This Content Downloaded From 182.255.0.242 On Fri, 27 Nov 2020 04:27:54 UTC

The Effectiveness of Knowledge Transfer Portfolios in Software Process Improvement:
A Field Study
Author(s): Sandra A. Slaughter and Laurie J. Kirsch
Source: Information Systems Research , September 2006, Vol. 17, No. 3 (September
2006), pp. 301-320
Published by: INFORMS
Stable URL: https://www.jstor.org/stable/23015892
REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/23015892?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms
INFORMS is collaborating with JSTOR to digitize, preserve and extend access to Information
Systems Research
This content downloaded from

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
All use subject to https://about.jstor.org/terms
Information Systems Research
Vol. 17, No. 3, September 2006, pp. 301-320
doi 10.1287/isre.l060.0098
issn 1047-70471 eissn 1526-55361061170310301 ©2006 INFORMS
The Effectiveness of Knowledge Transfer

Portfolios in Software Process
Improvement: A Field Study
Sandra
Sandra A.
A. Slaughter
Slaughter
David A. Tepper School of Business, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213,
sandras@andrew.cmu.edu
Laurie J. Kirsch
Joseph M. Katz Graduate School of Business, University of Pittsburgh, Pittsburgh, Pennsylvania 15260,
lkirsch@katz.pitt.edu
Because ofimprovement
process challenges often experienced
(SPI) initiatives. when to
Critical deploying software,
the success of these many firms
initiatives have
is the embarked
transfer on sof
of knowled
across individuals who occupy a range of roles in various organizational units involved in software product
Prior research suggests that a portfolio of different mechanisms, employed frequently, can be required for e
tive knowledge transfer. However, little research exists that examines under what situations differing portfo
of mechanisms are selected. Further, it is not clear how effective different portfolio designs are. In this stu
we conceptualize knowledge transfer portfolios in terms of their composition (the types of mechanisms u
and their intensity (the frequency with which the mechanisms are utilized). We hypothesize the influence
organizational design decisions on the composition and intensity of knowledge transfer portfolios for SPI.
then posit how the composition and intensity of knowledge transfer portfolios affect performance improveme
Our findings indicate that a more intense portfolio of knowledge transfer mechanisms is used when the sour
and recipient are proximate, when they are in a hierarchical relationship, or when they work in different un
Further, a source and recipient select direction-based portfolios when they are farther apart, in a hierarchic
relationship, or work in different units. In terms of performance, our results reveal that the fit between th
composition and intensity of the knowledge transfer portfolio influences the recipient's performance impro
ment. At lower levels of intensity direction-based portfolios are more effective, while at higher levels of inten
routine-based portfolios yield the highest performance improvement. We discuss the implications of our f
ings for researchers and for managers who want to promote knowledge transfer to improve software proces
in their organizations.
Key words: knowledge transfer mechanism portfolios; portfolio intensity; portfolio composition; knowledge
transfer; software process improvement; management of information systems
History: V. Sambamurthy, Senior Editor. This paper was received on February 15, 2005, and was with the
authors 7\ months for 1 revision.
Introduction Process improvement of any kind—including SPI—

is not simply a matter of individuals embracing incre
In recent years, firms have invested heavily in soft
ware as information technology (IT) plays a critimental changes. Rather, armed with new knowledge,
individuals
cal role in many aspects of the value chain. How in various roles and units fundamen
ever, firms often fail in their attempts to build tally

and rethink work patterns and relationships, de
deploy software (Gaudin 2003), prompting a grow oping new cognitive frameworks and schemas (Mi
ing interest in software process improvement et (SPI)al. 1997, Spencer 1994) and embed these n
initiatives (Herbsleb et al. 1997). These initiatives
structures into their work practices (Ravichand
involve the introduction of specific practices and and Rai 2003). SPI transcends discrete projects
measures designed to improve the deployment and systems, and comes from recognizing patterns ac
subsequent maintenance of information systems (IS) projects and systems to design and institutiona
(Grady 1997). new work practices (Mathiassen and Pourkomey
301

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
Slaughter and Kirsch: The Effectiveness of Knowledge Transfer Portfolios in Software Process Improvement
302 Information Systems Research 17(3), pp. 301-320, ©2006 INFORMS
2003). Improvement emanates from aWe next and

deep reviewbroad
the relevant literatures on SPI
andand
understanding of current processes on knowledge transfer, and develop our mod
practices,
and hypotheses.
their patterns, and implications. Developing this We
kindthen describe the methodology,
of understanding requires knowledge present the results
transfer about of our analysis, and discuss their
implications.
software processes and practices between We conclude by identifying the contri
individuals
butions
occupying various organizational roles andand limitations
located in of our study and suggest area
different work units. for future research.
Knowledge transfer is a dyadic exchange in which

a recipient learns and applies knowledge transmit Relevant Literature
ted from a source (Ko et al. 2005, Argote and Ingram
Software Process Improvement
2000). Knowledge is transferred via mechanisms such
Improving software development capabilities requires
as technology, personnel movement, and blueprints
knowledge creation (the accumulation of knowledge)
(Darr et al. 1995, Argote 1993, Epple et al. 1991, Adler
and knowledge embedding, absorbing individually
1990). Recently, researchers have argued that effective
and collectively held knowledge into organizational
knowledge transfer is dependent, in part, on repeat
processes (Ravichandran and Rai 2003). While the
edly using combinations of mechanisms rather than
knowledge required for process improvement may
relying on the single use of a single mechanism (Rulke
include some simpler or more explicit aspects, to
et al. 2000, Malhotra et al. 2001, Argote 1999). Extend
effectively embed and apply the knowledge requires a
ing these arguments, we define a portfolio of knowledge
more complex understanding of the context and task
transfer mechanisms as the set of mechanisms used by
(Mathiassen and Pourkomeylian 2003). For example,
a source to transfer knowledge to a recipient. Fur
consider an IS manager who wants to adapt devel
ther, we characterize portfolios by their composition
opment processes to facilitate the maintenance of sys
(the types of mechanisms used) and their intensity (the
tems. To enable this SPI, the manager can assign a
frequency with which the mechanisms are utilized).
maintainer to a development team to transfer knowl
The purpose of our study is to examine the role
edge about which coding practices yield code that is
that knowledge transfer portfolios play in SPI. In par
difficult or costly to maintain. Knowledge is trans
ticular, we examine factors that influence the struc
ferred to the developer from the maintainer over the
ture of the portfolio in terms of its composition and time period of the SPI initiative as the developer reg
intensity, as well as the effectiveness of various port ularly interacts with and observes the maintainer. In
folios. Our study is situated in the context of an due course, this knowledge transfer may yield new
SPI initiative undertaken by three distinct units—processes like peer design reviews that improve main
Development, Maintenance, and Quality Assurance tenance efficiency by identifying problematic designs
(QA)—within a large IS department in one firm. early in the software life cycle.
To the best of our knowledge, our study is one Improving the practice of software development
of the first to conceptualize and empirically investi and maintenance therefore requires the transfer of
gate knowledge transfer portfolios. We elaborate the knowledge between source and recipient dyads who
oretically grounded dimensions of knowledge trans are located within and across organizational units
fer portfolios—composition and intensity. Our focus (Mitki et al. 1997). Moreover, it requires continuous
on portfolios draws attention to the range of mechinteraction between source and recipient over time;
anisms used, and the way in which the mechanismsmultiple transfers utilizing multiple mechanisms may
are combined, to transfer knowledge about software be required so that a recipient understands the value
processes across roles and units. Our study also pro of the innovations and begins to incorporate them
vides insight into the effectiveness of various knowl into specific work practices (Ravichandran and Rai
edge transfer portfolio designs by examining the fit 2003). Institutionalizing these new work practices,
between portfolio composition and intensity and per however, takes time, suggesting that individual per
formance improvement. formance improvement may not be observed after a

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
Information Systems Research 17(3), pp. 301-320, ©2006 INFORMS 303
single knowledge transfer event. Though

