Content uploaded by Sungho Kim
Author content
All content in this area was uploaded by Sungho Kim on Apr 25, 2018
Content may be subject to copyright.
KNOWLEDGE COMPLEXITY AND THE PERFORMANCE OF INTER-UNIT KNOWLEDGE
REPLICATION STRUCTURES
Running head: Inter-unit Knowledge Replication Structures
Sungho Kim*
Assistant Professor of Strategy and Organization
School of Business
Southern Illinois University
Edwardsville, IL, USA 62025
sukim@siue.edu
Jaideep (Jay) Anand
William H. Davis Chair and Dean's Distinguished Professor of Strategy
Fisher College of Business
The Ohio State University
Columbus, OH, USA 43210
anand_18@osu.edu
*Corresponding author
Forthcoming in the Strategic Management Journal
https://doi.org/10.1002/smj.2899
April 2018
Keywords: Replication, knowledge complexity, inter-unit structure, information processing, bounded
rationality
Acknowledgement: We are grateful to associate editor Jeff Reuer for his guidance and anonymous
reviewers who helped improve the manuscript. We also thank Vikas Aggarwal, Dan Levinthal, Hart Posen,
Kannan Srikanth and Jane Zhao, seminar participants at the University of Florida, HKUST, Indiana
University, NYU, Ohio State University, University of Pennsylvania, Purdue University, SUNY Buffalo and
participants in the 2014 Academy of Management Annual Meeting for their helpful comments on the earlier
versions of the manuscript.
ABSTRACT
Research Summary: Intra-firm replication of complex knowledge is difficult yet critical to firm growth and
the exploitation of competitive advantage. Inter-unit organizational structures can facilitate the replication of
complex knowledge between a source unit and a recipient unit. This study examines how inter-unit
organizational structures perform at different levels of knowledge complexity. We dimensionalize the
patterns of information-processing interactions according to three specific factors: the degree of inter-unit
connectivity, the extent of mirroring between the structure and the knowledge configuration, and
coordination mechanisms. Simulation analyses offer a set of novel findings on how the information-
processing and bounded-rationality concerns of organizational design impact the replication performance of
the structures. We derive optimal structures for different levels of knowledge complexity and articulate their
theoretical and practical implications.
Managerial Summary: The growth of firms often involves redeployment of their complex knowledge to
new sub-units or markets, in the context of acquisitions, alliances, or the creation of multinational
subsidiaries. Complex knowledge is difficult to imitate and thus serves as a source of competitive advantage.
However, it is also challenging to replicate within a firm, which limits firms’ ability to redeploy their
capabilities in pursuit of new opportunities. A proper design of inter-unit structures can facilitate the
replication of complex knowledge between intra-firm units. This study examines how the design of inter-unit
structures affects the outcome of this replication. Our results suggest that managers in charge of
redeployment efforts should be mindful of the connectivity between units, coordination mechanisms,
information overload, and the level of knowledge complexity.
Replication of complex knowledge is critical for firms’ successful growth, which involves redeploying their
relevant capabilities to new sub-units, markets, or other opportunities (Hansen, 2002; Penrose, 1959; Winter
and Szulanski, 2001). In order for organizational knowledge to underlie and sustain a firm’s competitive
advantage, it should be redeployable within the firm yet difficult for competitor firms to imitate (Grant,
1996; Spender, 1996). Redeployment of knowledge is an important aspect of using scale-free and fungible
resources (Anand, Kim, and Lu, 2016). Consequently, this issue has important implications for resource
redeployment, which may take the form of acquisitions, alliances, or multinational subsidiaries (Anand and
Delios, 2002; Anand and Singh, 1997; Anand, Oriani, and Vassolo, 2010; Barney, 1997; Capron, Mitchell,
and Swaminathan, 2001; Markides and Williamson, 1994). Complex knowledge is difficult to imitate, and
thus it serves as a plausible basis on which to establish a firm’s competitiveness (Reed and Defillippi, 1990;
Rivkin, 2000). On the other hand, complex knowledge is also challenging to replicate within a firm, which
limits firms’ ability to redeploy their relevant knowledge in pursuing new opportunities (Argote, Ingram,
Levine, and Moreland, 2000; Pil and Cohen, 2006; Szulanski, 1996).
While acknowledging the ease or difficulty of knowledge replication, previous research has seldom
unpacked the replication process. Little research has focused on understanding the implications of the design
of inter-unit organizational structures and knowledge complexity on replication outcomes. The design of the
structure influences the pattern of information-processing activities and the coordination of
interdependencies (Hansen, Mors, and Løvås, 2005; Puranam, Raveendran, and Knudsen, 2012; Szulanski,
2000), and thus, it can affect replication performance. Discriminating alignment between inter-unit structures
and knowledge complexity can be critical to ensuring that complex knowledge is replicated in an accurate
manner.
Various inter-unit structures for knowledge replication exist in real organizational settings. For
example, an inter-unit structure was developed for the redeployment of capabilities between a group of
Chinese engineers of Volkswagen’s joint venture in Shanghai and the German engineers of Volkswagen’s
home R&D unit (Zhao and Anand, 2009), between Banc One and its acquired units (Szulanski, 2000), and
between units of various auto manufacturers (Kenney and Florida, 2000; Prochno, 2003). While previous
research has identified specific inter-unit structures such as boundary spanners (Allen, Tushman, and Lee,
1979; Leifer and Delbecq, 1978), we do not yet have a good understanding of how different inter-unit
organizational structures will perform with varying degrees of knowledge complexity.
The purpose of this study is to examine how the design of inter-unit organizational structures affects
the performance of complex knowledge replication. Building upon two foundational theories—the
information-processing view of organizations, and bounded rationality (Galbraith, 1973; Puranam et al.,
2012; Tushman and Nadler, 1978)—as well as phenomenological studies (e.g., Allen, 1977; Galbraith, 1994;
Egelhoff, 1993; Gupta and Govindarajan, 1991; Hansen, 1999; Szulanski, 1996, 2000), the present study
dimensionalizes the structures according to their inter-unit connectivity (Aral and Alstyne, 2011), the extent
of mirroring between the organizational structure and knowledge configuration (Sanchez and Mahoney,
1996), and their coordination mechanisms (Ghoshal, Korine, and Szulanski, 1994; Kogut and Zander, 1996;
Puranam, Singh, and Chaudhuri, 2009). In order to single out the effects of these parameters, this study does
not involve other possible determinants of replication outcomes, such as incentives, opportunism, and trust.
We have adopted a formal modeling and agent-based simulation approach to better understand how the
design of inter-unit structures may affect the distribution of information load imposed on individuals as well
as the coordination of interdependence, and thus, replication accuracy.
THEORETICAL OVERVIEW
Replication of complex knowledge
In our study, knowledge replication means the redeployment of knowledge from a source unit to a recipient
unit. Figure 1a provides an illustration of the replication of complex knowledge between units. In this
example, there are four members in each unit. The unit can be a subunit such as a team, a department, or a
division. A large circle represents a unit, and small shaded circles signify the individual members of a unit.
Knowledge resides in a source unit’s members, who possess different areas of expertise. The members of a
unit may interact with each other through intra-unit ties, which are denoted by dotted lines between
individuals. To focus on the design of inter-unit structures, we assume that within each unit all
interdependencies are understood by its unit members and thus intra-unit ties are established between all unit
members that have knowledge interdependencies. With a simplifying assumption that each individual is
responsible for one area of knowledge expertise, the dotted lines also represent interactions among
knowledge domains. A shaded rectangle between the two units denotes an inter-unit organizational structure
for knowledge replication, and various designs of organizational structures are possible. To replicate
knowledge from a source unit to a recipient unit, the members of the units interact through an inter-unit
structure.
--- Insert Figure 1a about here ---
Knowledge can reside in an individual—known as discrete knowledge—but knowledge can also
consist of a set of interdependent components of knowledge that is held by multiple individuals (Spender,
1996). For example, in Figure 1a, knowledge at a source unit consists of four expertise areas. Each individual
member of the unit carries each specialized knowledge component, and interdependencies exist between
some of the knowledge components (e.g., between knowledge components 1 and 2 and 1 and 3). Replication
of such collectively held knowledge is more challenging than replication of discrete knowledge, primarily
because of the complexity of the knowledge, which stems from the interdependencies between knowledge
that is held by multiple individuals (Zhao and Anand, 2013).
Knowledge complexity is determined by the number of knowledge elements (N) and the total
number of interdependencies (Q) among the knowledge elements (Simon, 1962). In Figure 1a, among the
four knowledge components in a source unit, there are two pairs of interdependencies: one between s1 and s3,
and one between s1 and s4. Thus, N = 4 and Q = 2. For instance, the Pratt and Whitney’s design project for a
jet engine involves 569 interdependencies among 54 components (Sosa, Eppinger, and Rowles, 2007). The
interdependencies among expertise areas arise when one expertise area affects the contribution of other areas
of expertise to the overall outcome (Faraj and Sproull, 2000). The complexity of knowledge could be a
source of competitive advantage, as it raises a barrier to imitation (Reed and Defillippi, 1990; Rivkin, 2000).
However, for the same reason that complex knowledge is hard for competitors to imitate, it is also difficult to
scale up or redeploy within a firm (Rivkin, 2001).
A salient component of such complex knowledge is trans-specialist or cross-expertise knowledge
(Postrel, 2002; Zhao and Anand, 2013). Cross-expertise knowledge is an understanding of how an
individual’s decision in one expertise area may impact the effectiveness of another individual’s decision in a
different expertise area when they are involved in an interdependent task (Reagans and McEvily, 2003).
Cross-expertise knowledge is developed through information-processing interactions between individuals
with different expertise areas, and such interactions cannot often be reduced to codified interactions. In the
context of Figure 1a, due to the interdependence between s1 and s3, the decision that s1 makes affects the
optimality of decision s3 in that it influences the contribution of s3 to the overall effectiveness of the source
knowledge. Lack of cross-expertise knowledge can be a significant barrier for knowledge replication. Even
when individuals are collocated, the absence of cross-expertise knowledge erects communication barriers and
thus prohibits proper coordination between individuals with different areas of expertise (Allen, 1977; Sosa et
al., 2004).
Replication of complex knowledge often involves replicating coordination routines, not just
transferring functional knowledge (Postrel, 2002). For example, to transfer R&D and manufacturing
knowledge from a European auto manufacturer to a China-based auto facility, a group of engineers from the
source unit worked with a group of engineers with diverse expertise from the recipient unit through a joint
venture (Zhao and Anand, 2009). Such knowledge comprises not only technical and functional knowledge
but also knowledge on how to interact with unit members to solve problems or execute routines and how to
recombine interdependent knowledge (Carlile, 2002; Fiol and Lyles, 1985). Because of contextual
differences (e.g., customer preferences, road conditions, supply chains, and safety and environment
regulations), the recipient unit needed to recombine transferred knowledge by adapting coordination routines
as well as the source’s technical knowledge to the recipient’s context.
