You are on page 1of 25

!

Academy of Management Review


2007, Vol. 32, No. 1, 195–218.

THE KNOWLEDGE-BASED VIEW,


NESTED HETEROGENEITY, AND NEW VALUE
CREATION: PHILOSOPHICAL
CONSIDERATIONS ON THE LOCUS OF
KNOWLEDGE
TEPPO FELIN
Brigham Young University

WILLIAM S. HESTERLY
University of Utah

At what level is new value created, or, put differently, what is the locus of knowledge?
While knowledge and capabilities-based researchers argue that the locus of new
value and knowledge lies at the firm level, we challenge this conceptualization and
theoretically build toward more individualist foundations. We explicate the underly-
ing philosophical assumptions of extant knowledge and capabilities-based work and
discuss attributional problems. Nested (individual-level, a priori) heterogeneity may
provide a better explanation of collective heterogeneity.

There is really no need for the firm to be the value or, put differently, the locus of knowl-
fundamental unit of organization in invention; edge.1 That is, at what level is new value cre-
there is plenty of reason to suppose that individ-
ual talents count for a good deal more than the ated?
firm as an organization (Arrow, 1962: 624). Nahapiet and Ghoshal (1998: 246), among oth-
ers (e.g., Grant, 1996: 112), have pointed out that
The knowledge and capabilities-based views scholars have clearly taken sides in this debate
(KBV) in strategy have largely extended re- and given primacy to one of two different com-
source-based reasoning by suggesting that peting sources of value or knowledge loci—
knowledge is the primary resource underlying individuals or collectives. While a few scholars
new value creation, heterogeneity, and compet- have argued for the primacy of the individual
itive advantage (Barney, 1991; Grant, 1996; Kogut (Grant, 1996: 112; Simon, 1991: 176), most have
& Zander, 1992). However, despite the recent pro- focused on a collective locus of knowledge (e.g.,
liferation of research into knowledge-based ar- Adler, 2001; Brown & Duguid, 2001; Eisenhardt &
guments, a number of fundamental constructs Martin, 2000; Kogut, 2000; Kogut & Zander, 1992,
and questions have yet to be clearly defined and 1995; Nahapiet & Ghoshal, 1998; Nelson & Win-
explored (e.g., see Kaplan, Schenkel, von Krogh, ter, 1982; Powell, Koput, & Smith-Doerr, 1996;
& Weber, 2001, for a recent overview). A critical, Spender, 1996; Tsoukas, 1996; Winter, 2003; Zollo
implicit debate underlying much knowledge & Winter, 2002).
and capabilities-based work is whether the in- The theoretical and practical implications of
dividual or the collective is the source of new the debate between an individual and collective
locus of knowledge are far from pedantic for our
conceptualization of new value creation. Sev-
We thank Aya Chacar, Russ Coff, Karin Fladmoe-
Lindquist, Nicolai Foss, Anoop Madhok, Anita McGahan, Jim
Robins, Jamal Shamsie, Steve Tallman, and Dave Whetten
1
for their helpful comments. Thanks to participants at the We roughly concur with Kaplan’s definition of the locus
PDW on the RBV (organized by Rich Makadok and Doug problem and apply the question to knowledge-based work:
Miller) at the 2003 annual meeting of the Academy of Man- “The locus problem may be described as that of selecting the
agement for their feedback. Participants at the 2003 DRUID ultimate subject-matter for inquiry in behavioral science, the
summer conference provided helpful comments on earlier attribute space for its description, and the conceptual struc-
drafts. Any mistakes in this manuscript are the sole respon- ture within which hypotheses about it are to be formulated”
sibility of the authors. (1964: 78).

195
196 Academy of Management Review January

eral fundamental questions are implicated. For value in the context of the biotechnology and
example, is new, value-creating knowledge fun- pharmaceutical industries. In conclusion, we
damentally about organizational processes, or discuss the immediate next steps necessary to
is it rooted more in the attributes and abilities of disentangle the locus of knowledge question or
the individuals involved? Are innovations the individual- and collective-level effects and as-
result of new value and ideas that are created sociated implications for new value creation.
as a collective process that is independent of
individuals, or does new value creation start
INDIVIDUAL VERSUS COLLECTIVE
with the individual? Moreover, is knowledge
KNOWLEDGE IDEAL TYPE
transfer more about structural and organization-
al processes, or is it about the underlying abil- To facilitate our discussion, we explore the
ities of the respective individuals to absorb collectivist and individualist traditions of
knowledge? Finally, from a strategic perspec- knowledge (cf. Nahapiet & Ghoshal, 1998: 246) as
tive, are knowledge-based advantages the re- pure or ideal types (Weber, 1949; cf. Adler, 2001;
sult of an organization’s ability to acquire and Ouchi & Johnson, 1978), or as Mertonian para-
build strategic assets, or are these advantages digms. We first discuss our justification for
the result of the underlying individuals’ abilities employing ideal types and then review the es-
to make astute resource acquisition decisions? sential positions of the two perspectives. There-
Overall, the practical implications for how a after, we contrast the collective and individual
firm creates new value are radically different, ideal types on important philosophy of science
depending on the underlying assumption about dimensions. These dimensions include episte-
the locus of knowledge.2 mology, ontology, and mereology, which deal
To foreshadow our conclusions, we argue that with issues of causal directionality and part-
a coherent theory of new value creation must whole relations. We then contrast the two per-
start with a consideration of the individuals who spectives in light of their underlying assump-
make up the organization. While collectivist tions with regard to levels-of-analysis issues
knowledge and capabilities-based work as- and human nature.
sumes that individuals a priori are homoge-
neous, infinitely malleable, or randomly distrib-
Note on Methodology and Parameters
uted into organizations, we argue and provide
evidence that this conceptualization is flawed For our purposes, the comparison between in-
and has lead to theoretical, locus-related misat- dividual and collective ideal types serves two
tributions, with direct implications for our un- functions. First, research tends to eschew para-
derstanding of the origins of new value. Specif- dox and conflict in theory building in favor of
ically, in lieu of an alternative approach to tractable models (Poole & Van de Ven, 1989: 563).
understanding new value and knowledge cre- However, we make the implicit debate between
ation, we build on and discuss the implications an individual versus collective locus of knowl-
of accumulating empirical evidence from cogni- edge the very focus of our paper and our explicit
tive science, which gives significant importance strategy for comparing assumptions and key
to heterogeneous a priori individual-level theoretical dimensions. From the perspective of
knowledge. We also discuss the underlying at- new value creation, our underlying question is
tributional problems of collectivist approaches, “Where (primarily) is new value created?” In
which include not accounting for individuals as explicating these two ideal types, we use two of
initial conditions, focusing on proximate versus Poole and Van de Ven’s (1989: 566) theory-
final causes, and measurement problems. By building strategies: opposition and spatial sep-
way of grounding our analysis, we highlight aration. As suggested by the authors, “A great
locus-related concerns by revisiting empirical deal can be learned from juxtaposing contradic-
knowledge and capabilities-based work, which tory propositions and assumptions, even if they
has argued for contradictory sources of new are incompatible” (1989: 566). Second, while
ideal types do not replace theory, they can nev-
ertheless offer a stepping-stone for the develop-
2
We thank the editor and an anonymous reviewer for ment of subsequent theoretical arguments,
pointing out some of these implications. broadly addressing the question of which key
2007 Felin and Hesterly 197

variables and constructs theories should ad- Ghoshal, 1998: 246). For example, Nelson and
dress and which should be discarded (Bach- Winter state that
arach, 1989: 497; Weick, 1995).
the possession of technical “knowledge” is an
It is important to briefly discuss some param- attribute of the firm as a whole, as an organized
eters and limitations. First, given the vast liter- entity, and is not reducible to what any single
ature on knowledge in strategy and organiza- individual knows, or even to any simple aggre-
tion theory (including several recent journal gation of competencies and capabilities of all the
various individuals, equipments and installa-
special issues), we do not pretend to include all
tions of the firm (1982: 63).
of the knowledge-based literature in our analy-
sis but focus, rather, on key foundational arti- Kogut and Zander also point to a distinctly
cles. Several of these articles make an explicit collectivist perspective by arguing that “firms
choice about the key sources of new value, exist because they provide a social community
whereas others assume it implicitly. Thus, the of voluntaristic action structured by organizing
dichotomy we employ is not our creation. Fur- principles that are not reducible to individuals”
thermore, the discussion of extremes or polar (1992: 384). The authors later define the firm fur-
modes provides a compelling way to study these ther as “a social community specializing in the
arguments, since mixed, dualist, or interaction- speed and transfer of knowledge” (Kogut &
ist modes are permutations of the two extremes Zander, 1996: 503). Spender concurs and has fur-
and often suffer from problems of internal con- ther pointed out that the collectivist tradition
tradiction and incoherence (Rosenberg, 1995: 22; presumes (specifically referring to Nelson &
Sawyer, 2001: 570; cf. Conner & Prahalad, 1996: Winter, 1982) that “the firm has an ability to
478). Moreover, social analysis inherently be- know independently of its employees, or at least
gins with a focus on either individuals or collec- independently of their conscious reasoning”
tives as the key independent causal variables or (1996: 51). Researchers have also alluded to a
as the antecedent to social outcomes (e.g., “collective mind” (Weick & Roberts, 1993; also
Coleman, 1990: 2; Wrong, 1961: 183); in our case, see Durkheim’s [1962] notion of “collective con-
what specifically are the primary, fundamental science”). Furthermore, collective knowledge is
drivers of new knowledge and value creation? argued to be the most important kind of strate-
Second, there have been a number of parallel gic knowledge, and, thus, collectivist theories
developments and extensions in the knowledge- inherently must recognize that “by restricting
based literature, including research looking at the scope of our analysis only to social knowl-
competencies, routines, capabilities, and, more edge, we will be unable to capture the influ-
broadly, organizational learning (e.g., Eisen- ences that . . . individual knowledge may have
hardt & Martin, 2000; Henderson & Cockburn, on the intellectual capital of the firm” (Nahapiet
1994; Levitt & March, 1988; Nelson & Winter, 1982; & Ghoshal, 1998: 247; also see Spender, 1996).
Teece, Pisano, & Shuen, 1997; Zollo & Winter, Overall, only collective variables are allowed as
2002). We broadly subsume and discuss this re- explanans for outcomes, or as the antecedent to
search under the KBV heading since the work new value—such as culture, community, rou-
shares significant overlaps (cf. Argote & Ingram, tine, and environment.
2000; Madhok, 1997). An exception to the prevailing mode in knowl-
edge-based work is the theoretical work of Si-
mon and Grant. This individual-oriented tradi-
tion takes the position that individuals are the
Individual Versus Collective Locus of
primary locus of knowledge and should be the
Knowledge
basis for understanding new value creation and
Extant knowledge-based research largely fo- organizational outcomes. Grant emphasizes the
cuses on collectives as the source of new value role of the individual as the key knowledge lo-
or the locus of knowledge (e.g., Henderson & cus. He turns the collectivist logic discussed
Cockburn, 1994; Kogut & Zander, 1992; Nelson & above on its head and suggests that
Winter, 1982; Teece et al., 1997). From the collec-
the emphasis upon the role of the individual as
tivist perspective, knowledge is fundamentally the primary actor in knowledge creation and the
a social phenomenon that is different from the principle repository of knowledge, I believe, is
aggregation of individuals (Nahapiet & essential to piercing the veil of organizational
198 Academy of Management Review January

knowledge and clarifying the role of organiza- When viewed against philosophy of science
tions in the creation and application of knowl- dimensions such as epistemology, ontology, and
edge (Grant, 1996: 121).
mereology, stark contrasts are apparent be-
Simon has further argued that “all organization- tween the individualist and collectivist knowl-
al learning takes place inside human heads; an edge perspectives in terms of the sources and
organization learns in only two ways: (a) by the origins of new value (see Table 1). These con-
learning of its members, or (b) by ingesting new trasting assumptions have received little ex-
members who have knowledge the organization plicit attention from KBV scholars. Yet these dif-
didn’t previously have” (1991: 125). ferences are important given that, as noted by