Thus, muchin our
of the study,
empirical literature seems to
the portfolio of mechanisms assume
used by a source to of a sin
that knowledge transfer is composed
transfer software process knowledge
gle movement ofto a recipient
knowledge from a source to a recip
ient utilizing
represents the accumulation of single transfersa single mechanism,
between some researchers
a particular source and recipienthave begun the
over to take timeframe
a broader view (McGrath and
of the process improvement period. In (1999),
Argote 2001). Hansen ourforexam example, found that
ple above, the portfolio of mechanisms includes
repeated interactions team
between individuals with strong
membership, observation, and source-recipient inter
ties facilitated knowledge acquisition. The results of
actions. In this portfolio, observation and
Carlile's (2004) study interac
suggest that successful knowl
tion are utilized multiple times to
edgetransfer
transfer requires knowledge
multiple iterations of sharing
through which
about coding practices, representing individuals develop
repeated trans common lan
fers of this knowledge as the SPI initiative
guage, understanding,unfolds.
and interests. In an experi
As illustrated in our example, knowledge transfer
ment examining knowledge sharing,for
Paulus and Yang
the purpose of SPI is not restricted
(2000) to
foundathat
particular part among
more frequent interactions
of the IS department, but it occurs across
group members units
engaged in idea and
generation resulted
roles.
in better performance than when ideas were gener
ated by individuals but not shared with group mem
Knowledge Transfer Mechanisms and
bers. These Portfolios
studies highlight the importance of the
Knowledge is information possessed intensityby
of thean individual
portfolio; that is, the frequency with
that, when combined with other personal dimensions which the mechanisms are used.
such as experience and reflection, becomes
Not only a basis
does the literature suggest that a portfo
for action (Grover and Davenport 2001).
lio varies For but
in intensity, exam there is also evidence to
ple, information about softwaresuggest defect that patterns
its composition in a the types of mech
(i.e.,
particular system becomes "knowledge" anisms included) when matters. Argoteit isand Ingram (2000)
combined with an understanding noteofthatsystem develop knowledge by mov
though transferring
ment and business processes toing improve
members is often thedifficult,
design transferring knowl
of future systems. Just as knowledge about defect
edge with a combination of task and tool elements
patterns can become a basis for can beaction, so too
effective. Others can similar find
have observed
knowledge of other software product ings about theand use ofprocess
multiple kinds of mechanisms.
attributes (e.g., software size, productivity, Galbraith (1990), or for example,
effort) found that effective
spur changes in development and maintenance prac
knowledge transfers utilize a combination of people
tices. Moreover, transferring knowledge and technology mechanisms.
among keyIn their study, Rulke
stakeholders means that software development
et al. (2000) observed a positive pro relationship between
cesses can be shaped for improved the use ofefficiency
multiple mechanisms and of knowledge transfer
effectiveness above and beyond what
(including any newsletters,
training, one indiobservation, personal
vidual can accomplish (Ravichandran contacts, and and Rai 2003,
conference presentations) and learn
2000). ing outcomes. This research establishes the impor
Generally speaking, knowledge is transferred "by tance of using multiple types of mechanisms, but it
moving people, technology, or structure to the recip raises the question of how to meaningfully categorize
ient organization or by modifying the people (e.g., these mechanisms to better understand their use and
through training), technology, and structure of the impact. We turn to this question in the next section.
recipient organization" (Argote 1999, p. 145). Specific
transfer mechanisms include training, observation of A Typology of Knowledge Transfer Mechanisms
experts, tooling, assembly line layouts, routines, and In a conceptual paper, Grant (1996) argues that the
meetings (Contu and Willmott 2003, Thomas-Hunt primary role of the firm is knowledge integration, or
et al. 2003, Birkinshaw et al. 2002, Adler 1990). the process of combining the specialized knowledge

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
of a number of individuals. Hiswith

concept
its emphasis of integrastandards, met
on documentation,
rics, and training, provides
tion includes the transfer and replication of knowl ample opportunities to
edge from individual to individual
make use in an
of such organiza
mechanisms to transfer knowledge
tion. According to Grant (1996),across
thereroles and
areunitstwo
in the fun
IS department. On the
damental categories of mechanismsotherthat
hand, organizational
can be used routines are more implicit
means
to integrate knowledge: directions and of transferring knowledge, which emphasize
organizational
routines. Mechanisms classified as "directions" are ex
the movement of knowledge "without language"
plicit rules and instructions, such as standard (Nonaka
oper 1994, p. 19). Organizational routines can be
found in SPI in the form of observing coding experts
ating procedures or plans. In contrast, organizational
in their work, imitating the use of best design prac
routines are those mechanisms that allow for integrat
ing knowledge without explicitly communicatingtices,
it: and informal interactions between a system
"... the essence of an organizational routine is developer
that and a maintainer or a quality specialist.
The theoretical foundations, as well as the par
individuals develop sequential patterns of interac
simonious nature, of the directions—organizational
tion which permit the integration of their special
ized knowledge without the need for communicatingroutines typology make it attractive for classifying
knowledge transfer mechanisms in portfolios. More
that knowledge" (Grant 1996, p. 379). Surgical teams
over, given the fit of this typology with the SPI
and project teams use routines to transfer knowledge
when individuals embrace a common understand context (as also perceived by Ravichandran and Rai
2003),
ing of roles and interactions, bring their own we felt it was most suited to our objec
exper
tise to bear on the task, and work closely tives and adopted it to describe the composition of
together
to accomplish their objective. As explained below,transfer portfolios in our study. We next
knowledge
Grant's (1996) typology is well suited to the develop
contexta model (summarized in Figure 1) that exam
of our study. ines the antecedents of the intensity and composi
Directions transfer explicit knowledge relating to tion of a portfolio of knowledge transfer mechanisms,
SPI by coding the knowledge into rules and pro as well as the consequences for software process
cedures. The software process improvement context, performance.
Figure 1 Research Model
Organizational design Knowledge transfer Software process

characteristics portfolio choices performance gains
Nature of
relationship
Proximity
Work unit
Controls (source and recipient):

IT work experience
Organizational tenure
Level of education
Gender
Prior work relationship

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
Research Model and Hypotheses It is often the case that the greater the diversity
of knowledge among the specialists involved, the
As shown in Figure 1, we posit that organizational
design characteristics will influence the naturelower is the level of their shared knowledge, in part
of the
knowledge transfer portfolios selected by a because
source one specialist generally does not develop
deep
and recipient, and portfolio choices will, in turn, familiarity in another's area of expertise. Prior
influ
research suggests that unless the specialists have
ence the recipient's performance gains from software
process improvement. In the context of worked
SPI ini closely together on prior activities, devel
tiatives, where learning and applying new oped
knowlexpertise or fluency in the language of each
edge are essential, organizational design parametersother's specialty, or have worked in the same prod
have important implications for knowledge transfer uct line or discipline, it is unlikely they will develop
(Ravichandran and Rai 2003, 2000). Indeed, the sharedstruc knowledge (Malhotra et al. 2001, Nelson and
tural configuration of an organization has beenCooprider
identi 1996). This lack of common knowledge
makes
fied as a critical factor that can facilitate or inhibit it difficult for the different specialists to com
pro
cess improvement (Mitki et al. 1997) becausemunicate knowl and reduces the efficiency of their knowl
edge tends to move horizontally and verticallyedge
alongtransfer (Carlile 2004, Grant 1996, Demsetz
structural lines (Schulz 2001). Thus we examine1991).
how It is possible that a source and recipient in
portfolio intensity and composition may differ different
for work units who continually interact may
flows of knowledge across a source and recipient who
develop a shared understanding that facilitates their
reside in similar or different work units, and for knowledge
flows transfer. However, even with repeated
of knowledge between a source and recipient who priorare
interaction, it is not likely that one specialist will
in a hierarchical or nonhierarchical relationship. We
develop a complete understanding of another special
then consider how the proximity of the source and
ist's domain. Thus, a source and recipient who reside
recipient affects portfolio choices. To shed light on
in different work units with distinct kinds of exper
the effectiveness of portfolio intensity and composi
tise will generally need to try harder than a source
tion, we also examine performance outcomes associ
and recipient in the same work unit to transfer their
ated with various portfolio designs. knowledge. To improve software processes, develop
ment, maintenance, and quality specialists residing in
Work Units and Portfolio Choices
different work units must share their knowledge. This
To improve software processes requires understand
need to transfer knowledge across units, coupled with
ing the cause and effect relationships between pro
the specialized knowledge of the individuals in these
cess parameters and outcomes (Ravichandran and Rai
units, implies that interunit dyads will repeatedly and
2003). However, the individuals who have knowledge
frequently transfer knowledge, increasing the inten
about development processes could be in different
sity of their portfolio.
work units than the individuals who have knowledge
about process outcomes. For example, a maintenance Hypothesis 1A (H1A). The intensity of the knowl
specialist could have detailed knowledge about the
edge transfer portfolio selected will be higher when the
performance (such as the response time or error rate)
source and recipient work in different units than when they
of a particular IS over a period of time. A develop work in the same unit.
ment specialist could have detailed knowledge about
the processes used to create that system, and a QAWe further expect that specialists who are in dif
specialist could have knowledge about how to iden ferent work units will be more likely to use direc
tify the root causes of quality problems. To under tions than routines to transfer knowledge. This is
stand how the performance problems in this system because routines can only be leveraged when there
relate to the development processes used to create itare repeated and direct interactions between the
requires the development, maintenance, and QA spe source and recipient who share a common frame of
cialists to transfer their specialized knowledge aboutreference. As noted by Nonaka (1994, p. 19), "with
process attributes and outcomes. out some form of shared experience, it is extremely

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
difficult for people to share each formal other'shierarchy

thinking provides
prochannels through which
cesses. The mere transfer of information will oftenknowledge can easily flow. For example, if a project
make little sense if it is abstracted from nuanced manager is interested in improving the development
emotions and nuanced contexts that are associated process, she examines defects in current projects,
with shared experiences." Software developersidentifies
have trends in the defects, and determines
a shared understanding of the process and context
whether
in patterns in defects can be traced to specific
which user requirements and technology are development
trans processes. Knowledge about defect pat
terns,
formed into functional information systems. Thus, an which becomes one basis for improving devel
improvement to a software testing procedureopment
pro processes, is transferred from the manager
posed by one developer does not require extensive
to a developer using existing channels (mechanisms),
such as regularly scheduled meetings and formal
directions to be comprehended and used by another
developer. Instead, this knowledge can be transferred
memos. It is important to note that either person in
via organizational routines (e.g., the source could
the hierarchical relationship could be the source of the
demonstrate the new procedure while working withknowledge transfer; that is, the source is not always
the recipient on a project, or the recipient could
the superior. The project manager, in the above exam
observe the source using the new procedure).
ple, In
could transfer the knowledge about defect pat
contrast, even if they have interacted frequently in to her superior as well as to her subordinate.
terns
the past, a software developer and a QA specialist
Frequent interactions and knowledge flow along the
could still lack a complete shared understanding of
hierarchy, as compared to peer-to-peer knowledge
the nuances of the development context (becausetransfer,
the seems to accelerate the identification and suc
developer focuses on development activities while the use of best practices (Ravichandran and Rai
cessful
quality specialist focuses on quality-related activities).
2000). Thus we expect:
Thus the developer must be more explicit in docu
Hypothesis 2A (H2A). The intensity of the knowl
menting and codifying the new procedure for the QA
edge transfer portfolio selected will be higher when the
specialist (than he would be with another developer)
source and recipient are in a hierarchical relationship than
to understand its implications for software quality.
when they are not.
This implies:
With respect to portfolio composition, we anticipate
Hypothesis IB (H1B). The use of directions in the
that peers would rely more on routines while superi
knowledge transfer portfolio selected will be higher when
ors and subordinates would rely more on directions.
the source and recipient work in different units than when
One reason is efficiency. As Grant (1996) explains, the
they work in the same unit.
efficiency of knowledge transfer requires economizing
Hierarchical Relationships and Portfolio Choices on the amount of communication required for trans
An important aspect of the source-recipient relation fer to occur. Organizational structures and, in partic
ular, bureaucracies, are very efficient in transferring
ship is whether it is hierarchical or nonhierarchical.
Because organizational structures often dictate knowledge
the along the hierarchy because directions can
paths of communication, there typically are frequent be used as the predominant integrating mechanism,
information exchanges along the hierarchy (Hinds achieving higher levels of coordination with less com
and Kiesler munication. For example, a senior manager could
1995). In knowledge transfer, substantial
knowledge flows have been observed between develop
subor a new policy for software testing. Codifying
dinates and the policy into a set of explicit rules and instructions
superiors, especially when the knowledge
is new and unsettled (Schulz 2001). is an economical way to transfer knowledge about the
A great deal of the knowledge about software newpro policy to subordinates.
cess diagnosis and interpretations flows between Directions
a leverage the detailed articulation of in
project manager and a software developer or structions,
main while routines typically rely upon a much
tamer, or between senior management and subordi more limited set of cues and responses that serve not
nates, rather than between peers. This is becausesothe
much as to communicate knowledge, but rather

