You are on page 1of 13

09/12/21, 14:38 Lewis & al 2005

1. 1. Organization Science informs ® Vol. 16, No. 6, November–December 2005, pp. 581–598 doi
10.1287/orsc.1050.0143 issn 1047-7039 eissn 1526-5455 05 1606 0581 © 2005 INFORMS
Transactive Memory Systems, Learning, and Learning Transfer Kyle Lewis, Donald Lange, Lynette
Gillis Department of Management, University of Texas at Austin, 1 University Station B6300, Austin,
Texas 78712-0210 {kyle.lewis@mccombs.utexas.edu, donald.lange@phd.mccombs.utexas.edu,
lynette.gillis@phd.mccombs.utexas.edu} K nowledge embedded in a group’s structures and processes
can be leveraged to create sustainable advantage for orga- nizations. We propose that knowledge
embedded with a transactive memory system (TMS) helps groups apply prior learning to new tasks
and develop an abstract understanding of a problem domain, leading to sustained performance. We
present a framework for understanding TMSs as learning systems that affect group learning and
learning transfer, and we test the major outcomes of the framework in an empirical study. We found
that groups with a prior TMS and experience with two tasks in the same domain were more likely to
develop an abstract understanding of the principles relevant to the task domain—a critical factor for
learning transfer in general. We did not, however, find strong support for our contention that a TMS
facilitates learning transfer after experience with only a single task. Further examinations of our
findings showed that the extent to which members maintained expertise across tasks influenced the
degree of learning transfer, especially for groups whose members had previously developed a TMS
with another group. Our findings show that a TMS has broader benefits beyond the task for which it
first developed because a TMS affects members’ ability to apply prior learning and develop a
collective, abstract understanding of the task domain. More generally, our study demonstrates that
TMSs influence group learning and learning transfer. We discuss our study’s implications for practice
and for TMS and group learning theories. Key words: learning; transactive memory 1. Introduction
knowledge, coordinate members’ interactions more Group performance in contexts as varied as
product effectively, and perform at higher levels than do groups development, consulting, research and
development, and without a TMS (Liang et al. 1995; Moreland 1999; top management depends on the
collaborative processes Moreland et al. 1996, 1998; Moreland and Myaskovsky members use to
combine and integrate their unique 2000). knowledge. As members collaborate, they encode, inter-
These laboratory studies were instrumental in bring- pret, and recall information together, and in so
doing ing the TMS concept and its effects to the attention they create knowledge that becomes
embedded in of researchers and practitioners. An objective of these a group’s structures and processes
(Moreland 1999). studies was to show how a TMS enhances task perfor- Embedded knowledge is
difficult to recognize and mea- mance on the same task for which the TMS first devel- sure, but it is
also difficult to imitate, making it a key oped. However, that focus does not capture the fact that point
of leverage for organizations (Argote and Ingram most organizational workgroups perform a variety of
2000). The goal of this study is to explain how embed- tasks, either in the context of a single project,
or across ded knowledge can be leveraged to create sustained sequential streams of projects over time
(Waller et al. group performance. We examine knowledge embedded 2002). Knowing whether the
effects of a TMS persist in with a group’s transactive memory system (TMS), which dynamic task
environments is critical to understanding we argue influences group learning and performance the real
impact of TMSs in organizations. Several recent across several tasks. field studies (Austin 2003, Faraj
and Sproull 2000, Lewis A TMS is a collective memory system for encoding, 2003) provide early
evidence that TMSs may have long- storing, retrieving, and communicating group knowl- term value
in ongoing groups, but none of these stud- edge (Hollingshead 2001, Wegner 1986). TMSs develop ies
specifically examines whether the effects of a TMS over time as group members communicate,
observe each extend beyond the task for which it first developed. other’s actions, and come to rely on
one another to We offer an explanation for the positive effects of be responsible for different but
complementary areas of a TMS on group performance that generally has been expertise. Laboratory
studies of TMSs demonstrate that overlooked by group TMS research: TMSs help mem- in groups that
develop a TMS, members collectively bers learn, both individually and collectively. We con-
remember and apply a greater amount of task-critical ceptualize TMSs as learning systems that affect
group 581
2. 2. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer 582 Organization
Science 16(6), pp. 581–598, © 2005 INFORMS learning and learning transfer to produce sustained
predictions. We conclude by discussing the implications group performance. Drawing on TMS theory
(Wegner of these results and by offering recommendations for 1986, Wegner et al. 1985), and on
learning and learn- capitalizing on the value of TMSs in organizations. ing transfer theories (e.g.,
Reeves and Weisberg 1994, Singley and Anderson 1989), we develop a framework to explain: (1) how

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 1/13
09/12/21, 14:38 Lewis & al 2005

a TMS promotes cycles of learning 2. TMS-Learning Framework that produce not only knowledge that
is relevant for the Our TMS-learning framework shows that a TMS pro- current task, but also
transferable knowledge that can be duces cycles of learning with effects that extend beyond applied to
other tasks in the same domain, and (2) how a the task for which the TMS first developed, to other
TMS helps members collectively apply prior knowledge tasks that a group performs. We adopt
Argote’s (1999, to benefit performance in new task contexts. p. 131) definition of group learning as a
process wherein This learning perspective is useful for understanding members share their own
knowledge, generate new the value of TMSs in organizations, especially those knowledge, and
evaluate and combine this knowledge. organizations in which leveraging prior knowledge by Our use
of the term learning transfer is consistent with transferring learning across contexts or to different cus-
Singley and Anderson (1989) and Cormier and Hagman tomers is critical to firm performance (Argote
1999). (1987); learning transfer is defined as occurring when For such firms, leveraging experience
gained on one knowledge acquired in one situation affects learning or task to produce efficiencies and
higher-quality products performance in other situations. We refer to learning and services is critical to
both winning new business that occurs as a consequence of having developed a and increasing the
likelihood that future activities are TMS as TMS learning. We refer to the learning trans- profitable. fer
facilitated by a TMS as TMS-learning transfer. The In sections that follow, we present a framework for
TMS-learning framework is depicted in Figure 1. The understanding TMSs as learning systems that
includes framework describes the learning processes, knowledge predictions about the effects of TMSs
on group learn- outcomes, and transfer mechanisms for a group whose ing, learning transfer, and
performance, and we present members have no prior history together, as the group the results of an
empirical study designed to test our performs several tasks. Figure 1 TMS-Learning Framework
Activities: Perform Develop TMS Perform Task 1 Perform Task 2 subsequent tasks Learning Learning
Learning Cycle 1 Cycle 2 Cycle 3 (Section 2.1) (Section 2.2) (Section 2.3) 1 TMS 2 3 TMS Learning
TMS Learning TMS Learning Shared location information Refined location information Increasingly
abstract higher- More individual lower-order Additional shared higher-order order knowledge
information (expertise) information Understanding of underlying TMS processes for encoding,
Elaborated, contextualized principles of task domain storing, retrieving knowledge Shared higher-order
Patterns for communicating, information retrieving information Performance Mechanisms Transfer
Mechanisms Learning/Transfer Mechanisms Availability of task-relevant Transferable knowledge
Interactive cueing that facilitates expertise structures analogical encoding Retrieval, coordination,
Recognition, retrieval, and Shared higher-order knowledge utilization of expertise mapping of
transactive that facilitates collective knowledge induction Evidence of abstract learning/ Evidence of
TMS learning Evidence of TMS learning: transfer: transfer: Task 1 performance Strategic knowledge
of Task 2 performance (Tested by H1) task domain (Tested by H2 and H4) (Tested by H3 and H5)
3. 3. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer Organization Science
16(6), pp. 581–598, © 2005 INFORMS 583 Our TMS-learning framework applies to those groups
known as the location for software engineering infor- for which TMSs are especially helpful—groups
that mation. Similarly, Tim might come to be known as the perform complex, divisible tasks that
require consider- location for information about product sales and mar- able knowledge (Moreland et
al. 1996). Divisible tasks keting, and Mina might be known as the location for (Steiner 1972) allow
members to divide the cognitive customer support information. We return to this example labor for the
task and integrate knowledge possessed throughout this section. by different members. More generally,
our framework Also stored in the TMS structure are the specific facts applies to task-oriented
workgroups whose members and details, or lower-order information, that each mem- share
responsibility for producing group outcomes, and ber possesses about a particular topic (Wegner et al.
whose performance depends on coordinating and inte- 1985). In the product management group
example above, grating the various skills, knowledge, and activities lower-order information in the
TMS structure might of group members. Some examples of groups where include particulars about
recently implemented function- our framework applies include crews, product develop- ality and bug
fixes (lower-order information possessed ment teams, consulting and other project teams, research by
Joanne), data about product sales performance for the and development teams, self-managing teams,
and top last two quarters in each customer market (possessed by management teams. Our framework
does not apply Tim), and information relevant to complaints received to workgroups that have loosely
defined membership, from customers (possessed by Mina). no definable collaborative task, or low
coordination The location information and lower-order information and specialization needs. Groups
with one or more of that make up the initial TMS structure affect what and these attributes include
informal groups, interest groups, how much each member decides to learn. In particu- advice groups,

