PERFORMANCE DIFFERENTIALS BETWEEN DIVERSIFYING …terpconnect.umd.edu/~rajshree/research/31 Ganco, Agarwal - 2009.pdfCOMPLEXITY APPROACH MARTIN GANCO RAJSHREE AGARWAL ... Survival,

PERFORMANCE DIFFERENTIALS BETWEENDIVERSIFYING ENTRANTS AND

ENTREPRENEURIAL START-UPS: ACOMPLEXITY APPROACH

MARTIN GANCORAJSHREE AGARWAL

University of Illinois at Urbana-Champaign

We investigate the relationship between firms’ entry characteristics and their subse-quent performance as contingent on environmental turbulence and stage of industry lifecycle by simulating industry as an NKC landscape. Diversifying entrants differ fromentrepreneurial start-ups in terms of the complexity of their routines. We posit thatdiversifying entrants outperform entrepreneurial start-ups when turbulence is high. Fur-ther, learning—possible in later industry stages—disproportionately favors entrepre-neurial start-ups.

An enduring theme of research centers on theperformance consequences of heterogeneity infirms’ characteristics at their time of entry intoan industry (Helfat & Lieberman, 2002). Whetherexamined via an industry evolution focus ontiming of entry (Agarwal & Gort, 1996; Lieber-man & Montgomery, 1988) or via a capabilitiesapproach (Helfat & Lieberman, 2002; Klepper &Simons 2000) or an organizational ecology ap-proach (Carroll, Bigelow, Seidel, & Tsai, 1996;Carroll & Khessina, 2005), firms’ attributes at thetime of entry have a persistent effect on theirlater organizational form, and such “imprinting”explains much of the heterogeneity in the pop-ulation of firms in an industry (Levinthal, 1997;Stinchcombe, 1965). However, the consensusabout the long-term performance implications ofinitial firm characteristics is considerablyweaker, particularly when viewed in the contextof rapidly evolving industries (Bayus & Agarwal,2007; Klepper & Simons, 2000). Indeed, there is con-

flicting evidence, with some evidence indicatingthat firms with pre-entry experience have “domi-nance by birthright” and some evidence indicat-ing that entrepreneurial start-ups can “creativelydestruct” the status quo and gain from advan-tages associated with the structural differencesbetween them and the diversifying entrants.

To investigate how pre-entry experiencemight confer performance advantages, we usean agent-based simulation approach that hasbeen extensively used and tested in manage-ment research (Ethiraj & Levinthal, 2004; Flem-ing & Sorenson, 2001, 2004; Levinthal, 1997;Rivkin, 2000; Sorenson, 1997, 2003). Althoughthese models typically have been used to exam-ine within-firm dynamics, they are also relevantto questions of industry evolution.

Since performance is a consequence of the fitbetween organizational capabilities and environ-mental conditions (Helfat & Lieberman, 2002), wefocus on how pre-entry experience interacts withenvironmental conditions measured along twoimportant dimensions of an industry: degree ofturbulence and maturity. Our simulation modelsuggests that diversifying entrants tend to outper-form start-ups in a more turbulent environment;however, start-ups benefit disproportionately fromlearning when entering a mature industry.

We begin by reviewing work on the perfor-mance differences between diversifying entrantsand entrepreneurial start-ups. The review enablesus to identify empirical regularities that have re-ceived significant support, as well as less estab-

Both authors contributed equally; our names are in reversealphabetical order. We thank the Ewing Marion KauffmanFoundation for grant funding that supported this research. Themanuscript has benefited significantly from the comments ofspecial issue editor Olav Sorenson, the three anonymous re-viewers, Adrian Caldart, Glenn Hoetker, Suresh Kotha, JohnMiller, Scott Page, Shawn Riley, Jaume Villanueva, CharlesWilliams, Govert Vroom, and participants at seminars at the2007 Academy of Management Meetings, 2007 Strategic Man-agement Society Meetings, 2006 Understanding Complex Sys-tems Symposium, 2006 Graduate Workshop in ComputationalSocial Sciences, and the University of Illinois at Urbana-Champaign. All remaining errors are our own.

� Academy of Management Review2009, Vol. 34, No. 2, 228–252.

228Copyright of the Academy of Management, all rights reserved. Contents may not be copied, emailed, posted to a listserv, or otherwise transmitted without the copyrightholder’s express written permission. Users may print, download, or email articles for individual use only.

lished areas that require theoretical attention. Wethen develop a complexity model by mapping theNKC model parameters to our specific context anddeveloping propositions based on the simulationresults. The final section provides a discussion ofthe key contributions, limitations, and avenues forfuture research.

LITERATURE REVIEW

Pre-entry Experience and Performance

Pre-entry experience has been identified asan important source of heterogeneity among in-dustry entrants that has consequences forpostentry performance (for a review see Helfat &Lieberman, 2002). Diversifying entrants— de-fined as firms that already existed before enter-ing the focal industry—differ from entrepreneur-ial start-ups—defined as firms founded in thefocal industry context— on two important di-mensions. First, diversifying entrants typicallyenter with more financial, managerial, and re-lated technological or marketing resources andcapabilities (Carroll et al., 1996; Klepper & Si-mons, 2000; Lane, 1989). These resource advan-tages often provide diversifying entrants atleast a short-term performance advantage, partic-ularly if they enter from related industries, so thatthe degree of fit between firm resources and focalindustry requirements for success is higher (Car-roll et al., 1996; Klepper & Simons, 2000; Mitchell,1991). For instance, Carroll et al. (1996) found thatdiversifying entrants with relevant specializedand transferable knowledge outperformed allother automobile industry entrants. Similarly,Mitchell (1991) related the performance advantageof diversifying entrants in new diagnostic imag-ing markets to their access to distribution chan-nels and complementary assets.

A second dimension of difference relates tostructural inertia (Hannan & Freeman, 1984). Di-versifying entrants possess more inert and es-tablished processes, such as “documents, un-derstandings and agreements intended asblueprints for future action” (Carroll et al., 1996:120). Entrepreneurial start-ups have more fluidand organic structures that change relativelyeasily (Hannan & Freeman, 1984). This distinc-tion between the two types of firms is akin to theorganizational ecology distinction between gen-

eralist and specialist firms (Aldrich, 1990; Brit-tain & Freeman, 1980; Hannan & Freeman, 1977).1

While superior access to resources provides anadvantage to diversifying entrants, the perfor-mance consequences of structural inertia are notclear. The more inert processes and structures ofdiversifying entrants can enhance their legiti-macy and make them appear more reliable tocustomers (Hannan & Freeman, 1984). However,the less inert structures of start-ups (Hannan &Freeman, 1984) and their higher rate of new prod-uct innovations (Khessina, 2002, 2003) may enablethem to span niches and quickly reorient activi-ties. These firms are better equipped for explora-tion, because their fluid and organic structuresand routines allow them to avoid the myopiclearning (Levinthal & March, 1993) and compe-tency traps (Levitt & March, 1988) endemic to themore established firms. The absence of ties toexisting processes (Carroll et al., 1996) or existingcustomer needs (Christensen, 1997) may enablestart-ups to engage in creative destruction(Schumpeter, 1934), as was evidenced in the diskdrive industry (Agarwal, Echambadi, Franco, &Sarkar, 2004; Christensen, 1997) and the laser in-dustry (Klepper & Sleeper, 2005).

Thus, theoretically, the “main effect” of pre-entry experience is ambiguous, particularlywhen one is examining long-term consequenceson performance. Not surprisingly, there is con-flicting empirical evidence regarding the maineffect of pre-entry experience on performance.As Table 1 summarizes, evidence exists for di-versifying entrant advantage (Barnett & Free-man, 2001; Freeman, 1990; Hannan & Freeman,1988; Rao, 1994), start-up firm advantage (Agar-wal et al., 2004; Carroll & Khessina, 2005;Khessina, 2002, 2003), and convergence betweenthe two (Carroll et al., 1996; Hannan, Carroll,Dobrev, & Han, 1998; Khessina & Carroll, 2008).The contradictory evidence leads us to formu-late our main question, as follows.

Question 1: Under what conditions willentrepreneurial start-ups outperform di-versifying entrants, and vice versa?

1 The generalism/specialism dichotomy does not map pre-cisely onto diversifying/start-up entrants since both entrantsmay compete in a same-width niche in the focal industry.The comparison is applicable, though, to the extent thatdiversifying entrants can be thought of as occupying a widerrange of activities than the focal industry alone.

2009 229Ganco and Agarwal

One potential route to reconciliation of thesecontrary findings is to look for contingency con-ditions that affect this relationship. In this paperwe focus on two industry-level factors that mayimpact the pre-entry experience–performance re-lationship: (1) the extent of environmental turbu-lence and (2) the accumulated stock of knowledgein an industry. We briefly review the relevant lit-

erature on each factor, with a focus on currentknowledge of potential contingency effects.

Environmental Turbulence

Industry environments differ in their degreesof turbulence, or dynamism. Turbulent or high-velocity industries typically have ambiguous

TABLE 1Literature on Performance Differentials Between Diversifying Entrants and Start-ups

Type ofRelationship

Performance Measureand Comparison

Best-Performing Entrant Type

Diversifying Entrants Start-ups

Main effects Survival, fitness Barnett & Freeman (2001), Freeman(1990), Hannan, Carroll, Dobrev,& Han (1998), Hannan & Freeman(1988), Klepper & Simons (2000),Rao (1994)

Agarwal, Echambadi, Franco,& Sarkar (2004)—forspinouts

Convergence: Carroll, Bigelow, Seidel, & Tsai (1996), Hannan,Carroll, Dobrev, & Han (1998), Khessina & Carroll (2008),Mitchell (1991)

Innovativeness (frequentchanges—also leads tohigher variance and risk)

Carroll & Khessina (2005),Khessina (2002, 2003)

Contingency: Turbulence

High turbulence (high C) Survival, fitness Aldrich (1990), Brittain & Freeman(1980), Hannan & Freeman (1977),Sastry (1997), Sine, Mitsuhashi, &Kirsch (2006)

Low turbulence (low C) Survival, fitness Aldrich (1990), Brittain &Freeman (1980), Hannan &Freeman (1977), Sastry(1997), Sine, Mitsuhashi, &Kirsch (2006)

Contingency: Industry Cycle

Growth stage Survival, short-run postentry Agarwal (1997), Carroll, Bigelow,Seidel, & Tsai (1996), Klepper(2002a,b), Klepper & Simons(2000), Mitchell (1991)

Survival, long-run postentry Klepper & Simons (2000)—in theTV industry

Convergence: Carroll, Bigelow, Seidel, & Tsai (1996), Klepper(2002a)—in the car and tire industries

Mature stage Survival, market share, lateentrants only

Mitchell (1991) Agarwal (1997), Bayus &Agarwal (2007)

Survival, market share, lateentrants versus earlyentrants

Early diversifying entrants:Klepper & Simons (2000)—in theTV industry; Mitchell (1991)

Late start-ups: Bryman (1999),Klepper (2002b), Klepper &Sleeper (2005)

Note: To map the findings of the organizational ecology literature onto our categories of diversifying and start-up entrants,we assume that diversifying entrants are generalists and start-up firms are specialists. To map the findings of Sine,Mitsuhashi, and Kirsch (2006), we assume that diversifying entrants have a more formal and routinized organizationalstructure than start-ups. To map the findings of Sastry’s (1997) simulation model, we assume that the firm with more frequentchanges corresponds to a start-up.

