good fences make good neighbors: a longitudinal analysis...

GOOD FENCES MAKE GOOD NEIGHBORS: A LONGITUDINALANALYSIS OF AN INDUSTRY SELF-REGULATORY INSTITUTION

MICHAEL L. BARNETTUniversity of South Florida

ANDREW A. KINGDartmouth College

We extend theories of self-regulation of physical commons to analyze self-regulation ofintangible commons in modern industry. We posit that when the action of one firm cancause “spillover” harm to others, firms share a type of commons. We theorize that theneed to protect this commons can motivate the formation of a self-regulatory institu-tion. Using data from the U.S. chemical industry, we find that spillover harm fromindustrial accidents increased after a major industry crisis and decreased followingthe formation of a new institution. Additionally, our findings suggest that the institu-tion lessened spillovers from participants to the broader industry.

Interest in self-regulatory institutions, wherebyfirms in an industry create and voluntarily abide by aset of governing rules, has gone through a renaissanceof late (cf. Prakash & Potoski, 2006). Scholars haveinvestigated self-regulatory institutions in industriesas diverse as chemicals (King & Lenox, 2000; Lenox,2006), hospitality and recreation (Rivera & de Leon,2004), nuclear power (Rees, 1994), and maritimeshipping (Furger, 1997). Drawing from the work ofElinor Ostrom (1990) and Douglas North (1981),many of these scholars have argued that self-regula-tory institutions arise to constrain individual actionsthat might harm an industry as a whole. Ostrom’s(1990) work on community self-regulation has beenparticularly influential in framing recent research onindustry self-regulation. She demonstrated that thosewho share in common pool resources like fisheries orforests can unite to create an institution that helpsthem avert the “tragedy of the commons” (Hardin,1968), wherein individuals overuse and destroy com-monly held resources.1

However, the theoretical and empirical founda-tions of this growing stream of research on industryself-regulation remain uncertain and contradictory.First, the common pool resource dilemmas thatOstrom considered are not apparent in many indus-tries in which self-regulation has arisen. Researchhas yet to establish whether the same logic thatOstrom applied to the governance of shared phys-ical resources can be extended to modern indus-tries. Second, empirical studies of self-regulationhave often fallen victim to what Granovetter (1985)dismissed as “bad functionalism”—the tendency toinfer the function of an institution by assumingconditions it might serve to ameliorate rather than byactually observing conditions before and after theinstitution’s creation. Third, research on some fre-quently studied institutions seems to provide contra-dictory evidence with respect to their functions. Forseveral important self-regulatory institutions, schol-ars have failed to find any evidence that they limit theharmful practices of member firms, yet studies ofthese same institutions show that they provide a ben-efit to firms in their industries (King & Lenox, 2000;Lenox, 2006; Rivera & de Leon, 2004).

In this article, we address some of the deficien-cies in previous research and so strengthen thetheoretical and empirical foundation of the litera-ture on self-regulatory institutions. First, we drawattention to a novel type of “commons problem”that exists in many industries. We argue that afirm’s error can harm other firms in its industry andthus cause all firms in the industry to share apooled risk. Second, we avoid bad functionalismby measuring this shared risk over a time periodthat spans both the emergence of our hypothesized

1 The concept of the tragedy of the commons arose as arebuttal to the common belief that the “invisible hand” ofthe market causes the pursuit of individual self-interest toaggregate into improved public welfare. In contrast, Hardin(1968), building on ideas advanced by Lloyd (1833), arguedthat because an individual’s gains from increased con-sumption of a public good exceed the individual’s costs(the individual captures all the gains, but the costs areshared by the members of the commons), individuals havea dominant incentive to overexploit unregulated publicgoods. As a result, “Ruin is the destination toward whichall men rush, each pursuing his own best interest in asociety that believes in the freedom of the commons. Free-dom in a commons brings ruin to all” (Hardin, 1968: 1244).

� Academy of Management Journal2008, Vol. 51, No. 6, 1150–1170.

1150

Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyright holder’s expresswritten permission. Users may print, download or email articles for individual use only.

commons problem and the formation of the self-regulatory institution. Finally, by more preciselyidentifying the mechanism by which the institutionprovides benefits to the industry, we provide in-sight on how inconsistencies in the existing litera-ture may be resolved.

THEORY AND HYPOTHESES

Institutions are the “humanly devised constraintsthat structure political, economic and social inter-action” (North, 1990: 97). North separated institu-tions into those that operate through formal con-straints (e.g., rules, laws, and constitutions) andthose that operate through informal constraints(e.g., norms of behavior, conventions, and self-im-posed codes of conduct). Ingram and Clay (2000)refined North’s typology by suggesting that institu-tions should be classified as (1) public or privateand (2) centralized or decentralized. Public institu-tions are usually compulsory and are often run bythe state. Private institutions—those run by organ-izations or individuals—are voluntary in nature,because actors can opt in or out. In the centralizedform of these institutions, a central authority setsrules, incentives, and sanctions for noncompliance.For example, many private institutions (e.g., for-profit firms) have a principal that is ultimately incharge of internal procedures. Decentralized formsof these institutions lack a powerful central author-ity and thus rely on the action of numerous inde-pendent actors to encourage compliance with insti-tutional rules. In many industrial settings, antitrustlaws forbid centralized industry bodies from con-trolling and sanctioning member firm behaviors,and so industry self-regulation tends to take theform of a private and decentralized institution.

Until recently, many scholars dismissed the via-bility of self-regulatory institutions. Influentialanalyses by Olson (1965) and Hardin (1968) sug-gested that since participation is voluntary and freefrom enforcement by a central authority, each actorhas an incentive to defect from agreements and thatconsequently, such institutions should never arise.As a result, absent government regulation or privat-ization, public goods should generally fall victim tothe tragedy of the commons.

However, widespread skepticism about self-reg-ulatory institutions began to change as a result of aseries of investigations in the 1980s and 1990s (cf.Acheson, 1988; Berkes, 1989; Ostrom, 1990; Wade,1988). Ostrom’s work, in particular, changed per-ceptions of the potential for self-regulation.Through a series of comparative case studies, shedemonstrated that individuals could, in fact, orga-nize institutions to cope with overuse of commonly

held resources such as fresh water aquifers, fisher-ies, and forests (Ostrom, 1990). When actors couldnegotiate, observe, and enforce compliance withcommon rules, she argued, self-regulatory institu-tions could protect commonly held resources, andthe benefits provided by protection of these com-mon resources could spur actors to create and par-ticipate in these institutions. In her own assess-ment, her work helped “shatter the convictions ofmany policy analysts that the only way to solvecommon pool resource problems is for external au-thorities to impose full private property rights orcentralized regulation” (Ostrom, 1990: 182). Draw-ing on experimental and field research conductedby her workshop, she concluded that “individualsin all walks of life and all parts of the world vol-untarily organize themselves so as to gain the ben-efits of trade, to provide mutual protection againstrisk, and to create and enforce rules that protectnatural resources” (Ostrom, 2000: 138).

Although self-regulatory institutions have oftenbeen overlooked in the management literature, re-search on the topic is not without precedent (e.g.,Gupta & Lad, 1983; Maitland, 1985). Yet it is onlyin recent years that growing awareness of researchin other fields and increasing recognition of theimportance of protecting common resources hascaused management scholars to take a more activeinterest in self-regulatory institutions (Furman &Stern, 2006; Jiang & Bansal, 2003; King & Lenox,2000). Most of these studies have focused on thedeterminants or consequences of participation in aself-regulatory institution. Few have explored theconditions both before and after an institution’sformation. In their review, Ingram and Clay (2000)identified only one study in the management liter-ature that analyzed antecedent and consequentconditions longitudinally, as is necessary to under-stand the relationship between the existence of acommons problem and self-regulation. In the iden-tified study, Ingram and Inman (1996) showed thatwhen faced with the threat of potential damage to acommonly valued resource, Niagara Falls, local ho-teliers were able to form a self-regulatory institu-tion that limited development and so protected thescenery of the Falls, thereby increasing the proba-bility of survival of nearby hotels.

Ingram and Inman (1996) followed Ostrom(1990) in arguing that the need to protect a sharedphysical resource such as water or land can act asthe catalyst for effective self-regulation. Yet manymodern industries engage in self-regulation, eventhough few are challenged by dwindling stocks of aphysical resource openly shared with rivals. Forexample, the Institute of Nuclear Power Operationsdid not arise in the face of overuse of shared stocks

2008 1151Barnett and King

of uranium in the nuclear power industry, nor didthe Beer Institute Code arise in the face of threat-ened shortages of communal supplies of barley orhops in the brewing industry. What might explainthe frequent presence of self-regulation in settingssuch as these?

One explanation, as we elaborate next, is thatfirms in an industry share an intangible commonsthat binds them to a shared fate. As with a physicalcommons, when the intangible commons is dam-aged, it can pose a serious threat to the success andsurvival of the firms that share it. We hypothesizethat industry self-regulation in modern industriesmay function as a means of resolving this type ofcommons problem.

Industry Commons: A Shared Fate throughShared Sanctions

Firms are considered to be members of the sameindustry when the outputs they produce are closelysubstitutable. To produce closely substitutable out-puts, firms in an industry tend to have similarcharacteristics and make use of similar processes.As a result, when new information is revealedabout the characteristics of one firm, it reflects tosome degree on all firms within its industry. Suchinterdependence can be favorable if, for example,one firm’s success helps to legitimize an emergingindustry and so eases all such firms’ access to re-sources (cf. Hannan & Carroll, 1992), but it can alsobe problematic. Just as one firm’s successes can“spill over” to other firms, so too can its problems.For example, recent news of contaminated spinachharmed the sale of all salad products—not just theproducts of those firms where the contaminationwas found (Galvin, 2007).

Tirole (1996) developed “a theory of collectivereputations” to explain how the reputation of agroup and those who compose the group, past andpresent, are intertwined. Building from the premisethat a group’s reputation is only as good as that ofits members, Tirole (1996) argued that imperfectobservability of individual behavior is the underly-ing cause of collective reputations. Because indi-vidual characteristics are observed with noise, agroup cannot separate itself from the behaviors ofits individual members, and these past behaviors ofindividuals within the group establish expectationsthat others hold of the entire group. As a result,“new members of an organization may suffer froman original sin of their elders long after the latter aregone” (Tirole, 1996: 1).

Whereas Tirole (1996) defined a group at the firmlevel, explaining how the behaviors of individualswithin a firm culminate in a “collective” reputation

for the firm, management scholars have since ap-plied similar logic in pushing collective reputationto the interfirm level, particularly to the industrylevel (Barnett & Hoffman, 2008). King, Lenox, andBarnett (2002) argued that firms in an industryshare a “reputation commons.” A firm’s reputationis based on observers’ judgments of the actions ofthat firm over time (Fombrun, 1996). If observerscan judge the actions of a firm independently of theactions of its rivals, no commons exists, but whenone firm’s actions influence the judgments observ-ers make of another firm or an industry as a whole,a commons arises. This reputation commons inter-twines the fates of firms in an industry because allfirms suffer when any firm engages in actions thatdamage the industry’s shared reputation.