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 2/13
09/12/21, 14:38 Lewis & al 2005

ad hoc committees, and communities of lar, an understanding of others’ expertise affects a mem-
practice. ber’s choice to learn in an area other than those already The TMS-learning transfer effects
described by our associated with another member (Hollingshead 2001, framework are bounded by the
limits on learning trans- Wittenbaum et al. 1998). As a result, individual members fer in general.
Learning transfer is limited to settings come to specialize in different areas, and the group’s in which
the tasks are functionally similar (Singley and knowledge becomes more differentiated. Furthermore,
Anderson 1989)—that is, when the tasks share similar when members rely on other members to be
responsible task elements and when the strategies used to perform for information in their respective
specialty areas, each one task are applicable to the other. Furthermore, learn- member is free to develop
more knowledge in his or her ing transfer is possible only when an individual’s prior own specialty
area (Hollingshead 1998, Wegner et al. knowledge is in some way relevant to the transfer task. 1991).
In these ways, the initial TMS structure affects Thus, our framework applies to tasks that are function-
the content and extent of each member’s learning. ally similar, and to tasks for which members’
learning Transactive processes (Wegner et al. 1985), the second can be relevant. Finally, because our
specific interest is component of a TMS, are established during Learning in the effects of TMSs on
learning and performance, Cycle 1 through a group’s early interactions. Transactive other social or
attitudinal factors that may influence processes function through the interaction and commu- group
processes and performance are not explicitly inte- nication among members to encode, store, and
retrieve grated into the framework. The TMS-learning frame- knowledge relevant to the group’s task.
When members work is represented in terms of three TMS-learning first communicate, they rely on the
initial TMS structure cycles, each of which is described below. to establish transactive processes. For
example, mem- bers query others about information they presume to be 2.1. Learning Cycle 1: Initial
TMS Learning associated with each member and allocate new informa- The first learning cycle
produces a TMS, consist- tion encountered by the group to the appropriate mem- ing of both an initial
TMS structure—an organized ber. Using location information, a member can retrieve store of
knowledge that is contained within members’ information quickly and efficiently when needed for the
memories—and a set of transactive processes that mem- task, without having to possess that
information him or bers use to encode, store, and retrieve that knowledge herself. (Wegner et al. 1985).
A TMS begins to develop when Transactive processes, in turn, affect the TMS struc- group members
start to associate individual members ture. First, communicating helps members gain a more with
specific areas of knowledge. Information about accurate understanding of what members know (or do
members’ expertise is stored in the TMS structure as not know), and over the course of repeated
interactions, location information (Wegner et al. 1985).1 Take, for makes members’ location
information more similar and example, a group tasked with managing a software prod- more accurate.
Second, interacting can lead to what uct, composed of members Joanne, Tim, and Mina. Wegner refers
to as integrations of members’ knowledge Members might come to associate Joanne with informa-
(Wegner et al. 1985). Integrations result when mem- tion about software and design—Joanne would
then be bers discover links between members’ knowledge and
4. 4. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer 584 Organization
Science 16(6), pp. 581–598, © 2005 INFORMS create new knowledge that no member had previously
performance. TMS learning occurs, for example, while possessed. For example, suppose Joanne, Tim,
and Mina the group performs its task. When a group performs its are discussing lagging sales of their
software product. task, members encode and store new information about During their discussion,
members explore their respec- the task, other members, and other members’ knowl- tive sets of lower-
order information (i.e., details about edge (Brandon and Hollingshead 2004). Learning dur- software
engineering, sales, and customers) that are rele- ing task performance comprises Learning Cycle 2 of
our vant to information about sales performance. Integrating TMS-learning framework. Learning by
doing is espe- their views may lead the members to a group-generated cially important when the
context is integral to perfor- solution—for example, the recognition that customers mance (Argote
1999), for example, when learning occurs from a particular market segment have been complain- in the
presence of other group members. Learning by ing about product functionality alterations in the
current doing affects both parts of a TMS—the TMS structure product release. As a consequence of
their collective and the set of transactive processes that operate on that discovery, members integrate
relevant details about soft- structure. ware engineering, sales, and customers, and encode that
Performing a group task affects the information information into their TMS. Integrated information is
encoded and stored in the TMS structure in at least three encoded as shared higher-order information,
defined as ways. First, seeing what works and does not work and the “topic, theme, or gist” of some set
of lower-order observing how individuals perform individually and col- information (Wegner et al.

