causation, counterfactuals, and competitive …rodolphedurand.com/wp-content/uploads/2012/09/... ·...

Strategic Management JournalStrat. Mgmt. J., 30: 1245–1264 (2009)

Published online EarlyView in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/smj.793

Received 30 June 2005; Final revision received 10 June 2009

CAUSATION, COUNTERFACTUALS,

AND COMPETITIVE ADVANTAGE†

RODOLPHE DURAND1* and EERO VAARA2

1 HEC School of Management, Paris, Jouy en Josas, France2 HANKEN School of Economics, Helsinki, Finland

Causation is still poorly understood in strategy research, and confusion prevails around keyconcepts such as competitive advantage. In this paper, we define epistemological conditionsthat help dispel some of this confusion and provide a basis for more developed approaches. Inparticular, we argue that a counterfactual approach—one that builds on a systematic analysisof ‘what-if’ questions—can advance our understanding of key causal mechanisms in strategyresearch. We offer two concrete methodologies—counterfactual history and causal modeling—asuseful solutions. We also show that these methodologies open up new avenues in research oncompetitive advantage. Counterfactual history can add to our understanding of the context-specific construction of resource-based competitive advantage and path dependence, and causalmodeling can help to reconceptualize the relationships between resources and performance.In particular, resource properties can be regarded as mediating mechanisms in these causalrelationships. Copyright 2009 John Wiley & Sons, Ltd.

INTRODUCTION

Causation is a central, often debated issue in strat-egy research (Cockburn, Henderson, and Stern,2000; King and Zeithaml, 2001; Powell, Lovallo,and Caringal, 2006). Classically, causation oper-ates under the principle that events have causes andconsequences (De Rond and Thietart, 2007; Pow-ell et al., 2006). For some researchers, causationinvolves empirical inquiry that verifies or falsi-fies law-like relationships between key variables(Camerer, 1985; Montgomery, Wernerfelt, andBalakrishnan, 1989). For others, such causal rela-tionships are more the discourse of researchers and

Keywords: causation; counterfactuals; competitive advan-tage; resource properties; history; causal modeling*Correspondence to: Rodolphe Durand, HEC School of Man-agement, 1 rue de la Liberation, 78351 Jouy en Josas, France.E-mail: [email protected]† This is a fully co-authored paper; the authors are listed inalphabetical order

do not necessarily correspond to anything concreteor lasting in the post-modern world (Løwendahland Revang, 1998). Still others argue that cau-sation should not be reduced to correlation, butshould be analyzed at the level of structures andprocesses (Godfrey and Hill, 1995; Tsang, 2006;Tsang and Kwan, 1999). Finally, there are thosewho see causation as beliefs that have instrumen-tal value above anything else (Mahoney, 1993;Powell, 2003; Powell et al., 2006).

Our key concern is that current disparate inter-pretations of causation generate wide-open ques-tions that plague the future development of strat-egy as a scientific and applied discipline. Thesequestions include the imprecise nature of concepts(Godfrey and Hill, 1995), the ambiguous meaningsof measures and the misuse of statistical tech-niques (Bergh and Holbein, 1997; Boyd, Gove,and Hitt, 2005), and the generalizability of findings(Boyd, Finkelstein, and Gove, 2005). Hence, thereis a need for clarification of what causation is and

Copyright 2009 John Wiley & Sons, Ltd.

1246 R. Durand and E. Vaara

what it means for strategic management research.In particular, we need methodological solutionsto the dilemmas faced by research focusing oncompetitive advantage.

In this paper, we therefore advocate a counter-factual approach to causation in strategy research.Counterfactuals—questions regarding what wouldhave happened otherwise (Collins, Hall, and Paul2004; Lewis, 1973; Woodward, 2003)—can beseen as key parts of causal analysis, but theyhave seldom received explicit attention in strategyresearch. We begin with a review of different epis-temological perspectives on causation to clarify thebasis of causation. We highlight the problems asso-ciated with the positivist and constructionist views,which tend to dominate discussions on causationin the strategy field. We then take up realist andpragmatist perspectives, which provide us with theinsights needed to outline four conditions for anepistemological position that dispels some of theconfusion around causation and serves as the basisfor our counterfactual approach.

On this basis, we proceed by drawing on philo-sophical studies of counterfactuals (Collins et al.,2004; Lewis, 1973, 1986) and their applicationsin qualitative (Tetlock and Belkin, 1996, Tetlock,Lebow, and Parker, 2006) and quantitative anal-ysis (Morgan and Winship, 2007; Pearl, 2000) toadvance our understanding of causation in concreteterms. We propose two methodological solutionsfor such research: counterfactual history and causalmodeling. We show how counterfactual reasoningand methods can in particular advance research

on competitive advantage. We argue that coun-terfactual methods can open up new avenues inhistorical analysis of constructions of resource-based competitive advantage and path dependence,but maintain that one should also focus atten-tion on the cognitive biases of causal reasoning.Furthermore, we show that applications of causalmodeling provide opportunities for new conceptu-alizations and empirical testing of the relationshipsbetween resources and performance. In particular,we suggest that resource properties can be regardedas mediating mechanisms in these causal relation-ships. Finally, we conclude by emphasizing therole of causation in strategy research, and make aplea for strategy research that will focus on com-monalities rather than exacerbate epistemologicaldifferences.

AN EPISTEMOLOGICAL BASISFOR CAUSATION

A proper understanding of the current debatesaround causation must start with a review of thecurrent dominant perspectives. In this section, wefirst review the positivist and constructionist per-spectives on causation that still tend to dominatediscussions on causation in strategy studies, andthen take up the realist and pragmatist alternatives.This review provides an epistemological basis forour counterfactual approach. The key character-istics of these perspectives are summarized inTable 1. This table shows fundamental differences

Table 1. Perspectives on causation in strategy research

Causation Competitive advantage Research objectives

Positivism Nomothetic view An object, the existence ofwhich is validated by studiesreporting positive effects onperformance

Empirical validation revealing thestatistical associations and causalrelationships between industrialconditions, resource position, andperformance

Constructionism Rejection of causation A social and discursiveconstruction with no obviouscausal status

Analysis of how social actors makesense and elaborate on competitiveadvantage as a construct in specificsettings

Realism Focus on generativestructures and causalmechanisms

A causal mechanism; cannotusually be observed orstudied directly

Analysis of the causal mechanismscreating (impairing) competitiveadvantage in specific settings

Pragmatism Instrumental view ofcausation; focus on itseffects on action

A concept of instrumentalvalue

Analysis of how the notion ofcompetitive advantage and relatedknowledge can effectivelycontribute to strategic action

Copyright 2009 John Wiley & Sons, Ltd. Strat. Mgmt. J., 30: 1245–1264 (2009)DOI: 10.1002/smj

Causation, Counterfactuals, and Competitive Advantage 1247

in conceptions of causation and the implicationsfor analyses of competitive advantage.

Causation is a central concept in positivism,which emphasizes nomothetic (law-like) regulari-ties between causes and effects (Hume, 1955). Pos-itivist researchers follow a hypothetico-deductivelogic: theoretical and falsifiable propositions areformulated and then tested for veracity usingappropriate empirical methods. In terms of the con-ception of causation, positivism is based on thedeductive-nomological model (Hempel, 1965). Byand large, positivism is the dominant view on cau-sation in strategy research (Bergh and Holbein,1997; Blaug, 1980; Boyd et al., 2005a; Camerer,1985; Montgomery et al., 1989; Simon, 1947).

The positivist view on causation, however, hasacknowledged limitations. First, because many fac-tors in the theories used in strategy are not directlyobservable or measurable, the status of these‘unobservables’ is a central problem (Godfrey andHill, 1995). While many positivists believe thatsuch unobservables are necessary for making pre-dictions but need not be included in empirical tests(Friedman, 1953; Godfrey and Hill, 1995), othersargue in favor of only observable, positive, fac-tors. However, in some cases, absence seems tohave important consequences.1

Second, positivist research relies on statisti-cal methods and tests. Nevertheless, observationalbiases, measurement errors, model misspecifica-tion, and the ambiguity of findings compromisethe ascertainment of causation (Bergh and Hol-bein, 1997; Boyd et al., 2005b; Denrell, 2003;Shaver, 1998). For instance, a direct empiricalassociation of resources with superior financialperformance does not preclude omission or mis-specification of links in the complex causal chain,such as industrial conditions, resource properties,competitive advantage, or superior performance(Cockburn et al., 2000; see also Tsang, 2006 aboutassumption-omitted testing). There are also empir-ical and methodological problems related to thedirection of causality and the reciprocal effectsbetween, for instance, ability and performance

1 For instance, Trevino and Weaver (2003) argue that positivistperspectives on organizational ethics are limited by the factthat ‘one of the major challenges of studying business ethicsis that success is often evidenced, in a more negative sense,by the “absence” of unethical or illegal conduct; but empiricalresearchers generally wish to account for increases in somephenomenon. It is difficult to explain variance in something thatis absent. . .’ (Trevino and Weaver, 2003: 331).

(Boyd et al., 2005a; Denrell, 2003). In this spirit,March and Sutton (1997) argue that financial per-formance may be both a consequence of variousbehaviors and an explanation for them.

Third, positivism is also confronted with thequestion of the ‘double hermeneutic’; that the phe-nomena under study have already been conceptu-alized (Giddens, 1984; Numagami, 1998). In strat-egy studies, we are dealing not only with naturalreality, but also with values, beliefs, and inter-pretations regarding, for instance, what constitutescompetitive advantage, which are then reflected inobservable behavior. In fact, competitive advan-tage was not discussed before this discourse gainedpopularity in the 1960s, along with the emergenceof strategy studies. This is obviously also true ofother disciplines: MacKenzie and Millo (2003) insociology, and Ferraro, Pfeffer, and Sutton (2005)in economics warn us against this risk of self-fulfilling theories. Such concerns of constructedcausality lie at the heart of the constructionistperspective, which we explore next.