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
leverage knowledge transfer mechanisms that rely

to permit a sequencing of an individual's application
of knowledge inputs (Grant 1996). As weon personal interactions more frequently and readily
reasoned
earlier, when a source and recipient share athan a quality specialist who is located in another part
common
understanding, they can more easily rely on of routines
the building or another city This suggests:
to transfer knowledge. We expect that peers share a
Hypothesis 3A (H3A). The intensity of the knowl
more common frame of reference than superiors and
edge transfer portfolio selected will be higher the closer the
subordinates, and thus are more likely to use routines
source and recipient.
than a superior and subordinate:
We also expect that proximity influences the com
Hypothesis 2B (H2B). The use of directions in the
position of the knowledge transfer portfolios selected.
knowledge transfer portfolio selected will be higher when
A proximate source and recipient can leverage infor
the source and recipient are in a hierarchical relationship
mal procedures and implicit communications of rou
than when they are not.
tines to transfer knowledge, while a distant source
Proximity and Portfolio Choices and recipient may be restricted to codified directions
for
Proximity is an important organizational designeasy transport to remote locations. For example,
a QA specialist who is located close to developers and
parameter. The literature suggests that collocating in
dividuals is vital to developing shared goals, deep could leverage routines such as observa
maintainers
tion of experts, face-to-face meetings, and informal
understanding, and information sharing, particularly
activities,
when individuals are in different disciplines or when while the more distant quality specialist
may have
the task is uncertain or complex (e.g., Argote 1999,to use directions such as computer-based
Crowston and Kammerer 1998). In software develop repositories that can be accessed across
reports and
large distances. Thus we posit:
ment, Teasley et al. (2002, p. 671) have documented
the beneficial effects of radical collocation or "warHypothesis 3B (H3B). The use of directions in the
rooms." Proximity has been shown to increase opporknowledge transfer portfolio selected will be lower the closer
tunities for interactions between individuals (Kraut
the source and recipient.
and Streeter 1995). In addition, the literature sug
Portfolio Choices and Performance Improvement
gests that the beneficial effects of collocation fade very
The characteristics of the knowledge transfer portfolio
quickly, at distances beyond 30 meters (Allen 1977).
used—its composition and intensity—are expected to
Thus, even relatively short distances between individ
impact the performance of the recipient. The portfo
uals can significantly affect their communication and
information-sharing behaviors. Indeed, despite therepresents the set of mechanisms used by a source
lio
to transfer knowledge to a recipient over the life
advantages of new collaboration technologies, experts
of the SPI initiative. A successful knowledge trans
concur that "distance still matters" (Olson and Olson
2000, p. 139). fer event implies that the recipient has applied the
knowledge supplied by the source. Realizing per
Because proximity fosters personal contact and
formance gains requires that the transferred knowl
communication, it seems likely that a colocated source
edge be embedded in new work practices, which
and recipient will transfer knowledge more intensely
are sustained over time to become institutionalized
than a source and recipient who are more distant:
(Ravichandran and Rai 2003). It is the institutional
Individuals who are physically proximate have more
opportunity to use personal mechanisms of trans ization of these work practices—their routinization
fer (Argote 1999), in addition to technology and and regular use—that drives improvement in perfor
mance. As the recipient receives an increasing amount
structural mechanisms that represent the most likely
of knowledge from a variety of sources over the life of
choices for a source and recipient who are separated
the process improvement period, he is learning tech
in space. All else being equal, then, it seems likely
that a proximate source-recipient dyad will trans niques and approaches to enhance his own software
fer knowledge more frequently. For example, adevelopment
QA practices, and therefore his performance
should
specialist who is located close to developers could improve as the SPI initiative goes forward.

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
However, successful knowledge transfer,

Hypothesis in 4the
(H4). The higher the inten
sense that a recipient applies knowledge
knowledgefrom a portfolio selected by th
transfer
source, does not guarantee that new practices
greater are
will be the performance improvement o
institutionalized and performance gains are realized.
We also expect portfolio composition to
Generating positive impacts on performance in the
formance. Directions represent knowledg
long term would seem to depend on whether two
codified and articulated, and would seem to facili
conditions are met: (1) that the correct knowledge is
tate the source's ability to transfer the correct knowl
transferred by the source and (2) that the recipient
edge. By their very nature, directions are less ambigu
correctly applies the knowledge (Mitki et al. 1997).
ous and more precise than routines, which lack such
These ideas are explored next as we develop spe
explicit specification. When knowledge is transferred
cific hypotheses that relate portfolio composition and
via directions (such as standard operating procedures,
intensity to performance improvement.
manuals, and policy statements), it seems reasonable
We expect that the intensity of the portfolio of
to expect that the recipient will understand the knowl
knowledge transfer mechanisms will have a posi
edge in a relatively precise manner, and thus be able
tive impact on performance improvement. An intense
portfolio suggests that the sourcetois apply it correctly. For example, knowledge about
repeatedly and
how to count function points can be transferred in
frequently transferring knowledge to a recipient. The
source is not relying on a singlea knowledge
manual, and can be applied directly by recipients.
trans
fer event, but instead transferringInthe
contrast, understanding
knowledge to knowledge that is com
municated via routines
the recipient on multiple occasions. When knowledge requires common understand
ing and
is rich and contextualized, such as theshared experiences between the source and
knowledge
required to improve and institutionalize workit pro
recipient. Moreover, implies that recipients need to
cesses, repeated transfers may be necessary to gen the knowledge to apply
interpret and contextualize
erate a deep understanding of theit, suggesting
process thaton
performance
the gains may be realized
more
part of the recipient and to enable slowly
the Thus, on average,
recipient to we expect to see that
portfolios
embed the new process into ongoing workcomposed
practicesof a greater proportion of direc
tions will yield
(Ravichandran and Rai 2003, Alavi 2000). A single more performance gains.
transfer of knowledge may not be as effective
Hypothesis as
5 (H5). The greater the use of directions
repeated transfers (Rulke et al. 2000). For example,
in the knowledge transfer portfolio by the source, the greater
consider a scenario where a quality will
specialist is trans
be the performance improvement of the recipient.
ferring knowledge about how to use root cause anal
We also posit
ysis to determine the source of software that composition
defects. The and intensity inter
act to affect performance.
initial transfer could involve a lecture, and the recipi As Grant (1996) argues,
directions are an
ent may be able to apply the concept after that lecture. efficient means of transferring
knowledge. This suggests
However, it is likely that if the lecture is followed by that if a portfolio is largely
composed
additional lectures, demonstrations, and of directions,
tutorials, the there is no need to transfer
this knowledge
recipient may be better able to understand repeatedly
and more(that is, intensity should
be low).into
effectively integrate root cause analysis For example,
software a procedures manual (a direc
tion) does not
design practices. Thus, in the SPI context, it seemsneed to be transferred multiple times
over the life of
likely that more knowledge transfers will result in a the SPI initiative; once should suf
recipient who truly understands the ficeknowledge
(low intensity). On the other hand, if routines are
being
used to
transferred, increasing the likelihood transfer
that theknowledge,
recipi the knowledge is tacit,
and thus effectively,
ent will be able to apply that knowledge more amorphous. Successfully transferring
this type of
and thus generate positive performance knowledge
gains. would seem to require mul
This
tiple transfersthat
argument is consistent with prior research (i.e., intensity
has should be high), such as
found a positive association betweenrepeated observation
frequent knowl of experts or closely working
with specialists.
edge transfer or sharing and performance (PaulusConsistent
and with this argument, Alavi
Yang 2000). This leads us to predict:
has noted that "... tacit knowledge is best transferred

