Williams TransferContextReplication 2007

Transfer in Context: Replication and Adaptation in Knowledge Transfer Relationships
Author(s): Charles Williams

Source: Strategic Management Journal, Vol. 28, No. 9 (Sep., 2007), pp. 867-889
Published by: Wiley
Stable URL: https://www.jstor.org/stable/20141955
Accessed: 31-05-2024 14:32 +00:00
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
Wiley is collaborating with JSTOR to digitize, preserve and extend access to Strategic
Management Journal
This content downloaded from 112.207.190.2 on Fri, 31 May 2024 14:32:21 +00:00
All use subject to https://about.jstor.org/terms
Strategie Management Journal
Strat. Mgmt. J., 28: 867-889 (2007)
Published online 4 May 2007 in Wiley InterScience (www.interscience.wiley.com) DOI: 10.1002/smj.614
Received 4 October 2004; Final revision received 6 November 2006
TRANSFER IN CONTEXT: REPLICATION AND

ADAPTATION IN KNOWLEDGE TRANSFER
RELATIONSHIPS
CHARLES WILLIAMS*
Fuqua School of Business, Duke University, Durham, North Carolina, U.S.A.
This paper explores the role of replication and adaptation in knowledge transfer relationships. I
develop a model of knowledge transfer in which firms replicate because knowledge is ambiguous
and adapt because knowledge depends on context. In the model, firms replicate more when
knowledge is discrete and adapt more when they understand the interactions between different
areas of knowledge. Replication and adaptation lead to successful knowledge transfer, which
leads to improved performance of the receiving unit. The predictions are tested using a survey of
cross-border knowledge transfer relationships among firms in the telecommunications industry.
The results are largely consistent with the model and point to potential areas for future
research, such as the drivers of replication, the depreciation rate of knowledge, and the role
of understanding in organizational knowledge. Copyright ? 2007 John Wiley & Sons, Ltd.
This paper presents a model of knowledge trans The idea that organizational knowledge is tacit
fer that follows from the dual nature of knowl and ambiguous has played a central role in research
edge?the fact that it is both ambiguous and on strategy and organizations (Kogut and Zander,
context-dependent. Recognizing these two aspects 1992; Nelson and Winter, 1982). Causal ambiguity
of knowledge has several implications for knowl is inherent to most complex production processes
edge transfer relationships. First, knowledge trans (Lippman and Rumelt, 1982), so firm members
fer will usually involve knowledge that requires often do not understand the root causes of firm
replication and knowledge that requires adaptation. performance or the interaction between individ
Second, if firms combine replication and adapta ual activities. Replication, which is effort towards
tion in knowledge transfer, we need to change our exact copying of a set of activities, enables the
understanding of replication in the transfer rela transfer of those activities without the need to
tionship as the absence of adaptation. I develop an understand their causes, consequences, and inter
approach to knowledge transfer that incorporates dependence. Thus, researchers have proposed that
these insights and test this model on a sample of firms replicate knowledge to transfer it in the face
knowledge transfer relationships in telecommuni of ambiguity (Winter and Szulanski, 2002).
cations services.
Management scholars have also recognized the
context-dependent nature of knowledge in organi
zations (Argote and Ingram, 2000; Kostova and
Keywords: knowledge transfer; replication; adaptation; Roth, 2002; Prahalad and Doz, 1987). The rich
evolution
connection between organizations and their envi
^Correspondence to: Charles Williams, Fuqua School of Busi
ness, Duke University, Box 90120, Durham, NC 27708, U.S.A. ronment is a long-standing concern of organiza
E-mail: charlesw@duke.edu tional scholars (Henderson and Mitchell, 1997;
* ""*" ?WILEY
\ InterScience^
DISCOVER
Copyright ? 2007 John Wiley & Sons, Ltd.
868 C. Williams
Nelson and Winter, 1982; Tushman and Anderson, relationships with industry leaders. To gather data
1986), and these links determine the composition on these relationships, I surveyed relationship
of knowledge in the organization and its fit with partners.
the new environment. To transfer knowledge effec The data from these surveys are largely consis
tively in the face of context dependence, firms tent with the hypotheses presented in the paper,
must adapt knowledge to a new setting. suggesting that replication and adaptation con
These insights into knowledge have arisen in tribute to knowledge transfer, and that transfer
separate streams, however, so studies tend to contributes to improvements in receiver perfor
emphasize either the need for replication or the mance. The results also suggest that the most sig
need for adaptation. In this study, I find that firms nificant factor for facilitating adaptation is expe
are often engaged in both careful copying of prac rience within the relationship, rather than prior
tices from a partner and significant adaptation of transfer experience. Robustness analysis suggests
practices. Accounting and reporting measures, for that the performance gains from knowledge trans
example, are frequently replicated because they are fer depreciate rapidly. The evidence is not con
highly interdependent with the transferring partner. sistent with the hypothesized relationship between
On the other hand, customer service practices and discrete knowledge and replication: it suggests that
marketing are frequently adapted because of the firms manage interdependence not by reducing
extent to which they interact with the unique envi replication, but by increasing adaptation.
ronment of each location. In other cases, replica
tion and adaptation are combined within the same
set of practices. A set of Latin American telephone A MODEL OF KNOWLEDGE
companies balance replication and adaptation in TRANSFER
their human resource practices. The firms replicate
a set of common programs for the rotation, devel This section lays out five hypotheses, which form
opment, and compensation of managers. Each firm, the basis of a model of knowledge transfer. I begin
however, adapts the exact positions and levels of by outlining the primary aspects of organizational
the hierarchy to its own institutional environment. knowledge described by prior research, and how
The model of knowledge transfer presented in they suggest the need to replicate and adapt knowl
this paper recognizes the joint role that replica edge in the transfer process. I then discuss how to
tion and adaptation play in a knowledge transfer reconcile the use of replication and adaptation in
relationship. In the model, replication rises when the knowledge transfer process. I propose two con
knowledge is more discrete, while adaptation rises tingencies that lead to higher levels of replication
when the receiving firm understands the knowl and adaptation, respectively. Finally, I hypothe
edge it is getting. Replication and adaptation then size that replication and adaptation contribute to
lead to increased knowledge transfer, which in turn knowledge transfer, which then leads to perfor
leads to greater performance gains at the receiving mance improvements at the receiving unit.
unit. Most theorists locate organizational knowledge
I study knowledge transfer relationships in the in the actions of the organization. Nelson and Win
context of the telecommunications industry. A ter (1982: 99) 'propose that organizations remem
radically new set of tools?digital switching, fiber ber by doing' Similarly, Spender (1996) empha
optic transmission, data transmission protocols, sizes that knowledge is inextricably linked to col
and wireless communications?have transformed lective action in organizations. Nonaka (1994: 16)
the telecom sector. These new technologies are also suggests that 'knowledge is deeply rooted in
necessary to build a modern communications action.' Thus, firms possess knowledge only if they
industry, but companies in less developed can put it into action.
markets have found that technology alone is Other approaches to knowledge focus on the
not sufficient. These firms have found they integration of diverse components of knowl
also require complementary knowledge, in the edge. Argote and Ingram (2000) propose that
form of practices and information, to provide knowledge resides in three reservoirs within the
modern services. To acquire this complementary firm?members, tools, and tasks?and in the net
knowledge, many firms used the impetus of works that link them. In their framework, firms'
regulatory reform to form knowledge transfer performance depends on the extent to which these
Copyright ? 2007 John Wiley & Sons, Ltd. Strat. Mgmt. J., 28: 867-889 (2007)
DOI: 10.1002/smj
Transfer in Context 869
networks are compatible with each other and with relationships between replication and adaptation,
the external environment. In a similar vein, Grant knowledge transfer, and performance. Data from
(1996) argues that knowledge resides in the orga 2001 are used to compare the contingencies driving
nization's ability to take individual knowledge and replication and adaptation early in the relationship
integrate it. to later periods.
These aspects of knowledge suggest two key This approach to knowledge transfer differs
characteristics of organizational knowledge: causal from recent studies in two key ways: the focus is
ambiguity and context dependence. Causal ambi the knowledge transfer relationship rather than the
guity (Lippman and Rumelt, 1982) arises because transfer of a single practice, and knowledge trans
knowledge is embodied in the repeated activi fer is treated as the outcome of the process rather
ties of the organization, known as routines. Rou than a distinct point in the process. While recent
tines link together the actions of organization studies of knowledge transfer focus on the trans
members, who may not understand, or even be fer of a single, targeted practice (e.g., Kostova and
aware of, actions elsewhere in the chain. Since Roth, 2002; Szulanski, 1996), knowledge transfer
these chains are long and incompletely understood, relationships frequently encompass a wide variety
no member of the organization will completely of areas targeted for transfer (Argote, 1999). For
understand the relationship between an organiza instance, the firms in this study face simultane
tion's actions and outcomes (Nelson and Win ous pressure to expand networks, improve service
ter, 1982). On the other hand, context depen quality, establish positive customer relationships,
dence arises because knowledge integrates com and eliminate waiting lists for service, so they turn
ponents of knowledge?such as people, personal to their partner for many types of knowledge. To
networks, or information?which vary between manage the complexity of these multiple links in
settings. Since compatibility between components the knowledge transfer relationship, firms need to
is necessary for knowledge to be effective (Argote approach the transfer relationship with a consistent
and Ingram, 2000), context will lead knowledge to strategy, and the nature of these strategies is the
vary in order to achieve similar outcomes while focus of this study.
integrating differing components. When a firm is engaged in a knowledge trans
Since ambiguous causality and context depen fer relationship, it must decide how to allocate its
dence are fundamental aspects of organizational effort with regard to two competing goals: chang
knowledge, knowledge transfer requires firms to ing the receiving unit's operations to be more like
resolve a number of uncertain relationships: the its partner's?replication?and changing its oper
relationship between an activity and its outcome, ations to integrate with local context?adaptation.
the relationship between an activity and the rest Theoretical approaches to replication have empha
of the organization, and the relationship between sized the costly mistakes created by adaptation
an activity and the firm's environment. Resolution (Winter, 1995; Winter and Szulanski, 1998, 2002).
of these relationships will involve a continuous From this perspective, replication reduces or rules
process of modification and observation: change a out adaptation, which poses problems if, in addi
practice, observe outcomes, observe interactions, tion to causal ambiguity, organizational knowledge
change again. By applying iterative changes at exhibits context dependence. Though replication
the receiving unit followed by observations at the and adaptation are typically understood as oppos
source and receiving units, firms can gradually ing approaches to transfer, this study examines the
copy practices from the source and integrate them possibility that firms might use them together.
with the new context, thus accomplishing knowl The joint use of replication and adaptation sug
edge transfer. gests an apparent contradiction between copying a
To study firms' approach to this process, I mea practice exactly and modifying it: at some fun
sure their transfer activities (replication and adap damental level, exact copying and modification
tation) and the outcome of those activities (knowl might be mutually exclusive. I resolve this con
edge transfer) in two defined periods: the first flict by treating replication and adaptation as effort
year of the knowledge transfer relationship, which towards a goal. Thus, I define replication as effort
varies between the firms, and the year of the sur aimed at creating activities at one location that
vey, 2001. Data from the first year are used to test are identical to those at another location. Sim
the structural model, which includes simultaneous ilarly, I define adaptation as effort toward the
DOI: 10.1002/smj
870 C. Williams
goal of modifying or combining practices from a When knowledge is more discrete, the costs of
source unit. Replication and adaptation are modes, replication are lower because the unit may copy
then, by which organizations accomplish knowl a set of activities with less consideration of their
edge transfer, which I define as the acquisition external links. It will be harder and more expensive
from another unit of useful information or prac to replicate knowledge that is less discrete because
tices. This approach allows for combined modes it is intertwined with areas outside the firm. If a
of transfer: an organization can invest substantial firm replicates highly interconnected knowledge
effort towards exact copying, but still invest some without context or related areas of knowledge,
effort in adaptation in areas where exact copying it will find this knowledge less well integrated,
is difficult or impossible. This might arise if a and less useful. It follows, then, that the cost of
firm is working to copy a complex and successful replicating knowledge falls as knowledge becomes
set of practices for dispatching network mainte more discrete. As these costs fall, I expect that
nance teams quickly, but also needs to adapt them firms will invest more in replication.
to the nature of local infrastructure?for instance,
whether lines are strung above ground on poles or Hypothesis 2: When knowledge to be transferred
underground in sewers. is more discrete, replication by the receiving unit
Whether firms typically combine these modes will be higher.
of transfer or whether replication arises separately
from adaptation is a question that can be empiri Adaptation, on the other hand, can be costly
cally resolved. To distinguish whether replication because of the challenge of predicting interactions
and adaptation are two ends of a single spectrum of between interdependent knowledge in the organi
transfer practices or two separate modes of trans zation (Winter and Szulanski, 2002). When a firm
fer, we can test whether they vary together or sep makes more accurate predictions of the interaction
arately. If replication and adaptation are mutually between a potential business practice and other
exclusive transfer strategies then the two will vary parts of the organization, adaptation will be less
inversely?as replication rises, adaptation will fall, costly. In interviews, some participants in transfer
and vice versa. In this case, they can be represented relationships pointed to the ability to understand
as opposite indicators of a single construct of trans interdependence as an essential practice for suc
fer mode. In contrast, when replication is defined cessful knowledge transfer. I define understand
as effort they may vary separately?replication ing of knowledge to represent two capabilities: the
may rise without affecting adaptation and vice ability to predict how a change in one area of
versa. If the two vary independently, a factor knowledge will affect other parts of the organi
model with two transfer modes will fit the data zation and the ability to identify the essential ele
better than one with a single construct of transfer ments of a set of practices. Understanding enables
mode. This leads to my first hypothesis. the firm to tailor changes by anticipating inter
dependence and thereby reduces the unintended
Hypothesis 1: Replication and adaptation by the problems from adaptation. Increasing the effective
receiving unit represent two separate constructs ness of adaptation reduces the costs of adaptation.
rather than a single continuum of transfer mode. Thus, I predict adaptation will increase with under
standing.
Levels of replication and adaptation