Inter-unit organizational structures
Due to knowledge complexity, the communication, coordination, and integration among individuals holding
interdependent knowledge is essential for the accurate replication of knowledge. An appropriate design for
inter-unit organizational structures may facilitate inter-unit replication of complex knowledge. Inter-unit
organizational structures are channels of communication and conduits for sharing information and
knowledge between individuals. Prior studies have found that firms may develop inter-unit ties, or utilize
existing ties formed through either formal arrangements or informal relationships, for knowledge replication.
For instance, as Ghoshal et al. (1994: 96) state, “interpersonal relationships developed through lateral
networking mechanisms such as joint work in teams, taskforces, and meetings have significant effects on the
frequency of both subsidiary-headquarters and inter-subsidiary communication.” Also, Szulanski (1996: 28)
finds that “transfer-specific social ties between the source and the recipient are established and the
transferred practice is often adapted to suit the anticipated needs of the recipient.”
Furthermore, field studies on knowledge transfer observe that formal arrangements such as joint
ventures, co-locations, structural integration, or job rotation training result in the formation of inter-unit ties
and informal socialization, facilitating knowledge transfer (Kenney and Florida, 2000: 92; Prochno, 2003).
Banc One, a U.S. super-regional bank, was known for its effective replication of operational routines within
newly acquired affiliates. A key event that occurred at the beginning of its replication process was an
overview meeting in which senior representatives from different departments of the organization made
presentations and developed relationships with the senior managers who dealt with the same range of
functions within the acquired affiliates (Szulanski, 2000). The outcome of such formal arrangements or
informal relationships transcends the immediate transfer of knowledge through direct contact. A broad set of
relationships enables future inter-unit knowledge sharing and replication (Hansen and Nohria, 2004).
The design of inter-unit structures influences the replication of complex knowledge because it
dictates the pattern of information-processing interactions between units (Puranam et al., 2012; Szulanski,
2000; Tushman and Nadler, 1978). For instance, like a boundary-spanner structure (Leifer and Delbecq,
1978), an inter-unit structure may be composed of a highly-centralized pattern of information-processing
interactions between units
1
(Tushman and Scanlan, 1981a, 1981b). Aside from the boundary-spanner
structure, a wide spectrum of organizational designs may exist.
Three dimensions of inter-unit structures
The information-processing view (Galbraith, 1973; Lawrence and Lorsch, 1967; Tushman and Nadler, 1978)
has been a well-received theoretical perspective underlying knowledge-transfer and management literature
(Carlile, 2002, 2004). However, despite the fact that the information-processing view of organizations is
anchored in bounded rationality and the Carnegie School (March and Simon, 1958), we find that influences
of bounded rationality on knowledge replication outcomes have been largely ignored.
By combining these theories with insights from phenomenological and field studies in knowledge
transfer, we identify the salient dimensions of inter-unit structures that determine fundamental organizational
design considerations such as the amount of information-processing interactions, the pattern of information-
processing interactions, and the distribution of information load (Galbraith, 1973; Puranam, et. al., 2012).
1
The boundary-spanning literature primarily pays attention to vertical knowledge transfer along value chains. The
context of this study is horizontal knowledge transfer between units.
Those dimensions are inter-unit connectivity, the extent of mirroring between knowledge and structure, and
coordination mechanism. Each of these dimensions will be discussed in the following sections.
Inter-unit connectivity between source and recipient units
A central tenet of organization theory is that organizational design should provide an adequate information-
processing capacity for organizations to meet the demands of tasks and environmental contingencies
(Galbraith, 1973; March and Simon, 1958). Prior studies rooted in this information-processing perspective of
organizations have largely emphasized lateral linkages between subunits for effective knowledge sharing and
better information processing (Egelhoff, 1993; Galbraith, 1994; Lawrence and Lorsch, 1967; Gupta and
Govindarajan, 1991; Tushman and Nadler, 1978). Carlile (2002: 444) summarizes an overarching conclusion
within the prior work on the impact of inter-unit connectivity: “Many extended this boundary spanning and
information processing framework and focused on what internal communication patterns, planning, and
prioritizing processes were determinants of product success (Keller, 1986; Joyce, 1986; Ancona and
Caldwell, 1992a). Others focused on the importance of external communication patterns and boundary
spanning activities in successful product development (Katz and Tushman, 1981; Ancona and Caldwell,
1992b). The overall insight from this type of research is that more information is better, more communication
is better, and more team strategies are better.”
Other related phenomenological studies make similar observations that are largely consistent with
the information-processing view. Extensive communication between units, or direct ties between individuals,
facilitates the inter-unit transfer of complex or tacit knowledge (Gupta and Govindarajan, 1991; Hansen,
1999, 2002). The more such ties exist between a source and recipient unit, the larger the amount—and the
higher the diversity—of information that can flow between the units. Increased access to information and
knowledge can help individuals more effectively solve complex problems by providing new perspectives on
those problems and enabling individuals to conceive of alternative solutions (Ancona and Caldwell, 1992;
Turner and Makhija, 2012). In addition, multiple inter-unit ties can allow individuals to draw upon multiple
sources of information and knowledge, and thus identify and locate an initial search position and relevant
knowledge (Hansen, Mors, and Lovas, 2005).
In those studies, however, it has been largely assumed that information transfer concomitantly occurs
as information availability increases (Aral and Alstyne, 2011). The availability and access to pertinent
information and knowledge are necessary but not sufficient conditions for accurate replication to occur. In
particular, even though one of the key deterrents of the accurate replication of complex knowledge is the
bounded rationality of individuals (Simon, 1991; Zhao and Anand, 2013), prior work has considered either
information-processing requirements or bounded rationality concerns without simultaneously considering the
influences of these two elements. Given the limited cognitive capacity of individuals, replication accuracy
decreases as the information load to individuals increases (Aldrich and Herker, 1977; Marrone, Tesluk, and
Carson, 2007).
In all, with a larger number of inter-unit ties, individuals are likely to process a higher volume or
broader scope of information, improving the accuracy of knowledge replication. However, from a bounded
rationality standpoint, with a large number of inter-unit ties, individuals may be information-overloaded in
terms of the volume or scope of information. This overload may result in a diminished level of knowledge-
replication accuracy. Given that high knowledge complexity demands more information-processing activities
to facilitate knowledge replication, which can cause information overload, the optimal extent of inter-unit
connectivity can be contingent on the level of knowledge complexity.
Mirroring between knowledge and inter-unit organizational structures
Given the configurational aspect of complex knowledge, the focus of organizational design shifts from the
amount of information processing to the pattern of information processing between units. The mirroring
hypothesis, which is rooted in the modularity literature and the information-processing view of organizations,
casts light on this issue. The mirroring hypothesis involves how product architectures affect organizational
designs (Baldwin, 2008; Colfer and Baldwin, 2016; Henderson and Clark, 1990; Schilling, 2000).
Knowledge about how the components of a product interact and are integrated into a coherent whole is often
embedded in an organization’s structure and its information-processing routines (Henderson and Clark,
1990). The mirroring hypothesis therefore posits that an organization’s structure mirrors the architecture and
interface of a product. In other words, the information-processing requirement mandated by a product’s
architecture is congruent with the information-processing pattern of the organization’s design.
The mirroring hypothesis has been examined primarily in the context of product modularity and
organizational design at the inter-firm level (Cabigiosu and Camuffo, 2012; Sosa, Eppinger, and Rowles,
2004). We translate the mirroring hypothesis from the product architecture domain to the knowledge domain
at the firm level. For exogenously given source knowledge, we posit that the design of structures, and thus
the pattern of information-processing interactions (Tushman and Nadler, 1978), should correspond to the
configuration of knowledge interdependence to allow for accurate knowledge replication.
2
While better inter-
unit connectivity increases the probability of accurate knowledge replication by providing more information-
processing interactions between units, what matters more for the accuracy of replication is the degree to
which inter-unit ties offer direct access to relevant knowledge (Hansen et al., 2005).
A misalignment between an organizational structure and knowledge occurs due to deficient or
redundant ties. Deficient ties exist if the organization lacks ties that mirror a given knowledge configuration.
Redundant ties are any ties that exist in addition to mirroring ties. Both kinds of misalignment are caused by
either the failure to perceive existing knowledge interdependencies or the incorrect identification of existing
knowledge interdependencies. A misalignment between the organizational structure and knowledge
configuration can decrease replication accuracy, as it leads to missing information or the delivery of
unnecessary information (Sosa, Eppinger, and Rowles, 2004). Such information-processing views on the
misalignment of organizational structure and knowledge are useful in predicting the impact of deficient ties
on replication accuracy. However, they cannot correctly predict the impact of redundant ties on replication
accuracy, because the mirroring hypothesis literature has largely ignored the influences of bounded
rationality on replication. Redundant ties may invoke information overload or the delivery of conflicting
information, degrading replication accuracy (Aldrich and Herker, 1977; Zhao and Anand, 2013).
The collective bridge structure has been proposed as a mirroring structure (Zhao and Anand, 2013).
Consisting of a set of within- and cross-expertise ties, the collective bridge structure has a topology that
emulates a knowledge configuration. Within-expertise ties are established between the counterparts of the
source and recipient units who hold expertise in the same areas, so that independent knowledge can be
transferred from the members of a source unit to the members of a recipient unit. Cross-expertise ties are
instituted between counterparts holding interdependent expertise areas so that trans-specialist knowledge
(Postrel, 2002) in the source unit can be replicated by the members of a recipient unit. Within the collective
bridge structure, all relevant members of a recipient unit communicate with and learn from all relevant
2
Given the concept of design interface, establishing a parallel between the product architecture and knowledge
configuration is deemed reasonable (Sosa, Eppinger, and Rowles, 2004).
members of a source unit, with a minimum possible number of ties. An examination of the performance
implications of the mirroring hypothesis, from both information-processing and bounded rationality
standpoints and at different levels of knowledge complexity, may shed light on the optimal design of inter-
unit structures for the replication of complex knowledge.
Coordination mechanisms
Coordination of interdependence is one of the key information-processing activities in organizations (March
and Simon, 1958; Tushman and Nadler, 1978). The replication of complex knowledge requires a recipient
unit to replicate the coordination routines between individuals with different areas of expertise (Postrel,
2002). The coordination of interdependencies can be achieved either by formal mechanisms such as
structural integration and common authority, or by informal coordination through the achievement of
common ground in the form of shared knowledge or ongoing communications between individuals (Pentland
and Rueter, 1994). Instances of common ground in the context of complex knowledge replication are cross-
expertise knowledge, which is a trans-specialist understanding between individuals with different knowledge
domains (Postrel, 2002), and within-expertise knowledge, which is shared knowledge between individuals
with the same expertise area. Mirroring ties help create and develop common ground between different
knowledge domains.