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
through collaboration, shared experience,

improvement.and rich to improve the quality of
For example,
interpersonal interactions over time (e.g.,design
a software a mentor
process may require the transfer of
ing relationship)" (2000, p. 20). Thusknowledge
we posit about
that
design
the processes from Development
fit between composition and intensity to Maintenance,
of the knowl and the transfer of knowledge about
edge transfer portfolio will together system influence
quality from the Maintenance to Development.
recipient's performance improvement, Three
such distinct
thatroleshigh
in the work units are salient to
intensity and low (i.e., routine-based the SPI effort: senior

portfolio) com management, project manage
position, or low intensity and high (i.e.,
ment, andastaff.
direction
Senior managers establish the overall
based portfolio) composition will yieldstrategic goals of the IS department as well as the
performance
improvement. tactical goals of their respective units. Project man
agers are responsible for day-to-day management of
Hypothesis 6 (H6). The intensity of the knowledge
the software projects in their respective units. Mem
transfer portfolio interacts with the composition of the
bers of the technical staff perform the requirements
knowledge transfer portfolio to affect the performance
gathering, programming, testing, and implementation
improvement of the recipient, such that at a lower inten
tasks for the software projects. Managers in the IS
sity a portfolio with a greater use of directions will be
department were interested in our study because they
more effective, while at a higher intensity a portfolio with
wanted to improve software development and main
a greater use of routines will be more effective.
tenance processes. They viewed the use of quality
improvement techniques, software metrics, and best
Methodology practices in development and maintenance as essen
We empirically evaluated our conceptual model in a tial for improved performance. Although a vision
detailed field study of an SPI initiative in an IS depart for SPI had been articulated, the managers were
ment within a large firm. The IS department supports concerned about whether this initiative would have
all centralized computing activities for the firm and its intended impact on performance throughout the
is located at headquarters. It is organized using the department.
"L" organizational design (Swanson and Beath 1990)
in three distinct work units: Development, Mainte Data Collection and Coding
nance, and QA. The work units in an "L" organiza Data were collected over one year, beginning when
tional design are distinct but interdependent. Devel the senior managers had articulated their quality
opment designs and implements new IS and major visions, and the SPI initiative commenced. During
enhancements to existing systems supported by Main this time period, we tracked knowledge transfers via
tenance, while Maintenance maintains systems that observation, interviews, meetings, mail, documents,
Development has built, makes minor enhancements and manuals. Our data analytic strategy involves a
to them, and supports the users of the systems. mixed-methods approach, with a quantitative assess
QA assesses the quality of IS that are developed ment of the qualitative data. To support this strategy,
and maintained, implements software quality initia we converted the qualitative information obtained
tives, and publishes and disseminates software qual from the various data sources into numerical codes
ity information to Development and Maintenance. and numbers that could be statistically analyzed;
Although the work units are interdependent, they Tashakkori and Teddlie (1998) refer to this data trans
do not have significant overlapping knowledge. formation process as "quantitizing." The following
Indeed, separating Development, Maintenance, and sections describe our data collection and quantitizing
QA into different work units could erect knowledge process.
barriers (for example, the units use different project Semistructured interviews were conducted in the
management systems, metrics, and software engineer three work units within the IS department. In par
ing tools). At the same time, the interdependencies ticular, all senior managers, project managers, and
between the work units motivate the need for knowl QA staff were interviewed. In addition, we randomly
selected 12 staff members (6 in Development and
edge transfer across units to facilitate software process

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
6 in Maintenance) to interview. Before conducting

Figure 2 A Coded the
Knowledge Transfer
interviews, protocols were developed that included

"My team [Source: Development, Staff X, Y, and Z] reports and
questions about the SPI initiative. Interviews were
analyzes development defects every month. We use a Tsuda
either tape-recorded or copious notes were
matrix taken dur
[Mechanism (Directions): Tsuda Matrix] to determine
whether there
ing the interview and were transcribed immediately are any patterns in the defects that can be traced to
development processes. For example, last month my team found
afterward. The observation process that revealed addi
test plans were inconsistent and often not written. Based on
tional behaviors and environmental conditions (such
this, I [Recipient: Development, Project Manager M] developed
a test plan related
as the posting of SPI data on bulletin boards) template."
Interview Transcript, {name redacted), Development,
to knowledge transfer. The first author, as a passive
Project Manager
observer, attended a number of staff meetings about
CODING
SPI, software metrics counting meetings, and soft
Portfolio Portfolio
ware quality training sessions during the data collec
Observation Dyad intensity composition
tion period. Detailed and extensive notes were taken
during these meetings and sessions, and
1 they were
XM 1 1.00 (directions)
2 YM 1 1.00 (directions)
transcribed immediately afterward. The
3
first author
ZM 1 1.00 (directions)
also kept a daily diary to record general observations
about the organizational environment.
To supplement and validate the interviewCategorizing Knowledge
and Transfer
ob Mechanisms.
In the second
servational data, relevant archival data (documents) round of coding, a research assis
tant who had not participated in the data collection
that spanned the time period of the SPI initiative
identified the particular mechanisms used in each
were collected. Examples include a quality metrics
knowledge transfer. The mechanisms include a broad
white paper that defines particular metrics, identifies
range such as training, observation of experts, trans
who uses them, and describes how they are used,
ferring of personnel, meetings, reports, standards,
and memos about the SPI initiative that describe
and IS. In all, 69 unique mechanisms were used in
the implementation of various improvements (such
the knowledge transfers. A representative sample of
as training on quality or the use of new templates).
those mechanisms is shown in Table 1.
We created an annotated electronic bibliography We initiallyin categorized each of the mechanisms
chronological order to briefly identify and describe
as either "directions" or "routines," following Grant
each archival document.
(1996) and Nonaka and Konno (1998). Specifically,
Identifying Knowledge Transfers. In the firstmechanism was coded as a direction if it rep
a
resented the codification of tacit knowledge into
round of coding, we examined the interview, observa
explicit rules and instructions that provide guidance.
tion, and archival data, and collaboratively identified
Examples include written rules, formal procedures,
each knowledge transfer relating to SPI. In identifying
formulae, IS, policies, guidelines, templates, check
knowledge transfers, we did not include regular commu
lists, formal training, plans, and directives. A mech
nications between a source and recipient. Rather, following
anism was coded as a routine if it represented an
Argote (1999), we ensured that recipients applied the
implicit transfer of knowledge without explicit com
knowledge. Consider an example shown in the munication
top such as through interaction or actual
half of Figure 2 where a recipient receives informa
performance of a task. Examples include an actual
tion about defect patterns, and applies the knowledge
activity or process, commonly understood roles, a
to develop a test plan template. Transmitting infor
demonstration, a meeting, or informal procedures.
mation about defects does not by itself constituteWea then asked an independent coder to also catego
knowledge transfer, but understanding the cause rize
and the knowledge transfer mechanisms to provide
effect relationship between defects and development
an independent assessment of our coding. The coder
processes and applying that knowledge to devise wasa given a list of the mechanisms and the coding
test plan template does. instructions and coded the mechanisms. The coder

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
knowledge
Table 1 Mechanisms for Knowledge Transfer transfer using organization charts and
Mechanism Category How used seating charts to distinguish personnel. A unique code
was assigned to each individual in the IS department,
Defect reporting Direction
form Used to document and describe
of which there were 73 at the time of our study (the
defects that developers observe in
their own or others' software CIO, the Software Development Director, the Software
development processes and to Maintenance Director, the QA Director, 3 Project Man
suggest causes and solutions.
agers in Development and 3 in Maintenance, 32 Staff
Development Routine An ongoing meeting in which
attendees review and discuss
in Development, 29 Staff in Maintenance, and 2 Staff
procedures meeting
defects reported by developers in QA). Any of these individuals could assume the
and brainstorm solutions to "source" or "recipient" role in a knowledge transfer.
improve development processes.
Based on the organization chart, the research assis
MIS dashboard Direction A bulletin board next to the senior
tant identified the individual's work unit (Develop
display managers' offices that displays
reports depicting the quality ment, Maintenance, or QA) and role (Senior Manager,
of systems and processes (for Project Manager, Staff). In a few situations, an indi
example, a graph on defect trends). vidual described a knowledge transfer between other
Quality training class Direction A class held by QA to train developers
individuals in the IS department or the recipient of
and maintainers on total quality
management concepts and how the the knowledge transfer was not immediately obvious.
concepts apply to software In these situations, the first author aided the research
development and maintenance. assistant in identifying the individuals involved.
Service-level IS Direction A set of database queries that
periodically calculates the uptime Source-Recipient Dyad as the Unit of Analysis.
for all applications in operation and The unit of analysis in this study is the source-recip
reports on trends in the
ient dyad, i.e., a unique source-recipient pair. For
percentage of uptime for different
applications. Used to identify example, information about dyad AB [source A and
problematic applications and to recipient B] is an observation, information about dyad
trace the impact of process BA [source B and recipient A] is an observation, infor
improvements.
mation about dyad AC [source A and recipient C] is
System turnover Direction A set of metrics against which
standards
an observation, and so on. Note that AB and BA are
important characteristics of a new
system created by Development different dyads because the sources and recipients are
are evaluated before the system different in each dyad.
can be transferred to Maintenance.
Measures of each dyad's portfolio characterize di
Transfer of personnel Routine Experienced personnel are transferred
mensions of the accumulated set of mechanisms used
and use their experience to help
develop better quality applications. for knowledge transfer between the respective source
Informal demo of Routine Developers observe other developers and recipient over the time period of the SPI initia
function point using function points to count tive. Alavi (2000) argues that contextualized knowl
counting their projects.
edge takes time to transfer using rich mechanisms
such as collaboration and mentoring; prior empir
agreed with our categorizations of the mechanisms in
ical studies have found that multiple mechanisms
all instances but one. The disagreement over the cod
are often needed to transfer such knowledge (Carlile
ing of that mechanism was resolved by going back
2004, Rulke et al. 2000). Therefore, to examine the way
to the transcript of an interview in which the mech
in which a source uses a group of mechanisms in
anism was described to better understand its context
a portfolio to transfer knowledge to a recipient, we
of use. Based on the description in the transcript, the
focus our analysis on the accumulated set or port
mechanism was categorized according to our original
folio of knowledge transfer mechanisms. Portfolios
categorization. Table 1 shows the category to which
included for analysis are those that result in a recip
each of the representative mechanisms was assigned.
ient's application of knowledge from a source. This
Identifying the Source and Recipient. A research
application might occur after one exchange between
assistant identified the source and recipient for each
source and recipient, or after many such exchanges.

