been ratedby the GRP. They find that poor GRP ratings did in fact spur significant negative abnormalreturns. Several other papers examine the same broad question but focus on types of disclo-sure other than PERPs, namely pollutant release transfer registries and ad hoc announcementsof environmental accidents and regulatory violations. These papers also find that public dis-closure affects stock prices (Dasgupta et al. 2001, 2005; Laplante and Lanoie 1994). Theaforementioned papers test whether disclosure affects stock prices but do not attempt toassess whether changes in stock prices, in turn, affect firms’ pollution control decisions.However, both Konar and Cohen (1996) and Khanna et al. (1998) test for such effects, andboth conclude that they are significant.

Regarding channel (r), regulatory pressure, García et al. (2009) analyze the correlationbetween the characteristics of firms participating in Indonesia’s PROPER and changes intheir ratings over time. They find that foreign-owned plants have been more responsive toPROPER ratings and attribute this result partly to these plants’ lack of close connections toruling elites, which would immunize them from regulatory pressure.

Finally, one paper finds evidence for channel (t), managerial information. Blackman etal. (2004) surveyed managers of plants participating in PROPER to shed light on the impor-tance of the six channels listed above. Somewhat surprisingly, the responses suggested thatPROPER ratings spur abatement mainly by improving poorly performing plants’ informationabout their own pollution and abatement opportunities, a phenomenon the authors term theenvironmental audit effect.

Beyond the studies cited thus far, empirical research on how public disclosure per se hasan impact is quite limited. However, the literatures on “voluntary regulation” and “informalregulation” are relevant because they examine the effect on firms’ abatement decisions of

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

many of the same channels, including pressures generated by regulators (r), output markets(g), courts (j), and communities (c). The literature on voluntary regulation examines the driv-ers of firms’ decisions to overcomply with mandatory environmental regulation (see Lyonand Maxwell 2002; Khanna 2001 for reviews). Research has focused on the effects of regu-latory pressure (Maxwell et al. 2000; Sam and Innes 2006), output market pressure (Aroraand Gangopadhayay 1995) and judicial pressure (Videras and Alberini 2000; Vidovic andKhanna 2007). The literature on informal regulation focuses on pressures to abate generatedby private sector agents in developing countries where state regulation is weak (see Blackman2010 for a review). For example, Pargal and Wheeler (1996) examine the environmental per-formance of plants in Indonesia at a time when regulatory enforcement was negligible (beforePROPER was initiated) and find that emissions were lower in communities with higher percapita income and higher levels of education, implying that such communities effectivelypressure plants to abate (see also Blackman and Bannister 1998).

2.3 Conceptual Framework

This section presents an heuristic graphical model of public disclosure to underpin the econo-metric analysis (see Appendix 1 for an analytical version of the model). It draws on the stan-dard representation of a plant’s abatement decision in the environmental economics literature(see, e.g., World Bank 1999). We assume marginal abatement costs (MAC) are increasingin abatement while private marginal abatement benefits (MAB) are decreasing in abatement(Fig. 1). Note that MAB are private benefits, not the social benefits used in many graphicalanalyses of abatement. They arise from the pecuniary and nonpecuniary penalties levied onthe plant by regulators (e.g., fines), communities (e.g., reputation costs), consumers (e.g., lostmarket share), capital markets (e.g., lower stock prices), and other stakeholders. The plantchooses the level of abatement, α∗, such that MAC equals MAB.

Drawing on the literature surveyed in Sect. 2, we define six channels through which publicdisclosure might affect the firm’s abatement decision. The first five channels have to do withthe costs that can be imposed on dirty firms by green consumers (g), input (capital and labor)markets (k), courts (j), regulatory authorities (r), and communities (c). The sixth channel is

$

(MAC,MAB)

MAC1

MAC2

MAB1

MAB2

*1 *2α α α

Fig. 1 Marginal abatement cost (MAC) and marginal abatement benefit (MAB) schedules; optimal abatementlevel, α∗

123

N. Powers et al.

related to the costs of pollution abatement arising from the plant’s need to acquire informationabout abatement technologies and its own pollutant discharge (t). We use Fig. 1 to illustratehow each channel might affect the firm’s abatement decision. First, public disclosure couldeither reduce or enhance consumers’ demand for the firm’s output (g), depending on whetherthe firm is relatively clean or dirty. Either effect, in turn, implies that the marginal benefitto the firm of cutting discharges will be greater regardless of the actual level of abatement.For a clean firm, disclosure enhances its demand, but for a dirty firm, disclosure decreasesits demand. Graphically, in either case, the MAB curve shifts up, and in equilibrium the firmchooses a higher level of abatement. Similarly, depending on whether the plant is dirty orclean, public disclosure could either raise or lower costs imposed by regulators (r), input mar-kets (k), communities (c), and courts (j). Once again, regardless of whether the plant is cleanor dirty, each of these effects shifts the MAB curve up and results in a higher equilibriumlevel of abatement. Finally, public disclosure could reduce the cost to firms of acquiringinformation about abatement (t), lowering the marginal cost at every level of abatement.In this case, graphically, the MAC curve shifts down, and the end result is a higher level ofabatement.

3 Background

3.1 Green Rating Project

The Centre for Science and Environment (CSE), one of India’s best-known and most influen-tial environmental nongovernmental organizations, began work on the GRP in 1997. Accord-ing to CSE background materials, the program was urgently needed to shore up India’s weakenvironmental regulatory institutions and was inspired by the Council on Economic Prior-ities, a now-defunct US nongovernmental organization that provided investors with annualratings of the environmental performance of US companies.

To date, the GRP has rated the environmental performance of large plants in four pol-lution-intensive industrial sectors: pulp and paper, chlor-alkali, cement, and automobiles.Plants in the pulp and paper sector have been rated twice (once in 1999 and again in 2004);plants in the other three sectors have been rated just once. In each rating, plants are assigned anumerical score from 0 to 100 and are awarded symbolic “leaves” depending on their score:five leaves for scores of 75 and above, four for 50–74, three for 35–49, two for 25–34, onefor 15–24, and none for 14 and below. The GRP scores are based on an evaluation of theplant’s life-cycle environmental impacts, from the sourcing and processing of raw materialsto the manufacture, use, and disposal of products. The exceptionally detailed data neededto conduct this cradle-to-grave analysis are collected from questionnaires administered toparticipating plants, along with secondary data provided by local environmental regulatoryinstitutions and other sources. Both the questionnaires used to collect the data and the meth-odology used to analyze them were designed by a panel of leading technical experts in eachrated sector. In addition, to ensure objectivity and transparency, the entire GRP program issupervised by a panel comprising high-level representatives of industry, government, thejudiciary, academia, and nongovernmental organizations. Lastly, self-reported data from thefirms are carefully checked by GRP inspectors and compared with the secondary data.

In addition to informing the public about plants’ environmental performance, the GRPalso informs plants about their pollution and pollution abatement options. The programuses the primary and secondary data it collects to construct a detailed environmental pro-file of each plant and sends it to the facility for review before releasing the ratings to

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

the public. The program also publishes specific recommendations for improving environ-mental performance in each sector. Finally, the ratings are released at a high-profile publicevent by leading public figures. For example, some ratings were released by the late K. R.Narayanan, India’s former president, and some by Manmohan Singh, India’s current primeminister.

3.2 Pulp and Paper Sector

Pulp and paper is a notoriously dirty industry. The environmental performance of millsin North America and Scandinavia, which tend to be much larger and more modern thantheir developing-country counterparts, has improved considerably over the past few decades.Indian mills lag behind, with per unit measures of industrial pollution 5–10 times higher(CSE 2004).

Making paper involves four main steps, all of which generate water pollution: raw mate-rial processing, pulping, bleaching, and papermaking.4 To reduce pollution loadings, plantsundertake both pollution control and prevention. All the plants rated by the GRP have waste-water treatment plants. In addition, to prevent pollution, mills have eliminated particularlydirty inputs, adopted good “housekeeping” measures, improved chemical recovery systems,and modified the pulping process.