In the presence of epistemic interdependence, successful coordination of interdependence requires
architectural knowledge, predictive knowledge, or their combinations. Architectural knowledge involves how
individuals or corresponding knowledge components interact and are integrated into a coherent whole.
Predictive knowledge is “knowledge that enables one agent to act as though he or she can accurately predict
another agent’s actions” (Puranam et al., 2012: 420). Two archetypical coordination mechanisms—dyad-
level and centralized coordination—rely on two distinct types of knowledge for coordination: predictive and
architectural knowledge, respectively. Whereas dyad-level coordination solely relies on predictive
knowledge, centralized coordination primarily uses architectural knowledge to achieve unity of actions. To
the extent that a central coordinator possesses and uses accurate architectural knowledge, the centralized
coordination mechanism minimizes coordination errors, which may occur with the dyad-level coordination
mechanism (Henderson and Clark, 1990).
Without compromising organizational performance, architectural and predictive knowledge may
substitute for each other in their roles of coordinating interdependencies (Puranam et al., 2012). Thus, the use
of accurate architectural knowledge decreases the need for predictive knowledge or direct ongoing
communications at the dyad level between a source unit and a recipient unit.
However, we argue that such substitutability between architectural knowledge and predictive
knowledge may not be complete under a certain condition. Such a condition can be acknowledged only when
the information-processing view and bounded rationality are considered simultaneously. If an agent
controlling architectural knowledge performs interdependence coordination, the agent can be subject to
information overload, precisely because interdependence coordination necessitates additional information-
processing activities that are imposed on the coordinator. Thus, the centralized coordination mechanism is
prone to escalating the information overload that is imposed on the central coordinator. Meanwhile,
coordination mechanisms that rely only on predictive knowledge (e.g., dyad-level coordination) do not
escalate information overload, because they delegate information-processing activities across individuals.
Thus, we argue that the substitutability between architectural and predictive knowledge can be complete as
long as an agent controlling architectural knowledge delegates coordination tasks to other agents. When an
agent controlling architectural knowledge is information-overloaded, the performance of replication based on
architectural knowledge can be worse than that based on predictive knowledge.
In sum, a simultaneous application of the information-processing view and bounded rationality
underscores a tradeoff between facilitating interdependence coordination and managing information overload
when designing inter-unit organizational structures. Given the boundary condition for a complete
substitutability between architectural and predictive knowledge, it remains difficult to conjecture what would
be an optimal choice of coordination mechanisms at different levels of knowledge complexity.
MODEL AND SIMULATION
Formalizing the theoretical underpinning explicated above, we have developed an analytical structure to
model knowledge replication between two units. Modeling and simulations help us to develop more precise
and rigorous propositions, which are difficult to derive purely from theoretical arguments and verbal
reasoning (Csaszar and Siggelkow, 2010; Siggelkow and Rivkin, 2006). Our model strategically focuses on
the relationship between the design of inter-unit structures and knowledge replication performance, while
controlling for all other antecedents of replication performance. A distinctive strength of simulation
methodology is its ability to completely rule out potential alternative explanations. As illustrated in Figure
1a, the model involves two units, and each unit consists of N individuals. Knowledge to be replicated
involves N areas of expertise and Q interdependencies. Two knowledge attributes are considered in the
model: knowledge complexity and knowledge configuration. The design of inter-unit structures is
dimensionalized by inter-unit connectivity, the extent of mirroring, and the coordination mechanism. We
explain each element of the model in the following subsections.
Complex knowledge
Knowledge is represented by an N dimensional binary vector
. dk represents the knowledge component associated with kth expertise area of knowledge, and N
represents the total number of expertise areas associated with the knowledge. Such operationalization of
knowledge is similar to modeling approaches in previous simulation studies of knowledge replication or
organizational learning (Levinthal, 1997; Rivkin, 2000). A structure of knowledge interdependence is
denoted by an N x N matrix, E. when there is no interdependence between expertise areas i and j;
when i and j are interdependent. Q represents the total number of interdependencies between the
expertise areas. A simplifying assumption is made by considering each individual to be responsible for one
area of expertise. Accordingly, if there are N areas of expertise, an N number of individuals exists at each
unit, which is consistent with Zhao and Anand’s (2013) approach. Knowledge with a Q number of
interdependencies involves Q instances of cross-expertise knowledge.
Inter-unit organizational structures
The topology of an inter-unit structure is denoted by adjacency matrix A. Matrix element Ai,j denotes the
presence or absence of an inter-unit tie between the individual, i, of a source unit who holds expertise, di, and
the individual, j, of a recipient unit who holds expertise, dj, with 1 indicating its presence and 0 indicating its
absence. Matrix A can be symmetric or asymmetric, depending on the type of inter-unit structures. To focus
on the effects of inter-unit structures on knowledge replication performances, we control for the intra-unit
structure. In all simulation analyses, intra-unit structures for the same values of N and Q are identical. An
intra-unit structure always mirrors knowledge configuration, E. For example, in Figure 1b and Figure 1c,
given that E14 = 1, an intra-unit tie is established between individuals 1 and 4 at a source and recipient unit,
respectively.
Inter-unit connectivity
The extent of inter-unit connectivity is defined as the total number of inter-unit ties that are established
between a source and recipient unit:
.
The extent of mirroring
The extent of mirroring between knowledge and an inter-unit structure is operationalized by deficient and
redundant ties. We first define and formalize a collective bridge structure, which is an inter-unit structure that
completely mirrors a knowledge configuration with a minimum possible number of ties, as follows:
Figure 1b illustrates an instance of collective bridge structures for N = 4 and Q = 1. Suppose that
knowledge components d1 and d4 are interdependent. In other words, E14 = 1, and all other matrix elements Eij
= 0. Solid lines denote inter-unit ties, and dotted lines denote intra-unit ties. Each knowledge expertise area is
illustrated by a distinct pattern of shades. For example, since an individual at a source unit, s1, and a
counterpart individual, r1, have the same area of expertise, they are indicated by the same shading pattern.
For each pair of knowledge interdependencies, a set of mirroring ties consists of two within-expertise and
two cross-expertise ties. For example, for knowledge interdependency E14, there are two within-expertise
ties—a tie between s1 and r1, and one between s4 and r4—and two cross-expertise ties—a tie between s1 and
r4, and one between s4 and r1. Consequently, matrix A is always symmetric for all collective bridge structures.
For knowledge elements without interdependence, only within-expertise ties constitute mirroring ties.
--- Insert Figure 1b about here ---
In turn, to modulate the extent of mirroring, we progressively add ties to, or remove mirroring ties
from, the collective bridge structure, which serves as a baseline structure. A higher number of redundant or
deficient ties is associated with a lower degree of mirroring. Redundant ties are defined as all existing inter-
unit ties aside from mirroring ties. In the case of Figure 1b, E14 = 0, and thus only inter-unit ties A14 and A41
are required to mirror the knowledge configuration. An inter-unit tie A12 or A21 would be a redundant tie,
precisely because there is no knowledge interdependence between d1 and d2, and thus no cross-expertise
knowledge to replicate. Lastly, deficient ties exist if there is any lack of mirroring ties for a given knowledge
configuration, E. Figure 1c illustrates an example of a structure with deficient ties: Inter-unit ties A14 and A41
are missing. Figure 1d depicts a structure with redundant ties: Inter-unit tie A32 is redundant.
--- Insert Figure 1c and Figure 1d about here ---
Coordination mechanisms
Two types of coordination mechanisms are considered in our model: dyad-level and centralized coordination.
In the case of dyad-level coordination, coordination is achieved through predictive knowledge embedded
between individuals in a dyad. The collective bridge structure, illustrated in Figure 1b, is an example of an
inter-unit structure employing only the dyad-level coordination mechanism. Given knowledge
interdependence E14, individuals 1 and 4 at the source and recipient units coordinate with each other through
within- and cross-expertise inter-unit ties. Coordination is achieved without any central coordinating agent
overseeing individuals 1 and 4 at both units.
Meanwhile, with a centralized coordination mechanism, all knowledge interdependencies are
coordinated by a central agent holding architectural knowledge. Unit members report to the central
coordinator about their local conditions, and in turn the central coordinator makes decisions while also
allocating and scheduling tasks. For instance, in the boundary-spanner structure illustrated in Figure 1e,
boundary-spanning individuals, s1 at a source unit or r1 at a recipient unit, oversees all interdependencies
within each unit. Consequently, in our model, rather than interdependent knowledge components being
transferred pair-wise as is the case for the dyad-level coordination, the whole knowledge components of s1,
s2, s3, and s4 are transferred as a bundle by the boundary spanners s1 and r1.
--- Insert Figure 1e about here ---
A disadvantage of dyad-level coordination compared to centralized coordination is that certain
knowledge components can be updated redundantly or erroneously, potentially resulting in coordination
failure. For example, in the case of E12 = 1 and E23 = 1, knowledge component d2 can be updated via a cross-
expertise tie, s1-r2 (A12), or via another cross-expertise tie, s3-r2 (A32). If values of d2 that are transmitted
through those two inter-unit ties are different, coordination failure may occur. For instance, in the absence of
centralized coordination, what a software engineer (individual 1) recommends to a hardware engineer
(individual 2) can differ from what the product manager (individual 3) recommends to the hardware engineer
(individual 2). The sequencing of interdependency pairs in our simulation is such that individuals starting
from i=1 to N take turns and coordinate their interdependencies. Thus, in this example, d2 takes on the value
as suggested by an interdependency between individuals 2 and 3.
Distribution of information load
The model in this study incorporates two types of information load: information volume and scope loads.
This model assumes that the volume of each knowledge component is identical and equal to one unit of
volume, regardless of expertise areas. The total volume of knowledge that a unit holds is N+Q, which
consists of N units of volume stemming from N expertise areas and Q units of volume stemming from Q
number of cross-expertise knowledge components: .
The design of inter-unit organizational structures determines the distribution of information volume
and scope across individuals. The information volume imposed on an individual i, vi, is determined by the
number of ties that i has and the information volume that each tie carries:
, where ωk denotes
the volume of information that the individual i processes through the kth tie and n signifies the total number
of ties of an individual, i. Intra- and inter-unit ties contribute equally to the information volume. For
example, in a structure illustrated in Figure 1b, an individual, r1, processes information through ties s1-r1, s4-
r1, and r4-r1, leading to .
Information scope imposed on an individual, i—represented as—is defined as the total number of
distinct expertise areas that an individual, i, processes. For example, individual r1 in Figure 1b in a recipient
unit processes two distinct expertise areas, leading to : expertise areas 1 and 4. Thus, the total
information load imposed on individual r1 .
