RECONCEPTUALIZING ENTREPRENEURIAL · PDF filereconceptualizing entrepreneurial orientation...

Strategic Management JournalStrat. Mgmt. J., 36: 1579–1596 (2015)

Published online EarlyView 8 July 2014 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/smj.2298Received 19 November 2012; Final revision received 27 May 2014

RECONCEPTUALIZING ENTREPRENEURIALORIENTATION

BRIAN S. ANDERSON,1* PATRICK M. KREISER,2 DONALD F. KURATKO,3

JEFFREY S. HORNSBY,4 and YOSHIHIRO ESHIMA5

1 Leeds School of Business, University of Colorado, Boulder, Colorado, U.S.A.2 College of Business, Iowa State University, Ames, Iowa, U.S.A.3 Kelley School of Business, Indiana University, Bloomington, Indiana, U.S.A.4 Henry W. Bloch School of Management, University of Missouri, Kansas City,Missouri, U.S.A.5 Faculty of Business Administration, Osaka University of Economics, Osaka, Japan

Entrepreneurial orientation (EO)—a firm’s strategic posture towards entrepreneurship—hasbecome the predominant construct of interest in strategic entrepreneurship research. Despitethe ever-increasing volume of nomological research on EO, there remain ongoing conversationsregarding its ontology. Drawing from measurement theory, we outline an EO reconceptualizationaddressing the likely prevalence of Type II nomological error in the EO literature stemming frommeasurement model misspecification. Focusing on the question of whether EO is an attitudinalconstruct, a behavioral construct, or both, we propose a formative construction of EO viewingthe exhibition of entrepreneurial behaviors and of managerial attitude towards risk as jointlynecessary dimensions that collectively form the higher-order EO construct. We present anempirical illustration of our reconceptualization followed by a discussion of future researchopportunities. Copyright © 2014 John Wiley & Sons, Ltd.

INTRODUCTION

What makes a firm entrepreneurial, and how dowe distinguish entrepreneurial firms from thosemore conservatively managed? Miller’s (1983)foundational paper provided much needed clarityregarding this fundamental issue to managementscholars. After three decades of research build-ing from Miller’s (1983) insights, we delineatebetween entrepreneurial and conservative firmsaccording to their entrepreneurial orientation(EO): the decision-making practices, managerialphilosophies, and strategic behaviors that are

Keywords: entrepreneurial orientation; measurement the-ory; formative measurement; entrepreneurial behaviors;managerial attitudes towards risk*Correspondence to: Brian S. Anderson, 995 Regent Dr., Boulder,CO 80309, U.S.A. Email: [email protected]

Copyright © 2014 John Wiley & Sons, Ltd.

entrepreneurial in nature, with entrepreneurialreferring to three components—innovativeness,proactiveness, and risk taking (Anderson, Covin,and Slevin, 2009). While scholars have noted thatdefinitions and operationalizations of EO vary(Covin and Wales, 2012), a substantial body ofEO research has shown that entrepreneurial firmsgenerally outperform their more conservativelymanaged peers (Rauch et al., 2009).

Despite the burgeoning scholarly interest in thisarea, a number of ontological questions persist inthe EO literature. For example, there are ongoingconversations regarding the dimensionality of EO(Lumpkin and Dess, 1996); whether EO shouldbe measured formatively or reflectively (Covin andWales, 2012); whether EO is an attitudinal con-struct, a behavioral construct, or both (Covin andLumpkin, 2011; Miller, 2011); whether the dimen-sions of EO must necessarily covary (Lumpkin and

1580 B. S. Anderson et al.

Dess, 2001); and even whether the name “EO”is being appropriately used across studies (Georgeand Marino, 2011). Active and constructive schol-arly debate on the ontological assumptions of ourtheoretical constructs is healthy for scientific dis-course (Bagozzi, Yi, and Phillips, 1991). However,when such arguments remain salient despite a pro-liferation of research on the EO phenomenon, itbecomes apparent that fundamental questions sur-rounding what it means at the firm level to “beentrepreneurial” remain unsettled.

The current study applies a measurement the-ory perspective to identify how measurement modelmisspecification related to the question of whetherto construe EO as an attitudinal construct, a behav-ioral construct, or both, contributes to nomologi-cal error—specifically Type II error (i.e., a falsenegative)—in the EO literature. Further, we suggestthat certain ontological assumptions and measure-ment inconsistencies remaining unresolved underthe dominant conceptualization of EO in the lit-erature (i.e., Miller, 1983/Covin and Slevin, 1991)have limited the development of EO’s nomologicalnetwork. We also maintain that efforts to resolvemeasurement issues related to EO without address-ing these ontological issues (e.g., Covin and Wales,2012; George, 2011; George and Marino, 2011),while offering valuable insights, have limited util-ity. That is, scholars may inadvertently continue tomake nomological errors irrespective of adopting aparticular ex post measurement remedy.

We therefore offer a reconceptualization of EOaddressing Type II nomological error in the EOliterature. Under our reconceptualization, we defineEO in a manner consistent with Miller (1983) andCovin and Slevin (1991) as the joint exhibition ofobserved entrepreneurial behaviors and a manage-rial inclination at the strategic decision-makinglevel favoring actions with uncertain outcomes.Under our reconceptualization, entrepreneurialbehaviors and managerial attitude towards riskjointly and in totality comprise the conceptualdomain of firm-level EO. We then pair a formativemeasurement model ontologically congruent withour reconceptualization that enables the refinementand expansion of EO’s nomological network.Specifically, our paper (1) demonstrates the need torevisit prior research questions, particularly regard-ing EO’s contributing factors, because of likelynomological error; and (2) illuminates new researchquestions and new research designs that meaning-fully advance the EO conversation. We further offer

an empirical illustration of our reconceptualizationto provide practical guidance to EO scholars,and conclude by discussing new research designsand opportunities for strategic entrepreneurshipresearch leveraging our reconceptualization.

EO, MEASUREMENT THEORY, ANDNOMOLOGICAL CONCLUSIONS

What is entrepreneurial orientation (EO)?

We may trace research on an entrepreneurialmode of strategic decision making to the worksof Mintzberg (1973) and Khandwalla (1977),which argued that firm performance is largelypredicated on gestalts comprised of strategicchoices, organizational attributes (e.g., structure),and environmental exigencies. One such gestaltis entrepreneurial in nature, typified in part byproactive strategic moves and a willingness to takeon projects with uncertain outcomes (Khandwalla,1977). From this early work, Miller (1983) crystal-ized an entrepreneurial approach to strategy makingby suggesting that entrepreneurial firms are thosethat pursue innovation, aggressively enter new mar-kets, and accept a measure of strategic and financialrisk in the pursuit of new opportunities. Notably,Miller (1983: 780) observed that an entrepreneurialfirm should exhibit all three strategic componentswith some degree of simultaneity:

In general, theorists would not call a firmentrepreneurial if it changed its technology orproduct-line … simply by directly imitatingcompetitors while refusing to take any risks.Some proactiveness would be essential aswell. By the same token, risk-taking firmsthat are highly leveraged financially are notnecessarily entrepreneurial. They must alsoengage in product-market or technologicalinnovation.

Building from Miller (1983) Covin and Slevin(1989, 1991) posited the existence of a continuumused to plot a firm’s strategic behavioral proclivi-ties. The continuum ranges from more conservativeto more entrepreneurial, with the entrepreneurialend of the spectrum evidenced by innovativeness,proactiveness, and risk taking. Covin and Slevin(1991: 8) further suggested that the observation ofsustained entrepreneurial behaviors is a necessarycondition for being entrepreneurial: “Organizations

Copyright © 2014 John Wiley & Sons, Ltd. Strat. Mgmt. J., 36: 1579–1596 (2015)DOI: 10.1002/smj

Reconceptualizing Entrepreneurial Orientation 1581

with an entrepreneurial posture are those in whichparticular behavioral patterns are recurring.” Thus,under the Miller/Covin and Slevin conceptualiza-tion, a firm is entrepreneurial because it exhibitsentrepreneurial behaviors, and there is an elementof temporal consistency in this exhibition.

In the period following the Covin and Slevin(1991) conceptualization, scholars offered alter-native perspectives on the conceptual domainof a firm-level strategic orientation towardsentrepreneurship (e.g., Lumpkin and Dess, 1996;see Covin and Wales, 2012 for a discussion of thedifferences between the two conceptualizations).Nonetheless, as noted in two recent meta-analyses,the Miller/Covin and Slevin conceptualizationis by far the dominant perspective of EO in therelevant literature (Rauch et al., 2009; Rosenbusch,Rauch, and Bausch, 2013). We therefore groundour discussion in the Miller/Covin and Slevin viewsuggesting that entrepreneurial firms are those thatexhibit innovativeness (the introduction of newproducts, processes, and business models), proac-tiveness (actively entering new product/marketspaces and seeking market leadership positions),and risk taking (a willingness among strategicdecision makers to contribute resources to projectswith uncertain outcomes).

