When Credible Sources Share Complex Knowledge

by

Gabriel Szulanski and

Rossella Cappetta

WP 00-11

A Working Paper of the

Reginald H. Jones Center

The Wharton School

University of Pennsylvania

1

WHEN CREDIBLE SOURCES SHARE COMPLEX KNOWLEDGE

Gabriel Szulanski

Rossella Cappetta

Wharton School, University of Pennsylvania Department of Management

2033 SH-DH Philadelphia, PA 19104-6370

Tel: (215) 573-9627 e-mail: szulanski@wharton.upenn.edu

July 17, 2000

Draft, please do not cite without explicit agreement from the authors

------------------------- The authors acknowledge helpful conversations with and suggestions from Lotte Bailyn, Michael

Boyer, Paul Carlile, John Carroll, Sumantra Ghoshal, Rebecca Henderson, Witold Henizs, Dan Levinthal, John Van Maanen, Costas Markides, Marshall Meyer, Peter Moran, John Stopford, Sidney Winter, JoAnne Yates and participants of seminars at Wharton, London Business School and MIT. Financial support was graciously provided by the Reginald Jones Center and by the Hunstman Center at the Wharton School of the University of Pennsylvania. Errors and omissions are solely the authors’ responsibility.

2

WHEN CREDIBLE SOURCES SHARE COMPLEX KNOWLEDGE

Abstract

It is generally believed that the credibility of the source contributes to the

effectiveness of the transfer of knowledge. This belief, that could be traced back to

Aristotle’s seminal observation that the opinions of “good men” have more impact on

other men’s behavior, has received clear empirical support. Few cautionary notes exist in

the management literature. Likewise, feeble discord can be detected in the field of

communication studies, where occasional warnings are thinly supported by evidence. We

argue that whether or not credibility contributes to the effectiveness of the transfer

depends on how deeply what is being transferred is understood. We focus on situations

where the transfer is primarily an effort to reproduce the results of a successful working

example or template of a complex practice. We show that in the case of a causally

ambiguous template, i.e., when there is irreducible uncertainty about the workings of the

exemplary practice, a credible source might not contribute to the effectiveness of the

transfer. At the core of our argument is the notion of accuracy. We back empirically and

theoretically the notion that a transfer is generally more effective when it is structured

around an initial effort to replicate the template accurately. Our analysis then shows that,

under conditions of causal ambiguity, the credibility of the source is not associated with

the accuracy and hence the effectiveness of the transfer. We rely on primary data

collected through a two-step survey of 122 transfers of organizational practices within

eight firms.

3

Introduction

The problem of improving knowledge utilization continues to attract considerable

attention from academics (Garvin & Oliver, 2000; Hansen, 1999) and practitioners

(Dixon, 2000; O'Dell, Grayson, & Essaides, 1998; Pfeffer & Sutton, 2000). This seems to

be a reflection of the broader interest on the phenomenon of organizational learning, on

how organizations create, retain and transfer knowledge (Argote, 1999; Huber, 1991).

The most prevalent approach to improve knowledge use in organizations is to promote

transfers of better practices that alter, and often improve, the results achieved by those

who absorb them (O'Dell, Grayson, & Essaides, 1998).

It is generally believed that knowledge is transferred more effectively when it

stems from a credible source1. This belief, that could be traced back to Aristotle’s seminal

observation that the opinions of “good men” have more impact on other men’s behavior,

has received clear empirical support. Well known experiments have demonstrated that

immediately after the communication episode, a credible source substantially affects the

recipient’s attitude (e.g: Allen, 1989; Hovland, Lumsdaine, & Sheffield, 1949; Hovland

& Weiss, 1951; Kelman & Hovland, 1953; see for a review: Capon, 1973; Perry, 1996).

In his influential study of work redesign efforts in eight firms, Walton (1975) found that

credible pilot projects of work redesign diffused more rapidly throughout the

organization. Szulanski (2000) found that credibility was negatively correlated to the

difficulty of initiating transfers of best practice within organizations.

More generally, factors normally associated with credibility, such as

trustworthiness (Zaheer, McEvily, & Perrone, 1998), status (Benjamin & Podolny, 1999)

4

and social capital (Belliveau, O'Reilly, & Wade, 1996; Nahapiet & Ghoshal, 1998; Tsai

& Ghoshal, 1998), are believed to contribute to the efficiency of social exchange. Few

cautionary notes exist in the management literature which essentially extols the virtues of

credibility, reputation and trust. Likewise, feeble discord can be detected in the field of

communication studies, where occasional warnings that credibility could distract the

recipient from the contents of the message and thus undermine the effectiveness of the

transfer are thinly supported by evidence (Allen & Stiff, 1989).

In this paper, we show that the extent to which the credibility of the source

contributes to the effectiveness of knowledge transfer depends on the causal ambiguity

associated with that knowledge. We focus on situations in which the transfer is primarily

an effort to reproduce in another location the results of a successful but imperfectly

understood working example of a practice. We show that in the case of a causally

ambiguous example, i.e., when there is irreducible uncertainty about the workings of an

exemplary practice, a credible source might not contribute to and could potentially

detract from the effectiveness of the transfer. Our analysis hinges on the notion of

accuracy. We argue that a transfer is most effective when it is structured around an initial

effort to replicate the example accurately, and show that, under conditions of causal

ambiguity, a credible source does not contribute to the accuracy and hence to the

effectiveness of the transfer. Our analysis relies on primary data collected through a two-

step survey of 122 transfers of organizational practices within eight firms.

1 For individuals, credibility is largely a reflection of their expertise and trustworthiness (Sternthal, Phillips, & Dholakia, 1978; see for a review: chapter 6 Perloff, 1993).

5

Theory and Hypotheses

Transfer of Knowledge as Replication

The spatial reproduction of superior results across additional geographical

locations is a major motive behind the transfer of knowledge. This is variously described

as horizontal organizational learning (Huber, 1991), knowledge use (Glaser, Abelson, &

Garrison, 1983; Leibenstein, 1966), transfer of knowledge and best practices (O'Dell,

Grayson, & Essaides, 1998), sharing of common knowledge (Dixon, 2000) and

knowledge sharing (Hansen, 1999).

When the details of the successful working example can be observed by the agent

seeking to reproduce results, the process of transferring knowledge underlying superior

results could be conceived as a replication (Nelson & Winter, 1982; Rivkin,

forthcoming). Replication differs from imitation in that the replicating agent has access to

a template or working example of the practice to be replicated, both through the

interpretations of the source agent or through direct observation. The replicating agent

seeks to obtain similar results by creating an exact or partial replica of a web of

coordinating relationships connecting specific resources so that a different but similar set

of resources is coordinated by a very similar web of relationships (Winter, 1995).

The notion of replication extends the signaling metaphor, hitherto the

predominant analytical approach to knowledge transfer (Attewell, 1992; Shannon &

Weaver, 1949). While preserving the components of Shannon’s linear model of

communication, SMCR (source, message, channel and receiver), as the central figure of

6

analysis, and the notion that the exchange is directional2, the replication metaphor

augments the signaling metaphor in at least two significant respects.

First, it portraits knowledge transfer as a protracted iterative process rather than as

an instantaneous act as depicted by the signaling metaphor. This modification is

motivated by the empirical observation that complex knowledge must be recreated by the

recipient, rather than obtained through a single act of transmission and absorption

(Attewell, 1992; Rosenberg, 1982; Zander & Kogut, 1995). Replicating such knowledge

typically requires several iterations before an acceptable replica is produced.

Iterations are an inevitable consequence of the uncertainty and equivocality that

results from either causally ambiguous knowledge (Nelson & Winter, 1982) or distortions

and filtering in the communication process (Stohl & Redding, 1987). Causally ambiguous

knowledge would normally have features that are irrelevant or even detrimental to the

effectiveness of the replication and ones that, though desirable, are impossible to replicate

– such as unique human capital. Furthermore, some of its features, replicable or not, may

be tacit. Many of the things that are being done right are not obvious and unlikely to be

codified; other things are quite obviously being done but are correct in non-obvious ways.

This makes it challenging to forecast or even explain the performance of the replica and

hence the effectiveness of the transfer.

2 Replication is seen as a special kind of communication process. The general communication process entails convergence on a shared interpretation and meaning between actors (Boland & Tenkasi, 1995; Rogers & Kincaid, 1981; Rommetveit, 1974; Weick, 1979). In a replication, the resulting interpretation is expected to be close to the original one, because the recipient’s practice will be deemed a partial or complete replica of the original template. This defines specialized roles for the actors of the replication process. One actor provides the working example that anchors the ensuing interpretation of the practice. Thus, even though we acknowledge that the source may learn from the exchange of knowledge, the transfer is considered effective only if the recipient is able to apply specific knowledge that was in the possession of the source at the onset of the transfer. In other words, the significant portion of the learning must occur at the recipient side.