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
For example, if source A uses three were:

mechanisms (two
"To what extent do you agree with the following
directions and one routine) to transfer
statements:
the knowledge
Because {subordinate} uses software process
and recipient B applies the knowledge, dyad
improvements: AB's
(1) {subordinate} has improved the quality
portfolio has an intensity of three of
andhis a
or composition
her work, (2) {subordinate} is more productive in
of 67% directions. his or her work, (3) {subordinate} is more effective in his
Consider the knowledge transfer shown in Figure 2,or her work." For each question, a 7-point scale was
where the source is Development Staff. For the sakeprovided that ranged from 1 = strongly disagree to
of illustration, assume there are three development 7 = strongly agree. Having the immediate supervisor
staff involved in this knowledge transfer (with unique
evaluate the effectiveness of subordinates strengthens
identifiers of X, Y, and Z) and one project manager the internal validity of the study because it avoids
(with a unique identifier of M). We would code this the self-report bias that can affect performance ratings
as three dyads: XM, YM, and ZM. The knowledge(Crampton and Wagner 1994). The quality, productiv
transfer mechanism (Tsuda matrix) would be coded
ity, and effectiveness measures were highly correlated
as Directions as it represents the codification of tacit
knowledge about the meaning and significance of
Table 2 Construct Measurement
defects into explicit rules and instructions to provide
Construct Measure
guidance on testing. If this knowledge transfer is the
only transfer between these dyads, the portfolio inten
Performance Performance rating by immediate supervisor of the
sity would be one for each dyad, and the compositionimprovement recipient: "To what extent do you agree with the
following statements: Because {subordinate} uses
would be 100% directions.
software process improvements: (1) (subordinate!
has improved the quality of his or her work,
Construct Measurement (2) {subordinate} is more productive in his or her
Table 2 summarizes the measures of the constructs work, (3) {subordinate} is more effective in his
or her work." For each question, a 7-point scale
in the model. The measures of the control variables
was provided that ranged from 1 = strongly
are straightforward, and are summarized in this table. disagree to 7 = strongly agree.
We describe in more detail the measures of our Portfolio intensity The number of knowledge transfers from a particular
key constructs in the model: performance improve source to a particular recipient.

Portfolio The ratio of the number of directions in a source and
ment, portfolio intensity and composition, and the
composition recipient's portfolio divided by the total number
organizational design characteristics. We measured of knowledge transfers in their portfolio.
the SPI outcome of knowledge transfer mechanism
Proximity Number of cubicles between a source and recipient.
Note that cubicles are all the same size with the
portfolio choices—Performance improvement—by ask
exception of managers', which are twice as big.
ing the immediate supervisor of each recipient to
Nature of Dichotomous: 1 = source and recipient are in a
rate the extent to which the individual improved
relationship hierarchical (superior-subordinate or
the quality, productivity, and effectiveness of his or subordinate-superior) relationship; 0 = source
her work, because of that individual's adoption of and recipient are peers.
Work unit Dichotomous: 1 = source and recipient are in
SPI (i.e., Project Managers evaluated their respective
different work units; 0 = source and recipient are
Staff, Senior Managers evaluated the Project Man in the same work unit.
agers in their respective work units, andIS the ChiefNumber of years of IS work experience of the
work experience
Information Officer evaluated the three Senior Man source or recipient.
agers). Because SPI requires the institutionalization

Organizational Number of years that the source or recipient has
tenure worked for the firm.
of new work practices (Ravichandran and Rai 2003),
Education Highest level of education completed by t
it is unlikely that enhanced performance because of
or recipient: 1 = high school; 2 = associate's;
process improvement would be observable early in 3 = bachelor's; 4 = master's; 5 = Ph.D.
the SPI initiative but would most likely surface overGender Dichotomous: 1 = female, 0 = male. Refers to the
time. Thus the performance ratings were completed source's or recipient's gender.
at the end of our study, one year after the commence Prior work Dichotomous: 1 = source and recipient have worked
relationship together, 0 = otherwise.
ment of the SPI initiative. The specific questions asked

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
Analysis
(a = 0.95), so we averaged the three and Results
items together to
yield one measure of performance Ourimprovement.
coding identified 1,395 knowledge transfers be
The knowledge transfer mechanism portfolio
tween 355 unique source-recipient dyads. On average,
choices include intensity and composition. Our mea
there were 3.93 knowledge transfers per dyad (com
sure of Portfolio intensity captures the
puted frequency
as the of total number of knowl
average of the
knowledge transfers about SPI in edge
each dyad's
transfers portfo
between each of the 355 dyads), with a
lio and is assessed by counting the number
minimum of and
of 1 transfer knowl
a maximum of 11 transfers
edge transfers from a particular per
source
dyad. In to a particu
terms of the composition, the average
lar recipient over the time period of the
portfolio study.
consists of 73% Our
directions and 27% routines.
measure of Portfolio compositionSomeassesses the propor
source-recipient dyads (178) use portfolios with
tion of directions versus routines in a dyad's portfolio
no routines and 100% directions; some dyads (33) use
of knowledge transfers, and is expressed
portfolios withas100%
the ratio
routines and no directions; the
of the number of directions in the dyad's portfolio
remaining dyads (144) use a mix of directions and
divided by the total number of knowledge transfers
routines in their portfolios. Table 3 shows the descrip
in the dyad's portfolio. For example, ifand
tive statistics the portfo
correlations for the variables in our
lio of knowledge transfers for a source-recipient dyad
analysis.
includes eight transfers, and two of these transfers use
We analyzed the data using a simultaneous equa
directions while six use routines the intensity of the
tions framework of the form: j/l( = Y,( y + Xlj;/3 + /i, +
dyad's portfolio is 8, and it is composed of 25% (2/8)
directions. Vjj, where y,;- is the dependent variable (the perfor
mance improvement of recipient j in source-recipient
The antecedents of knowledge transfer mechanism
dyad ij), Yi; is a vector of observations for the endoge
portfolio choices include the nature of the relation
nous variables (the intensity and the composition of
ship, the proximity, and the work units of each source
knowledge transfer portfolios selected by source i
recipient dyad. The Nature of the relationship between
and recipient j), which are, in turn, predicted by
the source and recipient is a dichotomous variable
Xj;-—a vector of observations for the exogenous vari
that is set to "1" if the source and recipient are in a
ables (the nature of the relationship, work unit, and
hierarchical relationship (i.e., supervisor-subordinate
or subordinate-supervisor), and "0" otherwise. We proximity of source i and recipient j and the con
measured the Proximity between each source and trol variables), and y and (3 are vectors of coeffi
recipient dyad using a scaled, architectural render cients to be estimated in the different equations. The
ing of the IS department's floor plan. The distance equation for performance predicts the performance
between a source and recipient was measured by improvement of each recipient in a source-recipient
counting the number of cubicles between them, fol dyad based on the dyad's knowledge transfer port
lowing the corridors between room sections in the folio choices for SPI: Performance improvementj;- = yQ +
floor of the building where the IS department is ^(Portfolio intensityi;) + ^(Portfolio composition^) +
housed. Except for senior managers (whose cubicles y3(Portfolio intensity* Portfolio composition^) + /a, +
are twice the size of those for other personnel), all Vjj. The equations for knowledge transfer portfolio
cubicles in the department are square and of the same choices predict the intensity and composition of the
length; thus, counting cubicles measures the actual portfolios selected by source-recipient dyads, based
physical distance between the source and recipient on the nature of the relationship between the source
in each dyad. Measuring proximity in terms of the and recipient, the work units of the source and recip
actual distance between sources and recipients is con ient, and the proximity of the source and recipient,
sistent with literature on communication (e.g., Allen and controlling for source and recipient demographics:
1977). The Work unit construct is measured using aPortfolio intensity^ or Portfolio compositioni; = /30 + /3, (IS
dichotomous variable that is set to "1" if the source experience,) + f32 (Org tenure,) + p3 (Education,) +
and recipient are in different work units and is f3i(Gender,)
"0" if + /85(IS experience^) + /36(Org tenure;) +
they are in the same work unit. /37(Education;) + /38(Gender;) + /3(J(Prior worki;) +

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
Table 3 Descriptive Statistics and Correlations
Variable name Mean Std. dev. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)
1. Portfolio 3.93 3.07 -0.23"
intensity, 0.73 0.34 -0.23**
composition
2. Proximity 29.65 31.85 0.47** 1.00
-0.41**
3. Nature of 0.39 0.49 0.16** 0.33** 1.00
relationship3 -0.14*
4. Work unitb 0.52 0.50 0.05 -0.51** -0.82** 1.00