Information load and replication accuracy
Given the bounded rationality of individuals, the accuracy of information-processing activities of an
individual decreases as the information load imposed on the individual increases (Marrone, Tesluk, and
Carson, 2007).The likelihood of accurate information processing by an individual i, linearly decreases
as the information volume imposed on the individual i increases:
Cv, where denotes the
information volume imposed on i and Cv is a negative proportional constant. The magnitude of Cv determines
the extent to which replication accuracy deteriorates with respect to information-volume load. Likewise,
linearly decreases with respect to the information scope imposed on i:
Cs, where is the
information scope imposed on i, and Cs is a proportional constant denoting the rate at which replication
accuracy deteriorates with respect to information scope. Aggregating the effects of information-volume and
information-scope loads together, the exact equation used for simulation is Cs )
3
. In
turn, the probability that individuals i and j accurately process information is
Coordination failure
We assume that in the presence of mirroring ties, predictive knowledge is complete and accurate, leading to
accurate dyad-level coordination. Thus, a lack of mirroring ties causes coordination failure, leading to the
degradation of replication accuracy. As per the definition of mirroring ties, two within- and two cross-
expertise ties are necessary for accurate coordination of a given instance of knowledge interdependence.
Coordination accuracy linearly decreases with respect to the number of missing mirroring ties. More
specifically, if an inter-unit structure has x number of within-expertise ties and y number of cross-expertise
ties for a given interdependence, replication accuracy is degraded by a multiplicative coordination failure
factor of
, where CF is a coordination failure factor,
CPw is a coordination failure penalty for missing within-expertise ties, and CPc is a coordination failure
penalty for missing cross-expertise ties. For example, when there exists a set of completely mirroring ties
(i.e., x = 2 and y = 2), the coordination failure factor is CF = 1, leading to no degradation of replication
accuracy. On the other hand, when no inter-unit ties are instituted (i.e., x = 0 and y = 0), CF = 0, resulting in
a complete failure of interdependence coordination. The coordination failure factor and the information
overload factor independently affect replication accuracy:
A case of a structure with deficient ties
Figure 1c illustrates an example of an inter-unit structure with deficient ties: Cross-expertise ties, a tie
between and
and a tie between and , are missing from a collective bridge structure standpoint. With
this inter-unit structure, the probability that knowledge N = 4, Q = 1, and E14 = 1 is replicated accurately from
a source unit to a recipient unit is
, where
denotes the
probability of accurate replication of knowledge elements d1 and d4 with E14 = 1 from source individuals
3
All values of below 0 and above 1 are truncated and substituted with a value of 0 or 1, respectively.
and to recipient individuals and
. Because of a lack of two cross-expertise ties,
,
leading to
. Accordingly, the likelihood of accurate replication
is
. In turn, each probability, is calculated according to
the rules associated with information volume and scope overloads.
A case of a boundary-spanner structure
Figure 1e is a boundary-spanner structure, which is an archetypical centralized inter-unit structure. By virtue
of all necessary inter-unit ties, the dyad-level coordination failure factor . Thus, the probability that
knowledge tuple is replicated accurately is
. An advantage of this centralized
coordination is that the replication accuracy will be high across all of the knowledge elements that are
coordinated if the central agents, s1 and r1, are not information-overloaded. On the other hand, if central
agents are information-overloaded, assumes a low value. Thus, even if all other individuals except
for the central agent are not information-overloaded, the overall probability of replication accuracy becomes
low.
Measure of replication accuracy
Replication accuracy is operationalized as the extent to which the knowledge of a recipient unit matches that
of a source unit: , where ρ is replication accuracy that ranges between 0 and 1, N is the total number
of expertise areas, and M is the number of matching knowledge elements between two units that accurately
replicate knowledge interdependence, E. In other words, replication accuracy is determined not simply by
counting the number of digits that have identical values between source and recipient units, but by
considering whether interdependencies at the source unit have been successfully replicated at the recipient
unit. For example, consider the following setup: N = 3, Q = 1, source unit knowledge Ds = (0, 1, 1), E12 = 1,
and all other Eij = 0. Since knowledge elements d1 and d2 are interdependent, their values must be replicated
as a pair. Thus, if the knowledge tuple of a recipient unit after a replication attempt is Dr = (0,0,1), then
replication accuracy = 1/3 instead of 2/3, because, given E12, only d3 matches between the source and
recipient units. In this example, if Q = 0 and thus E12 = 0, replication accuracy is = 2/3. If no knowledge
interdependence exists (i.e., Q = 0), M is simply the total number of matching knowledge elements between
the knowledge tuples of a source and recipient unit. This measure of replication accuracy captures the
consistency between the knowledge at a source unit and the knowledge deployed to a recipient unit in a
simple way.
4
Simulation procedure
Based on the described model, we have conducted an agent-based simulation (Epstein, 1999). This
simulation analysis was conducted in the following sequence. First, parameters and initial conditions
pertaining to knowledge attributes and inter-unit structures were set up. This procedure included the
initialization of knowledge tuples of a source and recipient organization and knowledge configuration.
Second, during each time period, t, source and recipient units were engaged in a knowledge-transfer process
shaped by a set of behavioral rules for knowledge transfer. The knowledge-transfer process was then
repeated. In the third step, for a given fixed set of simulation parameters, the entire simulation process was
repeated and the average and probability distribution of knowledge-transfer performance measures were
calculated.
This simulation process is consistent with the standard Monte Carlo method (Law and Kelton, 1991)
as well as with other applications of simulation methodology in strategy research (e.g., Siggelkow and
Rivkin, 2005). To ensure the validity of average values, we calculated confidence intervals. For a given
average value of a performance measure, μ, a 95% corresponding confidence interval was defined as the
range between μ – 2σ/n and μ + 2σ/n, where σ is a standard deviation of the performance measure and n is the
number of simulation trials (Law and Kelton, 1991).
5
By simulating knowledge transfer on a wide range of inter-unit structures, this study identifies how
structural attributes of an inter-unit structure moderate the impact of knowledge complexity on knowledge
replication performance. Based on the results of these simulation experiments, we derived a set of
propositions. Lastly, we conducted sensitivity analyses and assessed how knowledge-transfer performances
are affected by information-overload parameters, coordination-failure penalty parameters, and, boundary-
spanner capability. Table 1 summarizes the parameters used for simulation analyses. Unless otherwise
indicated, default parameter values used for all simulations are Cv = 0.04, Cs = 0.04, CPW = 0.25, and CPc =
4
A very similar operationalization was used in some prior studies on organizational learning that used simulation
methodologies. For instance, the essence of such operationalization of interdependence is consistent with Fang, Lee, and
Schilling’s (2010) modeling approach for their payoff function.
5
Across all simulations in this study, σ/n values ranged from 0.00012 to 0.00205, while μ ranged from 0 to 1.
0.25.
--- Insert Table 1 about here ---
There are both deterministic and stochastic elements in our simulation. On the one hand, given a
structure (and thus given values of , , and ) as well as a given set of parameters, is computed.
On the other hand, the selection of deficient and redundant ties (e.g., which ties will be deficient or redundant
ties) is stochastic. As a result, (information volume) and (information scope) are determined only after
those stochastic elements are taken into account. In this aspect, is simulated.
RESULTS AND PROPOSITIONS
We conducted simulation experiments that examined how the accuracy of complex knowledge replication is
impacted by inter-unit connectivity, the mirroring between inter-unit structure and knowledge, and
coordination mechanisms. Table 2 summarizes the mechanisms underlying each simulation result and its
associated proposition. We provide detailed explications below.
--- Insert Table 2 about here ---
Inter-unit connectivity
We first assessed how inter-unit connectivity influences replication accuracy. This analysis was done by
gradually adding random inter-unit ties between a source and recipient unit. This simulation setup is
analogous to a situation in which a firm blindly increases connectivity between units to achieve a better
replication outcome, in the absence of an accurate architectural understanding of knowledge configuration or
information on the locations of relevant knowledge.
Figure 2(a) reports results for various values of Q. As a reference, the performance of the collective
bridge structure (CB) is indicated. In this experiments N = 10, and the results qualitatively remain intact for
other values of N. Across all Q, the association between the extent of inter-unit connectivity and replication
accuracy is characterized consistently as a curvilinear relationship. Additional inter-unit ties are beneficial to
knowledge-replication performances up to a certain threshold, and additional ties beyond the optimal point
degrade replication performances. This finding largely contrasts with the prediction of previous literature,
demonstrating a positive monotonic association between connectivity and replication accuracy. For example,
the knowledge replication literature suggests that denser connections support effective knowledge replication
(Reagans and McEvily, 2003). In a similar vein, the information-processing view emphasizes lateral linkages
between units for effective sharing and processing of knowledge (Galbraith, 1994; Lawrence and Lorsh,
1967; Tushman and Nadler, 1978). The simulation result, on the other hand, illustrates that an optimal design
of structures depends not only on information-processing requirements but also on bounded-rationality
concerns.
--- Insert Figure 2 about here ---
Figure 2(b) presents a decomposition of the replication performance. The dotted line represents
replication accuracy with only the effect of coordination failure included, the shaded line with only the effect
of information overload included, and the solid line containing both of the effects. As for information
overload, while additional inter-unit connectivity is translated into greater information flow between units or
a higher level of information diversity, irrelevant information generates a higher volume of information or
scope loads over individuals, diminishing the accuracy of replication. Thus, the information overload effect
leads to a concave curve for replication accuracy as a function of inter-unit connectivity. In regard to the
effect of coordination failure, each inter-unit tie carries partial knowledge from a complex-knowledge
standpoint. Thus, an additional inter-unit tie contributes to better replication accuracy to the extent that it
brings a relevant piece of complex knowledge to a recipient unit. Thereby, with the coordination failure
effect, replication accuracy monotonically increases with respect to inter-unit connectivity.
Proposition 1. The replication accuracy of inter-unit structures as a function of the number of inter-
unit ties follows an inverted-U curve. There exists an optimal point of inter-unit connectivity for a
given level of knowledge complexity.
The mirroring between structure and knowledge
The previous experiment considered the degree of connectivity between units, without considering the extent
to which inter-unit ties mirror knowledge configuration. In the following experiment, we delve into the
configurational aspect of inter-unit structures and examine the performance implications of the misalignment
between an inter-unit structure and knowledge configuration. In so doing, we compare predictions of the
information-processing view and bounded rationality.
The first experiment examined the impact of deficient ties on replication accuracy. We progressively
and randomly eliminated within- or cross-expertise ties from the collective bridge structure, which is a
completely mirroring structure. Figure 3 presents replication accuracy as a function of the number of
deficient ties across different values of knowledge complexity. A general observation across all cases of N
and Q is that as more inter-unit ties become deficient, the replication accuracy decreases consistently across
all levels of knowledge complexity. The deficiency of within- or cross-expertise ties results in a partial
delivery of complex knowledge. Coordination failures stemming from the lack of trans-specialist
understanding further deteriorate replication accuracy.