TABLE 1
Dimensions of Individual and Collective Knowledge Ideal Types

Dimensions Individual Ideal Type Collective Ideal Type

Locus of knowledge or Individual Collectives


source of new value
Methodological Methodological individualism Methodological holism or collectivism
tradition or
epistemology
Representative quote “. . . the belief in the empirical “Sociological method as we practice it
existence of social wholes or rests wholly on the basic principle
collectives, which may be described that social facts must be studied as
as naı̈ve collectivism, has to be things, that is, as realities external to
replaced by the demand that social the individual. There is no principle
phenomena, including collectives, for which we have received more
should be analyzed in terms of criticism; but none is more
individuals” (Popper, 1968: 341) fundamental” (Durkheim, 1952: 39)
Causal directionality Micro-micro, micro-macro; upward Macro-macro, macro-micro; downward
causation causation
Explanans or Individuals “Social facts”—community, collective,
independent routines, culture, environment,
variables organizing principles, capabilities,
etc.
Collective ontology Reducible to individuals—whole is the Not reducible to individuals—whole is
sum of parts; only individuals are greater than sum of parts or
“real” independent whole
Mereology—i.e., part- Resultant whole (supervenience) Emergent whole (multiple
whole or individual- realizability)
collective level
relationship
Level of analysis Individual heterogeneity, independence Individual homogeneity, higher-level
assumptions from higher-level interaction collective heterogeneity—e.g., firm,
culture, environment
Theory of knowledge Internalist Externalist
Source of knowledge A priori or innate Environmentally determined
Human nature Nature Nurture, blank slate
Key variables and Individual mobility, personnel Routines, competencies, capability,
mechanisms selection, appropriation, incentives, process, culture, community
HR practices, self-selection
Representative quote “The firm is in no sense a ‘natural unit’. “. . . the possession of technical
Only the individual members of the ‘knowledge’ is an attribute of the
economy can lay claim to that firm as a whole, as an organized
distinction. All are potential entity, and is not reducible to what
entrepreneurs. . . . The ultimate any single individual knows, or
repositories of technological even to any simple aggregation of
knowledge in any society are the competencies and capabilities of all
men comprising it . . . in itself the the various individuals, equipments
firm possesses no knowledge” (De and installations of the firm”
Graaf, 1957: 16) (Nelson & Winter, 1982: 63)
2007 Felin and Hesterly 199

Rosenberg, “Being clear about a discipline’s sidered real because they exhibit “downward
philosophy is essential because at the frontiers causation” or determine the behavior of lower
of the disciplines, it is the philosophy of science levels or individuals (Durkheim, 1952; cf. Saw-
that guides inquiry” (1995: 4). Indeed, collectivist yer, 2001). In his classic analysis of suicide, for
assumptions have guided most knowledge- example, Durkheim concluded that the phenom-
based work. Moreover, such assumptions have enon was a function of top-down, extraindi-
important implications for how we interpret the vidual “social facts” and cultural environment
findings of research and also for how the field (Durkheim, 1952: 324). That is, not only do these
thinks about new value creation. collective facts exist, but they also have primary
Epistemology: Methodological individualism causal influence on lower levels of analysis. The
versus collectivism. The individual and collec- strong methodological assumption is that struc-
tive knowledge ideal types map squarely onto ture and organization exist prior to individual
the epistemological traditions of methodologi- action and drive the behavior of individuals
cal individualism and collectivism. The peren- (Coleman, 1990; Rosenberg, 1995: 5). Individuals
nial debate between these two traditions has and collectives only behave according to a pri-
plagued a number of the disciplines that the ori routines, structure, roles, or organization.
field of strategy draws on, including economics, These are independent of individuals, or onto-
sociology, and philosophy (see O’Neill, 1973, for logically autonomous.
a compilation of key readings; see also Popper, Methodological individualists, however, have
1957; Rosenberg, 1995; Udehn, 2001). However, argued that collectives are inherently made up
the individualism-collectivism discussion has of and result because of individuals, and indi-
only rarely been alluded to in the management viduals should thus be the basic unit of analysis
literature (see Donaldson, 1990, for an excep- (Elster, 1989; Nagel, 1961; Popper, 1968). Popper,
tion). This omission is unfortunate, since the ba- for example, has argued that “the belief in the
sic values and questions suggested in this de- empirical existence of social wholes or collec-
bate are particularly critical regarding our tives . . . has to be replaced by the demand that
assumptions about new value creation, given social phenomena, including collectives, should
that knowledge-based arguments directly build be analyzed in terms of individuals” (1968: 341;
on these respective traditions. The crux of this see also Tuomela, 1990). Individualists deny the
debate has been whether individuals or collec- existence of the metaphysical and therefore ar-
tives should serve as the locus in explaining gue that only individuals should serve as ex-
and predicting social outcomes—micro-macro planans (cf. Hempel & Oppenheim, 1948). Collec-
versus macro-macro (Kincaid, 1997). We briefly tives such as organizations then result from
review each tradition in turn. individuals and their actions. The term method-
Methodological collectivism builds on a ological connotes an a priori, epistemological
Durkheimian sociological tradition, which pro- commitment to seeking individual-level expla-
vides the basis for much extant knowledge- nations or, perhaps more accurately, the belief
based work (e.g., Kogut & Zander, 1996). Method- that only individuals exist in any real sense and
ological collectivists argue that there is should provide the basis for all collective expla-
something emergent about collectives leading nation (Rosenberg, 1995: 125). Therefore, individ-
to “social facts,” which are sui genesis, or worth ual human action is the key level of analysis
studying in their own right (Durkheim, 1952: 39; (Elster, 1989: 74). Inherently, organizations and
Popper, 1957: 71; Rosenberg, 1995: 132).3 Individ- collectives are made up of individuals.
uals are, in effect, considered extraneous to col- Ontology and mereology. In line with method-
lectivist theories since collective facts (e.g., com- ological collectivism, knowledge-based schol-
munity, routines) largely determine outcomes ars have generally argued that organizational
(Durkheim, 1962: 103). Collectives must be con- knowledge is either emergent or even com-
pletely independent of the individuals or parts
3
that make up the whole or the organization. Col-
The terms collectivism and holism are considered syn- lective theorists argue, for example, that rou-
onyms for the purposes of our analysis. Some authors have
introduced fine-grained distinctions, although the two terms tines, the building blocks of organization, are
have been used interchangeably in most of the philosophi- “independent of the individual actors who exe-
cal and sociological literature (Popper, 1957; Udehn, 2001). cute them” (Levitt & March, 1988: 320; also see
200 Academy of Management Review January

Nelson & Winter, 1982). That is, the collective ure 1). In other words, supervenience is defined
cannot be understood by reducing it to its parts as higher-level dependence or determination
and is, thus, ontologically independent: emer- based on lower-level properties or facts (Kim,
gent (multiply realized) or even completely in- 1993; Sawyer, 2001). In terms of our individual
dependent of its parts (cf. Sawyer, 2001). Focus in knowledge ideal type, as argued by Simon
this perspective has been given to various su- (1991), the organization only learns based on the
praindividual structures, including routines individuals that it is composed of, or based on
(Nelson & Winter, 1982; Winter, 2003), interac- its lower-level supervenience base. Ontologi-
tions (Argote & Ingram, 2000), organizing princi- cally, only individuals are real and they there-
ples (Kogut & Zander, 1992), capabilities (Teece fore determine organization and outcome.
et al., 1997), and community (Brown & Duguid,
2001).
These a priori structures, similar to
Implicit Assumptions in Choosing a Level of
Durkheim’s (1952) analysis of suicide, determine
Analysis
individual behavior and thus become the key
antecedent or explanan for analyzing outcomes Related to our discussion of mereology, which
(see Figure 1). Independence and top-down focuses on the causal flow between different
(macro-micro, macro-macro) causality between levels, collectivist approaches raise very spe-
the two levels are critical, strong assumptions, cific theoretical and methodological concerns
which drive the analysis (Durkheim, 1962; cf. about assumptions that are implicitly adopted
Rosenberg, 1995). The same collective outcome in choosing a level of analysis (cf. Klein, Danse-
can be realized through a number of different reau, & Hall, 1994: 224; Rousseau, 1985). In spec-
individuals, given their homogeneity, and the ifying a level of analysis, scholars make strong
relation between the individual and organiza- assumptions about other levels, which are
tion is not determined by the individuals who rarely explicitly stated and perhaps not even
make up the organization—thus the emphasis recognized (Dansereau, Yammarino, & Kohles,
on routines and other extraindividual consider- 1999). While levels-related matters are often con-
ations such as routines and collectives (cf. Levitt sidered to be largely empirical matters (e.g.,
& March, 1988; Nelson & Winter, 1982). firm versus industry debate; Rumelt, 1991), or the
The notion of “supervenience” has been dis- stuff of statisticians, we concur that they are,
cussed extensively in the philosophy of science “first and foremost . . . the domain of theorists”
literature and provides the opposite prediction (Klein et al., 1994: 224). Two levels-related issues
in the part-whole relationship (cf. Kim, 1993; and assumptions, highlighted by Klein et al.
Sawyer, 2001). Supervenience means that any (1994: 199), are of primary interest to us with
changes at a higher level are strictly a function regard to our individual and collective knowl-
of changes at the level below and that all col- edge ideal types: specifying homogeneity and
lective outcomes can (or ultimately can) be ex- specifying independence. We discuss each of
plained with reference to individuals (cf. Elster, the two levels-related assumptions in turn,
1989). The whole results from the parts (see Fig- along with subsequently explicating what these

FIGURE 1
Mereology and Causal Directionality Between Individuals and Collective
2007 Felin and Hesterly 201

assumptions reveal about our understanding of typically simplify and idealize in the interest of
human nature, knowledge, and causality. parsimony. As noted by Nelson and Winter,
First, in specifying an organization or any Theorists should aim to tell the truth in their
kind of collective as the key level of analysis, theorizing, but they cannot aim to tell the whole
scholars implicitly attribute homogeneity to truth. For to theorize is precisely to focus on those
lower levels of analysis (Klein et al., 1994). This, entities and relationships in reality that are be-
lieved to be central to the phenomenon ob-
of course, is the assumption, although in reality,
served—and largely ignore the rest (1982: 134; see
given potential heterogeneity at all levels, the also Coleman, 1990).
question in theorizing is where most heteroge-
neity lies. In the field of strategy, the question of This simplification involves choosing a particu-
where heterogeneity resides has led to large- lar level of analysis, which is presumed to be
scale variance decomposition work comparing more heterogeneous than lower levels (for
the performance effects at differing levels of which homogeneity has to be specified). Thus, in
analysis, including industry, corporate, and choosing a level of analysis, critical trade-offs
business levels (e.g., McGahan & Porter, 1997; between heterogeneity and independence are
Rumelt, 1991). As a point of illustration, if indus- involved. Specifically, when knowledge-based
try is specified as the key level of analysis (e.g., scholars focus on the collective, they are implic-
itly assuming that some collective, firm-level
research on industry-level product markets),
dynamics (for e.g., routines or capabilities),
then homogeneity at the firm level is assumed.
rather than individual differences, account for
Resource and knowledge-based scholars have,
knowledge heterogeneity.
of course, challenged the assumption of firm-
level homogeneity and have argued that the
firm is a more important source of heterogeneity HUMAN NATURE, LEVELS, AND KNOWLEDGE
(cf. Rumelt, 1991). Just as the industry level of
The discussion of levels in the terms we have
analysis assumes firm homogeneity, so firm-
outlined above (homogeneity and indepen-
level heterogeneity (often implicitly) assumes
dence), while instructive and necessary, pro-
individual homogeneity.
vides a rather sterile and detached look at the
The second implicit assumption involved in
underlying choices that researchers make, and
choosing a level of analysis has to do with in-
therefore needs to be more carefully unpacked.
dependence. When the firm, for example, is
More specifically, in choosing a level of analy-
specified as the key level of analysis, it is as- sis, researchers unveil, or at least implicitly
sumed that all firms are independent of any make, critical assumptions about human nature.
higher-level interaction with other firms— These assumptions are rarely stated, but they go
industry, for instance (cf. Dansereau et al., 1999). to the very core of theorizing and therefore ought
Tying the levels assumption of independence to to be explicated more carefully. As forcefully
the firm versus industry debate, a firm-level fo- stated by Simon, “Nothing is more fundamental
cus suggests that there is nothing systematic in setting our research agenda and informing
about belonging to an industry. Therefore, firm- our research methods than our view of the na-
level analysis generally has controls for indus- ture of the human beings whose behavior we
try-level variables. Again, in reality, interaction are studying” (1985: 303).
and heterogeneity exist at several levels of When specifying a collective as the key level
analysis, and, thus, theorists inherently make of analysis, what specifically are the underlying
critical decisions and assumptions about where assumptions regarding human nature implied
most of the heterogeneity lies by specifying a by “homogeneity at the lower level”? Durkheim
level for their theory and attempting to control was very clear when he argued that “individual
for potential alternative explanations from dif- natures are merely the indeterminate material
fering levels of analysis. There is, of course, a that the social factor molds and transforms”
second-order problem in specifying another di- (1962: 106). The Durkheimian assumption, shared
visible collective as the key level. by collectivist knowledge-based scholars, is
In choosing a level of analysis for a theory, that individuals are indeterminate material or
one cannot completely rule out interaction and homogeneous (or a “blank slate”), which heter-
heterogeneity at alternate levels. Thus, theorists ogeneous social facts (such as culture, social
202 Academy of Management Review January

