the structure of competition: how competition between one ... · hsieh and vermeulen: the structure...

OrganizationScienceArticles in Advance, pp. 1–21ISSN 1047-7039 (print) � ISSN 1526-5455 (online) http://dx.doi.org/10.1287/orsc.2013.0832

© 2013 INFORMS

The Structure of Competition:How Competition Between One’s Rivals

Influences Imitative Market Entry

Kai-Yu HsiehSchool of Business, National University of Singapore, Singapore 119245, [email protected]

Freek VermeulenLondon Business School, London NW1 4SA, United Kingdom, [email protected]

This paper investigates how the pattern of encounters between a firm’s competitors affects the firm’s inclination tofollow its competitors into a new market. We theorize that direct encounters between a firm’s rivals lead to a herding

effect, making imitative market entry more likely. Past mutual forbearance between a firm’s competitors (resulting fromasymmetric multimarket competition) further strengthens this herding effect, by enhancing the firm’s expectations of marketattractiveness. In contrast, aggressive past rivalry between the competitors (resulting from symmetric multimarket contact)dampens these expectations, producing a competition effect that makes herding less probable. We test our idea in twodistinct contexts—the Chinese pharmaceutical industry and the Taiwanese computer hardware industry—and find consistentsupport in both settings. We discuss how our analysis of what we call the “structure of competition” can be extended toresearch on other forms of firm behavior.

Key words : structure of competition; imitation; market entry; multimarket competition; herding theoryHistory : Published online in Articles in Advance.

1. IntroductionUnder uncertainty, firms often seem to be inclinedto imitate each other’s actions (Lieberman and Asaba2006). Empirical studies have documented the influenceof imitation on a wide variety of strategic decisions,including diversification (Fligstein 1991, Haveman1993), organizational structure (Fligstein 1985), adop-tion of management practices (Abrahamson 1997, Burnsand Wholey 1993), innovations (Greve and Taylor 2000,Rogers 1995), and choice of location (Henisz and Delios2001), among others. Imitation is alleged to play a rolein influential societal phenomena such as acquisitionwaves, the Internet bubble, and the surge in usage ofcredit default swaps in the banking sector that led upto the 2008 financial crisis (e.g., McNamara et al. 2008,Pernell 2010).

Among the many reference organizations that firmscan potentially imitate, those who compete closely witha focal firm in the same industry segment have receivedparticular scholarly attention. Although organizationalscholars generally expected a firm to imitate its directcompetitors, empirical findings have been inconclusiveor even conflicting. Although in several studies firmswere found to imitate their close competitors (e.g.,Garcia-Pont and Nohria 2002, Guillen 2002, Yang andHyland 2006), other studies showed that competingorganizations’ actions had insignificant or mixed influ-ence (e.g., Greve 1996, Guillen 2003, Semadeni 2006),

and still others found that firms tended to steer awayfrom—rather than imitate—their direct competitors (e.g.,Baum et al. 2000, Delios et al. 2008, Han 1994). In anattempt to shed light on these conflicting findings, weargue and show that a firm’s propensity to imitate orinstead steer away from its direct competitors is deter-mined by the extent and nature of rivalry between itscompetitors.

We focus on a specific form of firm behavior—marketentry—and analyze how a firm’s decision to followsuit is moderated by the rivalry between its competi-tors. Using insights from herding theory (Banerjee 1992,Bikhchandani et al. 1992), which analyzes perceivedmarket attractiveness, we theorize that competitiveinteractions between a firm’s rivals create a herdingeffect, which makes mimetic entry more likely. Onthe other hand, rivalry between a firm’s competitorsmay also make entry into the same market segmentseem less attractive because of intensified competition.We assess the latter effect by examining the nature ofthe competition between the firm’s rivals—specifically,whether there is multimarket competition (Bernheimand Whinston 1990, Jayachandran et al. 1999) andwhether it is the type that reduces or intensifies rivalry(Gimeno 1999). Thus, we theorize that the contagiouseffect of competitors’ prior entry is determined by aset of moderators that delineate the structural pattern of

1

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.

Published online ahead of print June 26, 2013

Hsieh and Vermeulen: The Structure of Competition2 Organization Science, Articles in Advance, pp. 1–21, © 2013 INFORMS

competition around a focal firm—whether a firm’s directcompetitors also directly compete with each other and,if so, what the nature of their rivalry is.

We tested our ideas in two distinct contexts. The firststudy concerned the Chinese pharmaceutical industryand domestic drug producers’ entry into new productmarkets. The second study concerned the Taiwanese per-sonal computer (PC) industry and hardware producers’entry into new geographical markets in China. Of course,product-market entry and geographic-market entry aretwo different strategic actions, and they are associatedwith different underlying motives and partially differentstreams of literature. However, by examining these dif-ferent contexts, we intend to show evidence of the appli-cability of our ideas across different settings and formsof market entry. Our findings reveal that a firm’s ten-dency to imitate its direct competitors is strongly influ-enced by the wider pattern of market encounters betweenthese rivals.

Our paper makes two main contributions. First, byexamining a set of moderators, we advance the literatureon imitation by revealing circumstances under whichfirms have a propensity to imitate their direct rivals andthose under which they are distinctly disinclined to fol-low suit. Our second contribution is a more generalone. Our independent variables concern characteristicsof what we call the structure of competition: the patternof who competes with whom in an industry, particu-larly as regards characteristics of competitive encoun-ters between a firm’s rivals. Our results indicate thatfirms’ behavior—in our study, imitative market entry—is strongly influenced by such competitive structuresbecause they can bring about herding effects. This opensup the intriguing possibility that other types of firmbehavior might also depend on and be moderated bysuch structural features of the wider pattern of competi-tive interactions in an industry.

2. The Structure of CompetitionEach firm within a particular industry faces a differentcompetitive context. This is because an industry typi-cally consists of multiple product segments operated bydifferent but overlapping sets of industry members. Twoindustry members can be regarded as direct competitorsto each other when they operate in at least one com-mon product segment (Chen 1996, Peteraf and Bergen2003). Commonality in market segments is consequen-tial because it is often used by firm managers as a basisof competitor identification and business strategy for-mulation (Porac et al. 1995). Each firm in an industrymight directly compete with some industry members forcertain product segments but not with others. Similarly,among the direct competitors of a focal firm, some might

directly compete with each other, whereas others do not.For example, a pharmaceutical company operating in thecardiovascular and the dermatological segments mighthave one direct competitor in the former segment andanother in the latter, but these two direct competitorsmay or may not directly encounter each other in anycommon segment. Mapping all the competitive encoun-ters between a firm and its rivals can reveal remarkablydifferent structures.

Consider the sample firm depicted in Figure 1. It has22 direct competitors. The lines between them indi-cate whether these competitors also compete directlywith each other. The 22 rivals could potentially form231 competitor pairs in the market. However, becausenot all firms encounter each other, there are only 94connections between them. Moreover, 15 of these con-nections (as indicated by the bold lines) concern firmsthat meet each other in multiple segments of the indus-try, a situation of multimarket contact (Bernheim andWhinston 1990, Jayachandran et al. 1999). We theo-rize that the properties of such a competitive structureinfluence how likely it is that the focal firm followsentrants from this set of competitors into a new market.This is so, we argue, because these properties determineexpectations of market attractiveness, possibly triggeringherding effects.

In particular, in this paper, we compare four basicsituations: a firm whose direct competitors do not com-pete with each other; a firm whose competitors alsoencounter each other; and a firm whose competitors areengaged in multimarket competition, wherein this lat-ter case we will distinguish between two different formsof multimarket rivalry. We argue that in an uncertainsituation such as new market entry, the presence ofdirect encounters between a firm’s current rivals cre-ates a group effect—the proverbial herd (Banerjee 1992,Bikhchandani et al. 1992)—making imitative entry morelikely. This results in our first hypothesis. Yet it is thenature of this competition—whether it is multimarketcompetition and, if so, of what form (following Gimeno1999)—that may counter this effect. Some forms of mul-timarket competition may lead the firm to expect aggres-sive rivalry in the new market resulting from entry byits direct competitors (i.e., so-called symmetric multi-market contact), making the market seem less attractiveand thus suppressing herding effects. In other cases, therivalry may be more subdued (i.e., so-called asymmetricmultimarket competition), augmenting the herding effectand, with it, a firm’s propensity to enter. This results inour second and third predictions. Hence, shaped by theextent and nature of the rivalry between its present com-petitors, a firm might be inclined to mimetically enter anew market or stay out of it altogether.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.

Hsieh and Vermeulen: The Structure of CompetitionOrganization Science, Articles in Advance, pp. 1–21, © 2013 INFORMS 3

Figure 1 Example of a Firm and the Market Encounters Between Its Direct Competitors

FOCALFIRM

Focal firm

Competitors of the focal firm

Direct encounter

Multimarket contact

3. Theory Development and Hypotheses3.1. Competition Between a Firm’s CompetitorsPrior market entry by a firm’s competitors may make afirm more inclined to enter that market too because of aninclination to follow “the herd.” Herding theory suggeststhat, under uncertainty (Avery and Zemsky 1998), firmsinterpret prior entry as a signal of market attractivenessin terms of the market’s potential size, consumers’ will-ingness to pay, and so on. These earlier entrants arethought to act based on information unavailable to thefirm (Bikhchandani et al. 1992).1 Especially when anincreasing number of its direct competitors appears toconclude that entry is favorable, the firm is likely to actbased on this information and enter too (Banerjee 1992,Bikhchandani et al. 1998). However, prior entry by afirm’s competitors can also reduce the firm’s expecta-tions of market attractiveness (and with it, its inclinationto enter) because of a competitive crowding effect. Thiseffect could be especially pertinent when it concerns thefirm’s direct competitors.

We expect that the positive herding effect will beparticularly strong when many of a firm’s direct rivalsalso encounter each other, because in this situation, itscompetitors start to act as a reference group (Greve1998). Encounters between rivals prompt the firm to

view them as forming a group of closely related orga-nizations, of which the focal firm is a part (Fiegenbaumand Thomas 1995). Organizations usually closely moni-tor their reference groups, which increases the probabil-ity that they will act in similar ways because, as Greve(1998) put it, “What you see is what you do.” Whenthe group of rivals starts acting as a herd and the firmmonitors them, it is likely to be drawn into followingtheir actions. According to herding theory, this effect isfurther strengthened by outsiders’ assessments of theirmovements. That is because firms that are a part of thisgroup of competitors but that do not follow suit—that is,they do not enter the new market when others do—willexpose themselves to reputational risk (Zwiebel 1995).In their seminal paper on herding theory, Scharfstein andStein (1990) explained how rational managers, eager toprotect their reputation as sound decision makers, fol-low the herd in the form of mimicking the decisions ofothers even in light of private information that suggeststhat a market is unattractive. Although the firm mightbelieve that the market is unlikely to be profitable, man-agers cannot take the risk of not entering because theyare “afraid of being perceived as lone fools for missingout on the ride” (Scharfstein and Stein 1990, p. 465); incontrast, the personal risk of entering a market that is

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Figure 2 Direct Encounters Between a Firm’s Competitors

Firm BFirm A

Competitor3

Competitor3

Competitor1

Competitor1

Competitor2

Competitor2

likely to be unprofitable while others also enter is com-paratively low “because an unprofitable decision is notas bad for reputation when others make the same mis-take” (p. 466). Hence, building on these insights fromherding theory, we predict that encounters between afirm’s direct competitors are likely to create a herdingeffect, thus drawing the firm into entering into the newmarket, too.