Hypothesis 3: When the units possess greater
If replication and adaptation vary separately, spe understanding of their knowledge, adaptation by
cific contingencies will lead them to rise and fall. I the receiving unit will be higher.
propose that replication will rise when knowledge
is more discrete and adaptation will rise when firms A structural model of transfer and
understand the interactions between different areas
performance
of knowledge.
A firm's knowledge is more discrete when it I have proposed that adaptation and replication are
is more self-contained, with fewer connections separate mechanisms for transfer and that firms
to actors outside the firm, such as customers, vary their use based on contingent factors of dis
suppliers, or local institutions and infrastructure. crete knowledge and understanding. Next, I present
DOI: 10.1002/smj
three hypotheses that complete a structural model. multinationals adapt human resource practices to
The full model begins in the previous section with local institutions (Kostova and Roth, 2002).
the contingencies leading to higher replication or Adaptation increases the total amount of knowl
adaptation. The next two hypotheses propose that edge transfer because it reduces the cost of trans
replication and adaptation will each contribute to ferring a valuable set of practices. Without adapta
knowledge transfer. The final hypothesis proposes tion, replication has inconvenient cascading impli
that knowledge transfer will lead to subsequent cations. When using high levels of replication
improvements in performance at the receiving unit. without any adaptation, a firm that targets a set
Replication leads to knowledge transfer because of practices for transfer will also need to replicate
it is necessary to create a new working any activities that are linked with the focal prac
copy of complex and ambiguous knowledge tices. These linked activities will have links of their
(Winter, 1995). When knowledge is complex and own, creating a widening set of practices that must
causally ambiguous, a firm may not be able to be replicated simply to acquire the original target.
predict which elements of the knowledge are Adaptation can reduce the cost of transfer by mod
essential for its effective operation. By investing ifying the connections between a set of practices
significant effort in copying a set of practices and the rest of the organization or environment,
exactly?replicating?a firm insures that the allowing those practices to be targeted without
transferred practices contain any elements of the replicating related areas. Adaptation enables firms
knowledge that might turn out to be essential. to focus on potentially valuable knowledge and
A set of practices with all essential elements integrate it with the new context.
is more likely to be effective, and when new In addition, adaptation can make transferred
practices are more effective they will be more knowledge more effective. The effectiveness of
valuable to the receiving partner. This represents organizational knowledge depends on the integra
increased knowledge transfer. Thus, the more a tion of diverse components of knowledge (Argote
receiving unit invests in replication the greater the and Ingram, 2000; Grant, 1996). Some of these
knowledge transfer it will accomplish. diverse components of knowledge must vary in
new contexts, so specific practices will need to
Hypothesis 4a: The greater the level of replica change to integrate them. For instance, if the
tion by the receiving unit, the greater the knowl legacy switching equipment for the existing tele
edge transfer to the receiving unit. phone network is made by a local supplier, then
the firm's management systems must be changed
Transfer only helps firms if the knowledge is to monitor and manage the legacy equipment.
suited to the new context. Many lines of research Thus, when firms engage in adaptive efforts they
on strategy, technology, and international business will be able to transfer more knowledge and make
suggest that knowledge depends on its context. that knowledge more effective in the new context.
Argote and Ingram (2000: 159) argue that com
patibility between networks of knowledge reser Hypothesis 4b: The greater the level of adapta
voirs is difficult to replicate in a new environ tion by the receiving unit, the greater the knowl
ment. Winter (1995: 154) points out that repli edge transfer to the receiving unit.
cation will often be accompanied by adaptation:
Tn many actual cases the objective is not exact Scholars generally posit that knowledge transfer
replication but partial replication, accompanied by is a key source of competitive advantage, but
adaptive or innovative change in some routines.' the hypothesized link between knowledge transfer
Allen (1977) proposes that firms require techno and performance is rarely verified. Kogut and
logical gatekeepers because technical knowledge Zander (1992) argue that successful firms must
from other organizations needs to be translated be able to replicate their organizational knowledge
for use within the firm. Leonard-Barton (1988) to grow. Winter proposes that knowledge is a
finds that the implementation of a new production strategic asset (Winter, 1987) and a source of rents
technology requires mutual adaptation between the for firms (Winter, 1995). Argote and Ingram (2000)
technology and organization to such an extent that hypothesize that knowledge transfer is a basis
it is best understood as an extension of the inno for the competitive advantage of firms, since the
vation process. Empirical studies have found that effectiveness of transfer varies considerably among
DOI: 10.1002/smj
872 C. Williams
organizations. Still, most studies of knowledge effort toward copying and modifying practices,
transfer either focus on transfer alone through which allows for replication and adaptation in
survey measures (Gupta and Govindarajan, 2000; knowledge transfer. In the model, replication rises
Szulanski, 1996; Winter and Szulanski, 1998) or when knowledge is more discrete while adaptation
use performance improvements in the presence rises when firms understand interactions between
of transfer relationships to proxy for knowledge knowledge and context. Replication and adaptation
transfer (Ahuja, 2000; Argote, 1999; Baum and each contribute to knowledge transfer, and knowl
Ingram, 1998; Darr, Argote, and Epple, 1995). edge transfer leads to higher performance at the
Knowledge transfer leads to performance receiving unit.
improvements because the valuable new practices
and information enable the firm to accomplish
its goals more quickly and at lower cost, and
DATA AND METHODS
even accomplish new goals. Nelson and Winter
(1982: 123) point out that access to a source
unit's template is a key ingredient for successful This section explains my empirical approach. I
knowledge transfer, otherwise '[w]hen problems discuss the advantages of the empirical setting
arise in the copy, it is not possible to resolve for studying knowledge transfer. I explain the
them by closer scrutiny of the original.' Without data collection methods and statistical analysis.
access to the template, a firm that wishes to Finally, I describe the variables used to represent
acquire knowledge has little information to work the primary constructs.
with. Since the most important elements of
organizational knowledge tend to be tacit, the
ability to observe, ask, articulate, and check is an Empirical setting
essential element of improving the effectiveness of
I test the theoretical model in the context of inter
practices through knowledge transfer relationships.
Because of the ability to clarify and explore national transfer relationships among telecommu
complex relationships in the template, knowledge nications service firms. These relationships have a
number of characteristics that make them an inter
transfer relationships help firms acquire the most
complex, and often most valuable, practices, which esting setting to study knowledge transfer. The
will improve the performance of the receiving unit. relationships were formed with the express purpose
Even when an organization is not engaged in of transferring knowledge, and the source partners
wide-scale copying of practices, information from were frequently chosen for their ability to transfer
a successful partner can help solve pressing prob knowledge. There is a clear source of knowledge
lems an organization faces. When firms face prob and receiver of knowledge in these settings, so the
lems, they search their information environment outcome of the relationship is more straightforward
for potential solutions (Cyert and March, 1992; than in other relationships where transfer can be
Nelson and Winter, 1982). The knowledge transfer non-cooperative and in both directions. In addi
relationship gives the receiving firm more infor tion, the telecom service industry features a high
mation on potential solutions, which improves the level of tacit knowledge that firms must employ
changes it makes. to provide service effectively. There is an array
of advanced equipment, but complementary orga
Hypothesis 5: Greater knowledge transfer will nizational knowledge is required to operate the
lead to higher performance at the receiving unit. equipment effectively.
The rapid diffusion of new telecommunications
When we recognize that ambiguity and context equipment also means that the operating routines
dependence arise from the fundamental nature of of advanced telecom firms have some relevance
organizational knowledge, a more complete picture to firms in less developed markets. On the other
of the knowledge transfer process emerges. Firms hand, the industry still features multi-domestic
replicate because some knowledge is ambiguous competition, in which coordinated international
and adapt because some knowledge depends on strategies do not play an important role. Thus,
context. While the two transfer modes have been replication is less likely to be imposed by the need
treated as mutually exclusive, I define them as for coordination.
DOI: 10.1002/smj
Data collection of responses and target firms for telephone pen
etration. In addition, the Kolmogorov-Smirnov
I use a survey and additional archival measures test, which detects clustering in the response
as sources of data on the firms and settings for sample (Conover, 1971), found that the survey
these transfer relationships. The population of rela responses cluster in some areas of GNI even
tionships studied here are cross-border investments though the means did not differ. Overall, the
in fixed-telephone and cellular service providers. I response sample is similar to the target in basic
targeted firms in these relationships using infor country characteristics of population and income,
mation from Pyramid Research (a former division but it differs in terms of telecommunications devel
of the Economist Intelligence Unit that performs opment (slightly higher) and income levels (clus
market research on telecommunications markets tered data).
around the world) and RCR Wireless, a pub
lisher focused on the wireless industry. I defined
the relationships as those in which a multina Analysis
tional telecommunications firm invested (minor
I test the hypotheses in the context of a struc
ity or majority stake) in a local telecommunica
tural equation model (SEM), which represents the
tions firm. I gathered contact information for 324
simultaneous relationships between replication and
unique relationships using RCR's Global Wire
adaptation, knowledge transfer, and performance
less Database, Dunn & Bradstreet, and company
of the receiving unit. This allows me to test the
reports.
relationships in a framework that accounts for
I developed the survey through a modified
the direct and indirect relationships between these
version of Dillman's Tailored Design Method variables. The structural model also includes a
(Dillman, 2000). The survey asked respondents measurement model, which corrects for measure
to rate each item on a five-point Likert scale ment error in latent constructs (e.g., knowledge
(agree/disagree) for the current year (2001) and transfer) by using multiple observed indicators of
for the first year of the relationship (which varies them.
by firm). These items served as indicators for the The number of observations is low by the field's
constructs of replication, adaptation, and knowl rules of thumb for SEM. Small samples are known
edge transfer (see Variables section). I mailed the to bias fit statistics downward, leading to over
surveys to the top decision-maker at each of the rejection of true models (Hu and Bentler, 1999). A
firms in two batches in late summer and early fall study, which examined the effect of sample size on
2001. Follow-up to the survey included a post card parameter estimates, however, found no evidence
3 weeks after the survey mailing, phone calls to the of bias; most of the problems from small sam
targeted respondent, and regular emails to those ples arose because of non-convergence of the esti
who planned to complete the survey but had not mation algorithms (Marsh et al., 1998). Since the
yet returned it. focus of this study is the relationship between key
At the end of this process, I received 62 com constructs, which are represented by the param
pleted surveys for a final response rate of 19 eter estimates, I use SEM to test the hypotheses
percent. Two of those responses were from rela while accounting for measurement error, which is
tionships begun 30-40 years earlier, so I did not a common problem with survey studies. I also con
include them in the final analysis. I compared the duct considerable robustness analysis to check if
respondents to the 324 targets for three country the findings are stable across different specifica
characteristics: gross national income (GNI), pop tions that reduce estimation problems or increase
ulation, and telephone penetration (lines per 100 discriminatory power. I present alternative results
inhabitants). Based on a comparison of means test, based on ordinary least squares (OLS) regression,
I did not find a significant difference between the which has better small sample estimation proper
responses and the targeted firms for GNI or popu ties than SEM. The small sample limits the dis
lation. I also performed a Kolmogorov-Smirnov criminatory power of the study in any specifica
nonparametric comparison of the samples and tion, however, so the results are most likely to be
found that they did not differ based on popula valid for large effects but may under-report the
tion. I did find a difference between the means significance of small effects.
DOI: 10.1002/smj
874 C. Williams
Tests of performance, such as Hypothesis 5, vary with different scales assigned to a parame
are fraught with selection problems because firms ter. Generally, this approach reduced the level of
make investment decisions based on anticipated significance of the relationships compared to the
performance benefits (Masten, 1993; Shaver, p-values from Wald tests.
1998). Selection problems arise when a population
is not randomly selected from among the possible Variables
population parameters (Greene, 2003; Heckman,
1979; Maddala, 1983). In this case, selection bias Appendix 1 presents descriptions and descriptive
could arise because the firms choose their efforts statistics for all 42 observed variables in the study,
toward replication and adaptation with some fore including the survey items used to measure the
sight. When higher performance from transfer is latent constructs of replication, adaptation, and
anticipated, the firms are likely to invest more knowledge transfer. Appendix 2 presents the cor
relation of these variables.
in replication and adaptation. A structural model
can include this effect by modeling the selection Table 1 presents the items representing the latent
mechanism directly and allowing the error terms to variables of knowledge transfer, replication, and
correlate between the selection criteria and the out adaptation, as well as the results of the measure
come (Greene, 2003; Maddala, 1983; Muthen and ment model using the responses referring to the
first year of the relationship (the reliability of the
Joreskog, 1983). This is analogous to maximum
likelihood estimation of selection in simultaneous constructs is even higher for the 2001 data). Each
construct is represented by three or four survey
equations (Greene, 2003). Conveniently, Hypothe
items, which span different aspects of the con
ses 2 and 3, along with a number of controls for
struct. Each of the items loads significantly on
country attractiveness and firm size, directly model
the construct, and the reliability indicator (Cron
firms' propensity for replication and adaptation.
bach's alpha) is well above the standard 0.7 cut-off
Identification of the parameters is a serious issue
for all three constructs. In addition, the measure
in structural equation models (Bollen, 1989; Gold
ment model demonstrates very close fit with the
berger, 1991; Greene, 2003), since not all specified
observed data and a non-significant x2 test- Thus,
models may be estimated in the population. The
confirmatory factor analysis suggests that the mea
structural model presented in this paper is identi
surement model demonstrates high reliability and
fied by Bollen's two-step rule (Bollen, 1989: 328): a close fit with the observed data.
the measurement model is identified by the three The additional variables are described below.
indicator rule and the structural model is identified
by the rank and order conditions for non-recursive
models. Discrete knowledge
The models were estimated using full-infor Three measures represent the extent of links bet
mation maximum likelihood in AMOS 5.0, which
ween knowledge in the receiving firm and out
uses all available data to estimate the model. This
side actors, institutions, and infrastructure?that
approach provides consistent and efficient estima is, between the firm and the outside environment.
tors if the data are missing completely at random The variable cell, which was coded from survey
and is consistent if the data are missing at random responses, indicates whether the firm operates only
(Arbuckle, 1996; Arbuckle and Wothke, 1999). in the cellular telecommunications market. Cellu
There are 40 missing values out of 2400 data points lar firms have fewer links with the environment
in this set. The missing values are clustered in the because regulators do not impose universal service
performance measure (nine missing values) and the obligations and because the infrastructure does not
subscriber data (seven missing for receiver, four require installation in every neighborhood. The
missing for source). variable new indicates whether the receiving firm
I conducted all significance and hypothesis tests was founded at the same time as the transfer
using likelihood ratio tests, which are scale inde relationship, which was coded from surveys. His
pendent (Gonzalez and Griffin, 2001; Greene, torical routines will acquire connections to the
2003). This is a significant issue for SEM, since environment as they face new contingencies over
all latent variables are scaled, but defaults in most time, which makes them less discrete. The vari
statistical packages rely on Wald tests, which may able template indicates whether the source firm
DOI: 10.1002/smj
Table 1. Measurement of latent constructs
Latent variable Survey items (item number, refers to initial year of relationship) Estimates S.E.
(Cronbach's a)
Knowledge transfer (a = 0.89)