As previously mentioned, the pulp and paper sector is the only sector that GRP ratedtwice. As a result, survey data on environmental impact indicators and other relevant vari-ables are available for several years before and after the first (1999) rating. This allows us toconstruct a counterfactual—that is, an estimate of what pollution would have been absent theprogram—needed to identify the impact of the GRP. Because the necessary data on environ-mental indicators and firm behavior are not available after the second rating, we can analyzethe effectiveness of only the first rating.

The manner in which GRP selected pulp and paper plants for rating and collected therequisite data determined the composition of our regression sample. GRP’s only criterionfor including pulp and paper plants in their rating program was facility size. All 28 plantsin India with a production capacity exceeding 100 tons per day in fiscal year 1998 wereinvited to participate.5 (India’s fiscal year ends on March 31; all years referred to in thispaper and accompanying figures are fiscal years). Although GRP is nominally a voluntaryinitiative, the 28 plants were informed that a failure to participate would result in a defaultrating of zero leaves. Ultimately, all invitees chose to participate. Collectively, these 28 plantswere responsible for 59% of pulp and paper production in India. They were contacted forthe survey toward the end of fiscal year 1998 and asked to provide retrospective annualdata—information they normally track and record in any case—for 1996 through 1998. Theratings for these plants were released by Manmohan Singh, India’s current prime minister,on July 18, 1999. All major Indian daily newspapers covered the event. GRP conducted asecond rating of pulp and paper plants in 2004. It was based on retrospective responses to a2003 GRP questionnaire covering the years 1999–2003.6 Of the 28 plants rated in 1999, 22were also rated in 2004. (For the remainder of the paper we will refer to years 1996–1998as “predisclosure” and 1999–2003 as “postdisclosure.”) We restrict our regression sample

4 See CSE (2004) or Schumacher and Sathaye (1999) for more detail.5 By way of comparison, the average plant in the United States has a capacity of nearly 600 tons/day (CSE2004). GRP’s size criteria excluded more than 500 smaller plants.6 Of the six plants that were rated in the first period but not the second, five were permanently closed and onewas temporarily closed after the first rating.

123

N. Powers et al.

Table 1 1999 and 2004 GRP ratings for pulp and paper plants in empirical analysis; number of plants in eachrating categorya

1999 rating

All 0-leaf 1-leaf 2-leaf 3-leaf 4-leaf 5-leaf

2004 rating All 22 0 7 12 3 0 0

0-leaf 0 0 0 0 0 0 0

1-leaf 7 0 5 2 0 0 0

2-leaf 11 0 2 8 1 0 0

3-leaf 4 0 0 2 2 0 0

4-leaf 0 0 0 0 0 0 0

5-leaf 0 0 0 0 0 0 0

a Includes BILT Graphics, known as Sinar Mas Pulp and Paper in 1999, which received a score in the firstrating but not a leaf categorization. Because BILT’s score was in the three-leaf range in 1999, it is includedhere as such

to the 22 plants rated in both 1999 and 2004 and for which the full 8 years’ panel data areavailable (because these plants completed GRP surveys for 1996–2003).

Table 1 presents the 1999 and 2004 ratings for the 22 pulp and paper plants in our regressionsample. Although the GRP’s scoring system allows plants to receive zero to five leaves, all 22sample plants received one, two, or three leaves in both ratings; that is, no plants received zero,four, or five leaves. In both ratings, the majority of plants—at least 11—received two leaves.

3.3 Trends in Ratings and Pollution Indicators

The first goal of our empirical analysis is to determine whether GRP ratings had a significantimpact on our sample plants’ pollutant loadings between 1996 and 2003. Simple summarystatistics suggest that ratings may have spurred reductions for the lowest-rated plants. One-leaf plants were the only category that demonstrated unambiguous improvement between1999 and 2004 (Table 1). Of the 7 plants that received one leaf in 1999, 2 received a betterrating and none received worse ratings in 2004. However, of the 12 plants that received twoleaves in 1999, 2 earned worse ratings and 2 got better ratings in 2004. Finally, of the 3 plantsthat received three leaves in 1999, 1 earned a lower rating and none received a better ratingin 2004.

Average annual wastewater discharge data for 1996–2003, as shown in Figs. 2 and 3, tellan analogous story. They suggest that the environmental performance of particularly dirtyplants—those that received one leaf in the 1999 rating—did in fact improve significantly dur-ing 1999, the year of the first GRP rating. Figure 1 shows trends in chemical oxygen demand(COD), and Fig. 2 shows trends in total suspended solids (TSS)—two common measures ofwater pollution.7 Both pollutants declined dramatically after 1998 for plants that received aone-leaf rating, but not for plants with higher ratings.

7 COD is a measure of the amount of oxygen needed to fully oxidize the organic compounds in water tocarbon dioxide; it is a linear function of the chemical composition of the organic pollutants in a sample ofwater. TSS is the dry weight of particles suspended in a sample of water and, like COD, is typically expressedin mg/L. To properly scale for water use, we convert both measures and express them as kg per bone-drymetric ton (bdmt) of product. This is done by multiplying first by a constant, then by the amount of water usedto produce a bdmt of product.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

20

40

60

80

100

2/3 leaves

1 leaf

All

20032002200120001999199819971996

Fiscal Year

CO

D (

kg/B

DM

T o

f pr

oduc

t)

Fig. 2 Annual unweighted average discharges of chemical oxygen demand (COD) from 22 pulp and paperplants participating in India’s Green Ratings Project, by performance rating

5

10

15

20

25

30

2/3 leaves

1 leaf

All

20032002200120001999199819971996

Fiscal Year

TSS

(kg

/BD

MT

of

prod

uct)

Fig. 3 Annual unweighted average discharges of total suspended solids (TSS) from 22 pulp and paper plantsparticipating in India’s Green Rating Project, by performance rating

The fact that the most dramatic reductions in discharges took place in 1999, the year thatthe ratings were released, suggests that the rating prompted improvements in the environmen-tal performance of poorly performing plants. These reductions, however, could have beencaused by contemporaneous changes in technological, market, and regulatory conditions thatinfluence the plants’ pollution loadings. For example, they could have been caused by reduc-tions in the relative price of cleaner inputs that happened to coincide with the release of theGRP ratings.

123

N. Powers et al.

4 Empirical Framework

The first goal of our empirical analysis is to determine whether the declines in COD and TSSlevels depicted in Figs. 2 and 3 were caused by the GRP and not by confounding factors. Inaddition, if we find that the GRP was, in fact, responsible, we seek to identify the channelsthrough which it had this effect.

4.1 Econometric Model

4.1.1 GRP’s Impact

To identify the effect of a policy like GRP ratings, a common strategy is to include in theanalysis both “treatment” entities affected by the policy and similar “control” entities notaffected by it. In our case, the control group would consist of nonparticipating pulp and paperplants that were similar to those that participated in GRP in 1999. We are not able to use thisapproach, however, because no such plants exist. As noted above, the GRP was restrictedto plants with a production capacity exceeding 100 tons per day in 1998, and every plant inIndia meeting this criterion participated. Using smaller plants to construct a control groupwould conflate effects on pollution loadings of GRP participation and plant size. Therefore,we employ a different strategy to identify the effect of GRP ratings: we use the exceptionallydetailed panel data that GRP compiles on all program participants to econometrically controlfor all factors other than GRP that might drive cross-sectional and intertemporal variation inpollutant discharges.8 Toward that end, we run separate fixed-effects ordinary least squares(OLS) panel-data models for COD and TSS of the form

yit = αi + β1 TREND + β2 POSTGRPt + β3 TREND ∗ ONELEAFi

+ β4 POSTGRPt ∗ ONELEAFi + β5xit + uit (1)

where yit is a measure of COD or TSS, αi is the plant fixed effect, TREND is a linear timetrend, POSTGRPt is a dummy variable that takes the value one for the postdisclosure period(1999–2003) and zero otherwise, ONELEAFi is a dummy variable for plants that receiveda one-leaf rating, and xit is a vector of confounding factors, including plant characteris-tics, input prices, and measures of regulatory pressure (Table 2).9 POSTGRPt is includedto capture the effect of the 1999 GRP rating. TREND allows us to control for exogenousimprovements in technology that could affect pollution levels before and after the informationdisclosure.10 (We are not able to include fixed-year effects to control for this because year