Consider firms A and B in Figure 2. In this styl-ized example, both firms have three direct competitors.However, firm A’s competitors do not compete directlywith one another. This is because firm A encounterscompetitor 1 in one segment of the industry (e.g., cardio-vascular drugs), competitor 2 in another segment (e.g.,dermatological drugs), and competitor 3 in yet anotherone; these firms have no other segment in common.In contrast, the four firms on the right-hand side all meeteach other in some segment of the industry. The latterfirms (firm B and its competitors) are more likely tooperate as a herd because they all monitor each other.Therefore, we expect that firm B is more likely thanfirm A to follow its competitors into a new market.That is, we predict that direct encounters between anorganization’s existing competitors will positively mod-erate the relationship between prior entry by competitorsand a firm’s propensity to follow suit. Hence, a firmwill be even more inclined to match rivals’ market entrymoves if many of its current competitors also competedirectly with each other. Formally stated,

Hypothesis 1. The extent to which a firm’s directcompetitors also compete directly with each other willincrease the relationship between the number of directcompetitors that have entered a particular market andthe firm’s likelihood of entering the same market.

3.2. Asymmetric Multimarket ContactBetween a Firm’s Competitors

Prior entry may create a herding effect because itincreases a firm’s expectations of market attractiveness(Banerjee 1992, Bikhchandani et al. 1992). However,as stated previously, there could also potentially beincreased competition resulting from prior entry by thefirm’s direct competitors, which could lower the firm’smarket expectations. To what extent this competitioneffect occurs will depend on the nature of the rivalry

between the firm’s competitors. If the entrants stem froma group of direct competitors among which rivalry isfierce, the firm might expect this rivalry to spill over intothe new market, lowering market expectations. On theother hand, if existing competition is subdued, the firmmight expect entering firms to behave similarly towardeach other in the new context as well. One thing influ-encing the nature of the rivalry is whether there existsmultimarket contact between the competitors and whatform it takes.

Multimarket or multipoint competition (Bernheim andWhinston 1990, Jayachandran et al. 1999, Karnani andWernerfelt 1985) occurs when a firm’s competitorsencounter each other in multiple segments of the indus-try. Traditionally, scholars of multimarket contact haveargued that when two competitors have small footholdsin each other’s key market, an implicit threat of recip-rocal retaliation results. Competition thus becomes moresubdued, appearing in the form of nonaggressive behav-ior toward each other, a situation that has been referredto as mutual forbearance. For example, Gimeno (1999)showed that airlines use a relatively small presence intheir competitors’ main hubs to reduce rivalry in theirown hub. Thus, this form of multimarket contact servesas a deterrent, which limits the intensity of rivalry. It hasbeen associated with higher profits (Feinberg 1985, Scott1982), more favorable price setting (Evans and Kessides1994, Gimeno 1999), and lower rates of market exit(Barnett 1993, Baum and Korn 1996, Boeker et al.1997). Following Gimeno (1999), we refer to this situa-tion in which competitors have a small foothold in eachother’s key market as “asymmetric multimarket competi-tion,” in contrast to symmetric multimarket competition,according to which both players are largely dependenton the same segment. It is in the situation of asymmetricmultimarket contact that mutual forbearance is likely.

Hence, we predict that the market expectations ofprior entry are positively enhanced by asymmetricmultimarket competition between a firm’s competitorsbecause it makes the firm optimistic about the fiercenessof rivalry in the new market. Hence, it strengthens theherding effect. Owing to mutual forbearance, the rivalrybetween the firm’s competitors will have been charac-terized by nonaggressive behavior, which is likely tospill over into the new market (Baum and Korn 1999).Competitors that behaved in a nonaggressive way towardeach other in the past, as a result of asymmetric mul-timarket competition, are expected to also behave in anonaggressive way in the new market. This applies toprior entrants as well as potential future entrants fromthis group. Thus, because of the asymmetric multimarketcontact between its current rivals, the firm’s expectationsof rivalry in the new market will be relatively optimistic.

The firm will benefit from the mutual forbearancebetween its competitors, for example, because theirnonaggressive price setting will also be advantageous to

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


the firm. The rivals will abstain from initiating fierceprice competition so as to avoid retaliation, a state ofaffairs from which the focal firm will also profit (Barnett1993). Indeed, Haveman and Nonnemaker (2000), exam-ining the California savings and loan industry, found thatsingle-market competitors benefited from the mutual for-bearance created by multimarket competition betweentheir rivals. Therefore, a firm’s expectations of marketattractiveness will be more positive when the entrantscome from a group of rivals who are engaged in the typeof multimarket competition that implies a small footholdin each other’s key market. Having observed a history offorbearing behavior between them, the firm can expectthat rivalry in the new market will also be restrained,which strengthens the herding effect by making entrymore attractive.

Consider firms B and C in Figure 3, in whichfirm B’s competitors directly encounter one another,as do firm C’s competitors. Yet whereas firm C’scompetitors are engaged in multimarket competition,firm B’s competitors are not. Two of firm C’s competi-tors encounter one another in segments W and X. Com-petitor 1 depends 90% on segment W (e.g., because itgenerates 90% of its revenues there) and 10% on mar-ket X. The interests of competitor 2 in the two marketsare almost the reverse: it depends 12% on segment Wand 88% on segment X. Mutual forbearance is likelyto occur in such a situation because competitor 1 willlikely respect competitor 2’s interest in X in exchange forcompetitor 2’s restraint in W (Gimeno 1999). We expectfirm C to be more prone to follow the herd and imitateits competitors because these firms’ forbearing behaviortoward each other can be expected to spill over into thenew market. Therefore, we predict that this type of multi-market competition between a firm’s present competitorswill positively moderate the relationship between priorentry by competitors and the firm’s propensity to followsuit such that the firm will be even more inclined to matchits rivals’ entry moves if these rivals encounter each otherin multiple segments.

Hypothesis 2. Asymmetric multimarket competitionamong a firm’s direct competitors will furtherincrease the relationship between the number of direct

Figure 3 Asymmetric Multimarket Contacts Between aFirm’s Competitors

Firm CFirm B

Competitor3

Competitor1

Competitor2

Competitor2

Competitor1

Competitor3

90%

10%

12%

88%

MarketW

MarketX

Note. Bold figures highlight the market of major interest to eachcompetitor.

competitors that have entered a particular market andthe firm’s likelihood of entering the same market.

3.3. Symmetric Multimarket ContactBetween a Firm’s Competitors

As discussed previously, multimarket contact betweencompetitors can potentially lead to mutual forbearance.However, this does not always happen (for an overviewof empirical findings, see Baum and Korn 1999). Mutualforbearance is unlikely to occur when two multimarketcompetitors depend heavily on the same segment. Forexample, if two airlines share the same main hub, andall other routes are of marginal importance to both ofthem, competition is likely to be fierce (see Gimeno1999). Take Figure 4 as an example. Consider the caseof firm D. As in the case of firm C, competitor 1depends 90% on market Y and 10% on market Z.Here, however, competitor 2’s relative interests in thetwo markets are very similar: it depends 88% on mar-ket Y and 12% on market Z, a situation of highly sym-metric market dependence. In this situation, aggressivecompetition—not mutual forbearance—is more likelyto occur. Because both markets are of equal strate-gic importance to both firms, they will often competeintensely against one another. Each firm seeks to preventthe other from gaining the upper hand in either of thetwo markets (Knickerbocker 1973) to avoid cross-marketsubsidization (Chen and MacMillan 1992, Gimeno 1999,McGrath et al. 1998).

We contend that the herding effect of prior entry bycompetitors is negatively affected by such symmetricmultimarket contact between a firm’s competitors. Hav-ing observed the aggressive behavior of its competi-tors toward each other in the past, a firm might expectthat entrants from such a group will compete fiercelywith each other in the new market too, as will possiblefuture entrants from that group. Their aggressive behav-ior toward each other will likely incur negative spilloversbecause, for example, intense price competition betweenthem will put pressure on the profit margins of everyfirm in the market. Rather than being inclined to mimicthese rivals’ entry moves, the firm might choose not tojoin its competitors in the new market, in an attempt

Figure 4 Asymmetric vs. Symmetric Multimarket ContactsBetween a Firm’s Competitors

Firm DFirm C

Competitor3

Competitor1

Competitor1

Competitor2

Competitor2

Competitor3

90% 90%

10% 10%

12%

12%88%

88%MarketW

MarketX

MarketZ

MarketY

Note. Bold figures highlight the market of major interest to eachcompetitor.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


to steer clear of the rivalry. Therefore, we expect thatfirm D will be less likely to follow its competitors intoa new market. Put differently, we predict that symmetricmultimarket contact will negatively moderate the rela-tionship between prior entry by competitors and a firm’spropensity to follow suit, so a firm will not be inclinedto imitate its competitors’ entry decisions.

Hypothesis 3. Symmetric multimarket competitionamong a firm’s direct competitors will decrease the rela-tionship between the number of direct competitors thathave entered a particular market and the firm’s likeli-hood of entering the same market.

4. Methods4.1. Study 1: The Chinese Pharmaceutical Industry

Research Setting. Our first database consisted of alldomestic producers of active pharmaceutical ingredientsin China, observed during the period 1992–2001. Beforethe start of economic reform in China in the 1980s,drug producers operated solely in accordance with theproduction plans set by the government’s public healthagency, whose primary concern was ensuring the supplyof basic pharmaceuticals rather than efficiency, profit, orinnovation. Consequently, the industry became charac-terized by bulk capacity of generic drugs and a focus onoutput volume (White and Liu 1998). Firms were pri-marily engaged with meeting the output targets that hadbeen set centrally for them.

In the late 1980s and early 1990s, a series of economicreforms moved China toward a market economy. Fol-lowing these developments, China’s national healthcaresystem underwent various significant changes. Amongthese, the old cooperative medical system was dis-solved; price regulations were abolished and replacedby a system distinguishing between “ethical” and“over-the-counter” drugs; medical consultation and drugdispensing were separated; and a partial national insur-ance scheme, including an expanding list of approveddrugs and mandatory price caps, was introduced. Fur-thermore, various policy measures and certification pro-grams aimed at improving the efficiency and qualityof drug production and distribution, similar to those indeveloped countries, were introduced by the Ministryof Health. These developments presented managers inthe pharmaceutical industry with various challenges anduncertainties. Managers had to move beyond their tradi-tional role of overseeing production and assume respon-sibility for strategic decision making, which previouslyhad been handled by governmental agencies (Guthrie1997, White 2000). It is for this latter period that wetested our predictions regarding imitative market entry.

Data. We compiled our longitudinal data for theperiod 1992–2001 using the Pharmaceutical IndustryYearbook, an industry directory published annually by the

Chinese National Bureau of Statistics. We chose 1992 asa starting date because the types of data reported in theyearbooks before 1992 were inconsistent with those forthe years thereafter. Moreover, before this date, becauseof the central planning economy, individual firms hadrelatively little control over which products they pro-duced. The Pharmaceutical Industry Yearbook combinesrecords of various public healthcare agencies and cov-ers every drug manufacturer in China, providing detailedproduction data on all drugs produced by each manufac-turer in a given year. Because the pharmaceutical industryis highly regulated and monitored, these official recordsare comprehensive. Using this information source, wewere able to identify the whole population of drug manu-facturers in China as well as every product line expansionevent during the observation period.