Our partner provided valuable information for our business (pbl) 1.07*" 0.15
We learned a lot from our partner (pb2) 1.40*" 0.19
Our partner provided examples of new ways to manage our business (pb3) 1.00
Replication (a = 0.89)
We tried to manage our business exactly like our partner (pel) 0.77**' 0.13
We tried to implement practices from our partner exactly as they existed (pc5) 1.11** 0.12
We tried to copy practices from our partner down to smallest detail (pc6) 1.00
We spent substantial time making sure practices we adopted from our partner worked 0.90**' 0.12
just as they did there (pc7)
Adaptation (a = 0.84)
We usually modified practices from our partner when we 1.00
implemented them in our business (pel)
We usually combined ideas from our partner with other ideas when we adopted them (pe4) 0.88*** 0.13
We spent substantial time modifying practices from our partner to 0.64*** 0.15
make them work in our business (pe5)
We carefully selected practices from our partner to adopt in our business (pe6) 0.96*** 0.15
Model fit
X2 (d.f.) 53.22 (41)
IFI (delta 2) 0.97
CFI 0.97
Significant at * 0.10 level; ** 0.05 level; *** 0.01 level (by one-tailed likelihood ratio test).
had operations in its home country and was drawn variable represents understanding built through
from company reports. Firms without home oper experience in this relationship.
ations are more likely to transfer knowledge from
multiple source locations, introducing connections
with multiple source environments. As a result, Performance
firms with a home template possess more discrete
The performance of the receiving firm is measured
knowledge.
as improvements in efficiency in the year after the
knowledge transfer relationship was founded. Effi
ciency is measured as the number of subscribers
Understanding
per employee. This figure is calculated for the end
of the first year of the relationship and the end of
Three measures represent the degree to which
the second year of the relationship. The variable
the transfer partners build understanding of their
edeltH is defined as the percentage change in this
knowledge through experience. The variable scou
measure between the first year and the second year
ntries represents the number of countries where
of the relationship.
the source firm provided telecom services in 2001
(from company reports). Source units build under
standing of connections through repeated trans Controls
fer. The variable srelat represents the number of
source-firm relationships within the same country A number of controls for country attractiveness
as the receiving firm (from company reports). This and firm characteristics are included in the model
variable represents source-unit understanding of to account for the relative propensity of firms to
the specific country setting built through multiple invest in replication and adaptation. Year repre
relationships. The variable time indicates the num sents the year the relationship was founded, which
ber of years the knowledge transfer relationship was coded from survey responses. Lnrecsubs rep
had existed in 2001 (from survey responses). This resents the natural logarithm of receiving firm
DOI: 10.1002/smj
876 C. Williams
subscribers in 2001, which was coded from sur the target country in 2001, while telpenJ is the
vey responses and company reports. Lnsrcsubs telephone penetration in the first year of the rela
represents the natural logarithm of source-firm tionship. The telephone penetration data were gath
ered from the International Telecom Union. Polcon
subscribers (in all countries, equity adjusted) in
2001, which was coded from company reports. is a measure of political risk for the receiving coun
LngniOl is the log of gross national income per try in 2001, while polcon J is the same measure of
political risk in the initial year (Henisz, 2002). Cur
capita in 2001, while IngdpJ is the log of GDP
rown is the source firm's ownership stake in the
per capita for the first year of the relationship.
These variables were coded from the World Bank receiving firm in 2001, while firstown is the source
firm's ownership at the beginning of the relation
Development Indicators. LnpopOl is the log of the
ship. These were coded from survey responses.
national population of the receiving firm's country
in 2001, while InpopJ is the log of the population
in the first year of the relationship. The popula RESULTS AND DISCUSSION
tion variables were coded from the World Bank's
(2002) World Development Report. TelpenOl is the Tables 2 and 3 present the results of the analysis.
number of telephone lines per 100 inhabitants in The models build from a regression with latent
Table 2. Levels of replication and adaptation (2001)
Hypothesis Model 1 Model 2 Model 3