context, and environment) shape and determine. “window to the mind,” and thus allowing us to
Or, put in light of the knowledge locus problem, understand sources of knowledge and learning.
no individual-level, a priori knowledge exists Overall, this externalist learning tradition ar-
without environmental stimulus and learning. gues that language acquisition and learning in
That is, heterogeneity in stimulus, context, and general is socially constructed, emergent, con-
environment determines differential individual- text dependent, and, for the most part, environ-
and collective-level outcomes. mentally determined—providing the theoretical
The above sentiments for homogeneity are basis for assuming homogeneity and malleabil-
strongly echoed and built on in collectivist work ity at the individual level and the primacy of
in the KBV (e.g., Brown & Duguid, 2001; Hender- heterogeneity in social context, environment,
son & Cockburn, 1994; Nahapiet & Ghoshal, and collectives in determining knowledge out-
1998). Specifically, emphasis in collective comes.
knowledge-based work is placed on heteroge- The environmentally oriented, externalist,
neous collective process, social construction, sit- and collectivist learning tradition discussed
uation, and experience (Eisenhardt & Martin, above has been sharply criticized and funda-
2000; Teece et al., 1997; Weick & Roberts, 1993). mentally challenged by the Chomskyan cogni-
Spender, for example, states that “we must ar- tive revolution in linguistics, cognitive psychol-
gue that organizations learn and have knowl- ogy, and philosophy. Given that this literature
edge only to the extent that their members are directly challenges the very foundations of col-
malleable beings whose sense of self is influ- lectivist knowledge-based arguments and re-
enced by the organization’s evolving social lates to the above levels assumptions, it is
identity,” and, thus, learning is “primarily inter- important to carefully explicate the key implica-
nalized from the social context” (1996: 53; em- tions as they relate to the knowledge locus ques-
phasis added). tion. This is also warranted given that, quite
Collective knowledge-based scholars justify surprisingly, Chomsky’s key arguments for the
the emphasis on individual “malleability,” even primacy of a priori and innate, individual-level
to the point of “questioning the concept of the knowledge have received virtually no attention
individual” by citing work in psychology, which in the management literature, despite their rad-
emphasizes that cognitive abilities and knowl- ical implications for the very notion of learning.4
edge in general are context dependent and en- Beginning with his harshly critical review of
vironmentally determined (Spender, 1996: 52–54; Skinner’s (1957) Verbal Behavior, Chomsky (1959)
also see Brown & Duguid, 2001). The explicit argues and empirically shows that external en-
theoretical foundation for collectivist knowl- vironmental input and socialization play only a
edge-based work (along with Durkheim) is the minimal triggering role in language acquisition,
work of externalist and behaviorist psycholo- while he gives primacy to a priori, individual-
gists and philosophers, such as Bandura, level knowledge (Chomsky, 1986, 2000). More
Bruner, James, Piaget, Polanyi, Ryle, and Vy- specifically, human beings have a genetically
gotsky (e.g., Brown & Duguid, 1991, 2001; Harga- determined “initial state,” competence, or en-
don & Fanelli, 2001; Kogut & Zander, 1996; Na- dowment, which is individual, intensional, and
hapiet & Ghoshal, 1998; Nonaka, 1994; Spender, internal (Chomsky’s “I”-language) and provides
1996; Tsoukas, 1996; Weick & Roberts, 1993). the fundamental basis for learning. Chomsky-
Given that this literature provides the explicit ans, in fact, are uncomfortable with the very
foundation for much of collectivist knowledge notion of “learning,” given the “poverty of stim-
and capabilities-based work in strategy, it is
worth highlighting that most of the above exter-
nalist scholars (both in psychology and philos- 4
While Chomsky occasionally has been cited in the man-
ophy) developed a substantial part of their ar- agement literature, a search of key management journals
guments for the primacy of environment and found no discussion, reference, or citation of his key contri-
social context based on their studies of lan- bution with regard to the primacy of individual-level knowl-
edge or his challenge of behaviorism and empiricism. Pent-
guage acquisition (e.g., Rylean linguistic behav-
land (e.g., 1995) does extensively cite Chomsky, although, as
iorism) and child development. Language argu- he himself admits, his arguments are far from Chomsky’s
ably is one of the most basic and critical, but original intention, and he furthermore makes an externalist,
also creative, acts of human beings, providing a social structural, and collectivist argument.
2007 Felin and Hesterly 203

ulus” (i.e., inputs scarcely reflect outputs); they outcomes (see Bouchard & McGue, 2003, and
argue, rather, that individual knowledge devel- McGue & Bouchard, 1998, for reviews). The twin
ops from within or grows or matures with rela- studies have largely included monozygotic
tively minimal external influence or environ- twins (who, by definition, share the same exact
mental causal importance (Chomsky, 2002). genetic, innate material) living in shared or non-
We should, of course, note that the emphasis shared (separated) environments, and have also
on individual innate cognitive abilities sans en- included large-scale empirical analysis and
vironment does not imply that outcomes are comparisons of other reference groups (adopted
completely determined by individual competen- siblings, siblings, etc.).
cies but that they are simply significantly more McGue and Bouchard (1998; also see Bouchard
important than environmental influences, thus & McGue, 2003), in their meta-analyses of twin
reversing the causal arrows of previous exter- and adoption studies, show that, depending on
nalist and collectivist learning theory. For the study, 50 to 85 percent of variance (75 percent
Chomsky and many linguists, languages (e.g., weighted average over numerous studies) in
Chinese, English, Finnish—Chomsky’s “E”- cognitive abilities and intelligence (percent-
language) are simply surface structures or epi- ages are also similar, although somewhat
phenoma, and thus artifacts, while the most sig- lower, for various dimensions of personality) is
nificant driver of language acquisition is the genetically and individually determined, while
universal deep structure or innate syntax, which the environment only, at best, accounts for a
is independent of semantics (Chomsky, 2000, marginal percentage (at most, one-fourth of the
2002). genetic explanation) of the variance. Most as-
Further evidence for a priori and innate tonishingly, empirical studies of separated
knowledge is provided by numerous other psy- twins raised in nonshared or different environ-
chologists (e.g., Susan Carey, Alan Leslie) and ments show little or no environmental influence
philosophers (e.g., Peter Carruthers, Jerry Fodor), (McGue & Bouchard, 1998). Furthermore, adop-
including, for example, Elizabeth Spelke’s pio- tive siblings raised together (same environment/
neering experimental work with infants, which socialization, etc.) show no correlation in cogni-
shows strong support for a priori innate knowl- tive abilities and intelligence (or other
edge in infants in other (other than language psychological dimensions), thus significantly
acquisition) cognitive domains (e.g., Spelke, questioning the environmental and blank slate
Breinlinger, Macomber, & Jacobson, 1992). hypothesis (see Scarr, 1997, for an excellent
The Chomskyan theoretical arguments and methodological overview). Overall, despite crit-
empirical findings are not alone in giving pri- icism, the robust findings for individual-level,
macy to human nature and innate knowledge, innate, invariable differences in cognitive abil-
as well as in challenging social constructionist ities and intelligence have withstood numerous
learning theory. While much of the above has critiques and are among the most robust empir-
focused on innate versus external knowledge, ical studies in psychology (see Bouchard, 1997;
Chomskyans have said relatively little about Lubinski, 2004; Scarr, 1997, for further discussion
the heterogeneity of innate knowledge. Recent and key references).5
empirical studies in the cognitive sciences and The Chomskyan cognitive revolution in lin-
psychology have provided strong evidence not guistics, the twin studies, and cognitive psy-
only for the innateness of knowledge but also for chology research suggest some critical implica-
its substantial heterogeneity and invariance tions for the KBV. First, findings in the cognitive
over time, thus further challenging the primacy sciences directly challenge the externalist
of learning theories and collective and hetero- learning theories that much of knowledge-
geneous environmental influence (Bouchard & based work builds on. Specifically, a priori and
McGue, 2003; Lubinski, 2000; also see House, heterogeneous individual-level knowledge may
Shane, & Herold, 1996). Variance decomposition provide a significant alternative explanation for
studies of separated twins (replicated by numer-
ous databases throughout the world) have pro- 5
The twin studies have, of course, been criticized, al-
vided a natural experiment for testing the rela- though these criticisms have been responded to persua-
tive importance of the individual-level versus sively. The interested reader is referred to excellent discus-
environmental and collective factors in learning sions by Bouchard (1997) and Scarr (1997).
204 Academy of Management Review January

the primacy given to context, environment, and the initial conditions cannot be ascertained, the
collective in learning outcomes. Collectivist scientific way of predicting breaks down” (1959:
knowledge-based theories may wrongly be at- 198; see also Nagel, 1961: 32). In experimental
tributing learning to the organization, while a design, for example, it is unacceptable to con-
more simple, parsimonious explanation has not sider an experiment valid unless initial condi-
been controlled for—the individual level. At the tions are specified, measured, or controlled. The
very least, the findings for a priori knowledge randomization of subjects accomplishes this in
suggest that the individual level needs to be studies where a manipulation is involved and
carefully controlled for, measured, and under- provides the control for potential confounds
stood. based on individual differences in a given con-
struct (although even randomization may mask
key drivers based on a priori individual differ-
ATTRIBUTIONAL PROBLEMS IN COLLECTIVIST
ences; Stanovich, 1999). The Chomskyan cogni-
APPROACHES
tive revolution highlighted above strongly
In previous sections we have shown how col- points out the need to better understand initial
lectivist knowledge-based approaches make conditions, or the individual “initial state,”
strong assumptions, either explicitly or implic- which, prior to Chomsky, was assumed to be
itly, about important methodological, epistemo- homogeneous and largely environmentally de-
logical, and ontological considerations. We termined. Specifically, prior learning and so-
have also demonstrated that collectivists adopt cialization studies have confounded heteroge-
assumptions about the homogeneity and mal- neous socialization and learning effects with
leability of individuals that are at odds with nested individual effects (Chomsky, 2000; Scarr,
growing theoretical consensus and rapidly ac- 1997).
cumulating empirical evidence in fields such as In the context of new knowledge and value
cognitive psychology. We have also raised sev- creation, the question is “What are the individ-
eral issues that suggest the strong possibility ual-level initial conditions in terms of a priori
that individual heterogeneity may account for abilities to learn and create knowledge, which
much of what is attributed to collectives in may fundamentally drive (or confound) collec-
knowledge-based theorizing and research. tive-level outcomes?” For example, March’s
There are, however, three more fundamental (1991) highly cited model of organizational
concerns about how collectivist approaches learning fundamentally is dependent on initial
make attributional errors, arising from method- knowledge and abilities (rates of learning) at
ological individualists in philosophy and the individual level and subsequent turnover—
sociology. The concerns about collectivist that is, who is joining and leaving the organiza-
approaches follow from the underlying assump- tion is absolutely fundamental to overall out-
tions with regard to individual homogeneity and comes (also see Arrow, 1974, and Castanias &
human nature discussed above, but, neverthe- Helfat, 2001). Schneider’s (1987) classic attrac-
less, they deserve more explicit treatment. Our tion-selection-attrition framework also points
concerns relate to the attributional problems in out the need to understand the question of who.
collectivist knowledge-based work in account- Specifically, Schneider’s perspective “rests on
ing for initial conditions, problems in identify- the idea that people are not randomly assigned
ing proximate as opposed to final causes of phe- to settings. It argues that it is the people who are
nomena, and, finally, misattributions and attracted to, are selected by, and remain in a
problems in measurement. setting that determine the setting” (1987: 440).
Thus, the underlying causal driver is the initial
condition of who composes the organization—
Individuals As Initial Conditions
the competencies, propensities, personalities,
Few things are as critical in theory building or, more broadly, the individual-level general
and testing as understanding initial conditions, abilities to create and absorb knowledge.
given that initial conditions may fatefully deter- The initial conditions or a priori propensities
mine outcomes and, if not accounted for, may of individuals within an organization of course
result in confounded, nonscientific findings and have fundamental implications for performance
underspecified models. As noted by Popper, “If heterogeneity and new value creation. While
2007 Felin and Hesterly 205