A shared reputation is just one mechanism thatmay cause firms in an industry to share a commonfate. King et al. (2002) further noted the role ofcollective stakeholder sanctions or rewards. Firmsmay be grouped together because it is easier toadminister a common policy over a number offirms. For example, U.S. water pollution regulationoften follows categorical guidelines for each indus-try (Environmental Protection Agency [EPA],1999). Such common policies also reduce the po-tential for regulatory corruption (Blackman & Boyd,2002).

Regulators and other stakeholders also may im-pose common sanctions because they are unable todiscriminate between high- and low-performerfirms in an industry. For example, unable to ascer-tain which firms had contributed to toxic wastedumps, the U.S. government imposed a fee on allchemical producers to fund the Superfund cleanupeffort. Nongovernmental stakeholders often haveeven greater difficulty evaluating the relative per-formance of firms, because such stakeholders usu-ally lack the access and financial resources avail-able to regulators. As a result, these stakeholdersmay advocate a general boycott of certain types ofgoods, or they may select individual firms arbi-trarily for sanctioning, thereby putting all firms inthe industry at risk (Spar & La Mure, 2003). Forexample, activist discontent with working condi-tions in the coffee, athletic shoe, and apparel in-dustries led to high-profile protests and boycottsagainst individual firms in these industries (Star-bucks, Nike, and Kathy Lee Gifford, respectively)that used suppliers with working conditions thatwere no worse than their rivals’—and were in somecases superior (Hornblower, 2000; Malkin, 1996).Although these sanctions were individually tar-geted, the arbitrariness with which firms were tar-geted for sanction created a risk that all firms inthese industries shared.

1152 DecemberAcademy of Management Journal

Though it has not isolated the mechanisms thatproduce such effects, empirical research in finan-cial economics has established that an error2 attrib-utable to a single firm can indeed have adversefinancial consequences for an entire industry. Jar-rell and Peltzman (1985) found that a drug recall byone pharmaceutical firm caused a portfolio of 50rival firm stocks to drop by 1 percent. They foundan even stronger industrywide effect in the auto-mobile industry. When Ford or Chrysler initiateda recall, General Motors actually experienced alarger loss than the recalling firm. Hill andSchneeweis (1983) reported a loss in marketvalue of a portfolio of all electrical utility stocksfollowing the 1979 accident at the Three MileIsland nuclear plant. Mitchell (1989) concludedthat the firms in the over-the-counter pharmaceu-tical industry lost $4.06 billion following thedeadly Tylenol tampering incident. Accordingly,we hypothesize:

Hypothesis 1. An error at one firm harms otherfirms in the same industry.

Industry Commons as a Problem

An industry commons cannot be physically de-pleted in the same manner as a fishery, but it canbecome damaged in a way that significantly harmsfirms and even threatens the industry’s ongoinglegitimacy. For example, the crisis at Three MileIsland in 1979 sparked such deep and enduringpublic concern about nuclear safety that regulatorshave not since approved any new nuclear powerplants in the United States. Major crises like thiscan be a catalyst for shifts in stakeholder percep-tions of an industry (Hoffman & Ocasio, 2001;Meyer, 1982; Rees, 1997). Yu, Sengul, and Lester(2008) argued that crises alter stakeholders’ mentalclassifications of firms. These mental classifica-tions are simplistic and so can produce broad-brushed responses. As a result, a crisis stemmingfrom the actions of one firm can cause stakeholders

to update their beliefs about the reliability and ac-countability of other firms in the same industry.Greenwood, Suddaby, and Hinings (2002) de-scribed a similar process in which “jolts” (Meyer,Brooks, & Goes, 1990) deinstitutionalize estab-lished industry practices and set in motion a pro-cess of “theorization” that determines how observ-ers will view future industry practices.

Hoffman argued that stakeholder perceptionsof an industry are based on metaphors. Theseperceptions can change “suddenly and unpre-dictably” (1999: 366) as significant events influ-ence taken-for-granted assumptions and createnew metaphors about the industry. These newmetaphors influence the interpretation of futureevents, and they can cause even minor events todraw attention and raise the threat of greatersanctions across an industry. Consider the airlineindustry in the aftermath of the events of Septem-ber 11, 2001. These events shifted how observersviewed the airline industry, causing many to as-sess airplanes not merely as a means of transpor-tation, but also as a means of terrorism. Underthis shifted mind-set, observers focused muchmore attention on airline activity and interpretednew events in light of their potential to be part ofa terrorist plot. As a result, minor breaches ofsecurity that had previously gone unnoticed orunquestioned now drew media attention and en-gendered public calls for more stringent securityprotocols that raised costs and sometimes low-ered demand for the entire industry.

Reports of executives in the petroleum and nu-clear power industries validate the perspective thata major crisis can alter perceptions of an industryand, as a result, future problems within the indus-try carry the risk of more severe industrywideharm. As Hoffman (1997) recounted, following theExxon Valdez oil spill in 1989, an Amoco executivenoted that now his firm would “have to live withthe sins of our brothers. We were doing fine untilExxon spilled all that oil. Then we were painted withthe same brush as them” (Hoffman, 1997: 189). Ac-cording to the founding chairman of the Institute ofNuclear Power Operations, in the aftermath of theThree Mile Island crisis, “It hit us that an event at anuclear plant anywhere in our country . . . could andwould affect each nuclear plant. . . . Each licensee is ahostage of each other licensee” (Rees, 1994: 2). There-fore, we hypothesize the following:

Hypothesis 2. A major crisis increases the de-gree to which an error at one firm harms otherfirms in the same industry.

2 We use the term “error” to describe an event associ-ated with a firm that carries adverse consequences. Welater operationalize errors as industrial accidents. How-ever, we do not use the terms “accident” or “mistake”because our theoretical framework deals with eventsthat, though unintended and unplanned, may reveal in-tentional actions by managers and so reflect on the char-acteristics of the firm and, in particular, similar otherfirms. For example, an accident may reveal a managerialdecision to underinvest in safety systems, and the dis-covery of child labor in a supplier’s factories may reveala managerial failure to adequately oversee suppliers.


Industry Self-Regulation as the Solution

How might an industry respond to a commonsproblem? Until recently, many scholars wouldhave predicted there would be little response at all.Publications by Lloyd (1833/1968) and Hardin(1968) on the tragedy of the commons firmly estab-lished the dominant expectation that actors sharinga common resource cannot effectively self-governtheir use of this resource. In the last few years,research on common pool resources has begun tochange such expectations by suggesting that therisks from inefficient use of common resources canmotivate members of a commons to create a self-regulatory institution (Ostrom, 1990). Yet it is alsoclear that forming a self-regulatory institution canbe difficult. Competition, inertia, and the inherentcost of forming such an institution inhibit rivalfirms from coming together (Barnett, 2006).

Research has suggested that a sudden worseningof a commons problem, as described in the priorsection, is often the catalyst that brings actors to-gether to form a self-regulatory institution (Gunder-son, Holling, & Light, 1995; Gunningham & Rees,1997). For example, self-regulation in the Mainelobster industry arose after a collapse of the fish-ery caused the closure of important canneries(Acheson & Knight, 2000). Similarly, environmen-tal emergencies in New Brunswick, the Everglades,and the Chesapeake Bay precipitated changes togoverning institutions (Gunderson, Holling, &Light, 1995). Crises such as these can help actorsovercome cognitive barriers and recognize the ex-istence and importance of a commons (Weber, Ko-pelman, & Messick, 2004), and they can also changeexpectations of the value of taking action (Vasi &Macy, 2003).

Institutions created in response to a threat to ashared natural resource, scholars have argued, tendto reduce the shared threat (Acheson & Knight,2000; Ostrom, 1990). This prediction matches a“functionalist” interpretation of institutions thatsuggests they arise to facilitate more efficient socialexchange (cf. North, 1981; Williamson, 1985).However, evidence supporting the functionality ofself-regulatory institutions remains largely lacking,in part because scholars often cannot access theinformation about conditions prior to the creationof the institution. As a result, many attempts atempirical validation have been soundly criticizedas “bad functionalism.” In an influential critique,Granovetter (1985) quoted Schotter’s statement thatmany studies begin with a theory of the function ofan institution and then infer “the evolutionaryproblem that must have existed for the institution

as we see it to have developed” (Schotter [1981: 2],quoted in Granovetter, 1985: 489).

Herein, we advance a functional theory of self-regulatory institutions, but rather than inferring theproblem that must have existed for such institu-tions to have arisen in modern industry, we usetheory to hypothesize the creation of a particulartype of problem: a major crisis exacerbates an in-dustry commons, placing firms at heightened riskof harm from errors at other firms in the industry.To validate our functional theory, then, we nexthypothesize that the institution formed in responseto this crisis indeed functions to reduce the height-ened commons problem.

Hypothesis 3. An industry self-regulatory insti-tution created following a major crisis reducesthe degree to which an error at one firm harmsother firms in the same industry.

By specifying the existence of a particular com-mons problem and then testing whether or not anindustry self-regulatory institution alleviated thisproblem, the prior set of hypotheses constitutes aproper test of a functional theory of self-regulationin modern industry. However, a thorough analysisof a self-regulatory institution should include aspecification of the means through which itachieves its function. We next explore the institu-tion’s mechanisms.

Exploring the Mechanisms ofIndustry Self-Regulation

As we have theorized thus far, firms in an indus-try are subject to “spillover” harm from the errors ofother firms, and an industry self-regulatory institu-tion functions to lessen this harm. We have not yetspecified how the institution might accomplishthis objective. The literature on industry self-regu-lation suggests two possible mechanisms: the insti-tution might forestall the threat of industrywidestakeholder sanctions (Stefanadis, 2003), or itmight direct sanctions away from it members (Ter-laak & King, 2006).

Forestalling sanctions to an industry. One wayan institution might help firms forestall industry-wide stakeholder sanctions would be by facilitatingcollective performance improvements relative tocriteria of concern to stakeholders (Barrett, 2000;Dawson & Segerson, 2005). In doing so, the institu-tion would impose costs on its members, but itwould create greater value by reducing the risk ofstakeholder action (e.g., government regulation).Although it is influential in setting research agen-das, the general applicability of this model hasbeen called into question by empirical evidence


that participants in industry programs for environ-mental improvement do not have better perform-ance than nonmembers, nor do they seem to im-prove more rapidly (Howard, Nash, & Ehrenfeld,2000; King & Lenox, 2000; Rivera & de Leon, 2004).