Constructionist thinking has been embraced byphilosophers and scientists in various disciplines(Berger and Luckmann, 1966; Gergen, 1999;Knorr-Cetina, 1983; Latour and Woolgar, 1989),including management and strategy research (John-son and Duberley, 2003; Weick, 1989). Whilethere are different versions of constructionism(e.g., ‘constructivism’), most constructionists sharesome central assumptions. One such assumptionis that our knowledge of the world—and thusof organizations and strategies—is produced orinvented in social processes where linguistic ele-ments in particular play a key role (Gergen,1999). At the heart of this view is a widespreadphilosophical position that emphasizes the dis-tinctive interpretative nature of social phenom-ena and science. For example, Winch (1958) hasargued that the natural sciences deal only withexternal relations and the social sciences withinternal relations. According to this view, causa-tion (which, following Hume (1955), deals withexternal relations) has no relevance in social sci-ences. For von Wright (1971), the natural sci-ences deal with explanations—and thus causa-tion—whereas social sciences focus on under-standing and interpretations. Such views have alsobeen reproduced in epistemological discussionsin strategy research that point to the futility oftrying to establish causal relationships that are



context-dependent and complicated by the ongo-ing constructions of the social actors (Løwendahland Revang, 1998).

The constructionist position is, however, prob-lematic. In general, constructionism involves therisk of ontological relativism, that is, an inabilityto distinguish between more or less true theories orpropositions. At the extreme, this can mean that acompletely imaginary theory is accepted as true aslong as it is coherent and accepted by the scientificcommunity.2 Furthermore, it is difficult to ana-lyze any phenomena without notions such as causeor causal explanation. Constructionists themselvesuse causal language despite an explicit rejection ofcausal research. Thus, some notion of causation isalso needed for constructionist analyses. This hasled others, such as the scientific realists, to searchfor new epistemological formulations.

Discussions about realism, which began in the1970s, can be seen as an attempt to developan alternative to positivism and constructionism,especially in the case of causation (Bhaskar, 1975,1979; Mahner, 2001; Harre, 1970; Harre andMadden, 1975; Hull, 1988). This perspective hasalso received increasing interest in managementresearch in general (Ackroyd and Fleetwood, 2000,2004; Bacharach, 1989; Tsoukas, 1989; Van deVen, 2007) and strategy research in particular(Godfrey and Hill, 1995; Kwan and Tsang, 2001;Tsang, 2006; Tsang and Kwan, 1999). Accord-ing to the realist view, a constant conjunction ofevents is neither a sufficient nor a necessary con-dition for the manifestation of causality. Instead,realists focus on generative mechanisms or causalpowers and their effects. To ascribe causal powerto an object is to describe something that it willor can do in appropriate conditions by virtue ofits intrinsic nature (Harre and Madden, 1975). Insocial life, however, objects are often part of socialstructures. Thus, generative mechanisms reside instructures that endow them with specific causalpowers (Fleetwood, 2004). According to such atranscendental view, these mechanisms generatethe structure of causal associations between fac-tors. They exist and have causal potential even

2 As Huber and Mirosky (1997) put it, ‘Does every version ofan event have as much validity as every other version? This lastquestion is crucial because if the answer is yes, then scientificconfirmation or replication is pointless’ (1997: 1426). ‘The beliefthat social research is only one of many possible narratives takessociologists altogether out of the business of trying to gathervalid data. What would be the point?’ (1997: 1428).

when they are not actualized, because the effects ofthe many other causal processes and mechanismsat play may overshadow the particular effects ofthe causal mechanism under scrutiny (Bhaskar,1975; Kakkuri-Knuuttila and Vaara, 2007).

Although some realists incorporate ideas aboutthe socially constructed nature of observable events(Kwan and Tsang, 2001; Mir and Watson, 2000;Tsang, 2006), the capacity of the clear-cut real-ist position to embrace constructionist ideas islimited (Ackroyd and Fleetwood, 2004). In par-ticular, it would be unacceptable for a realist toconcur with radical constructionists and negate thecausal power of generative mechanisms, demot-ing them to discourse or fiction with no materialimplications. Moreover, for scientific realists, thedeepest level of understanding requires both theo-retical analysis and empirical studies focusing onunderlying structures and mechanisms.

However, the realist position also has its limi-tations. First, researchers cannot directly observethe underlying structures, processes, and mecha-nisms at play. Events could result from a com-bination of causes, some effective and othersimpaired by countervailing mechanisms. Disen-tangling effective causal powers from ineffectivemechanisms in observable and actual phenomenarequires demanding analytical and methodologicalcapabilities. Radical and critical realists are thussuspect of inoperative transcendentalism (Morganand Winship, 2007: 235–237). Second, the realistepistemology does not seem to provide the meansto reflect upon the role of researchers vis-a-viswhat and how they investigate their research objectand intervene in it.

This is why we need to consider insights pro-vided by pragmatist scholars interested in causa-tion. While there are different versions of prag-matism (Dewey, 1988; James, 1975; Peirce, 1992;Rorty, 1989), we draw here on the central ideasinitially promoted by Peirce (1992) regarding theinstrumental value of causation. In pragmatism,people’s conceptions and their sensations, expec-tations, and beliefs about the value of both knowl-edge and the inquiry process have a central role(Evered and Louis, 1981; Kaplan, 1964; Mahoney,1993; Mahoney and Sanchez, 2004; Powell, 2001,2002; Wicks and Freeman, 1998). Consequently,causation cannot be simply distinguished from aninterpretation of a succession of events, as themeaning people give to events in their descriptionsand narratives dictates comprehension. Hence,



descriptive and normative accounts and facts can-not be as easily separated as positivists or realistswould argue. Whereas some causal forces impingeon our freedom to act, they do not encroach uponthe meanings facts have for us or for differentcommunities of people (scientists or practition-ers). Hence, the unmediated causality and truthenvisioned by positivism is unreachable.

For pragmatists like Peirce (1992), causationprovides a satisfactory explanation for specificproblems because a causal representation helpshuman agents make successful inferences. To thisend, followers of Peirce’s pragmatism have devel-oped a stepwise method of inquiry that relieson problem identification and ‘abduction’ (Peirce,1992). Abduction combines the elaboration of spe-cific propositions with empirical observations. Ifthese propositions help to explain observations,our knowledge increases. Abduction allows forconstant movement between theory and empiricalinformation, for example in analyses of key causalprocesses and associations. This knowledge, how-ever, is not objective in the traditional sense, butis context-specific and dependent on the perspec-tive of the researcher. Thus, new observations andnew propositions can lead to different views thatchange existing knowledge.

However, critics of pragmatism denounce itsinstrumental use of causation and its acquiescenceto relativism vis-a-vis different versions of real-ity. In this view, a weakness of pragmatism wouldbe its inability to weigh the utility of differ-ent ideas, concepts, and propositions against eachother. Hence, the accumulation, transferability, andreproducibility of knowledge would appear sec-ondary in this approach, which might hinder con-struction of a generalizable corpus of theories andknowledge.

We do neither pretend in this paper to resolvedecades of arguments and disputes in and aroundcomplex epistemological notions such as causa-tion, nor do we propose a panacea for these lin-gering issues. However, drawing on the discussionabove, we define a basis for developing a coun-terfactual approach to causation. In particular, weenounce four conditions that constitute a commonground on which to develop our understanding ofcausation in strategy research:

1. Causation must be distinguished from mereconstant conjunction or statistical association;

by going beyond the regularity view on cau-sation, one can avoid some of the pitfalls ofpositivist tenets and deal with the construction-ist critique that targets the regularity view ofcausation.

2. Causation results from a complex activationof mechanisms and countervailing forces. Asargued by realists, there are three levels of cau-sation: generative mechanisms, actualization,and observable phenomena.

3. However, the standard transcendental view ofthe realists must be expanded to acknowledgethe importance of the interventions and con-structions of social actors. Actors can trigger orhinder the actualization of causal mechanisms,leading to expected (or unexpected) outcomes.They cannot change the causal mechanisms perse, but influence conditions that favor theiroccurrence. Actors’ interpretations also affecttheir actions and consequently the social andstrategic phenomena in question.

4. Causal explanations have instrumental value,which depends on their explanatory power, e.g.,whether these interpretations lead to increas-ingly better understanding of the phenomena inquestion.

This epistemological position also serves as thebasis for our counterfactual approach that focuseson a crucial but often neglected aspect of causa-tion: evaluation of whether a specific factor actu-ally causes a particular outcome by the construc-tion of counterfactual ‘what-if’ scenarios. Coun-terfactual reasoning, as we explain below, con-forms to the conditions just mentioned. It doesnot confound statistical association with causation,stresses degrees in causal relationships rangingfrom what is observable (the lowest degree) toimmutable mechanisms of event sequencing (thehighest degree), acknowledges human interventionin activating/deactivating causal paths; and is even-tually amenable to the interpretations of actors.Hence, we now proceed to outline a counterfactualperspective on causation that provides a concretemeans for advancing rigorous causal research instrategy.

A COUNTERFACTUAL APPROACHTO CAUSATION

Counterfactuals are conditional statements thatprobe the direction and stability of a relationship



between an event and its consequence (Collinset al., 2004; Lewis, 1973; Woodward, 2003). Theyattest to whether a change in a factor or eventis causally associated with changes in another.In other words, counterfactuals test different casescenarios and play a central, although somewhatvaried, role in both qualitative and quantitativeanalyses of causation by testing whether a spe-cific factor is necessary to produce a particulareffect (Woodward, 2003). While there are differenttheories of counterfactuals, Lewis’s (1973, 1986,2000) counterfactual theory of causation is prob-ably the most widely discussed in the philoso-phy of science. We start with this view, but alsoinclude insights of others on ‘possible worlds,’most notably those of political scientists and his-torians (Tetlock and Belkin, 1996, Tetlock et al.,2006), and work on causal modeling and inference(Morgan and Winship, 2007; Pearl, 2000).