230 AprilAcademy of Management Review

structures, blurred boundaries, fluid businessmodels, ambiguous and shifting players, non-linear and unpredictable change, short productcycles, and rapidly shifting competitive land-scapes (Brown & Eisenhardt, 1997; Eisenhardt &Martin, 2000). In contrast, industries with lowlevels of turbulence represent less uncertain en-vironments and relatively stable competitivelandscapes.

Overall, scholars examining the effects of en-vironmental turbulence emphasize the need fororganizational change in more turbulent envi-ronments. Hailing back to Burns and Stalker(1961), a core theme of organization and innova-tion studies is that “mechanistic organizations”(those with more centralized and formal struc-tures) perform better in stable markets that re-ward efficiency, whereas “organic organiza-tions” (those with more decentralized andinformal structures) perform better in turbulentmarkets (Eisenhardt & Tabrizi, 1995; Pisano,1994). This distinction arises because the imper-ative to change in response to environmentalshifts (Haveman, 1992) implies that fluid struc-tures and simpler rules that enable change pro-vide performance benefits as turbulence in-creases (Brown & Eisenhardt, 1997; Davis,Eisenhardt, & Bingham, 2007; Rindova & Kotha,2001).

However, recent theory and research showthat excessive change can be detrimental (Sas-try, 1997; Sine, Mitsuhashi, & Kirsch, 2006; Soren-son, McEvily, Ren, & Roy 2006). For instance,Sorenson et al. (2006) disentangled the behav-ioral and positional effects of change in orga-nizational scope and found that althoughchange results in positional benefits, the ten-dency of generalists to constantly change andexpand can be temporarily detrimental. Sine etal. (2006) examined how organizational structurerelates to the performance of small ventures in aturbulent environment. Drawing on a body ofprior research, these scholars maintained thatthe highly formal and routinized structures ofmature firms inhibit change and yield lowerperformance. However, they showed that, forsmall ventures in very turbulent environments,formal structures actually help performance,and they concluded that small ventures haveflexibility but need stability. Similarly, follow-ing the tradition of systems dynamics (Forrester,1961), Sastry (1997) built a single-firm model ofpunctuated organizational change, focusing on

the key trade-off between inertia that promotescompetence and change that destroys inertia. Ifan organization attempts to change in responseto environmental shifts, it restarts the “inertiaclock” and destroys competence. Consequently,being too responsive to change is not beneficialin turbulent environments since it diminishescompetence.

The distinction made above parallels that be-tween generalist and specialist firms. Organiza-tional ecologists (e.g., Aldrich, 1990; Brittain &Freeman, 1980; Hannan & Freeman, 1977) as-sume that organizations sacrifice some perfor-mance when they compete in a wider niche.Consequently, specialists outperform general-ists in environments within the narrow band oftheir specialty, and generalists fare better when“environments vary across different states withsome degree of uncertainty” (Aldrich, 1990: 13).

Although the literature on the effects of envi-ronmental turbulence on firm performance isrelatively silent on performance differentialsdue to pre-entry experience, the above researchraises the intriguing possibility that the degreeof environmental turbulence may explain theconflicting empirical findings. Diversifying en-trants arguably have more structural links andgreater organizational inertia than entrepre-neurial start-ups; thus, performance implica-tions differ under varying degrees of environ-mental turbulence. Such a notion is consistentwith the work of Khessina (2002), who found thatstart-ups introduce product innovations at ahigher rate than diversifying entrants but none-theless also fail at a higher rate (Carroll &Khessina, 2005). In line with this conjecture, oursecond question of interest is as follows.

Question 2: How does an industry’s en-vironmental turbulence affect the per-formance differential between entre-preneurial start-ups and diversifyingentrants?

Stage of Industry Life Cycle

Multiple research streams—in technologymanagement, evolutionary economics, and or-ganizational ecology—examine the evolution ofindustries over time (e.g., Gort & Klepper, 1982;Hannan & Freeman, 1977, 1988; Utterback & Ab-ernathy, 1975). Reconciling these three bodies ofscholarly work, Agarwal, Sarkar, and Echam-


badi (2002) noted that a common theme is theidentification of growth and mature life cyclestages, although scholars differ on whether theunderlying mechanism relates to “pre and post”dominant design (Abernathy & Utterback, 1978),legitimacy versus competitive effects of popula-tion density (Carroll & Hannan, 2000), or entre-preneurial versus routinized regimes (Au-dretsch, 1997; Nelson & Winter, 1982).

Importantly, although the literature is un-equivocal on the survival benefits of entering inthe early rather than later stages of an indus-try’s life cycle (Agarwal & Gort, 1996; Hannan &Freeman, 1988; Suarez & Utterback, 1995), thereis less agreement on whether different types ofentrants are differentially advantaged by entrytiming. Table 1 illustrates this lack of agree-ment. Klepper and Simons (2000) found evidencethat diversifying entrants entering early enjoy a“dominance by birthright,” yet Carroll et al.(1996) and Bayus and Agarwal (2007) found thatthese advantages erode over time and may bereversed. Similarly, while Agarwal (1997) docu-mented diversifying entrant advantage duringthe growth stage across thirty-three industries,Mitchell (1991) found that diversifying entrantsconsistently outperformed entrepreneurial en-trants regardless of time of entry. Among thelate cohorts, Klepper (2002a) found no significantperformance differential among the two types ofentrants, Mitchell (1991) documented support fordiversifying entrant advantage, and Agarwal(1997) and Bayus and Agarwal (2007) found sup-port for entrepreneurial start-up advantage.Conflicting evidence is also present amongstudies that compare early and late cohorts.Some studies show support for diversifying en-trants, whether entering early (e.g., Klepper &Simons, 2000) or later (e.g., Mitchell, 1991). How-ever, researchers examining employee entrepre-neurship (e.g., Agarwal et al., 2004; Klepper &Sleeper, 2005) maintain that the incumbent ad-vantage can successfully be challenged. Agar-wal et al. (2004) found that start-up firms createdby ex-employees of incumbent firms outperformedall other entrants, including diversifying andearly entrants. Similarly, Klepper (2002b) showedthat, in the automobile industry, late start-upentrants founded by former employees of incum-bent firms outperformed all other cohorts.

One reason for the conflicting findings may bethe extent of later entrants’ learning from anindustry stock of knowledge. Indeed, a key fea-

ture that distinguishes the “stable” maturestage of the industry life cycle from the “fluid”growth stage is resolution of product standardsand architecture issues; often, a dominant de-sign has evolved (Abernathy & Utterback, 1978;Anderson & Tushman, 1990; Gort & Klepper,1982; Murmann & Frenken, 2006). Gort and Klep-per (1982) described the higher development ofindustry-specific knowledge in the later stageas a critical difference between stages. Aber-nathy and Utterback’s (1978) fluid and specificphases and Anderson and Tushman’s (1990) erasof ferment and incremental change are similardistinctions.

Consequently, in the fluid phase or fermenta-tion era, entrants engage in significant experi-mentation regarding product design and stan-dards (Abernathy & Utterback, 1978; Anderson &Tushman, 1990). In the mature stages of an in-dustry, in contrast, entrants have the option oflearning from the established stock of knowl-edge and premeditating their entry strategies.The entrants may access learning via simpleimitation (Rivkin, 2000; Schumpeter, 1934), pat-ents (Almeida & Kogut, 1999), collaborative ar-rangements (Rosenkopf & Almeida, 2003), andemployee mobility (Agarwal et al., 2004). Thus, iflearning is an important differentiating factorbetween early and late industry stages, the fol-lowing question becomes an important one toinvestigate.

Question 3: How does the ability oflate entrants to learn from an indus-try-specific stock of knowledge affectthe performance differentials betweenentrepreneurial start-ups and diversi-fying entrants?

A COMPLEXITY APPROACH TO MODELINGPERFORMANCE DIFFERENTIALS BASED ON

ENTRANT CHARACTERISTICS

To reconcile previous findings and to shedlight on understudied contingency conditions,we employ an agent-based model to simulatehow organizational structures and processesmay interact with environmental conditions toexplain performance differentials.

The NKC Model

Kauffman’s NKC model (1995) is an extensionof the NK model (Kauffman, 1993: Chapter 6),


which is widely used in research on strategy(Ethiraj & Levinthal, 2004; Levinthal, 1997; Rivkin,2000). The NKC model was developed to modelthe coevolution of species and, thus, can beadapted to model an evolving industry with het-erogeneous entrants. In particular, the frame-work permits us to model an industry with dif-ferentiated products and each firm occupying acertain exogenous niche. A salient aspect of themodel is that it incorporates interfirm interac-tion: each firm’s choices have an impact on thepayoffs of the choices of the other firms. Such astructure allows us to focus on environmentalcharacteristics caused by the coevolution of het-erogeneous firms.

In the NKC model there are N elements of adecision vector in each firm. Like previous re-searchers (Levinthal, 1997; Rivkin, 2000; Rivkin &Siggelkow, 2003), we assume that the binary bitsof each firm’s decision vector represent organi-zational attributes or routines (Nelson & Winter,1982). We adopt Rivkin’s (2000) interpretationthat the binary bits represent broadly definedorganizational decisions related to firm strat-egy, organizational form, product design, and soforth. We use the term routines to broadly repre-sent all internal firm processes. The value ofeach bit—0 or 1—represents a decision aboutroutines (e.g., routine A versus routine B is cho-sen).

The parameter K measures the degree of in-terdependence or intrafirm coupling betweenthe N elements of the decision vector—that is,the performance contribution of each element ofthe vector xi, i � 1, . . . , N, is affected by K otherelements xj, where j is not equal to i. The perfor-mance contributions of the decision elementsare determined by the payoff function, whichworks as follows: if a coupling exists betweenthe decision element x1 and element x3 in whichx1 is the focal element (x3 affects x1), then achange in x3 (decision B instead of decision A ischosen) changes the payoff contribution of x1(the value is simply redrawn from the underly-ing distribution). When x1 is coupled to many (K)other decisions, its payoff is redrawn wheneverany of the coupled decisions change. The over-all payoff of the entire decision vector is themean of the payoff contributions of its individ-ual elements. A high value of K implies a “rug-ged landscape” arising from the underlyinghigh interdependence of decision elements(high interdependence is also described as high

coupling or complexity). Following prior work,we define the “peaks” or local optima on the NKlandscape as configurations of the elements of adecision vector that are such that it is not pos-sible to improve the decision’s overall payoff byaltering any single decision element.

Describing the model formally, the NK land-scape is characterized by the correspondencemapping of the vector x in the decision space tothe outcomes (payoffs). The landscape is a map-ping from the set X � {0, 1}N to R�. An elementx � X is a vector of binary digits of length N. Themapping assigns to each x � X a payoff, �(x) �R�. The mapping � depends on the parameter K,with �(x, K) reflecting the interdependence of theindividual components of x. The change in thepayoff contribution of the ith component is influ-enced by the change in the ith decision xi, andby the changes in the K other components of x. IfK � 0, there are no interdependencies and the�(.) function is additive. The mapping is gener-ated by assigning a payoff �i(.), which is a ran-dom number from a standard normal distribu-tion, to each decision xi, i � 1, . . . , N, and eachinstance in which either xi changes or some ofthe K decisions that are associated with xichange.2 The mapping is then given by

�(x, N, K) �1N�

i�1

N

�i( xi; xj(i)1 . . . xj(i)

K ), i � j(i),

where for any i we obtain a vector of indexes j(i)mapping from N to NK. None of the indexes of j(i)can be equal to i. The notation xj(i)

k means thatthe index of x is the kth element of the vector j(i).To create an overall mapping, we need to ran-domly generate 2K�1N payoff values. The struc-ture of the mapping �(.) is often depicted as amatrix called the “interaction” or “influence ma-trix.” The rows and columns represent individ-ual decision elements. The matrix has a 1 ineach entry affected by a particular decision el-ement (typically, row element affected by col-umn element). For instance, for K � 0, the inter-

2 Kauffman (1993, 1995) used a uniform [0, 1] distributionfor the draws. Our use of a standard normal distribution—forcomputational reasons and to better represent economicpayoffs where unusually high and low payoffs are veryunlikely—has no qualitative effect on the results (it does notchange ordering) and is asymptotically (for large N) equiv-alent to the uniform distribution owing to the central limittheorem.


action matrix is an N � N identity matrix, and forK � N � 1, it is an N � N matrix of ones. As in theoriginal Kauffman (1995) models, in our modelthe interdependencies within each firm are as-sumed to be randomly distributed: the elementsof the index vector j(i) are generated randomlyfrom the interval [1, N], with i � j(i).