Even if a self-regulatory institution failed to im-prove average industry performance, it might stillreduce the threat of industrywide sanction by co-ordinating a unified “non-market strategy” amongfirms in the industry (Baron, 1995). For example,by acting together, firms might more efficiently andeffectively lobby state regulators and so counterthreats of increased regulation. Even if the targetstakeholder is more diffuse than a state regulator(e.g., consumers or stakeholder groups), the insti-tution might still aid in forestalling sanctions byassisting in the creation of a consistent message orby allowing firms to pool resources and so accesseconomies of scale in communication (e.g., televi-sion advertisements).

Some mechanisms of forestalling stakeholdersanctions could differentially reduce the spilloverharm from errors occurring at a member of an in-stitution. One common case occurs when an insti-tution facilitates the transfer of reassuring informa-tion following an error at a member firm. Researchhas demonstrated that open provision of informa-tion about an error such as an accident, spill, ordrug-tampering incident can reduce the degree towhich stakeholders sanction the focal firm (Shriv-astava & Siomkos, 1989). Thus, when member firmsdo suffer errors, their coordinated efforts at com-munication could provide an efficient and effectivemeans of reducing spillover harm.

A self-regulatory institution could also reducespillover harm from the errors of member firms bycoordinating communication before an error oc-curs. From the perspective of information econom-ics, if a stakeholder already understands the pro-pensity for an error, the occurrence of an errorshould provide no new information. From a psy-chological perspective, information provided in ad-vance of an error may also influence how theerror is interpreted by reducing the degree towhich stakeholders view any observed error asinformative about unobserved dangers, thus re-sulting in a soothing effect. Slovik and Weber(2002) reported that:

The informativeness or signal potential of a mishap,and thus its potential social impact, appears to besystematically related to the perceived characteris-tics of the hazard. An accident that takes many livesmay produce relatively little social disturbance (be-yond that caused to the victims’ families andfriends) if it occurs as part of a familiar and well-understood system (e.g., a train wreck). However, a

small incident in an unfamiliar system (or one per-ceived as poorly understood), such as a nuclearwaste repository or a recombinant DNA laboratory,may have immense social consequences if it is per-ceived as a harbinger of future and possibly cata-strophic mishaps. (2002: 13)

Thus, an industry self-regulatory institution mayfulfill its function of reducing spillover harm bydisclosing key information about its member firms,so that future errors at these firms reveal little or nonew information of relevance and so draw few orno industrywide stakeholder sanctions. Consis-tently with this notion, the first code of one in-fluential self-regulatory institution, the chemicalindustry’s Responsible Care Program, was focusedon stakeholder outreach. It obliged managers offacilities of participating firms “to identify and re-spond to community concerns, [and] inform thecommunity of risks associated with company oper-ations” (Canadian Chemical Producers’ Associa-tion [CCPA], 2007).

If a self-regulatory institution forestalls broadstakeholder sanctions by coordinating improve-ment, lobbying, or public relations for an industry,it should provide a general benefit to the industryas a whole.3 However, if it forestalls industrywidesanctions by providing additional informationabout its members only, either before or after errors,spillover harm from members’ errors should be re-duced. Thus, we hypothesize the following:

Hypothesis 4. An industry self-regulatory insti-tution decreases the degree to which an error ata member firm harms other firms in the sameindustry.

Diverting sanctions from members. In contrastto functioning as a means of forestalling industry-wide sanctions, a self-regulatory institution mightinstead function as a type of “market signal” thatdirects sanctions away from members by helpingstakeholders distinguish the superior but unob-served characteristics of member firms from thoseof nonmember firms.

As discussed earlier, stakeholders often cannot

3 A prediction consistent with this method of indus-trywide stakeholder forestalling would be that the insti-tution would reduce spillover harm evenly (to membersand nonmembers), wherever an error occurs (at memberor nonmember firms). We do not develop this predictioninto a formal hypothesis because it requires a test of thenull hypothesis of indistinguishable coefficient estima-tions for both variable pairs. Confirmation of Hypothesis3 and disconfirmation of Hypotheses 4 and 5 wouldrepresent consistent but insufficient evidence of this in-stitutional mechanism.


directly observe important attributes of a firm. Forexample, they may not be able to see the degree ofaccident preparedness in a particular petrochemi-cal facility. Lacking such information, stakeholdersmay assume that all such firms tend to performsimilarly. If participation in an institution acts asa market signal, it would credibly reveal to stake-holders information about desirable characteris-tics they could not otherwise observe. As a result,members of the institution should be less at riskof spillover harm than nonmembers, as onlymembers are believed to possess these superiorcharacteristics.

For an institution to allow firms to credibly sig-nal their superior attributes, it must meet severalrestrictive conditions (Spence, 1973). First, theremust be some way to keep firms with lower perform-ance from joining the institution. This is commonlyaccomplished by setting rules for participation thatmake membership too costly for firms with inferiorattributes. Second, high-performing firms mustwish to participate. Stakeholders must be able todifferentially reward participants, and the cost ofparticipation must not exceed these benefits. Third,there must be a credible mechanism, usually athird-party auditor, for evaluating and certifyingcompliance with the institution’s rules. Finally,there must be some way for stakeholders to sanc-tion or reward individual firms (e.g., by boycottingor buying their products). Evidence supporting sig-naling theories has been found for institutions thatseem to target buyers and suppliers, such as theInternational Organization for Standardization(ISO) management standards (Corbett, Montes-San-cho, & Kirsch, 2005).

Hypothesis 5. An industry self-regulatory insti-tution decreases the degree to which an error atanother firm in its industry harms a memberfirm.

RESEARCH SETTING AND METHODOLOGY

To test our hypotheses, we needed an industrythat experiences frequent errors of varying signifi-cance, has suffered a major crisis, and has created aself-regulatory institution to recover from this ma-jor crisis. The U.S. chemical industry meets all ofthese requirements. Members of the industry suffermultiple errors each year, and these errors vary insignificance. Most are small and involve the un-planned release of potentially toxic chemicals.More serious errors injure or kill employees or localcitizens. Industry experts and industry membersreport that one error precipitated a major crisis forthe industry. On December 3, 1984, methyl isocya-

nate leaked from a Union Carbide facility in Bho-pal, India, and killed between 3,000 and 10,000people. Many thousands more were injured (Shriv-astava, 1987). It remains the most deadly industrialaccident on record.

Anecdotal reports from managers in the chemicalindustry around the time of this event support theperspective on industry commons we have hypoth-esized. Numerous respondents reported that theincident at Bhopal created a crisis for the entirechemical industry by changing how observersviewed the risks of chemical manufacturing. Ro-nald Lang, then executive director of the SyntheticOrganic Chemical Manufacturers Association,noted, “Bhopal focuses concern on something thathad not been adequately addressed before—thepossibility of catastrophe” (Gibson, 1985a: 21). An-other industry leader described the post-Bhopalenvironment as “chemophobia” (Gunningham,1995: 72).

Contemporaneous reports provide evidence thatindustry participants now had a greater sense ofbeing part of a commons. Then-chairman of UnionCarbide, Warren Anderson, remarked, “This is nota Carbide issue. This is an industry issue” (Gibson,1985a: 21). Others noted that with such a closefocus on the industry, further accidents at anychemical firm would have significant conse-quences for all chemical manufacturers (Gibson,1985b). Industry managers also noted that onestakeholder in particular—insurance companies—formally implemented this increased perception ofrisk in a way that affected all firms in the industry,regardless of their individual characteristics: “Nowthe Bhopal tragedy has reinforced the new conserva-tism of insurance underwriters and, as one brokerputs it, given them ‘an excuse to say no.’ . . . Whenthey see any operations associated with chemicals—even chemical operations posing no hazard to thepublic—[underwriters] are ready to paint them withthe same brush as Bhopal’s” (Katzenberg, 1985: 30).

Respondents also reported that the crisis droveindustry members to create a prominent self-regu-latory institution: The Responsible Care program ofthe American Chemistry Council (ACC).4

“More than anything else,” recalls [then] Union Car-bide CEO Robert Kennedy, “it was Bhopal that fi-nally put us on the path that would lead to Respon-sible Care.” “Bhopal was the wake-up call,” says[then] Dow Chemical vice president Dave Buzelli.“It brought home to everybody that we could have

4 At the time of the events reported in this study, theAmerican Chemistry Council was known as the Chemi-cal Manufacturers Association.


the best performance in the world but if anothercompany had an accident, all of us would be hurt, sowe started to work together.” (Rees, 1997: 485)

The first element of the program, CommunityAwareness and Emergency Response (CAER), ap-peared designed to reduce concern about accidents.Responsible Care encouraged firms to “open thedoors and let the fresh air flow through” (Coombes,1991) so that a skeptical public would be con-vinced that the dangers that Bhopal had brought tolight were being rigorously dealt with and thatthere was no imminent danger. Responsible Carerequired extensive outreach efforts with local offi-cials in communities where plants were located. Aspart of CAER, firms conducted thousands of planttours. The ACC also spent millions on advertisingcampaigns to humanize the industry (Heller, 1991).

The potential of the program to actually changethe operations of members has been questionedfrom its inception. Critics noted that the programdid not include a mechanism for third-party certi-fication of compliance with the rules and thus ar-gued that it was unlikely to help stakeholdersaccurately determine those firms with higher per-formance (Ember, 1995). Others argued, however,that close connections within the industry couldallow members to police adherence to the rules(Rees, 1997). As discussed earlier, empirical stud-ies have suggested that member firms did not re-duce their pollution any faster than nonpartici-pants but that the industry still benefited from theexistence of the program (King & Lenox, 2000;Lenox, 2006).

Sample

Our sample included all firms in the Center forResearch on Security Prices (CRSP) database thatreported any operations in the chemical industry(SIC 2800–2899) in the Compustat business seg-ment database between 1980 and 2000 or reportedto the Environmental Protection Agency (EPA) thatthey operated a production facility in these sectors.We chose this time period to allow a nearly 5-yearpretest window before the Bhopal crisis (1980–84),a 5-year interval between the Bhopal crisis and thecreation of Responsible Care (1985–1989), and an11-year posttest window after the creation of Re-sponsible Care (1990–2000). We chose the sampleto include all firms that reported any chemical op-erations, rather than only those with a primarydenomination in the chemical industry, so as toinclude diversified firms with significant thoughnondominant chemical operations. Our final sam-ple included 735 unique firms.

We obtained data about our sample firms fromthe CRSP database, the Compustat business sectordatabase, and the EPA’s Toxics Release Inventory(TRI). Reporting to the TRI began in 1987 and cov-ers facilities with ten or more employees that pro-duce, store, release, or transfer more than a thresh-old amount of any of more than 600 listedchemicals. We obtained data about errors by per-forming keyword searches of the major interna-tional and U.S. regional newspapers and wire da-tabases within the Lexis-Nexis service. In somecases, we supplemented information from news ar-ticles with information from the Hoovers onlinedatabase and Dialogweb.