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 3/13
09/12/21, 14:38 Lewis & al 2005

1985, p. 264). In the prod- lectively may cause revisions or refinements to mem- uct management
group example, a shared higher-order bers’ understanding of who knows what (i.e., location topic,
“determinants of declining sales,” might repre- information). Second, as members share and discuss
sent members’ newly discovered knowledge about the information, they may discover new ways that
individu- causes of sales problems in a particular market segment. als’ lower-order information can be
integrated as shared The shared higher-order topic points to specific lower- higher-order information
(Wegner et al. 1985). Third, order information about product functionality, sales fig- observing
interactions and taking part in discussions ures, and complaining customers that all members can
during which knowledge is exchanged helps members retrieve. Thus, relying on an initial TMS
structure and develop a more elaborated, contextualized understand- set of transactive processes can
produce new collective ing of their own knowledge. A semantically elaborated knowledge that is
stored as higher-order information in memory (cf. Anderson and Reder 1979, Wegner et al. the TMS.
1985) results when a member draws inferences about an Articulating the processes involved in
developing a item of information and considers its meaning in rela- TMS reveals the links between
TMSs and individual tion to other information. Observing other members and and collective learning.
By the end of Learning Cycle 1, taking part in group discussions help a member build members will
have learned who possesses what exper- elaborated knowledge structures that represent how a tise,
developed new member-level knowledge in the form member’s own knowledge fits with and builds on
other of specialized expertise, and developed new collective members’ task-related knowledge. Marks
et al. (2002, knowledge in the form of shared higher-order informa- p. 4) refer to this type of
knowledge as “interrole knowl- tion. Ultimately, the effects of Learning Cycle 1 are evi- edge,” an
understanding of the content of and interre- dent in the group’s performance on the task for which
lationships among members’ knowledge. Their research the TMS developed. Past TMS research
demonstrates found that when members were aware of one another’s that groups perform better when
they develop and rely jobs, roles, and expertise, they developed shared concep- on established TMS
structures and processes (e.g., Liang tualizations of interrole knowledge, which in turn pos- et al.
1995). Learning Cycle 1 produces those struc- itively influenced group coordination and performance.
tures and processes, as well as TMS learning, to make a We expect that by simultaneously refining
location infor- greater amount of relevant knowledge available for task mation and facilitating a
contextualized understanding processing. To replicate past research findings and lay of members’
knowledge, learning by doing will produce the foundation for predictions about subsequent learning
shared conceptualizations of interrole knowledge. cycles, we hypothesize that: In addition to affecting
the TMS structure, learning by Hypothesis 1. Groups with a TMS (groups that have doing affects
transactive processes. Performing the task completed Learning Cycle 1) will demonstrate higher
provides feedback about the efficacy of interactions for task performance than will groups with no
TMS. retrieving and sharing information and helps set patterns for future interaction. Research on
habitual routines in 2.2. Learning Cycle 2: Learning by Doing groups suggests that patterned
interactions do develop in Our learning systems perspective suggests that TMS groups, and that they
develop very quickly (cf. Gersick learning continues after a TMS has developed, and that and
Hackman 1990, Hackman and Morris 1975). Even this learning has effects that extend beyond initial
task while a group performs a single task, there are likely
5. 5. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer Organization Science
16(6), pp. 581–598, © 2005 INFORMS 585 to be many opportunities to execute patterns of com- to
learning transfer described above. We contend that munication and elicitation (Rulke and Rau 2000).
For Learning Cycle 2 influences the degree to which mem- example, a member who repeatedly queries
others dur- bers transfer their prior learning because: (1) learning by ing task performance might
initiate a pattern of interac- doing helps create abstract knowledge structures stored tion characterized
by members volunteering information within the TMS structure, and (2) utilizing an estab- only after
being asked. A different pattern might emerge lished TMS both helps members collectively recognize
in response to a member who withholds information functional similarities and underlying principles
com- from his/her expertise area—others might become more mon to tasks, and helps members
retrieve and map aggressive in asking for information from that mem- prior knowledge and problem-
solving strategies to the ber, or they might forego interacting with that member new task. altogether.
2.2.2. Transferable Knowledge Structures. Learning In sum, learning by doing during task performance
Cycle 2 produces three types of transferable knowledge helps members refine location information and
develop that can be relevant in new task contexts. First, indi- an elaborated, contextualized
understanding of how vidual knowledge produced from learning by doing is their own task-relevant
knowledge relates to others’ elaborated and contextualized as a result of members task-relevant

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 4/13
09/12/21, 14:38 Lewis & al 2005

knowledge. Learning Cycle 2 also helps learning more about their own specializations and more
establish patterns for communicating and retrieving about how their knowledge relates to others’
knowl- information, which reduce uncertainty about how group edge and to the task. Individuals with
more elaborate, interactions ought to proceed (Gersick and Hackman abstract representations of task-
relevant knowledge are 1990). Learning Cycle 2 has the effect of making groups more likely to
identify when tasks are indeed function- more effective and efficient at performing the same task ally
similar and recognize how their prior knowledge for which the TMS initially developed. We argue that
applies in a novel context. Second, shared location infor- Learning Cycle 2 also affects tasks other than
the task mation refined during Learning Cycle 2 will remain for which a TMS initially developed, by
facilitating useful in a new task context as long as membership transfer of learning across tasks. and
expertise specializations remain somewhat stable. 2.2.1. TMSs and Learning Transfer. Individuals are
Finally, shared higher-order information is likely to often unable to transfer learning from one situation
to remain relevant when members recognize that tasks another because they fail to notice the
functional simi- have underlying principles and elements in common. In larities and underlying
principles common across tasks our product management group example, market sales (Singley and
Anderson 1989). Research on individual- trends, software functionality, and customer feedback are
level learning transfer shows that whether a person key task elements relevant to the group’s initial
task. The who has acquired knowledge in one situation applies group members integrated these task
elements under the it to other situations depends largely upon that per- higher-order concept
“determinants of declining sales.” son’s mental representation of the knowledge (Reeves If the group
recognizes that these same task elements and Weisberg 1994). Individuals who have developed an also
apply to a different product context, they will be abstract mental representation of the problem domain
able to draw on the same shared higher-order concept are more likely to recognize when and how prior
learn- to retrieve specific sales, software engineering, and cus- ing will apply to a novel task. Without
an abstract tomer support information to diagnose sales problems understanding of the underlying
principles relevant to a with the new product. Thus, knowledge produced by domain, however,
individuals are more likely to focus TMS-learning cycles is likely to be useful across tasks on the
superficial features of a task and fail to recognize that share similar elements and underlying principles.
that previously learned procedures and strategies could 2.2.3. Recognition, Retrieval, and Mapping of
be used to solve the problem. Given a group bias for Transactive Knowledge. Utilizing an established
TMS discussing shared information (Stasser and Titus 1985), affects the degree to which members
actually transfer the more members that fail to notice functional simi- prior knowledge by helping
members recognize task larities across problems, the more likely it is that the similarities and by
facilitating retrieval and mapping of group will favor discussion of the superficial features prior
learning to a new task. Transactive retrieval pro- over the underlying principles relevant to both
problems. cesses refined in Learning Cycle 2 increase the like- In sum, the way that prior knowledge is
organized affects lihood that members recognize how prior lower-order two preconditions of learning
transfer: (1) recognition of and higher-order knowledge apply to the task. Transac- functional
similarities across problems, and (2) mapping tive retrieval processes are characterized by interactive
of prior knowledge and learned problem-solving strate- cueing of members’ recall, a sequential,
iterative pro- gies to the new problem (Bassok 1990). cess in which partners cue information from the
other’s We argue that having developed and utilized a TMS memory (Hollingshead 1998, Wegner et al.
1985). In a on a task helps groups overcome the impediments group context, interactive cueing
involves one member
6. 6. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer 586 Organization
Science 16(6), pp. 581–598, © 2005 INFORMS cueing recall of another member’s knowledge, which
in 2.3. Learning Cycle 3 and Beyond: Generalizing to turn helps members recall different information
relevant the Task Domain to the group’s task. Interactive cueing that occurs in a TMS learning occurs
not only as members perform novel task context can help members retrieve knowledge their initial
task, but also as they perform a subse- that they would not otherwise recognize as relevant to quent
(transfer) task. In particular, performing a second the new task. For example, suppose one member
recog- task in the same domain creates increasingly abstract nizes that the initial and new task share a
common ele- knowledge about the principles underlying both tasks. ment. When that member uses a
commonly understood Research on analogical encoding (e.g., Gentner et al. label for previously
encoded information related to the 2003, Loewenstein et al. 1999) shows that comparing task element,
it triggers associations in other members’ two different but analogous problems helps individuals
minds. Given this cue, other members can locate and understand the underlying structure common to
both. In retrieve detailed lower-order information relevant to the a recent study, Gentner et al. (2003)