7

Causal ambiguity increases the likelihood of iterations, even with distortion free

communication between agents. As Jensen and Meckling (1992) explain: “Uncertainty

about what specific piece of idiosyncratic knowledge is valuable enlarges transfer costs in

a subtle way. After the fact, it is often obvious that a specific piece of knowledge critical

to a decision could have been transferred at low cost (for example, particular quirks of an

organization, person, legal rule, or custom). But transferring this specific piece of

knowledge in advance requires knowing in advance that it will be critical.” (p. 255).

Thus, the initial attempt to replicate may be unsatisfactory because crucial details of time

and place have been left out from the initial replication attempt thus requiring further

iterations to correct oversights.

But, when the source and the recipient are active agents in the replication process,

additional iterations may be needed to correct oversights that result from inevitable

distortions that arise during the communication process. Message simplification, the most

general level of such distortions, encompasses ‘abbreviations, condensations, and loss of

detail’ (Stohl & Redding, 1987: 479). A major source of these and other forms of

distortion are manifestations of the limited information processing capacity of the agents

(Arrow, 1974).

Iterations may take one of two forms, whether they stem from causal ambiguity or

from distortions in the communication process. The recipient may directly compare the

replica to a working example in order to uncover flaws in the replica of the practice, or it

may challenge the source’s conception of the working example in light of the difficulties

being presented by the replication.

8

This leads to the second crucial extension of the signaling metaphor. Replication

acknowledges the relevance of social interaction. In particular, when source and recipient

are active agents, the process and the outcome of the transfer is affected by their identities

and by the nature of their relationship. The nature of social ties and of the organizational

context in which the transfer occurs (Hansen, 1999; Kostova, 1999; Szulanski, 1996) are

important contingencies of a replication process.

Effective Transfer and Accurate Replication

An important aspect of the problem of knowledge utilization is that successful

examples of existing practices frequently serve as referents to set aspirations for what

could be attained by applying existing knowledge (Cyert & March, 1963; Glaser,

Abelson, & Garrison, 1983); the benchmarking movement providing tangible dramatic

evidence of such expectation setting process (O'Dell, Grayson, & Essaides, 1998).

Because comparable results may be attained through different means, whether or

not existing modifications are made to the working example is a defining characteristic of

the knowledge utilization process.3 Indeed, Hammer (1999) claims that process

standardization versus process diversity is the key structural issue faced by process

enterprises. Likewise, Barley and Tolbert (1997) theorize that the choice of whether to

replicate or revise existing activities and patterns of interaction4 is a decisive moment in

the institutionalization process. Weick (1979) suggests that an organization’s survival

chances improve if existing practices are assigned little more weight than any other

3 That is because even though copying precisely an existing practice is often economical and functional, modifying the practice to meet the specific needs of a new situation may yield superior performance.

4 Barley and Tolbert (1997) use the term ‘script’ to connote the observable, recurrent activities and patterns of interaction characteristics of a particular setting.

9

version that could be fabricated. Axelrod and Cohen (1999) argue that a crucial

intervention in a complex adaptive system consists of consciously choosing between

multiplying a specific type of agent through errorless copying (to the extent such

replication is possible) or increasing the variety of types by activating recombination

mechanisms.

Fomenting variety in a system could be risky. As Axelrod and Cohen (1999)

explain, “most of the variants introduced into orderly systems by [uncontrolled forces,

external to the system] are deleterious—with occasional small improvements and a

sprinkling of very rare spectacular advances. Exploring for new possibilities, by nearly

random variation, can therefore be expensive. In fact, random variation is even slower

than enumerating all of the different possibilities, since random generation will add

duplication. With random variation, you examine each piece in the haystack and put it

back if it is not the needle, possibly to draw it again later” (p. 39). Indeed, Hammer

(1999) argues strongly for extensive standardization, even though he recognizes

explicitly that process diversity helps in meeting multiple customer demands and

sustaining an entrepreneurial culture.

Thus, when effective knowledge utilization is the goal, it may be preferable to

recreate existing knowledge, ignoring, at least initially, the temptation to accommodate

environmental differences; unless of course change is inescapable. Indeed, McDonalds

quickly realized that it could only be successful abroad if it stuck to the very same menu

and store design that worked in the US. For example, when McDonald's Australia finally

restored the standard American menu, its operation moved into the black after eight

consecutive losing years. Likewise, only when the McDonald's units in Germany began

10

to look more like those in the United States did they begin to build volume. The menu

also became standard McDonald's fare with the elimination of chicken and the addition of

the Quarter Pounder. As a result, the once low-volume German units began reaching the

sales averages of American stores. As a German executive concluded, “it seems that any

detour that we made from the standard McDonald's didn't work . . . we realized it was

better to stick with the system and, if necessary, wait for the German consumer to accept

it" (Love, 1995). Likewise, Bradach (1998: 23-24) found that franchise operators in his

sample quickly learned to conform to franchise formats, despite idiosyncratic local

pressures, because tinkering with isolated operating procedures of the complex

interlocking franchise operating system brought numerous unproductive distractions that

culminated in a re-discovery of franchise format. In a similar vein, Knott (1997) shows

that franchisors add value by enforcing compliance with prescribed routines implying

that adaptations to idiosyncratic needs undertaken by individual franchisees are on

average counter-productive.

The risk of modification seems to increase with causal ambiguity. As Adler

(1990) explains, it is desirable to stick to the original design of a technologically

sophisticated practices as much as possible because such practices rely often on poorly

mastered process techniques to such an extent that any substantial divergence from the

existing, functioning design of the process risks multiplying operational problems beyond

manageable levels. Intel’s “Copy Exactly” philosophy for building semiconductor plants

provides a tangible example of the practical meaning of such logic. Recognizing that

semiconductor production processes have enormously complex and opaque causal

structures, Intel requires that every change to the specifications of a semiconductor plant

11

be approved by a central committee, and, when approved, the change must be

implemented across all of the fabs built to that specification. Emphasis on precision is

such that Intel personnel joke that “even the height of the process technicians must be

identical at all fabs” (Iansiti, 1998). Likewise, Rank Xerox, a benchmarking pioneer,

allows a business unit to adapt a model process only after it has raised its performance to

the same level achieved by the benchmark unit. That is because it found out that it is

frequently difficult to pin down exactly what makes a particular approach effective, and

that business units that, despite a limited understanding, decide to modify a practice

before succeeding in replicating it rarely achieved satisfactory outcomes ("Xerox makes

copies," 13/07/1997).

Axelrod and Cohen (1999) comment on the importance of a standard of reference

in the presence of ambiguity, “When agents are not able to predict the effects of various

possible courses of action, they may resort to imitating the observable behavior of agents

who seem to be successful, or who at least have more experience with the new

environment (Cialdini, 1984). Imitating others who are successful or experienced is a

form of implicit attribution of credit that certainly has its disadvantages. When features

that are copied are only superficially relevant the results can be wasteful or even comical.

Nevertheless, following the practices of those with more experience or success is often a

good strategy in an uncertain world” (pp.148-149). Helper, MacDuffie and Sabel (1999)

argue that deliberate benchmarks sharpen error detection and correction and hence

mutual learning in inter-firm transfers of complex automobile manufacturing practices.

As Winter and Szulanski (forthcoming) explain, having a detailed template is of

particular value when clues are required to solve problems that arise in other locations:

12

“When creating a replica, the value of an explicit working example as a point of reference

naturally decreases with each step away from that explicit example . . . Due to differences

in environmental conditions, modifications introduced to adapt the established template

may create new problems; problems that will have to be solved in-situ through a costly

process of trial and error, since they cannot be solved through reference to the established

template (pp. 18-22)”. Accurate reproduction of the template could speed up the climb

towards satisfactory results.

Unsatisfactory results may be attributed to environmental differences or to

deficient execution5. Because the goal of the replication is to attain comparable results by

creating a replica of the practice, a replicator that achieves unsatisfactory results will seek

to establish whether or not the copy of the practice qualifies as a replica. For that, it must

perform a detailed comparison of the replica with the original practice, of which the most

specific manifestation is the available template. Accurate reproduction of the specific

details of the template thus shortens the time and effort required to pinpoint and correct

differences that may exist between the replica and the original practice6.

There is therefore a sequential link between accuracy of the replication and the

effectiveness of the transfer. Without precise attribution of fault it is harder to pinpoint

the obstacles that must be overcome to improve results. In turn, inaccurate reproduction

5 Of course, it may very well be that the communicated interpretation of the practice by the source, which includes a recommendation to replicate, rather than modify, is severely deficient. While it may be hard to rule out different interpretations of causally ambiguous practices, we assume that severe misinterpretation by the source is unlikely.