Dependent Variable

Managers in the chemical industry consider in-dustrial accidents to be serious errors (Greening &Johnson, 1997). Industry experts further claim thatthe Bhopal crisis changed the degree to which in-dustrial accidents harmed the industry (Gibson,1985b). Thus, we used industrial accidents atchemical plants as our measure of error.

To find industrial accidents, we performed key-word searches using terms such as “fire,” “gasleak,” “explosion,” “chemical spill,” “chemical re-lease,” and “chemical discharge.” These are termsthat top managers in the chemical industry associ-ate with serious accidents (Greening & Johnson,1997). Our search uncovered 359 possible acci-dents. As with any keyword search, however, wenetted numerous inappropriate events, such aschlorine burns in swimming pools or ammoniaspills on restaurant floors. We also found numerousaccidents related to transportation (e.g., a tankertruck flipped and exploded into flames on a rampto the Capital Beltway) and accidents related topetroleum transport and refining. We included ac-cidents at refineries, since the petroleum andchemical industries are closely associated (i.e., pet-rochemicals; see Hoffman [1997]),5 but we ex-cluded leaks and spills from crude oil pumping ortransport (e.g., the Exxon Valdez accident) becausetransportation is often subcontracted, and it is un-clear if accidents in transport would reflect on thetransporter, the producer, or the wholesaler/re-tailer. Incomplete reporting about key aspects ofaccidents further narrowed the sample. In total, wewere able to qualify and determine the date, mag-

5 In analyses not reported here, we created a dummyvariable to indicate accidents that occurred at refineries(rather than at chemical plants) and found it to be insig-nificant in various analyses.


nitude, location, and responsible firm for 123 of theraw events.

Problems with the size of events or contempora-neous firm actions caused removal of an additional31 accidents. Fifteen accidents resulted in neitheran injury nor death. Our preliminary analysisshowed that such accidents were too minor to af-fect the stock price of even the firms directly re-sponsible for them, and so we excluded theseevents from our study of the spillover effects. At theopposite end of the scale, we removed the Bhopalevent because of its extreme nature. Finally, eventsshould not be confounded with endogenous ac-tions that might bias coefficient estimates (McWil-liams & Siegel, 1997). We excluded 15 events be-cause other significant activities were mentioned inthe newspaper accounts of them (e.g., leadershipchanges). The exclusions left us with 92 accidents.

To measure the degree to which these accidents“harmed” other firms in the industry, we evaluatedthe stock price movements of other firms withchemical operations following each accident. Mar-ket theory suggests that stock prices reflect the bestassessment of future cash flows for firms. If inves-tors think that an accident might decrease a firm’sfuture cash flows, perhaps as a result of decreaseddemand resulting from increased stakeholder sanc-tions or increased costs resulting from increasedgovernment regulation, then that firm’s stock priceshould fall.

Appendix A describes how we captured theharm of each accident with our dependent variable,cumulative abnormal return (CAR) on day 5. CARis a measure of how much a stock’s value deviatesfrom its expected value over a particular period oftime. Though our study focuses on the extent towhich an accident at one firm causes harm to otherfirms, we first explored the effect of these accidentson CAR for the firms directly responsible for theevents in order to validate that our search uncov-ered a set of accidents that might influence stockvalue. As Appendix B shows, in model 1, the stocksof the firms responsible for these 92 events lost anaverage of 1.01 percent of their expected values inthe five days immediately following their acci-dents. If we use instead the point estimates frommodel 2, we find that an average accident (one thatinjured about 3.5 employees) caused about a 1.4percent reduction in a firm’s stock price. An acci-dent that killed an employee caused an additionalloss of 2.6 percent of market value. In model 3, weshow how other variables influence the direct ef-fect of an accident. For example, the positive coef-ficient for the variable “assets of perpetrator”shows that losses are greatly reduced for largerfirms that experience accidents. We explain these

variables below and return to them for comparativeanalysis later in the article.

Independent Variables

We hypothesize that one firm’s error can spillover to other firms within the same industry. Schol-ars commonly use the Standard Industrial Classifi-cation (SIC) code as the definition of an industry.Since we were examining spillovers from accidentsin the chemical industry, we limited our set offirms to those that reported at least one segment inthe 2800 SIC range, which encompasses chemicals.We also created a more refined measure of opera-tions in the same industry by creating a binaryvariable, same SIC, that captures whether a firmowned a facility that was in the same four-digit SICcode as the one that had an accident.6

Our second hypothesis predicts that a major cri-sis will increase this spillover effect. As previouslydiscussed, industry insiders assert that Bhopalcaused a major crisis that affected all firms withchemical operations. To capture the changescaused by Bhopal, we created a dummy variable,pre–Bhopal, that captures the time period from thebeginning of our sample through the end of 1984,when the Bhopal leak occurred. Events occurringin this period were coded 1, and those occurringlater were coded 0.

Our third hypothesis predicts that industry self-regulation decreases this spillover effect. Aspectsof the Responsible Care program began shortly afterthe Bhopal crisis, but no elements of the programwere promulgated until late in 1989. Thus, to cap-ture the post–Responsible Care period, we created adummy variable coded 1 if after 1989 and 0otherwise.

Finally, in order to explore the mechanisms of in-dustry self-regulation, we had to distinguish Respon-sible Care participants from nonparticipants. To cap-ture this effect, we measured membership in theparent association, the American Chemistry Council.We assigned this variable (ACC member) a value of 1if the ACC listed the company as a member in a given

6 We investigated whether a percentage measurewould provide similar results or results with greater ex-planatory power. We found that a continuous specifica-tion provided no significant increase in explanatorypower. It may be that spillover risk is not linearly relatedto the scale of a firm’s operations in a given industry orthat our continuous specification (the percentage of afirm’s total employees employed in any of the firm’sfacilities that share the same four-digit SIC as the firm atfault) does not precisely measure relative scale acrossindustries. Thus, we used the binary measure.


year, and 0 otherwise. From 1990 onward, participa-tion in Responsible Care was a condition of member-ship in the ACC. We further separated out the uniqueattributes of the Responsible Care program by creat-ing a dummy variable, Responsible Care member,that indicated ACC membership during a year whenthe Responsible Care program was in existence. Wealso sought to understand whether Responsible Caremembership might have different effects, dependingon the identity of the firm responsible for an accident.Thus, we created two variables, perpetrator in ACCand perpetrator in Responsible Care, that capturedwhether or not a firm that had an accident was amember of the American Chemistry Council or of theResponsible Care program.

Control Variables

Variation in the magnitude of an event shouldcause variation in the market’s response, so we in-cluded two measures of each accident’s severity. Em-ployee killed was a binary variable assigned a value of1 if an employee was killed in the accident. Onlyabout 25 percent of the accidents in our study in-cluded a fatality, so the binary parameterization isappropriate and truncates little information. In con-trast, we measured the number of employees injured(employees injured) using a continuous measure(log[number injured � 1]). We used the log paramet-ric form to reduce the effect of outliers and to accountfor possible diminishing effects. Alternative paramet-ric forms (binary and linear) for our measure of inju-ries reduce model fit but do not change the sign andsignificance of reported results.

The size of both a perpetrator (firm responsible foran accident) and a recipient firm (another firm in theindustry that is subject to spillover) may influence themarket response to accidents. Larger perpetrators of-ten have better public relations and so may be able to

diffuse public reaction and response to their ownaccidents and to the accidents of others that mightreflect on them. Moreover, larger firms tend to bemore diversified, and so their overall stock pricewould suffer less from adverse events in any singleindustry. Finally, the size of a firm has been shown toinfluence the variability of its stock performance(Fama & French, 1992). We measured firm size as thelog of total assets reported in Compustat for the yearan accident occurred, and we further added assets ofperpetrator, the log of the total assets of the firmwhere the accident occurred.

Table 1 provides descriptive statistics and corre-lations. We used several approaches to reduce thepotential for unobserved firm differences or endog-enous managerial choice to bias our coefficient es-timates. We included fixed effects of different typesto control for unobserved (but constant) differencesin our sample (e.g., accident industry, spilloverindustry, and the year of the accident). To controlfor unobserved differences among the industries inwhich accidents occurred, we included fixed ef-fects for all industries in which there was morethan one accident. To control for unobserved firm-level differences, in some models we includedfixed effects for a firm itself. These fixed effectsaccount for constant attributes of any firm thatmight influence spillover effects. Endogenouschoice processes might also bias our sample, par-ticularly if they are based on changing firm charac-teristics not captured by our fixed effects. To helpaccount for these, we also conducted a Heckmantwo-stage treatment model.

Analysis

Event study methodologies are commonly used tounderstand how stockholders interpret a single event(Blacconiere & Patten, 1994; Brown & Warner, 1980,

TABLE 1Descriptive Statistics and Correlations for Spillover Analysisa

Variable Mean s.d. 1 2 3 4 5 6 7 8 9 10 11

1. CAR, day 5 0.14 9.802. Pre-Bhopal 0.04 0.19 �.013. Post–Responsible Care 0.79 0.41 .05 �.354. Same SIC 0.44 0.50 �.05 �.06 .045. Employee killed 0.27 0.45 �.03 .11 �.29 �.106. Employees injured 1.44 1.17 �.05 .09 �.11 .08 .037. Assets of perpetrator 9.52 1.43 .07 .05 .04 �.33 �.08 �.208. Perpetrator in ACC 0.74 0.44 .01 .12 �.04 .13 �.01 �.15 .309. Perpetrator in RC 0.57 0.50 .05 �.21 .59 .13 �.18 �.21 .30 .71

10. ACC member 0.17 0.38 �.01 .09 �.11 .03 .04 .03 �.01 .01 �.0611. Responsible Care member 0.12 0.32 �.00 �.06 .16 .04 �.03 .00 �.001 .00 .10 .8212. Assets 4.87 2.36 .00 .04 .02 �.07 �.02 �.01 �.003 .01 .02 .52 .44


1985; Hamilton, 1995; Patten & Nance, 1998). McWil-liams and Siegel (1997) criticized the use of thismethod when firms might be able to alter or timefocal events so that they occur concurrently withother announcements. Accidents, however, are bytheir very nature unplanned and not amenable tostrategic timing. Thus, such manipulation is not aconcern in this study.