The core idea in Lewis’s counterfactual analysisis the existence of ‘possible worlds.’ We can imag-ine alternative realities that bear varying degrees ofsimilarity to the actual one. Following Lewis’s log-ical demonstrations, depending on the conditionaloperators ‘might’ and ‘would,’ one can exploreimagined consequences of actions and of the pres-ence or absence of antecedents in more or lesssimilar worlds. Some logical principles such asnecessity, conditional strictness, and the similarityof possible worlds determine the status of the coun-terfactuals. For instance, it is logically impossiblefor two possible worlds to disagree with respect toa particular causal fact while agreeing completelywith respect to all noncausal facts. According toLewis, (1973: 9) ‘Si... [is] the set of all worldsthat are similar to at least a certain fixed degreeto the world i.’ In general, a counterfactual state-ment is true if ‘the material conditional of itsantecedent and consequent is true throughout Si’(Lewis, 1973: 9–10). An event Y depends causallyon a distinct event X if and only if both X and Yoccur, and if X had not occurred, then Y would nothave occurred either. Lewis put it as follows: ‘Wethink of a cause as something that makes a differ-ence, and the difference it makes must be a dif-ference from what would have happened withoutit. Had it been absent, its effects—some of them,at least, and usually all—would have been absentas well’ (Lewis, 1986: 160–161). In addition, Xis a cause of Y if the counterfactual conditionalsapplied to X and Y are of the proper type, that

is they neither regress ad infinitum nor imply fur-ther nonindependent conditions. Comparing propo-sitions in quasi-similar worlds thus enables one toseparate causal facts from noncausal ones and toreveal to which degree Y is dependent on X.

While Lewis’s initial views attribute a strongontological status to possible worlds, most philoso-phers would think of these alternative realities as aconvenient means of contrasting an actual courseof events with other possibilities. When it takesless of a departure from actuality to make theantecedent X true along with the consequent Ythan to make the antecedent X true without theconsequent Y, the counterfactual ‘if X were thecase, Y would be the case’ is true. Hence, toogreat a departure from the actual world obscuresthe comparison of causal relationships. In the for-mulation and actual analysis of counterfactuals, theconstructed possible worlds must provide a plau-sible alternative to the actual world. This principleallows researchers to link counterfactuals to fac-tors that ‘matter,’ not to just any kind of possiblecausal relation.

Historians (Ferguson, 1997; Fogel, 1964) andpolitical scientists (Tetlock and Belkin, 1996, Tet-lock et al., 2006) have used counterfactual reason-ing to examine events and phenomena of histori-cal significance and compared them with alterna-tive, imaginary realities. Tetlock and Belkin (1996)explain the basis for counterfactual research insocial studies as follows: ‘Counterfactual reason-ing is a prerequisite for any form of learning fromhistory. . .. To paraphrase Robert Fogel’s (1964)reply to the critics of “counterfactualizing” in the1960s, everyone does it and the alternative to anopen counterfactual model is a concealed one’(Tetlock and Belkin, 1996: 4). Ferguson (1997)argues that cautiously elaborated counterfactualsplay an important role in overcoming determinism,inevitability, and heroism in traditional historicalresearch. He stresses the central issue of select-ing which counterfactuals to focus on: ‘In short,by narrowing down the historical alternatives weconsider to those which are plausible —and henceby replacing the enigma of “chance” with the cal-culation of probabilities— we solve the dilemmaof choosing between a single deterministic pastand an unmanageably infinite number of possiblepasts. The counterfactual scenarios we thereforeneed to construct are not mere fantasy: they aresimulations based on calculations about the rela-tive probability of plausible outcomes in a chaotic



world (hence “virtual history”)’ (Ferguson, 1997:85). We will use these historians’ ideas in the nextsection to spell out how counterfactual history canbe applied in strategy studies.

Important and relevant work on the statisticalimplications of counterfactual reasoning has in turnfocused on causal modeling (Morgan and Win-ship, 2007; Pearl, 2000). For two decades philoso-phers and computer scientists from Carnegie Mel-lon University (Spirtes, Glymour, and Scheines,2000) and elsewhere (Morgan and Winship, 2007;Pearl, 2000; Salmon, 1998) have worked on math-ematical causal models that build on counterfactuallogic. These scholars have created an extensivecorpus centered on the study of causal models,namely, graphs that relate factors on which logi-cal, mathematical, and probabilistic properties canbe assessed. In a graph, arrows indicate the causalorder (see Figure 1). For instance, X → Y meansthat X causes Y. In the mediation chain X → Z→ Y, X, and Y are unconditionally associated,namely having knowledge on X will give someinformation on the likely value of Y. In mutualdependence, X ← Z → Y, X, and Y are alsounconditionally associated, but this time becausethey mutually depend on Z. In mutual causation, X→ Z ← Y, X, and Y are not unconditionally asso-ciated. Having knowledge on X does not provideinformation on how Y looks. Z is said to block thepossible causal effects of A and B on each other.

In this tradition, the determination of a causalinfluence relies on how a counterfactual situationcontrasts with an actual situation, that is, the com-parison of two individual situations, one of whichis observable and the other is not. Suppose thateach individual in a population is exposed to twoalternative states of a cause. If the outcome iscorporate performance, the population of interestcould be firms belonging to a given industrial sec-tor, and the two states whether or not a firm pos-sesses a strategic resource. The ‘treatment state’ isto possess such a resource; the ‘control state’ is

not to have it. Distinct sets of conditions charac-terize each alternative state that impacts the out-come of interest. For instance, the possession ofstrategic resources depends on distinct surroundingconditions, and each state influences corporate per-formance in different ways. Counterfactual logiccompares potential outcomes for each individual(namely a firm in our example) in each treatmentstate. Indeed, only one observation for each indi-vidual is possible at any point in time. Hence,one needs to construct counterfactual ‘what-if’ out-comes. For example, firms possessing strategicresources have a ‘what-if’ performance when theypossess only generic resources, whereas firms withgeneric resources have a ‘what-if’ performancewhen they possess a strategic resource. Therefore,for each individual, distinct outcomes result fromrelative exposure to the alternative treatment. Letus suppose that y1

i and y0i correspond to the poten-

tial outcomes for individual i in the treatment state(superscript 1) and in the control state (superscript0). The theoretical difference or contrast betweenthese two values enables one to approach the dis-tinctive influence of the treatment (namely possess-ing a strategic resource) on the outcome. However,y0

i is unobservable for individuals belonging to thetreatment group, whereas y1

i is unobservable forindividuals belonging to the control group. Hence,since these two values cannot be observed at theindividual level, one needs to contrast cause andeffects at the population level in order to completea full graph of relationships between the cause X(possessing a strategic resource), the outcome Y(corporate performance), the surrounding condi-tions Z, their relationships with the observed X andY, and potential unobservable variables that com-plicate the causal explanation of Y by X. As weexplain below, the combination of causal graphsand counterfactual testing and evaluation providevery useful tests of causal relationships that aremore powerful than classical statistic methods.

Simple path Mediation Mutualdependence

Mutualcausation

X Y X Y X Y

X Y

Z

Z

Z

Figure 1. Basic relationships



TWO METHODOLOGIESFOR COUNTERFACTUALCAUSAL ANALYSIS

How then can such counterfactual research be con-ducted in practice? In the following, we presenttwo methodologies that can be used in counterfac-tual causal strategy research: counterfactual historyand causal modeling. While there are other meth-ods that can be used in counterfactual analysis,these two provide useful examples of qualitativeand quantitative analyses that have been success-fully used elsewhere but seldom applied to strategyresearch.

Counterfactual history

Strategy research has been criticized for a lack ofhistorical perspective; that is, for making poor useof longitudinal comparative analyses (Booth, 2003;Kieser, 1994). This is also the case with researchon competitive advantage. The reasons for the lackof historical analysis include the ambiguity con-cerning the nature of causation in historical stud-ies and the problems encountered in dealing withalternative histories and retrospective biases. Wesee counterfactual historical case study researchas a methodology that can help in resolving thesechallenges.

The kinds of tests conducted in the natural sci-ences are not possible in social research in generaland strategy research in particular. Thus, there is aneed to use specific methods such as the construc-tion of counterfactuals to be able to examine whatcould have happened had the initial event not takenplace. As noted above, historians and politicalscientists (Ferguson, 1997; Fogel, 1964; Griffin,1993; Tetlock and Belkin, 1996) have developedthe use of counterfactual reasoning in qualitativecase studies. This method builds on the idea ofimagining the suppression of the occurrence of anevent to test the significance of causal mechanismsand paths. This testing involves the construction ofalternative scenarios and worlds, in other words,what could have happened had X not occurred(Tetlock and Belkin, 1996; Tetlock et al., 2006). Infact, counterfactuals are like thought experiments(De Mey and Weber, 2003; Lewis, 1973) or fiction(Tetlock et al., 2006; White, 1987) in the sense thatthey require the construction of ‘possible worlds.’Examples of such analyses range from reinterpre-tations of the industrial revolution and its causes

and consequences (Fogel, 1964) to reconstructionsof World War II (Ferguson, 1997) or our Westerncivilization (Tetlock et al., 2006).

There is no reason why strategy scholars couldnot follow these examples and use counterfactualhistory to advance understanding of causation instrategy research. In particular, research on com-petitive advantage prompts questions about whatwould happen if specific resources or capabilitiesdid not exist, if others existed, or whether thecompetitive advantage under scrutiny is neededto produce an outcome (e.g., superior financialperformance). Even though qualitative counterfac-tual analyses have been relatively rare in stud-ies of competitive advantage, they can elucidateprecisely such crucial questions.