(predicted sign) Contingencies Contingencies Contingencies plus
plus time time, source, and size
currown -> R 0.32** 0.32*' 0.34***

IngniOl -> R -0.42* -0.42* -0.28
InpopOl ? R 0.05 0.05 -0.09
telpenOl -> R -0.17 -0.18 -0.35
polcon -> R 0.22 0.23 0.22
cell -> R H2(+) 0.19 0.19 0.17
new ? R H2(+) -0.14 -0.14 -0.07
template ?> R H2(+) 0.18 0.18 -0.10
i/me ?> R -0.02 -0.09
scountries -> R -0.10
sre/?tf -> R -0.14
Insrcsubs ? R 0.48
Inrecsubs -> R 0.06
currown ? A 0.37**: 0.39*** 0.35***
IngniOl -> A -0.10 -0.16 -0.19
InpopOl ? A 0.12 0.16 0.01
telpenOl ? A 0.05 0.20 0.19
polcon ? A 0.31* 0.25 0.20
ce// ?> A -0.47**' -0.40*** -0.59***
new ? A 0.37**: 0.23 0.44***
template ?> A 0.10 0.10 0.01
iime -> A H3(+) 0.31** 0.27**
scountries ?> A H3(+) 0.02
sre/?tf -> A H3(+) -0.08
Insrcsubs -> A -0.06
Inrecsubs ?> A 0.41***
X2 (d.f.) 155.69 (108) 149.94 (106) 132.74 (98)

IFI (delta 2) 0.93 0.94 0.95
cn 0.92 0.92 0.94
n = 60; standardized coefficients; significance by one-tailed likelihood ratio tests; 2001 data; A represents latent construct
for adaptation; R represents latent construct for replication; relationships between indicators and latent variables and
between independent variables (IVs) are suppressed for brevity; models are nested with all IVs included in each model.
Significant at * 0.10 level; **0.05 level; ***0.01 level.
DOI: 10.1002/smj
dependent variables (replication and adaptation) Model 2 with data from the first year of the
to a full structural model in which adaptation knowledge transfer relationship (year substitutes
and replication lead to knowledge transfer, which for time, but the two variables have a perfect,
in turn leads to performance improvements. The inverse correlation). Model 5 adds the effect of
models are nested by including all variables in each adaptation and replication on knowledge transfer.
estimated model, but gradually adding the rela Model 6 adds the effect of knowledge transfer on
tionships between the variables. Table 2 presents performance improvements. Model 7 incorporates
Models 1-3 using 2001 data to test Hypotheses 2 potential selection bias arising when firms invest
and 3. Table 3 presents Models 4-7 using data more in transfer mechanisms based on expectations
from the beginning of each knowledge transfer of performance improvements. Finally, Table 4
relationship to test Hypotheses 4 and 5. Model 1 recreates the components of Models 3 and 6
examines the relationship between adaptation and using ordinary least squares. Overall, the results
replication and eight exogenous variables. Model are consistent with a model of replication and
2 adds the exogenous variable time to Model adaptation as separate constructs that both con
1. Model 3 adds several source and receiving tribute to knowledge transfer, which leads to per
firm variables to Model 2. Model 4 replicates formance improvements. The results are mixed
Table 3. Structural model of knowledge transfer and performance (first year)
Hypothesis Model 4 Model 5 Model 6 Model 7

(predicted sign) Adaptation and Adaptation, Transfer and Full model
replication replication performance w/selection
contingencies and transfer
firstown - R 0.09 0.08 0.08 0.09

IngdpA - R 0.01 0.00 0.00 0.00
InpopA - R -0.12 -0.12 -0.12 -0.12
telpenA - R -0.28 -0.26 -0.26 -0.27
polcon A R -0.05 -0.05 -0.05 -0.05
cell -> R -0.04 -0.03 -0.03 -0.03
new ?>? R 0.14 0.14 0.14 0.14
template ? R 0.16 0.16 0.16 0.16
fir sty ear ? R -0.21 -0.21 -0.21 -0.21
firstown ? A 0.17 0.16 0.16 0.14
IngdpJ -> A -0.27 -0.28 -0.28 -0.29
InpopA ? A 0.27 0.27 0.27 0.30*
telpenA ?> A 0.45* 0.46* 0.46* 0.46*
polcon_i -> A 0.04 0.05 0.05 0.08
ce// ?> A -0.44*' -0.42*' -0.42** -0.36*
new ?> A 0.41*' 0.40*' 0.40** 0.43**
template ? A -0.03 -0.02 -0.02 -0.02
fir sty ear -+ A -0.00 -0.00 -0.00 0.04
R^ KT H4a (+) 0.40*' 0.40**' 0.40***
A^ KT H4b (+) 0.40*' 0.40**' 0.41***
KT -> ??te/i72 H5(+) 0.25* 0.34*
errorR errorp 0.02
error A errorp -0.23
X2 (d.f.) 225.82 (145) 204.23 (143) 201.26 (142) 199.75 (140)
IFI (delta 2) 0.88 0.91 0.91 0.91
CFI 0.86 0.89 0.90 0.90
n = 60; standardized coefficients; significance by one-tailed likelihood ratio test; data from initial year of relationship; A represents
latent construct for adaptation; R represents latent construct for replication; KT represents latent construct for knowledge transfer;
errorR?errorp and errorA?errorp represent the correlation of the error terms for replication and adaptation (respectively) with the
error term for edelt!2\ relationships between indicators and latent variables and between independent variables suppressed for brevity;
models are nested with all IVs included for each model.
Significant at * 0.10 level; ** 0.05 level; *** 0.01 level.
DOI: 10.1002/smj
878 C. Williams
Table 4. OLS regression of components of the structural models (Models 6 and 3 from Tables 2 and 3)
Initial (1) (2) (3) (4) 2001: (5) (6) (7)

year: DV: DV: adapt DV: Ktx DV: DV: DV: adapt DV: Ktx
replicate edeltl2 replicate
firstown -0.001 -0.013 curr_own 0.098** 0.083***