Our event study analysis allowed us to connectan accident with abnormal stock price movements.To understand the causes of variance within thesemovements, we used a linear regression:

CARij � a � BXij � eij � �p � ul � vi � �t,

where CARij is the cumulative abnormal return forfirm i five days following event j and Xij is a vectorof independent variables for firm i at the time ofevent j. Clearly, we could not measure every possi-ble factor that might influence the effect of an ac-cident or the spillover from that accident to otherfirms. We used a series of fixed effects to try toreduce potential problems from unobserved heter-ogeneity. First, we attempted to account for unob-servables in the industries in which the accidentsoccurred. We included a fixed effect (�p) for all pindustries in which we had more than a singleaccident (16 of 22 industries). These effects help tocontrol for unobserved industry differences (e.g.,the propensity to use subcontractors), which mightaffect the number or type of accidents. Second, weaccounted for potential differences among indus-tries that were affected by an accident (but notnecessarily the industry in which it occurred) byincluding fixed effects for every two-digit SIC code(ul). Alternatively, when we considered issues ofvariable spillover among firms and were interestedin the effect of variables that were not collinearwith firm identity, we included both firm (�i) andyear (�t) fixed effects.

Clearly, the decision to join the American Chem-istry Council and participate in Responsible Care isendogenous to our analysis; that is, managers makedecisions about participation conditional on thecharacteristics of their organization. To the extentthat we capture the important firm characteristicsthrough the inclusion of a direct measure or by theuse of fixed effects, our coefficient estimate shouldbe unbiased. However, these methods will fail tocapture the effect of endogenous choice based onchanging and unmeasured factors. To account forsuch factors and test the robustness of our analysis,we performed a two-stage Heckman treatmentmodel.

The first stage consists of the model predictingparticipation in Responsible Care. We based the

treatment selection function on a model of mem-bership proposed by King and Lenox (2000), whofound that participation in Responsible Care is in-fluenced by a firm’s relative emissions, the relativeemissions of the industries in which it operates, thedegree to which it is focused in chemicals, its size,and its reputation. Following their study, we usedTRI data to estimate the median emissions for eachindustry (four-digit SIC) and create a weightedmeasure of this value (based on percentages of salesin this SIC) for each firm.7 The log of this valuebecame our measure of the degree to which a firmoperated in sectors with many toxic chemicals (in-dustry emissions). We measured the degree towhich a firm operated in chemicals (chemical fo-cus) by calculating the percentage of sales fromchemical sectors (as estimated from Compustatdata). We already had a measure of firm size (as-sets). Because of the limitation of the TRI, we couldnot estimate relative performance prior to 1987. Wealso could not develop contemporaneous measuresof reputation for all of the firms in our sample. Weprovide descriptive statistics for these variables inAppendix C.

RESULTS

To explore the effect of spillovers from accidents,we evaluated the abnormal stock price movementsof each firm in our sample (excluding the perpetra-tor) after each accident in our sample. To get asense of the average scale of such spillovers, inTable 2 we first specify a simple model withoutfixed effects. In support of Hypothesis 1, model 1shows that a firm that operated a facility in thesame industry in which an accident occurred in-deed experienced a negative spillover. We alsofound that accidents in which employees werekilled or injured resulted in additional negativespillovers. Note that the effects are smaller thanthose to the focal firm (see Appendix B) but are stillsignificant. Following an accident that injured anaverage number of employees (3.5), a chemical firmwith operations in the same industry as that inwhich an accident occurred could expect to lose0.15 percent of its stock price. After an accidentthat caused the death of an employee, the firmcould expect to lose an additional 0.83 percent.

To explore whether the accident at Bhopal or theformation of Responsible Care influenced spill-overs from accidents, we specified a new model(model 2) with the two dummy variables respec-

7 For the years 1980–86, we estimated industry emis-sions on the basis of the 1987 TRI data.


TA

BL

E2

Sp

illo

ver

Eff

ect

ofA

ccid

ents

onF

irm

sN

otA

tF

ault

Var

iabl

eM

odel

1M

odel

2M

odel

3M

odel

4M

odel

5M

odel

6M

odel

7M

odel

8

Pre

-Bh

opal

0.75

**(0

.32)

1.01

***

(0.3

3)1.

06**

*(0

.34)

1.16

***

(0.3

4)1.

21**

*(0

.42)

Pos

t–R

esp

onsi

ble

Car

e1.

02**

*(0

.15)

0.92

***

(0.1

7)0.

79**

*(0

.19)

0.27

(0.3

1)0.

18(0

.35)

Acc

iden

tin

sam

eS

IC�

0.88

***

(0.1

1)�

0.87

***

(0.1

1)�

0.60

**(0

.27)

�0.

58**

(0.2

7)�

0.60

**(0

.27)

�0.

66**

(0.3

2)�

0.57

**(0

.27)

�0.

57**

(0.2

7)E

mp

loye

eki

lled

�0.

83**

*(0

.13)

�0.

59**

*(0

.13)

�0.

38**

*(0

.14)

�0.

37**

(0.1

4)�

0.41

***

(0.1

4)�

0.43

***

(0.1

6)�

0.27

*(0

.16)

�0.

27*

(0.1

6)E

mp

loye

ein

jure

d�

0.36

***

(0.0

5)�

0.34

***

(0.0

5)�

0.38

***

(0.0

6)�

0.38

***

(0.0

6)�

0.36

***

(0.0

6)�

0.39

***

(0.0

6)�

0.24

***

(0.0

7)�

0.24

***

(0.0

7)A

sset

sof

per

pet

rato

r0.

37**

*(0

.05)

0.37

***

(0.0

5)0.

33**

*(0

.06)

0.34

***

(0.0

6)0.

23**

*(0

.06)

0.23

***

(0.0

6)P

erp

etra

tor

inA

CC

�0.

81**

*(0

.31)

�0.

97**

*(0

.36)

�1.

16**

*(0

.32)

�1.

16**

*(0

.32)

Per

pet

rato

rin

Res

pon

sibl

eC

are

1.08

***

(0.3

5)1.

25**

*(0

.41)

1.43

***

(0.3

6)1.

49**

*(0

.37)

AC

Cm

embe

r0.

21(0

.30)

0.04

(0.3

8)0.

00(0

.00)

0.00

(0.0

0)R

esp

onsi

ble

Car

em

embe

r�

0.53

(0.3

3)�

0.59

(0.4

3)�

0.30

(0.3

6)A

sset

s0.

01(0

.03)

�0.

002

(0.1

0)0.

02(0

.03)

�0.

03(0

.05)

0.02

(0.1

2)0.

02(0

.12)

Mil

lsra

tio

�0.

20*

(0.1

0)R

esp

onsi

ble

Car

ep

erp

etra

tor3

Res

pon

sibl

eC

are

mem

ber

�0.

40(0

.37)

Non

–Res

pon

sibl

eC

are

per

pet

rato

r3

Res

pon

sibl

eC

are

mem

ber

0.01

(0.4

7)

Con

stan

t1.

28**

*(0

.11)

0.34

*(0

.18)

n30

,751

30,7

5130

,751

30,7

5130

,751

26,1

3930

,751

30,7

51R

20.

010.

010.

01a

0.01

a0.

010.

010.

02a

0.02

a

Lik

elih

ood

-rat

ioch

i-sq

uar

e(c

omp

aris

onm

odel

)44

.59*

*(1

)16

1.63

**(2

)12

.49*

(3)

Acc

iden

t(f

our-

dig

it)

Yes

Yes

Yes

Yes

Yes

Ind

ust

ry(t

wo-

dig

it)

Yes

Yes

Fir

mY

esH

eckm

anY

esY

esY

ear

Yes

Yes

aW

ith

in-R

2(�

)re

por

ted

for

the

fixe

d-e

ffec

tsan

alys

is.

*p

�.0

5**

p�

.01

tively capturing the time period before the Bhopalaccident and after the formation of ResponsibleCare. In support of Hypotheses 2 and 3, our resultssuggest that spillovers from accidents indeed in-creased after Bhopal and diminished after the for-mation of Responsible Care. In the intervening pe-riod, an average accident could be expected toreduce the stock price of other firms with opera-tions in that industry by about 1.1 percent. Wefound that before Bhopal, this loss was only about0.3 percent, and that after the creation of Respon-sible Care it was negligible. The increased spill-overs after the Bhopal accident provide corroborat-ing evidence for the observations of industrymembers. Speaking after the Bhopal accident, DanBishop, then Monsanto’s director for environmen-tal communication, remarked, “Every chemical in-cident becomes a national story now. A minor spillbecomes front page stuff, and that tends to exagger-ate the event and reinforce the public’s concern”(Gibson, 1985b: 90).

In model 3, we included fixed effects and addi-tional variables to account for unobserved differ-ences among the industries in which the accidentsoccurred, as well as differences in the characteris-tics of the firms. In model 4, we conducted analternative analysis in which we added a fixedeffect for each firm. Because this would be collinearwith a measure of industry, we removed the indus-try effects. In both models, we again find supportfor Hypotheses 1, 2, and 3. Accidents do causespillover harm; this harm increased following acrisis in the industry; and it decreased after theformation of a self-regulatory institution.

We begin our exploration of the mechanism ofResponsible Care with model 5. To explorewhether Responsible Care provided a general ben-efit to the industry or only acted when one of itsmember firms was responsible for an accident (Hy-pothesis 4), we included additional interactionterms to capture spillovers from an accident at anACC member firm in both time periods. In supportof Hypothesis 4, we obtained a positive and signif-icant coefficient for the variable identifying a per-petrator as a member of the program, indicatingreduced harm to firms in our sample when an ac-cident occurred at an ACC firm after the creation ofthe Responsible Care program. Interestingly, theinclusion of this variable also reduced the signifi-cance of the coefficient for our variable denotingthe Responsible Care time period (post–Respon-sible Care). Thus, once we capture (through ourinclusion of “perpetrator in Responsible Care” onehypothesized mechanism through which the pro-gram could have provided benefit, we no longerfind significant evidence that the program provided

a general benefit to the industry through anothermechanism.

Our argument that information disclosure may bethe mechanism by which Responsible Care reducedspillover harm when one of its members experi-enced an accident is corroborated by the contem-poraneous reports of industry experts. After touringthe site of a chemical plant belonging to Exxon (aResponsible Care member) in his city, a city man-ager noted a change in how he viewed this plant:

I don’t harbor the fears that I had. I have learnedabout what they do. I hadn’t realized the safetyprecautions, the amount of testing of final products,the monitoring of air and water that goes on. But Idon’t think this will eliminate all skepticism. It’sludicrous to think that industry is going to be safe allthe time. But the fact that they have been open andhonest is extremely important to me as city man-ager. (Heller, 1991: 82)

Others reported that when accidents inevitably didoccur, Responsible Care coordinated a quick re-sponse. “Mutual assistance is on all Responsible Carepractitioners’ lips, with large firms helping out smallfirms an important dynamic” (Heller & Hunter, 1994:31). Richard Doyle, vice president of ResponsibleCare, quoted in Begley (1994), described such effortsas operating out of a “war room”:

The emergency response effort in the war room alsoled to an upgrade in Chemtrec, recognized as thechemical emergency response center in the U.S. Itwas established in 1972 to provide timely informa-tion and connect emergency responders with indus-try experts on the chemicals they were dealing with.After Bhopal, CMA set about upgrading Chemtrec’soperations and improving its mutual assistance ac-tivities. (Begley, 1994: 33)

We found no evidence to support Hypothesis 5,which states that an institution provides additionalprotection from spillover harm to its members. Wetested this hypothesis by including a dummy vari-able for participation in the Responsible Care pro-gram. As shown in model 5, the coefficient forResponsible Care member is negative and insignif-icant, indicating that members received no morebenefit than nonmembers. In the section below onrobustness testing, we discuss an alternative test ofthis hypothesis, in which we separate spillover toResponsible Care members from accidents at (1)other members and (2) nonmembers. This analysisalso failed to support Hypothesis 5.