Measurement theory

While numerous perspectives exist in the litera-ture for the ontological grounding of theoreticalconstructs—realist, critical realist, constructivist,social constructivist, instrumentalist, and so forth(Bagozzi et al., 1991; Borsboom, 2005)—there areessentially two ways with which to measure alatent construct: reflective measurement or forma-tive measurement (Diamantopoulos, Riefler, andRoth, 2008). Prior research has amply coveredthe differences between these approaches (Jarvis,MacKenzie, and Podsakoff, 2003; MacKenzie, Pod-sakoff, and Jarvis, 2005), although we will exploresome of those differences in depth in subsequentsections. However, the equations associated withboth perspectives merit review, and we refer read-ers to online supporting information Appendix S1for a more detailed discussion of measurement the-ory and construct development.

Reflective measurement, which is the most com-monly employed perspective in the managementliterature (MacKenzie, Podsakoff, and Podsakoff,

2011), follows classical test theory wherein mea-sures are a reflection—with error—of a latent con-struct. Under reflective measurement, variations inthe latent construct induce variance in the indica-tors, as specified in the following equation:

xi = 𝜆i𝜉 + 𝛿i (1)

In Equation 1, x is the ith indicator of the latentexogenous construct 𝜉, with 𝜆 representing theloading coefficient and 𝛿 representing measure-ment error (Diamantopoulos et al., 2008). WhileEquation 1 is generally associated with struc-tural equation modeling techniques, the ontologi-cal assumptions of the equation are congruent withthose made when researchers employ summated rat-ing scales, which are created by taking the meanvalue among a set of indicators or by simply addingthem together. As Spector (1992: 10) observed:“With a summated rating scale, each individual itemis designed to be an observation of the intended trait.Each item represents an individual assessment ofthe true score. If the average (or sum) of individ-ual items is calculated, the errors of measurementare assumed to average approximately zero, result-ing in an estimate of the true score.” Furthermore,because the indicators are all observable manifes-tations of the same underlying latent phenomenon,we a priori expect the indicators to covary and toshare the same antecedents (Borsboom, 2005).

In contrast, under a formative specification, thecollective variance of a latent construct’s indica-tors define the total variance in the construct, rep-resented by the following equation:

𝜂 =n∑

(i=1)𝛾ixi + 𝜁 (2)

The sum of the indicators, x, collectively definethe conceptual domain of the latent construct 𝜂, andchanges in the indicators, 𝛾 , induce changes in 𝜂.Here, the disturbance term 𝜁 represents the amountof variance in the latent construct not accounted forby its indicators (MacKenzie et al., 2005; note that𝜁 is not an error term; see Diamantopoulos et al.,2008).

EO, measurement model misspecification,and nomological error

As noted by Miller (2011) and Covin and Lump-kin (2011), a highly salient yet largely unexplored



issue is whether EO is fundamentally a behav-ioral phenomenon or whether it represents somekind of attitudinal, philosophical, or dispositionalcharacteristic among strategic decision makers. AsMiller (2011) observed, the Miller/Covin and Slevinconceptualization—and the scale commonly usedto measure it—includes behavioral and attitudi-nal components. For example, Covin, Green, andSlevin (2006: 57) offer the following definition ofEO, broadly consistent with the Miller/Covin andSlevin conceptualization: “EO is a strategic con-struct whose conceptual domain includes certainfirm-level outcomes and management-related pref-erences, beliefs, and behaviors as expressed amonga firm’s top-level managers.”

There are two specific challenges when mixingattitudinal and behavioral components within a sin-gle latent construct. The first challenge, as noted byKoslowsky et al. (1997), is that attitude and behav-ior are mutually reinforcing; behavior is an out-come of attitude and the sustained engagement ina particular behavior reinforces the underlying atti-tude. The second challenge when mixing attitudinaland behavioral components is that, while it may bereasonable to assume that an attitude and a behav-ior share similar consequences, it is less clear thatthey share similar antecedents. This issue is particu-larly problematic in EO research. For example, doesthe act of new product introduction (captured bythe innovativeness component) result from the sameunderlying cause(s) as a decision-making proclivitytowards high-risk projects with the chance for highreturns (the risk-taking component)?

Reflective measurement of the Miller/Covinand Slevin conceptualization defines EO as theintersection of, or shared variance between, theunderlying components of innovativeness, proac-tiveness, and risk taking (Covin and Wales, 2012).Within this constraint, the assumption is that thesecomponents share the same antecedents (Covin andLumpkin, 2011). Because, however, an attitude isa manifestation of underlying beliefs, perceptions,and cognitive schemas (Rosenberg and Hovland,1960), it is conceptually unlikely that the samephenomena encouraging behavior also causallyrelate to attitude (Fishbein and Ajzen, 1975). Thepreceding is salient because, as George and Marino(2011) note, with only rare exception (e.g., Millerand Le Breton-Miller, 2011), scholars generallymeasure EO reflectively, and most often using asummated indicator approach with the Covin andSlevin (1989) EO scale or some derivative thereof.

Recall that under a reflective measurement model,the researcher makes the ontological assumptionthat changes in the underlying phenomenon ofinterest induce changes in all of its indicatorssimultaneously and of the same magnitude. This isbecause the indicators all share the same commoncause; the indicators are interchangeable and per-fectly covary, assuming the indicators are equallyvalid (Bollen, 1989). As George and Marino (2011)comment, however, there are theoretical argumentsand supporting empirical evidence suggestingthat the relationships among EO’s indicators, andbetween EO’s indicators and the higher-order EO,are not congruent with a reflective specification.The most significant evidence is research showingthat innovativeness, proactiveness, and risk takingexhibit differing nomological relationships and donot perfectly covary (Lumpkin and Dess, 2001).

Thus, the question of attitude versus behaviormanifests in manifold research design challengesfor EO scholars. While Covin and Wales (2012)maintain that a reflective specification does notimply that each underlying component of EOmust share exactly the same antecedents, mea-surement theory dictates that there must existcommonalities across the components such thata specified antecedent links causally to all three(MacKenzie et al., 2011). It is this assumption,however, that induces Type II nomological errorin the specification of EO’s contributing factors.EO scholars have likely overlooked—or haveincorrectly rejected—research models where agiven antecedent failed to “link up” with all of EO’sunderlying components yet may be theoreticallymeaningful contributory factors to only one.

The broader implication for EO scholars is thata reflective specification of EO may be unneces-sarily limiting and, of even greater concern, mayhave resulted in biased and/or incorrect nomolog-ical conclusions. Thus, continuing with a reflectivespecification that assumes EO’s dimensions sharecommon antecedents may not be the most fruitfulconceptualization to advance the EO conversation.As such, we submit that the reconceptualization ofEO developed herein is both timely and valuable.

RECONCEPTUALIZING EO

We define EO as a second-order, firm-level con-struct comprised of two lower-order dimensions:entrepreneurial behaviors (encompassing



innovativeness and proactiveness), and man-agerial attitude towards risk (risk taking). Wedefine entrepreneurial behaviors as the firm-levelpursuit of new products, processes, or businessmodels (e.g., innovativeness) with the intendedcommercialization of those innovations in newproduct/market domains (e.g., proactiveness). Wedefine managerial attitude towards risk as an inher-ent managerial inclination—existing at the levelof the senior manager(s) tasked with developingand implementing firm-level strategy—favoringstrategic actions that have uncertain outcomes(Miller, 1983). Under our reconceptualization,we reorder the three existing components of EOinto two lower-order dimensions—risk takingas an attitudinal dimension, while innovativenessand proactiveness collapse to one behavioraldimension.

Under our reconceptualization, we positthat (1) EO is a multidimensional con-struct consisting of two noninterchangeabledimensions—entrepreneurial behaviors and man-agerial attitude towards risk; (2) there is positivecovariance between these two dimensions; and (3)both dimensions are fundamentally necessary forEO to exist. For a firm to be entrepreneurial underour reconceptualization, observed entrepreneurialbehaviors necessarily but do not sufficiently cap-ture EO’s conceptual domain; that is, a managerialproclivity towards pursuing projects with uncertainoutcomes is an additional necessary condition. Fur-ther, under our reconceptualization, entrepreneurialbehaviors and managerial attitude towards riskare conceptually and functionally distinct. Assuch, we suggest viewing EO through a formativespecification. We further posit that, while the twodimensions positively covary and hence share someantecedents, there is an a priori expectation thatthe majority of factors that predict entrepreneurialbehaviors and those that predict managerial attitudetowards risk differ.