--- Insert Figure 3 about here ---
The second experiment assessed the impact of redundant inter-unit ties on the performance of inter-
unit structures
6
. We implemented this experiment by gradually adding random inter-unit ties to the collective
bridge structure. This experiment is distinct from the random inter-unit connectivity experiment in that a
complete set of mirroring ties was already established between units regardless of the number of redundant
ties. Figure 4 illustrates the simulation results, graphing the replication accuracy with respect to the number
of redundant inter-unit ties across varying levels of knowledge complexity.
--- Insert Figure 4 about here ---
Two observations are apparent from Figure 4. First, a general observation across all values of N and
Q is that replication accuracy decreases as more redundant ties are established between a source and recipient
unit. Additional within- or cross-expertise ties linked to an individual are only made to increase unnecessary
information load imposed on the focal individual—particularly information scope load—if the additional ties
do not carry knowledge that is relevant to the expertise area of the focal individual. In terms of coordination
failure, because the structures with redundant ties always retain a complete set of mirroring ties, there is no
penalty to replication accuracy due to coordination failure.
Second, the results demonstrate that replication accuracy improves when N increases, for both a
given Q and a given number of redundant ties. Although greater N knowledge involves a higher volume of
information, additional information volume or scope loads caused by redundant ties are distributed more
evenly over a larger N number of individuals. The underlying mechanism is that while the number of
necessary mirroring ties linearly increases with respect to N and Q, the number of possible combinations of
inter-unit ties increases quadratically with N.
6
Deficient or redundant ties approach for manipulating the degree of mirroring does not control for the total number of
inter-unit ties. We also conducted an experiment in which inter-unit ties were rewired rather than deleted or added and
confirmed that findings and the proposition remain intact.
These findings cast light on Sosa et al.’s (2004) assertion about potential impacts of redundant ties
on knowledge-transfer outcomes. Sosa et al. (2004) classified instances of misalignment between product
architecture and organizational structure as unmatched design interfaces and unmatched team interactions.
The former corresponds to deficient ties in our study, and the latter to redundant ties. From a theory
standpoint, Sosa et al.’s (2004) reasoning is informed by the information-processing view, while completely
ignoring the relationship between misalignment and information overload. The result is that while their
prediction about the influence of deficient ties is accurate, their prediction about the impact of redundant ties
is misleading. This discrepancy occurred because the impact of deficient ties is fully captured by the
information-processing view, but redundant ties can invoke information overload, the effect of which is
captured by bounded rationality.
Proposition 2: The replication accuracy in inter-unit replication of complex knowledge improves as
the extent of mirroring between knowledge configuration and inter-unit structures increases.
Figures 5 and 6 illustrate how the distribution of information loads varies in response to variations in
knowledge interdependence, Q, for structures with deficient or redundant ties, respectively. The X axis
represents information volume or scope imposed on members of a source unit and a recipient unit, and the Y
axis represents probability density. In these experiments, N = 10. Table 3 shows the mean values of the
distributions. The results confirm that knowledge with a higher Q imposes a higher volume or scope of
information.
7
It is worthwhile to note that the structure with deficient ties has a higher level of information
volume, and the structure with redundant ties has a higher level of information scope. With a lower total
number of inter-unit ties due to deficient ties, each inter-unit tie tends to carry a higher volume of
information, resulting in a higher information volume being concentrated in a few individuals. Redundant
ties may bring irrelevant pieces of information to some recipient individuals, resulting in a higher
information scope being imposed on a few individuals.
--- Insert Figure 5 and Figure 6 about here ---
--- Insert Table 3 about here ---
The aforementioned simulation experiments incorporated simultaneous effects of information
processing and bounded rationality. To examine the effect of omitting the impact of bounded rationality on
7
ANOVA analyses confirm that the mean values across different Qs are statistically significant at α = 0.05.
knowledge replication, we performed additional experiments with an extended model. The baseline model,
which was used for all other simulations, incorporates both effects, as explicated in the model section. In the
extended model, while the effect of information processing is intact, the effect of information overload or
coordination failure is excluded. In terms of the simulation parameters, we set Cv and Cs (information
overload intensity) to zero to exclude the information overload effect or we set CPw and CPc to zero to
exclude coordination failure effect. Graphing predictions of the two models, Figures 7 and 8 illustrate cases
of redundant and deficient ties, respectively. Dotted and shaded lines denote results from the extended
model, with the former representing results with only coordination failure effects and the latter representing
results with only information overload effects. In these experiments N = 10, and the results qualitatively
remain intact for other values of N.
--- Insert Figure 7 and Figure 8 about here ---
On one hand, a prediction of the information-processing view is that as the number of mirroring ties
increases, replication accuracy improves. Information-processing requirements demanded by a particular
knowledge configuration are fulfilled by mirroring ties. Thus, according to the information-processing view,
while the presence of deficient ties deteriorates replication accuracy, redundant ties do not impact
knowledge-replication outcomes. On the other hand, a prediction of bounded rationality is that redundant ties
can degenerate replication accuracy because they may incur information volume or scope overload. Figures 7
and 8 illustrate that the predictions of the two models diverge further as knowledge complexity increases
8
.
An underlying mechanism is that whereas the marginal benefit of mirroring ties decreases with higher
knowledge complexity, information overload concomitantly increases with higher knowledge complexity.
The above findings and arguments lead us to advance the following proposition:
Proposition 3: Predictions of replication accuracy in the inter-unit replication of complex
knowledge solely based on the information-processing view become more biased as knowledge
complexity increases.
Coordination mechanisms
8
Figures 7 and 8 also shed light on how the replication performance is decomposed into two parts: information overload
and coordination problems. Figure 7 presents that information overload concern grows with more redundant ties while
the coordination problem portion is zero. Figure 8 shows that the information overload portion of the problem decreases
as more deficient ties exist and that the coordination problem portion quickly grows with more deficient ties.
We report the results of experiments on the optimal choice of coordination mechanisms at different levels of
knowledge complexity. Two archetypical coordination mechanisms—dyad-level and centralized
coordination—and two types of knowledge required for coordination—architectural and predictive
knowledge—are considered.
Decentralized and centralized coordination: Contingency framework
We conducted an analysis of the impact of knowledge complexity on the performance of a centralized inter-
unit structure in comparison to that of decentralized structures. To simulate a range of decentralized
structures, we considered structures consisting of mirroring ties in addition to structures with deficient or
redundant ties that had been considered in previous experiments. We progressively added mirroring ties
between two units to modulate the extent to which the structure mirrors a knowledge configuration.
Figure 9 illustrates the results, plotting the replication accuracy of decentralized inter-unit structures
with a varying degree of mirroring alongside that of a boundary-spanner structure. N = 10 was used for all
experiments. A central observation from the results is that as Q increases, the gap between the replication
accuracy of the centralized structure and that of the decentralized structures grows. When knowledge
complexity is low, the centralized structure is superior to decentralized structures. However, the replication
accuracy of the centralized structure leads to complete failure when Q is high. In contrast, decentralized
structures maintain a reasonable replication accuracy even when Q is very high.
--- Insert Figure 9 about here ---
Figure 10, in turn, confirms the difference in the distribution of information volume and scope loads
between the centralized and decentralized structures. Within a boundary-spanner structure, while the
majority of individuals are loaded with low information volume and scope, a small number of individuals
(boundary-spanning individuals) are loaded with a high level of information volume and scope. Even when
all individuals other than the boundary spanner are not information-overloaded, an information-overloaded
boundary spanner can make the overall replication accuracy very low (Elfring and Hulsink, 2007; Mariotti
and Delbridge, 2012). A stark contrast between the centralized structure and decentralized structures is the
presence of highly centralized nodes. A high level of centralization is translated into high information
volume and scope loads imposed on central agents. On the other hand, in the mirroring structure, both
information volume and scope are distributed much more evenly than in the boundary-spanner structure.
Using inter-unit direct ties, decentralized structures—including mirroring structures—evenly distribute
information loads over individuals.
--- Insert Figure 10 about here ---
Proposition 4. Under conditions of bounded rationality and complex knowledge, the replication
accuracy of decentralized structures is less negatively impacted by knowledge complexity than that
of centralized structures.
Architectural and predictive knowledge
The mirroring structure is a completely decentralized structure wherein coordination is achieved through
dyadic predictive knowledge embedded between individuals. On the other hand, the boundary-spanner
structure is an example of a completely centralized structure wherein coordination is achieved through
architectural knowledge. What, then, happens to replication accuracy if architectural and predictive
knowledge are deployed simultaneously for coordination? Will it be superior to that of either coordination
mechanism, and will it be contingent on knowledge complexity?
To further examine the performance implications of the coordination mechanisms, we performed
another simulation experiment. We presented two cases that had not been discerned by prior studies: a case
wherein an agent possessing architectural knowledge performs interdependence coordination as well; and a
case wherein an agent possessing architectural knowledge delegates coordination tasks to other agents. For
this purpose, we added a centralized coordination mechanism to a mirroring structure. A member of a source
unit and a recipient unit were randomly picked, and they were given the role of implementing centralized
coordination. The replication accuracy of this structure was compared to that of a mirroring structure—a
decentralized structure—and that of a boundary-spanner structure—an archetype of centralized inter-unit
structures (Leifer and Delbecq, 1978; Tushman and Scanlan, 1981a, 1981b).
Figure 11 illustrates the simulation results. Unlike the completely centralized structure—the
boundary-spanner structure—the combination of a collective bridge structure and centralized coordination
performs well up to a threshold level of Q. Additionally, up to a certain point of Q, centralized coordination
improves the replication accuracy of a mirroring structure, regardless of N. However, after the threshold
value of Q, the replication accuracy rapidly declines, eventually leading to lower replication accuracy than
that of a mirroring structure.
--- Insert Figure 11 about here ---
An underlying mechanism at play is the tradeoff between the advantage of centralized coordination
and the information overload imposed on a central coordinator. After the threshold is reached, the
coordination benefit of the centralized coordination mechanism—elimination of redundant or erroneous
coordination caused by decentralized coordination—does not outweigh the cost of the centralized
coordination, leading to a lower level of accuracy in information processing by the central agent due to
information overload.
When accurate architectural knowledge is used, the level of the information load imposed on
individuals decreases because of the diminished use of predictive knowledge or direct ongoing
communications that are required to coordinate the interdependencies. As a result, in the case of centralized
coordination, if a central agent holds accurate and pertinent architectural knowledge, replication accuracy can
be improved through the minimization of coordination errors. However, the simulation results confirm that
such an advantage stemming from a centralized coordination mechanism and architectural knowledge has its
limits. When knowledge complexity is high, the information-processing activities required for reducing
dyad-level coordination errors eventually result in poor replication accuracy. Essentially, knowledge-
replication accuracy depends both on how the structure manages information overloads and on how the
structure achieves an accurate coordination of interdependencies.