While such qualitative research does not usu-ally proceed in clear-cut stages, we propose threegeneric steps to be followed when applying thisapproach to strategy research:

1. Identify critical events. Historical case studyresearch builds on event-causality; that is onan analysis of how specific events may becausally linked. A careful mapping of eventsand a thorough analysis of how these eventsrelate to broader and more general facts isthe first step in such analysis. Typically, thismapping involves choices as to which fac-tors to emphasize and which to leave out-side the core model. Such challenges char-acterize all process-oriented qualitative strat-egy research where one works with, compares,and distills data coming from various sources(Huber and Van de Ven, 1995; Langley, 1999),but are accentuated in causal analysis due tothe need to assess the impact of specific pro-cesses, mechanisms, and factors on others. Grif-fin and Ragin (1994) provide a systematic nar-rative method that combines a thick descrip-tion of interpretative research and generaliz-able causal explanations. In particular, Griffin(1993) has proposed an ‘event-structure anal-ysis’ methodology that links historical narra-tives to broader structures, leading to a com-prehensive understanding of the case in ques-tion within a broader framework. In its sim-plest form, such analysis involves outlining anevent time line that links case events to moregeneral phenomena and structures. This event-structure analysis can be conducted by meansof software such as Heise’s ETHNO program



(http://www.indiana.edu/∼socpsy/ESA/home.html) (Heise, 1989). This program helps topose ‘yes’/‘no’ questions regarding the impactof antecedent events or actions on subsequentones, and has been used by sociologists andorganization scholars (for examples, see Griffin,1993; Stevenson and Greenberg, 1998, 2000;Pajunen, 2004). Such event models provide thebasis for the next step.

2. Specify causal processes and mechanisms. Thisnext step is to specify causal explanations onthe basis of the event models. The key challengeis to focus attention on particular theory-basedcausal processes and mechanisms in the case inquestion. The theoretical ideas about processesand mechanisms can be derived from existingresearch (in a more deductive research design)or emerge out of theory development alongsidethe case analysis (in a more inductive researchdesign). This specification involves ‘isolation’where attention is focused on only the keycausal processes and mechanisms and not oth-ers (Maki, 1990). As there are numerous impor-tant interconnected factors, such isolation is nota trivial step in qualitative causal analysis, butone that usually involves difficult choices as towhat to include and exclude. In any case, theobjective is to spell out theoretically and empir-ically grounded arguments concerning funda-mental causal processes and mechanisms. Theycan usually be expressed in terms of hypothesesor propositions that then need to be tested withcounterfactual analysis.

3. Use counterfactuals to establish causation.Based on the identified potential causal struc-tures, processes, and mechanisms, the thirdstage involves contrasting the hypothesizedcausal explanations with alternative explana-tions. As explained previously, ‘possible world’counterfactuals (‘what could have happenedhad X not occurred’) play a central role in con-trasting hypothesized causal explanations withalternative ones. In particular, they serve ascontrastive explanations that are used to val-idate the causal claims (Maslen, 2004). Cru-cially, these counterfactuals should never bepure imagination, since their premises need tobe supported with theoretical arguments andempirical data that are logically consistent withthe causal hypotheses and propositions devel-oped in step 2 above. Even though the useof counterfactuals can vary greatly depending

on the topic at hand, the following principlesprovide useful guidelines for constructing con-trastive counterfactual explanations (see alsoTetlock and Belkin, 1996):

a. Conceptual clarity: The antecedent and conse-quent variables must be specified so that theyare conceptually distinctive and consistent withthe initial model (the hypotheses and proposi-tions that are created in step 2 (see above)).For instance, historical analysis can focus on theimpact of a strategic decision (e.g., an invest-ment or acquisition) on competitive advantageand financial performance. In that case, thecounterfactual scenario must be based on theabsence of such decision or on an alternativedecision, keeping other key variables as simi-lar as possible to the initial model. Similarly,historical analysis can examine whether par-ticular resources were the source of superiorperformance in a given time period. In the coun-terfactual scenario, the starting point is thenthe absence of such resources, with other keyvariables remaining the same.

b. Cotenability: Cotenability requires that theantecedent must logically imply its consequentand that there must be compatibility betweenall known facts. For instance, in a study of theimpact of a strategic decision on performance,the implications of the absence of such a deci-sion or an alternative decision must be logicalgiven all the other information around that case.

c. Historical consistency (‘minimal rewrite rule’):The specifications of antecedents must alter asfew established historical facts as possible. Ide-ally, the possible ‘imagined worlds’ should startwith the ‘real world’ as it is known beforethe counterfactual was developed, not requirerewriting long stretches of history, and not devi-ate too much from what we already know aboutthe key actors and circumstances. For example,in an analysis of the impact of a strategic deci-sion on competitive advantage, the alternativecounterfactual scenarios must not alter otherconditions, only the key decision in question,implying for instance that no decision was made(if this is plausible) or that other conceivabledecisions were carried out.

d. Theoretical consistency: The connecting prin-ciples between the antecedent and consequentshould be consistent with the relevant theo-ries. These theories can be established ones or



new ideas offering the basis for the contrastiveexplanations. The point is that the more clearlythe counterfactual scenarios are linked withalternative but conceivable theoretical explana-tions, the better the result validates the proposedcausal theorems. In some cases, the counter-factuals can serve as means to spell out dif-ferent and competing theorizations, but obvi-ously not all counterfactual analysis needs to betheory-testing in nature (Tetlock et al., 2006).

e. Generalizations and projectability: Connect-ing principles should be consistent withwell-established generalizations relevant to theantecedent-consequent link. In principle, thelinkages between antecedent and consequentshould not be overly complex, but build on rea-sonable inferences about the likely and possibleeffects of specific factors. For example, avail-able historical information on industry growth,performance, and profit margins can provide thebasis for qualitative or quantitative estimates incounterfactual scenarios. Thus, in an analysis ofthe impact of a specific decision on competitiveadvantage, the outcomes of the counterfactualscenarios (no decision or alternative decisions),can be estimated by using precisely such infor-mation. Projectability is an overall principle thatshould be followed in counterfactual analysis:one should be able to distinguish between coin-cidental generalizations that just happen to betrue at a particular time and place (and arethus unprojectable) and robust general mecha-nisms that are valid in a range of circumstancesand permit projections into the past and future(Goodman, 1983).

These counterfactuals are then used to validateor falsify hypothesized causal relationships, forexample, those regarding the existence of com-petitive advantage in a given setting. It shouldbe emphasized that this kind of qualitative his-torical analysis often follows an abductive logicthat combines deductive and inductive reasoning:reformulations of the initial models (step 2) andcontrastive counterfactual explanations (step 3) areneeded until the final causal model provides acredible account of the historical processes andmechanisms in question.

Applications. Counterfactual historical analysiscan be applied to a variety of questions in strat-egy research. Historical analyses can enlighten us

regarding the causes and consequences of compet-itive advantage in ways that emphasize contextualissues and cultural dependencies (Kieser, 1994).This opens up new avenues for studying issuessuch as the historical construction of resource-based competitive advantage (Bogner, Thomas,and McGee, 1996; Priem and Butler, 2001) andpath dependence (Booth, 2003; Lamberg andTikkanen, 2006). In particular, counterfactual his-tory can be seen as a means to avoid oversimpli-fications and excessive determinism in our inter-pretations of the role of specific strategic decisionsand key decision makers (for analogous argumentsin political history, see Ferguson, 1997; Tetlockand Belkin, 1996).

Counterfactual history also involves specificchallenges. To some opponents of historical coun-terfactual analysis, the imagination exercise aboutpossible worlds looks hopelessly subjective andcircular, while to others it appears arbitrary, purelyspeculative, and self-serving. Another objectiondeals with the idea that variables in which onecannot intervene cannot be scientifically analyzed.Hence, counterfactuals will border on fanciful con-jectures. Furthermore, some reject the postulateaccording to which ‘what-if’ scenarios can con-tribute scientific value since they are inherentlynon-testable. We do not respond to each criticism,but concur with Tetlock and Belkin: “We do notconclude that things are hopeless—that it is impos-sible to draw causal lessons from history. Rather,we conclude that disciplined use of counterfactu-als—grounded in explicit standards of evidenceand proof—can be enlightening in specific his-torical, theoretical, and policy settings” (Tetlockand Belkin, 1996: 38). Furthermore, it should beemphasized that strategy scholars usually deal withchains of events, the consequences of which do notaffect entire civilizations and are less remote intime than in most historical research. Thus, theseconcerns should be more limited in strategy stud-ies than in some other areas of social or historicalresearch.

However, such analysis must take into accountthe cognitive biases that characterize both man-agers’ retrospective accounts and researchers’explanations. It is crucial to pay attention to thesebiases, especially in the case of competitive advan-tage, since ambiguity is a fundamental compo-nent of sustainable advantage (King and Zeithaml,2001) and self-serving attributions are an inher-ent part of causal claims in an organizational



context (Powell et al., 2006). For instance, man-agers’ decisions usually lead them to overempha-size their own actions in successful ventures andto downplay their role in failures (Salancik andMeindl, 1984; Vaara, 2002). Past research has alsoshown that the illusion of control at the organiza-tional level leads to optimistic biases in estimatingfuture trends and can cause resource misallocations(Durand, 2003; Powell et al., 2006).

In building counterfactual explanations, the min-imum requirement is to critically analyze datasources to determine the extent to which such cog-nitive tendencies characterize people’s accounts.In particular, Tetlock and Belkin (1996) challengeresearchers to focus attention on 1) what is omittedin models that usually focus on dramatic change atthe expense of normality, 2) whether the choice ofcounterfactuals often tends to favor overly simplis-tic and convenient counterfactuals, 3) whether pre-dictability and controllability are overemphasizedat the expense of other explanations, 4) whetherneeds to avoid blame and to claim credit distortcounterfactual analysis, and 5) whether needs forconsolation and inspiration bias explanations.