8 The goal of our analysis is limited to identifying the impact of GRP ratings on participating plants, whichby design are the country’s largest facilities. We do not purport to shed light on how public disclosure mightaffect smaller plants not in our sample. Therefore, the representativeness of our sample is not a concern.9 Although we expect the error terms in the COD and TSS regressions to be correlated, we do not use aseemingly unrelated regression (SUR) approach because there is no reason, a priori, to omit any explanatoryvariables from either of the two regressions. When the same explanatory variables appear in both regressions,SUR is equivalent to OLS (Greene 2002) and yields no efficiency gain. The only advantage of system OLS(SOLS) in this case is that it allows for joint hypothesis testing across the two regressions. The disadvantageis that SOLS routines in standard statistical packages do not allow for clustered standard errors. SOLS results(available from the authors upon request) comport with the OLS results presented in Tables 2 and 3: althoughthe individual coefficients of interest are not always significant, they have the same sign, and testing the jointsignificance of these coefficients in both equations rejects the null hypothesis that they are jointly zero.10 The choice of a linear trend term, as opposed to other functional forms, is based on preliminary regres-sions using predisclosure data. In these regressions, available from the authors upon request, the addition of aquadratic term adds little explanatory power.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

dummies would be perfectly collinear with POSTGRPt.) ONELEAF is used to identify theeffect of GRP ratings on dirty versus clean plants.11 TREND * ONELEAF allows exogenousimprovements in technology (both before and after the disclosure) to differ between dirtyplants and clean plants. Finally, POSTGRPt * ONELEAF allows the effect of POSTGRPt

to differ between dirty and clean plants.12 Note that because ONELEAF is time-invariant,it is perfectly correlated with our plant fixed effects and therefore cannot be included inour fixed-effects regressions (except as multiplicative terms in an interaction variable). Toallow for heteroskedasticity across plants and arbitrary serial correlation within each plant,we report standard errors that are clustered by plant (Wooldridge 2002).13

4.1.2 GRP Channels

To analyze the channels through which the GRP potentially affects firm behavior, we runseparate models for COD and TSS of the form

yit = αi + β1 TREND + β2 POSTGRPt + β3 TREND ∗ ONELEAFi

+ β4 POSTGRPt ∗ ONELEAFi+β5xit+β6TREND ∗ zi+β7POSTGRPt ∗ zi+uit

(2)

where zi is a vector of time-invariant plant and community variables, such as whether theplant is part of a conglomerate and the level of community wealth. Hence, our strategy foridentifying the channels through which GRP ratings have an effect is similar to the one weuse to test whether one-leaf plants were more responsive to ratings: we include a vector ofregressors, each of which interacts a dummy identifying the postrating years in our panel withvarious time-invariant plant or community characteristics (POSTGRPt ∗ zi).14 In addition,we control for the possibility that these plant and community characteristics affect the slopeof the trend term by including a second set of regressors that interact these characteristicswith our trend variable (TREND ∗ zi). These pairs of interaction terms allow us to determinewhether plants with certain characteristics and in certain communities responded differentlyto the program. This determination, in turn, sheds light on the channels through which GRPhas an effect.

11 We include a dummy identifying one-leaf plants—not plants in other ratings categories—for two reasons.First, as discussed above (Table 1), none of the our sample plants received zero, four, or five leaves in either the1999 or the 2004 rating. Second, including a dummy for one-leaf plants, thereby defining two- and three-leafplants as the reference group, elucidates the impact of GRP ratings on particularly dirty (one-leaf) plants—amajor theme in the literature.12 We also experimented with the possibility that the disclosure program induced a persistent increase inthe rate of environmental progress (which would be consistent with a change in the slope of the trend term)rather than in a one-time shift of the equilibrium path (consistent with a one-time decrease in the interceptterm). The most obvious way to identify such an effect would be to include an additional trend term that takesonly positive values after disclosure. However, given the sample size and a high degree of multicollinearitybetween the postdisclosure trend term and the original trend term, this renders identification difficult. Fur-thermore, regression results (available from the authors upon request) suggest that the one-time drop is moreconsistent with the data than is a change in the trend. Accordingly, we proceed without the postdisclosuretrend variable.13 Although the inclusion of fixed plant effects allows for heterogeneity in the form of a mean plant-specificunobservable component, we still need to allow the variance to differ across plants. Also, although it is rea-sonable to assume independence of the error terms across plants, we cannot, a priori, rule out persistence inthe error term within a plant.14 Some of the characteristics we use to construct these interaction terms, such as EFFRIVER, display someminor variation over time. For these, we calculate the predisclosure average and interact it with the postdis-closure dummy to create the corresponding interaction term.

123

N. Powers et al.

Table 2 Variables in econometric analysis and descriptive statistics

Variable Description Mean S.D. Min. Max.

AGLABOR Percentage of municipality workers in agriculture 0.17 0.21 0.002 0.89

CASTE Percentage of municipality population belonging toscheduled caste or tribe

0.23 0.16 0.02 0.73

COD Natural log of chemical oxygen demand (kg/bdmt) 3.38 0.90 0.04 5.43

COMPTOWN Percentage of full-time nonagricultural laborers insubdistrict working for plant

0.32 0.30 0.002 1

COMWATER Formal complaints to regulatory authorities (0/1) 0.53 0.50 0 1

CONS_ORIENT Average percentage of plant’s output final product,1995–1997

0.61 0.35 0 1

DISPOSAL Average percentage of effluents discharged directly intoriver, 1995–1997

0.69 0.41 0 1

EFFRIVER Percentage effluents discharged into river (vs. land, etc.) 0.69 0.41 0 1

ENFWATER Faced water pollution related regulatory enforcementactions (0/1)

0.15 0.36 0 1

EXP_ORIENT Average percentage output sold foreign markets1995–1997

0.05 0.05 0 0.20

FINALGOOD Percentage of output that is final (vs. intermediate)product

0.62 0.35 0 1

FORSHARE Percentage of sales derived from exports 0.05 0.05 0 0.20

GOV Government-owned (0/1) 0.24 0.43 0 1

LITERACY Percentage literate of municipality population 0.65 0.14 0.19 0.84

ONELEAF Received one leaf in 1999 rating (0/1) 0.31 0.47 0 1

PCTBAMBOO Percentage of fiber inputs from bamboo 0.29 0.34 0 1

PCTGRASSES Percentage of fiber inputs from grasses 0.02 0.08 0 0.56

PCTPULP Percentage of fiber inputs from market pulp 0.10 0.22 0 1

PCTRECYCL Percentage of fiber inputs from recycled materials 0.11 0.21 0 1

PERMIT Water effluent permit (0/1) 0.87 0.34 0 1

PERMIT2 Up-to-date water pollution permits 1995–1997 (0/1) 0.74 0.44 0 1

POST-GRP 1999 or later (0/1) 0.69 0.46 0 1

PRICE_CL Inflation-adjusted national average price of chlorine 1.00 0.11 0.83 1.25

PRICE_NAOH Inflation-adjusted national average price of sodiumhydroxide

0.84 0.13 0.68 1.08

PRICE_WAGES Inflation-adjusted national average daily wages fornonagricultural unskilled workers

0.35 0.02 0.32 0.39

PRICE_WOOD Inflation-adjusted national average price of wood 1.16 0.15 0.96 1.41

SCALE Natural log of output (kg/bdmt) 11.17 0.55 9.84 12.21

SINGLE Stand alone plant (0/1) 0.42 0.50 0 1

TREND Linear time trend term (1996 = 1, 1997 = 2, …, 2003 = 8) 4.74 2.23 0 8

TSS Natural log of total suspended solids (kg/bdmt) 2.13 1.07 −2.74 4.08

URBANPCT Percentage of population in subdistrict living inmunicipalities classified as urban

0.36 0.26 0 1

WEALTH Percentage households in subdistrict that own mopeds 0.42 0.20 0.12 0.79

Instead of including all of our POSTGRPt ∗ zi and TREND ∗ zi interaction terms inthe same model, we include pairs corresponding to a single characteristic (e.g., TREND *WEALTH and POSTGRP * WEALTH) in separate models. There are three reasons. First,several of the interaction variables are highly collinear (in part because all take a value of zerofor the three predisclosure years). In addition, including several sets of interaction variablesin the same regression creates a degrees-of-freedom problem. Finally, we are more interested

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

in identifying channels by which disclosure has an effect than in identifying the strongestamong several closely related channels.