Entrepreneurial behaviors

There are two primary reasons for collapsingthe innovativeness and proactiveness compo-nents of EO into a single latent construct labeledentrepreneurial behaviors. The first reason is thatthere is little face validity in the a priori assumptionof an attitudinal element of innovativeness andof proactiveness. This assertion stems from theobservations of Miller (1983) and Covin and Slevin

(1991) that what gives meaning to innovation andto entering new markets are the specific acts ofinnovativeness and proactiveness. One cannot haveinnovation absent developing new products, pro-cesses, or business models; similarly, proactivenessdoes not exist without a firm actually entering anew market ahead of competitors and “acting inanticipation of future demand” (Lumpkin and Dess,2001: 431). As such, entrepreneurial behaviorsmust be observable and derive their meaning onlybecause a firm is behaving entrepreneurially.

One may then posit that the entrepreneurialbehaviors dimension is itself formed by thelower-order components of innovativeness andproactiveness. We would suggest a different viewand submit that, under the conceptual domainof entrepreneurial behaviors, innovativenessand proactiveness are inextricably confounded.This confounding leads to our second reasonfor collapsing innovativeness and proactivenessinto a single dimension: while innovation is anecessary condition for entrepreneurship, it isnot sufficient, nor is it meaningfully independentfrom proactiveness in this context (Rosenbusch,Brinckmann, Bausch, 2011). For example, asSchumpeter’s classic works (1934, 1942) notedin their treatment of creative destruction, it is thecombination of innovation (the creation of the new)with the process of commercialization (creatingnew markets and destroying old markets) that isthe defining characteristic of entrepreneurship.The argument is that entrepreneurial firms do notsimply create; entrepreneurial firms create withthe intent of employing those creations to estab-lish market leadership positions, to develop newmarkets, and to preempt competitors (Schumpeter,1942). It is therefore conceptually inconsistentto create a theoretical distinction between thetwo: innovativeness and proactiveness are func-tionally equivalent reflections of the underlyingentrepreneurial behaviors dimension.

Managerial attitude towards risk

Drawing a conceptual distinction between man-agerial attitude towards risk and the exhibition ofentrepreneurial behaviors, however, recognizes theinherent differences between managerial percep-tions of risk and the behaviors pursued by man-agers that result from those perceptions (March andShapira, 1987). Our argument that risk taking is atti-tudinal builds from the observation of March and



Shapira (1987: 1406) that “attitudes towards riskare usually pictured as stable properties of indi-viduals, perhaps related to personality developmentor culture … ” In the context of Ajzen and Fish-bein’s (1977) work on the attitude-behavior linkage,a stable individual property such as risk taking isdistinct from the behavioral manifestations of thatproperty (March and Shapira, 1987). To that end,Douglas and Shepherd (2002), for example, findthat an individual’s attitude towards risk does notperfectly correlate with subsequent entrepreneurialaction.

How managers think about risk, and the orga-nizational actions taken that embody an elementof risk, are therefore conceptually distinct. Whilethe relevant literature thoroughly discusses thedifferences between attitude towards risk andrisk-taking behaviors (Shapira, 1986; Wright et al.,2007), the salient point in the context of EO isthat the entrepreneurial behaviors undertaken bya firm are not a perfect correlate to the strategicdecision maker’s attitude towards risk, as wouldbe implied under a reflective measurement model.While the acts of innovation and new market entryinherently embody risk, the causal factors thatencourage the exhibition of those specific strategicactions are a reflection of organization-level and/orenvironment-level phenomenon facilitating—ordiminishing—their employment. Hence, it isnecessary to treat entrepreneurial behaviors andmanagerial attitude towards risk as separatelower-order dimensions of the higher-order EOconstruct with differing antecedent relationships.

Notably, managerial attitude towards risk existsat the level of the senior-most decision maker taskedwith firm-level strategic decisions. In practice, thisindividual is typically the CEO or managingdirector of the firm, or the senior executive incharge of a strategic business unit with independentprofit-and-loss responsibility within a larger com-pany (Covin and Lumpkin, 2011). The rationaleunderlying this level of analysis is that EO is afirm-level construct, and the behaviors exhibitedunder the EO rubric are most frequently associatedwith firm-level strategic choices (Covin and Slevin,1991). Thus, managerial attitude towards risk existsat the level at which senior executives decide topursue entrepreneurial behaviors as part of theirfirm’s strategy. It is noteworthy that the strategicmanagement literature has long acknowledgedthat the attitudes, managerial philosophies, and

personal proclivities among senior managers reflectin firm-level strategic choices (Hambrick, 2007).

Dimensional covariance

Our reconceptualization also recognizes twocritical requirements fundamental to classifyingentrepreneurial firms (i.e., Covin and Lumpkin,2011). The first such requirement is that a firmmust engage in entrepreneurial behaviors thatinvolve the pursuit of new ideas, processes, andtechnologies, and that a firm must aggressivelyseek to commercialize those ideas by extending itsboundaries to new product/market domains. Thesecond requirement in classifying entrepreneurialfirms is temporal stability—firms must engage inentrepreneurial behaviors with some reasonableconsistency across time. We maintain that manage-rial attitude towards risk correlates with behavioralconsistency. The proclivity of senior managersto pursue projects with uncertain outcomes is afundamental covariate to the sustained pursuit ofentrepreneurial behaviors. As such, we posit amutualistic relationship between the dimensions.As an analogy, we find a similarly mutualisticrelationship between risk-taking disposition andsustained entrepreneurial action in research onserial entrepreneurs (Haynie et al., 2010). Whileat a different level of analysis, an assumptionoften made in serial entrepreneurship researchis that prior entrepreneurial action reinforces thedisposition to engage in future entrepreneurialaction (Haynie et al., 2010).

We note that a common assumption of forma-tive measurement models is the lack of an a prioriexpectation of dimensional covariance (Howell,Breivik, and Wilcox, 2007). As Jarvis et al. (2003)and MacKenzie et al. (2011) observe, however,the requirement under formative measurement isnot that the dimensions cannot covary but ratherthat they do not necessarily covary, as would berequired under reflective measurement. The choiceof dimensional covariance under a formative spec-ification is thus a decision of the researcher andbased on theoretical considerations. The preced-ing is an important distinction and represents akey strength of the formative approach suggestedby our reconceptualization. By specifying a priorithat we expect managerial attitude towards risk andentrepreneurial behaviors to covary—although notperfectly so—we allow for the possibility that thedimensions of EO may share certain antecedents,



but that there may also be other theoreticallymeaningful phenomena within EO’s nomologi-cal network that relate to only one of the twodimensions. Unlike the predominant reflective per-spective in the EO literature, under our approacha researcher need not accept a tradeoff betweenthe granularity of exploring EO’s antecedents whileignoring the broader contribution to the EO lit-erature. Our reconceptualization thus expands thepool of antecedent relationships beyond that ofthe current Miller/Covin and Slevin conceptual-ization, while addressing nomological errors inthe EO literature because of measurement modelmisspecification.

A revised measurement model

We propose a formative measurement model for ourreconceptualization of EO as depicted in Figure 1.Formative measurement models inherently sufferfrom an identification problem—it is impossibleto identify the construct-level disturbance term (𝜁 )absent additional steps taken by the researcher(Bollen and Lennox, 1991). The relevant litera-ture suggests several remedies for identification,although space precludes us from discussing allsuch possibilities (see MacKenzie et al., 2011, fora thorough discussion). We will mention one suchidentification method that not only allows a forma-tively measured construct to be estimated in bothstructural and confirmatory factor models, but alsoaddresses the criticism of formative measurementoffered by Howell et al. (2007) regarding interpre-tational confounding.1

As noted by MacKenzie et al. (2011), the pre-ferred technique for dealing with formative modelidentification is to measure the focal latent constructwith two additional reflective indicators intendedto capture the entirety of the construct’s theoreti-cal domain (i.e., “global” effect measures of EO).The two additional indicators allow estimation ofthe disturbance term from content-valid measuresof the latent construct that are theoretically con-gruent with, and included within, the construct’sconceptual domain (Bollen, 2007). For example,a researcher could ask the respondent to rate thefollowing statements regarding EO: “Overall, I

1 Interpretational confounding exists when there is a discrepancybetween the nominal meaning of a construct (i.e., its theoreticalmeaning) and its empirical meaning as derived from its opera-tionalization (see Jarvis et al., 2003).

Entrepreneurial Orientation

Managerial Attitude Towards Risk

Entrepreneurial Behaviors

ζ

δ δ δ δ δδ

δ

δ

Figure 1. Reconceptualized EO measurement model

would consider my firm to be entrepreneurial,” and“Generally speaking, my company is innovative,actively pursues new markets, and our managers arewilling to take on risk.” As MacKenzie et al. (2011)discuss, this procedure largely resolves the issue ofinterpretational confounding and facilitates the useof formatively measured constructs in a variety ofstructural and confirmatory factor models.