the term organization is used as shorthand to ment, may in effect wipe out emergent, collec-
describe the actions of a group of individuals, tive effects (see Nagel, 1961: 366 –397, for an ex-
nested within the collective level nevertheless cellent discussion; see also Oppenheim &
are heterogeneous individuals who take action Putnam, 1958). For example, Scarr (1997) has re-
and make critical decisions, which determine cently rerun the results of several classic social-
collective-level outcomes. Inherently, when talk- ization and learning studies in child develop-
ing about collective performance, the underly- ment and has found that individual-level
ing capabilities of the individuals must receive explanations (initial conditions in terms of in-
consideration as an antecedent to collective- nate cognitive ability) provide significantly
level outcomes. While it, in effect, “looks like more robust and parsimonious empirical find-
organizations determine behavior” (Schneider, ings (largely wiping out heterogeneous social-
1987: 440), the causal arrows can and should be ization effects), which previous studies had con-
reversed. founded with real collective-level learning
The implications of considering initial condi- effects. Similarly, Jencks (1972) has rerun large
tions for knowledge and capabilities-based data on school effects and found that, once ini-
scholars are clear. Failing to account for indi- tial conditions in terms of individual-level abil-
vidual effects may lead to spurious reasoning ity to learn were controlled for (i.e., intake test
about collectives. The implied rationale for the scores), school effects became marginal to non-
focus on collectives in management and strat- significant.6
egy research is that organizational environ-
ments represent “strong situations” (Davis-
Final Versus Proximate Causes and Artifacts
Blake & Pfeffer, 1989) or emergent, collective
environments (Nelson & Winter, 1982; Teece et Closely related to the logic of accounting for
al., 1997). Thus, it is implied that the individual initial conditions is another attributional prob-
homogeneity assumption (top-down causation) lem in collective theorizing that has been dis-
is warranted. Yet simply arguing that the field cussed by individualist-oriented sociologists,
has defined itself based on a given level does particularly Coleman (1990) and Boudon
not remove the fact that plausible and more (1998a,b). This problem is a tendency to focus on
powerful alternative explanations are often proximate causes and artifacts rather than more
readily apparent from the individual level (cf. fundamental or final causes. By finality, we do
Coleman, 1990: 2–5). Given measurement at the not mean a final theory to completely explain
organizational level, individuals often get side- heterogeneity at all levels but, rather, a theory
lined, although their heterogeneity may provide that provides more than cursory attention to
an alternative explanation for strategic out- whether the explanation provided does not
comes. Specifically, while the mantra of “the readily induce further questions as to more fun-
whole is greater than the sum of its parts” has damental antecedents and origins (Coleman,
readily been invoked in knowledge and capa- 1990: 15–16; see also Boudon, 1998a: 172). Put
bilities-based collectivist theories (e.g., Na- differently, does the theory simply explain arti-
hapiet & Ghoshal, 1998; Nelson & Winter, 1982), facts, or, perhaps, only proximate (versus more
the arithmetic has scarcely been resolved (Na- final) causes, thus “black boxing” what is theo-
gel, 1961: 380). retically most important? By artifacts, we refer to
The heart of the debate is whether collective correlated, though not causal, explanations for
outcomes are the result of the individuals who any given phenomenon. Causality therefore re-
make up an organization, or whether something
else, emergent, is realized (not reducible to in- 6
The confounds recently discussed in the child develop-
dividuals)—which should receive primary con- ment literature are instructive to note for scholars in other
sideration in accounting for outcomes. It is im- fields. While previous studies assume a blank slate in child
portant to note that scholars should also be development, and thus no need to control for individual
ready to concede that even if lower individual- differences, more recent theoretical and empirical work has
challenged these notions. Child learning is confounded by
level effects are controlled for and emergence is
the fact that innate, genetic differences exist between indi-
found, inaccurate measurement at the lower in- viduals, and given these differences, they may lead to the
dividual level, or subsequent scientific findings “creation” of different environments (Scarr, 1997; Scarr &
allowing for better individual-level measure- McCartney, 1983).
206 Academy of Management Review January

quires understanding the underlying “reasons” in the strategic analysis of established firms,
(Boudon, 1998a,b) and answering the theoretical since systematically uncovering the past com-
and causal question of why (Whetten, 1989; also positions and actions of individuals in organi-
see Nagel, 1961). zations is tedious, if not impossible. This limita-
Given that the notion of more final causes and tion, however, does not excuse scholars who
artifacts is fairly abstract, a brief illustration ignore lower levels of analysis that are accessi-
from knowledge-based work is warranted. Na- ble. Simply stated, scholars need to more care-
hapiet and Ghoshal discuss the role that social fully take account of individual-level explana-
capital plays in collective knowledge out- tions, before they cite emergence (cf. Coleman,
comes—in short, “’who you know’ affects ’what 1990: 3–5).
you know’” (1998: 252). However, this externalist
argument readily invites questions about the
Measurement
more final causal relation between the two con-
structs, or, more specifically, why do you know A final problem in collectivist theorizing is the
who you know? The causal relationship may be lack of empirical measures for key constructs,
reversed—that is, what you know affects who such as routines and capabilities (Collis, 1994;
you know. More specifically, some initial condi- Winter, 2003). As noted by Winter, “At present we
tion or a priori reason provides the basis for the lack an adequate approach for characterizing
relationship and subsequent interaction. For ex- routines in ways that would be useful for statis-
ample, a self-assortive mechanism based on in- tical analysis” (Murmann et al., 2003: 29). What
dividual-level characteristics and capabilities has partially contributed to this problem is the
can provide a more fundamental, underlying ex- broad and fairly vague definition of routines
planation (Zenger, 1994; also see Teece, 2003). As (e.g., routines as organizational truce, memory,
a practical example from a different context, genes, etc.; Nelson & Winter, 1982). Thus, as Ba-
college students with the highest standardized charach persuasively argues, if a theory “is not
test scores (or some other, perhaps better, objec- testable, no matter how profound or aestheti-
tive measure of individual ability) go to the best cally pleasing it may be, it is not a theory” (1989:
schools and, thus, have more talented peers and 512). While Nelson and Winter (1982: 72) explic-
faculty or a “better, more central” network posi- itly develop their arguments for the existence of
tion to draw on. While the origins of this indi- routines through a metaphor between individ-
vidual-level heterogeneity can be questioned ual skills and collective routines, “organization-
(e.g., socialization), recent work has pointed to al routines” have nevertheless received some-
the primacy of invariant, innate, individual- thing of a life of their own and central
level, and heterogeneous knowledge (Lubinski, theoretical status in knowledge-based theories
2000). Overall, assuming homogeneity at the in- (Eisenhardt & Martin, 2000; Teece et al., 1997),
dividual level implies that the causal mecha- despite not having moved beyond metaphorical
nism is knowing someone versus the more final status (cf. Bacharach, 1989). Given the problems
explanation that is based on why you know who with measurement and definition, giving pri-
you know. macy to these collective notions seems prema-
The notions of initial conditions and more fi- ture without an understanding of individual-
nal causes certainly have a temporal connota- level antecedents. The individual level is most
tion (Nagel, 1961). They raise the question “How certainly more measurable than higher collec-
far back do we go to find the theoretical, causal tive effects and provides a natural starting point
antecedents and explanations of a particular (Coleman, 1990).
phenomenon or event?” As argued by Lewis,
“Any particular event that we might wish to
EFFORTS AT RECONCILIATION
explain stands at the end of a long and compli-
cated causal history. We might imagine a world We should note that efforts to reconcile the
where causal histories are short and simple; but competing perspectives of an individual and
in the world as we know it, the only question is collective locus have been made (Giddens, 1979,
whether they are infinite or merely enormous” 1984, 1985; also see Bourdieu, 1977). Giddens
(1986: 214). Long causal histories are indeed most famously treats the individual-collective
problematic for the social sciences, particularly dilemma (or action-structure dilemma) as an in-
2007 Felin and Hesterly 207

separable dualism or, in his own words, “struc- intention is not to argue that the interactionist or
turation.” That is, the individual and collective process approach is wrong— only that it oper-
cannot be separated in any meaningful way be- ates from fundamental paradigmatic assump-
cause each simultaneously instantiates the tions that are difficult or impossible to reconcile
other. Thus, the ontological focus is on process with scientific realism (cf. McKelvey, 1997). We
and recursive social practice. This has led to a acknowledge that the rich description that flows
growing body of “knowledge practice” litera- from the knowledge practice approach can yield
ture, which builds on the arguments of both Gid- useful insights (Carlile, 2002; Orlikowski, 2002;
dens and Bourdieu (e.g., Carlile, 2002; Or- Weick & Roberts, 1993), but such an approach
likowski, 2002). seems to largely assume away our central ques-
The practice or structuration approach largely tion of interest—that is, what is the locus of
sides with more collectivist approaches (e.g., knowledge, or what are the sources of new
Brown & Duguid, 2001: 112), given its underlying value?
assumption of individual homogeneity. Specifi-
cally, the practice approach dismisses the very
KNOWLEDGE LOCUS PROBLEM IN PRACTICE
notion of levels by not only arguing for a funda-
mental, all-encompassing meta-“level” between Our efforts at finding a potential locus of
individual and collective—that is, practice knowledge and source of new value have a
(Brown & Duguid, 2001; Tsoukas, 1996)— but thus clear analog in strategic management, where
also implicitly suggests that the individual and the field’s raison d’être, as we have discussed, is
collective do not exist as real, separate entities to uncover the relative importance of different
(cf. Archer, 1995). The argument of level-“less” sources of performance heterogeneity (e.g., Bow-
and individual-“less” structure deflates the on- man & Helfat, 2001; McGahan & Porter, 1997;
tological depth of organizing by completely Rumelt, 1991). In past research scholars have
questioning, in postmodern fashion, the very ex- done variance decomposition studies in an ef-
istence of the individual, for the process, prac- fort to try to understand the key source or level of
tice, and context are inseparable from the indi- heterogeneity—for example, firm versus indus-
vidual (Giddens, 1979, 1985). try. Just as early industry studies that failed to
Furthermore, as noted by Archer, structuration account for lower-level (i.e., firm) effects errone-
theory “throws a blanket over the two constitu- ously overattributed causality to industry fac-
ents, structure and agency, which only serves to tors, we argue that knowledge studies that focus
prevent us from examining what is going on on collective levels without accounting for indi-
beneath it” (1995: 102; see Sawyer, 2002, for fur- viduals face a similar problem. Individual ef-
ther discussion). Overall, given our theoretical fects have not been ruled out as an alternative
arguments and empirical support for the critical explanation in most strategic studies of collec-
importance of a core, innate self, sans social tive or firm-level knowledge. Thus, current col-
construction, the structurationist perspective lectivist explanations may in some cases merely
does not resolve central problems for address- capture what are really the effects of differing
ing the locus of knowledge problem. Further- individual inputs in skills and knowledge.
more, we fundamentally concur with both Pop-
per (1957) and Simon (1985), for whom the notion
Locus Problem in Biopharma
of the individual seems uncontroversial and ab-
solutely fundamental in theory building. To ground our analysis of the locus problem
We should note that it appears the interac- and to explicate the attributional problems and
tionist approaches, such as the knowledge prac- associated levels-related assumptions dis-
tice literature, may be fundamentally incompat- cussed above, we highlight empirical studies
ible with the goals of scientific realism, as noted that give primacy to various levels of analysis in
by both Weick (1979) and McKelvey (1997). While the biotechnology and pharmaceutical industry
interactionist approaches focus on the context- (biopharma). Of particular note in this large
specific description of process, scientific realism body of work in the biopharma sector is the
focuses on generalizability, prediction, parsi- emphasis that different scholars have given to
mony, and objective measurement (e.g., Weick, disparate, contradictory levels as the key level
1979: 35)— our goal clearly being the latter. Our of analysis, or the key source of new value. The
208 Academy of Management Review January