Note that just as ONELEAF is time invariant and therefore cannot be included in our fixed-effects regressions (except as multiplicative terms in an interaction variable), the same is trueof the community and plant characteristic variables. This does not pose a major problem,however, since (unlike Pargal and Wheeler 1996) we are not interested in the direct effects ofplant and community variables on plants’ emissions. Rather, our goal is to determine whetherthey mediate the effect of public disclosure on plants’ abatement decisions.

4.2 Data

Most of our data are taken from GRP surveys. In addition, we use data from: the 2001 Indiancensus to construct proxies for several community characteristics; Prowess, an on-line busi-ness database, to construct company-level (versus plant-level) variables and a regional energyprice index; and Indiastat, an on-line database of Indian statistics, to construct price indicesfor several inputs used in pulp and paper manufacturing.15 The independent and dependentvariables used in the regressions are described below. See Table 2 for summary statistics.

4.2.1 Environmental Performance

Because all 22 plants in the sample have effluent treatment plants, COD and TSS measure-ments reflect posttreatment, end-of-pipe quantities. In India, as in most countries, regulatorystandards for these pollutants are specified in milligrams per liter of effluent. In India, how-ever, where water is inexpensive, the use of this metric creates an incentive for plants to dilutetheir liquid discharges. To control for this effect, we measure COD and TSS in kilograms (kg)per bone-dry metric ton (bdmt) of pulp and paper product. Our resulting measure, kg/bdmt,gives water pollution per unit of product produced. Finally, because we are interested in theresponses of both clean and dirty plants, we use logged values of both dependent variablesso that each regression coefficient measures the relative change in the dependent variable fora given absolute change in the explanatory variable.

4.2.2 Confounding Factors

We use a broad set of regressors to control for factors other than public disclosure that affectpollution loadings, including variables related to plant characteristics, input prices, and inter-actions with regulators. Unless otherwise indicated, these data are derived from GRP surveydata. Among the plant characteristics, SCALE is the log of the amount of final product pro-duced by a given plant in a given year (measured in bone-dry metric tons because moisturecontent can vary between different classes of products). FINALGOOD is the proportion of theplant’s output that represents final products (such as writing and tissue paper) used directly byconsumers. This variable is meant to proxy for consumer pressure for environmental quality.PCTBAMBOO, PCTGRASSES, PCTRECYCL, and PCTPULP measure the proportion offiber inputs derived from bamboo, grasses, recyclables, and market pulp, respectively (woodis the omitted category, and agroresidues are also dropped because this input is collinearwith plant fixed effects). FORSHARE is the share of sales derived from exports, a firm-levelvariable derived from the Prowess database. This variable is meant to capture pressure forimproved environmental performance generated by foreign investors and consumers.

15 For information on Prowess, see http://www.cmie.com/database/?service=database-products.htm, and forinformation on Indiastat, see http://www.indiastat.com.

123

N. Powers et al.

Among the input price variables, PRICE_NAOH and PRICE_CL are inflation-adjustedcountry-wide price indices for sodium hydroxide and chlorine, the major chemicals used inthe pulping and bleaching processes. These prices do not vary across plants in our sample.PRICE_WOOD is an inflation-adjusted national price index for wood. This serves as a proxyfor fiber input prices.16 Finally, PRICE_WAGES is an inflation-adjusted national price indexfor average daily wage rates for nonagricultural, unskilled workers.

Two variables are used to proxy for changes in regulatory pressure. PERMIT is a dummyvariable that takes the value one if the plant has been granted a water effluent permit by itsstate pollution control board and zero otherwise. EFFRIVER is the percentage of the plant’seffluent discharged into a river (versus on land or in the sea) in a given year. Effluent dis-charged into a river must, by law, be cleaner, so it is important to control for shifts in thedischarge destinations of effluents.

4.2.3 Channels

As discussed above, to shed light on the channels through which GRP has an effect onplants’ abatement decision, we include various interaction terms—POSTGRPt multipliedby time-invariant plant- and community-level variables—in Eq. 1. We use interaction termsthat proxy for five of the six channels discussed in Sects. 2 and 3: community pressure (c),regulatory pressure (r), output market pressure (g), input market pressure (k), and managerialinformation (t). We were not able to find a reasonably credible proxy for judicial pressure(j). The following discussion defines the 14 variables used to construct interaction terms andexplains briefly how they may proxy for each channel. It is important to point out that theseexplanations are not definitive hypotheses. Rather, they are necessarily somewhat subjective.Also, some variables proxy for more than one channel. Two of our interaction variables(CONS_ORIENTATION and DISPOSAL) are constructed from 3-year predisclosure aver-ages of the control variables discussed above. For ease of exposition, the discussion omitsthe “* POSTGRPt” part of each interaction term.

Community pressure (c). We use five proxies for community pressure. WEALTH is the per-centage of households in the plant’s subdistrict (similar to a US county) that own mopeds.17

LITERACY is the percentage of the population that is literate in the municipality where theplant is located. CASTE is the percentage of population that belongs to a scheduled casteor tribe in the municipality where the plant is located. In India, caste proxies for a host ofsocioeconomic characteristics, including access to credit and quality education. Wealthiercommunities, those that are more literate, and those with a lower percentage of populationbelonging to a scheduled caste may be better able to collect, digest, and act on informationabout plants’ pollution provided by the GRP. AGLABOR is the proportion of the workforcethat are cultivators or agricultural laborers in the municipality where the plant is located. Tothe extent that pulp and paper plants’ water pollution damages irrigated farms, communitiesthat are largely agricultural may have stronger incentives to act on GRP disclosure. URBAN-PCT is the proportion of the population in the subdistrict that lives in municipalities classified

16 We were unable to find sufficient data to construct similar indices for any of the other fiber inputs—bamboo,grasses, agroresidues, recycled paper, and market pulp. However, these prices are correlated with wood pricesin India, where the fiber inputs are substitutable (CSE 2004, p. 40). Also, we do not include a variable mea-suring water costs because industrial users pay a minimal price for water and there was no temporal variationin these prices for the period in question.17 The census data provide alternative measures of wealth, all involving the percentage of households owninga particular asset (or employing banking services). These measures are all highly correlated, and our regressionresults using other wealth measure are qualitatively identical.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

as urban. If urbanized communities have better access to communications and media and aremore densely populated, they may be better able to collect, digest, and act on informationabout plants’ pollution provided by the GRP.

Regulatory pressure (r). We use four regulatory pressure proxies, all of which aredichotomous dummy variables. ENFWATER identifies plants that faced water pollution-related enforcement actions by the regional regulatory authority (pollution control board)before the 1999 rating. COMWATER identifies plants that had been the subject of formalcomplaints to regulatory authorities about water pollution before the 1999 disclosure. GOVidentifies plants that are government-owned. Finally, PERMIT2 identifies plants that hadup-to-date water pollution permits in the 3 years preceding the 1999 rating. Regulators mon-itoring plants that have faced enforcement actions, been the subject of a formal complaint,are owned by government, and have met permit requirements may react differently to GRPinformation about plants’ pollution. In the case of enforcement actions and complaints, theregulators may ratchet up the pressure they place on plants. In the case of government own-ership and permit compliance, the effect on regulators’ actions is more uncertain.

Output market pressure (g). Our two output market proxies are plant-level averages oftime-varying variables for the 3 years before the 1999 rating. CONS_ORIENTATION is theaverage percentage of the plant’s output sold as final (versus intermediate) product—that is,the average of FINALGOOD in 1995, 1996, and 1997—and EXP_ORIENTATION is theaverage percentage of output sold in foreign markets. Buyers of final paper goods and buyersin foreign countries may react differently to GRP disclosure about a plant’s environmentalperformance. For example, buyers of final goods may be more apt to substitute out of goodsproduced by low-rated plants.