Such cognitive biases can also be seen as a spe-cial object of study for causal strategy research.In particular, tendencies such as self-serving bias,illusion of control, blame attachment, and scape-goating have been studied in various applica-tions of attribution theory (Bettman and Weitz,1983; Heider, 1958; Salancik and Meindl, 1984;Weiner, 1986). A rare theoretical example in strat-egy research is provided by Powell et al. (2006),who show how managers’ perceptions are a keypart of ‘real’ causal ambiguity; in particular, howthe above-average effect increases causal ambi-guity, which then decreases the ability to lever-age competence, increases barriers to imitation,and augments rival substitution. There are manyother ways in which causal beliefs relate to causalexplanations (MacKenzie and Millo, 2003), andwe think that researchers in strategic manage-ment have only started to make these connectionsexplicit.

Causal modeling

Causal modeling includes a set of different meth-ods that deal with causal graphs and counterfac-tual testing. As mentioned earlier, the combinationof causal graphs and counterfactual testing and

evaluation provides powerful tests of causal rela-tionships. This approach to causation overcomesmany of the criticisms that were addressed in thesocial sciences in the reign of regressions duringthe 1990s: ignorance of temporality and context,superposition of covariates, and oversimplificationof causal linkages in a quest to establish the nextinteraction effect (Abbott, 2001; Hedstrom, 2005;Ragin, 2000). Causal modeling is obviously not theonly approach that has been developed to deal withsuch problems. For instance, taking a Bayeasianperspective has been found to be useful in variousfields of social science, including strategy research(Hahn and Doh, 2006; Hansen, Perry, and Reese,2004; Powell, 2001). However, for space limita-tions, we focus here on causal modeling as a par-ticularly fruitful method of counterfactual analysis.

This approach distinguishes statistical associa-tions from causal relationships. As Pearl states: ‘Inow take causal relationships to be the fundamen-tal building blocks both of physical reality and ofhuman understanding of that reality, and I regardprobabilistic relationships as but the surface phe-nomena of the causal machinery that underlies andpropels our understanding of the world’ (Pearl,2000, xiii–xiv). In a nutshell, causal relationshipsare more directional, more stable, and less depen-dent on intervention than statistical associations(Pearl, 2000; Spirtes et al., 2000; Salmon, 1998).A first key difference between a causal model anda probabilistic association concerns the directionof the relationships. Reversing the direction of therelationship between, for instance, x1 and x2 doesnot alter the structure of the relationship froma probabilistic point of view. If integrated withother factors xi , reversing the direction between x1and x2, yields an observationally equivalent net-work of probabilistic dependencies among factors.However, reading the associations in the oppositedirection may be neither theoretically meaningfulnor causally accurate. For example, time, logicalconditions, or theoretical considerations determinewhether such reversibility is feasible. Hence, weneed more than probabilistic information to deter-mine the direction of the link x1 → x2 (Pearl,2000: 19).

Second, causal relationships are more stable anddepend less than statistical associations on addi-tional knowledge about other factors. For exam-ple, the addition of a variable in an incompletestatistical model can change the value and signif-icance of prior estimated coefficients. In contrast,



the addition of a factor in a causal model doesnot dissipate a preexisting causal association. Togive another illustration, when regressors X arestatistically dependent, the effect of a regressorXj linked with other regressors, but not with theoutcome variable Y , may bias estimates of regres-sors X. In fact, even a variable Xj with no influ-ence whatsoever on Y may be given significantregression coefficients (Spirtes et al., 2000: 191).Causal modeling techniques take explicit accountof both these spurious effects and the conditioningof relationships on additional antecedents.

Third, causal relationships are not altered byinterventions in the models. Causal relationshipspossess an ontological robustness to changesaround the causal variables—for instance re-sources and capabilities in our case—that proba-bilistic relationships lack. The intuition behind thenotion of intervention as exposed by Pearl (2000)or Spirtes et al. (2000, chap. 7) captures differ-ences among states and causal relationships involv-ing antecedents. In our example, we can assumethat an intervention would introduce or suppressthe strategic properties of a firm’s resources (inother words, turning this factor on or off). Althoughnot observable in statistical terms, the causal powerof the relationship existing before the interven-tion remains operative in determining the stateof the firm even when it is no longer in effect(turned off). Different techniques are available tocapture the counterfactual value of these ‘what-if’ situations, and to estimate the magnitude ofeffects in situations with or without an activeantecedent.

Simply stated, causal model analysis proceedsin three steps: prediction, counterfactual reason-ing, and estimation. These steps form an ascend-ing “natural hierarchy of causal reasoning tasks”(Pearl, 2000: 38). We use a classical resource-based view (RBV) example to illustrate, in simpleterms, the implications of these three steps.

1. Develop predictions. This first step can be con-ducted by using the covariance matrix andappropriate statistical methods. This is the nor-mal starting point in strategy studies that focuson statistical associations between factors. InRBV terms, a typical question would be thefollowing: Would performance be abnormal ifthe firm did not possess strategic resourcesand capabilities? The structural properties of

causal relationships as defined by causal mod-els (directionality, stability, dependence, andboundary conditions) are not at stake in thistype of inquiry for which simple probabilisticapproaches are well suited (Pearl, 2000: 31).

2. Counterfactual reasoning. In their simplestform, counterfactual statements rephrase pre-dictions, since they convey the logical implica-tions of the classical predictions formulated inthe first step. Hence, counterfactual statementsare of particular relevance when there is uncer-tainty or disagreement about the nature andstructure of causal chains. They are more thana roundabout way of stating sets of predictions,since they focus on the causal mechanisms atwork as well as on the prevailing boundaryconditions. This second step thus focuses oncounterfactual ‘what-if’ questions. We providehere examples of two different types of counter-factual reformulation. Our illustrative questioncan now be posed as follows: Would perfor-mance be abnormal had the firm not possessedstrategic resources and capabilities, given thatfirm performance is, in fact, average and thefirm possesses strategic resources and capabili-ties? Another way of counterfactually reformu-lating the question assumes intervention: Wouldperformance be abnormal if we intervened tomake sure that the firm does not possess strate-gic resources and capabilities? Intervention isa common mental counterfactual operation thatconsists of hypothesizing changes in the statesof a Y variable when there are changing val-ues in variable X, an antecedent of Y. Sincea causal relationship is stable and invariant,the evaluation of intervention is simpler thanresorting to conditional probabilities that maybe modified in unknown proportions by thecontext of intervention. Intervention acts on afunction of the model instead of on an entireset of conditional probabilities. The effect ofthe intervention can be predicted by modifyingthe corresponding equations representing thecausal model and computing new probabilityfunctions. In some instances (e.g., experimen-tation) these operations can be effective andobservable.

It should also be noted that based on thestructure of data, the presence of unobservedeffects can be examined by such counterfac-tual analysis. Pearl (2000) suggests using com-putational techniques and algorithms (e.g., the



2a: Joint dependence 2b: joint dependence of X and Yon unobservable variable Z

2c: back-door path (blocked byconditioning on F)

X Y X Y

FZ

X Y

Z

U

Figure 2. Representation of a simple back-door path

TETRAD program: http://www.phil.cmu.edu/projects/tetrad/) to exhaust the set of pos-sible relationships, including effects due tolatent or unobservable factors, to determinethe more parsimonious causal models captur-ing the causal associations within a series ofdata. Through a systematic exploration of theset of relations between a series of observa-tions, it is possible to establish the presence ofan unobservable factor. The new causal modelthat includes the unobserved factor can mimicthe series of observed associations better thanthe original causal model. Based on Occam’srazor principle, causal modeling assumes thatthe minimal model is superior to any other (theparsimony principle). Whereas statistical proce-dures necessitate the inclusion of as many con-trol variables as possible to limit the impact ofunobservable factors, causal model techniquescan infer the causal influence of unobservables,while relying on the researchers’ knowledge tospecify them. Hence, these techniques help todetermine the boundary conditions of a causalmodel.

3. Causal effect estimation. Figure 2a depicts acausal graph where single-headed arrows rep-resent a causal relationship between the vari-able at the origin and the variable at thearrow’s head. Dashed double-arrows indicatethe presence of common unobserved causes ofboth terminal variables. This graph is not afull causal model because some unobservablecausal antecedents affect X and Y, and cansubsequently explain variations in Y via longerpaths than direct and observable X → Y. Itshould be noted that causal graphs are non-parametric and acyclic (i.e., they do not permitrepresentation of circular causation whereby Ywould impact X).

The general principle of causal graph estimationis to eliminate ‘back-door paths,’ namely pathsthat can be viewed as entering X through theback door (Pearl, 2000: 79). To use graphicallanguage, any arrow’s head pointing to X canbe regarded as entering through the back door.Figure 2b illustrates a situation where X and Y aremutually dependent on an unobservable Z variable;the dot for Z is white, indicating unobservability.To satisfy the back-door criterion, Pearl shows that(i) no node in Z is a descendant of X, and (ii) Zblocks every path between X and Y that containsan arrow into X. In Figure 2b, the back-door pathis simply X ← Z → Y, whereas in Figure 2c thereis a longer back-door path: X ← Z ← U → F→ Y.