Robustness Testing

To ensure the robustness of our analysis, wespecified several alternative models. First, we ac-


counted for endogenous choice processes based onunobserved fixed firm differences through the useof a Heckman correction technique. In model 6, wereport the result for the second stage of a fullyspecified model. In the first stage of this specifica-tion we used a probit analysis to estimate the ten-dency of firms to participate in the ACC and Re-sponsible Care in each year. The probit for the firststage (using 1990 as the example) is shown in Ap-pendix C. The probit results are consistent withKing and Lenox (2000) and suggest that ACC mem-bers tend to be in sectors with more toxic emissionsand more focused in chemicals, and to have moreassets. We controlled for the effect of endogenousdecisions to participate in ACC and ResponsibleCare by including the Mills ratio obtained from theprobit analysis as a variable in the second-stageregression. The Mills ratio represents the selectionhazard for the treatment (participation in ACC/Re-sponsible Care) occurring for a given firm in a givenyear. The coefficient estimations from model 6 pro-vided confirmatory evidence of the robustness ofour findings. Similarly, conducting an identicalmodification to models 1, 2, and 3 confirmed thefindings already reported. To account for unob-served firm differences, we also ran all of the mod-els (except those with Heckman corrections) usingfirm fixed effects and obtained consistent results.8

In model 7, we did not analyze the effect of timeperiods so as to allow the use of year fixed effects toaccount for any underlying macroeconomic changesthat might distort our results. Once we included theseyear effects, we had to remove the time perioddummy variables. As shown, an analysis with yearfixed effects again suggests that Responsible Care pro-vided a benefit to the industry by reducing the spill-over effect of an accident at a firm participating in theResponsible Care program. Thus, we again confirmedour support for Hypothesis 4.

In model 8, we further explore our failure to findsupport for Hypothesis 5 by including a new opera-tionalization of spillover effects to ResponsibleCare members. These two variables separate spill-over harm to members from (1) accidents at anotherResponsible Care member firm and (b) accidents ata nonmember firm. The coefficients estimated forboth variables were insignificant, and so we againfound no support for Hypothesis 5.

We conducted additional robustness testing todetermine if our time period dummy variables (pre-Bhopal and post–Responsible Care) might be cap-turing some other temporal effect. First, we testedwhether differences in the frequency of reportingmight influence responses to accidents. To rule outthis possibility, we specified models that includedmeasures of the accident prior to the one underconsideration. We included the days since the ac-cident, whether an employee was killed, andwhether an employee was injured. Models withadditional variables confirmed the sign and signif-icance of the reported results. We also exploredwhether we were simply capturing a wearing off ofstakeholder concern as the Bhopal accident becamea distant memory. We tested alternative modelswith a variable measuring linear and logged timesince Bhopal in days. Coefficient estimates on bothvariables were not statistically significant. The logform of the time estimate is highly correlated withthe pre-Bhopal variable (� � �0.81) and thus itreduces the significance of this measure in somemodels. Neither variable provided significant addi-tional explanatory power.

Throughout our analysis, we explain little of theobserved variance in stock prices—between 1 and2 percent. This is not surprising. Our method es-sentially removes fixed differences in firm stockprices, leaving only noise and the effect of newinformation about a firm. When we conductedfixed-effects analysis, we further removed twotypes of industry effects, and we report only ourability to explain the remaining variance withingroups. We evaluate the effect of only one type ofnews; clearly, numerous additional factors play animportant role in determining stock prices. How-ever, so long as this other news is not correlatedwith our events and uncaptured by our variables,our estimates should be unbiased. As discussedearlier, we carefully screened out events that werecontemporaneous with other corporate news.Moreover, the robustness of our analysis to multi-ple specifications and controls suggests that wehave significant and stable estimates.

It might seem that the effects found in our anal-ysis are too small to justify the creation of aself-regulatory institution. Indeed, we are agnos-tic regarding whether or not these effects drovethe observed behavior. We do believe that firmswere concerned with the potential for larger ac-cidents to have a serious effect on market values,and industry experts confirmed that these fearsprovided some of the impetus for self-regulation.We also believe that the response of the stockmarket to each of the smaller accidents analyzed

8 Interestingly, the coefficient estimate for the inverseMills Ratio is consistently negative. This suggests thatfirms with unobserved attributes that tended to causethem to join RC also tended to experience greater spill-over harm from accidents at other firms.


in our study provides a useful test of the func-tioning of the institution.

In summary, our analysis suggests that Bhopalindeed increased the interdependence of firmswith chemical operations in such a way that anaccident at one would have a negative effect onanother. Our analysis also suggests that Responsi-ble Care reduced this spillover effect, but it did sonot by insulating its members from the conse-quences of accidents at other facilities but byreducing the industrywide consequences of acci-dents at Responsible Care facilities.

DISCUSSION AND CONCLUSION

In this article, we have explained industry self-regulation occurring when physical commons areabsent. We posited that firms in modern industriesshare in a nonphysical type of commons—whatsome have termed a “reputation commons” (King etal., 2002)—that stems from the difficulty that stake-holders face in distinguishing the relative perfor-mance of individual firms and from the applicationof arbitrary or industrywide sanctions. We hypoth-esized that the risks associated with this commonscan become particularly acute following a majorindustry crisis. We further hypothesized that in-dustries create self-regulatory institutions as ameans of ameliorating this heightened threat ofshared sanctions.

Through a longitudinal analysis, we found thatfirms in the U.S. chemical industry did face suchan industry commons and that the shared sanctionsstemming from this commons became more severeafter the industry suffered a major crisis caused byan accident in Bhopal, India. Furthermore, wefound that this increased risk of shared sanctionspreceded the creation of the industry’s self-regula-tory program, Responsible Care, and that Respon-sible Care was able to ameliorate industrywideharm from the errors of individual chemical firms.Thus, in finding that an aggravated commons prob-lem led to the formation of a governing institutionand that the institution operated to reduce thisproblem, our study provides a “good functional-ism” perspective on the role of self-regulation inmodern industry.

We also explored the mechanism through whichResponsible Care functioned. We found that it re-duced industrywide harm resulting from the errorsof Responsible Care members, but we found noevidence that it provided members more protectionfrom spillovers than nonmembers. Thus, memberfirms appear to have provided a public benefit tothe industry, and the need to coordinate this ben-

efit may have provided a key motivation for theinstitution’s creation.

In Mending Wall, the poet Robert Frost wrote of atype of self-regulation of a physical commons, inthe form of a tradition that caused neighbors tocooperate in the creation of stone walls betweentheir properties. Each winter, storms and huntersknocked down some stones in these walls, andevery spring Frost would “let [his] neighbor knowbeyond the hill” that they need to “meet to walk theline, and set the wall between us once again.” But,as he worked on the wall one year, he wonderedwhy they were remaking it and asked his neighbor:

He only says, “Good fences make good neighbors.”Spring is the mischief in me, and I wonderif I could put a notion in his head:“Why do they make good neighbors? Isn’t itwhere there are cows?”But here there are no cows.Before I built a wall I’d ask to knowwhat I was walling in or walling out, . . .

Our analysis suggests that firms in the chemicalindustry, like Frost and his neighbor, share in theconstruction of a wall between them. We foundthat they “wall in” the effects of their own acci-dents rather than “wall out” the effects of others’.Responsible Care appears to serve the function ofensuring that each firm maintains its walls and soprotects its neighbors from the harm its accidentscould otherwise cause the broader industry.

Though skepticism about cooperative solutionsto commons problems has a lengthy history inscholarship (Hardin, 1968; Lloyd, 1833/1968), asillustrated in Frost’s poem, traditions of mendingfences to prevent spillover harm have a lengthyhistory in practice. Societies all over the worldhave developed norms to ensure that each memberacts to protect his neighbor. Indeed, the very nor-malcy of such traditions represents the culminationof Frost’s poem. Speaking of his neighbor, Frostwrites, “He will not go behind his father’s saying/And he likes having thought of it so well/ He saysagain, ‘Good fences make good neighbors.’ ”

Our analysis suggests that analogs to such tradi-tional responses can be found in modern indus-tries. Firms unite with rivals to ensure that eachprotects the other from a future problem. Becausethey are “walling in” their own effect on theirneighbors, firms cannot achieve such results inde-pendently. Rather, at-risk firms must come togetherto create an institution that helps ensure that eachprotects its neighbor, so that, in the aggregate, allare subject to less harm.


Implications

Our study suggests a revised interpretation ofrecent research on industry self-regulation. As dis-cussed earlier, many of these studies show thatself-regulatory programs have not measurably im-proved firm performance (Howard et al., 2000; King& Lenox, 2000; Rivera & de Leon, 2004; Rivera, deLeon, & Koerber, 2006). As a result, scholars havetended to conclude that such programs are a failureand have blamed such failure on the inability ofindustry associations to control the behavior ofmember firms. Our analysis suggests that thisskepticism may be based on an incorrect assump-tion about the institutions’ function. Rather thanacting as a means of improving firm performanceor of signaling unobserved attributes, these insti-tutions may be acting to directly reduce the prob-ability of stakeholder sanctions by encouragingfirms to engage in greater outreach and commu-nication. From contemporaneous accounts, weknow that such outreach occurred following theformation of Responsible Care. Using quantita-tive data over a number of years, we found evi-dence consistent with a hypothesis that such out-reach and communication may have a positiveeffect on stakeholder relations, regardless of theperformance of the member firms, and so can aidin reducing spillover harm.

Our research also suggests that an industry canmaintain a self-regulatory institution, even when itprovides benefit to nonmembers. Our findings sug-gest that the institution examined here protectedall firms from the errors of its members. Thus,participants in the program provided a publicgood to the industry. Despite the incentive of afree ride, however, firms agreed to participateand (over the years of our analysis, at least) theprogram provided a benefit to the industry. Thus,in accordance with Ostrom’s (1990) work demon-strating that actors can self-regulate to avoid de-struction of physical commons, we have demon-strated that firms can voluntarily come togetherto protect an intangible industry commons, de-spite the risk of free riding.