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 5/13
09/12/21, 14:38 Lewis & al 2005

found that indi- new task. By relating the new task context to the con- viduals prompted by researchers
to compare two differ- text in which information was first encoded, interactive ent negotiation
problems not only recognized common cueing processes improve the chances that what mem- task
features, but also developed an abstract understand- bers learned on the initial task will transfer
(Singley and ing of the underlying principles of the problem domain. Anderson 1989). By promoting
the abstraction of concepts related to the Learning Cycle 2 also establishes patterns of inter- domain,
analogical encoding helps individuals recall and action that are likely to persist in a new task context,
transfer prior knowledge across tasks (Gentner et al. especially when members perceive that the initial
and 2003, p. 394). new tasks are functionally similar. Members are likely Related conclusions can be
drawn from the research to execute the same patterns established on an initial on collective induction
(e.g., Laughlin 1999, Laughlin task, even without explicitly discussing their applicabil- ity or efficacy
(Feldman and Rafaeli 2002, Gersick and and Bonner 1999, Laughlin and Hollingshead 1995).
Hackman 1990, Louis and Sutton 1991). Interaction pat- Collective induction refers to the processes
by which a terns thought to have been successful in the past should group infers some general principle
or rule from con- remain useful for guiding efficient transactive processes crete examples of that
principle. The process of col- and helping members retrieve and share task-relevant lective induction
involves members observing patterns, knowledge. Having developed a shared conceptualiza-
regularities, and relationships across tasks in a domain, tion of interrole knowledge is likely to further
reinforce proposing and evaluating hypotheses to account for these interaction patterns, because the
shared concep- those patterns, and eventually converging on the cor- tualizations contain knowledge
about the sequence of rect principle or rule that underlies the domain tasks interdependent activities
needed to accomplish a task (Laughlin 1999). Collective induction research shows (Marks et al. 2002).
Routinized interactions can be prob- that groups tend to be good at inferring general princi- lematic if
groups execute the same patterns in inappro- ples from several examples, in part because groups share
priate situations and without full deliberation about their a conceptual system of ideas that helps
members realize probable effects (Gersick and Hackman 1990, Louis when a proposed solution is
correct (Laughlin 1999). and Sutton 1991). When different tasks are functionally We propose two
reasons that having developed a TMS similar, however, the transfer of established interaction in the
past will facilitate both analogical encoding and patterns is not likely to be detrimental. On the con-
collective induction. First, a prior TMS is likely to facil- trary, we expect that when members recognize
that tasks itate analogical encoding because the interactive cue- share common elements and
underlying principles, and ing typical of established TMS processes helps members when they draw on
established interaction patterns, they recognize functional similarities across tasks and, in so will be
more likely to retrieve knowledge critical to the doing, prompts members to make comparisons across
new task. tasks. Prompting individuals to compare across prob- The above arguments suggest that
learning by doing lems is known to improve analogical encoding (Gentner in Learning Cycle 2 helps
members leverage their TMS et al. 2003). Second, prior TMS-learning cycles pro- learning and
transfer what they have learned to a new duce shared higher-order information, an abstract form task
context. The effects of prior learning and learn- of knowledge that links each member’s knowledge to
ing transfer are properly measured by performance on a specific knowledge about the task. When the
same transfer task (cf. Singley and Anderson 1989). Thus, we members experience a second task
together, their higher- hypothesize that: order knowledge becomes further elaborated, as mem-
Hypothesis 2. Groups that have previously devel- bers form associations between what they and others
oped and utilized a TMS on one task will perform better know and between the initial task and the
subsequent on a subsequent, similar task than will groups with no task. These higher-order knowledge
structures are the prior TMS. very types of shared conceptualizations that help groups
7. 7. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer Organization Science
16(6), pp. 581–598, © 2005 INFORMS 587 collectively induce general principles underlying tasks in
the experimenters explicitly assigned responsibility for the domain (Laughlin 1999). learning in
specific areas of expertise. Imposing a new In sum, a prior TMS leverages prior learning, mak-
division of cognitive labor on couples that had already ing it more likely that members encode more
abstract developed an implicit TMS structure seemed to impede knowledge relevant to the domain and
more likely that how much information they were able to learn and later members recognize when the
same general principles recall. apply across different tasks. Therefore, we hypothesize An interesting
laboratory study by Baumann (2001) that, given experience on tasks in the same domain: suggests that
the impact of a disruption to the TMS structure may depend on the extent of that disrup- Hypothesis 3.
Groups with a prior TMS will be more tion. Baumann found that when groups were constructed likely
to demonstrate abstract, generalized knowledge to preserve expertise categories and the distribution of

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 6/13
09/12/21, 14:38 Lewis & al 2005