0.42**
5. IS experience 10.2810.20 0.11* 0.45" 0.69" -0.57" 1.00

(source) -0.08
6. Org tenure 7.82 6.54 -0.10 0.50" 0.47" -0.48" 0.76" 1.00
(source) 0.01
7. Education0 3.09 0.49 -0.03 -0.11* -0.39" 0.44" -0.11* 0.15* 1.00
(source) 0.40**
8. Gender" 0.34 0.47 0.42" 0.13* 0.11* -0.01 0.14* -0.02 —0.18* 1.00
(source) -0.21"
9. IS experience 12.71 8.53 0.18** 0.11* 0.16" -0.11* 0.08 0.03 -0.08 0.10+ 1.00
(recipient) 0.06
10. Org tenure 8.16 5.60 0.12* 0.08 0.14* -0.08 0.10+ 0.06 -0.08 0.07 0.62" 1.00
(recipient) 0.06
11. Education0 2.85 0.74 -0.01 -0.02 -0.01 -0.04 0.01 0.02 -0.03 -0.07 —0.22*' -0.29" 1.00
(recipient) -0.06
12. Gender" 0.38 0.49 0.01 -0.06 -0.02 -0.02 -0.03 0.01 0.02 0.01 -0.05 -0.14* 0.08 1.00
(recipient) -0.01
13. Prior work 0.03 0.17 -0.19" 0.16* -0.14* -0.18*' 0.14* -0.01 -0.04 -0.13* -0.07 -0.08 0.01 -0.01
-0.30"
14. Performance 4.06 1.77 -0.14" -0.08 -0.12* 0.10+ -0.07 -0.05 0.06 -0.14* —0.25* -0.22** 0.01 -0.24"
-0.08 -0.25*'
Notes, n = 355. ]p < 0.10; *p < 0.05; "p < 0.01. Coding: a1 = hierarchical, 0 = nonhierarchical; b1 = different unit, 0 = same unit; °1 = high school,
2 = associate's, 3 = bachelor's, 4 = master's, 5 = Ph.D.; d1 = female, 0 = male. Pearson correlations are reported between pairs of continuous variables,
Spearman correlations between pairs of continuous and dichotomous variables, and Phi correlations between pairs of dichotomous variables.
/310(Nature of relationship,^) + fiu (Work unit;/) ent + but different sources. As noted, the knowledge
/312(Proximityj;) + S, + ej;-. transfer mechanism portfolio choices are endogenous
The natural logarithm of portfolio intensity is usedin this formulation. Thus we used a generalized two
in the portfolio intensity equation, as suggested by thestage least squares (G2SLS) random effects instrumen
Box-Cox procedure (1964) to achieve a more symmet tal variables regression procedure to estimate the sys
rical distribution of the error terms. Also, every varitem of equations for the data panel (Greene 2003).
able was standardized to its Z-score before entering Diagnostics suggested no problems with multi
it into the analysis to ease interpretation of estimatedcollinearity in any of the equations, as the condi
coefficients (as each variable is measured using a dif tion indices and variance inflation factors are all less
ferent unit scale), and to mitigate potential collinearthan 10, well within acceptance levels. In addition,
ity between the main effects of portfolio intensity we found no outliers in the Performance improve
and composition and their interactions (Aiken and ment equation, five outliers in the Portfolio composi
West 1991). Finally, we added random effects to the tion equation, and three outliers in the Portfolio inten
equations to address the nested data, i.e., the possity equation. However, the results do not change
sible correlation of residuals across dyads with the materially when the analyses are conducted without
same source but different recipients or same recipi the outlying observations, and we thus report the

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
ent work
Table 4 Results for Portfolio Intensity and Portfolio units (/3U
Composition = 0.70, z = 14.70, p < 0.01). Sim
Portfolio intensity
ilarly,
Portfolio composition H1B is supported as more directions are used
in the portfolio when the source and recipient are in
Coefficient Coefficient
Variable name Z-value Z-value
different work units (/3n = 1.17, z = 22.71, p < 0.01).
(Std. error) (Std. error)
Table 4 also shows that H2A is supported as a source
0.054 2.74" -0.200 -11.99*"
Constant [/30]
and recipient in a hierarchical relationship choose
(0.020) (0.017)
0.184 3.67*** -0.089 -2.50*
a more intense knowledge transfer portfolio than a
IS experience
(source) [0,] (0.050) (0.035) dyad not in such a relationship (J3U] = 0.84, z = 17.11,
Org tenure -0.547 -11.16*" 0.331 11.36"* p < 0.01). H2B is supported as a source and recipi
(source) [ft] (0.049) (0.029) ent in a hierarchical relationship use more directions
Education -0.028 -1.14 0.303 12.80***
in their knowledge transfer portfolio (/310 = 1.07, z =
(source) [ft] (0.025) (0.024)
21.01, p < 0.01). Finally, H3A is supported as a more
Gender (source) [ft] 0.147 7.77*** -0.062 -3.35"
(0.019) (0.018) proximate source and recipient transfer knowledge
IS experience 0.020 1.08 0.057 3.90*** more intensely than a source and recipient who are
(recipient) [ft] (0.018) (0.015) less proximate (/312 = 0.79, z = 33.73, p < 0.01). A more
Org tenure -0.002 -0.07 -0.030 -1.74+ proximate source and recipient also use more routines
(recipient) [ft] (0.021) (0.017)
in their knowledge transfer portfolio as predicted in
Education 0.018 1.04 -0.004 -0.27
H3B (/312 = —0.39, z = —16.62, p < 0.01). Together, the
(recipient) [ft] (0.017) (0.013)
0.038 2.44* -0.006 -0.41 nature of the relationship, proximity, and work units
Gender (recipient) [ft]
(0.016) (0.014) of the source and recipient explain substantial vari
Prior work [ft] -0.005 -0.24 0.045 2.03* ation in the design of knowledge transfer portfolios
(0.022) (0.022) (explaining 30% of the variation in the intensity of
Nature of 0.836 17.11*** 1.067 21.01***
the portfolio selected, \2 = 342.30, p < 0.01; explaining
relationship [ft0] (0.049) (0.051)
0.696 14.70*** 1.165 22.71***
35% of the variation in the composition of the portfo
Work unit [ft,]
(0.047) (0.051) lio selected, x2 — 234.02, p < 0.01).
Proximity [ft2] 0.787 33.73*** -0.390 -16.62*** Our final set of hypotheses concerns the perfor
(0.023) (0.023) mance effects of knowledge transfer portfolio choices.
Deviance 5,045.97™ 7,400.34"* Table 5 shows the results. To evaluate H4, we differ
difference Or2)
Pseudo-fl2 0.670 0.615
entiated the Performance improvement equation with
respect to portfolio intensity, and examined the change
Notes, n = 355. tp < 0.10; *p < 0.05; "p < 0.01; —p < 0.001. Dependent
in performance improvement with a unit change in
variable is the natural logarithm of portfolio intensity in the portfolio intensity
equation. Estimates obtained via generalized least squares regression (with intensity, holding portfolio composition constant at
a random effect for each source and recipient in a dyad and a correction for
heteroskedastic panels). Table 5 Results for Performance Improvement
Variable name Coefficient (Std. error) Z-value

results with all observations included. Finally, het
Constant [y0] 3.963 11.14—
eroskedastic error structures were found in the Port
(0.356)
folio composition and Portfolio intensity equations,
Portfolio intensity [^] 1.422 2.47*
and we corrected for heteroskedasticity in the estima (0.576)
tion of those equations. Table 4 presents the statistical
Portfolio composition [72] 0.770 1.61*
(0.479)
results from the estimation of the equations for port
Portfolio intensity -2.560 -3.46***
folio intensity and portfolio composition.
x Portfolio composition [y3] (0.740)
The results suggest strong support for our hypothe 17.65"*
Deviance difference (x2)
ses predicting the intensity and composition of
Pseudo-/?2 0.380
knowledge transfer portfolio choices. As shown in

Notes, n = 355.< 0.10; *p < 0.05; "p < 0.01; *"p < 0.001. Estimates
Table 4, H1A is supported as a more intense portfolio
obtained via G2SLS regression (with a random effect for each source and
is used when the source and recipient are in differ
recipient in a dyad).

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
its mean value (Greene 2003). We find

Figure 3 Portfoliothat
IntensityH4 is
x Composition Interaction
supported as a recipient of a more intense knowl

edge transfer portfolio has a greater performance gain
from SPI than a recipient of a less intense portfo
lio, when evaluated at an average use of directions
in the portfolio (dPerformance/dPortfolio intensity =
7i + 73(Portfolio composition^) = -y, evaluated at the
mean value for portfolio composition = 1.42, z = 2.47,
p < 0.05).
Using a similar approach, we evaluate the main
effect of portfolio composition (H5). We find sup
port, albeit somewhat weaker, for this hypothesis as a
recipient of a knowledge transfer portfolio with more
directions achieves a greater performance gain from
SPI than a recipient of a portfolio with more rou
tines, when evaluated at average levels of intensity in For our sample, a sensitivity analysis suggests tha
the portfolio (3Performance/<9Portfolio composition = at portfolio intensity equal to 4.85 transfers per dya
(the point on the graph where all three lines intersect
y2 + ^(Portfolio intensityi;) = y2 when evaluated at
the mean value of zero for portfolio intensity — 0.77, a portfolio composed of 63% directions and 37% rou
z = 1.61, p < 0.10). Finally, the coefficient of the inter tines is effective (the average portfolio has 73% dire
action of portfolio intensity and composition is nega tions and 27% routines). However, when portfol
tive and significant (y3 = —2.56, z = —3.46, p < 0.001), intensity is below or above 4.85 transfers per dyad,
suggesting support for H6. We discuss our findings then different compositions of directions and routin
and their implications next. are more effective. When portfolio intensity is great
than 4.85 transfers per dyad, portfolios compose
Discussion of more routines (>37% routines) are more effective
Several important insights can be gleaned fromwhile

ourat portfolio intensity less than 4.85 transfers p
dyad, portfolios composed of more directions (>63%
results. First and foremost, our findings indicate that
how organizational members transfer software pro are more effective. Note also, that in our
directions)
cess knowledge affects the level of individualsample,
perfor the highest performance gains (performanc
rating
mance gains from the SPI initiative. Both intensity of seven) are achieved for dyads whose por
and composition are significant dimensions of knowl composed of 100% routines, with an inten
folios are
sity level
edge transfer portfolios, and some combinations of of seven knowledge transfers. The highest
intensity and composition are more effective performance
than oth rating that can be achieved for dyad
whose
ers. Figure 3 graphs the interaction between portfo portfolios are composed of 100% directions is
5.86 (this
lio intensity and composition. As shown in Figure 3, occurs when their portfolio has the lowes
with greater portfolio intensity, a portfolio intensity
with a level of one transfer).
higher proportion of routines results in higher As Grant (1996) has suggested, directions are more
perfor
mance gains, while at lower levels of portfolio inten mechanisms for knowledge transfer becau
efficient
routines require significant amounts of repetitio
sity, a portfolio with a higher proportion of directions
is more effective.1 and interaction before they can be used effectively
Thus, it would follow that, at low portfolio inte
1 Note that "low" and "high" intensity correspond to one standard
sity, directions have a greater impact than routines
deviation below and above the mean intensity of the portfolio,
and this is consistent with our sensitivity analyse
respectively. For directions, "more directions" corresponds to one
However, to achieve higher levels of performanc
standard deviation above the mean percentage of directions in the
requires
portfolio, while "more routines" corresponds to one standard devi portfolios with a larger proportion of r
tines as well as greater intensity. Our finding th
ation below the mean percentage of directions in the portfolio.