Labor market pressure (k). Our labor market pressure proxy is COMPTOWN, the per-centage of nonagricultural laborers in a subdistrict that works for the plant. To the extentworkers in a plant are directly exposed to its pollution, they have added incentives to act onGRP ratings (e.g., by complaining to management or seeking employment elsewhere).18

Managerial information (t). Our managerial information proxy is SINGLE, a dummyvariable that identified standalone single-plant firms. Plants for which SINGLE equals zerowere either part of a larger multiplant pulp and paper company or a diversified conglomerate.Standalone plants may be more responsive to GRP disclosure than multiplant firms becausethey are not able to gather information about pollution and pollution abatement from sisterplants and, as a result, learn more about these topics from GRP disclosure.

Finally, we tested an interaction term that could reasonably be associated with severalchannels. DISPOSAL is the average percentage of a plant’s effluents discharged directly intoa river versus a sewer—that is, the average of EFFRIVER between 1995 and 1997. By law,effluents discharged directly into a river must be cleaner than those discharged into a sewer.This variable could proxy for regulatory pressure if regulators are particularly concernedabout direct discharges, community pressure if communities are particularly concerned, oreven managerial information, if managers of plants discharging into rivers know more abouttheir plants’ emissions.

18 Alternatively, however, one might argue that political economy considerations would cut in the oppositedirection: communities that are heavily dependent on a plant may be less likely to pressure it to cut emissions.

123

N. Powers et al.

5 Results

5.1 GRP’s Impact

As previously noted, Figs. 2 and 3 suggest not only that COD and TSS levels were lowerafter disclosure but also that reductions were greatest among the worst-performing plants.The first objective of our empirical analysis is simply to test whether these results are statis-tically significant. Two specifications for each dependent variable are presented in Table 3;Model 1 (for COD) and Model 2 (for TSS) include only regressors related to the disclosureprogram: TREND, POST-GRP, and ONELEAF, along with interaction terms for ONELEAF.Models 3 and 4 include all time-varying covariates as well.

In Models 2, 3, and 4, the POST-GRP dummy is not significant. In Model 1 it is positiveand significant, although this significance disappears when all covariates are included inModel 3. We conclude that the 1999 GRP rating did not have a significant pollution-reducingimpact on the average plant, including both good and poor environmental performers.

The interaction terms in Table 3—ONELEAF * TREND and ONELEAF * POST-GRP—allow us to test a second hypothesis suggested by Figs. 2 and 3: following the first rating,one-leaf plants improved their environmental performance more than two- and three-leafplants. The negative and statistically significant coefficients on ONELEAF * POST-GRP inall four models in Table 3 suggest that this hypothesis is correct.

The important question that follows from this pair of findings is whether the net effect ofGRP on one-leaf plants was significant. A simple Wald test of the sum of the coefficientson POSTGRP and ONELEAF * POSTGRP allows us to test this hypothesis. The results,reported in the bottom row of Table 3, suggest that GRP did drive reductions in TSS byone-leaf plants. The net effect for COD was not significant, however.

In all regressions TREND is negative and significant, suggesting that exogenous tech-nological improvements in the pulp and paper industry, independent of the GRP, are alsoresponsible for declines in pollution loads. The results for the other covariates unrelated tothe GRP rating shown in columns 3 and 4 are largely consistent with stylized facts aboutthe determinants of water pollution in the pulp and paper industry. In Model 4 (for TSS),SCALE, PCTRECYCL, FORSHARE, PRICE_NAOH, PRICE_CL, PERMIT, EFFRIVER,and PRICE_WOOD are negative and significant, and PRICE_WAGES is positive and signif-icant. In Model 3 (for COD), EFFRIVER is negative and significant. These results comportwith the conventional wisdom that pollution abatement entails economies of scale (SCALE),pollution loadings are decreasing in the share of nonwood inputs (PCTRECYCL), plantsthat export are subject to more pressure to improve their environmental performance (FOR-SHARE), pollution loadings are decreasing in the price of polluting inputs (PRICE_NAOH,PRICE_CL, and PRICE_WOOD), regulatory pressure spurs abatement (EFFRIVER andPERMIT), and labor and pollution abatement are substitutes in production (PRICE_WAGES).It is encouraging that coefficients on these regressors have the same sign, if not significance,in Models 3 and 4.

To assess the economic significance of the estimates, we analyze the effect of disclosureon a hypothetical plant that received one leaf in the first rating and had mean values for allother covariates, using the estimation results from Models 3 and 4. With GRP disclosure,such a plant’s COD discharges would decrease by 63% between 1996 and 2003. However,absent disclosure, the plant’s COD discharges would have decreased by only 54%. The effectof the disclosure program is stronger when we focus on TSS. The plant’s emissions woulddecrease 65% with the disclosure program but only 46% without it.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

Table 3 OLS regression results: Green Ratings Project impact (clustered standard errors in brackets; allregressions include plant fixed effects)

Model number 1 2 3 4Dependent variable COD TSS COD TSS

TREND −0.136 −0.12 −0.124 −0.154

[0.025]** [0.026]** [0.039]** [0.032]**

POST-GRP 0.308 0.021 0.201 0.087

[0.107]** [0.109] [0.129] [0.111]

ONELEAF*TREND 0.043 0.064 0.014 0.066

[0.038] [0.043] [0.033] [0.046]

ONELEAF*POST-GRP −0.446 −0.445 −0.411 −0.527

[0.142]** [0.154]** [0.185]* [0.215]*

SCALE 0.051 −0.274

[0.229] [0.146]+FINALGOOD −0.453 −0.242

[0.296] [0.260]

PCTBAMBOO −0.529 −0.336

[0.357] [0.344]

PCTGRASSES 0.545 −0.906

[1.247] [1.377]

PCTRECYCL −0.492 −0.927

[0.511] [0.369]*

PCTPULP −2.081 0.736

[1.524] [0.911]

PRICE_WAGES 1.648 4.033

[1.262] [1.769]*

FORSHARE −0.473 −2.252

[0.639] [0.727]**

PRICE_NAOH 0.028 −0.874

[0.527] [0.361]*

PRICE_CL −0.191 −0.489

[0.220] [0.238]+PERMIT −0.061 −0.133

[0.088] [0.077]+EFFRIVER −4.399 −4.551

[1.444]** [0.790]**

PRICE_WOOD −0.186 −1.093

[0.503] [0.363]**

Constant 3.85 2.695 6.89 10.577

[0.066]** [0.080]** [3.321]+ [1.981]**

Observations 155 153 154 152

R-squared 0.49 0.58 0.56 0.64

Wald statistic −1.46 −3.90 −1.10 −2.09

+ Significant at 10%; * Significant at 5%; ** Significant at 1%

123

N. Powers et al.

Finally, it is interesting to speculate about the effect on our results of our omission fromthe analysis of the six pulp and paper plants that participated in the first round of the GRP, butwere permanently or temporarily closed before the second rating in 2004. Our results werenot affected by a change over our study period of the composition of our regression samplebecause the six plants in question were completely excluded from our empirical analysis. Thequestion remains, however, whether our results might have been different if these six plantshad not closed and therefore not been excluded. That is, would these plants have respondedto GRP ratings in the same manner as the plants in our sample? Although an argument couldbe made either way, we believe both intuition and previous literature suggests that theseomitted plants would have been at least as responsive as the regression sample.19 At the endof the day, we do not have data to determine which view is valid (including by estimating aselection model that explains plant closure).

5.2 GRP Channels

We tested each of the 14 interaction term variables meant to proxy for five channels in amodel of COD emissions and in a model of TSS emissions—that is, in a total of 28 models.Of these interactions terms, only 2, WEALTH ∗ POSTGRPt, and SINGLE ∗ POSTGRPt,were significant in at least one model (both of TSS) and robust to minor changes in thecomposition of the regression sample (i.e., to the omission of certain plants). Three otherinteraction terms were significant (all in models of COD), but these result were not robust.20

The 12 insignificant or nonrobust interaction terms included both of our proxies for outputmarket pressure and all four of our proxies for regulatory pressure. Several of these proxieswere quite coarse and may not have picked up the intended effects. To the extent that theydid, however, these results hint that output market pressure and regulatory pressure may notplay a strong role in explaining plants’ responsiveness to GRP disclosure.