about the underlying principles relevant to the domain expertise across task trials, groups whose
membership than will groups that have never developed a TMS. had changed after developing a TMS
were able to quickly learn which members possessed what expertise and con- 3. Interference in TMS
Learning and struct a new TMS. Because the task as well as the divi- Transfer sion of cognitive labor
remained constant across trials, Thus far, our arguments have assumed that group mem- members may
have been able to apply some prior TMS bers will have full access to the two components of a learning
even though group membership had changed. TMS—a TMS structure and a set of transactive pro-
Together, the above studies suggest that significant cesses. We argued that once a group has developed
an disruptions to the TMS structure, defined as changes efficient TMS structure and effective TMS
processes, that affect location information and redefine the division members are better able to learn
and transfer task- of cognitive labor, are likely to affect members’ learn- relevant information. Suppose
instead that a group’s ing and, consequently, learning transfer across tasks. We TMS structure is
relatively inefficient, as might be the note that when the task has also changed, it is likely that case
when group members do not possess common loca- even minor disruptions to the TMS structure will
affect members’ ability to map prior knowledge. Therefore, we tion information. In that case, TMS
processes would also hypothesize that: be relatively inefficient. Without a shared understanding of who
is responsible for what, new information encoun- Hypothesis 4. Groups that experience a disruption
tered by the group might be encoded by more mem- to an established TMS structure will perform
worse on a bers than necessary, or might never be encoded (Wegner transfer task than will groups that
have never developed 1986). Discovering which members possess what exper- a TMS. tise and
deciding on the appropriate allocation of new Disruptions to an existing TMS structure are also
information takes time, reducing the efficiency of TMS likely to interfere with members’ higher-order
learning encoding processes. Furthermore, until members under- about the task and task domain.
Members are less apt stand which members possess what expertise, they will to develop contextualized
knowledge about how their be less efficient at retrieving information and commu- own knowledge fits
with other members’ task-relevant nicating about task elements that had previously been knowledge,
because members will no longer be certain organized as shared higher-order information. Members
what knowledge other members actually possess. Fur- must again develop shared higher-order
concepts before thermore, because disruptions prevent or delay access they can efficiently retrieve and
coordinate what mem- to lower-order information, integrations that produce bers know. Changes to the
TMS structure, if severe shared higher-order information will occur slowly, or not enough, could also
cause groups to abandon their habit- at all. Without contextualized knowledge at the individ- ual
routines and force members to learn new patterns of ual level, and without shared conceptual
knowledge at interaction (Gersick and Hackman 1990). Disruptions to the group level, members are
unlikely to identify and the TMS structure, then, should interfere with members’ recognize when tasks
are functionally similar, and are learning and learning transfer. unlikely to be able to abstract common
principles under- Indeed, TMS research examining TMS encoding and lying the tasks. Therefore, we
predict that: retrieval processes demonstrates that, when an existing TMS structure is changed, a prior
TMS does interfere Hypothesis 5. Groups that experience a disruption with learning and reduces group
performance. In a study to an established TMS structure will be less likely to comparing the encoding
and retrieval processes of inti- demonstrate abstract, generalized knowledge about the mate couples
with those of stranger couples, Wegner underlying principles relevant to the task domain than et al.
(1991) found that intimate couples (who had will groups that have never developed a TMS.
presumably already developed an implicit structure for We have argued that TMSs are learning
systems that learning and recalling information) performed worse produce transferable knowledge,
help members recog- than stranger couples on a knowledge recall test when nize the functional
similarities and common principles
8. 8. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer 588 Organization
Science 16(6), pp. 581–598, © 2005 INFORMS across tasks, and facilitate retrieval and mapping of
prior which groups could develop a TMS. Second, the initial knowledge across tasks. We further
argued that, given and subsequent tasks had to be different from each other additional task experience,
groups with a TMS are more in terms of superficial features, and yet be functionally likely to develop
an abstract understanding of the under- similar, such that they had task elements in common and lying
principles of the task domain. If a TMS structure such that the strategies used to complete the learning
is significantly disrupted, however, having developed a task were appropriate to the other tasks. For
prior learn- TMS in the past is expected to reduce members’ learning ing to transfer, members would
have to recognize the transfer and hamper the development of abstract knowl- common features of the
tasks and ignore the superficial edge about the task domain. We test our hypotheses differences

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 7/13
09/12/21, 14:38 Lewis & al 2005

(Singley and Anderson 1989). in an empirical study that is described next. Our study For the learning
task we chose an off-the-shelf elec- examines the key outcomes of our TMS-learning frame- tronics
assembly kit—a telephone kit—that is compa- work: (1) whether a TMS developed on one task facil-
rable in complexity to electronics-oriented kits used in itates learning transfer across tasks; (2) whether
a prior past TMS research (e.g., Liang et al. 1995, Moreland and TMS, combined with multiple task
experiences, helps Myaskovsky 2000), so we were confident that partici- groups develop abstract,
generalized knowledge about pants would be able to develop TMSs on the task. The the underlying
principles relevant to the task domain; telephone kit was composed of 47 parts, including circuit and
(3) the conditions under which having developed a boards, wires, screws, and buttons. We chose
another TMS in the past hampers, rather than facilitates, learning off-the-shelf electronics assembly kit
—a personal stereo and transfer. tape player—for our transfer task. The personal stereo kit was
composed of 31 parts, including earphones, 4. Methods play/stop buttons, tape guides, circuit board,
screws, We tested our hypotheses by conducting a longitudi- wires, and battery connections. We used a
third kit— nal experiment in which three-person groups performed an electronic stapler—as the basis
for testing whether electronic assembly tasks. In a series of three sessions, groups with experience with
the learning and transfer each separated by one week, groups were trained on an tasks developed
abstract, generalized knowledge about assembly task (Week 1), performed that task (Week 2), the
underlying principles relevant to the task domain. and then a week later performed a different assembly
The electronic stapler kit was composed of 45 parts, task and a knowledge task (Week 3). The
participants, including a circuit board, wires, buttons, a small motor, tasks and procedures, design,
manipulations, and mea- and screws. sures are described next. To confirm that these three tasks
differed superficially but were indeed functionally similar, we analyzed the 4.1. Participants tasks using
frameworks proposed in the literature (e.g., Participants were undergraduate students from a large
McGrath 1984, Steiner 1972, Wageman 1995). The tasks southwestern U.S. university who earned
extra credit differ in terms of superficial features in two ways. First, toward their course grades by
taking part in the study. while the kits have many common parts (e.g., screws, We began with 434
students randomly assigned to con- brackets, circuit boards, and wires), some parts differ ditions in
groups of three. Over the course of the three- across kits. For instance, the telephone assembly has no
week study, 47 participants were lost to attrition and motorized parts, while the personal stereo and
stapler 87 were excused because a member of their group did have gears and belts that regulate
operation. Second, the not show up to one of the three sessions (if even a sin- tasks differ in terms of
what Miller (1973, 1974) called gle member was absent, the group to which that mem- goal image, or
a mental picture of the task’s end state. ber was assigned became unusable). Excused students
Participants likely have a different goal image for each received full credit for participating. Three
hundred par- of the three products (telephone, personal stereo, and ticipants in 100 groups completed
the entire study. Attri- stapler) and are consequently likely to have preconcep- tion did not differ across
conditions, nor were there tualizations of the tasks that make them seem somewhat any demographic
differences among students who were dissimilar (Fleishman and Quaintance 1984). excused, dropped
out, or finished the experiment. The In spite of their surface differences, the three tasks final sample
averaged 21 years of age and was approx- are functionally similar in several ways. First, all three
imately 50% males and 50% females. Of the partici- tasks can be categorized as divisible rather than
unitary pants, 59% were Caucasian, 20% were Asian, 8% were (Steiner 1966, 1972), meaning that
each task has the Hispanic, 5% were African-American, and 8% did not potential to be accomplished
through a genuine division report their ethnicity. of labor. Second, all three tasks lend themselves to
inter- dependent rather than independent work by group mem- 4.2. Tasks and Materials bers (Wageman
1995). Indeed, when we reviewed pretest We selected tasks for this study with two criteria in
videotapes of groups completing the tasks, we observed mind. First, the initial (learning) task had to be
one on members working interdependently, rather than working
9. 9. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer Organization Science
16(6), pp. 581–598, © 2005 INFORMS 589 in isolation or relying on a single member to apply his full
factorial be carried out. We chose this design for or her expertise to accomplish the entire assembly.
The three reasons, described next. degree of interdependence needed to successfully com- First, to test
the effects of a TMS on transfer and plete the tasks requires the type of interaction that typi- learning,
we had to create a TMS in some groups and fies a TMS (Hollingshead 2001). not in others. Comparing
these groups on the learning Other functional similarities among the tasks can task would allow us to
confirm the effects of Learn- be assessed using McGrath’s (1984) task classifica- ing Cycle 1
(Hypothesis 1), and comparing these groups tion scheme. McGrath organized tasks into six types on
later tasks would allow us to test our predictions along two dimensions, cooperative versus conflictual,