Proposition 5. Centralized coordination mechanism improves the replication accuracy of mirroring
structures for low levels of knowledge complexity, but decreases replication accuracy for higher
levels of knowledge complexity.
Our findings suggest that the use of architectural knowledge for interdependence coordination as a
substitute for predictive knowledge pays off only up to a threshold level of knowledge complexity. In other
words, the substitutability between architectural and predictive knowledge, in terms of organizational
performance, is contingent on knowledge complexity. This boundary condition stems from the observation
that it is an agent who processes information, and thus, an agent performing coordination by using
architectural knowledge is subject to information overload when knowledge complexity is high.
Sensitivity analysis
We next performed a robustness check. First, we assessed the sensitivity of the results to the simulation
model parameters, repeating the entire simulation procedure by changing the value of each parameter. Table
1 summarizes the parameters and their corresponding values for a robustness check. In each of these
sensitivity analyses, the reported findings and propositions of this study remained intact and robust across all
ranges of the parameters. Some of the sensitivity analyses are notable. When the value of the information
load parameters—Cv and Cs—increases for a given value of information volume or scope, the replication
accuracy diminishes. Increasing Cv or Cs exerts analogous effects on information volume or scope with
increasing Q. In particular, when Cv or Cs is high, the replication accuracy of the boundary-spanner structure
is more negatively affected than in decentralized structures, such as the mirroring structure. This effect is
consistent with the effect of high N or Q on the replication accuracy. In addition, while our simulation model
and analyses are agnostic about whether intra- and inter-unit ties are formal or informal, the origin of ties
may affect the values of parameters associated with coordination failure and information overload: CPw and
CPc, and Cs and Cv,
9
respectively. Sensitivity analyses on the changes in these parameters indicate that the
findings and propositions remain intact.
Our model assumed a linear relationship between information load and replication accuracy. We
conducted a sensitivity analysis with a quadratic function representing the relationship. Results indicate that
the replication accuracy with the quadratic specification is lower than that with the linear specification,
which was expected, and the findings remain intact. This further reinforces our original findings, because
bounded rationality is further accentuated with a quadratic specification.
While the simulation model treated the information-processing capacity of boundary spanners as
equal to that of other individuals in the source or recipient unit, boundary spanners can be more capable than
other organizational members. In a sensitivity analysis, we introduced an additional parameter, boundary-
spanner capability, and investigated whether the inclusion of more capable boundary spanners would alter
the main findings. We operationalized the boundary-spanner capability as a multiplicative factor attenuating
the impact of information load on replication accuracy. Results revealed that boundary spanners needs to be
eight times more capable than other unit members for the boundary-spanner structure to outperform the
mirroring structure.
9
We speculate that CPw and CPc would be higher in general for informal ties than for formal ties, and that Cs and Cv
would be higher for formal ties than for informal ties. Unless predictive knowledge or common ground (e.g., shared
knowledge) is highly accurate, informal coordination would enable less accurate coordination than coordination through
formal mechanisms (e.g., common authority and integration). Meanwhile, formal coordination may generally
necessitate more deliberate information-processing activities than informal coordination would. Thus, we speculate that
Cs and Cv would be higher for formal ties than for informal ties. Future studies may examine the effects of the origin of
ties on replication outcomes.
DISCUSSION
Complex knowledge is a key element of organizational capability. Knowledge is a scale-free and fungible
resource (Anand and Delios, 2002; Anand and Singh, 1997; Anand et al., 2010, 2016; Levinthal and Wu,
2010; Penrose, 1959), and consequently it may be redeployed without loss in value and without
commensurate marginal costs. However, redeploying complex knowledge is more challenging than
transferring individually held knowledge. This study suggests that a proper design of inter-unit structures
helps facilitate the replication of complex knowledge. Thus, inter-unit structures have important implications
for the redeployment of complex knowledge to different units, multinational subsidiaries, alliances, or
acquisitions. Using an agent-based simulation, we examine the relationship between the design of inter-unit
structures and the accuracy of knowledge replication at different levels of knowledge complexity.
Theoretical implications
By simultaneously deploying the information-processing view and bounded rationality, we identified three
salient dimensions of inter-unit structures for knowledge replication and gained insights on the tension
between information-processing capacity and bounded-rationality concerns in designing inter-unit structures.
Simulation methodology proved a useful tool for theory development, as it aided in specifying assumptions
and analytically articulating relations between constructs, thereby helping to advance an integrated
theoretical framework. Our research contributes to the knowledge-replication literature and the information-
processing view of organizations in at least three ways.
First, whereas the extant literature largely suggests that additional inter-unit ties enhance the inter-
unit replication of complex knowledge, this study indicates that there is an optimal level of connectivity
between units. In the absence of accurate architectural knowledge, organizations undergo learning by
experimenting with alternative configurations of modules and interfaces (Baldwin and Clark, 1994; Sanchez
and Mahoney, 1996). Some knowledge interdependencies are not anticipated in the beginning of a project
and are uncovered only after learning by doing (Sosa et al., 2004). Additional inter-unit connectivity in our
simulation analysis can be interpreted as stemming from organizations that are experimenting to find an
effective design of an inter-unit structure in the absence of complete architectural knowledge. The results of
the simulation imply that experimenting to find an effective inter-unit structure is beneficial but pays off only
up to a certain point.
Second, we studied the mirroring hypothesis in the context of designing inter-unit structures for
redeploying complex knowledge. We examined the consequence of the misalignment between a structure
and knowledge by incorporating both information-processing and bounded-rationality factors.
Ignoring the association between the organizational design and the distribution of information load imposed
on organizational members, prior work, including that of Sosa et al. (2004), has examined the mirroring
hypothesis solely from the information-processing view. Consequently, although redundant ties have an
unequivocally negative influence on replication accuracy, this effect is not obvious from the extent literature.
By comparing the simulation results from the baseline and extended models, we disaggregated information-
processing and bounded-rationality factors in the impact of a misalignment between a structure and
knowledge on replication accuracy. In doing so, we found that the predictions of replication accuracy solely
based on the information-processing view become more biased as knowledge complexity increases. Thus, we
predict that Sosa et al.’s (2004) estimation of the effects of unmatched team interactions, which correspond
to redundant ties in our study, will become more biased as knowledge complexity increases (e.g.,
Martignoni, Menon, and Siggelkow, 2016).
Third, this study unpacks how replication accuracy is affected by coordination mechanisms
underlying the inter-unit structures at different levels of knowledge complexity. Our study speaks to Puranam
et al.’s (2012) call for study on the impact of the complexity of a task environment on the choice of
coordination mechanisms. Proposition 4 of our study sheds light on their hypothesis that delegation will be
greater when the task environment is complex. According to our contingency framework associated with
Proposition 4, when complexity is high, a decentralized inter-unit structure using predictive knowledge
performs better than a centralized inter-unit structure using architectural knowledge.
Proposition 5 extends the prior research on the substitutability between architectural and predictive
knowledge for interdependence coordination by distinguishing two cases that were not clearly discerned by
prior studies: a case wherein an agent possessing architectural knowledge performs interdependence
coordination, and a case wherein an agent possessing architectural knowledge delegates coordination tasks to
other agents. Simulation results showed that in the former case, architectural knowledge improves replication
accuracy only up to a threshold level of knowledge complexity. After this threshold, even in the presence of
architectural knowledge, replication accuracy sharply declines as knowledge complexity increases. An
underlying factor is that interdependence coordination in itself requires information-processing activities. In
contrast to the case of decentralized coordination using a set of dyadic predictive knowledge, in the case of
centralized coordination relying on architectural knowledge, such information-processing activities can
impose information overload on an agent performing interdependence coordination. This effect becomes
more pronounced when knowledge complexity is high. Thus, architectural knowledge can serve as a
complete substitute for predictive knowledge only when knowledge complexity is not high, or when
decentralized inter-unit structures are used.
Practical implications
This study also provides useful practical implications for designing inter-unit structures for redeploying
capabilities. First, when redeploying knowledge across units, it is important to differentiate the design of
inter-unit structures according to the knowledge complexity and knowledge configuration involved. Field
studies in the auto industry (Zhao and Anand, 2009) reveal that firms often deploy or develop inter-unit
structures that do not match existing knowledge. For example, it is common for firms to use ODT (only
direct ties) structures, in which only within-expertise ties are established between counterparts of source and
recipient units, for complex knowledge replication. Those structures with deficient or redundant ties,
including ODT structures, demonstrate a replication performance that is suboptimal to that of the collective
bridge structure. Collective bridge structures exist widely in real organizational settings. For example, a
structure similar to the collective bridge structure was developed between a group of Chinese engineers in
Volkswagen’s joint venture in Shanghai and the German engineers of Volkswagen’s home R&D unit when
the Chinese engineers went through an extensive on-site training in Germany (Zhao and Anand, 2009).
Meanwhile, in the absence of accurate and complete architectural knowledge, designing and instituting a
collective bridge structure may not be a feasible option. In that case, the second-best approach to designing
the structure is to increase the degree of connectivity between units. Although this approach is less efficient
than instituting a collective bridge structure, it increases replication accuracy up to a threshold level of
knowledge complexity. Managers should expect that knowledge with a higher number of modules or a
higher degree of interdependence between modules requires a higher degree of connectivity between units in
order to achieve optimal replication accuracy.
Second, when assessing the level of complexity in architectural knowledge for a new product design,
practitioners should consider both N and Q. Prior literature regarding complexity, with a few exceptions, has
focused primarily on the extent of interdependencies among components of a system (Q). Complexity is
somehow considered equivalent to the interdependence among components. However, this study reveals that
N has an independent impact on the performance of inter-unit structures. It is intuitive that when N and Q are
both large, as in the case of the Pratt and Whitney jet engine development project that involved 569
interdependencies among 54 components, knowledge complexity involved in the design capabilities of the
project is high (Sosa, Eppinger, and Rowles, 2007). Similarly, when N and Q are both low, as in the case of a
car door system (Rivkin and Siggelkow, 2006), the knowledge complexity involved in a project’s design
capabilities is low. However, our study suggests that one needs to be mindful of the potential interaction
between N and Q, in that when N is large, the performance impact of Q may be heightened. For example,
Boeing aimed to reduce the complexity of the design and production of its Dreamliner through
modularization, which in essence lowered the extent of knowledge interdependences (Q) among modules.
However, the sheer number of modules (N) still presents tremendous challenges for knowledge sharing and
integration (Kotha and Srikanth, 2013).