The parameter C specifies the extent to whichindividual firms’ “sublandscapes” are tied to-gether—that is, it specifies the extent of inter-firm coupling. A high value of C implies a highlyinteractive environment where each firm’s deci-sion affects many decisions of other firms. If C �0, the firms operate in completely independentlandscapes. C is assumed to be outside the controlof individual firms and is given by the character-istics of the environment. In the original modelspecification (Kauffman, 1995), the payoff of eachdecision within the decision vector of each firm, xi,is affected by its own value, by K other within-firmdecisions, and by C decisions of each other firm inthe environment. This implies that a change fromCKauffman � 2 to CKauffman � 3 increases the num-ber of couplings within a single C matrix for N �10 from twenty to thirty links. If a focal firm islinked to three other firms, an increase of thirtylinks is implied. The CKauffman basically controlsthe number of links in each row of the C matrix,assuming that all rows are filled.

Since such a specification is extremely coarsefor our purposes— changes in C matter at amuch finer level for the performance differen-tials—we modify the model by assuming thatonly some of the elements are linked to the otherfirms’ decisions. The assumption seems reason-able since the decision vector x represents a setof broadly defined organizational attributes ordecisions—that is, firms do not tend to interactalong all decisions with other firms. Our C thencontrols the number of decision elements xi ofeach firm that are linked to decisions of otherfirms—the number of rows of the C matrix thathave at least one entry. We then introduce a �parameter (and set it to 2 in all simulations buttest its robustness), defining the number of linksin each row (for C � 1 there are � entries per row;for C � 1 there is only one entry). These modifi-cations allow very refined changes. For in-stance, with � � 2, an increase in C from 2 to 3changes the number of links in a given C matrixby only two and changes links to all three firmsby six. The C matrix with our specification andparameter values C � 10, � � 2 is equivalent to

C matrix with CKauffman � 2 in the original spec-ification (C � 10, � � 10 would be equivalent toCKauffman � 10). As with the intrafirm coupling,we assume that the links are distributed ran-domly across the coupled decisions. Figure 1illustrates the full NKC interaction matrix for theparameter values N � 10, Kstart-up � 2, Kdivers. �5, C � 7, � � 2.

The level of interfirm coupling controlled by Calso determines the level of environmental tur-bulence. Parameter C controls the number offirm decision variables that are tied to variablesof other firms. However, the turbulence resultsfrom the changes of the tied variables of otherfirms. For example, if C � 5, five decisions arecoupled to another firm’s decisions. However, ifthe other firm optimizes by changing only onedecision and keeps the other four constant, theturbulence resulting from such a relationshipwill be small. As firms optimize their problemsand climb peaks, the turbulence that each firmfaces decreases over time, even as the level ofinterdependence remains constant. As firmsadapt, the variability of their decisions de-creases and, thus, also the effect on other firms’choices. As we show below, different values of Caffect firms differently depending on the com-plexity of their internal routines.

In keeping with prior work, we assume thatfirms learn primarily (or on average) throughlocal search and exploitation. The local searchis modeled by alteration of a single random bitof the decision vector. If this change implies astrictly greater payoff, the firm makes the move.Otherwise, the new vector is disregarded andthe system stays at the original position. Thefirms search for a Nash equilibrium of the sys-tem—that is, when one firm alters a decisionelement, it takes the positions of the other firmsas given.

Mapping the NKC Model to Our Context

As indicated earlier, our interest is in exam-ining how entrant characteristics interact withenvironmental characteristics to explain differ-ences in firm performance. We seek to model theeffect that environmental characteristics relatedto (1) interfirm coupling leading to industry tur-bulence and (2) industry evolution may have onthe relationship between pre-entry experienceand performance. We model differences in diver-sifying entrants and entrepreneurial start-ups


through K, differences in industry turbulencethrough C, and industry life cycle by permittingearly and late entry into the simulation model.Firms are assumed to have a constant size, repre-sented by a constant problem size N (i.e., all firmsare assumed to make the same number of deci-sions), and we exclude selection.

In line with the NK models (Kauffman, 1993,1995) and following the organizational ecologyliterature (Hannan & Freeman, 1977), we defineperformance as organizational fitness—that is,the probability that a given form of organization

persists in a certain environment. This defini-tion is also consistent with survival as the keyperformance measure in technology manage-ment and industry evolution (Agarwal & Gort,1996; Klepper & Simons, 2000). Our analysis fo-cuses on the performance differentials observedduring and at the end of the simulation periodand on the associated dynamics, such as fre-quency of changes or performance variance.

Mapping K to differences in entrant character-istics. We model differences in entrant charac-teristics by differences in the value of the pa-

FIGURE 1Example of Interaction Matrices

Note: The ones in C matrices are randomly distributed within each row of the partition of the size C � 7. The off-diagonalelements of the NK matrices are also randomly distributed within each row. The entries in all matrices are drawn indepen-dently for each C and NK matrix and each simulation run. The firms search by changing elements within their own matrices.


rameter K. This is, in spirit, similar to theassumption that differences in the degree of ver-tical integration correspond to differences in K(Sorenson, 1997). Following the Hannan, Polos,and Carroll (2003a,b) concept of organizationalintricacy and the notion of intrafirm coupling,we distinguish between entrepreneurial start-ups and diversifying entrants on the basis of theinterdependence of their choices. Start-ups havea lower value of K—or less coupled organiza-tional structures—than diversifying entrants.This is consistent with the characterization ofstart-ups as lacking strong idiosyncrasies thatrelate to capabilities suited for other industrycontexts and intradivisional links imposed bythe existence of multiple divisions. Similarly, itis consistent with the characterization of diver-sifying entrants as possessing predefined struc-tures and established routines in other industrycontexts with predetermined ways of resourcedeployment. We further assume that the level ofcoupling within both types of firms remains con-stant throughout the industry life cycle and thatfirms cannot easily adjust or imitate the level ofcoupling that would be most appropriate in agiven environment.3

Since the assumption that diversifying en-trants have more internal links than start-ups iscrucial to our model, we elaborate on how andwhy this may be the case. Given their opera-tions in other industries prior to entry into afocal context, diversifying entrants clearly pos-sess preexisting routines and capabilities. Asubstantial body of corporate strategy literature,hailing back to Penrose (1959), discusses valuecreation through related industry entry as ameans to leverage existing firm capabilities re-lated to technologies, knowledge, and organiza-tional routines (Farjoun, 1998; Markides & Wil-liamson, 1994; Miller, 2006; Teece, Rumelt, Dosi,& Winter, 1994). However, the new industry contexta firm enters often requires new capabilitiesand the combination of new skills, routines, andorganizational demands with extant capabili-ties. This combination of new and old increasescoupling, causing diversifying firms to havehigher Ks when represented in an NKC model.

Holbrook, Cohen, Hounshell, and Klepper(2000) have provided a rich description of the

differences between start-ups and diversifyingentrants in the semiconductor industry. There isa remarkable similarity between our assump-tion and their account of the differences in theearly histories of two diversifying entrants(Sprague Electric and Motorola) and two start-ups (Shockley and Fairchild Semiconductor). Indescribing the actions of Sprague, a producer ofelectrochemical transistors that diversified intosemiconductors, Holbrook et al. (2000) highlightcapabilities in capacitor production and experi-ence in making small electronic components.The existence of links between the new capabil-ities that Sprague developed for semiconductorsand its old electrochemical capabilities is evi-dent from the following statement:

In spite of late 1950s developments using siliconinstead of germanium and photolithographyrather than electrochemistry to make transistors,Sprague Electric stuck with its ceramic-based hy-brid circuits until well into the 1960s, trying tocapitalize on its historical expertise (Holbrook etal., 2000: 1022).

Similarly, Motorola focused on hybrid circuitsthat took advantage of Motorola’s capabilitiesand experience in the use of ceramic materialsand the design of rugged circuits.

In sharp contrast to these diversifying en-trants, which maintained links with existing ca-pabilities and technologies, start-ups Shockleyand Fairchild Semiconductor chose entirely newtechnologies and experimented with alternativematerials and processes. Fairchild Semiconduc-tor broke from tradition entirely to bet on siliconinstead of germanium and a photolithographicinstead of electrochemical process, which led toits invention of the monolithic integrated circuitthat manufactured all components on a singlepiece of silicon. Although the start-ups did ben-efit from what their founders had learned work-ing in related industries, their organizationalframeworks were created anew. As opposed tothe diversifying entrants, who maintained linksto existing technologies and production pro-cesses, the start-ups were able to sever whatwere perceived as constraining links to capabil-ities built to serve other industries.

Mapping C to industry characteristics that im-pact turbulence. Similarly, we model differencesin industry turbulence using the notion of inter-firm coupling. We assume that the exogenousparameter C in Kauffman’s (1995) NKC frame-work controls interfirm interdependence. Since

3 Our assumption is that, on average, imprinting lasts andmost firms retain their initial characteristics until they fail.


the draws in the NKC model are random, inter-firm coupling may result in either positive ornegative changes in payoffs in response to otherfirms’ moves. A high C implies high interdepen-dence resulting either from strong competitivedynamics or from a greater need for interfirmcollaboration. C parsimoniously captures boththe complementary and substitute effects of thecoupled moves of other firms.4 For instance, fre-quent introductions of both new complementaryand substitute products by other firms maytranslate into high turbulence and a frequentneed to change and adapt.

Several broad structural characteristics maycause differences in interfirm couplings and,thus, may serve as empirical proxies for C. Theliterature suggests that one characteristic inparticular—technological intensity—may corre-spond to differences in C, and we expect a pos-itive correlation. The level of technological in-tensity may, ceteris paribus, influence the needfor close links to other firms through collabora-tive relationships. The effect of knowledge spill-overs may also be more pronounced in technol-ogy-intensive industries. At the same time, hightechnological intensity may foster agglomera-tion, employee mobility, and pressures on factormarkets, thus increasing the likelihood of com-petitive links. For example, studies of the semi-conductor industry reveal that in highly techno-logically intensive industries, strong interfirmlinks driven by local proximity, employee mobil-ity, and alliances (Almeida & Kogut 1999; Rosen-kopf & Almeida, 2003) enable knowledge diffu-sion. Similarly, the history of the disk driveindustry provides examples of the competitivepressures firms exert on each other (Chris-tensen, 1997) and the many interfirm links result-ing from employee mobility and entrepreneur-ship (Agarwal et al., 2004; McKendrick, Doner, &Haggard, 2000). Collaborative and competitivecouplings also can be seen as endogenous tothe complexity of the problems posed in an in-dustry. Firms solving more complicated prob-lems also have more coupled interfirm land-scapes; technological intensity can be thoughtof as a proxy for this complexity. Supporting thisconjecture, Hoetker and Agarwal (2007) have

shown that disk drive manufacturers’ ability tobuild on prior innovation depended on whetherthe innovative firms still existed, because theirprivate knowledge was an important comple-ment to the public and codified knowledge.