Under our reconceptualization, we therefore sug-gest the following measurement model for EO asshown in Figure 1 to include: (1) at least three reflec-tive indicators for the two lower-order dimensionsto facilitate identification (see Bollen, 1989), (2) thelower-order dimensions forming the higher-orderEO construct, and (3) the inclusion of two addi-tional global reflective indicators used for iden-tification of the construct-level disturbance term.Notably, this measurement model is empiricallyequivalent to a MIMIC model (Multiple Indica-tors, Multiple Causes; see MacKenzie et al., 2011).Conceptually, however, the interpretation is betterunderstood as the latent variable EO defined bytwo lower-order dimensions, with two additionalcontent-valid effect indicators facilitating identifi-cation (MacKenzie et al., 2011). The strength of thepreceding interpretation is ontological consistencybetween our conceptual model of EO and the mea-surement model used for its operationalization.

AN EMPIRICAL ILLUSTRATION

Purpose

Drawing from measurement theory, we based ourEO reconceptualization on conceptual grounds,



consistent with Covin and Lumpkin’s (2011) obser-vation that questions of EO’s ontology are fun-damentally theoretical in nature. Notably, in theEO literature, there are examples of deriving onto-logical conclusions from empirical analysis (e.g.,Kreiser, Marino, and Weaver, 2002). We concurwith Covin and Lumpkin’s (2011) comment that,while such research offers valuable insights, it isinappropriate to derive ontological conclusions ofa theoretical construct through data analysis. Thatsaid, we believe it worthwhile to offer an empiri-cal illustration describing how EO scholars mightuse our reconceptualization in their research design.The purpose of this illustration is neither to empir-ically validate our reconceptualization nor to rig-orously test an existing nomological relationship.Rather, we simply wish to illustrate the value of ourreconceptualization using an empirical example.

Sample and data collection

We collected data from the senior-most executiveat 610 small to medium-sized South Korean busi-nesses randomly selected from the membershiprolls of the Korean Venture Business Association(KOVA). KOVA is a business trade organizationlocated in Seoul, South Korea, that operatesanalogously to a chamber of commerce in theUnited States. We collected data in partnershipwith KOVA, with the number of respondents drivenprimarily by budgetary constraints. Specifically,our budget allowed for the collection of roughly600 responses from KOVA’s 11,248 members, for aresponse rate of≈ 5 percent. We contacted memberbusinesses by phone until we obtained the predeter-mined number of responses. T-test comparisons offirm age and firm size between the responding andnonresponding firms revealed no significant differ-ences; neither did comparisons between early andlate respondents. The survey was originally writtenin English, translated into Korean by a nativeKorean language speaker fluent in English, andthen back-translated into English by a second nativeKorean speaker to verify that the translation processdid not materially alter the meaning of the indicators(Brislin, 1980).

Model specification

Our empirical illustration draws from the observa-tion of Covin and Slevin (1991) that environmentalexigencies, including environmental hostility,

causally influence EO. Environmental hostilityrefers to industry conditions characterized by“precarious industry settings, intense competition,harsh, overwhelming business climates, and therelative lack of exploitable opportunities” (Covinand Slevin, 1989: 75). Though other componentsof the firm’s task environment are a source ofcausal influence, environmental hostility playsa fundamental role in understanding a firm’sentrepreneurial strategic posture (Covin and Slevin,1991). The logic for causal adjacency betweenhostility and EO is that in hostile environments,where resources are scarce and growth opportu-nities limited, firms achieve superior performanceby following tried-and-true strategies that do notthreaten the firm’s survival (Miller and Friesen,1983). This logic suggests a negative relationshipbetween hostility and EO. Replicated in Figure 2,Rosenbusch et al. (2013) tested this relationshipdrawing from meta-analytic data and depicting EOmediating the relationship between hostility andfirm performance, and found no direct relationshipbetween hostility and EO (corrected r = -0.06,p> 0.05). Based on these findings, and the con-tradictory evidence in the broader EO literature,Rosenbusch et al. (2013) concluded that there islikely no meaningful causal connection betweenenvironmental hostility and EO.

When viewed through the lens of our recon-ceptualization, however, we posit that hostility iscausally adjacent to EO, although it does so byinfluencing entrepreneurial behaviors, while likelyexhibiting no significant influence on managerialattitude towards risk. We predicate our argument onthe observation that hostile environments constrainstrategic choice and favor the pursuit of moreconservative strategies (Miller and Friesen, 1983).However, these environmental constraints actprimarily to influence behavior; it is not likely thata firm’s task environment will induce discerniblechanges to a senior manager’s attitude towards risk,which is itself argued to be a more stable propertyof the senior manager (March and Shapira, 1987).Presenting a revised research model in Figure 3,we again suggest that EO mediates the relationshipbetween hostility and performance. However, wedecompose the mediation effect into hostility’srelationship with EO’s lower-order dimensions:EO mediates the relationship between environmen-tal hostility and firm performance such that theeffect of hostility on performance is transmittedthrough a negative relationship with entrepreneurial



Figure 2. Structural model—EO (Miller/Covin and Slevin first-order reflective measurement model). While this figuredepicts a first-order reflective measurement model, there is no material difference either conceptually or empiricallybetween this model and a first-order reflective, second-order reflective measurement model wherein indicators ofinnovativeness, proactiveness, and risk-taking load on lower-order reflective dimensions of a reflectively construed

first-order EO construct. See Bollen (1989)

Figure 3. Structural model—reconceptualized EO (first-order reflective; second-order formative measurement model).EntBehaviors= entrepreneurial behaviors; MATR=managerial attitude towards risk



behaviors, while hostility does not exhibit a sig-nificant relationship with managerial attitudetowards risk.

Measures

Given that we ground our reconceptualizationin the Miller/Covin and Slevin perspective, wemeasured EO using the widely employed StrategicPosture/EO scale developed by Covin and Slevin(1989). We chose the Covin and Slevin (1989) EOscale in part due to its popularity, but also to illus-trate that, with small modifications, EO scholarsmay employ the scale in a manner consistent withour reconceptualization. There are nine indicatorsin the Covin and Slevin (1989) EO scale—threeeach for the components of innovativeness, proac-tiveness, and risk taking—and Runyan et al.(2011) found that the scale is robust in internationalresearch settings. Please see supporting informationAppendix S2 for a description of all focal scalesused in our empirical illustration.

We added two indicators to the Covin and Slevin(1989) scale to capture EO’s global conceptualdomain. As discussed previously, these indicatorsare necessary for identification purposes whenspecifying EO formatively. Respondents wereasked on a seven-point Likert scale the extent towhich they agreed with the following items, withthe number 4 indicating a neutral position: (1) “Ingeneral, my business unit … 1—prefers to be cau-tious when considering new opportunities, to shyaway from overly risky initiatives, and prefers to letrivals take the lead on innovation in our industry,”to 7—“is on the cutting edge when it comes to

exploiting entrepreneurial opportunities becauseof our demonstrated ability to embrace novel andrisky initiatives;” and (2) “In general, I wouldconsider my business unit to be … 1—managedwith a more conservative, risk-adverse manage-rial philosophy,” to 7—“managed with a moreentrepreneurial, innovation-centric managerialphilosophy.” In so doing, our global indicatorstapped into both behavioral and attitudinal elementsof EO, making them each content-valid indicatorsof EO’s conceptual domain as defined in ourreconceptualization.

We measured environmental hostility using afive-item Likert scale commonly found in the EOliterature (Covin and Slevin, 1989). We measuredperformance using five indicators of the seniormanager’s satisfaction with total sales, cash flow,return on equity, net income ratio, and the abilityto fund growth from profits. Table 1 includes thesummary statistics and correlation matrix for ourfocal constructs including EO as a unidimensionalcomposite construct; entrepreneurial behaviorsas a lower-order, reflectively measured construct;managerial attitude towards risk as a lower-order,reflectively measured construct; and two globalindicators of EO. Note that we would expect strongcovariance between the two global indicatorsbecause they are both content-valid measurescapturing the totality of EO’s conceptual domain.Additionally, we included the following controls:the log of firm revenue, the log of firm employees,the log of firm age, a dummy variable indicatingwhether or not the firm was in a high-technologyindustry, and environmental dynamism.