Our database consisted of the entire population ofdomestic drug manufacturers in China, as listed in theyearbooks. During the observation period, 1,634 firmshad their own facilities for manufacturing active pharma-ceutical ingredients. Industry experts we consulted forthis project suggested that a relatively large number ofthese firms were likely to be short-lived, highly special-ized companies that focused on producing one specificingredient. Since we feared this might distort our statis-tical results, in the analysis presented below we includedonly the 742 firms that had ever produced at least twodistinct active ingredients during the observation period.However, we repeated the analyses on the data coveringall the firms and found consistent results.

In total, these drug manufacturers produced more thana thousand active ingredients. The Chinese State Foodand Drug Administration classified all active ingredi-ents into 22 therapeutic categories (see Table A.1 inthe appendix for details) similar to those used in othercountries. Ingredients that act on the same organ or sys-tem and that have similar therapeutic characteristics aregrouped into the same category. Producing ingredientswith different therapeutic effects requires drug produc-ers to acquire and exercise different technical know-howand to deal with different sets of distributors and physi-cians. Therefore, each therapeutic category delineates adistinct product market within the pharmaceutical indus-try. We used these categories to identify direct competi-tive relationships between firms and market entry events.Two drug producers were regarded as directly competingwith each other if they manufactured active ingredientsin at least one common category. A market entry eventwas defined as a firm starting to produce an ingredientin a therapeutic category that it had not served before.During the period 1992–2001, 331 of the 742 companiesmade 542 entries into 22 different product markets.

4.2. Study 2: The Taiwanese ComputerHardware Industry

Research Setting. Our second database consisted ofmanufacturers of PCs and ancillary hardware in Taiwan

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


and their direct investments into China during the period1999–2005. During this period, Taiwan had—and stillhas—a very substantial computer industry. For example,in 1998, Taiwanese producers registered a productionvalue of US$34 billion (Institute for Information Indus-try 2000). This made Taiwan the third-largest hardwaresupplier in the world (after the United States and Japan).Unlike Japan and Korea—whose computer industrieswere dominated by large, diversified conglomerates—Taiwan’s computer industry consisted of many medium-sized producers, located along a 50-mile stretch southof the capital, Taipei. These medium-sized companiesspecialized in rapidly adapting their production to tech-nological innovations by quickly moving into emergingproduct technologies and jumping onto the next technol-ogy available as soon as margins eroded (Dedrick andKraemer 1998).

Yet the business environment became increasinglychallenging in the late 1990s. First, the global PC mar-ket experienced dramatic price erosion. In the UnitedStates between 1997 and 1998, for example, the averageretail price of a PC dropped from $1,800 to about $1,000(Curry and Kenny 1999). In addition, soon after, mostPCs became so powerful that few applications required afaster machine, and corporations and consumers slowedtheir replacement cycles, which led to a drop in demand.As a consequence, for the first time since 1981, in 2000–2001 the market demand for PCs declined. During thesame period, while Taiwanese PC hardware producerswere struggling to adapt to the drop in demand, theChinese economy, which had been relatively shelteredfrom the 1997 Asian financial crisis, was experiencingsubstantial development as a result of its shift towarda market economy. For Taiwanese companies, Chinanot only represented an ideal site for setting up low-cost production facilities but also became an importantend market in itself. Economic wealth increased rapidlyin various regions, and local demand for PCs surged.Consequently, although relatively few companies had amajor presence in China until the mid-1990s, by the endof 2005, investments by our sample companies in Chinaaccounted for more than 60% of their total investmentsoverseas.

Data. The data used in this study concerned all com-panies listed in Taiwan that had ever produced PCs orancillary hardware using in-house manufacturing facili-ties during the period 1999–2005. Companies were iden-tified from two bourses in Taiwan: the Taiwan StockExchange (where larger and more-established compa-nies are listed) and the Gre Tai Securities Market (whichlists smaller, more-entrepreneurial companies). Regula-tions in Taiwan require listed companies to disclose inannual reports their sales of products accounting for 10%or more of their total revenue. Using Standard IndustryClassification (SIC) categories to define product mar-kets, we identified all firms that were reported to be

active in the following five product markets: computers,monitors and terminals, computer peripherals, audio andvideo equipment, and communication equipment. Thisyielded 205 companies, for which we proceeded to iden-tify all direct competitors. We regarded two companiesto be directly competing with each other if they oper-ated in at least one common product market. Since manyof the 205 companies had diversified into other relatedmarkets (e.g., optoelectronics components), we also col-lected information about their direct competitors in thesesegments. This in turn yielded another 344 companies inan additional 15 product markets (see Table A.1 in theappendix for details).

For these 549 companies, we traced all investmentsinto China back to 1982—before which Taiwaneseinvestments into China were virtually nonexistent—using two complementary sources. First, the InvestmentCommission (the regulatory agency of foreign invest-ment in Taiwan) reviewed and kept records of everyinvestment project greater than US$200,000. In addi-tion, Taiwan’s listed companies self-reported their activ-ities in China to the Market Observation Post System,an open information platform managed by the TaiwanStock Exchange. Together, these two sources providedcomprehensive coverage of all investments into Chinaby Taiwanese companies. These data were updated quar-terly. During the period 1999–2005, 389 of the 549 com-panies made 650 entries into 22 different geographicmarkets in China. A geographic market was defined asa province, a municipality, or an autonomous regionin China. This is the standard way of distinguishingbetween various regions within China (Chan et al. 2010)because each is overseen by a separate local govern-ment. A market entry move was defined as the first directinvestment made by a Taiwanese company in a particulargeographic market.

4.3. Dependent Variables and Method of AnalysisOur hypotheses pertained to the likelihood that a firmwill enter a specific new market, as influenced by itsdirect competitors’ prior entries. We estimated this entrylikelihood using Cox’s (1975) semiparametric propor-tional hazard model. The dependent variable in Coxmodels is the hazard rate, or the instantaneous probabil-ity of a firm entering a particular market by a specificpoint in time. A firm was at risk of entering any marketit had not operated in before. Letting h6t � Z4t57 be thehazard rate of entering a market at time t for a firm withcovariate vector Z4t5, our Cox models are as follows:

h6t �Z4t57= h0i4t5 exp6Â′Z4t570

In this equation, the hazard rate is represented as covari-ates modifying multiplicatively an unspecified baselinehazard function. Because both of our samples containedmultiple observations of each firm with regard to its

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


entry decisions in different markets at a given point intime, we were able to stratify our models by fitting anindividualistic baseline hazard function h0i4t5 for eachfirm (denoted by i).

Accordingly, estimates of the parameter vector Â wereobtained by maximizing the log-likelihood function:

logL4Â5=∑

i

logLi4Â51

where Li4Â5 is the partial likelihood function using onlythe data for those observations associated with firm i.In a few instances, a firm might have entered morethan one market at the same time, resulting in tiesbetween event times. We implemented Breslow’s estima-tor (1974) to account for ties. Specifically, we let di4t5equal the number of entry moves that firm i undertook attime t, let Di4t5 be the set of markets that firm i enteredat time t (each market is denoted by m), and let Ri4t5 bethe set of markets that firm i was at risk of entering at atime just prior to t (i.e., risk set). The partial likelihoodfunction corresponding to each firm can be expressed asfollows:

Li4Â5=∏

t

exp6Â′si4t576∑

m∈Ri4t5exp6Â′Zi4t577

di4t51

where si4t5=∑

m∈Di4t5

Zi4t50

Since the likelihood function is not dependent onthe baseline hazard function, we could make inferenceson the effects of covariates without specifying h0i4t5.As such, stratifying the baseline hazard function consti-tutes a convenient and highly effective way to controlfor unobserved firm-level effects. Indeed, since Li4Â5 isbased on contrasting observations in Dt4t5 (the set ofmarkets that firm i entered at time t) with observationsin Ri4t5 (the set of markets that firm i was at risk ofentering at a time just prior to t), the effects of any vari-ables that vary only at the firm level—including bothfixed and time-variant ones (e.g., firm size)—would becanceled out. Another important advantage of these Coxmodels is that left-truncated data can be easily accountedfor. When left-truncated data are present, the risk setRi4t5 is defined as observations of firm i that were understudy at a time just prior to t. With this simple modi-fication, the usual techniques for estimating regressioncoefficients in Cox models can be applied (see Klein andMoeschberger 2003).2

In Study 1 (the Chinese pharmaceutical industry),observations were determined by firms, years, and thera-peutic categories. A spell, or time-to-event, started eitherat the beginning of the period of observation (1992)—in which case the observation was left-truncated—or atthe point of founding of a specific firm (if after 1992).A spell ended in an event when the firm entered a partic-ular market or was right-censored either at the end of the

observation period (2001) or because the firm went outof business before that date. In Study 2 (the TaiwanesePC industry), observations were determined by firms,quarters, and geographic regions in China. A spell endedin an event if the firm entered the particular market; oth-erwise, the observation was right-censored, either in thequarter in which the firm was dissolved or because ithad not entered a market by the end of 2001. Our datacontained 66,540 firm-year-market observations associ-ated with 12,988 spells in Study 1 and 88,678 firm-quarter-market observations associated with 3,703 spellsin Study 2.

4.4. Explanatory VariablesCompetitors’ prior entry was measured as the number ofa firm’s competitors that had previously entered a partic-ular market. As a robustness check, we also computed analternative measure weighed by the size of each competi-tor (using the logarithm of annual sales) because priorresearch has indicated that firms might be more inclinedto imitate large organizations (Greve 2000, Haunschildand Miner 1997, Haveman 1993). This weighted mea-sure yielded highly consistent results. In the tables, wereport the results based on the count measure becauseits estimated coefficients are relatively easy to interpretin terms of their implied effects.

Direct encounters between a firm’s competitors indi-cated how often a firm’s direct competitors also previ-ously encountered one another in at least one commonmarket segment. In Study 1, a market segment wasdefined as one of the 22 therapeutic categories; inStudy 2, a market segment was identified as one of the 20SIC categories (as listed in Table A.1 in the appendix).Thus, we counted the number of occasions that com-petitors of the focal firm also encountered each otherand divided that by the maximum number of times theycould have encountered each other. For example, in Fig-ure 2, this measure takes on the value of 0 (no encountersbetween competitors) divided by 3 (number of times theycould have encountered each other) for firm A and wouldbe 3 divided by 3 for firm B. Thus, this variable indicatesthe extent to which a firm’s direct competitors also com-pete with each other.3 Its value ranges from 0 to 1, wherea higher number indicates more direct market encountersbetween a firm’s competitors.

We captured asymmetric (versus symmetric) mul-timarket competition through a combination of twovariables. The first variable indicates the extent towhich there is multimarket contact between a firm’scompetitors, regardless of whether it is symmetric orasymmetric, which is the traditional way to measuremultimarket competition. Specifically, we measured howmany of the direct encounters between a firm’s com-petitors consisted of multimarket contacts. Thus, again,this variable represents a proportion whose value rangesfrom 0 (no multimarket encounters between competi-tors) to 1 (all the competitors that directly encounter

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


each other are engaged in multimarket competition). Forexample, in Figure 3, it takes on the value 0 (no competi-tors meet in multiple markets) divided by 3 (the numberof times multimarket competition could have happened)for firm B, whereas it takes on the value 3 divided by 3for firm C.

The second variable (defined below) captured theextent of symmetry in these multimarket contacts.Hence, in our analysis, when we include (and con-trol for) both measures, the variable defined above willmodel the effect of asymmetric multimarket contact,whereas the symmetry variable will capture the influ-ence of symmetric multimarket competition. Alterna-tively, we also developed a set of dichotomous variablesthat directly distinguished between asymmetric and sym-metric multimarket contact. These variables and models,which we will discuss in the Results section, producedresults virtually identical to the ones above.