Other industry characteristics may also affectinterfirm coupling. For instance, capital inten-sity and need for vertical integration likely neg-atively correlate with C. Capital intensitybroadly captures economies of scale and scopeand has an effect opposite that of technologicalintensity. While producing high volumes and/orvertically integrating may increase intrafirmcoupling (Sorenson, 1997), these activities maydecrease interfirm coupling. Similar to Schmal-ensee’s work (1989: 978, 993), which states thathigher capital intensity (low C) results in higherprofits, our model predicts that C and perfor-mance are negatively related. Advertising inten-sity and the potential for product differentiationmay also negatively correlate with C.5 More dif-ferentiated products imply landscapes with lesscoupling with other firms and less competitiveinteraction and, at the same time, greater differ-ences between firms and less need for collabo-ration. Schmalensee (1989: 978) provided evi-dence that, in the consumer goods industries,advertising intensity positively correlates withprofits.

The above structural characteristics can thusbe proxies for the nature and strength of inter-firm linkages. In many cases these proxies arein tandem. For instance, the cement industryhas lower technological intensity, higher capitalintensity, and higher (spatial) differentiationthan semiconductors; we therefore expect ce-ment to have a lower C than semiconductors.

Modeling early versus late entry. We modelearly versus late industry entry by substituting,midway through the simulation, new entrants ofthe same type for half of each firm type and thencomparing the early and late cohorts. Impor-tantly, the early and late entrants differ fromeach other on ability to learn from an estab-lished stock of industry-specific knowledgeprior to their entry into the focal industry.

We assume that early entrants enter with ran-dom innovation. For later entrants, we modellearning by building on the notion that late en-

4 This is analogous to the discussion of K in Rivkin (2000),who describes the difference between Milgrom and Roberts’(1990) definition of “complementarity” and the notion of in-teractions in the NK models.

5 Advertising intensity can be seen as a proxy for productdifferentiation.


try allows them to sample from the industrystock of knowledge—a pool of recombinant in-formation generated by incumbents (firms thatentered in the prior stage). The late entrants canalso combine the information gathered withtheir own random innovation. Late entrants areable to enter at positions that maximize the pay-off of the recombination of the industry stock ofknowledge with their own innovation and thenproceed through local search. Thus, we permitentrants to engage in “offline” search (e.g.,Gavetti, 2005) regarding the decision vector andthe interaction matrix with which they enter.6

Specifically, while early entrants generate boththe decision vector and the interaction matrix atrandom, late entrants have the additionalchoice of copying either or both from a randomlyselected incumbent of the same type. Our as-sumption of imitating an incumbent of the sametype exists to ensure no mismatch in the numberof links (robustness checks reveal that the re-sults are not sensitive to this assumption).Therefore, late entrants evaluate the entry per-formance vector of each of the four options re-sulting from the combination of random or cop-ied decision vector and interaction matrices,and they enter with the option yielding the bestpayoff. After entry, late entrants learn throughlocal search, just as the early entrants do.

To analyze the impact of learning on postentryperformance, we compare outcomes for a ran-dom late entry case with two learning algo-rithms—moderate and strong.7 For moderatelearning the late entrants sample only oncefrom the pool of interaction matrices used by theincumbents, and for strong learning the entrantssample ten times and select the best combina-tion with the copied or generated decision vec-tor.8 To ensure a sufficient recombinant pool, theentrants sample from ten prior runs and thebaseline model is run ten times before the learn-ing simulation starts. Regardless of the learning

algorithm employed, both types of entrants canevaluate the same mix of pre-entry options.

Model Simulations

Like prior researchers, we ran the models withspecific values for N, K, and C. Since the modelis probabilistic and dynamic, many runs wereneeded to ensure the significance of results(�15,000 runs). The simulations were coded inMATLAB 7. Table 2 summarizes and describesthe parameters used in the model.

We represent two types of entrants—diversi-fying entrants with high decision interdepen-dence (Kdivers. � 5) and start-ups with low deci-sion interdependence (Kstart-up � 2). We set thelength of the decision vector to N � 10. Each firmrepresents a cohort of identical firms that areassumed to move through the landscape in uni-son as if performing local search.

To accommodate learning during the latersimulations while enabling model comparisons,we ran all the models with two types but fourfirms, two firms of each type being identical. Inthe learning version of the model, we requiredfour firms to model two types of firms and twoentry periods— early and late. Although wecould simulate the turbulence effects model us-ing only two firms (one of each type), the rela-tionship between the particular parameters cho-sen and the model dynamics may not beinformative for the extended version of themodel. As we briefly discuss below, the turbu-lence is a function not only of the level of inter-firm coupling C but of the number of firmspresent. Thus, we chose to keep the number offirms constant at four as a way of controlling for“industry size” and focusing solely on the effectsof C and K. The firms are assumed to occupydifferent segments of the industry and to inter-act through some of their decisions. The level ofinterfirm coupling is controlled by the parame-ter C. Firms 1 and 3 (both Kstart-up � 2) and firms2 and 4 (both Kdivers. � 5) have the same level ofinternal coupling. To eliminate possible biasesassociated with the order in which firms searchfor their local peaks, we randomized this orderin each step.

Baseline simulations. Firms enter with ran-dom decision vectors and random interactionmatrices and adapt through local search afterthey enter. We obtained results by performing15,000 runs of the model, wherein all randomized

6 Our results should not change substantially if we in-stead assumed that late entrants entered at random posi-tions and then vicariously learned along the way. However,our assumption makes the analysis more transparent andfacilitates discerning the effects of learning.

7 In case of random late entry, the entry decision vector, aswell as the interaction matrix, is randomly generated.

8 The dimension on which the late entrants are more pow-erful is arbitrary (see the section on robustness).


variables, including the NKC landscape, wereredrawn for each run. Figures 2 through 5 pro-vide four sets of information regarding themodel dynamics. Figure 2 shows the mean ab-

solute performance for the two different firmtypes (the lines always show the mean perfor-mance of the two firms of the same type). Figure3 provides the variance of absolute performance

TABLE 2NKC Model Parameter and Assumptions Summary

Parameter DescriptionValue Used inText

UnderlyingAssumptions Robustness Tested

Possible EmpiricalProxies

N Problem size (numberof decisionelements)

10 Fixed firm size, fixedresources

10 Patent-based measures(Fleming & Sorenson,2001), firm size (crudeproxy)

K Number of linkageswithin a firmlandscape (for eachrow of theinteraction matrix),randomlydistributed withininteraction matrix

Kstart-up � 2,Kdivers. � 5

Diversifying entrantsare relativelymore coupledinternally; thedifferences incoupling arepersistent, andeasy adjustmentor imitation is notpossible

{0, . . . , 9} (all combinationsfor differences of 0 to 4between the K of start-up and divers.)

Patent-based measures(Fleming & Sorenson,2001), product designmatrix (Rivkin &Siggelkow, 2007),vertical integration(Sorenson, 1997)

C, � Number of linkagesacross firmlandscapes (C isthe number of rowsin the C matrixwith linkages, and� is the number ofinteractions ineach row of the Cmatrix), randomlydistributed withinthe relevantpartition of the Cmatrix

C � 1, 7, 9,� � 2

Fixed structuralcharacteristics ofan industrydetermine theinterfirm coupling

C � {0, . . . , 10}, � � {2, 3} Technologicalintensity, capitalintensity, andpotential for productdifferentiation(Schmalensee, 1989)

S Number of firms 4 Fixed industry size 4 Industry sizeNumber of

periodsTo observe

performancedifferentials;number of firms“walking” at theend ranges from0% to 60%depending mainlyon K and C

50, 100 Performance notmeasured in the“infinite” timelimit; firms facereal timeconstraints whenoperating indynamicenvironments

15–150

Simulationruns

To obtain 1%significance for themean

15,000 Mean is informativeof the qualitativeperformancedifferentials

15,000 � 20,000

Nature ofsearch

Local searchpostentry, randomentry versus pre-entry learning withdifferent strength,constantoptimization speed(number ofadaptationattempts) and noselection

Randomentry,moderateand stronglearning

Firms have theopporttunity tosampleinformation fromthe environment inmature industries

Different sources ofinformation, differententry periods, differentmixes for the learningalgorithm

Intellectual propertyregime, prevalenceof employeemobility, etc.


across the simulation runs. Figure 4 reports thelevel of turbulence that each type of firm faces.And Figure 5 provides information on the fre-quency of adaptation or change.9

As for the main effects discussed above, weobserve here that several patterns hold regard-

9 The turbulence is calculated as the mean number of deci-sion payoff components �i that are redrawn as a result of themoves of other firms (up to N components may be redrawn ineach step). When calculating the payoff values that change, wehold the position of the focal firm fixed and calculate it after all

other firms make their moves—that is, this turbulence measurecaptures the effect of changes in the coupled decisions. Thefrequency of adaptation is calculated as the mean number ofdecision changes at a given step over all runs. Within each runthe adaptation is coded 1 when the agent changes the decisionat time t � 1 compared to t (the measure starts at time period 2).Both turbulence and adaptation measures are averaged overall 15,000 simulation runs.

FIGURE 2Performance Mean

FIGURE 3Performance Variance


less of the level of C. Figure 2 shows that diver-sifying entrants always outperform start-ups ini-tially. Because of the diversifying entrants’higher internal coupling, their problem space ismore rugged and the slopes are steeper. Theinitial performance improvements are, thus,greater for them. The performance variance di-agram (Figure 3) shows that the variance of per-formance is significantly higher for the start-ups

(note that the variance in mean performancereflects the significance level for the given num-ber of runs and is close to zero). The less coupledstart-ups exhibit higher risk, given more fre-quent changes (Figure 5). Both types of firmschange and optimally adapt conditional on themyopic nature of their search (which is identicalfor both types of firms) and the level of intrafirmcoupling (which is different for the two types).

FIGURE 4Level of Environmental Turbulence

FIGURE 5Probability of Organizational Adaptation


However, the less constrained start-ups find itoptimal to change and adapt more—regardlessof the value of C. The degree of change and ad-aptation is, thus, endogenous to the nature of thelandscapes faced by each firm.10 The diversifyingentrants also face greater turbulence, since theycompete with firms that change more (Figures 4and 5). It is also interesting to note that the perfor-mance differentials are small relative to the abso-lute performance values. This outcome is relatedto the probabilistic nature of the framework and isa general feature of many NK models. It perhapsreveals the fact that relative differences betweenwinners and losers in evolution are rather smallcompared to their absolute fitness.

Effect of turbulence. We proceeded by varyingthe level of C.11 When C � 1, firms operate onalmost separate landscapes, resulting in lowturbulence and minimal impact of the decisionsof other firms on the shape of the landscapeeach firm faces. In this environment of relativelystationary peaks, firms can climb peaks effec-tively. As seen in Figure 2, while diversifyingentrants have an initial lead, start-ups eventu-ally achieve higher performance owing to theirhigher adaptation and potential; the averagepeak on the landscape of a start-up is greaterthan that on the landscape of a diversifyingentrant. However, it takes more steps for thestart-up to discover its peak because it occupiesa smoother space with flatter slopes. Over time,turbulence decreases and converges to zero(Figure 4).