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
the highest performance gains Table

ensue from
6 Integrated Results portfo
lios with 100% routines and a high level of inten Portfolio intensity
sity suggests that to achieve significant performance
Low High
improvements requires a more in-depth
Portfolio composition
understand
(<4.85 transfers) (>4.85 transfers)
ing, application, and "embedding" of software pro
More directions Nonproxlmate dyad Hierarchical
cess improvements in work practices (Ravichandran
(<37% routines) relationship
and Rai 2003), and that this embedding is more effec Different work units
tively realized when high intensity

More routines portfolios com
Nonhierarchicai Proximate dyad
posed of routines are used. (>37% routines) relationship
Same work unit
Our analysis of performance suggests the impor
tance of knowledge transfer mechanisms for software
process improvements. It also reveals the optimal
of more directions or at higher intensity with a port
combinations of intensity and composition in dyads'
folio composed of more routines.
knowledge transfer portfolios to maximize the perfor
Also shown in Table 6 are the organizational design
mance gains from SPI. A natural follow-on question is
choices (i.e., proximity, hierarchy, work unit structure)
to ask how to achieve these optimal levels of portfolio
that correspond to the intensity and composition lev
design. Our analysis of the antecedents of portfolio
els of the portfolio. Table 6 suggests that, for our
design choices provides insight into this question. In
sample, the only design choice that reflects an opti
particular, we find that organizational design choices
mal knowledge transfer portfolio structure (in terms
strongly affect the nature of the knowledge transfer
of intensity and composition) is proximity: distant
portfolios used by dyads. Our results indicate that a
source-recipient dyads use directions infrequently,
more intense portfolio of knowledge transfer mecha
while proximate source-recipient dyads use routines
nisms is used in SPI when the source and recipient are
frequently. This suggests that proximity facilitates
proximate, when they are in a hierarchical the choice of a portfolio thatrelation
achieves composition—
ship or when they work in different units. In addition,
intensity fit, which, in turn, has a positive impact on
we find that more directions are used in SPI when the
performance.
source and recipient are farther apart, when they areIn contrast, those dyads in a hierarchical relation
in a hierarchical relationship, or when they work ship,in or those who work in different units, use too
different units.
much intensity given the level of directions in their
Our findings highlight the role that organizational portfolio composition, as seen in Table 6. That is, these
design decisions play in the transfer of knowledge sources are transferring knowledge via directions in
that is complex and contextual such as in SPI. Link an intense manner, but this portfolio design is inef
ing our results from the performance and portfolio ficient in terms of performance impact. For example,
design equations reveals how organizational design sources may be issuing frequent memos (directions)
decisions can be structured to yield higher perfor more often than is required to transfer the knowl
mance gains from SPI. Table 6 shows how organiza edge. The impact of this portfolio on the performance
tional design decisions in this firm relate to portfolio improvement of the recipient is not as successful as
choices and ultimate performance impacts. The opti it could be. Table 6 suggests that to increase perfor
mal cells in the table indicating the highest perfor mance, dyads in a hierarchical relationship or in dif
mance gains are highlighted in bold borders. Recall ferent work units can reduce the frequency of their
from our sensitivity analysis that portfolios lower knowledge transfers, given the high proportion of
in intensity (<4.85 transfers) and high in composi directions in their portfolio.
tion (>63% directions), or high in intensity (>4.85 Similarly, Table 6 indicates that dyads who are
transfers) and low in composition (>37% routines), peers, or who work in the same unit, use routines
are most effective. That is, high performance can be but too infrequently, resulting in suboptimal perfor
achieved at lower intensity with a portfolio composed mance improvement. For example, a developer on

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
We identified
one project team may make a one-time intensity and composition as two sig
demonstration
of an improved design technique to another
nificant dimensions of devel
knowledge transfer portfolios,
oper on a different project team.using the typology offered
However, becauseby Grant (1996) to analyze
portfoliofrequent
the recipient developer does not make composition. While
obser this typology is theoret
ically based understand
vations, he does not gain an in-depth and is well suited to the SPI context, it
ing of the knowledge, and thusdoes doesconstrain
notthe categorization
see many of mechanisms into
gains in his performance. Tabledirections
6 suggestsor routines.
thatFuture
toresearch could explore
increase performance, dyads that are peers or in the transfer portfolios,
other dimensions of knowledge
suchintensity
same work unit should increase the as the ratio of formal to informal mechanisms or
of their
of codified
knowledge transfers, given the high levelto personalized
of routines mechanisms, or the timing
in their portfolio. of the mechanisms deployed in the portfolio. Other
contexts should be studied as well.
SPI has been identified as an activity that is compli
In addition,
cated and difficult, and it can require years our study provides insight into the
of effort
for an organization to advance in effectiveness
software of different
capabil knowledge transfer portfolio
ity maturity (e.g., Harter et al. 2000). Further, SPI is goal of SPI is to improve
designs. Because the ultimate
an initiative that often fails to achieve individual andobjectives;
its organizational performance, linking
Hardgrave and Armstrong (2005) performance cite failure to portfolio
rates design
of choices demonstrates
the importance
up to 70%. Our findings suggest a practical strategy of knowledge transfer in this con
text.
for improving the performance gains from SPI. First, Further, our results imply that the fit between
the performance impact from the SPI composition
shouldand beintensity
mea of a dyad's knowledge
transfer
sured after some period of time. Managers should portfolio impacts performance improvement.
then identify where the highest performance gains demonstrates the signifi
Thus our analysis not only
are occurring and relate these cance gains of studying
to theknowledge
knowltransfer portfolios, but
edge transfer portfolios used to achieve also suggests the importance
them. Finally,of conducting further
research
managers should make any necessary modifications to deepen our understanding of the effective
to organizational design parameters to facilitate the transfer portfolio choices.
ness of different knowledge
selection of those portfolios. Finally, our findings highlight the importance of
organizational design decisions in influencing choices
of knowledge transfer portfolio intensity and com
Conclusions position, and thereby performance. While proximity,
Our objective in this study is to understand how orgaof relationship, and work units were found
nature
nizations can be more successful in improvingtotheir
be influential in our study, there are undoubtedly
software development processes. We approached this organizational design parameters that could
other
question from the perspective of knowledge transfer,
influence portfolio choices such as the span of control.
considering how software process knowledge is trans
Future research is needed to determine the most effec
ferred between source and recipient dyads tive
within
combination of mechanisms in a portfolio, given
an IS department, and relating how knowledge is
differences in organizational design and workplace
characteristics.
transferred to the recipients' performance gains from Individual differences could also moti
process improvement. In concluding, we note vate
the the
conchoice of different portfolios, but such factors
tributions and limitations of this study and suggest
may be more challenging for managers to influence.
directions for future research. Further research on other aspects of knowledge
Our study theorizes and provides empirical evitransfer portfolios would be instructive. For exam
dence about the design choices for a portfolio ofple, we were not able in a field setting to isolate and
knowledge transfer mechanisms. This focus on port measure different dimensions of the knowledge trans
folios offers a unique and important contribution toferred (such as amount and type). Experimental stud
the IS literature because it highlights how dyads uti ies that control for the nature of the knowledge being
transferred may be helpful to confirm and extend the
lize a set of mechanisms to transfer knowledge for SPI.