6This argument parallels Waver’s treatment of the general problem of communication, of which replication is a special case. Waver breaks the problem of communication into three sub-problems or levels, technical – how accurately can the symbols of the communication be transmitted, semantic – how precisely the transmitted symbols convey the desired meaning, and effectiveness – how effectively does the received meaning affect conduct in the desired way. Waver reaches a similar conclusion, “ . . . [semantics and

13

of the details of the template hinders the precise attribution of fault. As the chairman of

Rank Xerox (UK) confirms, “Following the best practice down to the last detail is crucial

. . . We lost a lot of best practices because people edited them before implementing them”

("Xerox makes copies," 13/07/1997).

In retrospect, it may turn out that the decision to replicate rather than modify a

practice may have been inappropriate. Wal-Mart’s blatant decision to enter Argentina

with the same basic US store model, disregarding local idiosyncrasies (a decision since

reversed), is a dramatic example of such a situation: “The meat counters featured

American cuts like T-bone steaks, not the rib strips and tail rumps that Argentines prefer.

Cosmetic counters were filled with bright colored rouge lipstick, though Argentine

women tend to like softer, more natural look. And jewelry displays gave prominent

placement to emeralds, sapphires and diamonds, while most women [in Argentina] prefer

wearing gold and silver. The first few stores even had hardware departments fool of tools

and appliances wired for 110 volt electric power; the standard throughout Argentina is

220” ("Selling to Argentina," 12/05/1999).

However, barring such extraordinary circumstances where copying is clearly

inappropriate, we conclude that an initial effort to copy the template accurately will

increase the likelihood of obtaining a satisfactory outcome. An accurate transfer is more

likely to be effective.

Credibility of the Source and Effective Knowledge Transfer

The re-creation of a template could be guided by a supplied conception of that

template or by inferences drawn directly from observing that template. Thus, the source’s

effectiveness] can make use only of those . . . accuracies which turn out to be possible when analyzed at

14

agent can assume two possible roles; it can act as gatekeeper to the template or it can

supply a conception of the template.

Direct observation of the template allows the recipient to infer quick and

arbitrarily precise answers; answers that are also presumed correct because the template

is the best-known manifestation of the practice. However, inferences drawn from the

template may in fact exceed the scope of the original conception of the practice thus

over-prescribing its functioning. In other words, the template contains de-facto solutions

to yet unspecified problems (Brooks, 1995). Conversely, a supplied conception of the

template may leave out practical detail that is needed to create a viable replica. For this

reason, a replication effort that combines direct observation of the template with

information about the practice supplied by the source is likely to result in a more accurate

replica of the actual underlying practices.

The credibility of the source affects how the conception of the practice supplied

by the source will influence the behavior of the recipient, and thus the effectiveness of the

replication. When the source is credible, i.e., perceived as knowledgeable and

trustworthy, the recipient will be less suspicious of the offered conception and thus more

open and receptive to its detail ( Hovland, Lumsdaine, & Sheffield, 1949; Hovland &

Weiss, 1951). This increases the amount of information that can be exchanged (Carley,

1991; Tsai & Ghoshal, 1998) and decreases the cost of the exchange (Curall & Judge,

1995; Zaheer, McEvily, & Perrone, 1998). More detail can be communicated to the

recipient, which can thus afford a better grasp of the source’s conception of the practice.

Credibility could thus enhance the accuracy of the transfer.

[the technical level] ” (Shannon & Weaver, 1949: 6).

15

On the other hand, a credible source could distract the recipient from the details of

the supplied conception of the practice, beyond obtaining an impression of the source’s

general idea (Allen & Stiff, 1989; Perry, 1996). As Petty and Cacioppo argue, a highly

credible source inhibits critical thinking, i.e., the processing and counterarguing that

would normally take place during the receipt of a counterattitudinal message (Petty &

Cacioppo, 1996). The recipient will expect little damage to ensue from interactions with a

highly credible source (Noteboom, 1997) and, consequently, will take fewer steps to

reduce the inherent uncertainty of the situation by monitoring closely the actions of the

source (Berber, 1983; Lewis & Weigert, 1985; McAllister, 1995) 7. For this reason, the

recipient is less likely to attend also to available details of the template.

Thus, in a transfer from a highly credible source the recipient is more likely to

assume that the source knows the object of the transfer and how to transfer it (because of

the expertise) and that it intends to approach the transfer in the best way possible

(because of the trustworthiness). When the recipient is distracted and a portion of the

available detail of the conception or the template is lost, the accuracy of the replication is

likely to suffer.

Causal Ambiguity and the Overall Effect of Credibility.

Credibility, thus, both contributes to and detracts from the accuracy of the

replication and hence from the effectiveness of the transfer. For simple, easy to

comprehend practices, where little uncertainty exists about the functioning of the

practice, the conception of the source is likely to approximate the essence of the original

7 In particular, McAllister (1995) argues that a high level of cognition-based trust, that is the trust derived from the evaluation of the positive characteristics of the other person, is associated with little control-based monitoring, that is the monitoring of the other person’s actions in order to control her or him.

16

practice. Hence, a receptive recipient is likely to produce an accurate replica. Further, the

negative effect of distraction is modest, because missed detail can be easily detected and

reconstructed. Hence the positive contribution of credibility is likely to out-weight the

negative contribution. This leads us to our first hypothesis:

H1: Under conditions of low causal ambiguity, a credible source contributes to

the effectiveness of knowledge transfer.

Conversely, for complex and ambiguous practices, when available information

does not allow discriminating among equally plausible conceptions of the practice, the

conception actually supplied by the source might not be the correct one. Further, it will

take longer for a receptive and trusting recipient to appraise the situation correctly

because a credible source may inhibit close scrutiny of available detail and missed details

of a complex practice may not easily reconstructed. In this case, the recipient will initially

produce an unsatisfactory replica. The consequences of initial oversight will be harder to

correct the longer they last, because mistakes become progressively costlier to rectify.

Hence the negative contribution of credibility will loom larger and possibly nullify the

positive contribution. This leads us to our second hypothesis:

H2: Under conditions of high causal ambiguity, a credible source does not

contribute to the effectiveness of knowledge transfer.

Method

Sample and Research Process

The transfer of best practices (O'Dell, Grayson, & Essaides, 1998) provides a

propitious setting to observe transfers of complex knowledge within organizations. Data

were collected through a two-step questionnaire survey. The first step of the survey asked

17

companies to provide a list of transfers for study that included sufficient detail about the

parties involved in those transfers (i.e., respondents). More than 60 companies, with

varying degrees of experience in the transfer of practices, expressed interest. Of that

group, 12 were able to provide such a list. Of the 12, only eight provided entries of

sufficient quality to warrant continuation of their involvement in the research. The eight

companies were: AMP, AT&T Paradyne, British Petroleum, Burmah Castrol, Chevron

Corporation, EDS, Kaiser Permanente, and Rank Xerox.

The second step of the survey was devised to analyze accuracy at specific

transfers. The final sample consisted of 271 returned questionnaires, spanning 122

transfers of 38 practices8, for a response rate of 61%. To obtain a balanced perspective on

each transfer, separate questionnaires were sent to the source, the recipient, and a third

party to the transfer. The respondents were comprised 110 sources units, 101 recipient

units and 60 third parties. Average item non-response was lower than 5%. An average of

7.3 questionnaires were received for each practice studied.

To provide practices for study, companies were directed to search for transfers of

important activities or processes that showed evidence of difficulty during the transfer

and in the adaptation of the practice by the recipient9. They were also instructed to rule

out practices that could be performed by a single individual and to choose only practices

that required the coordinated effort of many.

8 The sample contained both technical and administrative practices. Examples of technical practices are software development procedures and drawing standards. Examples of administrative practices are upward appraisal and activity-based costing (ABC). Full disclosure of the practices studied is precluded by a guarantee of confidentiality.

9 In an effort to increase the variance in the dependent variable, this directive was necessary to counter the inclination of firms to report only successful transfers.

18

Construction of Measures

Multiple-item scales were developed for all constructs to ensure the reliability and

validity of the measurement system. Little empirical precedent was available to guide the

development of the measures. A broad and thorough literature review informed the

generation of the initial constructs and the a priori assignment of items to measure those

constructs. In-depth clinical work, consultation with subject experts and feedback

obtained when piloting the questionnaire helped refine the choice of constructs, identify

the most relevant items for those constructs and select their proper wording given the

empirical context. Some items were discarded, but not re-assigned, after the full data set

was obtained.10

Unless otherwise stated, a balanced five-point Likert-type scale was used to

measure most items in the questionnaire: Y! = “Yes!”; y = “yes, but”; o = “no opinion”,

n = “no, not really”, N! = “No!” Following Nunnally’s (1978) recommendation,

construct scores were computed by adding up the standardized item scores.