When the level of interfirm coupling increasesto C � 7, the firms face dynamic landscapes. Thedecision of any one firm affects the performancecontribution of the decisions of the other firms.In such an environment the firms sometimes“chase chimeras,” since before they can dis-cover how to climb a peak and solve the prob-lem, the landscape changes its shape—thepeaks become valleys and vice versa—and theproblem changes into a new one. On average,the firms are much farther from the peaks thanthey are in the more stable environments de-picted by C � 1. In such a situation diversifyingentrants fare better, and the start-ups can matchtheir performance only much later. For the diver-

sifying entrants the higher interdependence oftheir decisions partially locks them in and al-lows them to get closer to the peaks along lessconstrained variables. The start-ups need to per-form more steps because they need to climbflatter slopes than the diversifying entrants toreach their (albeit on average higher) peaks. Thestart-ups attempt to adapt excessively to anyenvironmental change since it is myopically op-timal to do so. However, the problem changesbefore they can exploit it, causing them, on av-erage, to be positioned farther from the peaks.The performance achieved by all firms also con-verges to a lower value than in the C � 1 case.Although the additional interactions favor di-versifying entrants over start-ups, the interac-tions prevent the firms from climbing to theirhighest peaks. The interaction and turbulencedestroy some adaptation in the industry thatwould otherwise be achieved through localsearch. The performance variance again de-creases with simulation time but converges at ahigher value. It is also notable that the higherlevel of interfirm coupling dramatically shiftsthe level of turbulence upward; although it de-creases over time, it does not approach zerowithin the simulation time frame. The model’sprediction is consistent with the conventionalwisdom that change and environmental turbu-lence go hand-in-hand, since both types of firmsrespond to the greater turbulence by adaptingmore; however, the start-ups again exhibit ahigher degree of change. The lower frequency ofchange for the diversifying entrants provides aperformance benefit in more turbulent environ-ments.

In our third case we model stronger firm inter-dependence with C � 9, representing a highlycoupled, turbulent environment in which indi-vidual firms’ landscapes shift rapidly and theirabilities to optimize are seriously compromised.The diversifying entrants continue to outperformstart-ups, although both types of firms find theirperformance to be adversely affected by thehigher turbulence. The patterns of variance, tur-bulence, and adaptation are consistent with theC � 7 case but much more pronounced.

Our model thus shows that a decrease in in-terfirm coupling favors entrepreneurial start-upsin the long run. Specifically, we find that diver-sifying entrants—who tend to undertake orga-nizational changes less frequently than start-

10 We are grateful to the special issue editor for helping usframe the model.

11 The parameter � is set to 2 for all runs.


ups—perform persistently better when theenvironment is more turbulent.

Proposition 1: The likelihood that en-trepreneurial start-ups will outper-form diversifying entrants in the longrun decreases with the level of inter-firm coupling (or degree of environ-mental turbulence).

We note that environmental turbulence is alsoa function of the interplay between intrafirm (K)and interfirm coupling (C). Consider the case ofC � 7 and an increase in the internal coupling Kof the diversifying entrants from Kdivers. � 5 toKdivers. � 6. This interplay causes the diversify-ing entrants to achieve higher performance.However, as a consequence of fewer organiza-tional changes by the diversifying entrants(Kdivers. � 6), the environment facing the start-ups (Kstart-up � 2) is also stabilized. Notably, thestart-ups benefit more, with the performance dif-ferential increasing and the tipping point shift-ing to the left (by twelve periods). This high-lights the fact that an increase in internalcomplexity as a response to environmental tur-bulence may help increase performance butmay also disproportionately benefit less cou-pled firms. As with the decrease in C above, theincrease in K allows achieving a better fit inturbulent environments at the price of beingovertaken by less coupled firms sooner. A simi-lar situation emerges when we increase the in-ternal coupling of the start-ups (Kstart-up � 3).They benefit and achieve higher fit than in thecase above (Kstart-up � 2), but the diversifyingentrants (Kdivers. � 5) benefit as well. The in-crease in the coupling of start-ups as a responseto environmental turbulence is beneficial in theshort term for start-ups, but their performancedifferential with diversifying firms (Kdivers. � 5)decreases. Furthermore, all firms achievehigher performance than in the above cases(Kstart-up � 2, Kdivers. � 5 and Kstart-up � 2,Kdivers. � 6), and the tipping point shifts to anearlier period (period 17 as opposed to 20) thanwhen Kstart-up � 2, Kdivers. � 6. The internal cou-pling is now better distributed collectively, withboth types of firms changing less, making theenvironment more stable and allowing the firmsto get closer to their local optima.

Finally, since the number of firms can alsodrive turbulence, we experimented with sequen-tial firm entry. With interfirm coupling C held

constant, an increase in the number of firmsactivates more links and increases turbulence.A sufficient increase in the number of firmscauses a reversal of performance advantagefrom start-ups to diversifying entrants.

Modeling Late Entry and the Effect of Learning

We model late entry by choosing C � 7 andlate entry period t � 35.12 In period 35 of thesimulation, entrants (one of each type) replacetwo of the four incumbent firms.

Figure 6 provides information on the meanfitness of each of the four types of firms repre-sented by the interaction of entry timing andpre-entry experience, under the three learningconditions. For the no learning (random innova-tion) case, firms that enter late do not captureany of the external knowledge, since both start-ups (Kstart-up � 2) and diversifying entrants(Kdivers. � 5) generate their interaction matricesand initial decision vectors at random. Not sur-prisingly, the converged values are similar tothe results in Figure 2. The performance of lateentrants with random entry replicates the per-formance pattern of early entrants. However,late entrants are disadvantaged by their delayin entry and have a lower performance thanearly entrants.

In the moderate learning algorithm the perfor-mance patterns depicted in Figure 6 deviatefrom the baseline case depicted in Figure 2. Theperformance of the late start-up entrant con-verges at a significantly higher value than thatof the late diversifying entrant. Choosing frommultiple entry options permits the late start-upto enter in a significantly improved initial posi-tion than in the random entry case. On the otherhand, the late diversifying firm is not able tofully capitalize on the available stock of knowl-edge and ends up as the poorest performer. Wenote that this occurs despite the fact that bothtypes of entrants evaluate the same mix of pre-entry options. The implication is that the addi-tional internal coupling of the late diversifyingentrant results in a situation in which it is not

12 We assume that mature industries are not in the mostturbulent regime. As Figure 5 shows, the level of turbulencedecreases over time even for a constant C. The model at thetime of late entry represents an industry where initial peakturbulence has diminished and accumulated pertinentknowledge that the new entrants can exploit is present.


optimal for this entrant to fully internalize andincorporate the information available throughthe learning options. To establish the argumentfurther, we examine the performance of the lateentrants using the strong learning algorithm.Remarkably, the gap between the late diversi-fying and late start-up entrant is much widerthan the gap under moderate learning. Facing aless coupled internal environment, the latestart-up is able to capitalize on the improvedlearning mechanism and gain a stronger lead.

Our results also show that the late start-upbecomes the best performer in the moderate andstrong learning case. The late diversifying en-trant is either the worst or better than the earlydiversifying entrant. The ordering between theearly and late entrants is, thus, sensitive to thelearning algorithm employed. Additional simu-lations have shown that if we make the lateentrants even more powerful, the late diversify-ing entrants also outperform both types of earlyentrants. However, the increase in learning dif-ferentially favors the late start-up and, with suf-ficiently strong learning, it is likely to becomethe best performer. Increase in learning strengththerefore has the opposite effect of increase ininterfirm coupling. We can now put forward ournext propositions.

Proposition 2: If learning from a priorstock of external knowledge is possi-ble, entrepreneurial start-ups will out-perform diversifying entrants.

Proposition 3: The likelihood entrepre-neurial start-ups will outperform di-versifying entrants (among late en-trants) increases with the strength ofthe learning mechanisms the entrantscan employ.

Proposition 4: The performance order-ing between early and late entrantswill depend on the strength of thelearning mechanisms employed bythe late entrants. With sufficientlystrong learning, late entrepreneurialstart-ups are likely to become the best-performing cohort.

Robustness Analysis

To investigate the sensitivity of our resultsto the parameter space, we ran simulations formultiple parameter value combinations of K,C, and �. We report a representative subset ofthese simulations in Tables 3 and 4. The ob-served patterns are consistent with our modelpredictions with one boundary condition. Forvery low values of K (e.g., Kdivers. � 2, Kstart-up �1), diversifying firms outperform start-upseven when C is very low (e.g., C � 0). Thisexception relates to the underlying property ofthe NK landscape implying that, for low valuesof C, the relationship between K and the pay-off values takes an inverted-U shape. As de-

FIGURE 6Late Entry, Mean Performance (Entry Period 35, Performance Observed in Period 100, C � 7)


picted in Figure 7, the payoffs peak at aroundK � 3 for local search and N � 10, C � 0. For ouranalysis, the implication is that for K � 3 di-versifying entrants have a universal advan-tage over start-ups, since the payoff values areincreasing in K up to this point.

Our learning algorithm specification relieson a completely probabilistic structure of theNKC model without explicitly modeling simi-larity between the sublandscapes of the indi-vidual firms (e.g., Gavetti, 2005). The learningalgorithm can thus be seen as a way of con-sidering multiple options before entering, with

random innovation and the imitated decisionvector having equal likelihoods of success.Modeling similarity may generate additionalinsights but would not change our proposi-tions, which are related to very fundamentalproperties of the NKC model. As additionalchecks, we considered several alternativemixes of learning algorithms; for example, wevaried learning strength and type of informa-tion that is being sampled multiple times inthe strong version of learning, and we alsoassumed that late entrants could learn fromexiting incumbents, from incumbents of both

TABLE 3Model Robustness: Effect of Turbulence

Vary K and CKstart-up, Kdivers.