biopharma industry provides a good setting for degree, and perhaps a great degree, a covarying
highlighting important assumptions with re- reflection of the compositional quality of the
gard to the knowledge locus problem, since the nodes in a network, rather than the final causal
industry has created tremendous amounts of source of heterogeneity. To take an extreme ex-
new value over the past few decades and has ample to illustrate the point, consider a newly
also been the subject of significant academic emerging but promising high-technology com-
interest at several levels of analysis, including pany with a potential compound to cure cancer.
industry, network, firm, research program, and A firm with such valuable knowledge will at-
individual (e.g., Baum, Calabrese, & Silverman, tract a network commensurate with its own ca-
2000; Gulati & Higgins, 2002; Henderson & Cock- pabilities and future potential. That is, a priori
burn, 1994; Kaplan, Murray, & Henderson, 2003; firm-level heterogeneity may determine network
Pisano, 1994; Powell et al., 1996; Rothaermel, structure, which is only a proximal cause, and,
2001; Zucker, Darby, & Brewer, 1998)—see Table 2 thus, assuming lower-level homogeneity (as is
for a summary of representative articles. Our done in the networks and alliances literature)
purpose here is simply to highlight the diver- simply confounds the analysis for understand-
gent conclusions that scholars have come to in ing the sources of new value (cf. Dansereau et
analyzing this industry and to discuss these in al., 1999).
light of the knowledge locus problem. We spe- The practical implications of network analy-
cifically highlight that lower levels and individ- sis readily reflect the artifactual problems of the
uals can provide an alternative explanation to network and alliances literature. The prescrip-
numerous collective outcomes by briefly dis- tive advice for young biotechnology companies
cussing research at the level of alliances and is to ally with the top (or most central) pharma-
networks (e.g., Powell et al., 1996), firms (e.g., ceutical company (or vice versa). Yet, arguably,
Henderson & Cockburn, 1994), and individuals only the most promising new firms in terms of
(e.g., Zucker et al., 1998). either discoveries or scientists with outstanding
Networks and alliances. Powell et al. unequiv- track records are able to gain the most favorable
ocally argue that “the locus of innovation is network positions. Thus, this recommendation
found within the networks of interorganizational points out that network effects may simply re-
relationships that sustain a fluid and evolving flect underlying lower-level heterogeneity or,
community” (1996: 142; also see Kogut, 2000). specifically, what particular firms have to offer
Subsequent articles have further pointed out the each other, rather than an emergent opportu-
need for biotechnology companies to “not go it nity. Network position, while certainly increas-
alone” (e.g., Baum et al., 2000; Rothaermel, 2001). ing the chances for success or survival, is poten-
Specifically, Baum et al. (2000: 287) argue that tially another post hoc artifactual measure of
the establishment of diverse alliances with es- underlying firm capabilities, but most likely not
tablished rivals leads to increased performance. the true source of a potential advantage.7 Over-
Network and alliance relations have also re- all, while networks may provide a component in
ceived significant theoretical emphasis as the firm performance (e.g., Dyer & Singh, 1998; see
key locus and source of heterogeneity (Adler, Kogut, 2000, for a general overview), we argue
2001; Dyer & Singh, 1998; Kogut, 2000). Adler, for that there must be some a priori, hierarchically
example, suggests that “the consensus in the nested rationale for exchange, and, thus, these
field is that the proliferation [of interfirm rela- studies prove rather descriptive since they de-
tions] is driven in large measure by the chal- scribe relationships that are artifacts of compo-
lenge of growing knowledge intensity” (2001: sitional qualities. In other words, overall initial
224). conditions (Poincaré, 1903) lead to a path-
A full exegesis of the rapidly growing alli-
ances literature is not feasible, but a couple of
problems are readily apparent and related to 7
We should note that some network analysis does control
the locus of knowledge problem. First, networks for various firm effects—for example, age and size (Powell et
al., 1996)—but these controls do not properly account for the
can be construed as artifacts, or, put differently,
underlying heterogeneity in terms of ability and potential,
networks may only be proximate causes of new as our practical example readily points out. Thus, an unmea-
value, while more final explanations lie nested sured lower-level confound may drive the purported higher-
within the firm. That is, networks are, to some level (network) effects.
2007 Felin and Hesterly 209

TABLE 2
Nested, Alternative Sources of Knowledge Locus Heterogeneity in Biopharma

Locus or Key
Source of Independent
Heterogeneity Representative Articles Variables Representative Quote

Collectives— Baum, Calabrese, & Silverman Networks/alliances “The locus of innovation will be found
alliances/networks (2000), Powell, Koput, & and in networks of learning, rather than
Smith-Doerr (1996), communities of in individual firms” (Powell et al.,
Rothaermel (2001), Shan, learning 1996: 116).
Walker, & Kogut (1994)
“We have argued that in a field of
rapid technological development,
such as biotechnology, the locus of
innovation is found within the
networks of interorganizational
relationships that sustain a fluid
and evolving community” (Powell et
al., 1996: 142).
Collectives— firm/ Cockburn, Henderson, & Stern Capabilities—e.g., “Our results suggest that a focus on
organization (2000), DeCarolis & Deeds promotion based architectural or integrative or
(1999), Henderson & Cockburn on publication, combinative capabilities as a
(1994, 1996), Pisano (1994), R&D spending source of competitive advantage
Yeoh & Roth (1999) may provide useful insights into the
sources of competitive advantage”
(Henderson & Cockburn, 1994: 77).
Individuals— Higgins & Gulati (2003), Two streams: (1) “Our findings are consistent with the
managers Kaplan, Murray, & Henderson managerial hypothesis that would suggest that
(2003) cognition and managerial sensemaking
skills; (2) upper (recognition and interpretation) of
echelon the environment may be an
affiliations and additional explanatory factor in
ties, managerial understanding firm fate and
networks performance during periods of
technological discontinuity” (Kaplan
et al., 2003).
Individuals— Audretsch & Stephan (1996), Scientists’ “Scientific breakthroughs are created
scientists Darby & Brewer (1998), Darby abilities, skills, by, embodied in, and applied
& Zucker (2003), Lacetera, cognition commercially by particular
Cockburn, & Henderson individuals responding to
(2004), Liebeskind, Oliver, incentives and working in specific
Zucker, & Brewer (1996), organizations and locations” (Zucker
Zucker, Darby, & Brewer & Darby, 1998).
(1998) “Until recently, economists and
sociologists studying science and
technology have been averse to
viewing scientists—particularly top
scientists—as pursuing private
motives, viewing them instead as
disinterested contributors to a
shared common pool of knowledge.
Our results suggest that star
scientists often are better viewed as
entrepreneurial individuals who
value both financial rewards and
the pleasure, recognition, and
resources that come from being the
first to make a significant new
discovery” (Zucker & Darby, 1998).
210 Academy of Management Review January

dependent accumulation of resources in general that scientists’ publications are a function of the
and network relationships more specifically. extent to which organizations encourage publi-
The firm. Moving from networks to lower lev- cation (through ties to promotion) rather than the
els, we next consider work at the organizational individual scientist’s ability (again assumed to
level or, in the case of Henderson and Cockburn be homogeneous). Individual heterogeneity pro-
(1994), the program level. We first briefly revisit vides an important alternative explanation for
the highly cited and foundational article of Hen- both the independent variable (promotion based
derson and Cockburn (1994).8 The article demon- on publication) and the dependent variable (pat-
strates some of the potential confounding expla- enting). Stern (2004), for example, specifically
nations that can result when researchers do not points out the significant a priori innate hetero-
account for heterogeneous individual-level ef- geneity of these very scientists, who self-select
fects. As a counterpoint to Henderson and Cock- into particular organizations, which points out a
burn, we subsequently highlight some recent completely confounding explanation that needs
work by Zucker and colleagues (e.g., Darby & to be controlled for. Although numerous other
Zucker, 2003). The purpose here is not to address confounds exist (e.g., because of a policy to pro-
all aspects of Henderson and Cockburn’s paper mote based on publication, scientists may self-
but to briefly illustrate the assumptions made in select, or scientists simply may create their own
light of the levels literature and the knowledge environment or firms), the bottom line is that
locus problem. Henderson and Cockburn’s (1994) individual-level initial conditions play an im-
article serves as a good benchmark and point of portant confounding role in subsequent attribu-
comparison, because it is both widely cited and tions of innovation performance.
celebrated for its theoretical and empirical con- We strongly suspect that the underlying abil-
tributions (e.g., Barney, 2001: 46). ities of individual scientists may well be more
Building on earlier work highlighting the pri- important than firm capabilities. Thus, individ-
macy of organization-level capabilities in R&D- ual ability and self-selection on the part of sci-
intensive environments, Henderson and Cock- entists potentially provide a critical primary ef-
burn begin with a distinctly collectivist premise fect. In any case, empirically, it simply is not
by looking at heterogeneous firm-level compe- possible to make claims for collective knowl-
tencies as drivers of innovation (Henderson & edge or competencies without accounting for in-
Cockburn, 1994: 63). They proceed to show, for dividuals (particularly given the remarkable a
example, that the extent to which external pub- priori heterogeneity that is even anecdotally ev-
lication plays a key role in promotion is a sig- ident).
nificant organizational competence that leads to Second, the four organizational competencies
increased innovation (proxied by patenting). (independent variables) that are highlighted by
Henderson and Cockburn’s analysis illus- their work (Henderson & Cockburn, 1994)—(1)
trates several important points. First, they do not publication playing a key role in promotion, (2)
account for individual-level variation in scien- rich information flow, (3) single individual mak-
tists and, thus, explicitly assume a priori homo- ing resource decisions, and (4) worldwide re-
geneity in individuals (Henderson & Cockburn, search managed as an integrated whole and the
1994: 79). More specifically, they assume that dependent variable of the number of patents—
heterogeneity in organizational practices is the can all be feasibly reduced to or reinterpreted as
primary, collective-level heterogeneous driver a result of heterogeneity from lower levels.
of performance outcomes. The assumption is Thus, the individual-level explanation provides,
arguably, a more parsimonious theoretical solu-
8
tion.
Henderson and Cockburn’s (1994) paper is among the
most highly cited works in the knowledge and capabilities- The individual. The findings and conclusions
based literature, combining both theoretical and empirical of Henderson and Cockburn (1994), as well as
elements. The authors have further built on this work with those pointing to the primacy of alliances/
numerous related publications, also well received and cited networks (Powell et al., 1996), differ markedly
(e.g., Cockburn, Henderson, & Stern, 2000; Henderson & Cock-
from the arguments of Zucker and colleagues
burn, 1996). Furthermore, Henderson and Cockburn’s (1994)
paper has been highlighted by Barney (2001: 46) as an ex- (e.g., Darby & Zucker, 2003; Liebeskind, Oliver,
emplary empirical paper measuring resource-based argu- Zucker, & Brewer, 1996; Zucker et al., 1998). In an
ments. unusually comprehensive study of an industry
2007 Felin and Hesterly 211