There are three general strategies that can beused to estimate causal effects in this approach(see Figure 3 for a simple representation). The firststrategy is to condition on variables that block allback-door paths from the causal variable X to theoutcome variable Y. One first needs to calculatethe association between X and Y for different sub-groups’ values of the condition variable C. Then,averaging the associations of these specific val-ues over the marginal distribution of the values ctaken by C corresponds to the average treatmenteffect in the counterfactual causality literature. InFigure 3a, conditioning on C identifies the causaleffect of X on Y (Morgan and Winship, 2007: 38).In Figure 2c, conditioning on F also blocks theback-door path from X to Y. An important findingderived from this vein of research is that con-trolling for all the potential omitted direct causesof an outcome variable is not always efficient,although we regularly use this practice in strat-egy studies. It produces inaccurate results, sinceback-door paths—unobservable factors potentiallyaffecting one or more control variables—cannot



3a: Conditioning on C 3b: Instrumenting with T 3c: Mediating through M

Legend: Causal relationshipPresence of unobservable variableObservable variable

X Y

C

X Y

Z

T

X Y

Z

M

Unobservable variable

Figure 3. Three strategies for causal effect estimation

be taken into account by adding lines of con-trols in traditional statistical models. A properlyapplied conditioning strategy for a minimal num-ber of variables—the ones that block back-doorcausal paths—is more effective at revealing causalrelationships.

Strategy scholars have already used this strat-egy in producing estimates with treatment-effectmethods (Hamilton and Nickerson, 2003; Shaver,1998). For instance, when we observe the per-formance of firms that actually possess a certainresource, we do not examine what their perfor-mance might have been had they not had thisresource at their disposal. Because possession ofa specific resource is not randomly attributable tofirms and because resource possession and firmperformance are likely to depend jointly on unob-served factors (as in Figure 2b), a treatment effectsprocedure is often used to correct for the specifica-tion error and to avoid spurious causal associations(Greene, 2005; Maddala, 1983).

The second strategy is to use an instrument vari-able T for X to estimate its effect on Y. Accordingto this strategy, the pursued goal is not to blockback-door paths from X to Y, but to estimate indi-rectly the effect of T on Y, that of T on X, andthen deduce the effect of X on Y, all other factorsremaining unchanged. Hence, instrument variablesenable one to isolate the covariation between Xand Y (see Figure 3b). These instrumental variable(IV) techniques are becoming increasingly pop-ular in the strategy literature. The difficulty liesin finding an IV, that is, a variable T that has acausal effect on X without being causally related

to Y either directly or indirectly (via its effect ona mediating variable or the effect of unobserv-able factors that impact both the IV and Y dis-tributions). Another issue concerns the degree ofprediction of X by the IV, since ineffective predic-tion may produce incorrect estimates of the causaleffect (Morgan and Winship, 2007: 216).

The third generic strategy is to find a mediatorM that completely accounts for the causal effectof X on Y (Figure 3c). If one is able to findsuch a mechanism, M can be used to estimatethe causal relationship between X and Y eventhough it does not satisfy any of the back-doorcriteria. M is said to satisfy the front-door criteriawhen (i) M intercepts all directed paths from Xto Y; (ii) there is no back-door path from X toM; and (iii) all back-door paths from M to Y areblocked by X (Pearl, 2000: 82). In a nutshell, themechanism M needs to be isolated (it influencesY) and exhaustive (it captures all effects from X).

Applications. This kind of causal modeling canbe applied to various research questions, but itspecifically opens up new avenues in RBV re-search. For two decades, RBV has concentrated onexplaining a firm’s above-average returns throughdifferences between its resources and past per-formance, and industry resource and performanceaverages. In the presence of many plausible com-mon antecedents—that open back-door paths inthe causal diagram—researchers face highmethodological and empirical hurdles in demon-strating that differences in resource levels causesustained differences in performance. In otherwords, a firm’s resource differential vis-a-vis that



of its rivals covaries with past performances (boththe firm’s individual performance and the rivals’average performance) and is plagued with endo-geneity issues (common antecedents). On thisbasis, one cannot establish whether a statisti-cal association between resources and sustainedperformance is a causal relationship.

Recently, studies have tackled this problemusing a Bayesian approach to avoid the liabili-ties of classical statistical methods (e.g., that anaverage association is not specific to any givenobservation; that outliers must be eliminated fromempirical models on the grounds that they biasestimations even though RBV’s purpose is toexplain extreme performance). Bayesian method-ology allows a full estimation of individual effects,a prediction of ‘what-if’ results, and robust resultswith small samples or skewed data (for further dis-cussion, see Berry, 1996; Hahn and Doh, 2006;Hansen et al., 2004; Powell, 2001). This approachcan lead to promising results around the RBVresource-performance relationship, complementingthe causal modeling approach we focus on in thispaper.

We present a causal diagram in Figure 4 thatshows the relationships between past and currentperformance Y, both direct, mediated by resourcesR, and mutually dependent on unobserved vari-ables U. Resources and past performance mutuallydepend on observable variables O and unobserv-able variables V. Studying the statistical associa-tions between R and Y is doomed to fail, sincethere are many back-door paths via V, O, pastY, and U. The first strategy, conditioning on a

Legend: Causal relationshipPresence of unobservable variableObservable variableUnobservable variable

O

P Yt

Yt-1

U

R

V

Figure 4. Properties as mediating mechanisms in causalrelationship between resources and performance

variable that blocks all back-door paths, is inappli-cable. Conditioning on one of the Os leaves openthe back-door paths via the unobservable vari-ables U and V. Conditioning on past performancedoes not block paths via O and U. Instrumenta-tion, the second strategy, could work if we wereable to find a purely random instrument that is notrelated to other organizational variables (from O)or to performance Y. However, this condition isextremely restrictive, and it is highly unrealistic toassume that such an instrument could be found ordeveloped.

A promising avenue is to seek an isolated andexhaustive mechanism that respects the front-doorcriteria (the third strategy.) We could argue thatthe causation initially associated with the ‘strategicresource-competitive advantage’ relationship doesnot originate from the resources themselves, butfrom their properties: for example, rareness, immo-bility, low imitability, low substitutability, or pathdependence. In Figure 4, P is displayed as a mech-anism that blocks back-door paths and follows thefront-door path criteria. For this strategy to befruitful, one needs to assume that 1) the organi-zational factors O (that influence R and past Y)and P are independent, and 2) the unobservablefactors U and V do not influence P. Acceptingthese restrictive conditions, this solution representsa future research avenue for isolating and estimat-ing the impact of given resources on performance,as mediated by the properties of these resources.

Thus, to better understand the causal mech-anisms constituting competitive advantage, it isfruitful to shift attention from the mere possessionof specific resources to questions of how particularproperties mediate the impact of these resources onperformance. It follows that the RBV is noncausalif located only at the resource level, since it islogically impossible to prove causation between aresource and the purported advantage materializedat firm level. Traditional RBV studies may thusmistake resources for their properties (for a furtherdiscussion, see Durand, 2006). Thus, we can arguethat properties constitute a mediating mechanismthat could help to estimate the causal influence ofresources on firm performance with counterfactualanalysis. As a result, competitive advantage wouldbe defined as the conjunction of given resourcesand properties. This conjunction varies over timeand space, which makes it a fascinating objectof study. In particular, such a conceptualization



seems to be able to combine insights from a real-ist perspective that views competitive advantageas a causal mechanism (Durand, 2002; Wigginsand Ruefli, 2002) and a pragmatist perspective thatunderscores the role of local realities, perceptions,and interests (Powell, 2001; 2003).

DISCUSSION AND CONTRIBUTIONS

We began by stressing the urgent need to developour understanding of causation in strategy research.Hence, our aim has been to supply some of theelements needed to establish a common groundfor future research in strategy studies. We see adanger in constantly employing different paradig-matic approaches to the same questions that weaim at deciphering and modeling. We believe thatcloser attention to causal arguments is a requi-site condition for a better-grounded theorizationof strategy and for reestablishing strategy as aparadigmatic discipline rather than a garden fullof incommensurable flowers. Our discussion oncausation engages the field vis-a-vis the scientificvalue of our research, and our capacity to justifyand exploit it to address organizational and socioe-conomic issues. By probing deep into the notionof causation, the role of causal mechanisms, theapplication of counterfactual reasoning principles,the construction of historical counterfactuals, andthe representation of causal graphs, we are betterequipped to take up these challenges.

More specifically, this paper makes four con-tributions to our understanding of causation instrategy research: exposition of amenable episte-mological conditions for the study of causation,development of a counterfactual approach, detailsof two methodologies for conducting such analy-sis, and suggestions for future research on compet-itive advantage. First, we have outlined four con-ditions that provide a basis for reconciling someof the epistemological disputes and for advancingspecific approaches such as counterfactual anal-ysis. We believe that this stance helps us to gobeyond the empirically driven regularity view oncausation of standard positivist research by stress-ing the differences between statistical associa-tions and causal relationships, by revealing howunobservables modify causal dependencies, and bydemonstrating that adding more control variablesto equations does not make up for the problemsposed by inherent causal paths. This view also

facilitates dealing with the constructionist rejec-tion of causation that is based on the regularityview. We also think that our position advancesthe realist work on causation by giving a concretemeaning to the notion of causal mechanism, byconnecting series of observational data with evi-dence of the influence of unobservable variables,and by identifying relevant causal paths. Finally,our view acknowledges that causal research, suchas as human and social activity, is driven by inter-ests, interpretations, and a quest for satisfying andprovisional explanations. In this sense, our viewadvances pragmatist insights in causal analysiswithout regressing into relativism.