To keep the exposition manageable, this section presents regression results only forthe 2 channel proxies that were both significant and robust: WEALTH ∗ POSTGRPt, andSINGLE ∗ POSTGRPt. Results for the remaining 12 proxies are presented in a secondappendix available upon request from the authors. Table 4 presents results from eight models(Models 5–12) that include the WEALTH and SINGLE interaction terms. Models 5 and 6include WEALTH interaction terms, Models 7 and 8 include SINGLE interaction terms, andModels 9–12 include one or the other of these variables along with ONELEAF interaction

19 Note that these plants were relatively dirty. GRP ranked most of them at or near the bottom in the firstround. Given that, one might argue that these plants were relatively dirty because they had relatively highmarginal abatement costs and therefore would not have cut their emissions to the same extent as dirty plants inthe regression sample. But one might just as well argue the reverse: that because these plants were particularlydirty, they probably had not yet taken advantage of cheap abatement opportunities—that is, they had relativelylow marginal abatement costs—and partly as a result, would have been quite responsive to GRP ratings. Webelieve the latter argument is more persuasive because it comports with our finding that relatively dirty plantswere most responsive to public disclosure and with similar finding of other studies of pollution evaluationand ratings programs in developing countries (García et al. 2007, 2009; Dasgupta et al. 2007). In addition,one might point out that anecdotal evidence suggests environmental protests spurred by GRP ratings causedat least two of the six plants to close (personal communication, Monali Zeya Hazra, Center for Science andthe Environment, March 13, 2007). In effect, the GRP ratings completely eliminated these emissions.20 LITERACY * POST-GRP was positive and significant at the 5% level. However, this result is not robustto exclusion of two plants for which LITERACY is an outlier. AGLABOR * POST-GRP was negative andsignificant at the 10% level. However, this result is not robust to reestimating the model with any one plantremoved from the sample. Finally, ENFWATER * POST-GRP was significant at the 1% level. However, onlythree of the 22 plants in the sample had enforcement actions in the years prior to the 1999 GRP disclosure, sothis result must be taken with a grain of salt.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

terms in order to disentangle the effects of baseline environmental performance from that ofother plant and community characteristics.

Turning to the results, first note that even with the inclusion of these additional interactionterms, there is virtually no change in the qualitative results on the controls discussed in theprevious section. In both Models 5 (COD) and 6 (TSS), WEALTH ∗ POSTGRPt is negative,and it is significant at the 1% level in the Model 6. This result suggests that plants in wealthiercommunities were more responsive to disclosure than those in poorer communities. Severalexplanations are possible. As briefly noted in the previous section, residents of relativelywealthy communities may have been better equipped to pressure plants to improve theirperformance following GRP disclosure.21

Turning to Models 7 (COD) and 8 (TSS), SINGLE * POST-GRP is negative, and it issignificant at the 10% level in Model 8, implying that standalone plants were more responsiveto GRP ratings than multiplant or conglomerate firms. Again, several explanations are pos-sible. As briefly noted above, one is that standalone plants have worse baseline informationabout pollution and pollution abatement than plants that can exchange information with sisterinstitutions and therefore learn more from GRP ratings.

Although those explanations for results in Models 6 and 8 are plausible, it is also possiblethat plants in wealthier communities and standalone plants were more responsive to disclosurebecause they were dirtier to begin with. Indeed, our data do appear to reflect positive correla-tions between community wealth and baseline dirtiness and between the plant’s standalonestatus and baseline dirtiness.22 In Models 9 through 12, we add the ONELEAF interac-tion terms (ONELEAF * TREND and ONELEAF * POST-GRP) to disentangle the effectsof baseline environmental performance from other plant and community characteristics. Inthese regressions, the WEALTH and SINGLE interaction terms remain significant, althoughthe levels of significance are attenuated. These results suggest that the channels proxied forby WEALTH and SINGLE interaction terms facilitate environmental improvement aboveand beyond that driven by the simple fact that plants in wealthy communities and standalonetend to be dirtier.

5.3 Robustness Checks23

In this subsection we address some possible robustness issues. First, with an unbalancedpanel covering at most 22 plants over 8 years, we have a limited number of observations.Therefore, we need to make sure that our regression results are not being driven by outliers.To that end we performed an outlier check. We repeated each regression in Tables 3 and 4,22 times omitting one plant in each regression. The ONELEAF, WEALTH, and SINGLEinteraction terms are significant in each regression.

A second concern is that several of our regressors, including the composition of fiberinputs, as well as SCALE and FINALGOOD, could be endogenous if abatement decisionsand production decisions are made simultaneously. However, these variables display only

21 One might also argue that, perhaps as a result, regulators in such communities may have been more apt tocrack down on poorly performing plants following disclosure.22 The relationship between community wealth and baseline dirtiness is not as surprising as it may first appear.Certainly, we would expect a lone polluter to face more pressure to abate in a wealthy community, all otherthings equal. However, statistical analysis of our data shows that wealthy communities also tend to be moreurbanized and less reliant on agriculture. Hence, a positive correlation between wealth and dirtiness may existif a dirty plant located in a community where it was one among many such plants and therefore attracted lessattention from regulators and communities.23 These results are available from the authors upon request.

123

N. Powers et al.

Tabl

e4

OL

Sre

gres

sion

resu

lts:G

reen

Rat

ings

Proj

ectc

hann

els

(clu

ster

edst

anda

rder

rors

inbr

acke

ts;a

llre

gres

sion

sin

clud

epl

antfi

xed

effe

cts)

Mod

elnu

mbe

r5

67

89

1011

12D

epen

dent

vari

able

CO

DT

SSC

OD

TSS

CO

DT

SSC

OD

TSS

TR

EN

D−0

.104

−0.1

46−0

.13

−0.1

64−0

.102

−0.1

62−0

.128

−0.1

7

[0.0

42]*

[0.0

29]*

*[0

.034

]**

[0.0

31]*

*[0

.048

]*[0

.039

]**

[0.0

38]*

*[0

.035

]**

POST

-GR

P0.

214

0.38

40.

210.

150.

252

0.47

10.

233

0.19

2

[0.2

05]

[0.1

56]*

[0.1

64]

[0.1

36]

[0.2

28]

[0.1

69]*

[0.1

64]

[0.1

29]

ON

EL

EA

F*T

RE

ND

0.01

40.

064

0.00

70.

042

[0.0

36]

[0.0

46]

[0.0

37]

[0.0

35]

ON

EL

EA

F*PO

ST-G

RP

−0.3

79−0

.44

−0.3

49−0

.329

[0.1

82]*

[0.1

76]*

[0.2

11]

[0.2

11]

SCA

LE

0.06

7−0

.336

0.02

−0.3

430.

03−0

.383

0.03

5−0

.35

[0.2

49]

[0.1

73]+

[0.2

63]

[0.1

86]+

[0.2

34]

[0.1

57]*

[0.2

69]

[0.1

89]+

FIN

AL

GO

OD

−0.2

20.

013

−0.3

62−0

.224

−0.3

9−0

.105

−0.4

75−0

.286

[0.3

34]

[0.2

40]

[0.3

54]

[0.2

63]

[0.3

04]

[0.2

23]

[0.3

00]

[0.2

39]

PCT

BA

MB

OO

−0.2

81−0

.055

−0.4

62−0

.364

−0.4

05−0

.175

−0.5

51−0

.416

[0.3

99]

[0.3

19]

[0.3

93]

[0.3

63]

[0.3

54]

[0.2

99]

[0.3

54]

[0.3

51]

PCT

GR

ASS

ES

0.65

3−0

.249

0.26

2−1

.048

0.96

4−0

.408

0.53

−1.0

54

[1.1

15]

[1.3

31]

[1.1

82]

[1.3

63]

[1.2

96]

[1.3

88]

[1.2

47]

[1.3

86]

PCT

RE

CY

CL

−0.2

92−0

.291

−0.6

03−0

.977

−0.2

48−0

.251

−0.5

01−0

.916

[0.5

04]

[0.3

69]

[0.4

84]

[0.3

69]*

[0.5

81]

[0.3

43]

[0.5

39]

[0.3

33]*

PCT

PUL

P−1

.903

1.46

1−2

.10.

799

−1.8

461.

779

−2.1

070.

951

[1.6

02]

[0.9

03]

[1.6

20]

[1.0

32]

[1.6

53]

[0.8

16]*

[1.6

34]

[0.9

89]

PRIC

E_W

AG

ES

1.33

43.