Below we detail the operationalization of the central constructs for this paper.

Except for accuracy, all other constructs have been fully described in Szulanski (1996).

Causal Ambiguity

The construction of the measure of causal ambiguity was informed primarily

based the theory of uncertain imitability (Lippman & Rumelt, 1982), which posits that

efforts to reproduce results may be hindered by irreducible uncertainty about the elements

of production and how they interact to produce the desired outcome. Causal ambiguity

means that the precise reasons for success or failure cannot be determined, even ex-post,

19

and it is therefore impossible to produce an unambiguous list of factors of production,

much less measure their marginal contribution. For this reason, the accuracy of

reproduction is likely to decrease with increasing causal ambiguity about the functioning

of the working example.

Five of the eight items in the measure of causal ambiguity were derived from

Lippman and Rumelt’s theorizing (1982). The full text of those questions is: The limits of

the «practice» are fully specified; With the «practice», we know why a given action

results in a given outcome; When a problem surfaced with the «practice», the precise

reasons for failure could not be articulated even after the event; There is a precise list of

the skills, resources and prerequisites necessary for successfully performing the

«practice»; and It is well known how the components of that list interact to produce

«practice»’s output.

The remaining three items are designed to infer causal ambiguity from the degree

of tacitness of the practice (Nelson & Winter, 1982; Teece, 1977; Winter, 1987; Zander

& Kogut, 1995). The full text of these items is: Operating procedures for the «practice»

are available; Useful manuals for the «practice» are available; Existing work manuals and

operating procedures describe precisely what people working in the «practice» actually

do.

Credibility of the Source

Credibility of the source stems primarily from its expertise and trustworthiness

(Hovland, Janis, & Kelley, 1953; Perloff, 1993; Sternthal, Phillips, & Dholakia, 1978),

other dimensions receiving only fragmented support. Expertise is “the extent to which a

10 The a priori assignment of items was preserved for all constructs except accuracy. See

20

communicator is perceived to be a source of valid assertions” (Hovland, Janis, & Kelley,

1953: 21). Trustworthiness is “the degree of confidence in the communicator’s intent to

communicate the assertions he consider most valid” (Hovland, Janis, & Kelley, 1953:

21)11. A third dimension of relevance to our study is the degree of similarity between the

source and the recipient, which some treat as an autonomous dimension of credibility

(Perloff, 1993); and others consider it to be an adequate proxy or determinant of

credibility (Hovland, Lumsdaine, & Sheffield, 1949). Finally, our notion of agent or

organizational member, while typically referring to an individual, applies to any unit that

can "accomplish something on its own" (Nelson & Winter, 1982: 98). For this reason, we

relied also on Walton’s (1975) study of the determinants of the credibility of an

organizational unit to develop our measure.

Based on the above sources, we constructed eight items to measure credibility. To

assess knowledgeability we asked whether the «source»: had invented the «practice», was

the first unit to have experience with the practice or it had received the practice from

other unit; and whether the «source» was able to accommodate the needs of «recipient»

into «practice». To assess trustworthiness we asked whether the «source» had a hidden

agenda; whether superior results of the «source» were visible; and remained stable;

whether the «source» possessed the necessary resources to support the transfer; and

description below. 11 In management literature, the concept of trustworthiness is frequently used as synonymous of

credibility and includes dimensions of expertise and dimensions of intention. Mayer, Davis and Schoorman, e.g., write about three “factors of perceived trustworthiness”: the ability, such as “that group of skills, competencies, and characteristics that enable a party to have influence within specific domain” (1995: 717); the benevolence, such as “the extent to which a trustee is believed to want to do good to the trustor, aside from an egocentric profit motive” (pg: 718); and the integrity, such as “the trustor’s perception that the trustee adheres to a set of principles that the trustor finds acceptable” (pg: 719).

21

whether the «source» had a history of successful transfers. To assess similarity we asked

whether the «source» and the «recipient» have similar key success factors.

Accuracy of Transfer

The accuracy of the transfer of practice refers to the care invested in producing a

close replica of the template. Thus a measure of accuracy must be sensitive to differences

between the replica and the original template, i.e., to modifications introduced to the

original template, intentionally or otherwise (Eisenberg & Phillips, 1991; Stohl &

Redding, 1987). Communication scholars suggest two types of modifications.

Modifications can either be general, i.e. affecting the comprehensive meaning of the

practice. Alternatively, specific modules of the practice can be altered while preserving

the overall meaning of the practice.12

The measure of accuracy has eight items. We used six items to assess the level of

general modifications. We first asked whether compared to the source’s practice, the

recipient’s one is: 1 = “Exactly the same”; 2 = “Essentially the same”; 3 = “Slightly

modified”, 4 = “Markedly modified”, 5 = “Completely different”. Then we distinguished

between appropriate and inappropriate general modifications. Specifically, we asked

whether modifications were introduced to make the practice workable, and to adapt the

practice to different environment. We asked whether unnecessary modifications were

performed; whether the practice was modified in ways contrary to expert’s advice; and

whether, altering the practice, further problems have been created.

Then, we tried to assess specific modifications. At first, we evaluate the

incompleteness of the replication by asking whether: 1 = “All modules have been

22

transferred”; 2 = “Only selected, but all the essential modules have been transferred”; 3

= “Only the essential modules have been transferred”, 4 = “Only selected modules, some

essential some not, have been transferred”, 5 = “None of the modules have been

transferred”. Further, we asked whether original modules of the practice were replaced by

existing ones at the recipient.

Performance of the measurement model

Additional measures are introduced as control variables to complete the

specification of the model and control for unobserved heterogeneity. These factors

include the remaining elements of Shannon’s model of communication (Shannon &

Weaver, 1949), i.e., the source and recipient’s motivation, the recipient’s absorptive and

retentive ability, and proveness of the knowledge transferred; elements of the social

context, i.e., the ease of the relationship and the fertility of the context; as well as dummy

variables to control for firm specific effects. Finally, we added dummy variables to

control for the perspective of the respondent to the questionnaire who could either be a

source, a recipient or a third party to the transfer. Detailed specification of these items is

available in Szulanski (1996).

Table 1 summarizes the performance of the measurement model, including the

dependent variable, the predictors and the control variables.

----------------------------------

Insert Table 1 about here

---------------------------------

12 This distinction is similar to the distinction made by Henderson and Clark (1990) between architectural knowledge and component knowledge.

23

Convergent validity (reliability and unidimensionality) was evaluated separately

for each construct (Gerbing & Anderson, 1988). Cronbach’s alpha was used as a measure

of reliability because it provides a lower bound to the reliability of a scale and is the most

widely used measure (Nunnally, 1978). All but two scales had alpha greater than .70, thus

providing an adequate level of reliability for predictor tests and hypothesized measures of

a construct (Nunnally, 1978: 245-246). The two least reliable scales scored only

marginally below that standard. Unidimensionality was assessed through factor analysis

and computation of the theta coefficient (Armor, 1974; Carmines & Zeller, 1979; Zeller

& Carmines, 1980). The unidimensionality of all 10 scales was adequate. Finally, all

variables meet reasonable assumptions of normality (see Table 1 for skewness and

kurtosis values).

Discriminant validity was evaluated for all construct pairs by examining the

observed correlation matrix of the constructs. If the correlation between constructs i and j

is 1, (i.e., if constructs i and j are perfectly correlated), the observed correlation should be

(αi.5) * (αj

.5) where αi and αj are the reliability coefficients for the constructs. In practical

terms, testing for discriminant validity entails computing the upper limit for the

confidence interval of the observed correlations and testing whether this limit is smaller

than the maximum possible correlation between the scales as computed from their

reliability coefficients. Table 2 reports the correlations for all the variables. All construct

pairs met the discriminant validity test at p < .0012.

----------------------------------

Insert Table 2 about here

24

---------------------------------

In the design and administration of the questionnaire, several steps were taken to

minimize measurement error (Nunnally, 1978). Formulated only after extensive

fieldwork, the questionnaire was pre-tested with all the participating companies, with

experienced academics, and with respondents who volunteered to record their reactions

while completing it. Finally, the cognitive effort of the respondents was reduced by

minimizing the number of scales to be learned and by translating generic terms like

“source” or “recipient” into the specifics of a particular transfer.

Assumptions for the analysis.

Predictors are invariant throughout the transfer.