(including multiple methods, in over a decade of nies, which further commercialize these prod-
work), Zucker and Darby unequivocally settle on ucts. The primacy they give to individuals dif-
the primacy of a few key individuals in creating fers markedly from much of the collectivist
and embodying knowledge. biopharma research, which assumes individual
Scientific breakthroughs are created by, embod- homogeneity (Henderson & Cockburn, 1994: 79)
ied in, and applied commercially by particular and the primacy of collective heterogeneity (rou-
individuals responding to incentives and work- tines, network, competence, etc.). It should be
ing in specific organizations and locations; it is noted that not all of the causal links are com-
misleading to think of scientific breakthroughs as pletely disentangled in Zucker and colleagues’
disembodied information which, once discov-
ered, is transmitted by a contagion-like process work. To be sure, Zucker et al. partially infer that
in which the identities of the people involved are certain incentives (presumably organizational)
largely irrelevant (Zucker & Darby, 1995: 1). drive knowledge outcomes, although we suspect
that heterogeneity in individuals rather than
Zucker and colleagues emphasize heteroge-
heterogeneity in routines is driving new value
neous individuals as the key causal driver and
creation, particularly where these superstars
locus of knowledge in biotechnology (e.g.,
are largely responsible for starting these com-
Zucker et al., 1998).9 They specifically point to
panies, which, of course, further confounds the
the role that superstars play in creating new
analysis.
value by showing that certain scientist-entre-
preneurs are central in starting new companies
and are even the antecedent to the subsequent
formation of regional biopharma clusters (e.g., Nested Heterogeneity
Zucker et al., 1998). Based on their findings (e.g.,
Research and practice are replete with empir-
Zucker & Darby, 1998), in light of assumptions
ical and anecdotal evidence of the primacy of
about knowledge creation, they furthermore
individuals as the locus of knowledge and
suggest that
source of new value. However, decomposition
until recently, economists and sociologists study- studies that disentangle individual versus or-
ing science and technology have been averse to ganizational effects are lacking. Nevertheless,
viewing scientists—particularly top scientists—
as pursuing private motives, viewing them in- some relevant empirical work deserves consid-
stead as disinterested contributors to a shared eration. With respect to the remarkable hetero-
common pool of knowledge. Our results suggest geneity of individuals, it was Lotka (1926) who
that star scientists often are better viewed as first noted the highly skewed distribution and
entrepreneurial individuals who value both fi- heterogeneity of innovative output among scien-
nancial rewards and the pleasure, recognition,
and resources that come from being the first to tists. His early work, looking at scientific publi-
make a significant new discovery (1998). cations in chemistry over a seven-year period,
showed that roughly 5 percent of scientists were
Overall, the work of Zucker and colleagues responsible for more than half the scientific out-
provides evidence for the role of certain highly put. This analysis did not account for or weight
skilled scientists as the locus of knowledge, output in terms of quality, which points to sig-
since they not only play a significant role in nificantly higher concentrations in scientific
innovation outcomes but also in starting compa- output and, thus, substantial heterogeneity.
Narin and Breitzman (1995), in their analysis of
9
Jewkes, Sawers, and Stillerman, in their study of numer-
scientists in four large semiconductor compa-
ous innovations, concur with the above analysis and con- nies, showed that a few key scientists, akin to
clude that the Lotka distribution (but with an even higher
it is the practice of some writers to present a fuzzy concentration), were responsible for most of
picture of invention as a “social process”; to suggest those firms’ innovation output (as measured by
that, if one inventor had not done what he did when he patents). More recently, an analysis of forty-
did, someone else would have done it. . . . this attitude— three German companies in three disparate in-
that nothing can be understood unless all is under-
dustries (chemical, mechanical, electrical) also
stood, that by piling one unresolved enigma upon an-
other some all-comprehending solution is made the pointed to a similar distribution (Ernst, Leptien,
more likely—involves the error of “seeing depth in mere & Vitt, 2000; see also Huber, 1999, and Stolpe,
darkness”, as Sir Isaiah Berlin once put it (1969: 26 –27). 2002). Although the above evidence cannot con-
212 Academy of Management Review January

clusively be argued as undeniable support for the primary research setting for knowledge-
the primacy of the individual (given interaction based work. We should note that recent work by
effects), it nevertheless strongly suggests that Teece (2003) also points out the importance of
nested individual-level heterogeneity may pro- key individuals, such as experts and profession-
vide a significant alternative explanation and als, as the locus of knowledge in service-
confound to collective-level arguments, which oriented companies. Experts and professionals
presume individual-level homogeneity. do not rely on traditional firm-type organiza-
The above discussion leads to a hypothetical tions but, rather, are organizations unto them-
litmus test for the locus of knowledge, which selves, often structured in such disaggregated
may shed light on the locus problem and poten- forms as partnerships (cf. Zenger & Hesterly,
tial key sources of new value. That is, what 1997). This increased disaggregation indeed has
happens to the organization and its advantage been driven by an increased ability to measure
if a key individual leaves the organization? performance at lower levels, thus allowing more
Based on current collectivist arguments in the precise imputation of inputs, even in team pro-
KBV, organizations can withstand “consider- duction environments (Alchian & Demsetz, 1972).
able” turnover, given the emphasis placed on The incentives as well as autonomy and flexi-
heterogeneous, collective-level routines and, bility of disaggregated forms give control to
thus, homogeneous individuals (Levitt & March, these highly skilled individuals. Overall, this
1988: 320; also see Kogut & Zander, 1992: 383, and disaggregation and the relative importance
Nelson & Winter, 1982). given to individuals (Teece, 2003; Zenger & Hes-
However, some rather interesting, albeit pre- terly, 1997) are quite contrary to the communal
liminary, work has been done looking at the knowledge arrangements anticipated and dis-
turnover of key scientists in technology-moti- cussed by collectivist scholars such as Adler
vated acquisitions. This research sheds some (2001).
light on our litmus test, providing a natural ex-
periment to test the relative importance of indi-
IMPLICATIONS AND FUTURE DIRECTIONS
viduals versus collective structures. Granstrand
and Sjoelander (1990), for example, found that The full implications of the current collectivist
acquisitions failed to generate new value when knowledge-based research agenda, its logical
key scientists left the organization. In a more future directions, and assumptions about new
comprehensive analysis of scientist turnover, value creation and the locus of knowledge were
Ernst and Vitt (2000) further found that after key recently summarized and advocated by Howard
scientists left a company after an acquisition, Aldrich as follows: “If we truly focused on rou-
there was a significant overall decrease in the tines, competencies, practices and so on, we
innovative output for the organization as a would not follow people anymore in our re-
whole. Although these studies are preliminary search. Instead we would follow how competen-
and are missing some controls, they neverthe- cies spread, replicate, and insinuate themselves
less point toward the relative greater impor- into organizations. People would disappear
tance of individuals. from our equations” (Murmann et al., 2003: 27).
Finally, empirical work on “organizational dis- Given our arguments for the importance of a
economies” (the observation that small compa- priori, innate, and individual-level knowledge,
nies are proportionally much more innovative) we fundamentally question this projected path.
between large and small companies has also Quite contrary to collectivist work, we see prom-
pointed to the primacy of highly productive in- ising knowledge-based work explicitly focusing
dividuals. That is, the best engineering talent on the role that individuals play in creating and
readily moves to either start or join smaller com- embodying new value. We specifically focus on
panies (Zenger, 1994), which suggests that self- three critically important areas for future work
selection or an assortive mechanism based on and their implications with regard to new value
heterogeneous individual-level abilities is at creation: (1) property rights and appropriation,
the root of much new value creation. (2) individual mobility, and (3) tests of nested
The above examples largely have been drawn heterogeneity.
from R&D-intensive settings, such as biopharma First, our emphasis on the individual level
and semiconductors, given that these have been from the perspective of new value creation and
2007 Felin and Hesterly 213

knowledge highlights the need to understand strategy literature (Song, Almeida, & Wu, 2003:
such questions as who creates and owns new 352).10
value and who captures that value—implicating With regard to individual mobility, it should
both the property rights (e.g., Liebeskind, 1996) be noted that there is a readily apparent, locus-
and rent appropriation literature (Coff, 1999). related disconnect between collectivist theory
The shift to more carefully imputing new value and individual-level empirical measurement,
to nested sources within collective settings re- which our theoretical arguments can begin to
flects the fact that knowledge and associated alleviate. For example, Lacetera et al. (2004: 3– 4)
advantages inherently are created by and em- wrestle with problems of causality and mea-
bodied in specific individuals (lest we reify the surement related to the knowledge locus prob-
organization). Furthermore, knowledge-based lem by citing and attempting to build on collec-
tivist knowledge-based theory, yet arguing that
advantages are increasingly rooted in experts,
new capabilities reside in individuals (see also
superstars, and other highly productive individ-
Song et al., 2003). Thus, while there is a prevail-
uals (Teece, 2003). Thus, opening up the prover-
ing assumption in much collectivist work that
bial black box of the firm by explicating the
knowledge belongs to the firm (e.g., patents) and
underlying a priori capabilities and knowledge is the result of heterogeneous organizational ca-
of the individuals involved provides a natural pabilities (Henderson & Cockburn, 1994), never-
starting point and microfoundation for explain- theless, the underlying (perhaps more) heteroge-
ing the creation of new value. Moreover, re- neous abilities of the individuals creating this
source-based logic also is still firmly rooted in knowledge do not get measured or controlled.
the firm level (Barney, 2001), while more final Furthermore, quite counter to their theoretical
theoretical explanations (compared to proxi- assumptions, collectivist theories readily at-
mate explanations) of the origins of resources tribute new capabilities to specific individuals
would seem to lie at the individual level. For coming into the organization—an inherent con-
example, (purposeful) resource endowments tradiction. Overall, this research points out that
must inherently be the result of individual-level our collective theories need to be carefully re-
insights and abilities (cf. Coff, 1999). More visited, the dichotomy between collective theory
broadly, any organizational output should, in and individual empirical measurement being
theory, be imputable to specific individuals, and fundamentally incompatible. Our arguments for
understanding the various individuals’ respective a potential individual locus of knowledge pro-
contributions is paramount to explaining the col- vide at least an initial step toward more care-
lective outcome (cf. Alchian & Demsetz, 1972). fully understanding the underlying theoretical
The second critical implication of our argu- causalities of new value creation.
ments, given the emphasis we have placed on The third readily apparent implication and
the individual level in understanding knowl- future direction, which our arguments directly
edge and new value creation, concerns mobility. imply, is a comparative, empirical test of indi-
While collectivist knowledge and capabilities-
based work explicitly argues that individual 10
There has been some work on mobility, some of which
mobility is a “nonevent,” or, put differently, het- needs to be highlighted briefly. Almeida and Kogut (1999),
erogeneous routines are independent of the in- for example, show that the mobility of key engineers plays a
critical role in the diffusion of knowledge among firms. Spe-
dividuals who execute them (e.g., Kogut & cifically, knowledge is localized by region since key individ-
Zander, 1992: 383; Levitt & March, 1988: 320), we uals tend to be geographically constrained. Similar intuition
argue that this conceptualization is partially is also evident in the history of Silicon Valley, where knowl-
flawed. Specifically, who the organization is edge creation in the nascent semiconductor industry was
simply a function of certain individuals leaving their old
composed of is fundamental to overall out- companies and starting new ones, thus partially implying
comes. Thus, who turns over or who joins the that key capabilities, in fact, resided in these very employ-
organization has a significant impact on how ees (cf. Saxenian, 1994). Relatedly, Song et al. (2003; also see
the organization performs. The mobility of indi- Moen, 2004) empirically show how “learning-by-hiring”
plays a critical role in bringing in new knowledge and
viduals provides a natural opportunity for future capabilities into organizations. Furthermore, Lacetera et al.
research, and it overall has received little atten- (2004) show how new capabilities are built by bringing cer-
tion (given our collective-level theories) in the tain “star” individuals into an organization.
214 Academy of Management Review January