(0.057) (0.036) (0.052) (0.032)
IngdpA 1.243 -0.492 IngniOl -0.784 -0.290
(2.108) (1.337) (2.198) (1.362)
InpopA -1.339 0.543 InpopOl -0.180 0.339
(1.612) (0.731) (1.058) (0.656)
telpenA -0.174* 0.080 telpenOl -0.141 0.047
(0.113) (0.071) (0.115) (0.071)
polcon J -4.296 0.198 polcon 7.657 3.553
(6.883) (4.357) (6.944) (4.303)
cell 1.040 -2.404 cell 4.470* -5.810***
(3.244) (2.042) (3.281) (2.033)
new 1.610 2.769* new -3.415 3.871**
(2.999) (1.890) (3.203) (1.985)
template 1.177 -2.841 template -0.593 -1.259
(3.908) (2.477) (4.995) (3.095)
firstyear -0.465 -0.094 time -0.103 0.465**
(0.389) (0.245) (0.371) (0.230)
scountries -0.036 0.006
(0.135) (0.083)
srelat -0.838 -0.369
(0.770) (0.477)
Insrcsubs 2.456** -0.389
(1.404) (0.870)
Inrecsubs 0.137 1.065***
(0.651) (0.404)
replicated 0.165**' replicate 0.248***
(0.047) (0.49)
adapt A 0.221**' adapt 0.209***
(0.070) (0.062)
ktx A 0.078*
(0.054)
Intercept 945.673 209.668 3.336** -0.158 Intercept -19.528 9.237 1.361
(486.403) (1.489)
Observations
(772.880)
56 55 58 (0.621)
51 Observations
(25.145)
47
(15.581)
47
(1.457)
60
R2 0.16 0.13 0.35 0.04 R2 0.37 0.50 0.44
Data from 2001 survey; latent variables represented by sum of observed indicators; significance by one-tailed Wald tests.
Significant at * 10%; ** 5%; *** 1%.
for the contingencies affecting replication and at odds with the data (x2(42) = 67.4, p = 0.008).
adaptation. In addition, when the indicators are loaded on a
The results of confirmatory factor analysis are single factor, all the item loadings are positive,
consistent with Hypothesis 1. When the indicators while one would expect one set of loadings to
of adaptation and replication are constrained to a be negative in the one-factor case. Finally, in the
single factor in the measurement model, the model two-factor model adaptation and replication have
fit declines sharply over a model with two factors a positive, albeit non-significant, correlation. Thus,
that correlate freely. The difference between the the data are consistent with the proposition that
two models is highly significant, with a x2 (d.f. = adaptation and replication are separate constructs.
1) of 14.2. As described above, the baseline mea The proposition that replication increases with
surement model cannot be rejected by the data, discrete knowledge (Hypothesis 2) is not sup
while the model with a single factor is significantly ported, since the variables representing discrete
DOI: 10.1002/smj
knowledge (new, template, and cell) do not have a in this study if replication is largely dependent
significant relationship with replication. The vari on the policies of the source unit rather than the
able new, which indicates whether the receiving receiving unit. Secondary analysis using receiving
unit was founded at the same time as the transfer firm reports of source-firm investments in replica
relationship, does not significantly affect replica tion, however, found that none of these measures
tion. In addition, its effect varies between positive of discrete knowledge led to significant variation
(first year, Table 3) and negative (2001, Table 2), in source-firm replication.
so while it is possible that the effect is too small to The results are mixed with regard to the impact
be distinguished from a non-effect with this sam of understanding on adaptation (Hypothesis 3). In
ple, it does not appear to have a consistent impact Models 2 and 3, adaptation increases significantly
on replication. In contrast to its lack of effect as the length of the relationship rises (time). Thus,
on replication, the variable new is also associated understanding gained through experience in the
with significantly higher levels of adaptation in relationship does increase the use of adaptation by
Tables 2 and 3.
the receiving partner. Relationship-specific experi
Similarly, the existence of a clear home template ence appears to be necessary for building under
(template) does not significantly increase replica standing, since the source firm's prior relationships
tion among these firms. The relationship between (scountries) and country experience (srelat) do not
template and replication is not significant in any have a significant impact on adaptation in Model
of the models, though it is generally positive and 3. The lack of a relationship between source
could be small but significant with a larger sam firm experience and adaptation could arise because
ple. Finally, the variable cell does not lead to source firms that operate in many countries must
significantly higher replication in any of the mod impose uniform practices to reduce coordination
els, except in one instance in the OLS regression costs. One might also conclude that these variables
(Table 4, column 5) where it is weakly signifi affect source adaptation without changing receiver
cant. In addition, the effect changes from negative adaptation, but secondary analysis using receiving
(first year, Table 3) to positive (2001, Table 2). firm reports of source adaptation found that neither
The impact of cell on adaptation, however, is neg of these measures led to significant variation in
ative and significant in all models. While these source-firm adaptation.
firms do not significantly increase replication of The results are consistent with Hypotheses
more discrete knowledge in cellular markets, they 4a and 4b. Replication and adaptation each
do reduce adaptation in this setting. have a positive and highly significant impact
The prediction that replication would rise as on knowledge transfer (see Table 3, Models
knowledge became more discrete relied on a cost 5-7). Firms that engage in higher levels of
mechanism, in which firms would turn to repli replication and adaptation also report greater
cation as the cost of integrating knowledge fell. transfer of knowledge from the source partner.
Instead, these results suggest that firms choose a These relationships are robust to all specifications
level of replication based on the reward for the of the model and are significant at the .01 level.
transfer of ambiguous knowledge, and then choose The size of the effect is about the same for each
a level of adaptation to achieve integration con mechanism?a one standard deviation increase in
tingent upon the interdependence with the con adaptation or replication leads to a 0.4 standard
text. The lack of a relationship between indicators deviation increase in knowledge transfer.
of discrete knowledge and replication might also Finally, knowledge transfer has a positive,
arise if firms make knowledge more discrete by significant impact on efficiency improvements
accounting for contingencies within the knowledge (Hypothesis 5) in Models 6 and 7. This result
itself.1 Alternatively, the relationship between dis
crete knowledge and replication might be masked
then we might see fairly uniform replication across different rela
tionships with context. In this case, however, I would not expect
1 For instance, if a manufacturing process depends on tempera to see the reduction of adaptation in the face of more discrete
ture or humidity, the knowledge may include operating routines knowledge that is observed in all specifications. In addition, I
that are contingent upon these context variables. I am grateful to would expect firms to develop this contingent knowledge (which
a referee for pointing out this potentially endogenous aspect of can be more easily replicated) through the experience of repeated
discrete knowledge. However, if firms are generally successful in transfer, but there is not a significant relationship between past
incorporating contingencies into the knowledge to be transferred, transfer (scountries) and replication.
DOI: 10.1002/smj
880 C Williams
is somewhat larger when selection bias is On the other hand, replication has a gener
accounted for by allowing errors to correlate ally negative relationship with measures of coun
between investments in mechanisms and impact try attractiveness. GNI per capita weakly reduces
on performance in Model 7. Since firms are likely replication in the 2001 data. When telpen and pol
to invest more when they anticipate performance con, which are correlated with GNI per capita (0.80
pay-offs, this suggests that a selection bias may and 0.64, respectively), are not included in the
exist in the data, though the correlation between regression then the negative effect of IngniOl is
errors is not significant.2 strongly significant. This is consistent with a pull
Of the controls included in the study, the owner perspective on knowledge?whereby knowledge
ship variable has the strongest effect on replication is pulled into less developed countries because
local firms have more to learn. In the second time
and adaptation, though only in the 2001 data. Own
ership of the source firm has a positive and highly period (2001, Table 2), source size and political
significant impact on replication and adaptation in risk have a positive, non-significant effect on repli
2001. Firms appear to invest more in both replica cation.3 Finally, country size (Inpop) and political
tion and adaptation when the source owns more of risk (polcon) have a negative and non-significant
the receiving firm. This is consistent with gover effect on replication in the first time period but
nance theories that predict knowledge transfer haz a generally positive and non-significant effect in
2001.
ards are reduced by governance under a single firm
While model fit is not the focus of the anal
(Oxley, 1999). In the data from the initial year of
ysis in this paper, a number of fit indicators are
the relationship, however, source ownership does
included in Tables 2 and 3. Model fit is accept
not have a significant effect on adaptation or repli
able using 2001 data in Table 2. Models 1-3 all
cation. This might arise if source firms believed
have fit indices (incremental fit index (IFI or delta
that investments in knowledge transfer will allow
2) and comparative fit index (CFI)) between 0.90
them to acquire higher levels of ownership in the
and 0.95. Using data from the first year of the rela
future despite the risk of opportunism.
tionship the fit is less good, but above 0.90 for the
Country attractiveness has surprisingly little
full model (Models 5-7 in Table 3).
effect on adaptation. Telephone penetration is
associated with more adaptation in the initial
year of the relationship (telpen in Table 3, Robustness
Models 4-7), though not in the 2001 data
I analyzed the data under alternate approaches to
(Table 2, Models 1-3). Firms in countries
estimation in order to gauge the robustness of
with higher telephone penetration possess more
the SEM results given the small sample and to
technical understanding, which might increase the
explore alternative measures of performance. The
effectiveness of adaptation. Higher penetration results of these robustness checks are broadly con
is also associated with earlier liberalization, and
sistent with the results of the hypothesized model.
competitive pressures early in the relationship Time in the relationship increases adaptation in
may make adaptation essential for success. all specifications. Measures of discrete knowledge
Larger receiving firms adapt significantly more increase adaptation in most specifications. Replica
(Inrecsubs), possibly because they have more links tion and adaptation contribute to knowledge trans
to the environment than smaller firms. Country size fer in all specifications, while knowledge transfer
(Inpop) and political risk (polcon) have a positive, leads to performance improvements for measures
non-significant effect on adaptation in both time that emphasize short-term improvement, that is, in
periods, while GNI per capita (IngniOl in Table 2) the year following transfer. For measures of per
and GDP per capita (IngdpA in Table 3) have formance improvement over the length of the rela
negative but non-significant effects. tionship, I find knowledge transfer has no effect.
So, while the results are broadly consistent with the
hypotheses in the paper, some interesting nuances
2 In an alternative specification, I modeled selection at the level emerge from these analyses.
of knowledge transfer, which led to results consistent with those
in Model 7, though the selection effect was not significant in
this specification. In this specification, the relationship between 3 Except in the OLS analysis, where Insrcsubs has a significant,
knowledge transfer and performance becomes non-significant. positive effect on replication.
DOI: 10.1002/smj
In particular, OLS analysis is less sensitive to include efficiency and growth measures from pub
the small-sample estimation issues that might arise lic reports as well as self-reported improvements in
in the SEM method used as the primary analy five categories: service quality, efficiency, growth,
sis because OLS approaches asymptotic behavior innovation, and profitability. The lack of signif
more quickly than maximum likelihood estima icance for long-term performance improvements
tion (used in SEM). Table 4 shows the results of suggests that the value of knowledge depreciates
OLS analysis of each component (equation) within quickly (Argote, 1999; Darr et al, 1995). Knowl
the full SEM model for Model 3 and Model 6 edge has an impact one year after transfer, but
(Tables 2 and 3). In these models, the latent vari the effect does not appear to lead to compounding
ables are treated as composite indexes?averages improvements over the following years.
of indicator variables from the survey?so the
models do not account for measurement error in
Summary
estimating the variables of replication, adaptation,
and knowledge transfer. The results of these indi Overall, the results are consistent with the theo
vidual regressions closely match those reported retical arguments laid out in the paper. Adapta
in the SEM analysis (Tables 2 and 3). The main tion and replication are distinct transfer mecha
difference is in the OLS analysis for the initial nisms that firms use simultaneously when transfer
year, where the variables telpenA, cell, and new ring knowledge. Both contribute to transfer, which
have weak or no significant effect on adaptation, leads to improved efficiency. Adaptation rises with
which could arise because measurement error is
time, suggesting that time within the relationship is
a larger concern in data requiring significant his an important source of understanding. Replication
torical recall on the part of survey respondents.4 does not rise when knowledge is more discrete,
Together, the OLS regressions suggest that the though adaptation does fall. Finally, the effect of
SEM results presented in the tables do not arise governance on knowledge transfer changes over
because of errors in estimation from the maximum
the course of the relationship. The source firm's
likelihood estimation technique. ownership of the receiving firm does not affect
In the final robustness analysis, I explored alter transfer in the first year of the relationship, though
native measures of performance. The relation later both replication and adaptation rise when
ship between knowledge transfer and performance source ownership is higher.
remains significant for alternative short-term per
formance measures, but not for long-term mea
sures. Growth in subscriber base is a very impor CONCLUSION
tant metric for telecom firms, especially in emerg
ing markets where they face a pent-up demand This paper contributes to our theory of orga
from consumers. When Model 7 uses growth in nizational knowledge by bringing together two
subscribers from year 1 to year 2 in place of the streams of research that have emphasized differ
efficiency performance measure, the relationship ent aspects of knowledge: context dependence and
between knowledge transfer and performance is causal ambiguity. The implicit assumption of these
positive and significant at the 0.05 level. When I streams has been that replication and adaptation
substitute measures of performance improvement are mutually exclusive approaches to transfer. In
over the course of the relationship (first year until this study, I find that the actions of firms trans
2001), however, there is not a significant rela ferring knowledge are consistent with the theory
tionship between knowledge transfer and perfor that knowledge is simultaneously context depen
mance improvements. These long-term measures dent and causally ambiguous. Firms replicate more
when organizational knowledge is ambiguous and
4 Table 4 shows the following minor discrepancies with the must be copied exactly, and firms adapt more
SEM analysis. In two cases, variables (telpenJ and cell) in the
when organizational knowledge depends on con
initial year lose a significant impact on adaptation, presumably
because of higher measurement error with longer recall. In three text and must be modified for the new setting. I
cases, variables (firstown, cell, template) that are not significantly define replication and adaptation as effort towards
different from zero change sign in the OLS results. Finally,
three non-significant relationships between replication and other
a goal, which opens the possibility that firms repli
variables (telpenJ, cell, Insrcsubs) in the SEM analysis gain cate and adapt simultaneously, and I find that they
weak significance in the OLS. frequently do. In sum, I find that replication and
DOI: 10.1002/smj
882 C. Williams
adaptation vary separately, are used jointly, and Finally, some of the design choices in the study
contribute significantly to knowledge transfer. limit the extent to which we can distinguish the
This study is just a first step in developing these underlying mechanisms and compare alternative
insights about knowledge. Any conclusions must approaches to knowledge transfer. While I predict
be tempered by our understanding of the study's and find that firms are engaged in both replication
limits. The study's small sample size raises the and adaptation at the level of the relationship, this
question of whether the findings would be repli study cannot rule out the possibility that at some
cated in another study. Since SEM makes heavy sufficiently fine-grained level of individual prac
demands of small samples, it is possible that the tice firms only employ one or the other transfer
relationships between variables in this study might mechanism. The logic developed here suggests that
not represent the true asymptotic values of the even at a fine-grained level organizational knowl
parameters. Yet, the robustness analysis with OLS, edge is less like discrete atoms than like gears in a
which requires fewer observations, finds a simi machine, which must be adapted to work together
lar pattern of relationships, mitigating this concern. in a new setting. Future research will be needed,
Similarly, response bias could skew the results if however, to explore how firms balance, combine,
this is an unrepresentative sample. Comparing our and trade off between replication and adaptation at
sample to the population of relationships, however, the level of an individual practice. Similarly, the
uncovers only small differences between them. measurement of replication and adaptation as sus
Finally, the power of a test depends on the size of tained practices over a year of the transfer relation
the effect, so this study has less power to determine ship means that this study cannot distinguish fine
the significance of small effects than large ones. It grained differences in the timing of the two activ
is likely that another study with a larger sample ities. Szulanski's (2003) stage model of transfer
would find some of the non-significant parameters includes four stages of transfer?initiation, imple
to be significant, though likely smaller than the mentation, ramp-up, and integration?and Szulan
relationships discussed in this paper. ski and Winter (2002) recommend that adapta
In addition, the empirical setting raises two tion should only occur after implementation. I find
questions about how the results would generalize that, in the first year of the relationship, greater
to other firms. As service firms, the companies in investment in adaptation leads to more knowledge
the study interact with the environment throughout transfer. While the theory developed here suggests
their operations. In contrast, manufacturing firms that interdependent knowledge could not be imple
can buffer some of their operations within facil mented without some adaptation, this study cannot
ities where there is less interaction with outside rule out the possibility that the adaptation reported
actors (Thompson, 1967). With careful buffering, in the first year occurs entirely after the actual
the need for adaptation may be less than in ser implementation of new practices.
vice firms, and the effectiveness of replication may Despite these limitations, a number of interest
be increased. Service firms represent an interesting ing results emerge from the study that can help us
setting to study knowledge that is embodied in the develop our theory of organizational knowledge
routines of the firm, but it remains an empirical and suggest new paths for research. Adaptation
question if these results will generalize to settings emerges from the study as the primary tool by
where firms can buffer more of their core oper which firms calibrate transfer to a particular sit
ations. Another limitation is that the study uses uation. When firms transfer more discrete knowl
reports from the receiving partner to represent the edge to units in cellular markets, they adapt less
policies of the transfer relationship. The relation because there are fewer connections. Firms adapt
ship includes a source partner as well, and it is more when they possess understanding of the con
possible that the source's actions differ signifi nections between knowledge and its environment.
cantly from those of the receiver. Responses from The finding that firms are quite active in managing
receiving units suggest that source firm replication the levels of adaptation to fit these contingencies
is correlated almost 0.9 with the receiving unit and suggests that adaptation deserves a central role in
source firm adaptation is correlated nearly 0.6 with our theories of organizational knowledge.
receiving unit adaptation, but this study still lacks That we have much more to learn about adapta
the source unit's perspective and the impact of its tion is suggested by two surprising results: when
actions on transfer and performance. receiving units are newly founded they increase
DOI: 10.1002/smj
adaptation, but when source firms have prior expe the process of adaptation generate knowledge that
rience with transfer they do not shift adaptation. firms carry back to their core operations and what
Understanding is not something typically associ firm policies will facilitate this knowledge creation
ated with organizations, but I propose that orga and transfer?
nizations possess understanding when they can In addition, the appropriate governance of
predict how a given practice will interact with knowledge transfer appears to vary over the rela
other elements of the organization and environ tionship. Consistent with the transaction-cost per
ment. One respondent from a source firm pointed spective on the mitigation of hazards in knowl
to the management of interacting practices as an edge transfer (Oxley, 1999), source ownership has
essential aspect of transfer: 'Some organizations a sizeable and significant impact on replication
are very good at [managing] the individual slices, and adaptation later in the relationship. In the first
but they do not have a good idea of how those year, however, ownership does not have a signif
slices fit together ... [my company] used to move icant effect on investments in knowledge transfer.
people around and we developed a good feel for These results suggest that governance models for
how the pieces fit together, which is not typical knowledge transfer need to incorporate the stage
elsewhere.' Understanding interconnected knowl of relationship as a contingent factor shaping the
edge might well contribute to recombination and choice of ownership.
innovation within firms (Teece, Pisano, and Shuen, While replication is clearly an important mech
1997; Winter, 2003). A valuable future project anism for knowledge transfer, this study suggests
would be to study how firms build understanding more about which factors do not lead to higher
and where it must reside to be useful for knowl replication than those which do. Variables repre
edge transfer or recombination. senting prior experience of the source firm and
I find that knowledge transfer has a measurable the level of discreteness of the knowledge have
impact on firm performance the year after knowl negligible impacts on replication. Some other fac
edge is transferred. This effect is weak, but robust tors, such as country development and source size,
across several measures of performance, and it have large but non-significant effects, which might
is larger when the model accounts for selection be significant in larger samples. Still, most of the
bias. In addition, the results are likely to under contingencies for replication explored in this study
state the impact of transfer on performance since appear to have less influence on replication than
they do not include simultaneous transfer and per on adaptation. Perhaps replication is calibrated to
formance improvements in the same year. How the level of ambiguity in the targeted knowledge.
ever, the knowledge acquired by the receiving In that case, future studies at the level of individ
unit appears to depreciate rapidly, since there is ual practices might explain the contingent nature
no long-term performance impact from knowledge of replication by measuring the relative causal
transfer in the first year of the relationship. The ambiguity of different practices.5 Some of the con
finding that knowledge in organizations depreci tingencies that drive replication may also be the
ates at a fairly rapid pace also dovetails with other factors that lead firms to enter knowledge transfer
recent research (Argote, 1999; Darr et al., 1995). relationships. In this case, the decision to replicate
Our evolutionary theories of the firm need to be may be tightly bound with the decision to trans
extended to explain the friction that keeps success fer knowledge, after which adaptation is calibrated
ful early transfer from leading to later transfer, and to the context. This suggests an alternative, mul
empirical studies need to examine the mechanisms tilevel model, in which adaptation increases the
that lead to rapid depreciation of knowledge. effectiveness of replication when knowledge is less
From an evolutionary perspective, the picture of discrete or understanding is greater. Such a model
multinational firms that emerges from the study is is beyond the possibilities of these data, but might
of large firms evolving and changing as they move
to new contexts rather than replicating uniform
5 One would expect, however, that practices would become less
practices around the world. Thus, one mechanism ambiguous as firms gained experience transferring them to new
of change for firms might be to expand to new locations, but prior transfer has a negligible effect on replication
in the study. Still, this could arise if causal ambiguity is so
environments where existing routines will require
deep that it is not reduced through repeated transfer, which is
adaptation. From an evolutionary perspective, the the implicit assumption behind Intel's 'Copy Exactly!' policy
study suggests an important question: when will (McDonald, 1998).
DOI: 10.1002/smj
884 C. Williams
be fruitful for future exploration. In the final analy Behavior and Human Decision Processes 82(1):
150-169.
sis, there remains unexplained variance in replica
Baum JAC, Ingram P. 1998. Survival-enhancing learning
tion, which awaits explanation as we develop our in the Manhattan hotel industry, 1898-1980.
understanding of organizational knowledge. Management Science 44(7): 996-1016.
When firms set out to transfer knowledge, the Bollen KA. 1989. Structural Equations with Latent
theory and empirical analysis in this paper sug Variables. Wiley: New York.
Conover WJ. 1971. Practical Nonparametric Statistics.
gest they must assess the nature of the knowl
Wiley: New York.
edge to be transferred. Firms need to differenti
Cyert RM, March JG. 1992. A Behavioral Theory of the
ate between ambiguous knowledge, which must Firm (2nd edn). Blackwell Business: Cambridge, MA.
be copied exactly, and knowledge intertwined Darr ED, Argote L, Epple D. 1995. The acquisition,
with the environment, which requires modification. transfer, and depreciation of knowledge in service
While exact copying of knowledge is difficult in organizations: productivity in franchises. Management
Science 41(11): 1750-1762.
new settings, effort invested in copying will signif Dillman DA. 2000. Mail and Internet Surveys: The
icantly increase the transfer of valuable knowledge Tailored Design Method (2nd edn). Wiley: New York.
even if perfect copying is not achieved. In addition, Goldberger AS. 1991. A Course in Econometrics.
the interdependent nature of knowledge means that Harvard University Press: Cambridge, MA.
firms will need to identify the interfaces between Gonzalez R, Griffin D. 2001. Testing parameters in
structural equation modeling: every 'one' matters.
ambiguous knowledge and the context to target Psychological Methods 6(3): 258-269.
them for adaptation. Understanding the connec Grant RM. 1996. Prospering in dynamically-competitive
tions between knowledge, the organization, and its environments: organizational capability as knowledge
environment takes substantial experience in a spe integration. Organization Science 7(4): 375-387.
cific relationship, but it can improve knowledge Greene WH. 2003. Econometric Analysis (5th edn).
Prentice-Hall: Upper Saddle River, NJ.
transfer and may create capabilities that are useful
Gupta AK, Govindarajan V. 2000. Knowledge flows
in other parts of the firm. within multinational corporations. Strategic Manage
ment Journal 21(4): 473-496.
Heckman JJ. 1979. Sample selection bias as a
ACKNOWLEDGEMENTS specification error. Econometrica 47(1): 153-161.
Henderson R, Mitchell W. 1997. The interactions of
organizational and competitive influences on strategy
This research was supported by grants from the and performance. Strategic Management Journal,
University of Michigan's Interdepartmental Con Summer Special Issue 18: 5-14.
Henisz WJ. 2002. The institutional environment for
sortium on Organization Studies and Center for
International Business Education. infrastructure investment. Industrial and Corporate
Change 11(2): 355-389.
Hu L-t, Bentier PM. 1999. Cutoff criteria for fit indexes
in covariance structure analysis: conventional criteria
REFERENCES versus new alternatives. Structural Equation Modeling
6(1): 1-55.
Ahuja G. 2000. Collaboration networks, structural holes, Kogut B, Zander U. 1992. Knowledge of the firm,
and innovation: a longitudinal study. Administrative combinative capabilities, and the replication of
Sciences Quarterly 45(3): 425-455. technology. Organization Science 3(3): 383-397.
Allen TJ. 1977. Managing the Flow of Technology: Tech Kostova T, Roth K. 2002. Adoption of an organizational
nology Transfer and the Dissemination of Technolog practice by subsidiaries of multinational corporations:
ical Information within the R&D Organization. MIT institutional and relational effects. Academy of
Press: Cambridge, MA. Management Journal 45(1): 215-233.
Arbuckle JL. 1996. Full information estimation in Leonard-Barton D. 1988. Implementation as mutual
the presence of incomplete data. In Advanced adaptation of technology and organization. Research
Structural Equation Modeling: Issues and Techniques, Policy 17: 251-267.
Marcoulides GA, Schumacker RE (eds). Erlbaum: Lippman SA, Rumelt RP. 1982. Uncertain imitability:
Mahwah, NJ; 243-277. an analysis of interfirm differences in efficiency
Arbuckle JL, Wothke W. 1999. Amos 4.0 Users Guide. under competition. Bell Journal of Economics 13(2):
SmallWaters: Chicago, IL. 418-438.
Argote L. 1999. Organizational Learning: Creating, Maddala GS. 1983. Limited-Dependent and Qualitative
Retaining, and Transferring Knowledge. Kluwer Variables in Econometrics. Cambridge University
Academic: Boston, MA. Press: New York.
Argote L, Ingram P. 2000. Knowledge transfer: a basis Marsh HW, Hau K-T, Baila JR, Gray son D. 1998. Is
for competitive advantage in firms. Organizational more ever too much? The number of indicators per
DOI: 10.1002/smj
factor in confirmatory factor analysis. Multivariate Szulanski G. 2003. Sticky Knowledge. Sage: Thousand
Behavioral Research 33(2): 181-220. Oaks, CA.
Masten SE. 1993. Transaction costs, mistakes, and Szulanski G, Winter S. 2002. Getting it right the second
performance: assessing the importance of governance. time. Harvard Business Review 80(1): 62-69.
Managerial and Decision Economics 14(2): 119-129. Teece DJ, Pisano G, Shuen A. 1997. Dynamic capabili
McDonald CJ. 1998. The evolution of Intel's Copy ties and strategic management. Strategic Management
Exactly! technology transfer method. Intel Technology Journal 18(7): 509-533.
Journal 4th quarter: 1-6. Thompson JD. 1967. Organizations in Action. McGraw
Muthen B, Joreskog KG. 1983. Selectivity problems in Hill: New York.
quasi-experimental studies. Evaluation Review 7(2): Tushman ML, Anderson P. 1986. Technological discon
139-174. tinuities and organizational environments. Administra
Nelson RR, Winter SG. 1982. An Evolutionary Theory of tive Science Quarterly 31: 439-465.
Economic Change. Belknap Press: Cambridge, MA. Winter SG. 1987. Knowledge and competence as
Nonaka I. 1994. A dynamic theory of organizational strategic assets. In The Competitive Challenge:
knowledge creation. Organization Science 5(1): Strategies for Industrial Innovation and Renewal,
14-37. Teece DJ (ed). Ballinger: Cambridge, MA; 159-184.
Oxley JE. 1999. Governance of International Strategic Winter SG. 1995. Four R's of profitability: rents,
Alliances: Technology and Transaction Costs. Har resources, routines, and replication. In Resource-Based
wood: Amsterdam. and Evolutionary Theories of the Firm: Towards a
Prahalad CK, Doz YL. 1987. The Multinational Mission: Synthesis, Montgomery CA (ed). Kluwer Academic:
Balancing Local Demands and Global Vision. Free Boston, MA; 147-175.
Press: New York. Winter SG. 2003. Understanding dynamic capabilities.
Shaver JM. 1998. Accounting for endogeneity when Strategic Management Journal 24(10): 991-995.
assessing strategy performance: does entry mode Winter SG, Szulanski G. 1998. Replication as strategy.
choice affect FDI survival? Management Science Working paper, The Wharton School.
44(4): 571-585. Winter SG, Szulanski G. 2002. Replication of organi
Spender J-C. 1996. Making knowledge the basis of a zational routines: conceptualizing the exploitation of
dynamic theory of the firm. Strategic Management knowledge assets. In The Strategic Management of
Journal, Winter Special Issue 17: 45-62. Intellectual Capital and Organizational Knowledge:
Szulanski G. 1996. Exploring internal stickiness: imped A Collection of Readings, Bontis N, Choo CW (ed).
iments to the transfer of best practice within the firm. Oxford University Press: New York; 207-221.
Strategic Management Journal, Winter Special Issue World Bank. 2002. World Development Report. Oxford
17: 27-43. University Press: New York.
APPENDIX 1: DESCRIPTIVE STATISTICS