Third, we note the relevance of the collective bridge structure to practice based on its ease of
establishment. The most obvious method for creating a collective bridge structure across two units is doing it
in a centrally-planned manner, wherein a high-level manager has architectural knowledge and then
systematically builds an inter-unit structure to mirror a knowledge structure. However, field observations
suggest that rather than being formed by architectural knowledge in a centrally planned manner, the
collective bridge structure can also be developed tacitly through a simple heuristic with no need to articulate
the knowledge structure. The heuristic is that each person in the recipient unit forges a direct information-
processing interaction, through a direct inter-personal tie or co-location with counterparts in the source unit
who have the same and interdependent expertise areas. Once every individual follows this simple heuristic, a
collective bridge structure is established between two units, without the need for a central planner or
awareness of the entire knowledge structure. This simple implementation heuristic, which relies on a
decentralized set of predictive knowledge, makes the collective bridge idea much more practical.
10
10
We are grateful to Dan Levinthal for this insightful observation.
Limitations and future research
There are a few limitations of this simulation-based analysis that future studies—both simulation-based and
empirical—could address. First, the tacitness of knowledge, which has been identified as a major barrier to
knowledge replication (Kogut and Zander, 1996; Teece, 1986), was not explicitly considered. Although this
study did not include knowledge tacitness as a parameter, the operationalization of knowledge
interdependence in this study implicitly involves knowledge tacitness. Replication of knowledge
interdependences may involve extensive reciprocal adjustments and thus cannot be reduced to fully codified
and articulated interactions between individuals. Future studies may explicitly incorporate knowledge
tacitness and examine potential interactions between knowledge complexity and tacitness in determining the
impact of inter-unit structure on knowledge-replication outcomes.
In terms of research scope, this research assumed that source knowledge is exogenously given and
focused on replication outcomes, while ruling out factors associated with knowledge creation and retention.
With regard to two essential considerations in organization design—the alignment of incentives and
alignment of actions (Gulati, Lawrence, and Puranam, 2005)—this study presumed that incentives are fully
aligned between individuals in recipient and source units and that knowledge transfer occurs voluntarily. By
employing this assumption, this study was able to focus on how the design of inter-unit structures determines
the alignment of actions, and thus, replication accuracy. Future research can address the impact of the
misalignment of interests between units on the design of inter-unit structures and replication outcomes. For
example, if there is a lack of inter-unit trust, individuals in a source unit may not be motivated or willing to
share and transfer relevant knowledge to recipient unit members, potentially resulting in a poor replication
performance (McEvily, Perrone, and Zaheer, 2003). Further, this study assumed an alignment between
knowledge configuration and intra-unit structure, but Sosa et al. (2004) have documented the misalignment
between intra-unit communication channels and product architecture. Thus, the adoption of the assumption
of this alignment poses a boundary condition on our study. However, in light of the knowledge-based view of
the firm (e.g., Kogut and Zander, 1996), which suggests that misalignment is more likely to occur between
units than within a unit, our emphasis on inter-unit structures was a valid approach for the purpose of our
research question. Future research can investigate how intra-unit misalignment affects the optimal design of
inter-unit structure between source and recipient units.
Another fruitful avenue for future research is inter-temporal effects in designing inter-unit structures
for knowledge replication, since organizational design may be impacted by path dependence and
organizational inertia. For example, Rivkin and Siggelkow (2006) found that “unnecessary overlap across
departments” can sometimes help a firm to explore a broader range of choices, prohibiting premature lock-in
to suboptimal outcomes. An underlying mechanism is the effect of sequencing the decentralization and
centralization in organizing search over time. In this vein, as an extension of this study, it would be
interesting to develop a model of evolution of inter-unit structures and to identify the conditions under which
an inter-unit structure can evolve into an optimal structure.
Conclusion
We examined the impact of design factors underlying inter-unit organizational structures—inter-unit
connectivity, the extent of mirroring between the structure and knowledge configuration, and coordination
mechanisms—on the replication accuracy of complex knowledge. Through modeling and simulation
analyses, this study derived five propositions that offer a more nuanced understanding of the roles of inter-
unit structures in complex knowledge replication. Our study also sheds light on the tradeoff between dyad-
level and centralized coordination mechanisms, as well as the mirroring hypothesis, in the context of inter-
unit replication of complex knowledge. These issues involving the role of inter-unit structures in knowledge
replication have important implications for corporate growth, competitive advantage, and the redeployment
of resources to acquisitions, alliances, and multinational subsidiaries.
REFERENCES
Aldrich H, Herker D. 1977. Boundary spanning roles and organization structure. Academy of Management Review
2(2): 217–230.
Allen TJ. 1977. Managing the Flow of Technology. MIT Press: Cambridge, MA.
Anand J, Delios A. 2002. Relative and absolute resources as determinants of international acquisitions. Strategic
Management Journal 23(2): 119–135.
Anand J, Kim H, Lu S. 2016. Resource characteristics and redeployment strategies: Towards a theoretical
synthesis. Advances in Strategic Management 35: 155–184.
Anand J, Oriani R, Vassolo R. 2010. Alliance activity as a dynamic capability in the face of a discontinuous
technological change. Organization Science 21(6): 1213-1232.
Anand J, Singh H. 1997. Asset redeployment, acquisitions and corporate strategy in declining industries. Strategic
Management Journal 18(S1): 99–118.
Ancona DG, Caldwell DF. 1992. Bridging the boundary: external activity and performance in organizational
teams. Administrative Science Quarterly 37(4): 634–665.
Aral S, Alstyne MV. 2011. The diversity-bandwidth trade-off. American Journal of Sociology 117(1): 90–171.
Argote L, Ingram P, Levine JM, Moreland RL. 2000. Knowledge transfer in organizations: learning from the
experience of others. Organizational Behavior and Human Decision Processes 82(1): 1–8.
Argote L, McEvily B, Reagans R. 2003. Managing knowledge in organizations: an integrative framework and
review of emerging themes. Management Science 49(4): 571–582.
Baldwin CY. 2008. Where do transactions come from? Modularity, transactions, and the boundaries of firms.
Industrial and Corporate Change 17(1): 155–195.
Baldwin CY, Clark KB. 1994. Modularity-in-design: an analysis based on the theory of real options. Working
Paper, Harvard Business School.
Barney JB. 1997. Gaining and Sustaining Competitive Advantage. Addison-Wesley: Reading, MA.
Brown SL, Eisenhardt KM. 1997. The art of continuous change: linking complexity theory and time-paced
evolution in relentlessly shifting organizations. Administrative Science Quarterly 42(1): 1–34.
Cabigiosu A, Camuffo A. 2012. Beyond the “mirroring” hypothesis: product modularity and interorganizational
relations in the air conditioning industry. Organization Science 23(3): 686–703.
Capron L, Mitchell W, Swaminathan A. 2001. Asset divestiture following horizontal acquisitions: a dynamic
view. Strategic Management Journal 22(9): 817–844.
Carlile PR. 2002. A pragmatic view of knowledge and boundaries: boundary objects in new product development.
Organization Science 13(4): 442–455.
Carlile PR. 2004. Transferring, translating, and transforming: an integrative framework for managing knowledge
across boundaries. Organization Science 15(5): 555–568.
Centola D, Macy M. 2007. Complex contagions and the weakness of long ties. American Journal of Sociology
113(3): 702–734.
Colfer LJ, Baldwin CY. 2016. The mirroring hypothesis: theory, evidence and exceptions. Working Paper,
Harvard Business School.
Csaszar FA, Siggelkow N. 2010. How much to copy? Determinants of effective imitation breadth. Organization
Science 21(3): 661–676.
Egelhoff WG. 1993. Information-processing theory and the multinational corporation. In Organization Theory
and the Multinational Corporation, Ghoshal S, Westney DE (eds.). Palgrave: Basingstoke, UK, pp. 182‒
210.
Elfring T, Hulsink W. 2007. Networking by entrepreneurs: patterns of tie formation in emerging organizations.
Organization Studies 28(12): 1849–1872.
Fang C, Lee J, Schilling MA. 2010. Balancing exploration and exploitation through structural design: the isolation
of subgroups and organizational learning. Organization Science 21(3): 625–642.
Faraj S, Sproull L. 2000. Coordinating expertise in software development teams. Management Science 46(12):
1554–1568.
Fiol CM, Lyles MA. 1985. Organizational Learning, Academy of Management Review 10(4): 803–813.
Furlan A, Cabigiosu A, Camuffo A. 2014. When the mirror gets misted up: modularity and technological change.
Strategic Management Journal 35(6): 789–807.
Galbraith JR. 1973. Designing Complex Organizations. Addison-Wesley Longman Publishing Co., Inc.: Boston,
MA.
Galbraith JR. 1994. Competing with Flexible Lateral Organizations. Addison-Wesley: Reading, MA.
Ghoshal S, Korine H, Szulanski G. 1994. Interunit communication in multinational corporations. Management
Science 40(1): 96–110.
Grant RM. 1996. Toward a knowledge-based theory of the firm. Strategic Management Journal 17: 109–122.
Gulati R, Lawrence PR, Puranam P. 2005. Adaptation in vertical relationships: beyond incentive conflict.
Strategic Management Journal 26(5): 415–440.
Gupta AK, Govindarajan V. 1991. Knowledge flows and the structure of control within multinational
corporations. The Academy of Management Review 16(4): 768–792.
Hansen MT. 1999. The search-transfer problem: the role of weak ties in sharing knowledge across organization
subunits. Administrative Science Quarterly 44(1): 82–111.
Hansen MT. 2002. Knowledge networks: explaining effective knowledge sharing in multiunit companies.
Organization Science 13(3): 232–248.
Hansen MT, Mors ML, Løvås B. 2005. Knowledge sharing in organizations: multiple networks, multiple phases.
Academy of Management Journal 48(5): 776–793.
Hansen MT, Nohria N. 2004. How to build collaborative advantage. MIT Sloan Management Review 46(1): 22.
Henderson RM, Clark KB. 1990. Architectural innovation: the reconfiguration of existing product technologies
and the failure of established firms. Administrative Science Quarterly 35(1): 9–30.
Joyce WF. 1986. Matrix organization: a social experiment. Academy of Management Journal 29(3): 536-561.
Kenney M, Florida R. 2000. Venture capital in Silicon Valley: fueling new firm formation. In Understanding
Silicon Valley: the anatomy of an entrepreneurial region, Kenney M. (ed). Stanford University Press:
Stanford, CA, pp. 98–123.
Katz R, Tushman M. 1981. An investigation into the managerial roles and career paths of gatekeepers and project
supervisors in a major R & D facility. R&D Management 11(3): 103–110.
Keller R. 1986. Predictors of the performance of project groups in R&D organizations. Academy of Management
Journal 29(4): 715‒726.
Kogut B, Zander U. 1996. What firms do? Coordination, identity, and learning. Organization Science 7(5): 502–
518.
Kotha S, Srikanth K. 2013. Managing a global partnership model: lessons from the Boeing 787 "Dreamliner"
program. Global Strategy Journal 3(1): 41–66.