vidual versus collective effects in knowledge ply cannot rule out heterogeneous individuals
outcomes—similar to the competitive tests be- as an alternative explanation in explaining new
tween firm and industry (Rumelt, 1991). The find- value and knowledge creation. The observation
ing for the relative importance of the firm over that knowledge may to a significant extent be a
industry level has not ruled out nested sources priori, innate, and heterogeneous at the individ-
of heterogeneity arising from the individual ual level has important implications. Individual
level (cf. Bowman & Helfat, 2001). Just as the knowledge is not wholly (perhaps not even
observation that profitable companies belong to mostly) socially constructed or environmentally
certain industries was confounded by lower- determined, as is assumed in large part in the
level firm effects, which were assumed to be KBV, but, rather, there is a core self, which may
homogeneous, similarly, firm effects may be to a large degree determine learning and knowl-
confounded by heterogeneous individual-level edge outcomes. We have argued that this indi-
effects. The clear confounds arising from nested, vidual-level knowledge provides a significant
individual-level heterogeneity are readily ap- alternative explanation for much of collectivist
parent in our analysis of the biopharma indus- knowledge-based work, and we therefore chal-
try, where individuals are assumed to be lenge knowledge-based scholars to carefully re-
homogeneous and collective, heterogeneous ca- visit their underlying philosophical and theoret-
pabilities are given primacy (e.g., Henderson & ical assumptions about the primacy given to
Cockburn, 1994). collectives and to consider potential individual-
Overall, while recent work has pointed toward level explanations as antecedents to new value
a CEO effect (Bowman & Helfat, 2001), a more creation.
systematic approach that measures individual- Our analysis has limitations of course. First,
level performance more broadly is necessary to in many ways, we have raised some age-old
disentangle individual and collective effects. philosophical questions regarding the funda-
Empirical, multilevel data with individual-level mental origins of knowledge, which have yet to
performance measures (ideally, intake mea- be completely resolved. Given the one-sided na-
sures; e.g., Stern, 2004) or sufficient individual ture of the arguments in the extant knowledge
mobility from firm to firm would allow scholars and capabilities-based literature and recent
to disentangle the causal relationships that conflicting empirical findings in the cognitive
have proven so problematic. Based on our theo- sciences, we think our focus on the individual
retical arguments and recent findings in the level is justified, although future empirical work
cognitive sciences, we conjecture that much of sorting out individual and collective effects re-
the variance in knowledge can be attributed to mains to be done. Second, there has been a
relatively invariant and stable individual-level temptation on our part to attribute everything to
characteristics, which a priori are heteroge- initial conditions or the lowest level—that is,
neous or are significantly more heterogeneous individuals. We by no means doubt the exis-
than collective environments. We should briefly tence of collective-level effects, but, as we have
note here that our emphasis on the individual pointed out, individual-level controls are neces-
level is not to say that work at collective levels sary since they confound the analysis and po-
is unimportant. Just as findings of relatively tentially provide a more parsimonious explana-
more heterogeneity at the firm level (Rumelt, tion. Third, the present debates, both in the
1991) did not completely displace industry anal- disciplines that strategic management draws on
ysis, we do not expect or want collective-level and in the popular press, about the primacy of
analysis to be replaced. innateness (nature) versus environment (nur-
ture) in knowledge acquisition have obviously
led to heated discussions, which have had a
CONCLUSION
tendency to stir significant controversy and de-
In sum, we have argued that knowledge- bate. However, given that emphasis in knowl-
based theory and research must begin with in- edge-based work has been on nurture, collec-
dividuals rather than the collective level in un- tive, and environment (e.g., Nahapiet & Ghoshal,
derstanding new value creation. We have 1998: 247; Spender, 1996: 53), we have tried to
shown that collectivist approaches that do not objectively summarize and build on theoretical
account for individual-level heterogeneity sim- and empirical arguments from cognitive science
2007 Felin and Hesterly 215

that directly challenge collectivist conceptual- Bowman, E. H., & Helfat, C. E. 2001. Does corporate strategy
matter? Strategic Management Journal, 22: 1–23.
izations of knowledge and have critical implica-
tions for the future of the KBV and associated Brown, J. S., & Duguid, P. 1991. Organizational learning and
communities of practice: Toward a unified view of work-
efforts to understand new value creation.
ing, learning, and innovation. Organization Science, 2:
40 –57.
REFERENCES Brown, J. S., & Duguid P. 2001. Knowledge and organization:
A social-practice perspective. Organization Science, 12:
Adler, P. S. 2001. Market, hierarchy, and trust: The knowledge
198 –213.
economy and the future of capitalism. Organization Sci-
ence, 12: 215–234. Carlile, P. 2002. A pragmatic view of knowledge and bound-
aries: Boundary objects in new product development.
Alchian, A. A., & Demsetz, H. 1972. Production, information
Organization Science, 13: 442– 455.
costs, and economic organization. American Economic
Review, 62: 777–795. Castanias, R. P., & Helfat, C. E. 2001. The managerial rents
model: Theory and empirical analysis. Journal of Man-
Almeida, P., & Kogut, B. 1999. Localization of knowledge and
agement, 27: 661– 678.
the mobility of engineers in regional networks. Manage-
ment Science, 45: 905–917. Chomsky, N. 1959. A review of B. F. Skinner’s Verbal behav-
ior. Language, 35: 26 –58.
Archer, M. 1995. Realist social theory: The morphogenetic
approach. Cambridge: Cambridge University Press. Chomsky. N. 1986. Knowledge of language: Its nature, ori-
gins, and use. New York: Praeger.
Argote, L., & Ingram, P. 2000. Knowledge transfer: A basis for
competitive advantage in firms. Organizational Behav- Chomsky, N. 2000. New horizons in the study of language and
ior and Human Decision Processes, 82: 150 –169. mind. Cambridge: Cambridge University Press.
Arrow, K. J. 1962. Economic welfare and the allocation of Chomsky, N. 2002. Cartesian linguistics: A chapter in the
resources for invention. In R. R. Nelson (Ed.), The rate history of rationalist thought. Christchurch, New Zea-
and direction of inventive activity: 609 – 625. Princeton, land: Cybereditions.
NJ: Princeton University Press. Cockburn, I., Henderson, R., & Stern, S. 2000. Untangling the
Arrow, K. J. 1974. Limits of organization. New York: Norton. origins of competitive advantage. Strategic Manage-
ment Journal, 21: 1123–1145.
Audretsch, D. B., & Stephan, P. E. 1996. Company-scientist
locational links: The case of biotechnology. American Coff, R. W. 1999. When competitive advantage doesn’t lead to
Economic Review, 86: 641– 652. performance: Resource-based theory and stakeholder
bargaining power. Organization Science, 10: 119 –133.
Bacharach, S. B. 1989. Organizational theories: Some criteria
for evaluation. Academy of Management Review, 14: Coleman, J. S. 1990. Foundations of social theory. Cambridge,
496 –515. MA, London: Belknap Press of Harvard University Press.
Barney, J. B. 1991. Firm resources and sustained competitive Collis, D. J. 1994. How valuable are organizational capabili-
advantage. Journal of Management, 17: 99 –120. ties? Strategic Management Journal, 15: 143–152.
Barney, J. B. 2001. Is the resource-based view a useful per- Conner, K., & Prahalad, C. K. 1996. A resource-based theory
spective for strategic management research? Yes. Acad- of the firm: Knowledge versus opportunism. Organiza-
emy of Management Review, 26: 41–56. tion Science, 7: 477–501.
Baum, J., Calabrese, T., & Silverman, B. 2000. Don’t go it Dansereau, F., Yammarino, F. J., & Kohles, J. C. 1999. Multiple
alone: Alliance network composition and startups’ per- levels of analysis from a longitudinal perspective: Some
formance in Canadian biotechnology. Strategic Man- implications for theory building. Academy of Manage-
agement Journal, 21: 267–294. ment Review, 24: 346 –357.
Bouchard, T. J. 1997. IQ similarity in twins reared apart: Darby, M. R., & Zucker, L. G. 2003. Growing by leaps and
Findings and responses to critics. In R. J. Sternberg & inches: Creative destruction, real cost reduction, and
E. Grigorenko (Eds.), Intelligence, heredity, and environ- inching up. Economic Inquiry, 41: 1–19.
ment: 126 –160. Cambridge: Cambridge University Press. Davis-Blake, A., & Pfeffer, J. 1989. Just a mirage: The search
Bouchard, T. J., & McGue, M. 2003. Genetic and environmen- for dispositional effects in organizational research.
tal influences on human psychological differences. Jour- Academy of Management Review, 14: 385– 400.
nal of Neurobiology, 54: 4 – 45. DeCarolis, D. M., & Deeds, D. L. 1999. The impact of stocks
Boudon, R. 1998a. Social mechanisms without black boxes. In and flows of organizational knowledge on firm perfor-
P. Hedstrom & R. Swedberg (Eds.), Social mechanisms: mance: An empirical investigation of the biotechnology
An analytical approach to social theory: 1–31. Cam- industry. Strategic Management Journal, 20: 953–968.
bridge: Cambridge University Press. De Graaf, V. 1957. Theoretical welfare economics. Cam-
Boudon, R. 1998b. Limitations of rational choice theory. bridge: Cambridge University Press.
American Journal of Sociology, 104: 817– 828.
Donaldson, L. 1990. The ethereal hand: Organizational eco-
Bourdieu, P. 1977. Outline of a theory of practice. Cambridge: nomics and management theory. Academy of Manage-
Cambridge University Press. ment Review, 15: 369 –381.
216 Academy of Management Review January

Durkheim, E. 1952. Suicide: A study in sociology. London: Jencks, C. 1972. Inequality: A reassessment of the effect of
Routledge. family and schooling in America. New York: Basic Books.
Durkheim, E. 1962. The rules of the sociological method. Jewkes, J., Sawers, D., & Stillerman, R. 1969. The sources of
Glencoe, IL: Free Press. invention. New York: Norton.
Dyer, J., & Singh, H. 1998. The relational view: Cooperative Kaplan, A. 1964. The conduct of inquiry: Methodology for
strategy and sources of interorganizational competitive behavioral science. San Francisco: Chandler.
advantage. Academy of Management Review, 23: 660 –
Kaplan, S., Murray, F., & Henderson, R. 2003. Discontinuities
679.
and senior management: Assessing the role of recogni-
Eisenhardt, K., & Martin, J. 2000. Dynamic capabilities: What tion in pharmaceutical firm response to biotechnology.
are they? Strategic Management Journal, 21: 1105–1121. Industrial and Corporate Change, 12: 203–233.
Elster, J. 1989. Nuts and bolts for the social sciences. Cam- Kaplan, S., Schenkel, A., von Krogh, G., & Weber, C. 2001.
bridge: Cambridge University Press. Knowledge-based theories of the firm in strategic man-
Ernst, H., Leptien, C., & Vitt, J. 2000. Inventors are not alike: agement: A review and extension. Working paper No.
The distribution of patenting output among industrial 4216-01, Massachusetts Institute of Technology, Boston.
R&D personnel. IEEE Transactions on Engineering Man- Kim, J. 1993. Supervenience and mind. Cambridge: Cam-
agement, 47: 184 –199. bridge University Press.
Ernst, H., & Vitt, J. 2000. The influence of corporate acquisi- Kincaid, H. 1997. Individualism and the unity of science:
tions on the behaviour of key inventors. R&D Manage- Essays on reduction, explanation and the special sci-
ment, 30: 105–119. ences. Oxford: Rowman and Littlefield.
Giddens, A. 1979. Central problems in social theory: Action, Klein, K. J., Dansereau, F., & Hall, R. J. 1994. Levels issues in
structure and contradiction in social analysis. London: theory development, data collection, and analysis.
Macmillan. Academy of Management Review, 19: 195–229.
Giddens, A. 1984. The constitution of society: Outline of the Kogut, B., 2000. The network as knowledge: Generative rules
theory of structuration. Los Angeles & Berkeley: Univer- and the emergence of structure. Strategic Management
sity of California Press. Journal, 21: 405– 425.
Giddens, A. 1985. Marx’s correct views on everything. Theory Kogut, B., & Zander, U. 1992. Knowledge of the firm, combi-
and Society, 2: 167–174. native capabilities, and the replication of technology.
Granstrand, O., & Sjoelander, O. 1990. The acquisition of Organization Science, 3: 383–397.
technology and small firms by large firms. Journal of Kogut, B., & Zander, U. 1995. Knowledge, market failure and
Economic Behavior and Organization, 13: 367–386. the multinational enterprise: A reply. Journal of Interna-
Grant, R. 1996. Toward a knowledge-based theory of the firm. tional Business Studies, 26: 417– 426.
Strategic Management Journal, 17(Winter Special Issue): Kogut, B., & Zander, U. 1996. What firms do? Coordination,
109 –122. identity, and learning. Organization Science, 7: 502–518.
Hargadon, A., & Fanelli, A. 2002. Action and possibility: Rec- Lacetera, N., Cockburn, I., & Henderson, R. 2004. Do firms
onciling dual perspectives of knowledge in organiza- change capabilities by hiring new people? A study of
tions. Organization Science, 13: 290 –302. the adoption of science-based drug discovery. Advances
Hedstrom, P., & Swedberg, R. 1998. Social mechanisms: An in Strategic Management, 21: 133–159.
introductory essay. In P. Hedstrom & R. Swedberg (Eds.), Levitt, B., & March, J. 1988. Organizational learning. Annual
Social mechanisms: An analytical approach to social Review of Sociology, 14: 319 –340.
theory: 1–31. Cambridge: Cambridge University Press.
Lewis, D. 1986. Philosophical papers II. Oxford: Oxford Uni-
Hempel, C., & Oppenheim, P. 1948. Studies in the logic of versity Press.
explanation. Philosophy of Science, 15: 135–175.
Liebeskind, J. P., Oliver, A. L., Zucker, L., & Brewer, M. 1996.
Henderson, R., & Cockburn, I. 1994. Measuring competence? Social networks, learning and flexibility: Sourcing sci-
Exploring firm effects in pharmaceutical research. Stra- entific knowledge among new biotechnology firms. Or-
tegic Management Journal, 15: 63– 84. ganization Science, 7: 428 – 443.
Henderson, R., & Cockburn, I. 1996. Scale, scope and spill- Lotka, A. 1926. The frequency distribution of scientific pro-
overs: The determinants of research productivity in drug ductivity. Journal of Washington Academy of Science, 16:
discovery. Rand Journal of Economics, 27: 32–59. 317–323.
Higgins, M., & Gulati, R. 2003. Getting off to a good start: The Lubinski, D. 2000. Scientific and social significance of assess-
effects of upper echelon affiliation on underwriter pres- ing individual differences: “Sinking shafts at a few critical
tige. Organization Science, 14: 244 –263. points.” Annual Review of Psychology, 51: 405– 444.
House, R. J., Shane, S. A., & Herold, D. M. 1996. Rumors of the Lubinski, D. 2004. Introduction to the special section on cog-
death of dispositional research are vastly exaggerated. nitive abilities: 100 years after Spearman’s (1904) “Gen-
Academy of Management Review, 21: 203–224. eral intelligence,” objectively determined and mea-
Huber J. C. 1999. Inventive productivity and the statistics of sured. Journal of Personality and Social Psychology, 86:
exceedances. Scientometrics, 45(11): 33–53. 96 –111.
2007 Felin and Hesterly 217