Kdivers. � Kstart-up � 2 Kdivers. � Kstart-up � 3 Kdivers. � Kstart-up � 4

1, 3 3, 5 5, 7 7, 9 0, 3 2, 5 4, 7 6, 9 0, 4 2, 6 4, 8

C � 0Start-up � Divers. (period 1) �0.02 �0.02 �0.02 �0.01 �0.04 �0.03 �0.03 �0.02 �0.05 �0.04 �0.04Start-up � Divers. (period 5) �0.06 �0.02 0 0.02 �0.12 �0.05 0 0.02 �0.13 �0.04 0Start-up � Divers. (period 50) �0.05 0.04 0.06 0.07 �0.14 0.03 0.08 0.1 �0.12 0.06 0.11Start-up � Divers. �50 11 6 3 �50 16 6 4 �50 12 6Still Walking 1 1% 1% 1% 0% 1% 1% 1% 0% 1% 1% 1%Still Walking 3 2% 2% 2% 1% 2% 3% 2% 1% 2% 2% 2%

C � 7Start-up � Divers. (period 1) �0.03 �0.02 �0.02 �0.02 �0.05 �0.03 �0.03 �0.03 �0.06 �0.05 �0.04Start-up � Divers. (period 5) �0.05 �0.02 �0.01 0.01 �0.1 �0.04 �0.01 0 �0.11 �0.04 �0.01Start-up � Divers. (period 50) �0.04 0.01 0.03 0.04 �0.1 0 0.04 0.06 �0.1 0.02 0.06Start-up � Divers. �50 17 7 4 �50 32 9 5 �50 20 8Still Walking 1 24% 19% 14% 9% 25% 22% 15% 10% 24% 20% 14%Still Walking 3 54% 44% 32% 22% 56% 49% 35% 25% 54% 45% 33%

C � 10Start-up � Divers. (period 1) �0.03 �0.03 �0.02 �0.01 �0.05 �0.04 �0.03 �0.03 �0.07 �0.05 �0.04Start-up � Divers. (period 5) �0.04 �0.02 �0.01 0 �0.08 �0.04 �0.02 �0.01 �0.1 �0.05 �0.02Start-up � Divers. (period 50) �0.04 �0.02 0 0.02 �0.09 �0.03 0 0.02 �0.09 �0.03 0.01Start-up � Divers. �50 �50 14 5 �50 �50 �50 7 �50 �50 15Still Walking 1 36% 32% 27% 21% 37% 33% 29% 22% 36% 32% 27%Still Walking 3 72% 66% 56% 46% 73% 67% 60% 48% 71% 67% 58%

Vary � and C� 2 2 2 2 2 2 3 3 3 3 3C 0 2 4 6 8 10 2 4 6 8 10Kstart-up, Kdivers. 2, 5 2, 5 2, 5 2, 5 2, 5 2, 5 2, 5 2, 5 2, 5 2, 5 2, 5

Start-up � Divers. (period 1) �0.03 �0.03 �0.04 �0.03 �0.03 �0.04 �0.03 �0.04 �0.04 �0.03 �0.04Start-up � Divers. (period 5) �0.05 �0.03 �0.04 �0.04 �0.04 �0.04 �0.04 �0.04 �0.04 �0.04 �0.04Start-up � Divers. (period 50) 0.03 0.03 0.01 0.01 �0.01 �0.03 0.02 0.02 0 �0.01 �0.04Start-up � Divers. 16 16 17 24 �50 �50 16 20 23 �50 �50Still Walking 1 1% 3% 9% 15% 25% 33% 3% 11% 21% 32% 40%Still Walking 3 3% 9% 22% 38% 55% 68% 8% 27% 48% 64% 75%

Note: The cells report the difference in performance between start-ups and diversifying entrants (Start-up � Divers.) in agiven period or the first period when start-ups outperform diversifying entrants (Start-up � Divers.). The measures StillWalking 1 and 3 were calculated as the proportion of runs where the agent moved in the last one and three periods averagedover the four agents.


TABL

E4

Late

Entr

yM

odel

Rob

ustn

ess

Var

yK

,C,a

ndLe

arni

ngSt

reng

thK

sta

rt-u

p,K

div

ers

.

Lear

ning

(Ran

dom

,Wea

k,St

rong

)

Kd

ive

rs.

�K

sta

rt-u

p�

1K

div

ers

.�

Kst

art

-up

�3

0,1

R0,

1W

0,1

S4,

5R

4,5

W4,

5S

8,9

R8,

9W

8,9

S0,

3R

0,3

W0,

3S

3,6

R3,

6W

3,6

S6,

9R

6,9

W6,

9S

C�

0(e

ntr

y�

35)

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d35

)�

0.02

�0.

18�

0.34

�0.

010.

000.

01�

0.01

0.11

0.22

�0.

05�

0.20

�0.

35�

0.03

0.01

0.02

�0.

020.

120.

25La

test

art

-up

�La

ted

ive

rs.

(per

iod

40)

�0.

06�

0.18

�0.

310.

000.

010.

020.

010.

070.

15�

0.12

�0.

19�

0.30

�0.

020.

030.

050.

020.

100.

19La

test

art

-up

�La

ted

ive

rs.

(per

iod

100)

�0.

09�

0.17

�0.

270.

020.

030.

030.

030.

050.

09�

0.14

�0.

19�

0.24

0.07

0.09

0.10

0.10

0.12

0.16

Late

sta

rt-u

p�

Late

div

ers

.�

100

�10

0�

100

4336

3639

3636

�10

0�

100

�10

044

3636

3936

36La

test

art

-up

�E

arl

y sta

rt-u

p99

88�

100

�10

060

4598

5540

�10

0�

100

�10

0�

100

5744

9862

43La

test

art

-up

�E

arl

y div

ers

.�

100

�10

0�

100

7053

4258

4637

�10

0�

100

�10

060

4739

5042

36S

till

Wa

lkin

g3

0%0%

0%0%

0%0%

0%0%

0%0%

0%0%

0%0%

0%0%

0%0%

C�

7(e

ntr

y�

35)

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d35

)�

0.02

�0.

09�

0.16

�0.

010.

000.

00�

0.01

0.03

0.06

�0.

05�

0.12

�0.

17�

0.03

�0.

020.

00�

0.03

0.03

0.08

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d40

)�

0.05

�0.

09�

0.12

�0.

010.

000.

000.

010.

020.

03�

0.10

�0.

12�

0.14

�0.

020.

000.

020.

010.

030.

06La

test

art

-up

�La

ted

ive

rs.

(per

iod

100)

�0.

05�

0.07

�0.

110.

010.

010.

010.

020.

030.

03�

0.09

�0.

11�

0.11

0.03

0.04

0.04

0.06

0.07

0.07

Late

sta

rt-u

p�

Late

div

ers

.�

100

�10

0�

100

4641

3838

3636

�10

0�

100

�10

048

4238

4036

36La

test

art

-up

�E

arl

y sta

rt-u

p66

7070

�10

081

36�

100

9336

7070

66�

100

7236

�10

0�

100

36La

test

art

-up

�E

arl

y div

ers

.�

100

�10

0�

100

7554

3654

4436

�10

0�

100

�10

059

4436

4439

36S

till

Wa

lkin

g3

59%

59%

58%

37%

37%

37%

16%

15%

15%

55%

55%

55%

39%

38%

38%

21%

20%

20%

C�

10(e

ntr

y�

35)

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d35

)�

0.01

�0.

05�

0.07

�0.

01�

0.02

�0.

01�

0.01

0.00

0.02

�0.

05�

0.08

�0.

10�

0.03

�0.

03�

0.02

�0.

02�

0.01

0.01

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d40

)�

0.03

�0.

05�

0.05

�0.

01�

0.02

0.00

0.00

0.00

0.00

�0.

08�

0.09

�0.

09�

0.02

�0.

02�

0.01

0.00

0.01

0.01

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d10

0)�

0.04

�0.

04�

0.04

0.00

0.00

�0.

010.

020.

010.

01�

0.09

�0.

09�

0.09

�0.

01�

0.01

�0.

010.

020.

020.

02La

test

art

-up

�La

ted

ive

rs.

�10

0�

100

�10

054

4540

4136

36�

100

�10

0�

100

�10

0�

100

�10

042

3936

Late

sta

rt-u

p�

Ea

rly s

tart

-up

3836

3642

3636

4436

3648

3636

4136

3637

3636

Late

sta

rt-u

p�

Ea

rly d

ive

rs.

�10

0�

100

�10

054

3636

3936

36�

100

�10

0�

100

�10

036

3643

3636

Sti

llW

alk

ing

375

%75

%75

%63

%63

%63

%40

%40

%40

%73

%72

%73

%63

%63

%64

%47

%47

%47

%

Va

ryen

try

per

iod

,C,a

nd

lea

rnin

gst

ren

gth

C0

00

00

07

77

77

710

1010

1010

10E

ntr

yp

erio

d15

1515

5555

5515

1515

5555

5515

1515

5555

55K

sta

rt-u

p,K

div

ers

.2,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

52,

5

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d35

)�

0.04

0.00

0.02

�0.

030.

000.

02�

0.03

�0.

02�

0.01

�0.

04�

0.02

�0.

01�

0.04

�0.

03�

0.02

�0.

03�

0.03

�0.

03La

test

art

-up

�La

ted

ive

rs.

(per

iod

40)

�0.

040.

030.

05�

0.03

0.02

0.05

�0.

03�

0.01

0.00

�0.

04�

0.01

0.01

�0.

04�

0.04

�0.

03�

0.04

�0.

04�

0.04

Late

sta

rt-u

p�

Late

div

ers

.(p

erio

d10

0)0.

030.

060.

090.

040.

060.

090.

010.

020.

030.

000.

010.

03�

0.03

�0.

03�

0.03

�0.

03�

0.03

�0.

03La

test

art

-up

�La

ted

ive

rs.

3216

1669

5656

4226

2178

6959

�10

0�

100

�10

0�

100

�10

0�

100

Late

sta

rt-u

p�

Ea

rly s

tart

-up

�10

026

16�

100

7563

9419

16�

100

8256

1916

1663

5656

Late

sta

rt-u

p�

Ea

rly d

ive

rs.

4724

1689

7061

7332

16�

100

7760

�10

0�

100

16�

100

�10

056

Sti

llW

alk

ing

30%

0%0%

1%1%

1%43

%43

%43

%46

%46

%45

%71

%67

%70

%67

%70

%66

%

Not

e:T

he

cell

sre

por

tth

ed

iffe

ren

cein

per

form

an

ceb

etw

een

dif

fere

nt

cate

gor

ies

ofen

tra

nts

ina

giv

enp

erio

dor

the

firs

tp

erio

dw

hen

sta

rt-u

ps

outp

erfo

rmot

her

entr

an

ts.T

he

mea

sure

Sti

llW

alk

ing

3w

as

calc

ula

ted

as

the

pro

por

tion

ofru

ns

wh

ere

the

ag

ent

mov

edin

the

last

thre

ep

erio

ds

ave

rag

edov

erth

efo

ur

ag

ents

.


types, or from only the opposite type. Allyielded identical performance dynamics.

Our model explicitly excludes the effect ofpopulation variance and selection on perfor-mance. Per March (1991), while only mean per-formance matters when comparing a randomdiversifying entrant to a random start-up, whencomparing the best firms in each populationwith each other, variance grows in importanceas the number of firms increases. To illuminatethis problem more formally, we grouped the re-sults of independent simulation runs into popu-lations and then selected best performers withineach group. As in March (1991), greater varianceof the start-ups has a positive impact on theranking of extreme performers of this group vis-a-vis the diversifying entrant population, andsuch benefit increases with population size.However, we found that, consistent with Propo-sition 1, an increase in C always causes a re-duction in the performance of start-ups relativeto that of diversifying entrants. Since the vari-ance of start-ups exceeds the variance of diver-sifying entrants, measuring performance usingbest performers shifts the tipping point in favorof the start-ups, but it is unlikely to change thequalitative predictions of our model.

The notion of selection is closely related.Many underperforming firms are selected outbefore they can match and overtake the betterperformers. If selection operates strongly on thelower tail of the distribution, the higher variance

is both a blessing and a curse for the start-ups.It shifts the tipping point to the left, but only forthe surviving start-ups. Since their lower tail isthicker than that of the diversifying entrants,start-ups have higher hazard rates, consistentwith the empirical literature (e.g., Carroll &Khessina, 2005). Importantly, we found that,across population sizes, the performance of thebest start-ups temporarily fell short of the per-formance of the best diversifying entrants. Thissuggests that the mean is informative about thedifferences between the two types of firms. Al-though incorporation of both variance and se-lection likely affects the tipping points, the qual-itative effects of C and the different types oflearning are likely to remain unchanged.13

DISCUSSION AND CONCLUSION

Strategy, organizational studies, and entre-preneurship scholars have all emphasized theimportance of studying how firm characteristicsat the time of industry entry may have enduringeffects on performance, and pre-entry experi-ence has been identified as a critical determi-nant of firm success. However, the empirical ev-idence on performance differentials between

13 Whether the points would shift to the left or right de-pends on the number of factors, such as the strength ofselection pressures, population sizes, and the speed of se-lection relative to adaptation.