Kretschmer T, Puranam P. 2008. Integration through incentives within differentiated organizations. Organization
Science 19(6): 860–875.
Law AM, Kelton WD. 1991. Simulation Modeling and Analysis. McGraw-Hill: New York.
Lawrence PR, Lorsch JW. 1967. Differentiation and integration in complex organizations. Administrative Science
Quarterly 12: 1–47.
Leifer R, Delbecq A. 1978. Organizational/environmental interchange: a model of boundary spanning activity.
Academy of Management Review 3(1): 40–50.
Levinthal DA. 1997. Adaptation on rugged landscapes. Management Science 43(7): 934–950.
Levinthal DA, Wu B. 2010. Opportunity costs and non-scale free capabilities: profit maximization, corporate
scope, and profit margins. Strategic Management Journal 31(7): 780–801.
March JG, Simon HA. 1958. Organizations. John Wiley & Sons, Inc.: New York.
Markides CC, Williamson PJ. 1994. Related diversification, core competences and corporate performance.
Strategic Management Journal 15(S2): 149–165.
Mariotti F, Delbridge R. 2012. Overcoming network overload and redundancy in interorganizational networks: the
roles of potential and latent ties. Organization Science 23(2): 511–528.
Marrone JA, Tesluk PE, Carson JB. 2007. A multilevel investigation of antecedents and consequences of team
member boundary-spanning behavior. Academy of Management Journal 50(6): 1423–1439.
Martignoni D, Menon A, Siggelkow N. 2016. Consequences of misspecified mental models: Contrasting effects
and the role of cognitive fit. Strategic Management Journal 37(13): 2545–2568.
McEvily B, Perrone V, Zaheer A. 2003. Trust as an organizing principle. Organization Science 14(1): 91–103.
Morgan SL, Sørensen AB. 1999. Parental networks, social closure, and mathematics learning: a test of Coleman’s
social capital explanation of school effects. American Sociological Review 64: 661–681.
Morrison EW. 2002. Newcomers’ relationships: the role of social network ties during socialization. Academy of
Management Journal 45(6): 1149–1160.
Penrose ET. 1959. The Theory of the Growth of the Firm. Wiley: New York.
Pentland BT, Rueter HH. 1994. Organizational routines as grammars of action. Administrative Science Quarterly
39(3): 484–510.
Pil FK, Cohen SK. 2006. Modularity: implications for imitation, innovation, and sustained advantage. The
Academy of Management Review 31(4): 995–1011.
Postrel S. 2002. Islands of shared knowledge: specialization and mutual understanding in problem-solving teams.
Organization Science 13(3): 303–320.
Prochno P. 2003. Routine assembly: replicating and recreating practices in a new setting. PhD. diss., INSEAD,
Fontainebleau, France.
Puranam P, Raveendran M, Knudsen T. 2012. Organization design: the epistemic interdependence perspective.
Academy of Management Review 37(3): 419–440.
Puranam P, Singh H, Chaudhuri S. 2009. Integrating acquired capabilities: when structural integration is (un)
necessary. Organization Science 20(2): 313–328.
Reagans R, McEvily B. 2003. Network structure and knowledge transfer: the effects of cohesion and range.
Administrative Science Quarterly 48(2): 240–267.
Reed R, Defillippi RJ. 1990. Causal ambiguity, barriers to imitation, and sustainable competitive advantage.
Academy of Management Review 15(1): 88–102.
Rivkin JW. 2000. Imitation of complex strategies. Management Science 46(6): 824–844.
Rivkin JW, Siggelkow N. 2006. Organizing to strategize in the face of interactions: preventing premature lock-in.
Long Range Planning 39(6): 591–614.
Sanchez R, Mahoney JT. 1996. Modularity, flexibility, and knowledge management in product and organization
design. Strategic Management Journal 17(S2): 63–76.
Schilling MA. 2000. Toward a general modular systems theory and its application to interfirm product modularity.
Academy of Management Review 25(2): 312–334.
Siggelkow N, Rivkin JW. 2005. Speed and search: designing organizations for turbulence and complexity.
Organization Science 16(2): 101–122.
Siggelkow N, Rivkin JW. 2006. When exploration backfires: unintended consequences of multilevel
organizational search. Academy of Management Journal 49(4): 779–795.
Simon HA. 1962. The architecture of complexity. Proceedings of the American Philosophical Society 106(6):
467–482.
Simon HA. 1991. Bounded rationality and organizational learning. Organization Science 2(1): 125–134.
Singh J. 2005. Collaborative networks as determinants of knowledge diffusion patterns. Management Science
51(5): 756–770.
Sosa ME, Eppinger SD, Rowles CM. 2004. The misalignment of product architecture and organizational structure
in complex product development. Management Science 50(12): 1674–1689.
Sosa ME, Eppinger SD, Rowles CM. 2007. A network approach to define modularity of components in complex
products. Journal of Mechanical Design 129(11): 1118–1129.
Spender JC. 1996. Making knowledge the basis of a dynamic theory of the firm. Strategic Management Journal
17(S2): 45–62.
Szulanski G. 1996. Exploring internal stickiness: impediments to the transfer of best practice within the firm.
Strategic Management Journal 17(S2): 27–43.
Szulanski G. 2000. The process of knowledge transfer: a diachronic analysis of stickiness. Organizational
Behavior and Human Decision Processes 82(1): 9–27.
Teece DJ. 1986. Profiting from technological innovation: implications for integration, collaboration, licensing and
public policy. Research Policy 15: 285–305.
Turner KL, Makhija MV. 2012. The role of individuals in the information processing perspective. Strategic
Management Journal 33(6): 661–680.
Tushman ML, Nadler DA. 1978. Information processing as an integrating concept in organizational design.
Academy of Management Review 3(3): 613–624.
Tushman ML, Scanlan TJ. 1981a. Boundary spanning individuals: their role in information transfer and their
antecedents. Academy of Management Journal 24(2): 289–305.
Tushman ML, Scanlan TJ. 1981b. Characteristics and external orientations of boundary spanning individuals.
Academy of Management Journal 24(1): 83–98.
Winter SG, Szulanski G. 2001. Replication as strategy. Organization Science 12(6): 730–743.
Zhao Z, Anand J. 2009. A multilevel perspective on knowledge transfer: evidence from the Chinese automotive
industry. Strategic Management Journal 30(9): 959–983.
Zhao Z, Anand J. 2013. Beyond boundary spanners: the “collective bridge” as an efficient interunit
structure for transferring collective knowledge. Strategic Management Journal 34(13): 1513–
1530.
TABLES
Table 1. Model parameters
Category
Symbol
Representation
Parameter values for
simulation
Knowledge
N
dimension of knowledge tuple; i.e., total
number of expertise areas of complex
knowledge
5, 10, 15, 20
Q
total number of interdependencies between
expertise areas
1, 5, 10, 15
Information overload
Cv
intensity of information volume overload
0.02, 0.04, 0.1, 0.2
Cs
intensity of information scope overload
0.02, 0.04, 0.1, 0.2
Coordination failure
CPw
coordination failure penalty due to lack of
within-expertise ties
0.1, 0.25, 0.4
CPc
coordination failure penalty due to lack of
cross-expertise ties
0.1, 0.25, 0.4
Table 2. Summary of mechanisms underlying results and propositions
Information overload
Dyad-level
coordination
Centralized
coordination
Overall effect
P1. Inter-unit
connectivity
Additional interunit
ties may deliver
relevant knowledge,
but too many inter-unit
ties may cause
information overload.
Additional inter-
unit ties may
improve
interdependence
coordination.
n/a
Inverted-U relationship
between inter-unit
connectivity and
replication accuracy.
P2. Mirroring
between
structure and
knowledge
The same as above.
Mirroring leads to
better dyad-level
coordination.
n/a
Higher extent of mirroring
leads to higher replication
accuracy
P3. Information
processing view
and bounded
rationality
It is considered by the
bounded rationality
view.
It is considered by
the information
processing view.
It is considered by
the information
processing view.
Predictions solely based
on information processing
view become more biased
as knowledge complexity
increases.
P4.
Decentralized
vs. centralized
structures
A decentralized
coordination
mechanism has an
advantage.
Good fit with high
knowledge
complexity.
Good fit with low
knowledge
complexity.
The advantage of one
mechanism over another is
contingent on knowledge
complexity.
P5. Centralized
coordination
mechanism
Information-processing
activities for
centralized
coordination may
increase information
load.
n/a
A centralized
coordination
mechanism may
reduce
coordination
failure.
Centralized coordination
with architectural
knowledge improves
replication accuracy only
up to a threshold level of
knowledge complexity.
Table 3. Mean values of information volume and scope loads
Structure with deficient ties
Structure with redundant ties
Q
Information
volume
Information
scope
Information
volume
Information
scope
5
3.17
1.90
1.98
4.10
15
5.10
5.90
3.97
8.10
FIGURES
Figure 1a. An example of inter-unit knowledge replication
Figure 1b. An example of a collective bridge structure
Figure 1c. An example of an inter-unit structure with deficient ties
Figure 1d. An example of an inter-unit structure with redundant ties
Figure 1e. Boundary-spanner structure
Source Unit
ttUnit Unit
?
Inter-unit
structure
s2
s1
s4
s3
Recipient Unit
UUnitUnitugu
r4
r1
r2
r3
Recipient Unit
s2
s3
s1
s4
Source Unit
tUnit
r2
r4
r1
r3
s2
s3
s1
s4
Source Unit
Unit
r2
r4
r1
r3
Recipient Unit
Recipient Unit
s2
s3
s1
s4
Source Unit
UUUnit
r2
r4
r1
r3
Recipient Unit
s3
s4
s2
r3
r2
r1
r4
Source Unit
Unit
s1
Figure 2. Inter-unit connectivity and replication accuracy
Figure 3. Deficient ties and replication accuracy
Figure 4. Redundant ties and replication accuracy
(b) Decomposition of performance
(a) Effect of Ties and Q
(b) N=20
(a) N=10
(a) N=10
(b) N=20
Figure 5. Distribution of information load: Structure with deficient ties
Figure 6. Distribution of information load: Structure with redundant ties
Figure 7. Impact of bounded rationality: A case of redundant ties
(a) Information Volume: Deficient Ties
(b) Information Scope: Deficient Ties
(a) Information Volume: Redundant Ties
(b) Information Scope: Redundant Ties
(a) Q=5
(b) Q=15
Figure 8. Impact of bounded rationality: A case of deficient ties
Figure 9. Replication accuracy of decentralized structures and boundary spanner structure
Figure 10. Distribution of information load: Boundary-spanner and collective bridge structures
(a) Q=5
(b) Q=15
(a) Q=5
(b) Q=15
(a) Q=15
(b) Q=15
Figure 11. Centralized coordination and replication accuracy
(c) N=20
(a) N=10