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 8/13
09/12/21, 14:38 Lewis & al 2005

about the extent to which a TMS does or does not facil- and conceptual versus behavioral. All three of
our tasks itate learning transfer (Hypotheses 2 and 4) and con- correspond to the intersection of the
cooperative and tribute to the development of abstract domain knowledge behavioral categorizations,
and map well onto the two (Hypotheses 3 and 5). We created a TMS with a training tasks that McGrath
associated with that intersection— manipulation described in Moreland et al. (1996, 1998). planning
and performance/psychomotor activities. We trained all members on the learning task in groups of
three, expecting that group training would help TMSs Planning. Groups performing each of the three
tasks to develop (Moreland 1999), and then we “disabled” might benefit from devising an action-
oriented plan for the TMSs of half of the trained groups by reassigning completing the assembly of the
electronics kit. Assem- members to new groups before they performed the learn- bly planning that
applies to all three kits might include ing task. Groups remaining intact after training would
formulating a rough theory of operation, defining what have full access to their training TMS, while
groups suboperations comprise the whole theory of operation, whose members were reassigned after
training would identifying the assembly actions required to achieve sub- no longer have access to
elements of the TMS struc- operations, outlining the necessary sequence of actions, ture and processes
that were previously associated with considering how to organize parts, and planning the other
members. This manipulation created two compari- coordination of member actions and interactions.
son groups—those with a training TMS and those with- Performance/Psychomotor Activities. All three
kits out access to the TMS developed during training (for have small parts that can be described as
fasteners, but- simplicity, we refer to this comparison group as having tons, circuit boards and wiring,
moving parts, or station- “no prior TMS”). ary parts. The assemblies require similar ordered actions, A
second reason for our design choice is that we had such as placing and fastening an array of smaller
sta- to isolate the effects of the training TMS on learning tionary parts onto larger parts, placing small
moving and learning transfer across two subsequent tasks. This parts onto larger parts so that they
move and interact to meant that we had to control for the possibility that perform their designed
functions, placing and fastening groups might develop a useful TMS while performing parts that
secure the moving-parts assemblies, fastening one of the tasks, even if they had not developed a TMS
subassemblies to bases, and snapping and fastening cas- during training (Baumann 2001, Hollingshead
1998). ings. All the kits involve assembly with the use of small To demonstrate that transfer and
subsequent learning screwdrivers and screws. effects were caused by the training TMS, and not by All
told, the three tasks have functional similarities task learning or a newly developed TMS, we would
that permit us to expect that expertise gained on one task have to control for the confounding effects of
task and will be applicable to the other tasks. At the same time, group experience. Task experience was
controlled for by the task content of the kits differs substantially enough having all groups perform the
same tasks, in the same to allow us to infer whether learning transferred across sequence. Group
experience was controlled for by reas- tasks. signing members to new groups before they performed a
subsequent task. Reassignment should render any TMS 4.3. Design and Manipulations that developed
in a prior group less relevant to subse- To determine whether a TMS that developed on the quent task
performance (Moreland et al. 1996, 1998), learning task influenced learning transfer and the subse- and
keeping membership intact should provide a group quent development of domain knowledge, we
designed with the full advantages of its training TMS. Thus, we an experiment with three tasks
(learning task, trans- controlled for group experience on the telephone task by fer task, knowledge
task) performed in sequence. Each reassigning half of the members to new groups before task was
performed by two types of groups, created by they performed the stereo (transfer) task, and we con-
either reassigning members to new groups before they trolled for group experience on the stereo task
by reas- performed a task, or by keeping members in the same signing half of the members to new
groups prior to groups in which they completed the previous task, result- performing the stapler
(knowledge) task. Reassignment ing in a 2 × 2 × 2 factorial design. We intended to do enabled us to
isolate the effects of having developed a series of planned comparisons, which required that the a prior
TMS on learning transfer (Hypothesis 2) and
10. 10. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer 590 Organization
Science 16(6), pp. 581–598, © 2005 INFORMS on the subsequent development of abstract knowledge
participants were only allowed to watch and listen to the (Hypothesis 3). demonstration. The
experimenters then directed groups Finally, our design had to allow us to test whether to their separate
work areas and issued each group a disruptions to an established TMS structure impeded telephone
assembly kit on which to practice for approx- learning transfer (Hypothesis 4) and the development of
imately 30 minutes. Members were free to discuss the abstract knowledge (Hypothesis 5). This
requirement is task within their own groups and to call the trainer over satisfied by our 2 × 2 × 2 design

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 9/13
09/12/21, 14:38 Lewis & al 2005

because the reassign- to their work area for private questions. At the end of ment manipulation is itself
a significant disruption to the practice time the experimenters dismissed the partic- the established
TMS structure. Reassigning members to ipants and reminded them to return one week later. new
groups effectively disrupts any preexisting cogni- The same participants returned one week later to
per- tive division of labor and makes members’ prior location form the learning task (telephone
assembly) under timed information less relevant for task processing. Reassign- conditions. At the start
of this second experimental ses- ment allows us to compare groups that have experienced sion,
participants either remained in their training groups a disruption to a prior TMS structure with those
groups or were randomly reassigned to new groups in the man- whose full TMS structure remained
intact. Thus, our ner described earlier. The experimenters issued each longitudinal 2 × 2 × 2 design: (1)
produces a TMS in group an unassembled telephone kit identical to the one some groups but not
others, (2) controls for alterna- they had completed the week before and directed groups tive influences
on transfer and learning (i.e., by control- to complete the assembly to the best of their ability in a ling
for group and task experience) while isolating the maximum of 30 minutes. Upon finishing the group
task, effects of TMS learning, and (3) maintains or disrupts each participant completed a questionnaire
with items a group’s established TMS structure—all of which are about group cohesiveness and
motivation (Liang et al. necessary for testing our hypotheses about TMS learning 1995), items
measuring the extent to which a training and transfer. TMS had indeed developed (Lewis 2003), and a
fill- in-the-blank question asking members to describe each 4.4. Procedures member’s expertise.
Participants were reminded to return We conducted each of the three experimental sessions in one
week later and were dismissed. a large classroom. The room was equipped with tables In the first part
of the third and final session, groups spaced far enough apart that groups could not see other performed
the transfer task (stereo assembly). Partic- groups’ materials or hear their conversations. At the start
ipants either remained in their learning-task group or of the experiment, participants were told that the
study were randomly assigned to new groups before starting was being done to investigate how groups
work together. the transfer task. The experimenters gave each group a They were told that there would
be three sessions run preassembled personal stereo kit and allowed them one in three consecutive
weeks. Participants were not told minute to examine the assembly and components. Par- that some of
them would be reassigned to new groups in ticipants were not allowed to disassemble the stereo kit
subsequent sessions because group members who expect or alter the preassembled kit in any way. This
was the turnover may decide not to rely on transactive memory. first opportunity that the participants
had to examine Groups were randomly assigned to an experimental con- the assembled personal stereo
kit. The groups received dition, and participants were randomly assigned to their no other training or
instruction. After the one-minute initial groups by blindly drawing a group number from examination
period, the experimenters collected the pre- a hat. In subsequent sessions, the member composition
assembled kits, issued each group an unassembled per- of groups in the reassignment conditions was
also deter- sonal stereo kit, and instructed each group to assemble mined by random assignment,
constrained such that no the kit to the best of their ability in a maximum of members were regrouped
with people they had worked 30 minutes. Upon completing the transfer task, partic- with before. ipants
were given a questionnaire asking them to once During the first session, groups received training on
again describe each member’s expertise (the question- the learning task (telephone assembly). First,
partici- naire also included items about task difficulty and group pants completed a short survey asking
for demographic processes that were not analyzed for this study). information and previous experience
with electronics The second part of Session 3 was devoted to test- kit assembly. A graduate assistant
helping us with the ing the extent to which groups had developed a gen- experiment then performed a
15–20 minute task demon- eralized understanding of the underlying principles of stration in full view
of all of the groups. The trainer the electronics kit assembly domain (abstract knowledge used a script
and a demonstration model to describe the test). To control for the possibility that groups devel- step-
by-step procedures for constructing the telephone oped a TMS while performing the transfer task, we
once assembly kit. The script was used to ensure that all par- again reassigned half of the participants
to new groups ticipants received identical training instructions. Talk- before asking groups to complete
the knowledge test. ing among participants and note taking was prohibited; The abstract knowledge
test consisted of examining, but
11. 11. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer Organization Science
16(6), pp. 581–598, © 2005 INFORMS 591 not assembling, another electronics kit (electronic sta-
problem. Our measure is similar to explanation-based pler). Experimenters issued each group an
assembled measures used in past learning-transfer research to test electronic stapler and a survey that
asked the group to for the development of abstract knowledge (e.g., Gentner articulate a strategy for