Table 1. Summary statistics and correlation matrixa

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

1. EO 4.41 0.99 0.862. Entrepreneurial behaviors 4.52 1.02 0.79 0.953. MATR 4.19 1.19 0.85 0.85 0.654. Performance 11.80 4.31 0.91 0.22 0.24 0.145. Hostility 4.02 0.96 0.67 −0.08 −0.09 −0.04 −0.126. EO Global 1 4.45 1.60 — 0.61 0.55 0.57 0.13 −0.057. EO Global 2 4.40 1.42 — 0.58 0.50 0.58 0.12 −0.08 0.508. Age 9.70 7.25 — −0.06 −0.03 −0.09 0.10 0.06 −0.12 −0.159. Employees 2.80 1.15 — 0.09 0.10 0.05 0.26 -0.01 −0.05 −0.06 0.42

10. Turnover 7.72 1.84 — 0.08 0.09 0.04 0.30 0.05 −0.05 −0.07 0.44 0.7911. High tech industries 0.47 0.50 — −0.02 −0.02 −0.02 −0.05 0.03 −0.04 0.04 −0.05 0.03 −0.0112. Dynamism 4.24 0.95 0.69 0.17 0.21 0.07 0.26 −0.05 0.10 0.10 0.03 0.14 0.17 −0.10

N = 610a Correlations greater than± 0.08 are significant at p< 0.05. Age, employees, and turnover log transformed. EO= entrepreneurial orientation, measured asa first-order reflective construct; MATR=managerial attitude towards risk.



Analysis

We began our analysis by estimating a confirma-tory factor model (CFA) of our focal variables.We estimated all models using the structuralequation modeling module of Stata 12.1 (StataCorporation, 2012). For the initial CFA, we mod-eled EO following Covin and Slevin (1989) as aunidimensional composite construct wherein allnine indicators loaded on the latent EO construct.All indicators related positively and significantlyto their intended constructs (p< 0.001), andwe observed no significant cross-loadings. Theoverall model, however, exhibited relatively poorfit (𝜒2 = 660.148, p< 0.001; RMSEA= 0.075;CFI= 0.891; TLI= 0.875; SRMR= 0.064). Exami-nation of the modification indices suggested freeingthe covariance between the error terms of INN2and INN3, between PRO1 and PRO2, betweenINN1 and PRO1, between RISK2 and RISK3,and between PERF1 (Performance) and PERF3.These minor steps are reasonable approaches toimprove model fit and are appropriate given that themeasures are all indicators of the same latent con-struct (Bollen, 1989). The revised CFA exhibitedsignificantly improved fit (𝜒2 = 357.951, p< 0.001;𝜒2

DIFF = 302.197, p< 0.001; RMSEA= 0.049;CFI= 0.954; TLI= 0.946; SRMR= 0.058).

We then replicated the Rosenbusch et al. (2013)study by estimating a structural model with envi-ronmental hostility as a predictor of EO, and EOpredicting firm performance, shown in Figure 2. Weagain followed the Miller/Covin and Slevin concep-tualization and measured EO as a unidimensionalcomposite construct wherein all nine EO indicatorsloaded on the first-order factor. This model demon-strated acceptable fit (𝜒2 = 364.008, p< 0.001;RMSEA= 0.050; CFI= 0.953; TLI= 0.945;SRMR= 0.061). However, the standardized pathcoefficient between environmental hostility and EOwas nonsignificant (𝛽 = -0.087, p> 0.1), althoughwe did find a positive and significant relation-ship between EO and performance (𝛽 = 0.207,p< 0.001). Thus, our initial results largely mirroredthose as reported in Rosenbusch et al. (2013).

We then turned to our reconceptualization. Webegan with a CFA of performance, hostility, andEO following our proposed reconceptualizationas outlined in Figure 1. In this model, the threeinnovativeness indicators and the three proac-tiveness indicators load on the entrepreneurialbehaviors construct, and the three risk-taking

indicators load on the managerial attitude towardsrisk construct. In turn, entrepreneurial behav-iors and managerial attitude towards risk relatecausally to the higher-order EO construct, andwe identified the construct-level disturbanceterm with our two global reflective indicators.Maintaining the error covariance modificationssuggested in our initial analysis, the resultingCFA fit the data well (𝜒2 = 404.579, p< 0.001;RMSEA= 0.046; CFI= 0.957; TLI= 0.949;SRMR= 0.051).

It is important to note here that the eval-uation of construct validity for formativelymeasured constructs differs demonstrably fromthat for reflectively measured constructs. Forexample, the concept of internal consistencyreliability (i.e., Cronbach’s alpha) among for-mative indicators does not apply because ofthe presumption that the lower-order dimen-sions are conceptually independent (MacKenzieet al., 2011). However, we may evaluate thereliability of the two lower-order reflectivelymeasured dimensions separately. In this instance,the estimated reliability of the entrepreneurialbehaviors construct and the managerial attitudetowards risk construct are 𝛼 = 0.79 and 𝛼 = 0.85,respectively. We further examined the validityof our measurement model by evaluating thestandardized path coefficients between the twolower-order dimensions and the higher-order EOconstruct. Both parameter estimates were pos-itive and significant (𝛽ENT BEHAVIORS = 0.445,p< 0.001; 𝛽MATR = 0.505, p< 0.001). Thestandardized loadings of the two global reflec-tive indicators of EO were also positive andsignificant (𝜆EO GLOBAL 1 = 0.708, p< 0.001;𝜆EO GLOBAL 2 = 0.701, p< 0.001). Further, EO’scombined indicators accounted for 84.2 percentof EO’s total variance (𝜎2

EO = 0.9182), indicatingthat the specified measures were responsible forthe majority of the variance in the higher-orderEO construct (Diamantopoulos et al., 2008). Col-lectively, these results corroborate our specifiedmeasurement model.

We then estimated our structural model usingthe reconceptualized EO variable, as depictedin Figure 3 (error term covariance modificationsexcluded for parsimony). In this model, we esti-mated paths from hostility to entrepreneurialbehaviors and from hostility to managerial attitudetowards risk, which in turn are lower-order pre-dictors of the higher-order EO construct, with the



Table 2. Model estimation resultsa

Model 1:EOUnidemensional

Model 2:EOReconceptualization

Structural parameter 𝛽 s.e. 𝛽 s.e.

Hostility→EO −0.087 0.054EO→Performance 0.208*** 0.043Hostility→Entrepreneurial behaviors −0.150** 0.058Hostility→MATR −0.060 0.054Entrepreneurial behaviors→EO 0.549*** 0.132MATR→EO 0.416*** 0.128EO→Performance 0.213*** 0.044𝜎2 Entrepreneurial behaviors 0.022𝜎2 MATR 0.004𝜎2 EO 0.007 0.868𝜎2 Performance 0.043 0.045𝜒2 364.008*** 421.964***RMSEA 0.050 0.047CFI 0.953 0.954TLI 0.945 0.946SRMR 0.061 0.057

N = 610a Standardized coefficients; 𝜎2 = amount of explained variance, or equivalently, the R2; EO= entrepreneurial orientation;MATR=managerial attitude towards risk.*p< 0.05; **p< 0.01; ***p< 0.001

construct-level disturbance term identified by thetwo global reflective indicators. As before, we alsoestimated the relationship between EO and per-formance. This model demonstrated acceptable fit(𝜒2 = 421.964, p< 0.001; RMSEA= 0.047; CFI=0.954; TLI= 0.946; SRMR= 0.057). Consistentwith our expectations, we observed a negativeand significant relationship between hostility andentrepreneurial behaviors (𝛽 =−0.150, p< 0.01).Further, the path between hostility and man-agerial attitude towards risk was nonsignificant(𝛽 =−0.060, p> 0.1). We also found a posi-tive relationship between EO and performance(𝛽 = 0.213, p< 0.001). As a robustness test, were-estimated our model with our control variables.This analysis revealed no material changes inthe sign or significance of the estimated focalcoefficients. Lastly, it is worthwhile to note that ourmodel fit indices changed little between the modelsas shown in Figures 2 and 3. This is consistent withthe observation of Jarvis et al. (2003) that modelfit indices are rarely influenced by measurementmodel misspecification and cannot be relied uponwhen evaluating the influence of misspecification.We summarize the primary results from bothstructural models in Table 2.

Summary

Our empirical illustration highlights the problem ofnomological error in the EO literature due to mea-surement model misspecification. In our structuralmodel using a first-order, unidimensional conceptu-alization of EO, we found no significant relationshipbetween hostility and EO. Adopting our reconcep-tualization, however, we identified a significant neg-ative relationship between hostility and EO, withhostility transmitting that effect through a firm’sentrepreneurial behaviors, but not through man-agerial attitude towards risk. Hence, measurementmodel misspecification resulted in a Type II errorin the case of the unidimensional model, as wouldbe expected (MacKenzie et al., 2005). The findingof a Type II error carries significant ramificationsfor EO research, although admittedly such impactis not directly quantifiable. Given the bias towardspublishing significant findings (i.e., non-null rela-tionships) in the management literature (Pfeffer,2007), we can only speculate as to the number of“true” antecedent relationships incorrectly rejectedby EO scholars during the early stage of data anal-ysis because of misspecification. Our belief is thatsuch occurrence is likely high, particularly given thesignificant quantity differential between research on



EO’s consequences and on its antecedents (Rauchet al., 2009).