Variable Description Range Mean S.D.
(1) bl Knowledge transfer survey item 2001 (Table 1) 1-5 3.47 1.27
(2) b2 Knowledge transfer survey item 2001 (Table 1) 1-5 3.32 1.38
(3) b3 Knowledge transfer survey item 2001 (Table 1) 1-5 3.38 1.14
(4) cl Replication survey item 2001 (Table 1) 1-5 2.65 1.12
(5) c5 Replication survey item 2001 (Table 1) 1-5 2.38 1.09
(8) el Adaptation survey item 2001 (Table 1) 1-5 3.50 1.14
(9) e4 Adaptation survey item 2001 (Table 1) 1-5 3.90 1.10
(10) e5 Adaptation survey item 2001 (Table 1) 1-5 2.80 1.10
(11) e6 Adaptation survey item 2001 (Table 1) 1-5 343 1.28
(12) pbl Knowledge transfer survey item initial year (Table 1) 1-5 3.88 1.10
(13) pb2 Knowledge transfer survey item initial year (Table 1) 1-5 3.64 1.32
(14) pb3 Knowledge transfer survey item initial year (Table 1) 1-5 3.51 1.25
(15) pel Replication survey item initial year (Table 1) 1-5 3.02 1.14
(16) pc5 Replication survey item initial year (Table 1) 1-5 2.71 1.22
(17) pc6 Replication survey item initial year (Table 1) 1-5 2.25 1.20
(continued overleaf)
DOI: 10.1002/smj
886 C. Williams
APPENDIX 1 (Continued)
Variable Description Range Mean S.D.
(18) pel Replication survey item initial year (Table 1) 1-5 2.47 1.15
(19) pel Adaptation survey item initial year (Table 1) 1-5 3.27 1.06
(20) pe4 Adaptation survey item initial year (Table 1) 1-5 3.47 1.06
(23) edeltl2 Percentage change in subscribers/employee -0.33-6.76 0.70 1.25
from year 1 to year 2
(24) time Years since beginning of relationship 0-12 4.77 3.33
(25) year First year of relationship 1989-2001 1996.23 3.33
(26) new 1 = Receiving firm founded initial year of 0-1 0.47 0.50
relationship
0 = Receiving firm founded before relationship
(27) template 1 = Source firm has home country operation 0-1 0.85 0.36
0 = Source firm has no home country operation
(28) cell 1 = Receiving firm operates only in cellular 0-1 0.52 0.50
markets
0 = Receiving firm operates in fixed service
mkts
(29) curr_own Ownership stake of source firm, 2001 7-100 50.37 23.63
(30) first _own Ownership stake of source firm, initial year 6-100 41.55 22.81
(31) telpenOl Telephones per 100 inhabitants, country 2001 0.20-90.71 23.65 19.45
(32) polcon Political risk 2001 0.00-0.78 0.36 0.22
(33) telpen J Telephones per 100 inhabitants, country initial 0.20-90.71 23.52 19.44
year
(34) polcon J Political risk initial year 0.00-0.78 0.36 0.22
(35) lngdp_i Log(e) gdp per capita, initial year 5.72-10.47 8.05 1.14
(36) Inpop A Log(e) population, country initial year -2.81-5.10 2.55 1.53
(37) InpopOl Log(e) of population, country 2001 -2.81-5.15 2.54 1.57
(38) IngniOl Log(e) of gross national income, 2001 5.56-10.55 8.18 1.10
(39) Insrcsubs Log(e) of source firm subscribers, 2001 10.52-18.78 17.11 1.63
(40) Inrecsubs Log(e) of receiving firm subscribers, 2001 7.94-16.91 13.22 2.10
(41) scountries Count of countries where source firm provides 1-52 17.98 12.88
telecom services
(42) srelat Number of source firm relationships within 1-8 2.38 2.07
same country as receiving firm
Copyright ? 2007 John Wiley & Sons, Ltd. Strat. Mgmt. /., 28: 867-889 (2007)
DOI: 10.1002/smj
Transfer in Context
s*-h" O O O O ? ?
O <-h CO O <N <N
> o r- o r- o o ô
? r- o o (N t-H ^h en
!?O??OO?
oooTtasomr-Hr? so
Ot"-l>00<NOO(N
hoooooooo
Hododododo
s do?d??dddo
r-^???????????
oooHHomÔNinoÔh
Ooor-cNcnm(Nm^t^Hm(Si<N
Hoddddddddddd
^?????????????
II
Hoddddddddddddd
0?ninTtTfmo^^H(NmTj-rtmom
^ddddddddddddddd
ô???????o????o??
r?->nÔTtTtmoôo?nc^(N^-om
II III
OOHHfSM^r<nnininÔOHO(NH
^00000000000000000 I
OtsOOHH(NHHn^hlT)HOHOH(N
Hoddddddddddddddddd
II
III
oc^cxir^^^vocnwoaN?n?sor--r-cnTtr-cnaNcn
Hodddddddddddddddddd
Orhoornoor^^c^cnr--Ti-rtininaN^H,or^cnON?ri
^????????????????????
0(S\?ooMxcnoooo\HO(Srn(Nin^H(N(N(no\
^?????????????????????
O^-mmTfTj-mfn^Hininm?sioo^Hr-Hr-H?siom^H^H
Hoddddddddddddddddddddd
^'00000000000000000000000
OONi^-',^-^t^^-mTt(Ninminm^HO(N'-H'-H(NOm^H^H
î <N rri
-Ci -Ci -Ci
i-HCsimTfrinôr-oocs
Ltd. Strat. Mgmt. J., 28: 867-889 (2007)