Market dependence is most often measured as the pro-portion of a firm’s sales generated in a particular market(Bothner 2003, Burt 1992, Chen and MacMillan 1992),which is what we used in Study 2. In Study 1, however,data on firms’ sales per segment were not always avail-able. Therefore, we assessed the level of market depen-dence as the proportion of a firm’s total drug portfoliothat it offered in the particular market. Subsequently,using these numbers for both studies, we computed theEuclidean distance between each pair of competitorsto indicate the extent of symmetric market dependencebetween them. A high distance score indicates two firmsthat have very dissimilar market dependence, whereasa low distance score between two companies impliesthat they have highly symmetric market dependence. Forexample, in Figure 4, firm C’s competitors are located atthe coordinates 0.90, 0.10 (firm 1) and 0.12, 0.88 (firm 2),which leads to a Euclidean distance of 1.103, indicat-ing asymmetric market dependence. Firm D’s competi-tors are located at the coordinates 0.90, 0.10 (firm 1) and0.88, 0.12 (firm 2), which leads to a Euclidean distance of0.028, indicating highly symmetric market dependence.Formally stated,

dk1 l4t5=

√

∑

m

[

sk1m4t5

sk4t5−

sl1m4t5

sl4t5

]2

0

In the above equation, sk1m denotes the number ofactive pharmaceutical ingredients (Study 1) or total sales(Study 2) that competitor k generated in market m, andsk denotes the total number of ingredients or sales ofcompetitor k. We reverse-coded dk1 l (by subtracting dk1 l

from a constant equal to the maximum distance score inthe sample) to indicate the symmetry score between kand l, denoted by qk1 l. The firm-level variable symmetriccontact between a firm’s competitors was then computedas the average of qk1 l across competitors’ multimarketcontacts, so a high score indicates a set of multimarketcompetitors that are highly symmetric in terms of theirmarket dependence.

4.5. Control Variables

Study 1. Although we focused on the market rela-tionships between a firm’s direct competitors, we alsocontrolled for the market relationships between the firmitself and its competitors through the inclusion of twointeraction terms, a firm’s multimarket contact with itscompetitors× competitors’ prior entry and a firm’s sym-metric contact with its competitors × competitors’ priorentry. Including these variables should enable a directcomparison between the effect sizes of a firm’s encoun-ters with its direct competitors and the effect sizes ofencounters between its competitors.

Moreover, companies might sometimes enter the samemarket not because they imitate one another but becausethe target market has certain characteristics that makeit highly attractive. We accounted for market attractive-ness using a number of control variables. For one, weaccounted for density-dependent legitimation and com-petition dynamics (Hannan and Freeman 1989) by con-trolling for market density, measured as the total numberof firms that were active in a market at a time, andmarket density2 (divided by 100 for rescaling). Priorliterature suggests that the market entry rate wouldincrease with density but decrease with density squared(Carroll and Hannan 1989, Greve 2000, Haveman 1993).

In all our models, we included the natural logarithmof total market output to account for market size. Wealso controlled for market exits, measured as the num-ber of companies that left a particular product market,and incumbents operating at a loss, measured propor-tionally as the number of incumbents operating at aloss divided by the total number of incumbents in aproduct market. We expected the market entry rate todecrease with these two variables because they mightsignal a decline in market attractiveness (Chan et al.2006). Furthermore, literature in industrial economicshas suggested that excess capacity and high market con-centration can deter entry (Scherer and Ross 1990). Toconstruct a measure for excess capacity in a market, wecumulated the firms’ production capacity, subtracted themarket’s total production volume, and then divided thisnumber by the total production volume to arrive at ameasure of relative excess capacity. We expect accesscapacity to deter entry because it makes the market morecompetitive and less attractive. We measured the marketconcentration ratio as the proportion of market outputby the four largest manufacturers in a market (the CR4index). Whereas in some settings, market concentrationcan stimulate entry—especially by small specialists whooffer differentiated products that satisfy the needs of par-ticular niche customers better than large generalists can(Carroll 1985)—in our setting we expected the entry rateto decrease with market concentration. This is becausethe possibility for product differentiation is very lim-ited (the vast majority of Chinese producers focused on

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


generic drugs), and the economies of scale enjoyed bylarge incumbents created entry barriers.

Study 2. As in Study 1, in Study 2 we controlledfor competitors’ prior entry and included the interactionterms a firm’s multimarket contact with its competitors×

competitors’ prior entry and a firm’s symmetric contactwith its competitors× competitors’ prior entry to isolatethe effects of the market relationships between a firmand its competitors. Market density and market density2

were controlled for using the total number of foreignand domestic firms involved in the computer hardwareindustry in each of the geographic markets in China.The information on market density was obtained fromthe Bureau of Statistics of China. As before, we alsoincluded market exits as a control.

To reflect other factors that might affect the attrac-tiveness of a geographic market, we constructed fouradditional control variables: the level of international-ization of the particular geographic market (captured by

Table 1 Summary Statistics and Correlations

Variable Mean SD 1 2 3 4 5 6 7 8 9 10 11

Study 1: Chinese pharmaceutical industry1. Direct encounters between a

firm’s competitors0085 0015

2. Multimarket contact between afirm’s competitors

0031 0015 0009

3. Symmetric contact between afirm’s competitors

0054 0002 0019 0019

4. A firm’s multimarket contact withits competitors

0014 0015 −0076 −0010 −0006

5. A firm’s symmetric contact withits competitors

0072 0027 0082 0000 0016 −0065

6. Competitors’ prior entry 18076 18062 −0019 −0016 −0003 0018 −00207. Market density 0056 0052 0006 0010 0007 −0005 0004 00638. Market exits 2071 2055 0006 0005 0005 −0005 0004 0051 00679. Incumbents operating at a loss 0023 0013 0001 −0002 −0005 −0003 0002 0005 0020 0018

10. Excess capacity 0002 0006 −0002 −0001 −0003 0000 −0001 −0016 −0019 −0013 −002311. Market concentration ratio (CR4) 0064 0023 −0006 −0005 −0006 0005 −0004 −0059 −0076 −0062 −0017 002812. Market size 7076 2076 0003 0001 −0002 −0004 0002 0036 0057 0043 0031 −0046 −0070

Study 2: Taiwanese PC industry1. Direct encounters between a

firm’s competitors0086 0019

2. Multimarket contact between afirm’s competitors

0004 0003 0016

3. Symmetric contact between afirm’s competitors

0043 0009 −0001 −0016

4. A firm’s multimarket contact withits competitors

0001 0002 −0076 −0018 −0007

5. A firm’s symmetric contact withits competitors

0073 0023 0078 0017 0008 −0078

6. Competitors’ prior entry 2028 6094 −0011 −0010 −0005 0013 −00147. Market density 5027 5023 0002 −0002 0003 −0002 0003 00558. Market exits 0050 0053 −0002 0001 0000 0002 −0002 −0007 −00249. Internationalization −0009 0094 −0001 0004 −0004 0001 −0001 0044 0085 −0026

10. Wealth per capita 0009 0095 0006 −0011 0011 −0008 0008 0020 0018 −0020 000011. Skilled labor 0008 1000 0005 −0010 0010 −0007 0006 0011 0027 −0007 −0004 −000912. Transportation infrastructure 0000 0099 −0001 0002 −0002 0001 −0001 0006 −0010 −0017 0003 −0003 0001

imports, exports, and total foreign capital), wealth percapita (captured by the average gross domestic prod-uct, disposable income, and household expenditure percapita), the availability of skilled labor (captured bythe proportion of the population with college degrees,the number of recent college graduates, and the numberof professionals in the region), and the market’s trans-portation infrastructure (captured by highway densityand railway density). These data were obtained from thevarious yearly issues of the China Statistical Yearbook.Table A.2 in the appendix shows all these indicators andtheir discriminant validity using exploratory factor anal-ysis. These measures, representing favorable local condi-tions, were expected to be positively associated with themarket entry rate. Table 1 reports descriptive statisticsof the variables used in the two studies.

5. ResultsThe Cox models testing our hypotheses are presentedin Tables 2 and 3. In both tables, Model 1 includes

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


only the control variables. Table 2 shows the resultsfor the Chinese pharmaceutical industry. In line withprior research (Hannan and Freeman 1989, Nickel andFuentes 2004), the market entry rate increases with den-sity but decreases with density squared, indicating acompetitive crowding effect when many firms come tobe active in the market. Table 3 displays the resultsfor the Taiwanese PC hardware industry. Here, too, themarket entry rate initially increases with density butdecreases at higher levels of density. In both sets of mod-els, the effect of prior entry by the firm’s direct competi-tors is positive and significant, which suggests that, onaverage, firms are inclined to imitate the market entrymoves by their direct competitors. Our hypotheses testsbelow concern moderators of this relationship.

5.1. Hypotheses TestsHypothesis 1 predicted that the propensity of a firm toimitate its direct competitors will be greater if its directcompetitors also directly compete with one another. Thishypothesis is tested through the interaction of com-petitors’ direct encounters and competitors’ prior entrybecause the former is expected to positively moderatethe influence of the latter on the market entry rate. Cor-roborating Hypothesis 1, the effect of the interactionterm is positive and significant in both studies. Follow-ing Cleves et al. (2010), we assessed the magnitudeof the interaction effects by taking the partial deriva-tives of the relative hazard with respect to the variablecompetitors’ prior entry under different levels of directencounters between a firm’s competitors. In Study 1,when direct encounters between a firm’s competitorsare prevalent (one standard deviation above the sam-ple mean), an additional competitor’s entry will increasethe relative hazard of entry by a factor of 1.04. Yetwhen direct encounters between a firm’s competitors arerare (one standard deviation below the sample mean),an additional entry by a competitor has only a marginalimpact on the relative entry hazard (multiplier ≈ 1.00).In Study 2, these multipliers are 1.08 and 1.02, respec-tively. These results are in line with the idea that themimetic effect of prior entry is stronger in cases wherea firm’s rivals also encounter each other.

Hypothesis 2 predicted that a firm’s propensity toimitate its direct competitors’ market entry moves willbe even stronger if these firms have small footholdsin each other’s markets: the classic case of multimar-ket competition. This prediction rests on the idea thatover and beyond the positive herding effect due toencounters between a firm’s rivals (tested in Hypothe-sis 1), such asymmetric multimarket contact between afirm’s competitors can lead to even more positive marketexpectations. In line with these ideas, in both studies, thecoefficient of the interaction term of (asymmetric) mul-timarket contact between a firm’s competitors and com-petitors’ prior entry is positive and significant, whichcorroborates this hypothesis.

In terms of the size of this effect, in Study 1,when multimarket contact between a firm’s competitorsis prevalent (one standard deviation above the samplemean), an additional competitor entering the market willincrease the relative hazard of entry by the firm by afactor of 1.03. Yet when multimarket contact betweena firm’s competitors is rare (one standard deviationbelow the sample mean), an additional competitor enter-ing has a negligible impact on the firm’s entry rate(multiplier ≈ 1.00). In Study 2, these multipliers are 1.09and 1.01, respectively. Thus, when a firm’s competitorshave footholds in each other’s main markets—referred toas asymmetric multimarket contact—the firm’s inclina-tion to mimic its direct competitors’ market entry movesis significantly higher.