Madhok, A. 1997. Cost, value and foreign market entry mode: Popper, K. R. 1957. The poverty of historicism. London: Rout-
The transaction and the firm. Strategic Management ledge & Kegan Paul.
Journal, 18: 39 – 61. Popper, K. R. 1959. The logic of scientific discovery. London:
March, J. G. 1991. Exploration and exploitation in organiza- Hutchinson.
tional learning. Organization Science, 2: 71– 87. Popper, K. R. 1968. Conjectures and refutations: The growth of
McGahan, A. M., & Porter, M. 1997. How much does industry scientific knowledge. New York: Harper & Row.
matter, really? Strategic Management Journal, 18(Spe- Powell, W. W., Koput, K. W., & Smith-Doerr, L. 1996. Interor-
cial Issue): 15–30. ganizational collaboration and the locus of innovation:
McGue, M., & Bouchard, T. J. 1998. Genetic and environmen- Networks of learning in biotechnology. Administrative
tal influences on human behavioral differences. Annual Science Quarterly, 41: 116 –145.
Review of Neuroscience, 21: 1–24. Rosenberg, A. 1995. The philosophy of social science. Boul-
McKelvey, B. 1997. Quasi-natural organization science. Or- der, CO: Westview Press.
ganization Science, 8: 352–380. Rothaermal, F. 2001. Incumbent’s advantage through exploit-
Moen, J. 2004. Is mobility of technical personnel a source of ing complementary assets via interfirm cooperation.
R&D spillovers? Journal of Labor Economics, 23: 81–114. Strategic Management Journal, 22: 687– 699.
Rousseau, D. M. 1985. Issues of level in organizational re-
Murmann, J. P., Aldrich, H., Levinthal, D., & Winter, S. 2003.
search: Multi-level and cross-level perspectives. Re-
Evolutionary thought in management and organization
search in Organizational Behavior, 7: 1–37.
theory at the beginning of the new millennium. Journal
of Management Inquiry, 12: 1–19. Rumelt, R. P. 1991. How much does industry matter? Strategic
Management Journal, 12: 167–185.
Nagel, E. 1961. The structure of science: Problems in the logic
of scientific explanation. New York: Harcourt, Brace & Sawyer, R. K. 2001. Emergence in sociology: Contemporary
World. philosophy of mind and some implications for sociolog-
ical theory. American Journal of Sociology, 107: 551–585.
Nahapiet, J., & Ghoshal, S. 1998. Social capital, intellectual
capital, and the organizational advantage. Academy of Sawyer, R. K. 2002. Unresolved tensions in sociocultural the-
Management Review, 23: 242–266. ory: Analogies with comtemporary sociological debates.
Culture & Psychology, 8: 283–305.
Narin, F., & Breitzman, A. 1995. Inventive productivity. Re-
search Policy, 24: 507–519. Saxenian, A. 1994. Regional advantage: Culture and compe-
tition in Silicon Valley and Route 128. Cambridge, MA, &
Nelson, R. R., & Winter, S. 1982. An evolutionary theory of London: Harvard University Press.
economic change. Cambridge, MA: Harvard University
Press. Scaltsas, T. 1990. Is the whole identical to its parts? Mind, 99:
583–598.
Nonaka, I. 1994. A dynamic theory of organizational knowl-
Scarr, S., & McCartney, K. 1983. How people make their own
edge creation. Organization Science, 5: 14 –37.
environments: A theory of genotype-environment ef-
O’Neill, J. (Ed.). 1973. Modes of individualism and collectiv- fects. Child Development, 54: 424 – 435.
ism. London: Heinemann Educational Books.
Scarr, S. 1997. Behavior-genetic and socialization theories of
Oppenheim, P., & Putnam, H. 1958. Unity of science as a intelligence: Truce and reconciliation. In R. J. Sternberg
working hypothesis. In H. Feigl, M. Scriven, & G. Max- & E. Grigorenko (Eds.), Intelligence, heredity, and envi-
well (Eds.), Minnesota studies in the philosophy of sci- ronment: 3– 41. Cambridge: Cambridge University Press.
ence: 3–36. Minneapolis: University of Minnesota Press.
Schneider, B. 1987. The people make the place. Personnel
Orlikowski, W. J. 2002. Knowing in practice: Enacting a col- Psychology, 40: 437– 454.
lective capability in distributed organizing. Organiza- Shan, W. J., Walker, G., & Kogut, B. 1994. Interfirm coopera-
tion Science, 13: 249 –273. tion and startup innovation in the biotechnology indus-
Ouchi, W. G., & Johnson, J. B. 1978. Types of organizational try. Strategic Management Journal, 15: 387–394.
control and their relationship to emotional well being. Simon, H. A. 1985. Human nature in politics. American Polit-
Administrative Science Quarterly, 23: 293–317. ical Science Review, 79: 293–304.
Pentland, B. 1995. Grammatical models of organizational Simon, H. A. 1991. Bounded rationality and organizational
processes. Organization Science, 6: 541–556. learning. Organization Science, 2: 125–134.
Pisano, G. 1994. Knowledge, integration and the locus of Skinner, B. F. 1957. Verbal behavior. New York: Appleton-
learning: An empirical analysis of process development. Century-Crofts.
Strategic Management Journal, 15: 85–100.
Song, J., Almeida, P., & Wu, G. 2003. Learning-by-hiring:
Poincaré, H. 1903. Science and hypothesis. London: Walter When is mobility more likely to facilitate interfirm
Scott. knowledge transfer? Management Science, 49: 351–365.
Poole, M. S., & Van de Ven, A. H. 1989. Using paradox to build Spelke, E. S., Breinlinger, K., Macomber, J., & Jacobson, K.
management and organization theories. Academy of 1992. Origins of knowledge. Psychological Review, 99:
Management Review, 14: 562–578. 605– 632.
218 Academy of Management Review January

Spender, J. C. 1996. Making knowledge the basis of a dy- Whetten, D. A. 1989. What constitutes a theoretical contribu-
namic theory of the firm. Strategic Management Journal, tion? Academy of Management Review, 14: 490 – 495.
17(Winter Special Issue): 45– 62.
Winter, S. G. 2003. Understanding dynamic capabilities.
Stanovich, K. 1999. Who is rational: Studies of individual Strategic Management Journal, 24: 991–995.
differences in reasoning. Mahwah, NJ: Lawrence Erl-
Wrong, D. 1961. The oversocialized conception of man in
baum Associates.
modern sociology. American Sociological Review, 26:
Stern, S. 2004. Do scientists pay to be scientists? Manage- 183–193.
ment Science, 50: 835– 853.
Yeoh, P. L., & Roth, K. 1999. An empirical analysis of sus-
Stolpe, M. 2002. Determinants of knowledge diffusion as ev- tained advantage in the U.S. pharmeceutical industry:
idenced in patent data: The case of liquid crystal tech- Impact of firm resources and capabilities. Strategic
nology. Research Policy, 31: 1181–1198. Management Journal, 20: 637– 653.
Teece, D. J. 2003. Expert talent and the design of (professional
Zenger, T. R. 1994. Explaining organizational diseconomies
services) firms. Industrial and Corporate Change, 12: 895–916.
of scale in R&D: The allocation of engineering talent,
Teece, D. J., Pisano, G. P., & Shuen, A. 1997. Dynamic capa- ideas and effort by firm size. Management Science, 40:
bilities and strategic management. Strategic Manage- 708 –729.
ment Journal, 18: 509 –534.
Zenger, T. R., & Hesterly, W. 1997. The disaggregation of
Tsoukas, H. 1996. The firm as a distributed knowledge sys- corporations: Selective intervention, high-powered in-
tem: A constructionist approach. Strategic Management centives, and molecular units. Organization Science, 8:
Journal, 17(Winter Special Issue): 11–25. 209 –222.
Tuomela, R. 1990. Methodological individualism and expla- Zollo, M., & Winter, S. G. 2002. Deliberate learning and the
nation. Philosophy of Science, 57: 133–140. evolution of dynamic capabilities. Organization Sci-
Udehn, L. 2001. Methodological individualism: Background, ence, 13: 339 –352.
history and meaning. London & New York: Routledge.
Zucker, L., & Darby, M. 1995. Virtuous circles of productivity:
Weber, M. 1949. Max Weber on the methodology of the social Star bioscientists and the institutional transformation of
sciences. (Translated and edited by E. Shils & H. Finch.) industry. Working paper No. 5342, National Bureau of
Glencoe, IL: Free Press. Economic Research, Cambridge, MA.
Weick, K. E. 1979. The social psychology of organizing. Read- Zucker, L., & Darby, M. 1998. Entrepreneurs, star scientists
ing, MA: Addison-Wesley. and biotechnology. NBER Reporter Online, www.nber.
Weick, K. E. 1995. What theory is not, theorizing is. Adminis- org/reporter/fall98/zucker-darby_fall98.html, accessed
trative Science Quarterly, 40: 385–390. December 2002.
Weick, K. E., & Roberts, K. H. 1993. Collective mind in orga- Zucker, L., Darby, M., & Brewer, M. B. 1998. Intellectual hu-
nizations: Heedful interrelating on flight decks. Admin- man capital and the birth of U.S. biotechnology enter-
istrative Science Quarterly, 38: 357–381. prises. American Economic Review, 88: 290 –306.

Teppo Felin (teppo.felin@byu.edu) is an assistant professor of organizational leader-


ship and strategy at the Marriott School, Brigham Young University. He received his
Ph.D. in organization theory from the University of Utah. He studies organizational
learning and capabilities, epistemology, social theory, and the microfoundations of
organization.

William S. Hesterly (mgtwh@business.utah.edu) is the Ezekiel R. and Katherine W.


Dumke Professor of Management at the University of Utah’s David Eccles School of
Business, where he also serves as associate dean of academic affairs. He received his
Ph.D. from the University of California, Los Angeles. He studies the creation and
appropriation of economic value.