Second, and most importantly, we have devel-oped a counterfactual approach to causal strategyresearch. While this approach is well known in thephilosophy of science (Collins et al., 2004; Lewis,1973; Woodward, 2003) and applied in areas suchas historical analysis (Ferguson, 1997; Tetlock andBelkin, 1996) and causal modeling (Morgan andWinship, 2007; Pearl, 2000), it has not been givenmuch attention in strategy research. Nevertheless,counterfactual ‘what-if’ scenarios do play a cen-tral role in causal reflection, and we argue thatthis reasoning should be as explicit as possible,including the deliberate construction of ‘alterna-tive worlds.’ Our approach focuses on contrastingspecific causal mechanisms with alternative coun-terfactual explanations. Pearl’s description is accu-rate: counterfactuals ‘rest directly on the mecha-nisms . . . that produce those worlds and on theinvariant properties of those mechanisms” (Pearl,2000: 239). Without a deeper comprehension ofthese mechanisms and their construction and test-ing, we as strategy scholars risk misinterpretingand misusing our findings and those of othersas well. In the worst case, this may lead to thereproduction of fallacious causal ideas and claims.

Third, we have presented two methodologies forcounterfactual-based strategy research. While thereare other useful methods, we have promoted coun-terfactual historical analysis and causal modelingas alternatives for researchers conducting causalstrategy research. In counterfactual history, onecan combine thick empirical description with asystematic analysis of event causality. This canand should lead to explicit presentation of propo-sitions that are then contrasted with alternativehistories (counterfactuals). With the help of theimaginary counterfactuals, one can eventually val-idate or invalidate the proposed arguments. In



causal modeling, one can focus on the nature ofthe relationships between key variables: causalrelationships, statistical associations, or covaria-tions (in a decreasing order of directionality, sta-bility, and dependence). This helps us to thinkmore about whether we have identified the causalfactors appropriately and whether we have cor-rectly assessed the closure of a chain of causalrelationships (i.e., identification and treatment ofthe back-door paths). In addition, we have shownhow causal modeling based on counterfactualsintegrates some of the most popular econometrictechniques (e.g., treatment effect and instrumentvariables).

Counterfactual analysis is not a panacea. It mustalways be based on sound theoretical argumentsand careful empirical analyses. While we have pro-moted counterfactual historical analysis and causalmodeling as useful methodologies, we stress thatboth involve limitations, problems, and challenges.As discussed above, counterfactual reasoning doesinvolve dealing with alternative chains of events,missing or at least incomplete data, retrospectiverecall, and other cognitive biases. However, theseare precisely the kinds of issues that we have todeal with to advance causal strategy research. Webelieve that it is far better to tackle these issueshead-on than to avoid them altogether.

Finally, our analysis also has implications forour understanding of competitive advantage. Theconceptualization of competitive advantage as acausal mechanism is one way to deal with theambiguity surrounding this crucial issue. Weargued that historical counterfactual analysis couldadvance something that has been scarce in ourfield: historical research on competitive advantage.In particular, counterfactual history can add to ourunderstanding of the context- and culture-specificdependencies in the construction of resource-basedcompetitive advantage and path dependence. Wealso illustrated the ways in which causal mod-eling elucidates causal associations around com-petitive advantage. In particular, we offered newinsights into the lingering dispute around the causalrelationship between resources, competitive advan-tage, and performance. In particular, we arguedthat resource properties could be seen as mediatingmechanisms in these causal relationships.

In conclusion, the fundamental questions exam-ined by strategy research revolve around the actualeffects of strategic action. Without sufficient agree-ment on the notion of causation, however, we risk

losing relevance, wasting our efforts, and failingto accumulate knowledge. In this paper we haveexamined traditional views on causation, outlined areconciliatory epistemological position, introducedcounterfactual reasoning, and detailed two meth-ods to probe causation. Our goal is to pursue thisfundamental discussion in ways that reduce dis-tance between current postures instead of stressingtheir differences. While we are not saying thatall strategy scholars should now engage in coun-terfactual analysis, a lack of epistemological andmethodological discussion on causation hampersthe development of our discipline. It is time tomove forward and be explicit about what we meanby causation and how it impacts research on keyissues such as competitive advantage.

ACKNOWLEDGEMENTS

We are grateful to Professor Dan Schendel, whoseadvice and support have been crucial in this pro-cess. We also thank our two anonymous reviewersfor their excellent comments and insights. In addi-tion, we are indebted to Juha-Antti Lamberg, SakuMantere, Kalle Pajunen, and JP Vergne for use-ful comments and for helping with cumbersomeissues, and to David Miller for language revision.A first version of this paper was presented at theAtlanta Competitive Advantage Conference, wherewe received extremely valuable comments fromMargaret Peteraf and other participants.

REFERENCES

Abbott A. 2001. Time Matters: On Theory and Methods .University of Chicago Press: Chicago, IL.

Ackroyd S, Fleetwood S (eds). 2000. Realist Perspectiveson Management and Organisations . Routledge:London, UK.

Ackroyd S, Fleetwood S (eds). 2004. Critical RealistApplications in Organisations and ManagementStudies . Routledge: London, UK.

Bacharach SB. 1989. Organizational theories: somecriteria for evaluation. Academy of ManagementReview 14: 496–515.

Berger PL, Luckmann T. 1966. The Social Constructionof Reality: A Treatise in the Sociology of Knowledge.Doubleday: Garden City, NY.

Bergh DD, Holbein GF. 1997. Assessment and redirec-tion of longitudinal analysis: demonstration with astudy of the diversification and divestiture relationship.Strategic Management Journal 18(7): 557–571.

Berry DA. 1996. Statistics: A Bayesian Perspective.Duxbury: Belmont, CA.



Bettman JR, Weitz BA. 1983. Attributions in theboardroom. Administrative Science Quarterly 28:165–183.

Bhaskar R. 1975. A Realist Theory of Science. LeedsBooks: Leeds, UK.

Bhaskar R. 1979. The Possibility of Naturalism: APhilosophical Critique of Contemporary HumanSciences . Harvester Press: Brighton, UK.

Blaug M. 1980. The Methodology of Economics (Cam-bridge Surveys of Economic Literature) CambridgeUniversity Press: Cambridge, UK.

Bogner WC, Thomas H, McGee J. 1996. A longitudinalstudy of the competitive positions and entry paths ofEuropean firms in the U.S. pharmaceutical market.Strategic Management Journal 17(2): 85–107.

Booth C. 2003. Does history matter in strategy: thepossibilities and problems of counterfactual analysis.Management Decision 41: 96–104.

Boyd BK, Finkelstein S, Gove S. 2005a. How advancedis the strategy paradigm? The role of particularism anduniversalism in shaping research outcomes. StrategicManagement Journal 26(9): 841–854.

Boyd BK, Gove S, Hitt MA. 2005b. Construct mea-surement in strategic management research: illu-sion or reality? Strategic Management Journal 26(3):239–257.

Camerer C. 1985. Redirecting research in business policyand strategy. Strategic Management Journal 6(1):1–15.

Cockburn IM, Henderson RM, Stern S. 2000. Untanglingthe origins of competitive advantage. StrategicManagement Journal , October–November SpecialIssue 21: 1123–1145.

Collins J, Hall N, Paul LA (eds). 2004. Causation andCounterfactuals . MIT Press: Cambridge, MA.

De Mey T, Weber E. 2003. Explanation and thoughtexperiments in history. History and Theory 42: 28–38.

Denrell J. 2003. Vicarious learning, under-sampling offailure, and the myths of management. OrganizationScience 14: 227–243.

De Rond M, Thietart R-A. 2007. Choice, chance,and inevitability in strategy. Strategic ManagementJournal 28(5): 535–551.

Dewey J. 1988. The Middle Works of John Dewey, Volume12, 1899–1924: 1920, Reconstruction in Philosophyand Essays (Collected Works of John Dewey),Boydston JA (ed). Southern Illinois University Press:Carbondale, IL.

Durand R. 2002. Competitive advantages exist: a critiqueof Powell. Strategic Management Journal 23(9):867–872.

Durand R. 2003. Predicting a firm’s forecasting ability:the roles of organizational illusion of controland organizational attention. Strategic ManagementJournal 24(9): 821–838.

Durand R. 2006. Organizational Evolution and StrategicManagement . Sage: London, UK.

Evered R, Louis MR. 1981. Alternative perspectivesin the organizational sciences: ‘Inquiry from theinside’ and ‘inquiry from the outside.’ Academy ofManagement Review 6: 385–395.

Ferguson N (ed). 1997. Virtual History: Alternatives andCounterfactuals . Picador: London, UK.

Ferraro F, Pfeffer J, Sutton R. 2005. Economics languageand assumptions: how theories can become self-fulfilling. Academy of Management Review 30: 8–25.

Fleetwood S. 2004. An ontology for organisation andmanagement studies. In Critical Realist Applica-tions in Organisation and Management Studies , Fleet-wood S, Ackroyd S (eds). Routledge: London, UK;27–54.

Fogel R. 1964. Railroads and American EconomicGrowth: Essays in Econometric History . John HopkinsUniversity Press: Baltimore, MD.

Friedman M. 1953. Essays in Positive Economics .University of Chicago Press: Chicago, IL.

Gergen KJ. 1999. An Invitation to Social Construction .Sage: London, UK.

Giddens A. 1984. The Constitution of Society . Polity:Cambridge, UK.

Godfrey PC, Hill CWL. 1995. The problem of unob-servables in strategic management research. StrategicManagement Journal 16(7): 519–533.

Goodman N. 1983. Fact, Fiction, and Forecast . HarvardUniversity Press: Cambridge, MA.

Greene WH. 2005. Econometric Analysis . PearsonEducation, Upper Saddle River: NJ.

Griffin LJ. 1993. Narrative, event-structure analysis,causal interpretation in historical sociology. AmericanJournal of Sociology 98: 1094–1133.

Griffin LJ, Ragin CC. 1994. Some observations of formalmethods of qualitative analysis. Sociological Methodsand Research 23: 4–21.

Hahn ED, Doh JP. 2006. Using Bayesian methods instrategy research: an extension of Hansen, et al.Strategic Management Journal 27(8): 783–798.

Hamilton BH, Nickerson JA. 2003. Correcting forendogeneity in strategic management research.Strategic Organization 1: 51–78.