Hypothesis 3 predicted that a firm’s propensity tomimic its rivals’ market entry moves will be lower ifits direct competitors do not have small footholds ineach other’s markets but instead have highly similarstakes in the same markets. Corroborating this hypoth-esis, in both studies the interaction term of symmetriccontact between a firm’s competitors and competitors’prior entry is negative and significant, which supportsthe idea that the former negatively moderates the influ-ence of the latter. In both studies, the influence of priorentry by a firm’s direct competitors would even turn neg-ative at very high levels of symmetric multimarket con-tact (two standard deviations above the mean): in Study1, each additional competitor’s entry would decrease therelative entry hazard by a factor of 0.92, and in Study2, the corresponding multiplier was 0.95. This findingimplies that in such a situation, firms actually avoid themarkets entered by their direct competitors. It indicatesthat a firm might also become less attracted to a certainmarket as a result of competitors’ prior entry—namely,if its rivals have highly symmetric market dependence.

5.2. Robustness Tests and Additional Analysis

Dichotomous Multimarket Contact Variables. Ourempirical approach to testing Hypotheses 2 and 3 wasto use one variable to measure the extent of multimar-ket contact and a second variable to measure the extentto which it was symmetric. Using this approach, in ourfull model, the former variable represents the case ofasymmetric multimarket competition and the latter vari-able picks up the influence of symmetric multimarketcontact. The advantage of this approach is that thesemeasures acknowledge that the extent of symmetry isa continuous variable. However, as a robustness check,we also created two dichotomous variables that directlyseparated asymmetric and symmetric multimarket con-tacts. This involved identifying a cutoff point of the dis-tance score (dk1 l) to distinguish between the two. Usinga grid search algorithm (see Bazaraa et al. 2006), theoptimal threshold that maximized model fit was 0.488

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Table 2 Cox Models for Study 1 (Chinese Pharmaceutical Industry)

Models

Variables 1 2 3 4 5

Competition between a firm’s competitorsDirect encounters between a firm’s competitors × 00131∗∗∗ 00183∗∗∗ 00140∗∗∗ 00233∗∗∗

Competitors’ prior entry (H1) 4000385 4000435 4000385 40004754Asymmetric5 multimarket contact between a firm’s competitors × 00147∗∗∗ 00313∗ 00164∗∗∗

Competitors’ prior entry (H2) 4000325 4001305 4000335Symmetric contact between a firm’s competitors × −20182∗∗∗ −00343† −20727∗∗∗

Competitors’ prior entry (H3) 4003165 4001835 4003965Symmetric and equal dependence × 00604∗

Competitors’ prior entry 4002985Symmetric and high share dependence × 00182∗

Competitors’ prior entry 4000925

Other controlsCompetitors’ prior entry 00017∗∗ 00016∗∗ 00009 00020∗∗ 00018∗

4000065 4000065 4000075 4000065 4000075A firm’s multimarket contact with its competitors × −00035 00043 00056 00041 00074∗

Competitors’ prior entry 4000215 4000315 4000355 4000325 4000375A firm’s symmetric contact with its competitors × 00005 −00023 00001 −00016 −00006

Competitors’ prior entry 4000145 4000175 4000185 4000175 4000195Market density 10354∗ 10303∗ 10157∗ 10270∗ 00991†

4005305 4005355 4005575 4005395 4005585Market density2 −00334∗∗ −00318∗∗ −00393∗∗ −00355∗∗ −00375∗∗

4001155 4001175 4001275 4001215 4001285Market exits −00003 −00001 −00001 −00004 −00003

4000285 4000285 4000295 4000285 4000295Incumbents operating at a loss −20845∗∗∗ −20731∗∗ −20636∗∗ −20681∗∗ −20534∗∗

4008605 4008555 4008755 4008575 4008745Excess capacity −30377 −30550 −20868 −30368 −20959

4207605 4207835 4207665 4207795 4207665Market concentration ratio (CR4) −10397∗∗ −10493∗∗ −10573∗∗ −10508∗∗ −10565∗∗

4005375 4005385 4005445 4005385 4005485Market size −00041 −00037 −00014 −00031 −00013

4000415 4000415 4000435 4000425 4000435Stratified baseline rate (by individual firms) Specified Specified Specified Specified Specified

Log likelihood −89706 −89100 −85707 −88702 −85107LR �2 against null model 20001∗∗∗ 21302∗∗∗ 27909∗∗∗ 22008∗∗∗ 29109∗∗∗

Akaike information criterion 1,815.1 1,804.0 1,741.3 1,800.4 1,733.3

Notes. N = 12,988 (spells). Standard errors are in parentheses. Model 4 is based on alternative dichotomized variables. LR, likelihood ratio.†p < 001; ∗p < 0005; ∗∗p < 0001; ∗∗∗p < 00001 (all two-tailed tests).

in Study 1 and 0.236 in Study 2. These two alterna-tive variables, as reported in Model 4 in Tables 2 and 3,also led to significant support for our predictions inboth studies: asymmetric multimarket contact between afirm’s direct competitors positively moderated the rela-tion between prior entrants and the firm’s entry rate,whereas symmetric competition negatively moderatedthe same relationship.

Multimarket Contact Accounting for Market Share.In our analysis above, we followed Gimeno (1999)using the proportion of the firm’s sales or portfolio inits various markets to distinguish between symmetricand asymmetric multimarket contact. However, althoughtheory pertaining to multimarket contact predominantly

revolves around relative market dependence (e.g., Both-ner 2003, Burt 1992, Chen and MacMillan 1992), onecould also argue that proportion of market share, in addi-tion to proportion of the firm’s total revenue, could playa role, too. For example, two rivals engaged in sym-metric multimarket competition could each have a lowmarket share in their main market (e.g., just 5% each)but could also each have a high market share in the samesegment (e.g., 40% each). Our measure does not distin-guish between the two situations. Therefore, in addition,we developed a variable taking into account the size ofeach multimarket competitor’s market share. Formallystated, for competitors k and l in common market m,let wk1m denote the product of k’s dependence on andmarket share in m, and let wl1m denote the product of

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Table 3 Cox Models for Study 2 (Taiwanese PC Industry)

Models

Variables 1 2 3 4 5

Competition between a firm’s competitorsDirect encounters between a firm’s competitors × 00160∗∗∗ 00143∗ 00169∗∗ 00091

Competitors’ prior entry (H1) 4000475 4000595 4000535 40006454Asymmetric5 multimarket contact between a firm’s competitors × 10430∗∗∗ 30402∗∗ 10545∗∗∗

Competitors’ prior entry (H2) 4003875 4100415 4004435Symmetric contact between a firm’s competitors × −00522∗∗ −60445† −10366∗∗∗

Competitors’ prior entry (H3) 4001595 4304415 4003295Symmetric and equal dependence × 10429∗∗∗

Competitors’ prior entry 4003775Symmetric and high share dependence × 00109∗

Competitors’ prior entry 4000495

Other controlsCompetitors’ prior entry 00053∗∗∗ 00041∗∗∗ 00044∗∗ 00069∗∗∗ 00011

4000115 4000125 4000165 4000155 4000205A firm’s multimarket contact with its competitors × −00031 00409 00240 00365 −00190

Competitors’ prior entry 4002875 4003355 4003595 4003485 4003975A firm’s symmetric contact with its competitors × −00067† −00062 −00069† −00053 −00096∗

Competitors’ prior entry 4000365 4000385 4000375 4000375 4000425Market density 00347∗∗∗ 00351∗∗∗ 00305∗∗ 00309∗∗ 00339∗∗∗

4000955 4000965 4000975 4000975 4000985Market density2 −00009∗∗∗ −00010∗∗∗ −00009∗∗∗ −00009∗∗∗ −00010∗∗∗

4000025 4000025 4000025 4000025 4000025Market exits 00034 00057 00079 00107 00150

4001375 4001395 4001425 4001425 4001475Internationalization 00383 00393 00444 00430 00397

4003215 4003205 4003255 4003255 4003265Wealth per capita 00357∗ 00383∗ 00428∗∗ 00426∗∗ 00480∗∗

4001505 4001515 4001565 4001555 4001565Skilled labor 00259 00280 00341† 00342† 00335†

4001795 4001805 4001875 4001865 4001865Transportation infrastructure 00309∗∗∗ 00320∗∗∗ 00268∗∗ 00273∗∗∗ 00312∗∗∗

4000795 4000795 4000825 4000825 4000845Stratified baseline rate (by individual firms) Specified Specified Specified Specified Specified

Log likelihood −56006 −55309 −54200 −54607 −53302LR �2 against null model 64703∗∗∗ 66008∗∗∗ 68405∗∗∗ 67501∗∗∗ 70200∗∗∗

Akaike information criterion 1,141.2 1,129.7 1,110.1 1,119.5 1,096.5

Notes. N = 3,703 (spells). Standard errors are in parentheses. Model 4 is based on alternative dichotomized variables. LR, likelihood ratio.†p < 001; ∗p < 0005; ∗∗p < 0001; ∗∗∗p < 00001 (all two-tailed tests).

l’s dependence on and market share in m; the extent towhich both parties are highly dependent on and have alarge share in market m is measured as the square rootof wk1m × wl1m. This score is then summed across allcommon markets shared by each pair of a firm’s directcompetitors. Finally, the measure symmetric and highshare dependence is computed as the average across allpairs of competitors facing a focal firm. See panel B ofTable A.3 in the appendix for some illustrative examples.

Model 5 in Tables 2 and 3 displays the estimateswith the inclusion of the interaction between this vari-able and prior entry by a firm’s direct competitors.As shown, in both our samples the estimate is positiveand significant, whereas the estimate of the interactionbetween symmetric multimarket competition and prior

entry remains negative and significant. This implies thatwhereas symmetric multimarket competition betweenone’s rivals usually deters imitative entry (as per Hypoth-esis 3), this appears not to be the case when it is accom-panied by high market shares. In this case, firms dofollow their rivals into the new market. It is somewhatspeculative what is driving this result, but two potentialexplanations come to mind: First, this finding is consis-tent with the industrial organization economics literatureon market concentration, which argues that high mar-ket concentration renders rivalry between market leadersmore benign (Scherer and Ross 1990). Our theory wouldindeed predict that such more subdued rivalry betweena firm’s competitors makes mimetic entry more likely,as we observe here. Furthermore, it may also be that

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


the signaling value of prior entry is not the same forfirms with low and high market shares. It seems possi-ble that entry by a large and dominant competitor mightbe interpreted by the firm as a stronger signal of mar-ket attractiveness than when a relatively minor playerenters (Haunschild and Miner 1997). Hence, in spite ofpossible past rivalry between the firm’s direct competi-tors, the firm may decide to enter anyway because of thestrong signal of market attractiveness. Future research onmultimarket competition might want to examine thesedifferent effects in more depth.

Multimarket Contact Accounting for Equal Depen-dence. Another shortcoming of our market dependence-based measure of symmetric multimarket competition isthat it does not take into account the absolute levels ofmarket dependence, just the extent to which the levelof dependence on a particular segment is equal to thatof its competitors. For instance, in the stylized exam-ple as displayed in Figure 4, our measure would notdistinguish between the situation in which two competi-tors each have a 10% and 90% dependence on theirtwo segments and the situation in which two competi-tors each have a 50% and 50% dependence on the samesegments; the variable just indicates that their depen-dence is the same—that is, symmetric. However, per-haps one might expect that the 50:50 situation is lessclear-cut in terms of how likely it is to foster aggressivecompetition versus mutual forbearance.4 Therefore, weconstructed an additional variable dedicated to captur-ing how similar the firms’ dependence is on the variousmarkets. Formally stated, for competitors k and l whohave Nk1 l markets in common (Nk1 l ≥ 2), the situation ofperfectly equal dependence would be when both parties’dependence on each common market is equal to 1/Nk1 l

(i.e., 50:50 with two common markets, or 25:25:25:25with four common markets). We first capture the extentto which k and l’s relationship deviates from the situ-ation of perfectly equal dependence using a Euclideandistance score and then reverse-code the distance scoreto obtain the equal dependence score. Finally, the mea-sure symmetric and equal dependence is computed asthe average across all pairs of competitors facing a focalfirm. See panel A of Table A.3 in the appendix for someillustrative examples.