FIGURE 7Absolute NK Performance for N � 10 As a Function of K, C � 0


entrepreneurial start-ups and diversifying en-trants is mixed, partly because of trade-offs inthe theoretical mechanisms that underlie suchdifferentials. The overarching question in thispaper is whether contingent conditions canswitch the performance advantage from diversi-fying entrants to start-ups. Using an NKC model,we highlighted industry turbulence and life cy-cle stages as contingencies that may explainthe performance differentials.

Our model predictions conform to the empiri-cal patterns established in the literature and,importantly, provide a potential rationale for thecontrary findings identified in the literature re-view section (Table 1). Our model replicates theunequivocal empirical finding that diversifyingentrants have a short-run performance advan-tage (Agarwal, 1997; Carroll et al., 1996; Klepper,2002a,b; Klepper & Simons, 2000; Mitchell, 1991).Further, in keeping with studies of organization-al change (Haveman, 1992; Sastry, 1997; Sine etal., 2006) and organizational ecology (e.g., Al-drich, 1990; Brittain & Freeman, 1980; Hannan &Freeman, 1977), the model predicts that organi-zational change in response to environmentalshifts is not always beneficial. In line withKhessina (2002) and Khessina and Carroll (2008),our model predicts that start-ups change morefrequently and face greater performance variance.

The model also provides an answer for howenvironmental turbulence impacts the perfor-mance differential between start-ups and diver-sifying entrants. The main effect of turbulence isdetrimental to firm performance for both types ofentrants, yet increases in turbulence are moreadvantageous for diversifying entrants than forstart-ups. Even though both types of entrantsadapt more frequently in more turbulent envi-ronments, the effect is stronger for diversifyingentrants, consistent with studies emphasizingthat moderate levels of structures are best indynamic markets (Brown & Eisenhardt, 1997). Di-versifying entrants, by definition, have exhib-ited the ability to adapt to new market opportu-nities by entering a new industry, and our modelhighlights that in more turbulent environmentstheir coupling provides them with additionalstability benefits that the less coupled entrepre-neurial start-ups lack. Thus, for instance, whendiversifying entrants have access to accumu-lated process innovation in related industries,the coupling provides long-term benefits, whichmay have played an important role in some in-

dustries like television and penicillin (Klepper &Simons, 2000). Similarly, Mitchell’s (1991) findingof diversifying entrant advantage due to strongcomplementary assets may be linked to the ben-eficial effects of coupling in turbulent industries.

Our model also provides a potential resolu-tion to the conflicting findings on performancedifferentials due to pre-entry experience basedon entry timing. The key question that we an-swer is whether the ability to learn from estab-lished industry-specific knowledge advantagesstart-ups over diversifying entrants. Our modelpredicts that this is indeed the case: diversifyingentrants retain their long-term advantage in in-dustry environments in which knowledge spill-overs are constrained and learning is limited.This may explain why diversifying entrants inthe television receiver industry (Klepper & Si-mons, 2000) and in new markets of medical di-agnostic imaging (Mitchell, 1991) enjoyed a“dominance by birthright.” In other industry con-texts start-ups may benefit from some learningpotential, resulting, for instance, in the observedconvergence of performance between the twotypes of firms in the automobile and tire indus-tries (Carroll et al., 1996; Klepper, 2002a).

Importantly, our model predicts that later-stage start-ups can outperform early-stage di-versifying entrants in instances of high learningpotential. This prediction is consistent with theempirical evidence from industries where em-ployee mobility and entrepreneurship havebeen identified as critical for learning (e.g.,Agarwal et al., 2004; Klepper & Sleeper, 2005).14

We note that an important condition for such areversal is that the mature stages of the industryhave low to moderate levels of turbulence. Highturbulence implies that the landscape changesrapidly, thus reducing the value and usefulnessof learning. If an environment is persistentlyturbulent as a result of denser interfirm link-ages, the usefulness of the learning algorithmrapidly deteriorates.15

14 Additional effects not currently modeled that may affectperformance differentials between early versus late en-trants by favoring early entry are the increasing returnsassociated with technological adoption.

15 For instance, when we model the case of C � 9 and amoderate-strength learning algorithm, the ability of late en-trants to learn does not help. The effect of turbulence over-powers the effect of learning, and the competencies arequickly destroyed. Although the late entrants enter at betterpositions than in the random entry case, their advantage is


Limitations and Future Research

Our study has several limitations. We as-sumed that interfirm coupling (C) and the inter-dependence of internal firm decisions (K) areexogenous and constant throughout the simula-tion. It is possible that, with the establishment ofa dominant design and a shift from product toprocess innovation (Anderson & Tushman, 1990;Gort & Klepper, 1982), some decisions acrossfirms are decoupled or that technological inten-sity decreases as an industry matures. Firmsmay also become more internally coupled as aresponse to an initially turbulent environmentand as a way to manage their growth. Asidefrom unforeseen interactions, such dynamicsshould magnify the proposed effects; both thedecrease in C and the increase in K of incum-bents will lead to a stronger effect of learning,with more positive benefits accruing to less cou-pled firms. Further, our assumption that a firm’ssize and the niche width it faces are constantputs more emphasis on the positive aspects ofinterfirm interactions.16

In addition to the avenues for future researchthat the limitations of our study open, severalfruitful possibilities stem from our results. Forinstance, future in-depth examination of the rel-ative importance of learning for late entrantsand of the contingencies that impact this rela-tionship is needed. Further, the implication ofour model that the performance differential be-tween late start-ups and late diversifying en-trants should be larger in industries wherelearning is easy—in industries with weak pro-tection of intellectual property rights and fre-quent employee mobility—merits empirical in-vestigation. The study may also stimulate morestudies on the long-term performance effects ofpre-entry firm characteristics through its em-phasis on contingencies.

Contributions

Multiple research streams have examined thequestion of whether pre-entry experience resultsin persistent differences in performance. Thefindings have been mixed. Strategy scholarshave found support for diversifying entrant ad-vantages (Klepper & Simons, 2000; Mitchell,1991), entrepreneurship scholars have shownperformance benefits for start-ups (Cooper, Wil-lard, & Woo, 1986; Levinthal & March, 1993; Levitt& March, 1988; March, 1991), and organizationalecologists have found convergence in long-runadvantages (Carroll et al., 1996; Hannan et al.,1998). We contribute to these literature streamsby specifying the contingencies in the perfor-mance differentials due to pre-entry experience.Our modeling approach allows us to abstractfrom resource endowment differences and to fo-cus on the trade-offs that result from differencesin the underlying structures of the two types offirms (Hannan & Freeman, 1984). Further, ourstudy explicitly brings the notion of learningfrom an external stock of knowledge into thediscussion on entry. Late start-ups can overcomethe disadvantage due to lack of their own pre-history (Helfat & Lieberman, 2002) by tapping theindustry-specific stock of knowledge. Thus, ourstudy highlights the potential compensation forinternal lack of knowledge by external sources.

The paper also contributes to the literature onNK modeling. To our knowledge, this is the firstsimulation study that uses the coevolutionaryNKC model to show industry evolution with stra-tegic firm interaction. Our work is similar inspirit to Sorenson’s empirical work (1997, 2003)on the relationship between vertical integrationand performance contingent on environmentalturbulence. Sorenson tested Kauffman’s (1995)original predictions by assuming that verticallyintegrated firms are more complex. Although thecontext is different, our paper contributes byrefining the theory used in the empirical studies.

Finally, our model contributes to the organi-zational ecology literature (Brittain & Freeman,1980; Hannan & Freeman, 1977) by building amodel of the dynamics between generalists andspecialists using coupling (as opposed to nichewidth) as a primitive and modeling factors likeinertia and change as endogenous to the level ofcoupling. If one assumes that links in the NKmodel are analogous to intricacy in models ofcascading organizational change (Hannan et

not built on. Only the relative stability arising from internalcoupling matters, and the diversifying entrants dominate.

16 We also note that although our prediction that diversi-fying entrants always outperform start-ups in the short runcan be explained by size (or resource endowment) differ-ences, such differences do not explain why start-ups outper-form diversifying entrants in less turbulent environments; itis theoretically unclear why endowment effects are condi-tional on environmental conditions without resorting to cou-pling.


al., 2003a,b), we contribute to this stream of or-ganizational ecology by emphasizing the bene-fits of intricacy.

The purpose of our study was to refine thetheory of industry evolution through applicationof a complexity simulation model. We have pro-vided a framework that helps to explain themixed empirical evidence on the performancedifferentials between diversifying and start-upindustry entrants. Our approach emphasizes theroles of environmental turbulence and learningpotential in later stages of the industry life cycleas potential contingency factors. Our modelhighlights that the relative success of entrantsdepends on the entrants’ ability to capture priorknowledge, the strength of intrafirm linkages,and the strength of firms’ interactions. Althoughindustry turbulence favors diversifying entrants,start-ups can capture greater benefits from ac-cumulated industry-specific knowledge. Thus,the relative performance of firms based on pre-entry experience is conditional on the environ-mental conditions that favor particular underly-ing mechanisms over others.

REFERENCES

Abernathy, W. J., & Utterback, J. M. 1978. Patterns of indus-trial innovation. Technology Review, 14: 39–48.

Agarwal, R. 1997. Survival of firms over the product life cycle.Southern Economic Journal, 63: 571–583.

Agarwal, R., Echambadi, R., Franco, A. M., & Sarkar, M. B.2004. Knowledge transfer through inheritance: Spinoutgeneration, development, and survival. Academy ofManagement Journal, 47: 501–522.

Agarwal, R., & Gort, M. 1996. The evolution of markets andentry, exit and survival of firms. Review of Economicsand Statistics, 78: 489–498.

Agarwal, R., Sarkar, M. B., & Echambadi, R. 2002. The condi-tioning effect of time on firm survival: An industry lifecycle approach. Academy of Management Journal, 45:971–994.

Aldrich, H. A. 1990. Using an ecological perspective to studyorganizational founding rates. Entrepreneurship Theoryand Practice, 14: 7–24.

Almeida, P., & Kogut, B. 1999. Localization of knowledge andthe mobility of engineers in regional networks. Manage-ment Science, 45: 905–917.

Anderson, P., & Tushman, M. L. 1990. Technological discon-tinuities and dominant designs: A cyclical model oftechnological change. Administrative Science Quar-terly, 35: 604–634.

Audretsch, D. B. 1997. Technical regimes, industrial demog-raphy and the evolution of industrial structures. Indus-trial and Corporate Change, 6: 49–82.

Barnett, W. P., & Freeman, J. 2001. Too much of a good thing?Product proliferation and organizational failure. Orga-nization Science, 12: 539–558.

Bayus, B., & Agarwal, R. 2007. Product technology strategiesand firm survival: The personal computer industry 1974–1994. Working paper, University of Illinois, Champaign.

Brittain, J., & Freeman, J. H. 1980. Organizational prolifera-tion and density dependent selection. In J. Kimberly &M. Miles (Eds.), Organizational life cycles: 291–338. SanFrancisco: Jossey-Bass.

Brown, S. L., & Eisenhardt, K. M. 1997. The art of continuouschange: Linking complexity theory and time-paced evo-lution in relentlessly shifting organizations. Administra-tive Science Quarterly, 42: 1–34.

Bryman, A. 1997. Animating the pioneer versus late entrantdebate: An historical case study. Journal of Manage-ment Studies, 34: 415–438.

Burns, T., & Stalker, G. M. 1961. The management of innova-tion. London: Tavistock.