As a first approximation, predictors are assumed to remain invariant for the

duration of the transfer. When such assumption holds true, the timing of the measurement

of the independent variables is not critical. This assumption is deemed reasonable

because most of the predictors typically change slowly. However there may be

exceptions. Some predictors such as the motivation of the source, the motivation of the

recipient and the nature of the relationship between the units may be affected by the

expected outcome of the transfer. Pre-existing relationships between source and recipient

sub-units did exist for al least two years prior to the beginning of the transfer.

Cross-sectional comparison of transfers is warranted.

Leonard-Barton (Leonard-Barton, 1990: 259) argues that it is necessary to

measure multi-item constructs at a “defined point” in time if meaningful comparisons are

wanted, because the meaning of complex constructs depends on when during a transfer

they are measured. As point of reference for her study she selected the “very first use of

25

the technology in a routine production task” as the anchor point. She chose that point

because it could be identified with a “satisfactory degree of precision”. In this study, all

questionnaires were completed within a narrow13 band of 3.5 months, which started 5

months after the first day that knowledge was first put to use by the recipient. Thus, all

transfers are at a defined and comparable point in time. Comparison across transfers is

thus considered appropriate.

Analysis

We used hierarchical regression to test the hypotheses with the following model:

Accuracy = β0 + β1 Causal Ambiguity + β2 Credibility + β1x2 Causal Ambiguity x Credibility + βc1 Control Variable4+ …+βcn Control Variablen

Model 1 included only six control variables pertaining to the main characteristics

of the transfer. Model 2 firm adds firm-specific dummy variables; We then remove firm

dummies and add causal ambiguity and credibility (β1-β2) in Model 3 and their

interaction (β1x2) in Model 4. In Model 5 we add firm dummies to Model 4. Finally, in

Model 6 we add perspective dummies to Model 4.We report partial F tests to significance

of the added variables.

Results

Table 3 displays the findings from the regression analyses for the six models.

----------------------------------

Insert Table 3 about here

---------------------------------

13Such a band of 3.5 months can be considered narrow, because it means that all transfers were sampled early on in the integration stage which has been documented to last between 1.5 to 2 years.

26

Model 1 is strongly significant (F= 8.771; p< .001) with adjusted R-square of

.231. As the partial-F test shows, the addition of a block of firm dummy variables in

Model 2 is significant raising the Adjusted R-square to .282. However these variables

were individually insignificant.

Models 3 and 4 clearly demonstrate the predictive power of the three predictors of

accuracy. In introducing the first two predictors to the regression, Adj. R-square increases

to .418 for Model 3 and then to .428 when the interaction term is added in Model 4. That

the addition of causal ambiguity and of their interaction is significant is confirmed by the

results of the Partial-F tests. The addition of the interaction term is significant at the (p <

.05) level.

Neither firm dummies (Model 5) nor perspective dummies (Model 6) are

significant additions to Model 4, either individually or as blocks as shown by the partial F

tests. Adj. R-square increases from .428 in Model 4 to .446 in Model 5 and decreases to

.425 in Model 6. The modest gain in predictive power of Model 5 does not seem to be

justified by the loss of degrees of freedom.

The control variables, when significant, are relatively stable and have the

expected sign. A supportive context, and an easy relationship could be expected to

contribute to accuracy. The recipient’s absorptive capacity (Cohen & Levinthal, 1990)

contributes to accuracy because it relates the capacity of the recipient to adopt the

perspective of the source and better absorb his or her knowledge thus contributing to the

effectiveness of the exchange (Rommetveit, 1974). The source’s motivation contributes

to accuracy, because it reflects the desire of the source to exchange knowledge thus

increasing the effective amount of opinion and explanatory information transferred to the

27

recipient (Berger & Kellerman, 1983). Retentive capacity, which becomes almost

significant in the fully specified models, should be negatively related to accuracy. This is

consistent the interpretation of this variable as indicative of the existence of barriers to

unlearning (see discussion in Szulanski, 1996). It is also consistent with the argument that

a recipient that is capable of ‘switching’ from one perspective to another will increase the

effectiveness of the exchange (Rommetveit, 1974). A motivated recipient tends to

introduce unnecessary changes to preserve status and ownership thus detracting from

accuracy (see discussion in Szulanski, 2000).

The coefficients of causal ambiguity and credibility (Models 3-6) are highly

significant and stable. As expected, the coefficient of causal ambiguity (β1 in the equation

1.1) was negative and significant (in Model 4: -.341; p< 0.001) and the coefficient of

credibility (β2 in the equation 1.1) was positive and significant (in Model 4: -.341; p<

0.001). The coefficient of their interaction (β1x2 in equation 1.1) is negative and

significant (in Model 4: -.145; p< 0.05) suggesting that the connection between

credibility and accuracy weakens with increasing levels of causal ambiguity.

To better understand the nature of the connection between credibility and

accuracy we plotted the relationship between these three variables. The graph, using

quadratic smoothing on the raw data, clearly shows a strong positive relationship between

credibility and accuracy for low levels of the causal ambiguity, a relationship that

weakens with increasing for levels of the causal ambiguity.

----------------------------------

Insert Figure 2 about here

---------------------------------

28

To complete the analysis of the nature of the interaction between causal ambiguity

and credibility we analyzed the sign and the magnitude of the total effect that credibility

has on accuracy, conditional on the level of the causal ambiguity following procedures

detailed by Aiken and West (1991).

The total effect of credibility on accuracy is derived from equation [1.1] as

follows:

βCredibility tot = β2 + β1x2 Causal Ambiguity [1.2] Table 4 reports the total values of βCredibility tot and the associated p-values14 with

causal ambiguity set to different values. All computations rely on the estimates of Model

4.

These results clearly support the hypotheses advanced in this paper. The total

effect of the credibility on accuracy is positive and significant at low levels of causal

ambiguity, suggesting that credibility will be associated with accuracy at low levels of

causal ambiguity. However the total effect becomes non-significant when causal

ambiguity reaches one standard deviation above the mean and negligible at two standard

deviations above the mean.

--------------------------------

Insert Table 4 about here

---------------------------------

14 P-value for the total effect are calculated using the following t-statistic: tCredibility tot = βCredibility tot / SE(βCredibility tot), where SE(βCredibility tot) = √ [VAR(β2) + Causal Ambiguity2VAR(β1x2 )+ 2Causal Ambiguity COV(β2, β1x2)].

29

Robustness of the Results

Further analyses were conducted to explore the stability of the coefficients.

Missing data were handled in three different ways. First, regressions were run with

missing data deleted case-wise, then with missing data deleted pair-wise and finally by

substituting the missing value of the constructs with the mean value of the construct.

Results remain stable also when company dummy variables and perspective dummies are

included in the four regression equations.

Further, the results reported are based on an analysis in which each questionnaire

is treated as a discrete data point. In other words, identical questionnaires completed by

the source, by the recipient and by the third party pertaining to a same transfer are each

treated as a singular data point. Thus, each transfer – the unit of analysis – is sampled

three times15. This raises the problem of non-independence of data. To confirm the

stability and robustness of the findings, additional analyses were conducted. A single

observation was created from the three questionnaires for the same transfer with two

methods: by discarding all but the best16 questionnaire for each transfer (highest quality

of response) and by averaging the three questionnaires. In all these analyses, the models

remain highly significant with adj. R-square >= .420, samples sizes ranging from 102 to

112 observations. The analyses revealed that all coefficients of the predictors are stable.

Discussion and Conclusion

Against a background of almost unqualified confidence in the value of credibility,

we show that the extent to which the credibility of the source contributes to the

15 Unless one or more questionnaires for that sample have not been returned. 16 The questionnaires were selected based on the completeness and on the accuracy of the

responses.

30

effectiveness of knowledge transfer depends on the causal ambiguity associated with that

knowledge. We focus on situations in which the transfer is primarily an effort to

reproduce the results of a successful but imperfectly understood working example of a

practice in another location. We show that in the case of a causally ambiguous example,

i.e., when there is irreducible uncertainty about the workings of an exemplary practice, a

credible source might not contribute to the effectiveness of the transfer. Our analysis

hinges on the notion of accuracy. We argue that a transfer could be seen as an instance of

replication. A replication is most effective when it is structured around an initial effort to

replicate the example accurately. For a high level of causal ambiguity, the accuracy of the

transfer is not related to the credibility of the source. Under such conditions, a credible

source does not seem to contribute to the effectiveness of the transfer.

Our robust findings support the notion that credibility is a double edge sword. On

the one hand, it fosters receptivity; on the other, it may inhibit critical appraisal. For

transfers of simple, easy to comprehend knowledge, credibility contributes to the

effectiveness of the transfer because receptivity translates into accuracy and critical

appraisal plays a modest role because there is little room for error. However, when the

object of transfer is a complex practice – typical of benchmarking, reengineering, total

quality management and other quests for synergy – credibility may not help.