Our empirical illustration does have certainlimitations. For example, the analyses usedcross-sectional data and subjective performanceindicators. However, the primary purpose of ourempirical example was not to focus on the nuancesof the hostility-EO relationship—althoughour findings may prompt further study in thisregard—but rather to illustrate how our recon-ceptualization alleviates nomological error in EOresearch. Ultimately, our illustration underscoreswhy congruence between the theoretical meaningof a construct and the approach used to measureit is vital for drawing nomological conclusions(Bagozzi, Yi, and Phillips, 1991).

DISCUSSION AND CONCLUSION

As Miller (2011) recently noted, while theMiller/Covin and Slevin EO conceptualization hasserved the field well, the time has come to revisitthe core ontological assumptions underpinning theconceptual domain of firm-level EO. In the spiritof Miller’s (2011) observation, our reconceptual-ization shows that we can recast EO in a mannerthat remains largely consistent with its originalformulations while addressing the ontological andmeasurement inconsistencies hampering knowl-edge creation in the field. We suggest that ourproposed reconceptualization offers a number ofimportant advantages to EO scholars.

Revisiting existing nomological relationships

Our empirical illustration identified a Type IIerror—a failure to reject a false nullhypothesis—resulting from measurement modelmisspecification. One implication from this findingis that there is a distinct possibility that such mis-specification is the reason for the preponderanceof null findings in hostility-EO research (Rosen-busch et al., 2013). However, the implicationof nomological error from measurement modelmisspecification extends beyond the hostility-EOrelationship. Unfortunately, absent empiricallyreplicating past EO-antecedent research, we cannotconclusively determine the extent of such error inthe literature, although simulation-based researchon measurement model misspecification suggeststhat nomological error rates may be worrisomely

high (George, 2011; Jarvis et al., 2003; MacKenzieet al., 2005).

Confounding the preceding problem is thatscholars may be tempted to jettison insignificantrelationships during the early stages of data anal-ysis in favor of reporting supported hypotheses(Pfeffer, 2007). As such, EO scholars may haveinadvertently ignored theoretically meaningfulcontributing factors by erroneously believing thosefactors to be inconsequential or insignificant.Importantly, simply suggesting that measuringEO formatively versus reflectively does not fullyaddress the ontological challenges discussed at thebeginning of the current paper. Addressing thesechallenges requires reexamining EO’s ontologicalassumptions to recognize that a firm’s strategicposture arises through the observation of thefirm’s behaviors, and that how managers viewrisk is a necessary covariate with those behav-iors. Our reconceptualization accomplishes theserequirements, and suggests that viewing existingantecedent relationships through a new lens mayfundamentally alter how scholars examine andinterpret EO’s contributing factors.

To illustrate, consider two prior studies incorpo-rating antecedent relationships to EO and employ-ing the common unidimensional conceptualizationof EO. While both studies draw conclusions con-sistent with the unidimensional conceptualizationof EO, we wish to illustrate how the results ofthese studies may have changed under our recon-ceptualization and when employing a formativemeasurement model. In their study of governancefactors and entrepreneurial orientation among non-profit organizations, Coombes et al. (2011) arguedthat board activism—activism referring to the vigorwith which a board embraces its governance andoversight duties—positively relates to EO, and“translates into intentional efforts to bring aboutchange via new approaches,” (Coombes et al. 2011:837). Hence, activist boards are more likely toencourage entrepreneurial activities. Building fromthe preceding logic, we might speculate that anactivist board’s desire to bring about change mayresult in a positive relationship between activismand the lower-order dimension of entrepreneurialbehaviors. However, we could envision a negativerelationship between board activism and managerialattitude towards risk, building on Gomez-Mejia andWiseman’s (1997) observation that activist boardsmay diminish a CEO’s attitudinal proclivity to com-mit resources to projects with uncertain outcomes.



Board activism may thus differentially relate toEO’s two lower-order dimensions, although this isan empirical question.

Consider further a more complicated causalmodel offered by De Clerq, Dimov, and Thongpa-panl (2013), wherein knowledge sharing within afirm positively relates to EO, routine formalizationrelates negatively to EO, and routine formalizationmoderates the relationship between knowledgesharing and EO. Specifically, De Clerq et al.(2013) hypothesized that the positive relationshipbetween knowledge sharing and EO is greaterwhen formalization is low (i.e., formalization neg-atively moderates the knowledge sharing-EO path).However, our reconceptualization suggests a morecomplex moderation model. For example, we mightargue that routine formalization strengthens theknowledge sharing-entrepreneurial behaviors rela-tionship, because formalization focuses managerialattention on higher-value opportunities (Ocasio,1997). Conversely, because of formalization’stendency to diminish managerial risk-taking, wemight posit a differential moderating relationship offormalization on the knowledge sharing-managerialattitude towards risk relationship. We could furtherspeculate that the main effect is actually contingenton formalization—there is little a priori reasonto suspect that knowledge sharing influencesmanagerial attitude towards risk except in thepresence of routine formalization. Again, our pointwith these examples is not to suggest that the pre-ceding studies are normatively wrong. Rather, byviewing these relationships through the lens of ourreconceptualization, scholars may recognize theacuteness of the attitudinal/behavioral distinctionand the advantage of a formative specificationusing our reconceptualization.

Causal inference, new indicators, and newdesigns

While revisiting past nomological relationships inthe EO literature has merit, we submit that the mostsignificant contribution of our reconceptualizationis to move the EO conversation forward in excitingand promising new directions.

Drawing causal inference

There is a distinct disadvantage in adopting theexisting unidimensional conceptualization of EOwhen designing studies to draw causal inferences

between EO and antecedent phenomenon. Becausestrategy and entrepreneurship scholars generallycannot randomly assign behaviors to a populationof companies (an experiment), researchers mustdeal with the endogeneity present in a researchmodel using other techniques to evaluate causaladjacency (Antonakis et al., 2010). While endo-geneity may arise from reverse causality, there aremyriad causes inducing a correlation between theerror term in a regression equation and the focalpredictor variables (we refer interested readers toAntonakis et al.’s, 2010, exemplary discussion ofendogeneity and its sources). However, we do wishto highlight one source of endogeneity—omittedvariable bias—and discuss why this is particularlyproblematic for EO antecedent research using theexisting unidimensional conceptualization.

Consider a standard regression equation account-ing for omitted variables:

EO = 𝛼 + Xt𝛽t + Xu𝛽u + 𝜖 (3)

In Equation 3, Xt represents an observed pre-dictor variable expected to causally influence thecriterion, EO. Xu represents a vector of variablesthat also causally influences EO, but we do notdirectly observe these variables. In a randomizedexperiment, conditions are ideally created to elimi-nate or to effectively marginalize Xu such that whatremains is a consistent and unbiased estimate of theeffect of Xt on EO (i.e., Xu𝛽u = 0). Such conditionsrarely present themselves in EO research, and with-out addressing Xu, scholars draw inappropriate con-clusions regarding EO’s causal factors because ofbias in the estimate of Xt (Antonakis et al., 2010).Unfortunately, the most common technique to dealwith omitted variable bias—the use of instrumentalvariables—presents its own challenges for scholarsusing the unidimensional conceptualization of EO.

Consider that for an instrument (z) to be valid,it must (1) significantly covary with the assumedendogenous variable (x) (e.g., COV(z,x)≠ 0);and (2) be strictly exogenous (i.e., COV(z,𝜖)= 0)(Wooldridge, 2010). For scholars adopting the uni-dimensional operationalization of EO, this meansfinding an instrument strongly correlated with allthree lower-order dimensions of the same directionand of the same magnitude, but not correlatedwith the equation disturbance term. Given theacknowledged difficulties in finding valid instru-ments (Semadeni, Withers, and Certo, 2014), it is,we suggest, very difficult to meet such a standard.



Under our reconceptualization, the researchermust identify more total instruments because ofthe decomposition of EO into its two lower-orderdimensions (at least two per endogenous variable;see Wooldridge, 2010). However, we maintainthat, by distinguishing between the behavioral andattitudinal components of EO, identifying suchinstruments is easier because we clearly delineatethe conceptual domain of each dimension. As such,our reconceptualization is more robust to researchdesigns seeking to draw causal inference (Bagozziet al., 1991).