Carroll, G., Bigelow, L., Seidel, M., & Tsai, L. 1996. The fatesof de novo and de alio producers in the American auto-mobile industry: 1885–1981. Strategic Management Jour-nal, 17: 117–137.

Carroll, G. R., & Hannan, M. T. 2000. The demography ofcorporations and industries. Princeton, NJ: Princeton Uni-versity Press.

Carroll, G. R., & Khessina, O. M. 2005. The ecology of entre-preneurship. In S. A. Alvarez, R. Agarwal, & O. Sorenson(Eds.), Handbook of entrepreneurship research: Disci-plinary perspectives: 167–200. New York: Springer.

Christensen, C. M. 1997. The innovator’s dilemma: When newtechnologies cause great firms to fail. Boston: HarvardBusiness School Press.

Cooper, A. C., Willard, G. E., & Woo, C. Y. 1986. Strategies ofhigh-performing new and small firms: A reexaminationof the niche concept. Journal of Business Venturing, 1:247–260.

Davis, J. P., Eisenhardt, K. M., & Bingham, C. B. 2007. Com-plexity theory, market dynamism and the strategy ofsimple rules. Working paper, Massachusetts Institute ofTechnology, Boston.

Eisenhardt, K. M., & Martin, J. A. 2000. Dynamic capabilities:What are they? Strategic Management Journal, 21: 1105–1121.

Eisenhardt, K. M., & Tabrizi, B. N. 1995. Accelerating adaptiveprocesses: Product innovation in the global computerindustry. Administrative Science Quarterly, 40: 84–110.

Ethiraj, S. K., & Levinthal, D. 2004. Modularity and innovationin complex systems. Management Science, 50: 159–173.

Farjoun, M. 1998. The independent and joint effects of theskill and physical bases of relatedness. Strategic Man-agement Journal, 19: 611–630.

Fleming, L., & Sorenson, O. 2001. Technology as a complexadaptive system: Evidence from patent data. ResearchPolicy, 30: 1019–1039.

Fleming, L., & Sorenson, O. 2004. Science as a map in tech-


nological search. Strategic Management Journal, 25:909–928.

Forrester, J. W. 1961. Industrial dynamics. Cambridge, MA:MIT Press.

Freeman, J. 1990. Ecological analysis of semiconductor firmmortality. In J. V. Singh (Ed.), Organizational evolution:New directions: 53–77. Newbury Park, CA: Sage.

Gavetti, G. 2005. Cognition and hierarchy: Rethinking themicrofoundations of capabilities’ development. Organi-zation Science, 16: 599–617.

Gort, M., & Klepper, S. 1982. Time paths in the diffusion ofproduct innovations. Economic Journal, 92: 630–653.

Hannan, M. T., Carroll, G. R., Dobrev, S. D., & Han, J. 1998.Organizational mortality in European and American au-tomobile industries. Part I: Revisiting the effects of ageand size. European Sociological Review, 14: 279–302.

Hannan, M. T., & Freeman, J. 1977. The population ecology oforganizations. American Journal of Sociology, 82: 929–964.

Hannan, M. T., & Freeman, J. 1984. Structural inertia andorganizational change. American Sociological Review,49: 149–164.

Hannan, M. T., & Freeman, J. 1988. The ecology of organiza-tional mortality: American labor unions, 1836 –1985.American Journal of Sociology, 94: 25–52.

Hannan, M. T., Polos, L., & Carroll, G. R. 2003a. Cascadingorganizational change. Organization Science, 14: 463–482.

Hannan, M. T., Polos, L., & Carroll, G. R. 2003b. The fog ofchange: Opacity and asperity in organizations. Admin-istrative Science Quarterly, 48: 399–432.

Haveman, H. A. 1992. Between a rock and a hard place:Organizational change and performance under condi-tions of fundamental environmental transformation. Ad-ministrative Science Quarterly, 37: 48–75.

Helfat, C. E., & Lieberman, M. 2002. The birth of capabilities:Market entry and the importance of pre-history. Indus-trial and Corporate Change, 11: 725–760.

Hoetker, G., & Agarwal, R. 2007. Death hurts, but it isn’t fatal:The postexit diffusion of knowledge created by innova-tive companies. Academy of Management Journal, 50:446–467.

Holbrook, D., Cohen, W. M., Hounshell, D. A., & Klepper, S.2000. The nature, sources, and consequences of firmdifferences in the early history of the semiconductorindustry. Strategic Management Journal, 21: 1017–1041.

Kauffman, S. A. 1993. The origins of order: Self-organizationand selection in evolution. Oxford: Oxford UniversityPress.

Kauffman, S. A. 1995. At home in the universe. Oxford: OxfordUniversity Press.

Khessina, O. M. 2002. Effects of entry mode and incumbencystatus on the rates of firm product innovation in theworldwide optical disk drive industry, 1983–1999. Work-ing paper, University of California, Berkeley.

Khessina, O. M. 2003. Entry mode, technological innovation

and firm survival in the worldwide optical disk driveindustry, 1983–1999. Unpublished doctoral dissertation,University of California, Berkeley.

Khessina, O. M., & Carroll, G. R. 2008. Product demography ofde novo and de alio firms in the optical disk driveindustry, 1983–1999. Organization Science, 19: 25–38.

Klepper, S. 2002a. Firm survival and the evolution of oligop-oly. RAND Journal of Economics. 33: 37–61.

Klepper, S. 2002b. The capabilities of new firms and theevolution of the US automobile industry. Industrial andCorporate Change, 11: 645–665.

Klepper, S., & Simons, K. L. 2000. Dominance by birthright:Entry to prior radio procedures and competitive ramifi-cations in the U.S. television receiver industry. StrategicManagement Journal, 21: 997–1016.

Klepper, S., & Sleeper, S. 2005. Entry by spinoffs. Manage-ment Science, 51: 1291–1306.

Lane, S. J. 1989. Entry and industry evolution in the ATMmanufacturers market. Unpublished doctoral disserta-tion, Stanford University, Stanford, CA.

Levinthal, D. A. 1997. Adaptation on rugged landscapes.Management Science, 43: 934–950.

Levinthal, D. A., & March, J. G. 1993. The myopia of learning.Strategic Management Journal, 14: 95–112.

Levitt, B., & March, J. 1988. Organizational learning. AnnualReview of Sociology, 14: 319–340.

Lieberman, M. B., & Montgomery, D. B. 1988. First-moveradvantages. Strategic Management Journal, 9: 41–58.

March, J. G. 1991. Exploration and exploitation in organiza-tional learning. Organization Science, 2: 71–87.

Markides, C., & Williamson, P. 1994. Related diversification,competencies and corporate performance. StrategicManagement Journal, 15(Summer Special Issue): 149–165.

McKendrick, D., Doner, R. F., & Haggard, S. 2000. From SiliconValley to Singapore: Location and competitive advan-tage in the hard disk drive industry. Stanford, CA: Stan-ford University Press.

Milgrom, P., & Roberts, J. 1990. The economics of modernmanufacturing: Technology, strategy and organization.American Economic Review, 80: 511–528.

Miller, D. J. 2006. Technological diversity, related diversifi-cation, and firm performance. Strategic ManagementJournal, 27: 601–619.

Mitchell, W. 1991. Dual clocks: Entry order influences onincumbent and newcomer market share and survivalwhen specialized assets retain their value. StrategicManagement Journal, 12: 85–100.

Murmann, J. P., & Frenken, K. 2006. Toward a systematicframework for research on dominant designs, techno-logical innovations, and industrial change. ResearchPolicy, 35: 925–952.

Nelson, R. R., & Winter, S. G. 1982. An evolutionary theory ofeconomic change. Cambridge, MA: Harvard UniversityPress.


Penrose, E. T. 1959. The theory of the growth of the firm.Oxford: Oxford University Press.

Pisano, G. P. 1994. Knowledge, integration, and the locus oflearning: An empirical analysis of process development.Strategic Management Journal, 15(Winter Special Issue):85–100.

Rao, H. 1994. The social construction of reputation: Certifica-tion contests, legitimation, and the survival of organiza-tions in the American automobile industry: 1895–1912.Strategic Management Journal, 15: 29–44.

Rindova, V. P., & Kotha, S. 2001. Continuous “morphing”:Competing through dynamic capabilities, form, andfunction. Academy of Management Journal, 44: 1264–1280.

Rivkin, J. W. 2000. Imitation of complex strategies. Manage-ment Science, 46: 824–844.

Rivkin, J. W., & Siggelkow, N. 2003. Balancing search andstability: Interdependencies among elements of organi-zational design. Management Science, 49: 290–311.

Rivkin, J. W., & Siggelkow, N. 2007. Patterned interactions incomplex systems: Implications for exploration. Manage-ment Science, 53: 1068–1085.

Rosenkopf, L., & Almeida, P. 2003. Overcoming local searchthrough alliances and mobility. Management Science,49: 751–766.

Sastry, M. A. 1997. Problems and paradoxes in a model ofpunctuated organizational change. Administrative Sci-ence Quarterly, 42: 237–275.

Schmalensee, R. 1989. Inter-industry studies of structure andperformance. In R. Schmalensee & R. Willig (Eds.), Hand-

book of industrial organization, vol. 2: 951–1001. Oxford:North Holland.

Schumpeter, J. A. 1934. The theory of economic development.Cambridge, MA: Harvard University Press.

Sine, W. D., Mitsuhashi, M., & Kirsch, D. A. 2006. RevisitingBurns and Stalker: Formal structure and new ventureperformance in emerging economic sectors. Academy ofManagement Journal, 49: 121–132.

Sorenson, O. 1997. The complexity catastrophe in the com-puter industry: Interdependence and adaptability in or-ganizational evolution. Unpublished doctoral disserta-tion, Stanford University, Stanford, CA.

Sorenson, O. 2003. Interdependence and adaptability: Orga-nizational learning and the long-term effect of integra-tion. Management Science, 49: 446–463.

Sorenson, O., McEvily, S., Ren, C. R., & Roy, R. 2006. Nichewidth revisited: Organizational scope, behavior andperformance. Strategic Management Journal, 27: 915–936.

Stinchcombe, A. 1965. Social structure and organizations. InJ. G. March (Ed.), Handbook of organizations: 143–192.Chicago: Rand McNally.

Suarez, F. F., & Utterback, J. M. 1995. Dominant designs andthe survival of firms. Strategic Management Journal, 16:415–430.

Teece, D., Rumelt, R. P., Dosi, G., & Winter, S. 1994. Under-standing corporate coherence: Theory and evidence.Journal of Economic Behavior and Organization, 32: 1–30.

Utterback, J. M., & Abernathy, W. J. 1975. A dynamic model ofprocess and product innovation. OMEGA, 3: 639–656.

Martin Ganco ([email protected]) is a doctoral candidate in strategy at the Uni-versity of Illinois at Urbana-Champaign. His current research focuses on understand-ing entrepreneurial processes and their impact on firm capabilities, performance, andindustry evolution.

Rajshree Agarwal ([email protected]) is the John Georges Professor of Technol-ogy Management and Strategy at the University of Illinois at Urbana Champaign. Shereceived her Ph.D. from the State University of New York at Buffalo. Her researchinterests focus on the implications of entrepreneurship and innovation for industryand firm evolution.


PERFORMANCE DIFFERENTIALS BETWEEN DIVERSIFYING …terpconnect.umd.edu/~rajshree/research/31 Ganco, Agarwal - 2009.pdfCOMPLEXITY APPROACH MARTIN GANCO RAJSHREE AGARWAL ... Survival,

Documents