Hansen MH, Perry LT, Reese CS. 2004. A Bayesianoperationalization of the resource-based view. Strate-gic Management Journal 25(13): 1279–1295.

Harre R. 1970. The Principles of Scientific Thinking .MacMillan: London, UK.

Harre R, Madden EH. 1975. Causal Powers: A Theory ofNatural Necessity . Basil Blackwell: Oxford, UK.

Hedstrom P. 2005. Dissecting the Social: On the Prin-ciples of Analytical Sociology . Cambridge UniversityPress: Cambridge, UK.

Heider F. 1958. The Psychology of InterpersonalRelations . Wiley: New York.

Heise D. 1989. Modeling event structures. Journal ofMathematical Sociology 14: 139–169.

Hempel C. 1965. Aspects of Scientific Explanation andOther Essays in Philosophy of Science. Free Press:New York.

Huber GP, Van de Ven AH. 1995. Longitudinal FieldResearch Methods: Studying Processes of Organiza-tional Change (Organization Science). Sage: ThousandOaks, CA.

Huber J, Mirosky J. 1997. Of facts and fables: replyto Denzin. American Journal of Sociology 102:1423–1428.



Hull D. 1988. Science and Selection . CambridgeUniversity Press: Cambridge, UK.

Hume D. 1955. An Enquiry Concerning Human Under-standing . Bobbs-Merrill: Indianapolis, IN.

James W. 1975. Pragmatism . Harvard University Press:Cambridge, MA.

Johnson P, Duberley I. 2003. Reflexivity in manage-ment research. Journal of Management Studies 40:1279–1303.

Kakkuri-Knuuttila, ML, Vaara E. 2007. Reconcilingopposites in organization studies: the case ofmodernism and post-modernism. Philosophy ofManagement 6: 97–114.

Kaplan A. 1964. The Conduct of Inquiry: Methodologyfor Behavioral Science. Chandler: New York.

Kieser A. 1994. Why organization theory needs historicalanalyses—and how this should be performed.Organization Science 5: 608–620.

King AW, Zeithaml CP. 2001. Competencies and firmperformance: examining the causal ambiguity paradox.Strategic Management Journal 22(1): 75–99.

Knorr-Cetina K. 1983. The ethnographic study ofscientific work: towards a constructivist interpretationof science. In Science Observed: Perspectives on theSocial Study of Science, Knorr-Cetina KD, Mulkay M(eds). Sage: Beverly Hills, CA; 115–140.

Kwan K-M, Tsang EWK. 2001. Realism and construc-tivism in strategy research: a critical realist responseto Mir and Watson. Strategic Management Journal22(12): 1163–1168.

Lamberg J-A, Tikkanen H. 2006. Changing sources ofcompetitive advantage: cognition and path dependencein the Finnish retail industry. Industrial and CorporateChange 15: 811–846.

Langley A. 1999. Strategies for theorizing from processdata. Academy of Management Review 24: 691–710.

Latour B, Woolgar S. 1989. Laboratory Life: TheConstruction of Scientific Facts . Princeton UniversityPress: Princeton, NJ.

Lewis D. 1973. Counterfactuals . Harvard UniversityPress: Cambridge, MA.

Lewis D. 1986. Philosophical Papers: Volume II . OxfordUniversity Press: Oxford, UK.

Lewis D. 2000. Causation as influence. Journal ofPhilosophy 97: 182–197.

Løwendahl B, Revang Ø. 1998. Challenges to existingstrategy theory in a postindustrial society. StrategicManagement Journal 19(8): 755–773.

MacKenzie D, Millo Y. 2003. Constructing a market,performing theory: the historical sociology of afinancial derivatives exchange. American Journal ofSociology 109: 107–145.

Maddala GS. 1983. Limited-Dependent and QualitativeVariables in Econometrics, Econometric SocietyMonographs . Cambridge University Press: New York.

Mahner M (ed). 2001. Scientific Realism: Selected Essaysof Mario Bunge. Prometheus Books: Amherst, NY.

Mahoney JT. 1993. Strategic management and determin-ism: sustaining the conversation. Journal of Manage-ment Studies 30: 173–191.

Mahoney JT, Sanchez R. 2004. Building new manage-ment theory by integrating processes and products ofthought. Journal of Management Inquiry 13: 34–47.

Maki U. 1990. Scientific realism and Austrian explana-tion. Review of Political Economy 2: 310–344.

March JG, Sutton RI. 1997. Organizational performanceas a dependent variable. Organization Science 8:698–706.

Maslen C. 2004. Causes, contrasts, and the nontransi-tivity of causation. In Causation and Counterfactu-als , Collins J, Hall E, Paul L (eds). MIT Press: Cam-bridge, MA; 341–358.

Mir R, Watson A. 2000. Strategic management and thephilosophy of science: the case for a constructivistmethodology. Strategic Management Journal 21(9):941–953.

Montgomery CA, Wernerfelt B, Balakrishnan S. 1989.Strategy content and the research process: a critiqueand commentary. Strategic Management Journal10(2): 189–197.

Morgan SL, Winship C. 2007. Counterfactuals andCausal inference: Methods and Principles for SocialResearch . Cambridge University Press: Cambridge,UK.

Numagami T. 1998. The infeasibility of invariant laws inmanagement studies: a reflective dialogue in defenseof case studies. Organization Science 9: 2–15.

Pajunen K. 2004. Explaining by Mechanisms: A studyof Organizational Decline and Turnaround Processes.Tampere University of Technology: Tampere, Finland.

Pearl J. 2000. Causality: Models, Reasoning, andInference. Cambridge University Press: Cambridge,UK.

Peirce CS. 1992. The Essential Peirce: Selected Philo-sophical Writings, 1893–1913 , Houser N, Kloe-sel CJW (eds). Indiana University Press: Blooming-ton, IN.

Powell TC. 2001. Competitive advantage: logical andphilosophical considerations. Strategic ManagementJournal 22(9): 875–888.

Powell TC. 2002. The philosophy of strategy. StrategicManagement Journal 23(9): 873–880.

Powell TC. 2003. Strategy without ontology. StrategicManagement Journal 24(3): 285–291.

Powell T, Lovallo D, Caringal C. 2006. Causal ambigu-ity, management perception, and firm performance.Academy of Management Review 31: 175–196.

Priem RL, Butler JE. 2001. Is the resource-based view auseful perspective for strategic management research?Academy of Management Review 26: 22–40.

Ragin CC. 2000. Fuzzy Set Social Science. University ofChicago Press: Chicago, IL.

Rorty R. 1989. Contingency, Irony and Solidarity .Cambridge University Press: New York.

Salancik GR, Meindl Jr. 1984. Corporate attributionsas strategic illusions of management control.Administrative Science Quarterly 29: 238–254.

Salmon W. 1998. Causality and Explanation. OxfordUniversity Press: New York.

Shaver JM. 1998. Accounting for endogeneity whenassessing strategy performance: does entry mode



choice affect FDI survival? Management Science 44:571–585.

Simon H. 1947. Administrative Behavior . Free Press:New York.

Spirtes P, Glymour C, Scheines R. 2000. Causation,Prediction and Search (2nd edn). MIT Press:Cambridge, MA.

Stevenson W, Greenberg, D. 1998. The formal analysisof narratives of organizational change. Journal ofManagement 24: 741–762.

Stevenson W, Greenberg, D. 2000. Agency and socialnetworks: strategies of action in a social structure ofposition, opposition, and opportunity. AdministrativeScience Quarterly 45: 651–678.

Tetlock PE, Belkin A (eds). 1996. CounterfactualThought Experiments in World Politics . PrincetonUniversity Press: Princeton, NJ.

Tetlock PE, Lebow RN, Parker NG (eds). 2006. Unmak-ing the West: ‘What-I’ Scenarios that Rewrite WorldHistory . Cambridge University Press: New York.

Trevino L, Weaver G. 2003. Managing Ethics in BusinessOrganizations: Social Scientific Perspectives . StanfordUniversity Press: Stanford, CA.

Tsang EWK. 2006. Behavioral assumptions and theorydevelopment: the case of transaction cost economics.Strategic Management Journal 27(11): 999–1011.

Tsang EWK, Kwan K-M. 1999. Replication and theorydevelopment in organizational science: a critical realistperspective. Academy of Management Review 24:759–780.

Tsoukas H. 1989. The validity of idiographic researchexplanations. Academy of Management Review 14:551–561.

Vaara E. 2002. On the discursive construction of suc-cess/failure in narratives of post-merger integration.Organization Studies 23: 213–250.

Van de Ven A. 2007. Engaged Scholarship: A Guidefor Organizational and Social Research . OxfordUniversity Press: New York.

von Wright GH. 1971. Explanation and Understanding .Cornell University Press: Ithaca, NY.

Weick K. 1989. Theory construction as disciplinedimagination. Academy of Management Review 14:516–531.

Weiner B. 1986. An Attribution Theory of AchievementMotivation and Emotion . Springer-Verlag: New York.

White H. 1987. The Content of the Form: Narrative Dis-course and Historical Representation . John HopkinsUniversity Press: Baltimore, MD.

Wicks AC, Freeman RE. 1998. Organization studiesand the new pragmatism: positivism, anti-positivismand the search for ethics. Organization Science 9:123–140.

Wiggins R, Ruefli T. 2002. Sustained competitiveadvantage: temporal dynamics and the incidenceand persistence of superior economic performance.Organization Science 13: 82–106.

Winch P. 1958. The Idea of a Social Science and ItsRelation to Philosophy. Routledge & Kegan Paul:London, UK.

Woodward J. 2003. Making Things Happen: A Theory ofCausal Explanation . Oxford University Press: Oxford,UK.


causation, counterfactuals, and competitive …rodolphedurand.com/wp-content/uploads/2012/09/... ·...

Documents