568

1.47

33.

948

1.33

23.

562

1.60

23.

964

[1.2

62]

[1.6

52]*

[1.2

94]

[1.8

27]*

[1.2

26]

[1.6

59]*

[1.2

62]

[1.8

15]*

FOR

SHA

RE

−0.3

8−2

.105

−0.3

65−2

.181

−0.2

63−1

.962

−0.4

22−2

.135

[0.7

23]

[0.5

73]*

*[0

.756

][0

.700

]**

[0.6

21]

[0.5

09]*

*[0

.677

][0

.671

]**

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

Tabl

e4

cont

inue

d

Mod

elnu

mbe

r5

67

89

1011

12D

epen

dent

vari

able

CO

DT

SSC

OD

TSS

CO

DT

SSC

OD

TSS

PRIC

E_N

AO

H0.

021

−0.8

930.

005

−0.9

20.

05−0

.906

0.02

8−0

.917

[0.5

30]

[0.3

83]*

[0.5

35]

[0.3

87]*

[0.5

24]

[0.3

70]*

[0.5

31]

[0.3

79]*

PRIC

E_C

L−0

.169

−0.4

41−0

.184

−0.4

65−0

.171

−0.4

62−0

.184

−0.4

8

[0.1

97]

[0.2

32]+

[0.2

03]

[0.2

37]+

[0.2

04]

[0.2

28]+

[0.2

21]

[0.2

33]+

PER

MIT

0.01

8−0

.052

−0.0

05−0

.083

−0.0

41−0

.09

−0.0

54−0

.113

[0.0

72]

[0.1

05]

[0.0

69]

[0.1

08]

[0.0

82]

[0.0

93]

[0.0

84]

[0.0

97]

EFF

RIV

ER

−2.0

59−2

.412

−3.5

2−4

.409

−4.0

14−3

.304

−4.5

3−4

.857

[1.2

12]

[0.8

29]*

*[1

.261

]*[1

.207

]**

[1.6

10]*

[0.9

93]*

*[1

.657

]*[0

.941

]**

PRIC

E_W

OO

D−0

.209

−1.0

8−0

.203

−1.0

92−0

.154

−1.0

53−0

.174

−1.0

78

[0.4

67]

[0.3

61]*

*[0

.469

][0

.359

]**

[0.4

89]

[0.3

60]*

*[0

.494

][0

.356

]**

WE

ALT

H*T

RE

ND

−0.1

30.

058

−0.1

360.

053

[0.1

93]

[0.1

69]

[0.2

03]

[0.1

85]

WE

ALT

H*P

OST

-GR

P−0

.631

−2.6

07−0

.288

−2.3

98

[1.0

33]

[0.8

12]*

*[1

.030

][0

.928

]*

SIN

GL

E*T

RE

ND

0.02

0.07

90.

018

0.06

4

[0.0

38]

[0.0

41]+

[0.0

43]

[0.0

31]+

SIN

GL

E*P

OST

-GR

P−0

.292

−0.5

76−0

.132

−0.4

46

[0.1

89]

[0.1

98]*

*[0

.198

][0

.229

]+C

onst

ant

4.91

99.

449

6.63

811

.241

6.74

210

.688

7.17

211

.673

[3.4

30]

[2.3

73]*

*[3

.832

]+[3

.095

]**

[3.2

98]+

[1.8

58]*

*[4

.081

]+[2

.886

]**

Obs

erva

tions

154

152

154

152

154

152

154

152

R-s

quar

ed0.

540.

670.

540.

650.

570.

690.

570.

66

+Si

gnif

ican

tat1

0%;*

Sign

ific

anta

t5%

;**

Sign

ific

anta

t1%

123

N. Powers et al.

minor temporal variation, so any endogeneity should be minimal. As a robustness check,we reestimated all regressions presented in Tables 3 and 4, omitting the potentially endog-enous regressors PCTBAMBOO, PCTGRASSES, PCTRECYCL, PCTPULP, SCALE, andFINALGOOD. In all cases, our results are qualitatively unchanged, and in Model 13, ONE-LEAF * POST-GRP becomes more negative and marginally significant. We conclude thatendogeneity is not an important practical concern in this context.

Finally, note that in the interest of preserving degrees of freedom and concise exposition,we have omitted regressors whose inclusion had no effect: real price of coal, vintage of theplant, other types of regulatory permits, share of sales spent on R&D, and diversity of theproduct mix produced by each plant.

6 Conclusion

We have used 8 years of exceptionally detailed survey data on 22 of India’s largest pulp andpaper plants to evaluate the GRP, an Indian environmental performance public disclosureprogram. We sought to determine whether a 1999 GRP rating caused plants to reduce theirwater pollution loadings. We have also attempted to shed light on the mechanism by whichthe rating may have done this. We found that the GRP drove significant reductions in pollu-tion loadings among dirty plants but not among cleaner ones. This result comports with thefindings in García et al. (2007) and Dasgupta et al. (2007) that performance ratings programsin Indonesia, Philippines, China, and Vietnam all led to improvements among plants withmoderately poor performance records, but not among those with either very bad or goodrecords.

We also found that pulp and paper plants located in wealthier communities were moreresponsive to GRP ratings, as were stand-alone plants. We hypothesized that both resultssuggest that environmental performance ratings programs may have an effect by mobilizinglocal communities and/or regulators to exert pressure for reductions in discharges and byimproving managers’ information about their own pollution and abatement opportunities.However, as discussed above, alternative explanations are plausible. Further research wouldbe needed to validate or reject these hypotheses.

What are the policy implications of these results? They add to a thin but fast-growingbody of evidence that public disclosure programs can be an effective environmental manage-ment tool, even in developing countries where weak regulatory institutions, limited politicalwill, and other problems hamstring conventional pollution control policies. In addition, ourresults hint that disclosure programs administered by nongovernmental organizations mayhold particular promise for developing countries. Whereas virtually all national-level PERPsin developing countries are administered by environmental regulatory agencies, the GRP isrun by a nongovernmental organization, the Centre for Science and Environment. Emergingevidence on PERPs suggests that public sector programs often face political challenges—spe-cifically, countering resistance to disclosure from industrial lobbies (Dasgupta et al. 2007).In principle, nongovernmental organizations—at least politically well-connected, techni-cally proficient ones like CSE—may be better than governmental regulators at overcomingthese challenges. Here too, however, our results are more suggestive than conclusive. Furtherempirical research would be needed to shed light on the promises and pitfalls of private sectordisclosure initiatives in developing countries.

Acknowledgments We are grateful to the Green Rating Project team at the Centre for Science and Envi-ronment (CSE), India, in particular Chandra Bhushan and Monali Zeya Hazra, for providing CSE survey data

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

on the pulp and paper industry and helping us better understand the industry. Thanks also to discussants andseminar participants at the University of Michigan, and the Southern Economics Association, Allied SocialSciences Association, and European Association of Environmental and Resource Economists meetings.

Appendix 1: Analytical model

This appendix presents a simple analytical version of the graphical model of public disclo-sure in Sect. 3. To keep the model as simple as possible and focus attention on pollutionabatement, we assume that the firm makes production and abatement decisions sequentially.First, it chooses a level of output, q, and a vector of levels of financial and human capital, k.Subsequently, it chooses a level of abatement, α, treating both q and k as fixed. We modelthe firm’s second-stage abatement decision only. Note that abatement here may also includepollution prevention. The firm chooses α to maximize profit, π, given by

π = P[g (α, d)

]q-C [α, t (d)] − W (α, d) k − H (α, d)

where

H (α, d) = r (α, d) + c (α, d) + j (α, d)

and

P(·) is the equilibrium price of output;g is an index of green consumerism—the sensitivity of P to the plant’s discharges;d is a measure of the public disclosure of information about the plant’s discharges;q is the quantity of output;

C(·) is the cost of abatement;t is the plant’s information about abatement technologies and its own discharges;

W(·) is a vector of the costs of two types of capital, financial and human;k is a vector of two types of capital, financial and human;

H(·) is the total cost of the plants’ discharges generated by external agents;r(·) is costs generated by formal regulatory authorities;c(·) is costs generated by communities; andj(·) is costs generated by courts.