New indicators

There is a growing call in the EO literature tocreate new operationalizations and measures of EO(George and Marino, 2011; Miller, 2011). We agreewith past conclusions that the Covin and Slevin(1989) psychometric scale has been exceedinglyvaluable to EO scholars (Covin and Wales, 2012).However, as Covin and Lumpkin (2011) observed,the popularity of the Covin and Slevin (1989)scale may also have been unintentionally limiting,as its widespread adoption likely discouraged thedevelopment of alternate measures of EO. Whileour empirical illustration suggests that scholarsmay use the Covin and Slevin (1989) scale ina manner consistent with our reconceptualization,we also add to the call for new EO indicators.Indeed, there is legitimate concern that retrospectivesummated scales may be inadequate in capturingthe conceptual domain of EO, and particularly inassessing managerial attitude towards risk.2 Underthe rubric of our reconceptualization, we submit thatEO scholars could embrace new indicators that offerexciting new possibilities.

For example, one area of interest is functionalmagnetic resonance imaging (fMRI) techniquesused to measure the respondent’s brain activitywhen evaluating risky or uncertain strategic deci-sions (Powell, 2011). One caveat to the precedingis that EO is fundamentally a firm-level constructand that managerial attitude towards risk resides atthe level of the senior manager ultimately respon-sible for strategic decision making, and access tothese individuals, particularly in larger firms, isadmittedly difficult. Nonetheless, we maintain thatresearch designs employing such an indicator of

2 We thank an anonymous reviewer for this insight.

managerial attitude towards risk, combined withsurvey and secondary data for the entrepreneurialbehaviors construct (e.g., patents, R&D expendi-tures, new product introductions), creates a pow-erful indicator set under our reconceptualization.Such indicators also address a number of concernsregarding the validity of single-respondent surveydata that currently dominates EO research (Miller,2011), and may further lend themselves well toexperimental and quasi-experimental designs.

Alternative research designs and estimators

Consider that, in our empirical illustration, weassumed EO as latent, and estimated followingEquation 2 with a disturbance term, 𝜁 , representingthe amount of variance in the latent construct notaccounted for by its indicators. An alternate methodto model a formative construct is to assume 𝜁 = 0,represented by the following equation:

𝜂 =n∑

(i=1)𝛾ixi (4)

Here, we again define 𝜂 by the weighted totalvariance of its indicators. However, because we alsoposit that the indicators are error free, 𝜂 effectivelybecomes an observed variable (Bollen, 2007). Werefer to this type of specification as a summative, orcomposite, index (Diamantopoulos and Winklhofer,2001).

A weakness of Equation 4 is the loss of the dis-turbance term (the amount of variance in 𝜂 notaccounted for by its indicators), rendering coeffi-cient estimates of variables inconsistent, even withoptimum weighting of the indicators (Bollen andLennox, 1991). Equation 4, however, does havecertain advantages. The first advantage is that theresulting value of 𝜂 does not suffer from the identifi-cation problem mentioned previously necessitatingadditional reflective indicators. The second advan-tage of Equation 4, if 𝛾 is set equal to one or ifusing weights derived from principal componentsanalysis, is that a researcher may use the resultingvalue of 𝜂 in a wide range of estimators includingregression, nonlinear models, dynamic models, andso forth (Bollen, 2007) as both an antecedent vari-able and as a criterion. Such a specification mayenable researchers to adopt new indicators and newresearch designs that push the EO conversation for-ward, assuming that such measurement limitationsare duly noted.



While admittedly not as preferred as thelatent variable structural equation modelingapproach, we further propose that a related struc-tural equation estimator—seemingly unrelatedregression—offers promise for EO scholars,particularly regarding contributory factors toEO’s lower-order dimensions (Wooldridge, 2010).Depicted in Equations 5 and 6, a researchersimultaneously estimates a model where differ-ent variables, represented as X and Z, predictentrepreneurial behaviors and managerial attitudetowards risk.3 The researcher also allows the errorterms, represented as 𝜖1 and 𝜖2, respectively, tocovary (e.g., COV[𝜖1,𝜖2]≠ 0). The explicit errorcovariance reflects the theoretical development ofour reconceptualization as a mutualistic relation-ship between the two lower-order EO dimensions.

Entrepreneurial Behaviors = 𝛼1 + 𝛽1X + 𝜖1 (5)

Managerial Attitude towards Risk = 𝛼2 + 𝛽2Z + 𝜖2(6)

A weakness of the preceding approach is thata researcher is unable to link the two lower-orderdimensions to the higher-order EO construct,limiting the theoretical utility of the research model(MacKenzie et al., 2011). However, this approachdoes have three important strengths. The firststrength of this approach is to accommodate a widevariety of indicators. While survey data could beused, secondary financial ratios and categorical,count, and dichotomous measures—which canbe difficult to employ using covariance-basedstructural equation modeling—could be used toconstruct the two lower-order EO dimensions.The second strength of this approach is to betteraccommodate longitudinal data where econometricmethods are often more robust to data of this type(Wooldridge, 2010). Additionally, with the inclu-sion of valid instruments, a researcher could easilyconvert a seemingly unrelated regression modelto a three-stage least squares model, which allowsthe researcher to account for potential endogeneitywithin the system and thus helps facilitate causalinference (Antonakis et al., 2010).

An additional strength of Equations 5 and 6,and one we believe encourages several promising

3 Note that X and Z could also be modeled as predicting bothentrepreneurial behaviors and managerial attitude towards risk in amultivariate seemingly unrelated regression (Wooldridge, 2010).

research opportunities, is the ability to estimatemodels with contextual effects influencing therelationship between a given antecedent and EO’slower-order dimensions. Such a specification mayprove fruitful for EO scholars because both directand contextual factors likely differentially influ-ence the two lower-order dimensions. For example,Wales, Patel, and Lumpkin (2013) posit thatCEO narcissism may increase a strategic decisionmaker’s risk-taking proclivities, although a firm’scompensation structure (March and Shapira, 1987),for example, might moderate this relationship. Itis certainly possible to estimate such contextualeffects using structural equation modeling. How-ever, given the significant interest in boundarycondition research in the EO literature (Covin andLumpkin, 2011), additional analytical techniquessuch as those proposed in Equations 5 and 6 opena number of new research streams for EO scholars.

To illustrate further, consider the nomologicalrelationship between EO and various manifestationsof organizational learning and knowledge manage-ment (e.g., Anderson et al., 2009; Kreiser, 2011).A frequently mentioned perspective in the EO con-versation is the reciprocally causal relationshipbetween EO and learning/knowledge exploitation;namely, that the enactment of entrepreneurial strate-gies generates knowledge that becomes the basisfor further entrepreneurial initiatives (Andersonet al., 2009). The theoretical mechanisms behindsuch knowledge generation and exploitation remainunderexplored, but, as Kreiser (2011) suggests, thetypes of knowledge generated by entrepreneurialstrategies and the types of knowledge necessaryto facilitate new market entry are likely to bedistinct. Under our reconceptualization, scholarsmay be able to peer further into this black boxby identifying knowledge typologies that facilitateentrepreneurial behaviors, the learning and psycho-logical processes influencing managerial attitudetoward risk, and the differing contextual factorsinfluencing both relationships in a single researchmodel using a variety of indicators and adoptingrecursive and nonrecursive designs.

CONCLUSION

While EO scholars generated significant insightsover the past three decades, much work remainsto expand our knowledge of entrepreneurial strate-gies and the causes and consequences of what it



means to “be entrepreneurial” at the firm level. Thispaper provides specific recommendations to facili-tate the resolution of nomological error in the EOliterature that may be limiting EO’s contributionto the strategic management and entrepreneurshipfield. By reconceptualizing EO in a manner congru-ent with its well-defined theoretical domain whilesimultaneously addressing relevant conceptual andmeasurement issues, we hope to propel EO researchalong an exciting new trajectory.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge the support andcounsel of Jeff Covin in the development of thismanuscript, and of Scott MacKenzie for technicalassistance. The authors also wish to thank JosephMahoney and two anonymous reviewers for theirpositive and developmental comments to improvethis study.

REFERENCES

Ajzen I, Fishbein M. 1977. Attitude-behavior relation: atheoretical analysis and review of empirical research.Psychological Bulletin 84: 888–918.

Anderson BS, Covin JG, Slevin DP. 2009. Understand-ing the relationship between entrepreneurial orienta-tion and strategic learning: an empirical investigation.Strategic Entrepreneurship Journal 3: 219–241.

Antonakis J, Bendahan S, Jacquart P, Lalive R. 2010. Onmaking causal claims: a review and recommendations.Leadership Quarterly 21: 1086–1120.

Bagozzi RP, Yi Y, Phillips LW. 1991. Assessing constructvalidity in organizational research. Administrative Sci-ence Quarterly 36: 421–458.

Bollen KA. 1989. Structural Equations with Latent Vari-ables. Wiley: New York.

Bollen KA. 2007. Interpretational confounding is due tomisspecification, not to type of indicator. PsychologicalMethods 12: 219–228.

Bollen KA, Lennox R. 1991. Conventional wisdom onmeasurement: a structural equation perspective. Psy-chological Bulletin 110: 305–314.