There is an important boundary condition to our analysis and findings. While

replication is a powerful way to utilize existing knowledge, it may not be appropriate

when drastic environmental differences or organizational rigidities demand the

introduction of significant modifications. When replication is not suitable, unsatisfactory

results occur because practices should have been re-designed rather than copied

31

accurately. In such situations, modifications made to a malfunctioning replica to improve

performance will instead yield further complication. Brooks (1995) depicts such

situations for the case of software maintenance, “All repairs tend to destroy the structure,

to increase the entropy and disorder of the system. Less and less effort is spent on fixing

original design flaws; more and more is spent in fixing flaws introduced by earlier fixes.

Maintenance is an entropy increasing process, even its most skilful execution only delays

the subsidence of the system into unfixable obsolescence” (pp. 122-123). Our analysis

holds exclusively for situations where replication is indeed warranted.

Our findings have important implication for understanding the relationship

between isolating mechanisms and tacit knowledge. Successful replication of knowledge,

in a different setting, may be compromised by idiosyncratic features of the new context in

which knowledge is put to use. The theory of uncertain imitability (Lippman & Rumelt,

1982; Rumelt, 1984) suggests that there may be irreducible uncertainty connected with an

attempt to replicate the use of knowledge. Replication falls short of complete success

because this irreducible uncertainty obscures how the features of the new context affect

the results of the replication effort. Modeling re-use of knowledge as the replication of a

production function, Lippman and Rumelt (1982) explain that uncertainty in the

replication effort is most likely to result from ambiguity about what the factors of

production are and how they interact during production. As Rumelt (1984: 562)

explicates, ". . . if the precise reasons for success or failure cannot be determined, even

after the event has occurred, there is causal ambiguity (italics added) and it is impossible

to produce an unambiguous list of the factors of production, much less measure their

marginal contribution." Therefore, he concludes, in the pure theory of uncertain

32

imitability, the fundamental factor that hinders the precise replication of results from the

use of knowledge is causal ambiguity17 (p.567). In the realm of knowledge management,

such difficulties to imitate or replicate are normally attributed to tacit knowing (Nonaka,

1994; Polanyi, 1966; Seely Brown, 1997). As our findings suggest, a recipient’s ability to

evaluate the credibility of a source of knowledge suffers when the subject of transfer is a

complex practice. In such situations, evaluation of the source’s credibility relies primarily

on an emotional or instinctive basis, because a source of causally ambiguous knowledge,

even if perceived as credible, could not really know the practice and be unable to transfer

it. The difficulties of transfer in such case stem not only from the source’s ‘tacit’

knowledge but also from the source’s ‘explicit’ ignorance. The difficulty to transfer

causally ambiguous knowledge stems from both of these reasons.

Our findings also suggest possible research questions for researchers of social

exchange processes. Under conditions of high causal ambiguity, factors normally

associated with credibility, such as trustworthiness (Zaheer, McEvily, & Perrone, 1998),

status (Benjamin & Podolny, 1999) and social capital (Belliveau, O'Reilly, & Wade,

1996; Nahapiet & Ghoshal, 1998; Tsai & Ghoshal, 1998), may contribute less to the

efficiency of social exchange than it is currently assumed. Conclusions drawn about the

effect of these factors should be examined under conditions of high causal ambiguity. A

possible way to extend the investigation of such factors is to measure the role that

accuracy plays in social exchange. Researchers currently measure the amount of

exchange or the cost of maintaining a relationship, measures that capture efficiency but

17Bohn (1994) has called causally ambiguous knowledge “incomplete”. He suggested a practical definition of complete knowledge as a “model that will predict output characteristics to an accuracy of one-tenth of the tolerance band, for changes in inputs across a 2:1 range, and including all interactions.” (p. 70)

33

not the effectiveness of exchange (Carley, 1991; Curall & Judge, 1995; Tsai & Ghoshal,

1998; Zaheer et al., 1998). The measure of accuracy could serve as a proxy for the

effectiveness of such a relationship and provides a fuller picture of how these factors

affect social exchange.

If our findings call for tempering the enthusiasm that exudes from the literature

with credibility and related factors, our theorizing suggests an even stronger cautionary

note. Indeed, it is theoretically possible, even if our findings do not show it, that a

credible source could detract from the effectiveness of knowledge transfer. Because

under conditions of high causal ambiguity it may be impossible to discriminate among

several a-priori plausible conceptions of a practice, a credible source may elect a

conception that actually detracts from the effectiveness of the transfer, compared to a

situation where replication occurs without input from the source. The knowledgeability of

a credible source is increasingly suspect, regardless of the source’s credentials, with

increasing causal ambiguity. In such circumstances, credibility is purely a reflection of

trustworthiness, a mostly subjective and possibly misleading criterion. A ‘credible’

source may not necessarily deserve its credibility. Thus our findings suggest a broader

question for further research: under what conditions will a credible source detract from

the effectiveness of social exchange?

Credibility is a powerful managerial lever to influence knowledge transfer and

hence to foment organizational learning. Take the example of IBM Rochester a flagship

manufacturing plant at IBM, which produced the AS/400 computer system. IBM’s

Rochester plant had the highest morale, lowest turnover and absenteeism rates in IBM,

won Minnesota’s Safety Award for 10 consecutive years, and IBM U.S. Market Driven

34

Quality Award for two consecutive years. Yet, it was not until it won the Malcom

Baldridge National Quality Award, that other units within IBM requested to benchmark

with Rochester. Indeed, it may take a strong signal to affect the perception of credibility.

Such signals, in the form of recognition and awards, are within managerial reach. Our

findings suggest that they should be used with caution. Indeed, credibility could be used

to highlight scarce information or direct the allocation of scarce attention. There are

dilemmas associated with that lever. Credibility could be misleading.

The management literature generally believes it to be desirable to foster factors

that increase fit and congruence between actors (Argote & Ingram, 2000; Argote &

Ophir, forthcoming). Such convictions stem from empirical findings that show a positive

relationship between similarity of actors and the amount of resource exchanged (Carley,

1991); between the trustworthiness of the actors and the amount of resource exchanged

(Tsai & Ghoshal, 1998); and between the strength of the relationship between the actors

and the speed of new product development (Hansen, 1999). Misfit among actors,

reflected in heterogeneity and weakness of social ties, is functional only for stimulating

the creation of and search for new knowledge (Argote & Ophir, forthcoming). Fit, in

contrast, increases the fluidity of knowledge transfer (Argote & Ophir, forthcoming).

Besides increasing the fluidity in certain kinds of knowledge transfer, lack of

credibility, our findings suggest, could play an important role in knowledge transfer. The

transfer of highly ambiguous knowledge provides an occasion to re-assess existing

practices. A credible source may inhibit proper sensemaking and bias convergence on the

conception of the source, thus hindering the effectiveness of knowledge utilization.

35

We wish to conclude by providing a recent example that captures, in our opinion,

the central message of this paper. When causal ambiguity is high, the knowledgeability of

the source is hard to assess and thus credibility reflects basically the level of trust that one

places on the source. Under extreme causal ambiguity, a credible source may not help,

indeed, it may even cause damage. Excerpts from two emails sent recently by our system

administrator in response to the outbreak of a computer virus provide a dramatic

illustration of the limits to the credibility of the source.

May 28th This is a virus alert. If you receive an e-mail

with an attachment that claims to be a resume do not open it unless you trust the source [italics added].

June 19th

The latest update to our virus software is unaware of some new virus variations. An update is due soon. Before then

do not open any attachments unless you are absolutely sure what they are. It doesn’t matter who they are from [italics added].

Stan Kasper, MIS

36

Table 1: Measurement Model

Construct Description Cronbach α

Items Valid N Avg. Inter item Corr.