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 10/13
09/12/21, 14:38 Lewis & al 2005

assembling the stapler. To com- et al. 2003). Groups that recognize the ways in which plete this task
well, participants would have to abstract the learning and transfer tasks are functionally simi- the
underlying principles common to both of the prior lar should be better able to apply abstract principles
of two tasks and recognize the general task strategies that the task domain to decompose the new
problem and apply to all three tasks. After the surveys were com- plan out a solution for assembling
the stapler (Singley pleted, experimenters thanked the participants and then and Anderson 1989).
Abstract knowledge scores were dismissed them. To avoid the possibility that participants obtained by
rating a group’s response to the question who had recently completed the experiment would dis- “What
sort of strategy would your group develop and cuss it with classmates who had not yet completed the
utilize if you wanted to conduct the assembly of this three-session series, we did not immediately
debrief par- kit efficiently and accurately? (How would you go about ticipants about the study
hypotheses. Instead, we invited doing it?).” Two of the coauthors, blind to the group’s participants to
debriefing sessions held at the end of the experimental condition, rated each group’s written strat-
semester. A small percentage of the participants (less egy description. We used a five-point scale to rate
the than 5%) ultimately chose to attend these sessions.2 quality of each group’s strategy description.
The scale was anchored at 1 = trivial and 5 = integrative, where 4.5. Measures “trivial” was interpreted
as relating to strategy descrip- tions that entailed the superficial, surface elements of Learning-Task
Performance. Performance on the the task—elements that would differ between the sta- learning task
(telephone assembly) was measured by pler task and the other two tasks; and “integrative” was
assembly accuracy—the number of assembly opera- interpreted as relating to strategy descriptions that
tran- tions done correctly. Similar accuracy-based measures scended the superficial elements and
corresponded to of performance have been used in prior TMS research underlying principles relevant
to how such problems can (e.g., Liang et al. 1995, Moreland and Myaskovsky be solved. The two
raters independently rated the same 2000). Distinct operations were defined according to 20 groups
and, after determining that interrater reliability steps described in printed instructions included with
was high (ICC(2) = 0 98), split up the remaining groups the assembly kit. We pretested the telephone
assembly and rated them. The average score of the raters was used using the printed instructions and
confirmed 38 distinct for the first 20 groups. A higher strategy-quality score operations in the
telephone task. A trained experi- indicates a greater degree of abstract knowledge than menter
examined each group’s completed telephone and does a lower score. counted misplaced or
misconnected components accord- Control Variables. We controlled for prior expertise ing to these
instructions. We deducted one point from with electronics or electronics kits, anticipating that a
group’s performance score for each inaccurate opera- members who had experience on tasks similar to
the tion, such that higher scores indicate higher learning-task tasks we used in our study would be able
to achieve performance. higher performance. Prior expertise was measured with two items that
appeared on the pretraining survey: Learning Transfer Transfer-Task Performance . The “Based on past
experience, I would rate my overall extent to which groups transferred learning across tasks knowledge
level of electronics as” and “Based on past was measured in terms of transfer-task performance—
experience, I would rate my skill level with electron- the number of assembly operations done
correctly for ics kit assembly as.” Each of those items had a five- the personal stereo assembly kit.
Pretests using the point response format, anchored at 1 = beginner and 5 = kit’s printed instructions
indicated that there are 34 dis- expert. The interitem correlations r = 0 69 justified tinct operations in
the personal stereo task. Points were summing the item scores to form a composite. Group deducted
from a group’s score for each inaccurate oper- scores were computed as the sum of member compos-
ation, such that higher scores indicate higher learning ite scores in each group, in each condition.
Because transfer to the stereo task. each member’s prior expertise is independent of other Abstract
Knowledge. We measured the extent to members’ prior expertise, there was no need to check which
groups developed an abstract, generalized under- for within-group agreement before summing
members’ standing of the underlying principles common to the scores. A higher group score indicates
a greater level of tasks by asking groups to articulate a strategy for assem- prior expertise in the group
than does a lower score. bling a third electronics kit (electronic stapler), a task in the same domain as
the learning and transfer tasks. The 5. Results abstract knowledge measure takes into account knowl-
We conducted all hypotheses tests at the group level edge about the task and how the group might
solve the using ANOVA and planned contrasts. Initial checks
12. 12. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer 592 Organization
Science 16(6), pp. 581–598, © 2005 INFORMS showed that group gender composition (computed as
groups had significantly higher mean scores on the TMS percent of female members) was significantly
negatively scale, compared with groups composed of reassigned related to prior expertise for each of

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 11/13
09/12/21, 14:38 Lewis & al 2005