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
findings from the field. In addition, some

Argote, L. 1999.research
Organizational has
Learning: Creating, Retaining and
Transferring Knowledge. Kluwer Academic Publishers, Boston,
suggested that politics can play a MA. role in knowledge
transfer (e.g., Carlile 2004). However,
Argote, L.,our study
P. Ingram. was transfer: A basis for compet
2000. Knowledge
not designed to capture data about power
itive advantage inrelations,
firms. Organ. Behav. Human Decision Processes
82(1) 150-169.
and we cannot discern how political considerations
Birkinshaw, J., R. Nobel, J. Ridderstrale. 2002. Knowledge as a con
may have influenced knowledge transfer inDo this
tingency variable: con of knowledge predict
the characteristics
organization structure?
text. This would be an interesting direction forOrgan. Sci. 13(3) 274-289.
future
research. Box, G., D. Cox. 1964. An analysis of transformations. J. Roy. Statist.
Soc. B 26 211-243.
The implications of our findings for SPI transcend Carlile, P. 2004. Transferring, translating, and transforming: An
discrete software projects and instead speak to the integrative framework for managing knowledge across bound
aries. Organ. Sci. 15(5) 555-568.
need for a deep holistic understanding of softwareContu, A., H. Willmott. 2003. Re-embedding situatedness: The
practices and processes that span individuals and importance of power relations in learning theory. Organ. Sci.
14(3) 283-296.
their immediate responsibilities. Our findings high
Crampton, S., J. Wagner. 1994. Percept-percept inflation in microor
light the importance of carefully designing IS depart ganizational research: An investigation of prevalence and
ments so that knowledge is transferred effectively effect. J. Appl. Psych. 79(1) 67-76.
Crowston, K, E. Kammerer. 1998. Coordination and collective mind
across roles and units, facilitating the improvement
in software requirements development. IBM Systems ]. 37(2)
of software processes for the department as a whole. 227-245.
Additional research targeted at this metalevel can Darr, E., L. Argote, D. Epple. 1995. The acquisition, transfer, and
depreciation of knowledge in service organizations. Manage
yield further insights into how IS departments can ment Sci. 41(11) 1750-1762.
continually improve their ability to consistently buildDemsetz, H. 1991. The theory of the firm revisited. O. Williamson,
and deploy software that adds value to their firms. S. Winter, eds. The Nature of the Firm. Oxford University Press,
New York, 159-178.
Epple, D., L. Argote, R. Devadas. 1991. Organizational learn
Acknowledgments ing curves: A method for investigating intra-plant transfer of
knowledge acquired through learning by doing. Organ. Sci.
Prior versions of this research were presented in research 2(1) 58-70.
seminars at the University of Pittsburgh, University of Min
Galbraith, C. 1990. Transferring core manufacturing technologies in
nesota, and Carnegie Mellon University. The authors grate high technology firms. California Management Rev. 32(4) 56-70.
fully acknowledge the research assistance of Ranjit TinaikarGaudin, S. 2003. Study: Many major IT projects still fail. Data
and Mark Haney of the University of Pittsburgh for their mation (June 16). http://www.itmanagement.earthweb.com/
help with data coding. The authors thank Ronald Slaugh it_res / article, php /2222391.
ter for his detailed architectural rendering of the IS DepartGrady, R. 1997. Successful Software Process Improvement. Prentice
Hall, Upper Saddle River, NJ.
ment's floor plan. They are also grateful to the senior editor,
Grant, R. 1996. Prospering in dynamically-competitive environ
associate editor, and anonymous reviewers who provided ments: Organizational capability as loiowledge integration.
helpful comments and suggestions on earlier versions of Organ. Sci. 7(4) 375-387.
this manuscript. Greene, W. 2003. Econometric Analysis, 3rd ed. Prentice-Hall, Upper
Saddle River, NJ.
Grover, V., T. Davenport. 2001. General perspectives on knowledge
management: Fostering a research agenda. J. MIS 18(1) 5-21.
References Hansen, M. 1999. The search-transfer problem: The role of weak
Adler, P. 1990. Shared learning. Management Sci. 36(8) 938-957. ties in sharing knowledge across organization subunits. Admin.
Sci. Quart. 44(1) 82-111.
Aiken, L., S. West. 1991. Multiple Regression: Testing and Interpreting
Hardgrave, W., D. Armstrong. 2005. Software process improve
Interactions. Sage Publications, Thousand Oaks, CA. ment: It's a journey, not a destination. Comm. ACM 48(11)
93-96.
Alavi, M. 2000. Managing organizational knowledge. R. W. Zmud,
ed. Framing the Domains of IT Management. Pinnaflex Educa
Harter, D., M. Krishnan, S. Slaughter. 2000. Effects of process matu
tional Resources, Cincinnati, OH, 15-28. rity on quality, cost and cycle time in software product devel
opment. Management Sci. 46(4) 451-466.
Allen, T. 1977. Managing the Flow of Technology: Technology Transfer
Herbsleb, J., D. Zubrow, D. Goldenson, W. Hayes, M. Paulk. 1997.
and the Dissemination of Technological Information within the R&D
Software quality and the capability maturity model. Comm.
Organization. MIT Press, Cambridge, MA.
ACM 40(6) 30-40.
Argote, L. 1993. Group and organizational learning curves: Hinds,
Indi P., S. Kiesler. 1995. Communication across boundaries: Work,
vidual, system and environmental components. British J. Soc. structure, and use of communication technologies in a large
Psych. 32 31-51. organization. Organ. Sci. 6(4) 373-393.

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC
Ravichandran,
Ko, D., L. Kirsch, W. King. 2005. Antecedents T., A. Rai. 2000.
of knowledge Quality management in systems
transfer
from consultants to clients in enterprise development: An organizational system perspective. MIS
system implementa
tion. MIS Quart. 29(1) 59-85. Quart. 24(3) 381-415.
Kraut, R., L. Streeter. 1995. CoordinationRavichandran,
in software T., A.development.
Rai. 2003. Structural analysis of the impact
Comm. ACM 38(3) 69-81. of knowledge creation and knowledge embedding on software
Malhotra, A., A. Majchrzak, R. Carman, V. Lott. 2001. Radical
process capability. inno
IEEE Trans. Engrg. Management 50(3)
vation without collocation: A case study at Boeing-Rocketdyne.
270-284.
MIS Quart. 25(2) 229-249.
Rulke, D., S. Zaheer, M. Anderson. 2000. Sources of managers'
Mathiassen, L., P. Pourkomeylian. 2003. Managing knowledge in a
knowledge of organizational capabilities. Organ. Behav. Human
software organization. J. Knowledge Management 7(2) 63-80.
Decision Processes 82(1) 134-149.
McGrath,J., L. Argote. 2001. Group processes in organizational con
Schulz, M. 2001. The uncertain relevance of newness: Organiza
texts. M. Hogg, R. Tindale, eds. Blackwell Handbook of Social
tional learning and
Psychology: Group Processes. Blackwell Publishers, knowledge
Maiden, flows. Acad. Management J. 44(4)
MA,
603-627. 661-681.
Spencer, B. A. 1994. Models of organization and total quality man

Mitki, Y., A. Shani, Z. Meiri. 1997. Organizational learning mech
anisms and continuous improvement: A longitudinal study. agement: A comparison and critical evaluation. Acad. Manage
/. Organ. Change Management 10(5) 426-446. ment Rev. 19(3) 446-471.
Nelson, K., J. Cooprider. 1996. The contribution of shared knowl
Swanson, E., C. Beath. 1990. Departmentalization in software devel
edge to IS group performance. MIS Quart. 20(4) 409-432. opment and maintenance. Comm. ACM 33(6) 658-667.
Nonaka, I. 1994. A dynamic theory of organizational knowledge
Tashakkori, A., C. Teddlie. 1998. Mixed Methodology: Combin
creation. Organ. Sci. 5(1) 14-37.
ing Qualitative and Quantitative Approaches. Sage Publications,
Nonaka, I., N. Konno. 1998. The concept of "ba": Building a foun Thousand Oaks, CA.
dation for knowledge creation. California Management Rev. 40(3)
40-54. Teasley, S., L. Covi, M. Krishnan, J. Olson. 2002. Rapid software
Olson, G., J. Olson. 2000. Distance matters. Human-Comput. Interac development through team collocation. IEEE Trans. Software
tion 15(2/3) 139-178. Engrg. 28(7) 671-691.
Paulus, P., H.-C. Yang. 2000. Idea generation in groups: A basis Thomas-Hunt, M., T. Ogden, M. Neale. 2003. Who's really shar
for creativity in organizations. Organ. Behav. Human Decision ing? Effects of social and expert status on knowledge exchange
Processes 82(1) 76-87. within groups. Management Sci. 49(4) 464-477.

182.255.0.242 on Fri, 27 Nov 2020 04:27:54 UTC

This Content Downloaded From 182.255.0.242 On Fri, 27 Nov 2020 04:27:54 UTC

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

This Content Downloaded From 182.255.0.242 On Fri, 27 Nov 2020 04:27:54 UTC

Uploaded by

Copyright:

Available Formats

The Effectiveness of Knowledge Transfer Portfolios in Software Process Improvement:

Stable URL: https://www.jstor.org/stable/23015892

This content downloaded from

The Effectiveness of Knowledge Transfer

Introduction Process improvement of any kind—including SPI—

ever, firms often fail in their attempts to build tally

This content downloaded from

2003). Improvement emanates from aWe next and

Knowledge transfer is a dyadic exchange in which

This content downloaded from

single knowledge transfer event. Though

This content downloaded from

of a number of individuals. Hiswith

Figure 1 Research Model

Organizational design Knowledge transfer Software process

Controls (source and recipient):

Prior work relationship

This content downloaded from

This content downloaded from

difficult for people to share each formal other'shierarchy

This content downloaded from

leverage knowledge transfer mechanisms that rely

This content downloaded from

However, successful knowledge transfer,

This content downloaded from

through collaboration, shared experience,

intensity and low (i.e., routine-based the SPI effort: senior

This content downloaded from

6 in Maintenance) to interview. Before conducting

interviews, protocols were developed that included

This content downloaded from

This content downloaded from

For example, if source A uses three were:

key constructs in the model: performance improve source to a particular recipient.

agers). Because SPI requires the institutionalization

This content downloaded from

This content downloaded from

Table 3 Descriptive Statistics and Correlations

4. Work unitb 0.52 0.50 0.05 -0.51** -0.82** 1.00

5. IS experience 10.2810.20 0.11* 0.45" 0.69" -0.57" 1.00

This content downloaded from

Variable name Coefficient (Std. error) Z-value

knowledge transfer portfolio choices. As shown in

This content downloaded from

its mean value (Greene 2003). We find

supported as a recipient of a more intense knowl

Several important insights can be gleaned fromwhile

This content downloaded from

the highest performance gains Table

tively realized when high intensity

This content downloaded from

This content downloaded from

findings from the field. In addition, some

This content downloaded from

Spencer, B. A. 1994. Models of organization and total quality man

This content downloaded from

You might also like

4. Work unitb 0.52 0.50 0.05 -0.51 -0.82 1.00