For policy makers, our research reveals that pri-vate institutions may substitute in part for publicregulation on information disclosure. The need toprevent spillover harm can help drive the creationof institutions that require the disclosure of infor-mation to stakeholders. Thus, government pro-grams on information disclosure should be ana-lyzed by considering both their direct effect on firmbehavior and their indirect effect on the formationand function of self-regulatory institutions.

Limitations and Future Research

Although our study addresses several long-standing issues, it also raises several new ones.We used stock price movements to measurechanges in stakeholder expectations about firmsover time, following specific industry events.This methodology did not allow us to observe themechanisms that created changes in stakehold-ers’ expectations of a firm’s future performance.We did not directly observe, for example, theprovision of information from the firm to thesestakeholders. In our particular empirical setting,we noted that the CAER program, which is at thecore of Responsible Care, requires members toengage in significant communication with stake-holders, but we did not actually measure thedegree to which members abided by these re-quirements. In future research, we hope to di-rectly evaluate differences in communicationrates and styles among participants and nonpar-ticipants. We encourage others to further inves-tigate how outreach and communication reducespillovers rather than focusing on performancedifferences between members and nonmembersof self-regulatory industry institutions.

In our study, we described industry self-regula-tion as a private decentralized institution, but ittypically involves some central authority. In ex-amples such as Responsible Care, there exists agoverning body that oversees compliance withthe program’s codes. However, compliance is of-ten gauged through self-reporting, and punish-ment tends to be limited or nonexistent, givenantitrust concerns, as well as the institution’sdesire to retain as many members as possible.Thus, even in exemplary and robust instancessuch as Responsible Care, industry self-regula-tion tends toward a decentralized model, relyingon peer pressure for compliance. Nonetheless, wecould also have categorized Responsible Care as ahybrid form of private institution, since it con-tains both centralized and decentralized aspects.

This argument suggests that Ingram and Clay’s(2000) centralized-decentralized dichotomy for in-stitutions might better be treated as a continuum,and it raises the question of how institutionschoose to position themselves along this contin-uum, both initially and over time. Our study ad-dresses temporal changes, but it does not addresschanges within the ACC program itself. Responsi-ble Care started as a primarily decentralized insti-tution, but over time it has shifted toward morecentralized authority. The program evolved overthe 1990s as new codes were hammered out and


promulgated to members. After our study timeframe ended, Responsible Care’s leadershipchanged—in part because of several articles thatsuggested the program had not reduced the pollu-tion generated by its members—and reportedly,with the new leadership, the “velvet glove cameoff” and some members were disciplined for theirlax behavior (Reisch, 1998). Future research shouldinvestigate how self-regulatory programs balancecentralization and decentralization and the strin-gency of enforcement in order to attract memberswithout diminishing the legitimacy of the institu-tion, as well as how this balance changes over time.The drivers of such changes are poorly understood.Weak enforcement may be necessary to attractmembers, yet the appearance that standards areenforced may be essential to maintaining the legit-imacy of a program in the eyes of observers. Infuture research, we hope to explore the drivers andmechanisms of these changes.

Finally, our study does not resolve the issue ofwhy firms choose to participate in self-regulatoryprograms. If participating firms essentially “wallin” their spillovers, safeguarding the rest of theindustry, then an institution provides a pure publicgood. Given the dominating incentive to free rideon pure public goods (Olson, 1965), how then doesthe institution hold together? Scholars have sug-gested that the incestuous nature of the chemicalindustry allows bilateral sanctions that enforce par-ticipation. According to Rees, “The chemical in-dustry is its own best customer” (1997: 489), andlarge firms use their power over subordinate sup-pliers as “leverage” (Gunningham, 1995: 85) to ob-tain their compliance. Yet, aside from a few anec-dotes, such sanctions remain unobserved. In futureresearch, we hope to further explore the centripetalforces that hold this and similar institutionstogether.

Despite these limitations, our research makes asignificant contribution to emerging scholarship onself-regulatory institutions and broadens under-standing of what constitutes a commons problemand how such problems may be resolved. It showsthat exchange problems caused by shared reputa-tion and risk of sanction exist in modern industries.It shows that a major crisis can intensify problemsassociated with this industry commons. Finally, itshows that when faced with exchange problemscaused by these commons, the choice is not be-tween “Leviathan or oblivion” (Ophuls, 1973: 215).Industry members can take matters into their ownhands and repair shared problems by forging a newinstitution.

REFERENCES

Acheson, J. M. 1988. The lobster gangs of Maine.Hanover, NH: Dartmouth University Press.

Acheson, J. M., & Knight, J. 2000. Distribution fights,coordination games, and lobster management. Com-parative Studies in Society and History, 42(1):209–238.

Barnett, M. L. 2006. Finding a working balance betweencompetitive and communal strategy. Journal ofManagement Studies, 43: 1753–1773.

Barnett, M. L., & Hoffman, A. J. 2008. Beyond corporatereputation: Managing reputational interdependence.Corporate Reputation Review, 11: 1–9.

Baron, D. P. 1995. The nonmarket strategy system. SloanManagement Review, 37(1): 73–85.

Barret, S. 2000. International environmental agreements:Feasibility, efficiency, stability. In H. Siebert (Ed.),The economics of international environmentalproblems: 111–124. Tubingen: Mohr Siebeck.

Begley, R. 1994. After Bhopal: A CMA “war room,” newprograms, and changed habits. Chemical Week, De-cember 7: 32.

Berkes, F. 1989. Common property resources: Ecologyand community-based sustainable development.London, U.K.: Belhaven Press.

Blacconiere, W., & Patten, D. 1994. Environmental dis-closures, regulatory costs, and changes in firm value.Journal of Accounting and Economics, 18: 357–377.

Blackman, A., & Boyd, J. 2002. Tailored regulation: Willvoluntary site-specific environmental performancestandards improve welfare? Southern EconomicJournal, 69: 309–326.

Brown, S., & Warner, J. 1980. Measuring security priceperformance. Journal of Financial Economics, 8:205–258.

Brown, S., & Warner, J. 1985. Using daily stock returns:The case of event studies. Journal of Financial Eco-nomics, 14: 3–31.

CCPA. 2007. http://www.ccpa.ca/ResponsibleCare/, ac-cessed July 1.

Coombes, P. 1991. Union Carbide. Chemical Week, July17: 44.

Corbett, C. J., Montes-Sancho, M. J., & Kirsch, D. A. 2005.The financial impact of ISO 9000 certification in theUnited States: An empirical analysis. ManagementScience, 51: 1046–1059.

Dawson, N. L., & Segerson, K. 2005. Voluntary environ-mental agreements with industries: Participationincentives with industry-wide targets. Working pa-per, University of Connecticut.

Ember, L. R. 1995. Responsible Care: Chemical makersstill counting on it to improve image. Chemical andEngineering News, May 29: 10–18.


EPA. 1999. Introduction to the National PretreatmentProgram (EPA-833-B-98–002). Washington, DC:U.S. Government Printing Office.

Fama, E., & French, K. 1992. The cross-section of expectedstock returns. Journal of Finance, 47: 427–465.

Fombrun, C. 1996. Reputation: Realizing value from thecorporate image. Boston: Harvard Business SchoolPress.

Furger, F. 1997. Accountability and systems of self-gov-ernance: The case of the maritime industry. Law andPolicy, 19: 445–476.

Furman, J., & Stern, S. 2006. Climbing atop the shoul-ders of giants: The impact of institutions on cumu-lative research. National Bureau of EconomicResearch working paper no.12523, Cambridge, MA.

Galvin, L. 2007. Outbreak linked to spinach forces reas-sessment of food safety practices. Amber Waves:The Economics of Food, Farming, Natural Re-sources and Rural America, 5(3): 25–31.

Gibson, W. D. 1985a. After Bhopal: First aid for chemi-cals’ public image. Chemical Week, February 20:21–22.

Gibson, W. D. 1985b. Olive branch activity prevails afterInstitute. Chemical Week, October 16: 90–91.

Granovetter, M. 1985. Economic action and social struc-ture: The problem of embeddedness. AmericanJournal of Sociology, 91: 481–510.

Greening, D., & Johnson, R. 1997. Managing industrialand environmental crises: The role of heterogeneoustop management teams. Business & Society, 36:334–361.

Greenwood, R., Suddaby, R., & Hinings, C. 2002. Theo-rizing change: The role of professional associationsin the transformation of institutionalized fields.Academy of Management Journal, 45: 58–80.

Gunderson, L. H., Holling, C. S., & Light, S. S. 1995.Barriers broken and bridges built. In L. H. Gunder-son, C. S. Holling, & S. S. Light (Eds.), Barriers andbridges to the renewal of ecosystems and institu-tions: 489–533. New York: Columbia UniversityPress.

Gunningham, N. 1995. Environment, self-regulation, andthe chemical industry: Assessing Responsible Care.Law & Policy, 17: 57–108.

Gunningham, N., & Rees, J. 1997. Industry self-regula-tion: An institutional perspective. Law & Policy, 19:363–414.

Gupta, A., & Lad, L. 1983. Industry self-regulation: Aneconomic, organizational, and political analysis.Academy of Management Review, 8: 416–425.

Hamilton, J. 1995. Pollution as news: Media and stockmarket reactions to the TRI data. Journal of Environ-mental Economics and Management, 28: 98–113.

Hannan, M., & Carroll, G. 1992. Dynamics of organiza-tional populations: Density, legitimation, and com-petition. New York: Oxford University Press.

Hardin, G. 1968. The tragedy of the commons. Science,162: 1243–1248.

Heller, K. 1991. Public outreach: The stakes are high.Chemical Week, July 17: 81–84.

Heller, K., & Hunter, D. 1994. Responsible Care. Chemi-cal Week, July 6: 31.

Hill, J., & Schneeweis, T. 1983. The effect of Three MileIsland on electric utility stock prices: A note. Jour-nal of Finance, 38: 1285–1292.

Hoffman, A. 1997. From heresy to dogma: An institu-tional history of corporate environmentalism. SanFrancisco: New Lexington Press.

Hoffman, A. 1999. Institutional evolution and change:Environmentalism and the U.S. chemical industry.Academy of Management Journal, 42: 351–371.

Hoffman, A., & Ocasio, W. 2001. Not all events are at-tended equally: Toward a middle-range theory ofindustry attention to external events. OrganizationScience: 414–434.

Hornblower, M. 2000. Wake up and smell the protest.Time, 155(15): 58.

Howard, J., Nash, J., & Ehrenfeld, J. 2000. Standard orsmokescreen? Implementation of a voluntary envi-ronmental code. California Management Review,42(2): 63–82.

Ingram, P., & Clay, K. 2000. The choice-within-con-straints new institutionalism and implications forsociology. In K. S. Cook & J. Hagan (Eds.), Annualreview of sociology, vol. 26: 525–546. Palo Alto, CA:Annual Reviews.