Model 5 in Tables 2 and 3 includes the estimateswith the addition of the interaction between this variableand prior entry. As shown, in both samples, the esti-mate is positive and significant while leaving the orig-inal estimates testing our hypotheses intact. This resultindicates that a firm is indeed less likely to enter mimet-ically into a new market if its rivals are engaged insymmetric multimarket competition with unequal marketdependence—in other words, the situation as depicted inFigure 4. This confirms our hypothesis. However, this isnot the case when the firm’s competitors depend equally

on the different segments in which they operate. Hence,our hypothesis is especially relevant and supported incases where both competitors depend to a large extent onthe same segment (e.g., the 90:10 situation), not whendependence is much more equal (e.g., 50:50). Indeed,it seems plausible that competition is more fierce whenthe multimarket competitors depend heavily on the samesegment (e.g., both have a 90:10 dependence) rather thanwhen their interests are spread out over different seg-ments (e.g., both have a 50:50 spread over their twosegments). This is an important nuance to our theoryand perhaps for the study of multimarket competition ingeneral.

Main Effects. As discussed in the Methods section,in each of our models we controlled for all unobservedfirm-level variables using a separate baseline hazardfunction for each firm. The advantage of this approachis that it provided a superb control for all firm-leveleffects, both time-variant and time-invariant. The disad-vantage, however, is that the main effects of our pre-dictors (direct encounters between a firm’s competitors,multimarket contact between a firm’s competitors, andsymmetric contact between a firm’s competitors) cannotbe separately included and estimated in these stratifiedCox models. To check for robustness, we also estimatedalternative models without the firm-level stratification,with the main terms included separately. In these alter-native models, all the interaction terms still came outstatistically significant with the same sign.

Proportional Hazard Assumption. An assumption ofCox models is that the hazard rate can be specified ascovariates multiplicatively modifying the baseline hazardfunction. Given two observations with particular valuesof covariates, the ratio of the estimated hazards over timewill be constant within each stratum. Yet this propor-tional hazard assumption might not be valid for certaincovariates. Because our covariates are all time varying,we tested the proportional hazard assumption for eachof them using Schoenfeld residuals (Schoenfeld 1982).The results indicated that the assumption is rejected foronly two of our control variables (market density2 inStudy 2 and market exits in both studies). Hence, theproportional hazard assumption is not a problem for anyof our hypotheses tests. The models’ global chi-squarestatistics are 15.76 (p = 00328) in Study 1 and 23.57(p = 00052) in Study 2. Although the proportional haz-ard assumption appears to be borderline violated in thelatter case, this situation is driven entirely by the con-trol variable market exits. Removing this one variable—which is insignificant in itself—reduces the global chi-square to 11.54 (p = 00566). Moreover, neither includingnor excluding it leads to any noticeable difference in theestimated interaction effects between our predictors; thesame goes for excluding market density2.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Level of Analysis. We have argued that the marketrelationships between a firm’s direct competitors willinfluence the firm’s inclination to imitate the marketentrants among them. However, one question that mightarise is whether it is the encounters between all of afirm’s competitors that matter or only the encountersbetween those prior entrants. Our theory pertained to allof a firm’s direct competitors, mainly because nonen-trants may enter at some future point in time. To assessthis question empirically, we constructed alternative vari-ables that captured the direct encounters and the asym-metric and symmetric multimarket contacts betweenonly those competitors that have already entered the spe-cific target market. If a firm focuses mostly on priorentrants but largely ignores the nonentrants, these alter-native variables should perform better than the variablesreported in our tables. However, our empirical estimatessuggested otherwise: in both studies, these alternativevariables remained largely insignificant in terms of theireffect on the entry rate. Apparently, our results are notdriven by the competitive encounters between entrantsonly. We will return to this issue in the next section.

6. DiscussionIn this study, we examined the structure of competi-tion surrounding a firm, specifically in terms of theencounters between its competitors. Our main thesis isthat the extent and nature of rivalry between a firm’sdirect competitors determines to what extent a herd-ing effect occurs. Accordingly, our empirical analysisindicated how the structure of competition influencesa firm’s inclination to imitate its rivals—in particular,their market entry moves—with very consistent resultsacross our two settings. Dependent on its structuralproperties, competition may increase or decrease thefirm’s likelihood of mimetically entering a particularmarket. Although previous studies usually expected afirm to imitate its competitors, empirical findings hadbeen mixed or even conflicting. We contribute to theimitation literature by identifying this set of underre-searched factors that appear to be crucial moderators ofa firm’s imitative tendencies.

The general picture that emerges from our findings isthat a firm is especially inclined to imitate—in terms ofnew market entry—when its competitors form a densecluster of rivals, especially if these rivals hold footholdsin each other’s key segments. In such a situation, whena firm observes entry from this group into a new mar-ket, it is inclined to follow suit. Past mutual forbear-ance further raises expectations about the relative ease ofcompetition in the new market, increasing that market’sattractiveness even more. In contrast, imitative entry isless likely when the firm’s rivals do not encounter eachother and are scattered across different segments, orwhen they do encounter each other but engage in fiercehead-to-head competition because they depend heavilyon the same segment. In that case, herding is far less

likely because entry by competitors reduces a firm’sappetite for the new market.

Thus, another important contribution of our paper liesin the exploration of various characteristics of the com-petitive structure that different firms in an industry face.The types of structures we examined in this paper couldpotentially influence a range of other aspects of firmbehavior. We applied our ideas to explain imitative mar-ket entry; however, such competitive structures couldpotentially influence other aspects of firm behavior suchas price setting, the propensity to acquire and ally, andthe diffusion of practices and innovations (see Semadeniand Anderson 2010). Ultimately, the varying structuresof competition could affect organizations’ performanceand chances of survival. Few studies have explored howcompetitive linkages between particular competitors caninfluence other firms. Haveman and Nonnemaker (2000)indicated that mutual forbearance as a result of mul-timarket contact can spill over to other firms. Baumand Korn (1999) showed that multimarket competitionbetween two firms can stimulate other firms to also seekmultimarket contact—something they attributed to vicar-ious learning. Yet our study is the first to systematicallyanalyze and show how the particular pattern of compet-itive encounters between a firm’s rivals influences thechoices the firm makes.5 Our findings indicate that in itsactions, a firm is guided not only by its interactions withits competitors but also by the wider pattern of interac-tions among its competitors.6

6.1. Level of AnalysisInterestingly, our analysis showed that a firm’s deci-sion whether to enter a particular new market or notis moderated by the competitive encounters betweenall of its direct competitors, not just the subset of theprior entrants among them. Our theoretical argumentsincluded the suggestion that this is the case because thenonentrants will be considered potential entrants, whichinfluences the market expectations effect. Furthermore,to explore whether there are additional reasons why afirm’s entire structure of competition matters for imi-tative entry, and not just the encounters between thesubset of prior entrants, we conducted a series of 19face-to-face interviews with executives from six differentcompanies in the Taiwanese PC hardware industry (oursecond sample). The interviewees confirmed the firstpart of the mechanism: they pay attention to nonentrantsbecause they realize several of them may soon entertoo. For example, one executive commented, “For thosehotspots in China, our working assumption is that every-one is either already a player or will soon become one.”Another said, “They [firms that have not yet entered] arerelevant because they may enter any time.”

Yet the interviewees suggested an additional mech-anism. Their insights suggested that—at least in theirperception—firms become inherently more or less

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


aggressive because of the competitive interactions thatthey have experienced in the past. This inclinationtoward aggressiveness may persist even if a firm enters anew market and encounters a different competitive situa-tion. This implies that, for example, a firm that operatedin a situation with ample symmetric multimarket con-tact, which triggered fierce rivalry, may also be aggres-sive in a new market it enters, regardless of whether itsmultimarket contacts enter that market, too. One execu-tive commented, “Although their old rivals are not therein the new territory, their experience [surviving a pricewar] has taught them to stay tough no matter what.”This seems to constitute a form of competitive imprint-ing: the structure of competition determines whetherfirms become aggressive players or not, and they takethis aggressiveness with them into a new market, eventhough the situation there may be different. When fac-ing a new environment, adjusting their style may taketime. One executive, when confronted with the questionof whether a rival’s style may change in a different envi-ronment, commented, “They are what they are. True,a different market environment may eventually changetheir style, but it will be a slow process.” Thus, the struc-ture of competition may lead a firm to develop a certainstyle, through a process of imprinting, that it will takewith it into a new environment. This would also explainhow, in our setting and analysis, firms’ behavior is deter-mined by the full set of competitors around them, andnot just the subset of prior entrants, because the directrivals that did not enter also shaped the entrants’ currentcompetitive style. Using our notion of the structure ofcompetition, future research may focus more specificallyon the topic and process of competitive imprinting, thusenhancing our understanding of its impact.

6.2. Limitations and Future ResearchIn a general sense, prior research in management the-ory has emphasized how organizations are embedded inwider institutional and social contexts that influence theirbehavior and performance (Aldrich 1999, Burt 1992,Scott and Davis 2006, White 2002). In this paper, weendeavored to emphasize that firms are also embedded inwider structures of competitive interactions and to showhow these influence their behavior. We limited our analy-sis to examining the structure of competitive interactionsbetween a firm’s rivals. However, when mapping all thecompetitive encounters between firms within an industry,one could identify even wider structures. These patternsthrough which all firms connect to each other—what wecall the structure of competition—can differ significantlyacross industries and periods in time. In this paper, welimited our analysis to what could be considered the firstring of a firm’s structure of competition: the competi-tive relations between its rivals. In principle, analogousto social network theory, the concept could be extendedto examine much wider structures. For example, one

could imagine examining the influence of indirect tiesand structural holes in an organization’s competitive net-work (Burt 1992) or addressing the competitive linkagesbetween firms in a particular industry as forming a smallworld (Uzzi and Spiro 2005). Thus, the notion of thestructure of competition could be extended up to theindustry level, perhaps to compare the structures of anindustry at different points in time or to compare differ-ent industries altogether.

Studying the Chinese pharmaceutical industry, weexamined how the structure of competition influencesa firm’s entry into new product markets. In the sec-ond study, the Taiwanese computer hardware industry,the dependent variable was geographic market entry.In both studies, we mapped the independent variables—the structure of competition—using firms’ overlap interms of the product markets they serve. Research onmultimarket contact has used both product markets(Boeker et al. 1997, Stephan et al. 2003) and geo-graphic markets (Barnett 1993, Baum and Korn 1999,Greve 2000, Haveman and Nonnemaker 2000) to deter-mine market overlap. Similarly, one could imagine map-ping the structure of competition based on geography.Given the characteristics of our research settings, map-ping geographic structures might not be a fruitful courseof action.7 Nevertheless, future studies could potentiallyexplore geographic market structures, in addition to orperhaps in interaction with product market structures.

Our study showed that the structure of competitionwill influence firms’ decisions to enter mimetically intonew markets. However, this means that, in turn, firms’market entry might alter the competitive structure theyface. In other words, competitive market structures candrive firm behavior, but they also can result from firmbehavior. In this paper, we did not examine this recip-rocal loop. Similarly, market exit decisions might besubject to similar variables and processes (Boeker et al.1997, Bothner 2003, Greve 1995) that in turn influ-ence the structure of competition. Future research on thedynamic nature of the structure of competition mightprovide a more complete understanding of how marketstructures are shaped and evolve over time.