DOI: 10.1002/smj
\?OinH^ô\0000(Nirim(NHt^
IIIIIIIII
000000000000000000
(NÔO*-HO(N^H<r>^H<N^^H(N(NCNl^HO
II III I II
??????????????????
mm(NMOH(s)0(SOfÔO(NHOO(S
??????????????????
(N^hOOOOÔ^-h0^hOO*-h^hÔ<-<
Il I I I I I I I
^HOsmm^-(N(NcN(N(N'-Hm^-^H^-o^tO
oooooooooooooooooo
I I I I I I I I II
? ?cT
I III
?????????????????? ?>
<NOO^Hr-H(NO(NO(N^(N^H^H(N^H^HO c?
III I
^t-0(N^-ocNt^mr^mt^Ti-r-HCNas^Doo(N ^
dddddddddddddddddd ?>
t>cn^-H(NvsOÔNi>aNi>oin(Ni-HOOcn'-HON
??????????????????
(N(NOCN(N^^h0^hO'-"00<N^^h^hO
II III
??????????????????
(NOOOcn^-HOcsoc<iooo^HCN(Nimo
I I III I
oooMXJXôoomomHfnN^HOooH
??????????????????
(MO^^-HCslOr-H^-Hr-HT^^H^HÔ^TtO^ 00
III II
??????????????????
Cr)^H^HOC<l^H^HCN|'-<r<IOOOCN^Hr-H^Hl?I
mtmô^MHHHHHmcsO'?MiDH ^^
82
^d
^H so
I
dddddddddddddddddd ^
IIIIII
hOhOhMÔ(SOhhh(S(S(ShO
^j-vooovo^fi-HONvooN'sOoocNinT?ivooor^"3
dddddddddddddddddd ocsno
nodo
II III I I
ooMcshooi-inîn<nooMvoooo\Qm
MH(NHO(SWOnO(SOH(SHH?H^ ^
?????????????????? CS
cn^tinvorôsr-r-oc^ÔinmincNOCN
OOfSmo
^^???
??????????????????
HOMO(NHcnOcnO(SHH(SH(SMO
III I I I I I I
> o o en o Tt
Iooooo
m^HHOtrsooNoo^HOHomMH
IIIIIIII
00<Ncomm<NO(NOOOO'-H^HCNO<N
dddddddddddddddddd 8^>Ocm
O (N
Ôooooo
O oo
^H ?n
^Hvo^H
II lili I
^H0(N^cnm0(D0m0000'-HCMcN'-^ ,-h
dddddddddddddddddd ?},
oaoofSOMXiinooiriO^nHOôoo
r-HO'-H(NmmOÔ^H^^H^H^^H'-H(NO
8iDO(S(SHfth
H?dddddd
<T) 1-H ^H ,-H _i o O
^ ?nooocNoo
III
dddddddddddddddddd
II Hoddddddd
III
nOM^-mooOHOHinHOMnmtvo^
??????????????????
^T-Hr-H(N?nm^-H'-H^H'-H'-H'-HO'-H'-H(NCNO vomôo^H'-Hcnoo
'Hodddddddd
1^ *a ? %>
*?> *fc -g
O S O Cl Cl Cl 2
S 3e S "Sr ^ er ?
? ^"^ -S<^ 00 K
C?h _,
C ft, .K.?K? o S?
S S00
G^î ft, ^ ^-?s^^-^^s^ ?
1=
>r>vorôoo\O^H<NimT;?->r>vor-ooONO*-H<N
Strat. Mgmt. J., 28: 867-889 (2007)