Following the literature discussed in Sect. 2, we make the following assumptions about theprice and cost functions:

• the stronger is green consumerism, the lower is the equilibrium price the plant receivesfor its output (P is decreasing in g);24

• the less the plant abates and the more the public knows about its discharges, the strongeris green consumerism (g is decreasing in α and increasing in d), the higher are the costs offinancial and human capital (W is decreasing in α and is increasing in d), and the greaterare the costs imposed on the plant by external agents (r, c, and j are all decreasing in α

and increasing in d); and

24 To keep the exposition simple, we implicitly assume that the plant is an inherently dirty one—for example,an aged coal-fired power plant—so that regardless of its choice of α, green consumerism always reducesequilibrium price. We could just as easily assume that the plant is an inherently clean one whose equilibriumprice is always increased by green consumerism. Allowing green consumerism to increase or decrease theequilibrium price depending on the plant’s choice of α makes the model needlessly complex given our limitedgoal of illustrating how various channels discussed in the literature operate.

123

N. Powers et al.

• the less the plant abates and the more information it has about its discharges and abate-ment technologies, the lower is the marginal cost of abatement (C is increasing in α anddecreasing in t).

Finally, we make the reasonable assumptions that

• abatement has a diminishing marginal impact on green consumerism, capital costs, andcosts imposed by external agents; and an increasing marginal impact on abatement costs(g, W, H, and C are all convex in abatement).

The first-order condition for the choice of the optimal level of discharges, α∗, is25

{dP

dg

∂g

∂ αq − ∂W

∂ αk − ∂H

∂ α

}− ∂C

∂ α= 0 (A1)

The first term in brackets represents the marginal benefit of abatement due to an increase inthe equilibrium price of output (the first term in braces); a reduction in the costs of labor andcapital (the second term); and a reduction in costs imposed by formal regulatory authorities,communities, and the courts (the third term). We will refer to the sum of these three terms asthe marginal abatement benefit (MAB). The last term in (A1) is the marginal abatement cost(MAC). The plant chooses α∗ such that MAB is equal to MAC.

Using (A1), it is straightforward to show that the total derivative of α∗ with respect to dis unambiguously negative. Therefore, public disclosure will increase abatement. Figure 1makes this point graphically. Given our assumptions on P(·), C(·), W(·), and H(·), the MACschedule is increasing in α and the MAB schedule is decreasing in α. The plant chooses thelevel of discharges where these schedules intersect. An increase in d will cause t(·) to increaseand the MAC schedule to shift down. It will also cause g(·), W(·), and H(·) to increase andthe MAB schedule to shift up. Each of these shifts will cause α∗ to increase.

References

Arora S, Gangopadhayay S (1995) Toward a theoretical model of voluntary overcompliance. J Econ BehavOrg 28:289–309

Bennear L, Olmstead S (2008) The impacts of the “right to know”: information disclosure and the violationof drinking water standards. J Env Econ Manage 56(2):117–130

Blackman A. (2010) Alternative pollution control policies in developing countries. Rev Environ Econ Policy4(2):234–253

Blackman A, Afsah S, Ratunanda D (2004) How does public disclosure work? Evidence from Indonesia’sPROPER program. Hum Ecol Rev 11(3):235–246

Blackman A, Bannister GJ (1998) Community pressure and clean technology in the informal sector: an econo-metric analysis of the adoption of propane by traditional Mexican brickmakers. J Environ Econ Manage35:1–21

Bui L (2005) Public disclosure of private information as a tool for regulating emissions: firm-level responsesby petroleum refineries to the toxics release inventory. Working paper 05-13, US Census Bureau

Centre for Science and Environment (2004) All about paper: the life cycle of the Indian pulp and paperindustry. Centre for Science and Environment, New Delhi

Dasgupta S, Hong JH, Laplante B, Mamingi N (2005) Disclosure of environmental violations and stock marketin the Republic of Korea. Ecol Econ 58:759–777

Dasgupta S, Laplante B, Mamingi N (2001) Pollution and capital markets in developing countries. J EnvironEcon Manage 42:310–335

Dasgupta S, Wheeler D, Wang H (2007) Disclosure strategies for pollution control. In: Teitenberg T, FolmerH (eds) International yearbook of environmental and resource economics 2006/2007: a survey of currentissues. Edward Elgar, Northampton

25 The convexity of g(·), C(·), W(·), and H(·) guarantee that the second-order condition is met.

123

Does Disclosure Reduce Pollution? Evidence from India’s Green Rating Project

Delmas M, Montes-Sancho M, Shimshack J (2007) Information disclosure policies: evidence from the elec-tricity industry. Working paper, Department of Economics, Tufts University

Foulon J, Lanoie P, Laplante B (2002) Incentives for pollution control: regulation or information?. J EnvironEcon Manage 44:169–187

García JH, Afsah S, Sterner T (2009) Which firms are more sensitive to public disclosure schemes for pollutioncontrol? Evidence from Indonesia’s PROPER program. Environ Resour Econ 42(2):151–168

García JH, Sterner T, Afsah S (2007) Public disclosure of industrial pollution: the PROPER approach forIndonesia?. Environ Dev Econ 12:739–756

Greene W (2002) Econometric analysis, 5th edn. Prentice Hall, Upper Saddle RiverGreenstone M (2003) Estimating regulation-induced substitution: the effect of the clean air act on water and

ground pollution. Am Econ Rev 93:442–448Gupta S, Goldar B (2005) Do stock markets penalize environment-unfriendly behavior? Evidence from India.

Ecol Econ 52:81–95Kerret D, Gray GM (2007) What do we learn from emissions reporting? Analytical considerations and com-

parison of pollutant release and transfer registers in the United States, Canada, England, and Australia.Risk Anal 27(1):203–223

Khanna M (2001) Economic analysis of non-mandatory approaches to environmental protection. J Econ Surv15(3):291–324

Khanna M, Anton WRQ (2002) Corporate environmental management: regulatory and market-based pres-sures. Land Econ 78(4):539–558

Khanna M, Quimio W, Bojilova D (1998) Toxic release information: a policy tool for environmental infor-mation. J Environ Econ Manage 36:243–266

Koehler D, Spengler J (2007) The toxic release inventory: fact or fiction? A case study of the primary aluminumindustry. J Environ Manage 85(2):296–307

Konar S, Cohen M (1996) Information as regulation: the effect of community right to know laws on toxicemissions. J Environ Econ Manage 32:109–124

Laplante B, Lanoie P (1994) Market response to environmental incidents in Canada. Southern Econ J 60:657–672

Lyon TP, Maxwell JW (2002) Voluntary approaches to environmental regulation: a survey. In: Frazini M,Nicita A (eds) Economic institutions and environmental policy. Ashgate Publishing, Aldershot andHamsphire

Maxwell JW, Lyon TP, Hackett SC (2000) Self-regulation and social welfare: the political economy of cor-porate environmentalism. J Law Econ 43(2):583–618

Pargal S, Wheeler D (1996) Informal regulation of industrial pollution in developing countries. J Polit Econ104:1314–1327

Sam A, Innes R (2006) Voluntary pollution reductions and the enforcement of environmental law: an empiricalstudy of the 33/50 program. Working paper, Department of Agricultural, Environmental, and Develop-ment Economics, Ohio State University, Columbus

Schumacher K, Sathaye J (1999) India’s pulp and paper industry: productivity and energy efficiency. Workingpaper 41843, Lawrence Berkeley National Laboratories

Tietenberg T (1998) Disclosure strategies for pollution control. Environ Resour Econ 11:587–602Videras J, Alberini A (2000) The appeal of voluntary environmental programs: which firms participate and

why?. Contemp Econ Policy 18(4):449–461Vidovic M, Khanna N (2007) Can voluntary pollution control programs fulfill their promises? Further evidence

from EPA’s 33/50 program. J Environ Econ Manage 53:180–195Wang H, Bi J, Wheeler D, Wang J, Cao D, Lu G, Wang Y (2004) Environmental performance rating and

disclosure: China’s GreenWatch program. J Environ Manage 71(2):123–133Wooldridge J (2002) Econometric analysis of cross section and panel data. MIT Press, CambridgeWorld Bank (1999) Greening industry: new roles for communities, markets, and governments. Oxford Uni-

versity Press, New York

123