Following prior research on competitor analysis (Chen1996, Peteraf and Bergen 2003, Porac et al. 1995), weidentified direct competitors as firms competing in thesame product segments within a given industry. We usedthis to map the wider structure of competitive relation-ships between firms and analyzed how that influenced afirm’s propensity to imitate. Yet prior research has shownthat other forms of interfirm relationships, such as boardinterlock ties, might also influence imitation behavior(Haunschild 1993, Westphal et al. 2001) because marketactors are embedded in both social and competitive con-texts. Additionally, organizational decision makers mightsometimes identify firms operating in other industries astheir peers. For instance, Porac et al. (1999) showed thatcorporate boards might expand peer definitions beyond

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


industry boundaries when firms perform poorly, indus-tries perform well, CEOs are paid highly, and share-holders are powerful and active. Future research mightwant to consider multiple forms of interfirm relation-ships and categories jointly since it seems possible that,for instance, social structures and competitive structuresinteract.

Our analysis also uncovered some intriguing variantsof multimarket competition. Not only did we show—in conformity with Gimeno (1999)—that symmetric andasymmetric multimarket competition can have very dif-ferent, even opposing effects, our additional analysestaking into account separate measures of relative mar-ket share and measures of the absolute level of mar-ket dependence showed that these matter, too. Futureresearch on multimarket contact might further disentan-gle the effects of similarity in market dependence, abso-lute levels of dependence, and market share and developfurther theory in terms of what matters when. A limita-tion of our study is that we were not able to do all thiswithin the scope of this one paper.

Another limitation of our study is that we cannotentirely disentangle the various possible effects that leadto imitative behavior (see Lieberman and Asaba 2006)nor the different effects that could lead to competitivecrowding at higher levels of entry. The firm-specificbaseline hazard functions that we formulate in our mod-els may control for firm-specific tendencies, but there

Appendix

Table A.1 Product Markets Used in Our Studies

Panel A: Categories of active pharmaceutical ingredients (Study 1)

1. Anti-infectives for systemic use 12. Blood and blood-forming organs2. Musculoskeletal system 13. Solutions affecting the electrolyte balance3. Vitamins 14. Anesthetics4. Antiparasitic products, insecticides, and repellents 15. Antihistamines; antidotes5. Sex hormones; hormonal preparations 16. Biochemicals6. Antineoplastic and immunomodulating agents 17. Antiseptics and disinfectants7. Cardiovascular system 18. Sensory organs8. Respiratory system 19. Dermatologicals9. Nervous system 20. Diagnostic agents

10. Metabolism 21. Alimentary tract11. Genitourinary system 22. Solution additives

Panel B: Related product segments in the IT industry (Study 2)

1. Computers 11. Electro-medical equipment2. Monitors and terminals 12. Integrated circuits3. Computer peripherals 13. Discrete devices4. Audio and video equipment 14. Semiconductor packaging and testing5. Communication equipment 15. Electronic passive devices6. Telephones and cellular phones 16. Bare printed circuit boards7. Storage media 17. Printed circuit board assembly8. Cameras 18. Electronic parts and components9. Optical instruments 19. Liquid crystal panel

10. Measuring and control equipment 20. Optoelectronic materials and components

Note. The focal firms in our sample stem from segments 1 to 5, which appear in italics above.

could still be other influences when firms observe theirrivals enter. We observe the imitative tendency itself—or the absence thereof—but not exactly what is drivingit. Therefore, another limitation of our study is that wemust be careful in generalizing our results. We foundconfirmation for our predictions in two very differentcontexts but also realize that both contexts have theiridiosyncrasies. For example, in the Taiwanese sample,all the firms were listed. It is unclear to what extentthese results would hold for unlisted firms, which mightbe subject to smaller or different types of legitimacypressures. Furthermore, all of the Chinese pharmaceuti-cal sample concerned domestic firms. Although foreignfirms in China invariably operate in patented drugs, thevast majority of domestic firms operate solely in genericdrugs, which are the markets we were analyzing here.Future research comparing listed and nonlisted firms orthe influence of foreign on domestic firms, for exam-ple, might provide a fruitful avenue to further build ourunderstanding of the competitive interactions betweenfirms and imitative behavior in particular.

In summary, our study shows that firms might eitherfollow or steer away from their direct competitors in thecourse of market expansion depending on the patternof competitive relationships between their peers. In sodoing, it opens up a new avenue of research into howthe wider structures of competitive interactions that sur-round organizations affect organizations’ behavior.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Table A.2 Factor Analysis of Local Conditions in China (Study 2)

TransportationMeasures Internationalization Wealth per capita Skilled labor infrastructure

1. Total foreign capital 0 0898 00250 00250 001552. Exports 0 0871 00327 00289 001453. Imports 0 0829 00360 00270 002884. Gross domestic product per capita 00490 0 0625 00108 005515. Disposable income per capita 00450 0 0806 00209 002726. Household expenditure per capita 00437 0 0810 00176 002727. No. of professional personnel 00402 −00427 0 0745 −000678. Population with a college degree 00284 00202 0 0910 000819. No. of recent college graduates 00117 00306 0 0903 00020

10. Highway density 00437 00492 00061 0 065611. Railway density 00123 00178 −00009 0 0959Cumulative variance explained 00304 00539 00768 00944

Notes. Rotation: orthogonal varimax. Factor loadings displayed in italics indicate the measures used to construct thevariables.

Table A.3 Illustrative Examples of the Symmetric and Equal Dependence (A) and the Symmetric and High ShareDependence (B) Measures

Scenario I Scenario II Scenario III

Competitor 1 Competitor 2 Competitor 1 Competitor 2 Competitor 1 Competitor 2

Panel ADependence on market Y 0090 0088 0045 0052 0090 0012Dependence on market Z 0010 0012 0055 0048 0010 0088Perfectly equal dependence 1/2 = 0050 1/2 = 0050 1/2 = 0050

(no. of markets = 2)Distance score 0.780a 0.076b 0.780c

Equal dependence score 00780 − 00780 = 0 00780 − 00076 = 00704 00780 − 00780 = 0

Panel BDependence on market Y 0090 0088 0090 0088 0090 0088Share in market Y 0040 0040 0040 0004 0004 0004Dependence on market Z 0010 0012 0010 0012 0010 0012Share in market Z 0015 0015 0015 0015 0015 0015Symmetric and high share 0.382d 0.129e 0.052f

a√

40090 − 005052 + 40010 − 005052 + 40088 − 005052 + 40012 − 005052 = 00780.b√

40045 − 005552 + 40055 − 005052 + 40052 − 005052 + 40048 − 005052 = 00076.c√

40090 − 005052 + 40010 − 005052 + 40012 − 005052 + 40088 − 005052 = 00780.d√

40090 × 00405× 40088 × 00405+√

40010 × 00155× 40012 × 00155= 00356 + 00016 = 00382.e√

40090 × 00045× 40088 × 00405+√

40010 × 00155× 40012 × 00155= 00113 + 00016 = 001290f√

40090 × 00045× 40088 × 00045+√

40010 × 00155× 40012 × 00155= 00036 + 00016 = 000520

AcknowledgmentsThe authors thank Isabel Fernandez-Mateo, Gerry George,Ranjay Gulati, Brandon Lee, Olav Sorenson, Sandra Schruijer,and Ezra Zuckerman for comments on an early draft of thispaper. The authors also thank the anonymous reviewers andsenior editor Gautam Ahuja for their ideas and suggestions.The first author gratefully acknowledges financial support fromthe Sumantra Ghoshal Memorial Fund at the London Busi-ness School and the Singapore Ministry of Education [ACRF07/2008/133]. The second author acknowledges financial sup-port from the Centre for New and Emerging Markets at theLondon Business School.

Endnotes1Observing the entry moves does not reveal the actual informa-tion, but it does suggest to the firm that its competitors’ assess-ment of the market is positive. This does not mean that “theherd” is always correct in the sense that the market entered bythe competitors always turns out to be profitable, because theassessment of the early entrants could be flawed. It does implythat a firm’s estimation of the likelihood of market attractive-ness becomes more positive because of prior entry.2Left truncation is not an issue in our Taiwanese samplebecause if a spell started before 1999, we traced historicaldata back to 1991 (before which no firms in our sample wereoperating in China) so that we could determine the exact time

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


that had elapsed. And although there is left truncation in ourChinese sample, this should not introduce a bias because allproduct markets opened up gradually but at the same time.Furthermore, following Klein and Moeschberger (2003) andCleves et al. (2010), we deal with left truncation in our Coxmodels by adjusting the risk set at each point in time R4t5.3One could argue that highly diversified firms are less likelyto have many competitors that also encounter each other. Toaccount for this possible effect, we also created a variant of thevariable in which we divided the proportion of a firm’s directcompetitors that also compete with each other by the numberof markets in which the firm operates, thus controlling directlyfor the influence of diversification. In both samples, the esti-mates using this alternative measure were virtually identical tothe ones reported below, significant at p < 00001.4We thank an anonymous reviewer for suggesting both of theseimportant additions and lines of thought.5Our results, displayed in Tables 2 and 3, show that thevariables indicating the nature of the competitive encountersbetween a firm’s rivals had stronger moderating effects thandid the control variables indicating the nature of the compet-itive encounters between these rivals and the firm itself. Putdifferently, in our settings, how a firm’s competitors competedwith each other proved to be more influential than how thefirm itself competed with these rivals.6Note that according to our theory, firms do not need tobe aware of the exact shape of the structure of competitiveencounters around them but only need to observe the nature ofthe rivalry that results from it. This view was confirmed by anexecutive of one of our sample firms who said, “Except for afew big players, we rarely consider exactly who is competingwith whom and how. That would be way too complicated. Butwe do have an overall impression about how our competitorstypically behave toward one another.”7The Chinese pharmaceutical firms usually produce their drugsin only one location, and the headquarters of most TaiwanesePC hardware firms are clustered within one specific region—the Hsinchu Science Park.

ReferencesAbrahamson E (1997) The emergence and prevalence of employee

management rhetorics: The effects of long waves, labor unions,and turnover, 1875–1992. Acad. Management J. 40:491–533.

Aldrich HE (1999) Organizations Evolving (Sage, London).

Avery C, Zemsky P (1998) Multidimensional uncertainty and herdbehavior in financial markets. Amer. Econom. Rev. 88:724–748.

Banerjee AV (1992) A simple model of herd behavior. Quart. J.Econom. 107:797–817.

Barnett WP (1993) Strategic deterrence among multipoint competi-tors. Indust. Corporate Change 2:249–278.

Baum JAC, Korn HJ (1996) Competitive dynamics of interfirm rivalry.Acad. Management J. 39:255–291.

Baum JAC, Korn HJ (1999) Dynamics of dyadic competitive interac-tion. Strategic Management J. 20:251–278.

Baum JAC, Li SX, Usher JM (2000) Making the next move: Howexperiential and vicarious learning shape the locations of chains’acquisitions. Admin. Sci. Quart. 45:766–801.

Bazaraa MS, Sherali HD, Shetty CM (2006) Nonlinear Programming:Theory and Algorithms (John Wiley & Sons, Hoboken, NJ).

Bernheim BD, Whinston MD (1990) Multimarket contact and collu-sive behavior. RAND J. Econom. 21:1–26.

Bikhchandani S, Hirshleifer D, Welch I (1992) A theory of fads,fashion, custom, and cultural change as informational cascades.J. Political Econom. 100:992–1026.