Ingram, P., & Inman, C. 1996. Institutions, intergrouprivalry, and the evolution of hotel populationsaround Niagara Falls. Administrative ScienceQuarterly, 41: 629–658.

Jarrell, G., & Peltzman, S. 1985. The impact of productrecalls on the wealth of sellers. Journal of PoliticalEconomy, 93: 512–536.

Jiang, R. H. J., & Bansal, P. 2003. Seeing the need for ISO14001. Journal of Management Studies, 40: 1047–1067.

Katzenberg, D. 1985. Liability insurance: The added bur-den of Bhopal. Chemical Week, February 13: 30.

King, A., & Lenox, M. 2000. Industry self-regulation with-out sanctions: The chemical industry’s ResponsibleCare program. Academy of Management Journal,43: 698–716.

King, A., Lenox, M., & Barnett, M. 2002. Strategic re-sponses to the reputation commons problem. In A.Hoffman & M. Ventresca (Eds.), Organizations, pol-icy and the natural environment: Institutional and


strategic perspectives: 393–406. Stanford, CA:Stanford University Press.

Lenox, M. J. 2006. The role of private decentralized in-stitutions in sustaining industry self-regulation. Or-ganization Science, 17: 677–690.

Lloyd, W. F. 1833/1968. Two lectures on the checks topopulation. Oxford, U.K.: Augustus M. Kelley.

Maitland, I. 1985. The limits of business self-regulation.California Management Review, 27(3): 132–147.

Malkin, E. 1996. Pangs of conscience. BusinessWeek,July 29: 46–47.

McWilliams, A., & Siegel, D. 1997. Event studies in man-agement research: Theoretical and empirical issues.Academy of Management Journal, 40: 626–657.

Meyer, A. 1982. Adapting to environmental jolts. Admin-istrative Science Quarterly, 27: 515–537.

Meyer, A., Brooks, G., & Goes, J. 1990. Environmental joltsand industry revolutions: Organizational responses todiscontinuous change. Strategic Management Jour-nal (special issue): 93–110.

Mitchell, M. 1989. The impact of external parties on brand-name capital: The 1982 Tylenol poisonings and subse-quent cases. Economic Inquiry, 27: 601–618.

North, D. C. 1981. Structure and change in economichistory. New York: Norton.

North, D. C. 1990. Institutions, institutional change andeconomic performance. Cambridge, U.K.: Cam-bridge University Press.

Olson, M. 1965. The logic of collective action: Publicgoods and the theory of groups. Cambridge, MA:Harvard University Press.

Ophuls, W. 1973. Leviathan or oblivion? In H. E. Daly(Ed.), Toward a steady state economy: 215–230.San Francisco: Freeman.

Ostrom, E. 1990. Governing the commons: The evolu-tion of institutions for collective action. Cambridge,U.K.: Cambridge University Press.

Ostrom, E. 2000. Collective action and the evolution ofsocial norms. Journal of Economic Perspectives, 14:137–158.

Patten, D., & Nance, J. 1998. Regulatory cost effects in agood news environment: The intra-industry reactionto the Alaskan oil spill. Journal of Accounting andPublic Policy, 17: 409–429.

Prakash, A., & Potoski, M. 2006. The voluntary environ-mentalist: Green clubs, ISO 14001, and voluntaryenvironmental regulations. Cambridge, U.K.: Cam-bridge University Press.

Rees, J. 1994. Hostages of each other: The transfor-mation of nuclear safety since Three Mile Island.Chicago: University of Chicago Press.

Rees, J. 1997. The development of communitarian regu-

lation in the chemical industry. Law and Policy, 19:477–528.

Reisch, M. 1998. Industry ponders future of ResponsibleCare. Chemical and Engineering News, May 5: 13.

Rivera, J., & deLeon, P. 2004. Is greener whiter? Thesustainable slopes program and voluntary environ-mental performance of western ski areas. PolicyStudies Journal, 32: 417–437.

Rivera, J., de Leon, P., & Koerber, C. 2006. Is greenerwhiter yet? The sustainable slopes program after fiveyears. Policy Studies Journal, 34: 195–221.

Schotter, A. 1981. The economic theory of social insti-tutions. New York: Cambridge University Press.

Shrivastava, P. 1987. Bhopal: Anatomy of a crisis. Cam-bridge, MA: Ballinger.

Shrivastava, P., & Siomkos, G. 1989. Disaster contain-ment strategies. Journal of Business Strategy, 10(5):26–31.

Slovic, P., & Weber, E. U. 2002. Perception of risk posedby extreme events. Paper presented at the Risk Man-agement Strategies in an Uncertain World Confer-ence, Palisades, NY, April 12–13.

Spar, D. L., & La Mure, L. T. 2003. The power of activism:Assessing the impact of NGOs on global business.California Management Review, 45(3): 78–102.

Spence, A. M. 1973. Job market signaling. QuarterlyJournal of Economics, 87: 355–374.

Stefanadis C. 2003. Self-regulation, innovation, and thefinancial industry. Journal of Regulatory Econom-ics, 23: 5–25.

Terlaak, A., & King, A. 2006. The effect of certificationwith the ISO 9000 quality management standard: Asignaling approach. Journal of Economic Behaviorand Organization, 60: 579–602.

Tirole, J. 1996. A theory of collective reputations. Reviewof Economic Studies, 63: 1–22.

Vasi, I. B., & Macy, M. 2003. The mobilizer’s dilemma:Crisis, empowerment, and collective action. SocialForces, 81: 979–998.

Wade, R. 1988. Village republics: Economic conditionsfor collective action in South India. Oakland, CA:ICS Press.

Weber, J. M., Kopelman, S., & Messick, D. 2004. A con-ceptual review of decision making in social dilem-mas: Applying a logic of appropriateness. Personal-ity and Social Psychology Review, 8(3): 281–307.

Williamson, O. E. 1985. The economic institutions ofcapitalism: Firms, markets, and relational con-tracting. New York: Free Press.

Yu, T., Sengul, M., & Lester, R. 2008. Misery loves com-pany: The spread of negative impacts resulting froman organizational crisis. Academy of ManagementReview, 33: 452–472.


APPENDIX A

Calculation of Cumulative Abnormal Return (CAR)

To calculate CAR, we first calculated the relationshipbetween the value of each company’s stock and the mar-ket as a whole (measured by the CRSP value-weightedindex with dividends for the entire market):

Rit � ai � BiRmt � eit,

where Rit represents the value of the security i on day t,ai is a constant, Rmt represents the value of the marketportfolio for day t, Bi represents the beta of security i, andeit represents the error term. Beta is computed over theperiod t � �254 to t � �1, where t � 0 is the day of theevent.

The abnormal return of a stock is the difference be-tween the actual return of that stock and its expectedreturn. The abnormal return of security i at time t, ARit, is:

ARit � Rit � (ai � BiRmt).

The cumulative abnormal return for a firm, CARi, is thesum of abnormal returns over the event window:

CARi � �ARit.

To allow for continuous compounding when aggregatingthe abnormal returns, ln(1 � R) is used in place of R.Thus,

CARi � ��ln(1 � Rit) � (a � Bimln(1 � Rmt))].

Key to computation of CAR is determination of theevent window, the period of time over which to cumulateabnormal returns. Event studies commonly begin theevent window prior to the actual event announcement inorder to account for information leakage, but dangerousaccidents are, by nature, unanticipated events. Therefore,in this study, if the event occurred during trading hourson a trading day, the window begins that day; if not, thewindow begins with the next trading day.

While it is straightforward to choose the beginning ofthe event window for this study, it is less clear when toclose the window. The occurrence and magnitude ofevents sometimes take several days to become apparentto the market. News travels fast and markets update theirvalues nearly instantaneously, yet many characteristicsof major accidents take time to establish. For example,the enormous toll of Bhopal took many days to unfold,and the magnitude of the Exxon Valdez oil spill was notimmediately evident. Whereas a long event window in-creases the likelihood of capturing the full impact ofunfolding events, a long event window also increases theopportunity for intervening events to confound results(McWilliams & Siegel, 1997). Therefore, researchersshould use the shortest possible window that capturesthe fullest extent of an event. The bulk of the effectstended to occur within five days after the events. Thus,our dependent variable in this study is the cumulativeabnormal return on the fifth day after an event.

APPENDIX B

TABLE B1Direct Effect of Accidents on At-Fault Firmsa

Variable Model 1 Model 2 Model 3

Employee killed �2.60** (1.06) �2.25** (1.10)Employee injured �1.72 (1.15) �1.50 (1.15)Assets of

perpetrator0.73* (0.38)

ACC member �1.69† (1.20)Pre-Bhopal �1.33 (1.95)Post–Responsible

Care0.29 (1.17)

Constant �1.01* (0.51) 1.13 (1.06) �4.86 (3.92)

R2 .08 .16

a n � 92.† p � .10* p � .05

** p � .01

APPENDIX C

Probit Analysis of ACC Participation

TABLE C1Descriptive Statistics for All Years (1980–2000)a

Variable Mean s.d. 1 2

1. Industry emissions 9.58 2.522. Chemical focus 0.54 0.48 .103. Assets 4.49 2.37 .20 �.15

an � 5,073.

TABLE C2Results of Probit Analysis of ACC Membershipa

VariableMembership in

1990

Industry emissions 0.15** (0.05)Chemical focus 0.61* (0.30)Assets 0.40** (0.06)Constant �5.45** (0.64)

n 2762 74.24Pseudo-R2 .38

a We estimated separate probit models for each year from1980 to 2000 to obtain the Mills ratio and included it in thesecond-stage regression estimation to reduce the effect of non-random treatment. Inclusion of this term accounts for changes inthe expectation of the treatment coefficient but does not correctfor heteroskedastic changes in the error terms that result fromnonrandom treatment. We used a standard White’s method tohelp correct for these effects.

* p � .05** p � .01


Michael L. Barnett ([email protected]) is an associ-ate professor of strategic management and Exide Profes-sor of Sustainable Enterprise at the University of SouthFlorida. He received his Ph.D. in management and orga-nizational behavior from New York University. In hisresearch he tries to “enlighten” the notion of self-interestby investigating how, when, and if firms can use forecast-ing, real options, corporate social responsibility, industryself-regulation, trade associations, Zen teachings, and otherperspectives, techniques, and individual and collectiveforms of organizing to their long-term strategic advantage.

Andrew A. King ([email protected]) is a visiting associateprofessor in the Technology and Operations ManagementUnit at the Harvard Business School and an associateprofessor at the Tuck School of Business at Dartmouth.He received his Ph.D. from the Sloan School of Manage-ment at the Massachusetts Institute of Technology. Heholds degrees in engineering from the University of Cal-ifornia, Berkeley, and Brown University.


good fences make good neighbors: a longitudinal analysis...

Documents