Borsboom D. 2005. Measuring the Mind: ConceptualIssues in Contemporary Psychometrics. CambridgeUniversity Press: Cambridge, UK.

Brislin RW. 1980. Translation and content analysis of oraland written materials. In Handbook of Cross-CulturalPsychology, Triandis HC, Berry JW (eds). Allyn &Bacon: Boston, MA; 389–444.

Coombes SMT, Morris MH, Allen JA, Webb JW. 2011.Behavioral orientations of non-profit boards as a factorin entrepreneurial performance: does governance mat-ter? Journal of Management Studies 48: 829–856.

Covin JG, Green KM, Slevin DP. 2006. Strategic processeffects on the entrepreneurial orientation-sales growthrate relationship. Entrepreneurship: Theory and Prac-tice 30: 57–81.

Covin JG, Lumpkin GT. 2011. Entrepreneurial orien-tation theory and research: reflections on a neededconstruct. Entrepreneurship: Theory and Practice 35:855–872.

Covin JG, Slevin DP. 1989. Strategic management of smallfirms in hostile and benign environments. StrategicManagement Journal 10: 75–87.

Covin JG, Slevin DP. 1991. A conceptual model ofentrepreneurship as firm behavior. Entrepreneurship:Theory and Practice 16: 7–25.

Covin JG, Wales WJ. 2012. The measurement ofentrepreneurial orientation. Entrepreneurship: Theoryand Practice 36: 677–702.

De Clerq D, Dimov D, Thongpapanl N. 2013. Organiza-tional social capital, formalization, and internal knowl-edge sharing in entrepreneurial orientation formation.Entrepreneurship: Theory and Practice 37: 505–537.

Diamantopoulos A, Riefler P, Roth KP. 2008. Advancingformative measurement models. Journal of BusinessResearch 61: 1203–1218.

Diamantopoulos A, Winklhofer HM. 2001. Index con-struction with formative indicators. Journal of Market-ing Research 38: 269–277.

Douglas EJ, Shepherd DA. 2002. Self-employment as acareer choice: attitudes, entrepreneurial intentions andutility maximization. Entrepreneurship: Theory andPractice 26: 81–90.

Fishbein M, Ajzen I. 1975. Belief, Attitude, Intention andBehavior: An Introduction to Theory and Research.Addison-Wesley: Reading, MA.

George BA. 2011. Entrepreneurial orientation: a theoret-ical and empirical examination of the consequences ofdiffering construct representations. Journal of Manage-ment Studies 48: 1291–1313.

George BA, Marino L. 2011. The epistemology ofentrepreneurial orientation. Entrepreneurship: Theoryand Practice 35: 989–1024.

Gomez-Mejia L, Wiseman RM. 1997. Reframing execu-tive compensation: an assessment and outlook. Journalof Management 23: 291–374.

Hambrick DC. 2007. Upper echelon’s theory: an update.Academy of Management Review 32: 334–343.

Haynie JM, Shepherd D, Mosakowski E, Earley PC. 2010.A situated metacognitive model of the entrepreneurialmindset. Journal of Business Venturing 25: 217–229.

Howell RD, Breivik E, Wilcox JB. 2007. Reconsideringformative measurement. Psychological Methods 12:205–218.

Jarvis CB, MacKenzie SB, Podsakoff PM. 2003. A criticalreview of construct indicators and measurement modelmisspecification in marketing and consumer research.Journal of Consumer Research 30: 199–218.

Khandwalla PN. 1977. Some top management styles, theircontext and performance. Organization and Adminis-trative Sciences 7: 21–51.

Koslowsky M, Sagie A, Krausz M, Singer AD. 1997. Cor-relates of employee lateness: some theoretical consid-erations. Journal of Applied Psychology 82: 79–88.



Kreiser PM. 2011. Entrepreneurial orientation and orga-nizational learning: the impact of network range andnetwork closure. Entrepreneurship: Theory and Prac-tice 35: 1025–1050.

Kreiser PM, Marino LD, Weaver KM. 2002. Assessing thepsychometric properties of the entrepreneurial orienta-tion scale. Entrepreneurship: Theory and Practice 26:71–95.

Lumpkin GT, Dess GG. 1996. Clarifying theentrepreneurial orientation construct and linkingit to performance. Academy of Management Review21: 135–172.

Lumpkin GT, Dess GG. 2001. Linking two dimensionsof entrepreneurial orientation to firm performance.Journal of Business Venturing 16: 429–451.

MacKenzie SB, Podsakoff PM, Jarvis CB. 2005. The prob-lem of measurement model misspecification. Journal ofApplied Psychology 90: 710–730.

MacKenzie SB, Podsakoff PM, Podsakoff NP. 2011. Con-struct measurement and validation procedures in MISand behavioral research. MIS Quarterly 35: 293–334.

March JG, Shapira Z. 1987. Managerial perspectiveson risk and risk-taking. Management Science 33:1404–1418.

Miller D. 1983. The correlates of entrepreneurship in threetypes of firms. Management Science 29: 770–791.

Miller D. 2011. Miller (1983) revisited: a reflection onEO research and some suggestions for the future.Entrepreneurship: Theory and Practice 35: 873–894.

Miller D, Friesen PH. 1983. Strategy-making and environ-ment: the third link. Strategic Management Journal 4:221–235.

Miller D, Le Breton-Miller I. 2011. Governance, socialidentity, and entrepreneurial orientation in closely heldpublic companies. Entrepreneurship: Theory and Prac-tice 35: 1051–1076.

Mintzberg H. 1973. Strategy-making in three modes. Cal-ifornia Management Review 16: 44–53.

Ocasio W. 1997. Towards an attention-based view of thefirm. Strategic Management Journal 18: 187–206.

Pfeffer J. 2007. A modest proposal: how we mightchange the process and product of managerial research.Academy of Management Journal 50: 1334–1345.

Podsakoff PM, MacKenzie SB, Podsakoff NP, Lee JY.2003. The mismeasure of man(agement) and its impli-cations for leadership research. Leadership Quarterly14: 615–656.

Powell TC. 2011. Neurostrategy. Strategic ManagementJournal 32: 1484–1499.

Rauch A, Wiklund J, Lumpkin GT, Frese M. 2009.Entrepreneurial orientation and business performance.Entrepreneurship: Theory and Practice 33: 761–787.

Rosenberg MJ, Hovland CI. 1960. Cognitive, affective,and behavioral components of attitudes. In AttitudeOrganization and Change, Hovland CI, Rosenberg MJ(eds). Yale University Press: New Haven, CT; 1–14.

Rosenbusch N, Brinckmann J, Bausch A. 2011. Is inno-vation always beneficial? A meta-analysis of the rela-tionship between innovation and performance in SMEs.Journal of Business Venturing 26: 441–457.

Rosenbusch N, Rauch A, Bausch A. 2013. The medi-ating role of entrepreneurial orientation in the taskenvironment-performance relationship. Journal ofManagement 39: 633–659.

Runyan RC, Ge B, Dong B, Swinney JL. 2011.Entrepreneurial orientation in cross-cultural research.Entrepreneurship: Theory and Practice 35: 1–18.

Schumpeter J. 1934. The Theory of Economic Develop-ment. Harvard University Press: Cambridge, MA.

Schumpeter J. 1942. Capitalism, Socialism, and Democ-racy. Harper & Row: New York.

Semadeni M, Withers MC, Certo ST. 2014. The perilsof endogeneity and instrumental variables in strat-egy research. Strategic Management Journal 35:1070–1079.

Shapira Z. 1986. Risk in managerial decision making.Working paper, Hebrew University: Jerusalem, Israel.

Spector PE. 1992. Summated Rating Scale Construction.Sage: Newbury Park, CA.

Stata Corporation. 2012. SEM Reference Manual. StataPress: College Station, TX.

Wales WJ, Patel PC, Lumpkin GT. 2013. In pursuitof greatness: CEO narcissism, EO, and firm perfor-mance variance. Journal of Management Studies 50:963–1154.

Wooldridge JM. 2010. Econometric Analysis of CrossSection and Panel Data (2nd edn). MIT Press: Cam-bridge, MA.

Wright P, Kroll M, Krug JA, Pettus M. 2007. Influencesof top management team incentives on firm risk taking.Strategic Management Journal 28: 81–89.

SUPPORTING INFORMATION

Additional supporting information may be foundin the online version of this article:

Appendix S1. A brief discussion of measurementtheory and construct development.Appendix S2. Measures used in empiricalillustration.


RECONCEPTUALIZING ENTREPRENEURIAL · PDF filereconceptualizing entrepreneurial orientation...

Documents

Transcript of RECONCEPTUALIZING ENTREPRENEURIAL · PDF filereconceptualizing entrepreneurial orientation...