Bikhchandani S, Hirshleifer D, Welch I (1998) Learning from thebehavior of others: Conformity, fads, and informational cas-cades. J. Econom. Perspect. 12:151–170.

Boeker W, Goodstein J, Stephan J, Murmann JP (1997) Competitionin a multimarket environment: The case of market exit. Organ.Sci. 8:126–142.

Bothner MS (2003) Competition and social influence: The diffusion ofthe sixth-generation processor in the global computer industry.Amer. J. Sociol. 108:1175–1210.

Breslow NE (1974) Covariance analysis of censored survival data.Biometrics 30:89–99.

Burns LR, Wholey DR (1993) Adoption and abandonment of matrixmanagement programs: Effects of organizational characteris-tics and interorganizational networks. Acad. Management J.36:106–138.

Burt RS (1992) Structural Holes: The Social Structure of Competition(Harvard University Press, Cambridge, MA).

Carroll GR (1985) Concentration and specialization: Dynamics ofniche width in populations of organizations. Amer. J. Sociol.90:1262–1283.

Carroll GR, Hannan MT (1989) Density dependence in the evolutionof populations of newspaper organizations. Amer. Sociol. Rev.54:524–541.

Chan CM, Makino S, Isobe T (2006) Interdependent behavior in for-eign direct investment: The multi-level effects of prior entryand prior exit on foreign market entry. J. Internat. Bus. Stud.37:642–665.

Chan CM, Makino S, Isobe T (2010) Does subnational region matter?Foreign affiliate performance in the United States and China.Strategic Management J. 31:1226–1243.

Chen MJ (1996) Competitor analysis and interfirm rivalry: Toward atheoretical integration. Acad. Management Rev. 21:100–134.

Chen MJ, MacMillan IC (1992) Nonresponse and delayed-responseto competitive moves: The roles of competitor dependence andaction irreversibility. Acad. Management J. 35:539–570.

Cleves M, Gould W, Gutierrez RG, Marchenko YV (2010) An Intro-duction to Survival Analysis Using Stata, 3rd ed. (Stata Press,College Station, TX).

Cox DR (1975) Partial likelihood. Biometrika 62:269–276.

Curry J, Kenny M (1999) Beating the clock: Corporate responsesto rapid change in the PC industry. Calif. Management Rev.42:8–36.

Dedrick J, Kraemer KL (1998) Asia’s Computer Challenge: Threator Opportunity for the United States and the World? (OxfordUniversity Press, Oxford, UK).

Delios A, Gaur AS, Makino S (2008) The timing of internationalexpansion: Information, rivalry and imitation among Japanesefirms, 1980–2002. J. Management Stud. 45:169–195.

Evans WN, Kessides IN (1994) Living by the “golden rule”: Multi-market contact in the U.S. airline industry. Quart. J. Econom.109:341–366.

Feinberg RM (1985) “Sales-at-risk”: A test of the mutual forbearancetheory of conglomerate behavior. J. Bus. 58:225–241.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Fiegenbaum A, Thomas H (1995) Strategic groups as referencegroups: Theory, modeling and empirical examination of industryand competitive strategy. Strategic Management J. 16:461–476.

Fligstein N (1985) The spread of the multidivisional form amonglarge firms, 1919–1979. Amer. Sociol. Rev. 50:377–391.

Fligstein N (1991) The structural transformation of Americanindustry: The causes of diversification in the largest firms.DiMaggio PJ, ed. The New Institutionalism in OrganizationalAnalysis (University of Chicago Press, Chicago), 311–336.

Garcia-Pont C, Nohria N (2002) Local versus global mimetism:The dynamics of alliance formation in the automobile industry.Strategic Management J. 23:307–321.

Gimeno J (1999) Reciprocal threats in multimarket rivalry: Stakingout “spheres of influence” in the U.S. airline industry. StrategicManagement J. 20:101–128.

Greve HR (1995) Jumping ship: The diffusion of strategy abandon-ment. Admin. Sci. Quart. 40:444–473.

Greve HR (1996) Patterns of competition: The diffusion of a marketposition in radio broadcasting. Admin. Sci. Quart. 41:29–60.

Greve HR (1998) Managerial cognition and the mimetic adoptionof market positions: What you see is what you do. StrategicManagement J. 19:967–988.

Greve HR (2000) Market niche entry decisions: Competition, learn-ing, and strategy in Tokyo banking, 1894–1936. Acad. Manage-ment J. 43:816–836.

Greve HR, Taylor A (2000) Innovations as catalysts for organizationalchange: Shifts in organizational cognition and search. Admin.Sci. Quart. 45:54–80.

Guillen MF (2002) Structural inertia, imitation, and foreign expan-sion: South Korean firms and business groups in China,1987–95. Acad. Management J. 45:509–525.

Guillen MF (2003) Experience, imitation, and the sequence of for-eign entry: Wholly owned and joint-venture manufacturing bySouth Korean firms and business groups in China, 1987–1995.J. Internat. Bus. Stud. 34:185–198.

Guthrie D (1997) Between markets and politics: Organizationalresponses to reform in China. Amer. J. Sociol. 102:1258–1304.

Han SK (1994) Mimetic isomorphism and its effect on the audit ser-vices market. Soc. Forces 73:637–664.

Hannan MT, Freeman J (1989) Organizational Ecology (Harvard Uni-versity Press, Cambridge, MA).

Haunschild PR (1993) Interorganizational imitation: The impact ofinterlocks on corporate acquisition activity. Admin. Sci. Quart.38:564–592.

Haunschild PR, Miner AS (1997) Modes of interorganizational imi-tation: The effects of outcome salience and uncertainty. Admin.Sci. Quart. 42:472–500.

Haveman HA (1993) Follow the leader: Mimetic isomorphism andentry into new markets. Admin. Sci. Quart. 38:593–627.

Haveman HA, Nonnemaker L (2000) Competition in multiple geo-graphic markets: The impact on growth and market entry.Admin. Sci. Quart. 45:232–267.

Henisz WJ, Delios A (2001) Uncertainty, imitation, and plant loca-tion: Japanese multinational corporations, 1990–1996. Admin.Sci. Quart. 46:443–475.

Institute for Information Industry (2000) Information Industry Year-book (Ministry of Economic Affairs, Taipei, Taiwan).

Jayachandran S, Gimeno J, Varadarajan PR (1999) Theory of multi-market competition: A synthesis and implications for marketingstrategy. J. Marketing 63:49–66.

Karnani A, Wernerfelt B (1985) Multiple-point competition. StrategicManagement J. 6:87–96.

Klein JP, Moeschberger ML (2003) Survival Analysis: Techniquesfor Censored and Truncated Data, 2nd ed. (Springer-Verlag,New York).

Knickerbocker FT (1973) Oligopolistic Reaction and MultinationalEnterprise (Harvard Business School Press, Boston).

Lieberman MB, Asaba S (2006) Why do firms imitate each other?Acad. Management Rev. 31:366–385.

McGrath RG, Chen MJ, MacMillan IC (1998) Multimarket maneuver-ing in uncertain spheres of influence: Resource diversion strate-gies. Acad. Management Rev. 23:724–740.

McNamara GM, Haleblian JJ, Dykes BJ (2008) The performanceimplications of participating in an acquisition wave: Early moveradvantages, bandwagon effects, and the moderating influenceof industry characteristics and acquirer tactics. Acad. Manage-ment J. 51:113–130.

Nickel MN, Fuentes JM (2004) Relationship between legitimation,competition and organizational death: Current state of the art.Internat. J. Management Rev. 5/6:43–62.

Pernell K (2010) Innovation, imitation, and idiocy: Bank involvementwith CDO underwriting, 1996–2008. Working paper, HarvardUniversity, Cambridge, MA.

Peteraf MA, Bergen ME (2003) Scanning dynamic competitive land-scapes: A market-based and resource-based framework. Strate-gic Management J. 24:1027–1041.

Porac JF, Wade JB, Pollock TG (1999) Industry categories and thepolitics of the comparable firm in CEO compensation. Admin.Sci. Quart. 44:112–144.

Porac JF, Thomas H, Wilson F, Paton D, Kanfer A (1995) Rivalryand the industry model of Scottish knitwear producers. Admin.Sci. Quart. 40:203–227.

Rogers EM (1995) Diffusion of Innovations (Free Press, New York).

Scharfstein DS, Stein JC (1990) Herd behavior and investment. Amer.Econom. Rev. 80:465–479.

Scherer FM, Ross D (1990) Industrial Market Structure and Eco-nomic Performance (Houghton Mifflin, Boston).

Schoenfeld DA (1982) Partial residuals for the proportional hazardsregression model. Biometrika 69:239–241.

Scott JT (1982) Multimarket contact and economic performance. Rev.Econom. Statist. 64:368–375.

Scott WR, Davis GF (2006) Organizations and Organizing: Ratio-nal, Natural and Open Systems Perspectives (Prentice-Hall,Englewood Cliffs, NJ).

Semadeni M (2006) Minding your distance: How management con-sulting firms use service marks to position competitively. Strate-gic Management J. 27:169–187.

Semadeni M, Anderson BS (2010) The follower’s dilemma: Innova-tion and imitation in the professional services industry. Acad.Management J. 53:1175–1193.

Stephan J, Murmann JP, Boeker W, Goodstein J (2003) Bringing man-agers into theories of multimarket competition: CEOs and thedeterminants of market entry. Organ. Sci. 14:403–421.

Uzzi B, Spiro J (2005) Collaboration and creativity: The small worldproblem. Amer. J. Sociol. 111:447–504.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.


Westphal JD, Seidel MDL, Stewart KJ (2001) Second-order imita-tion: Uncovering latent effects of board network ties. Admin.Sci. Quart. 46:717–747.

White HC (2002) Markets from Networks: Socioeconomic Models ofProduction (Princeton University Press, Princeton, NJ).

White S (2000) Competition, capabilities, and the make, buy, or allydecisions of Chinese state-owned firms. Acad. Management J.43:324–341.

White S, Liu X (1998) Organizational processes to meet new perfor-mance criteria: Chinese pharmaceutical firms in transition. Res.Policy 27:369–383.

Yang M, Hyland M (2006) Who do firms imitate? A multilevelapproach to examining sources of imitation in the choice ofmergers and acquisitions. J. Management 32:381–399.

Zwiebel J (1995) Corporate conservatism and relative compensation.J. Political Econom. 103:1–25.

Kai-Yu Hsieh is an assistant professor of strategy andpolicy at the School of Business, National University ofSingapore. He received his Ph.D. from London BusinessSchool. His research interests include corporate strategy, for-eign market entry, and organizational learning, with particu-lar emphasis on the influence of competing organizations onstrategic decision making.

Freek Vermeulen is an associate professor of strategy andentrepreneurship at the London Business School. He receiveda Ph.D. in strategic management in 1999 from Tilburg Univer-sity and a Ph.D. in organization science in 2010 from UtrechtUniversity, the Netherlands. He studies the limitations of orga-nizations. His work has appeared in the Academy of Man-agement Journal, Administrative Science Quarterly, HarvardBusiness Review, Organization Science, and the StrategicManagement Journal, among others.

Copyright:

INFORMS

holdsco

pyrig

htto

this

Articlesin

Adv

ance

version,

which

ismad

eav

ailableto

subs

cribers.

The

filemay

notbe

posted

onan

yothe

rweb

site,includ

ing

the

author’s

site.Pleas

ese

ndan

yqu

estio

nsrega

rding

this

policyto

perm

ission

s@inform

s.org.

the structure of competition: how competition between one ... · hsieh and vermeulen: the structure...

Documents