Skewness Kurtosis

1 Source’s motivation

Motivation of the source unit to support the transfer

.93 13 271 .5 -.16 -1.34

2 Credibility of source

Degree to which the source of the best practice is perceived as reliable

.64 8 210 .19 -.29 -.28

3 Context

Degree to which the organizational context supports the development of transfers

.77 14 247 .2 -.09 -.03

4 Causal ambiguity

Depth of knowledge

.86 8 250 .45 .19 -.74

5 Knowledge proveness

Degree of conjecture on the utility of the transferred knowledge

.67 3 251 .4 -.67 -.27

6 Recipient’s motivation

Motivation of the recipient unit to support the transfer

.93 14 271 .48 -.31 -1.27

7 Recipient’s absorptive capacity

Ability of the recipient unit to identify, value and apply new knowledge

.83 9 252 .36 -.22 -.65

8 Recipient’s retentive capacity

Ability of the recipient unit to support the routinize the use of new knowledge

.81 6 249 .43 -.12 -.04

9 Relationship

Ease of communication and intimacy of the relationship

.71 3 237 .46 -.30 -.61

10 Accuracy Degree of similarity between the replica and the template.

.79 8 203 .32 -.08 -.30

* These scales are composed of binary items. Both scales qualify marginally as Guttman scales

37

Table 2: Correlations Between the Independent Variables, the Dependent Variables and the Control Variables (casewise)

1 2 3 4 5 6 7 8 9 10 11 1 Source’s motivation

1.00

2 Credibility of source

.42 1.00

3 Context

.23 .36 1.00

4 Causal ambiguity

-.28 -.45 -.36 1.00

5 Unproven knowledge

.23 .30 .30 -.47 1.00

6 Recipient’s motivation

.42 .36 .31 -.23 .11 1.00

7 Recipient’s absorptive capacity

.01 .31 .37 -.17 .05 .43 1.00

8 Recipient’s retentive capacity

-.08 .24 .41 -.23 .16 .24 .58 1.00

9 Relationship

.21 .37 .47 -.30 .29 .35 .33 .31 1.00

10 Interaction between causal ambiguity and credibility

-.10 -.02 -.27 .09 -.20 -.05 .00 -.06 -.30 1.00

11 Accuracy

.35 .53 .35 -.52 .28 .25 .32 .16 .36 -.20 1.00

38

Table 3: Regressions of accuracy

Model 1

Model 2 Model 3 Model 4 Model 5 Model 6

Causal ambiguity -.335** (-4.343)

-.341** (-4.474)

-.273** (-3.365)

-.319** (-4.026)

Credibility of source .260* (3.308)

.282** (3.606)

.271** (3.416)

.302** (3.744)

Interaction of causal ambiguity and credibility

-.145* (-2.180)

-.125^ (-1.873)

-.156* (-2.302)

Source’s motivation .285**

(3.587) .235*

(2.958) .144

(1.863) .136

(1.786) .102

(1.300) .108

(1.298) Context .152

(1.867) .189*

(2.233) .040

(.5183) .009

(.112) .027

(.324) .021

(.269) Knowledge proveness .077

(1.089) .027

(.363) -.012

(-.160) -.029

(-.405) -.036

(-.478) -.044

(-.599) Recipient’s motivation -.126

(-1.531) -.128

(-1.572) -.113

(-1.435) -.109

(-1.400) -.110

(-1.391) -.104

(-1.326) Recipient’s absorptive capacity

.279* (3.144)

.263* (3.010)

.255* (3.005)

.270* (3.213)

.244* (2.895)

.272* (3.214)

Recipient’s retentive capacity

-.006 (-.06886)

.034 (.388)

-.141 (-1.746)

-.139 (-1.739)

-.121 (-1.462)

-.136 (-1.704)

Relationship .117 (1.578)

.086 (1.148)

.122 (1.621)

.083 (1.10)

.077 (1.003)

.078 (.1.020)

Firm dummies

Not significant

Not significant

Perspectives dummies

Not significant

R-square .261

.333 .448 .466 .504 .470

Adj. R-square

.231 .282 .414 .428 .446 .425

F

8.771 6.459 12.992 12.472 8.7156 10.430

Partial-F

Model 1 vs 2 3.03*

Model 1 vs 3 21.80**

Model 1 vs 4 16.50**

Model 3 vs 4 4.75*

Model 1 vs 5 6.86**

Model 4 vs 5 1.78

Model 1 vs 6 10.07**

Model 4 vs 6 .581

Valid N 182 182 154 154 154 154

Notes to Table t-values in parentheses ** Significant at 1% level, * Significant at 5% level, ^ Significant at .063.

39

Table 4: The total effect of the credibility of the source on the accuracy of the transfer

Causal Ambiguity set to: Total effect

βcredibility tot (t-value)

Two standard deviation below the mean. .572** (3.643)

One standard deviation below the mean. .427** (4.082)

The mean. .282** (3.615)

One standard deviation above the mean. .137 (1.367)

Two standard deviation above the mean. .008 (.052)

Notes to Table Values calculated using estimates of Model 4. ** Significant at 1% level.

40

Figure 2: Relationship between accuracy of the transfer and credibility of the source

for different levels of causal ambiguity

-7.509 -5.558 -3.608 -1.657 0.294 2.245 4.196 6.147 8.098 10.049 aboveCREDIBILITYCAUSAL AMBIGUITY

ACCURACY

-14-10

-6-2

26

10

-12-8-40481216

-16

-12

-8

-4

0

4

8

12

41

Reference

Adler, P. S. (1990). Shared Learning. Management Science, 36(8), 938-957.

Aiken, L. S., & West, S. G. (1991). Multiple regression: testing and interpreting

interactions. Newbury Park, Calif.: Sage Publications.

Allen, M., & Stiff, J. (1989). Testing three models for the sleeper effect. Western

Journal of Speech Communication, 53, 411-426.

Argote, L. (1999). Organizational learning: creating, retaining, and transferring

knowledge. Boston: Kluwer Academic.

Argote, L., & Ingram, P. (2000). Knowledge transfer: A basis for competitive

advantage in firms. Organizational Behavior and Human Decision Processes, 82, 150-

169.

Argote, L., & Ophir, R. (forthcoming). Learning within Organizations: Factors

affecting the creation, retention, and transfer of knowledge. In J. Baum (Ed.), Companion

to Organizations.

Armor, D. J. (1974). Theta Reliability and Factor Scaling. In H. L. Costner (Ed.),

Sociological Methodology 1973-1974 (pp. 17-50). San Francisco: Jossey-Bass.

Arrow, K. J. (1974). The limits of organization. New York, NY: Norton.

Attewell, P. (1992). Technology Diffusion and Organizational Learning: The

Case of Business Computing. Organization Science, 3(1), 1-19.

Axelrod, R. M., & Cohen, M. D. (1999). Harnessing complexity: organizational

implications of a scientific frontier. New York: Free Press.

Barley, S., & Tolbert, P. (1997). Institutionalization and structuration: studying

the links between action and institution. Organization Studies, 18(1), 93-117.

42

Belliveau, M., O'Reilly, C., & Wade, J. (1996). Social Capital at the top: effects

of social similarity and status on CEO compensation. Academy of Management Journal,

39(6), 1568-1593.

Benjamin, B., & Podolny, J. (1999). Status, quality and social order in the

California wine industry. Administrative Science Quarterly, 44(3), 563-589.

Berber, B. (1983). The logic and limits of trust.

Berger, C., & Kellerman, K. (1983). To ask or not to ask: Is that a question? In R.

Bostrom (Ed.), Communication yearbook 7 (pp. 342-368). Newbury Park, CA: Sage.

Bohn, R. E. (1994). Measuring and Managing Technological Knowledge. Sloan

Management Review(Fall), 61-73.

Boland, R. J., Jr., & Tenkasi, R. V. (1995). Perspective making and perspective

taking in communities of knowing. Organization Science, 6(4), 350-372.

Bradach, J. L. (1998). Franchise organizations. Boston: Harvard Business School

Press.

Brooks, F. P. (1995). The mythical man-month: essays on software engineering

(Anniversary ed.). Reading, Mass.: Addison-Wesley Pub. Co.

Capon, N., Hulbert, J. (1973). The sleeper effect: an awakening. Public Opinion

Quarterly, 37(3), 333-358.

Carley, K. (1991). A theory of group stability. American Sociological Review,

56(3), 331-354.

Carmines, E. G., & Zeller, R. A. (1979). Reliability and Validity Assessment.

Beverly Hills: SAGE.

43

Cialdini, R. B. (1984). Influence: the new psychology of modern persuasion. New

York.

Cohen, W. M., & Levinthal, D. (1990). Absorptive capacity: a new perspective on

learning and innovation. Administrative Science Quarterly, 35(1), 128-152.

Curall, S., & Judge, T. (1995). Measuring Trust Between Organizational

Boundary Role Persons. Organizational Behavior and Human Decision Processes, 64,

151-170.

Cyert, R. M., & March, J. G. (1963). A behavioral theory of the firm. Englewood

Cliffs, N.J.,: Prentice-Hall.

Dixon, N. M. (2000). Common knowledge: how companies thrive by sharing

what they know. Boston: Harvard Business School Press.

Eisenberg, E., & Phillips, S. (1991). Miscommunication in organizations. In N.

Coupland, Giles, H., Wiermann, J. (Ed.), 'Miscommunication' and problematic talk (pp.

244-258). Newbury park, CA: Sage.