DOI: 10.1002/smj
on o cn on m en
?oooooo
^???????
CN <N un O i-! O ?
OOOOOOO
II
8 On *$
m m oo o o ? ^h
II I II
ooooooo
^0-00000000
enoencNeNino ? o ?
II
8S8 Os Tf? Tt o ? m r-- ô

m m oo o o ? r-H
II I II
???????
8:
III I
en in oo
odd
<N O O? t-h O O ? o (N
OOOOOOO O en
O vo
enOOOOCN
8eNÔ^*0^ ?'-hO<on ono m(NmTt
? ? On
mr
Hodddddddddddd m
o (N(N
8 ON
?OO
O
on in (N in (N g> MOOh On (N m ON
^H O ?? <N O ?
odd odd odd dodo
t?- ? o
III
(NONTI On rf On (N m m en ? o cn
?ON? \D
in CN
(N
oooooooooo
<T> ? ?? ^H eN o o <N en ? o ?
dodo
ON
en
? odd
I O ?S8
(N 00 (N 00 (N 00 ?
cnno? O O ' in Tt
?odd odd? ; d oo o Sen?oo? oo
O^f?
II II ? dodo
odd 288
O 00
odd o? Ooo I oI
o 00 ON 00 ON (N 00
odd HHff) ^3
m 8 O?? cn
r- cn
? ?^ en
????
r
ooo<
. 00 ON CN
tHHff) 288 O 00 ? 00
o o od dO o S
^H O 1-H T-H
II II II
odd odd
CN ^sO ON VO ON ON en <<t O
(N CN CN S2dO o
d
(N
CN 'S
X II
8?
ooooooooooo o o Q
(N ? <N HHff) ^H (N? o
r- (N o <N O <N ^H ON 00 1-H
odd odd O oS3

o o ?o ? o <
(N cn ? ^H ^H CO r- vo <nO Tf PU r- oo on o ? cn
odd oo ^H ^H O^h CN m m en rf Tf Tf
odd 23
823 o vo oo 1-H
odd odd o o o o
II
1-H 1-H O 1-H
?o
r- on in on in mO
<N O cn en
o m om o (NHHW
OOOOOOO idoodd II
lis - ?,_. ^ -S -S -I
b ? -^"K -^^ ?oft,ft,?o^ S? p ^
O ^ *S ft, *5 C^^ -^ ?^ -^ -^ ?^ oa <>?
ono ? cNenri-invorôooNO-cn
(Nmcnmmmmmmmmrj-'^-Tt
D0I: 10.1002/smj

Williams TransferContextReplication 2007

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Williams TransferContextReplication 2007

Uploaded by

Copyright:

Available Formats

Transfer in Context: Replication and Adaptation in Knowledge Transfer Relationships

Author(s): Charles Williams

TRANSFER IN CONTEXT: REPLICATION AND

Levels of replication and adaptation

Knowledge transfer (a = 0.89)

Table 2. Levels of replication and adaptation (2001)

Hypothesis Model 1 Model 2 Model 3

currown -> R 0.32** 0.32*' 0.34***

X2 (d.f.) 155.69 (108) 149.94 (106) 132.74 (98)

Table 3. Structural model of knowledge transfer and performance (first year)

Hypothesis Model 4 Model 5 Model 6 Model 7

firstown - R 0.09 0.08 0.08 0.09

Initial (1) (2) (3) (4) 2001: (5) (6) (7)

firstown -0.001 -0.013 curr_own 0.098 0.083*

APPENDIX 1: DESCRIPTIVE STATISTICS

Ltd. Strat. Mgmt. J., 28: 867-889 (2007)

Strat. Mgmt. J., 28: 867-889 (2007)

8S8 Os Tf? Tt o ? m r-- ^o

odd odd O oS3

You might also like