the tasks and signif- members (M = 55 47 versus M = 53 85), F 1 96 = icantly negatively related to
transfer-task performance. 4 06, p < 0 05. We also measured task recall in half of Therefore, we
controlled for gender composition in addi- the sample (34 intact groups, and 32 reassigned groups);
tion to controlling for prior expertise in all analyses. recall should be higher in groups with a TMS. The
number of task steps recalled was significantly higher 5.1. Learning-Transfer Check in intact groups,
compared with reassigned groups Singley and Anderson (1989) recommend checking that (M = 21 53
versus M = 14 83), F 1 62 = 33 03, p < learning transfer is even possible before testing hypothe- 0 001.
The bivariate correlation between TMS and recall ses about the extent of learning transfer between
tasks. N = 66 is 0.28, p < 0 05. These results suggest that Their recommended method compares
performance for keeping groups intact gave groups full access to the TMS groups that complete both
learning and transfer tasks developed during training, while reassigning groups dis- with the
performance of groups that only complete a abled any TMS that had developed. Thus, our manip-
transfer task. If the performance of the trained groups ulation for creating a TMS in some groups and
not in is higher than that of the untrained groups, transfer others was successful. has occurred from the
learning to the transfer task. We tested for learning transfer using a holdout sam- 5.3. Hypothesis Tests
ple from the same population as our study participants. Figure 2 depicts the experimental design
conditions and A total of 93 students in 31 groups comprised the hold- the results from ANOVAs and
planned contrasts for all out sample, 16 of whom were trained on the learning hypothesis tests.
Hypothesis 1 predicted that relative to task and performed the transfer (stereo) task, and 15 of groups
with no prior TMS, groups with a TMS would whom received no training before performing the trans-
demonstrate higher performance on the learning task. fer task. The difference between performance
means for Performance on the telephone assembly task was indeed the transfer task was significant, F
1 29 = 20 31, p < higher in groups with full access to their training TMS, 0 01 (M = 25 25 for trained
groups versus M = 16 66 compared with groups whose training TMS had been for untrained groups).
Higher transfer-task performance disabled (M = 32 22 versus M = 31 19), and this dif- for groups that
received training on the learning task ference is significant, F 1 96 = 3 80, p < 0 05. Thus, relative to
those that did not is evidence that transfer Hypothesis 1 is supported. occurred between the learning
task and the transfer task. Hypotheses 2 and 4 were tested together, using ANOVA and two planned
contrasts. We predicted that 5.2. Manipulation Check groups that had developed and utilized a TMS on
a We created a TMS in some groups and not others by previous task would perform better on a transfer
task keeping half of the participants in their training groups than groups with no prior TMS
(Hypothesis 2), and that and by reassigning the other half of the participants to groups that experienced
a disruption to an established new groups. We expected that groups remaining intact TMS structure
would perform worse on the transfer task after training would have full access to their TMS, and than
groups that had never developed a TMS (Hypoth- therefore higher TMS scores, than would groups
whose esis 4). ANOVA results show no significant differences members had been reassigned. We
measured TMSs with among any of the performance means, F 1 94 = 0 35, a 15-item scale developed
by Lewis (2003) and com- p = 0 55, providing no support for Hypotheses 2 and 4. puted a TMS
composite score for each member by sum- We discuss these findings in more detail below. ming scores
on the 15 items = 0 83 . To confirm that Hypotheses 3 and 5 were tested together. We pre- members’
scores could be aggregated to the group level, dicted that experience with two tasks would be more we
evaluated the rwg statistic (George 1990), which mea- likely to produce an abstract, generalized
understand- sures the degree to which individual ratings within a ing of the task domain when a group
had a prior TMS group are interchangeable. Mean rwg values of 0.70 or (Hypothesis 3). If, however, a
group experienced a dis- greater provide evidence of acceptable agreement among ruption to an
established TMS structure, a prior TMS member responses on a scale (George 1990). The aver- was
expected to interfere with learning and the develop- age rwg on the TMS scale for the learning task
was 0.97, ment of abstract knowledge (Hypothesis 5). An ANOVA with 100% of the rwg values above
0.70. These results with planned contrasts shows that both Hypotheses 3 indicate that group member
responses on the TMS scale and 5 are supported, F 1 89 = 5 14, p < 0 05. Groups were quite
homogeneous and that aggregating mem- with a training TMS that remained intact demonstrated bers’
scores to the group level of analysis is statistically a better understanding of the underlying principles
and justified. strategies relevant to the task domain, compared with A one-way ANOVA, with gender
composition and groups that had never developed a TMS (abstract knowl- prior expertise entered as
covariates, shows that intact edge score M = 3 41 versus M = 2 56), t 89 = 2 13,
13. 13. Lewis et al.: Transactive Memory Systems, Learning, and Learning Transfer Organization Science
16(6), pp. 581–598, © 2005 INFORMS 593 Figure 2 Experimental Design and ANOVA Results
Training Learning task (telephone) Transfer task (stereo): Knowledge task (stapler) Week 1 Week 2

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 12/13
09/12/21, 14:38 Lewis & al 2005

Week 3, Part 1 Week 3, Part 2 I I Maintain TMS 3.41H3 (0.28) Maintain TMS 22.73H2 (1.00) R 3.28
(0.32) (I) Intact Maintain TMS 32.22H1 (0.37) I 3.33 (0.25) R 23.28H4 (1.06) Disrupt TMS R 1.56H5
(0.31) Disrupt TMS N = 100 groups I I 3.72 (0.36) 23.33 (1.07) R 3.30 (0.29) (R) Reassigned 31.19H1
(0.37) Disrupt TMS I 2.55 (0.41) R 22.65H2,H4 (0.99) Disrupt TMS R 2.56H3,H5 (0.30) Disrupt
TMS Sample mean/s.d. 31.70/2.66 22.97/5.05 3.00/1.20 Notes. Mean scores are shown for each
condition. Standard errors are in parentheses. H1 comparison is significant, F 1 96 = 3 80, p < 0 05. H2
and H4 comparisons are not significant, F 1 94 = 0 35, p = 0 55. H3 and H5 comparisons are
significant, F 1 89 = 5 14, p < 0 05, t 89 = 2 13, p < 0 05, and t 89 = −2 31, p < 0 01. p < 0 05,
supporting Hypothesis 3. Groups with a train- expertise changed for the transfer task. Given these find-
ing TMS that later experienced a disruption to their ings, we decided to create a variable that measures
existing TMS structure had significantly lower abstract the extent to which perceived expertise
remained stable knowledge scores than did groups that had never devel- across tasks and to reexamine
Hypotheses 2 and 4, tak- oped a TMS M = 1 56 versus M = 2 56), t 89 = ing this new variable into
account. −2 31, p < 0 01. Thus, Hypothesis 5 is supported. Our results show that having developed a
TMS not 5.4. Expertise Stability Analysis only affects performance on the task for which the
Members’ consistent recognition of and agreement about TMS first developed, but also affects the
development of location information in both tasks is evidence that abstract knowledge about the task
domain, given experi- members did indeed maintain specializations across ence with an additional
task. Abstract knowledge facili- the different task contexts. In our surveys, we had tates mapping of the
task principles to other similar tasks asked participants to identify which members had what beyond
those the group has already completed (Reeves expertise following the completion of the learning and
and Weisberg 1994), suggesting that having developed a transfer tasks. Two of the coauthors, blind to
the TMS can indeed facilitate learning and learning transfer. conditions, independently coded
individual responses We also found evidence consistent with our prediction into nine expertise
categories: Mechanical/electrical, that a severe disruption to the TMS structure impedes general
assembly, small-parts assembly, large-parts abstract, conceptual learning about the task domain.
assembly, assembly strategy, recall, general assistance, We did not find evidence, however, of any
learning- motivation, and parts management. Raters indepen- transfer effects after groups had
experience with only dently categorized each member’s response (Cohen’s one task (Hypotheses 2 and
4). Additional examination kappa = 0 90, p < 0 01), discussed cases where there of the groups’ TMS
structures revealed some explana- was disagreement, and came to a consensus about the tions for these
findings. When we examined elements appropriate expertise categorization. The consensus cat- of the
TMS structures using members’ statements about egorizations were used as the basis for the expertise
sta- “who is expert at what” for each task, we found that bility scores. in more than 20% of the
reassigned groups, members’ A measure of expertise stability was derived from the perceived expertise
remained stable across tasks, while number of times members agreed about each member’s in nearly
half of the intact groups, members’ perceived expertise, both within and between tasks. For example,

https://pt.slideshare.net/IkramKASSOU/lewis-al-2005 13/13

You might also like