Garvin, D. A., & Oliver, R. W. (2000). Learning in action. Hinsdale, IL: Audio-

Tech Business Book Summaries Inc.

Gerbing, D. W., & Anderson, J. C. (1988). An Updated Paradigm for Scale

Development Incorporating Unidimensionality and Its Assessment. Journal of Marketing

Research, XXV(May), 186-192.

Glaser, E. M., Abelson, H. H., & Garrison, K. N. (1983). Putting knowledge to

use : facilitating the diffusion of knowledge and the implementation of planned change

(1st ed.). San Francisco: Jossey-Bass.

44

Hammer, M., & Stanton, S. (1999). How process enterprises really work. Harvard

Business Review, 77(6), 108-118.

Hansen, M. (1999). The search-transfer problem: The role of weak ties in sharing

knowledge across organization subunits. Administrative Science Quarterly, 44, 82-111.

Helper, S., MacDuffie, J., & C., S. (1999). Pragmatic collaboration: advancing

knowledge while controlling opportunism. Working paper.

Henderson, R. M., & Clark, K. B. (1990). Architectural Innovation: The

Reconfiguration of Existing Product Technologies and the Failure of Established Firms.

Administrative Science Quarterly, 35(1), 9-30.

Hovland, C., Lumsdaine, A., & Sheffield, F. (1949). Experiments in mass

communication. New York.

Hovland, C., & Weiss, W. (1951). The influence of source credibility on

communication effectiveness. Public Opinion Quarterly, 15(4), 635-650.

Hovland, C. I., Janis, I. L., & Kelley, H. H. (1953). Communication and

persuasion. New Haven, CT: Yale University Press.

Huber, G. P. (1991). Organizational Learning: The Contributing Processes and the

Literatures. Organization Science, 2(1, February), 88-115.

Iansiti, M. (1998). Technology integration : making critical choices in a dynamic

world. Boston, Mass.: Harvard Business School Press.

Jensen, M., & Meckling, W. (1992). Specific and general knowledge, and

organizational structure. In L. Werin & H. Wijkander (Eds.), Contract Economics.

Oxford UK and Cambridge USA: Blackwell.

45

Kelman, H., & Hovland C. (1953). Reinstatement of the communication in

delayed measurement of opinion change. Journal of Abnormal and Social Psychology,

45, 327-335.

Knott, A. M. (1997). The organizational routines factor market paradox: Wharton

School of the University of Pennsylvania.

Kostova, T. (1999). Transnational transfer of strategic organizational practices: a

contextual perspective. Academy of Management Review, 24, 308-324.

Leibenstein, H. (1966). Allocative Efficiency vs. X-Efficiency. American

Economic Review, 56(June), 392-415.

Leonard-Barton, D. (1990). A dual methodology for case studies: Synergistic Use

of a Longitudinal Single Site with Replicated Multiple Sites. Organization Science, 1(3

(August)), 248-266.

Lewis, J., & Weigert, A. (1985). Trust as a social reality. Social Forces, 63, 967-

985.

Lippman, S. A., & Rumelt, R. P. (1982). Uncertain Imitability: an analysis of

interfirm differences in efficiency under competition. Bell Journal of Economics, 13,

418-438.

Love, J. F. (1995). McDonald's: behind the arches (Rev. ed.). New York: Bantam

Books.

Mayer, R., David, J., Schoorman, F. (1995). An integration model of

organizational trust. The Academy of Management Review, 20(3), 709-734.

46

McAllister, D. (1995). Affect- and cognition-based trust as foundation for

interpersonal cooperation in organizations. Academy of Management Journal, 38(1), 24-

60.

Nahapiet, P., & Ghoshal, S. (1998). Social capital, intellectual capital and the

organizational advantage. Academy of Management Review, 32(2), 242-257.

Nelson, R., & Winter, S. (1982). An Evolutionary Theory of Economic Change.

Cambridge: Belknap Press.

Nonaka, I. (1994). A Dynamic Theory of Organizational Knowledge Creation.

Organization Science, 5(1, February), 14-37.

Noteboom, N., Berger, H., Noorderhaven, N. (1997). Effects of trust and

governance on relational risk. Academy of Management Journal, 40(2), 308-338.

Nunnally, J. C. (1978). Psychometric Theory (Second Edition ed.). New York:

McGraw-Hill Book Company.

O'Dell, C. S., Grayson, C. J., & Essaides, N. (1998). If only we knew what we

know : the transfer of internal knowledge and best practice. New York: Free Press.

Perloff, R. M. (1993). The dynamics of persuasion. Hillsdale, New Jersey:

Lawrence Erlbaum Associates.

Perry, D. K. (1996). Theory and research in mass communication: contexts and

consequences. Mahwah, N.J.: L. Erlbaum Associates.

Petty, R. E., & Cacioppo, J. T. (1996). Attitudes and persuasion: classic and

contemporary approaches. Boulder, Colo.: WestviewPress.

Pfeffer, J., & Sutton, R. I. (2000). The knowing-doing gap: how smart companies

turn knowledge into action. Boston, Mass.: Harvard Business School Press.

47

Polanyi, M. (1966). Personal Knowledge: Towards a Post-Critical Philosophy.

Chicago. IL: Chicago University Press.

Rivkin, J. (forthcoming). Imitation of complex strategies. Management Science.

Rogers, E. M., & Kincaid, D. L. (1981). Communication networks : toward a new

paradigm for research. New York, London: Free Press ; Collier Macmillan.

Rommetveit, R. (1974). On message structure; a framework for the study of

language and communication. London, New York: Wiley.

Rosenberg, N. (1982). Inside the Black Box: Technology and Economics.

Cambridge: Cambridge University Press.

Rumelt, R. (1984). Toward a strategic theory of the firm. In R. Lamb (Ed.),

Competitive Strategic Management (pp. 556-570). Englewood Cliffs: Prentice Hall.

Seely Brown, J. (1997). Seeing differently. Boston: Harvard Business School

Press.

Selling to Argentina(12/05/1999, Issue Date). The New York Times.

Shannon, C. E., & Weaver, W. (1949). The Mathematical Theory of

Communication. Chicago, Illinois: University of Illinois Press.

Sternthal, B., Phillips, L., & Dholakia, R. (1978). The persuasive effect of source

credibility: a situational analysis. Public Opinion Quarterly, 42(3), 285-314.

Stohl, C. & Redding, C. (1987). Messages and message exchange process. In F.

Jablin, Putnam, L., Roberts, K., Porter, L. (Ed.), Handbook of organizational

communication: an interdisciplinary perspective (pp. 451-502). Beverly Hills, CA: Sage.

48

Szulanski, G. (1996). Exploring Internal Stickiness: Impediments to the Transfer

of Best Practice Within the Firm. Strategic Management Journal, 17(Winter Special

Issue), 27-43.

Szulanski, G. (2000). The process of knowledge transfer: A diachronic analysis of

stickiness. Organizational Behavior and Human Decision Processes, 82, 9-27.

Teece, D. (1977). Technology Transfer by Multinational Corporations: The

Resource Cost of Transferring Technological Know-How. Economic Journal(87), 242-

261.

Tsai, W., & Ghoshal, S. (1998). Social capital and value creation: the role of

intrafirm networks. Academy of Management Journal, 41(4), 464-476.

Walton, R. E. (1975). The Diffusion of New Work Structures: Explaining why

Success Didn't Take. Organizational Dynamics(Winter), 3-21.

Weick, K. E. (1979). The Social Psychology of Organizing. Reading, Mass.:

Addison Wesley.

Winter, S. G. (1987). Knowledge and Competence as Strategic Assets. In D.

Teece (Ed.), The Competitive Challenge-Strategies for Industrial Innovation and

Renewal (pp. 159-184). Cambridge, MA: Ballinger.

Winter, S. G. (1995). Four Rs of Profitability: Rents, Resources, Routines and

Replication. In C. A. Montgomery (Ed.), Resource-based and evolutionary theories of the

firm: towards a synthesis (pp. 147-178). Norwell, MA: Kluwer Academic Publishers.

Winter, S. G., & Szulanski, G. (forthcoming). Replication as Strategy.

Organization Science.

Xerox makes copies(13/07/1997, Issue Date). Financial Times.

49

Zaheer, A., McEvily, B., & Perrone, V. (1998). Does trust matter? Exploring the

effects of interorganizational and interpersonal trust on performance. Organization

Science, 9(2), 141-159.

Zander, U., & Kogut, B. (1995). Knowledge and the speed of the transfer and

imitation of organizational capabilities: An empirical test. Organization Science, 6(1

(Jan/Feb)), 76-92.

Zeller, R. A., & Carmines, E. G. (1980). Measurement in the social sciences: The

link between theory and data. Cambridge: Cambridge University Press.