can informal communication networks disrupt … · sosa, gargiulo, and rowles: can informal...

This article was downloaded by: [175.176.173.30] On: 02 August 2015, At: 23:47Publisher: Institute for Operations Research and the Management Sciences (INFORMS)INFORMS is located in Maryland, USA

Organization Science

Publication details, including instructions for authors and subscription information:http://pubsonline.informs.org

Can Informal Communication Networks DisruptCoordination in New Product Development Projects?Manuel E. Sosa, Martin Gargiulo, Craig Rowles

To cite this article:Manuel E. Sosa, Martin Gargiulo, Craig Rowles (2015) Can Informal Communication Networks Disrupt Coordination in NewProduct Development Projects?. Organization Science 26(4):1059-1078. http://dx.doi.org/10.1287/orsc.2015.0974

Full terms and conditions of use: http://pubsonline.informs.org/page/terms-and-conditions

This article may be used only for the purposes of research, teaching, and/or private study. Commercial useor systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisherapproval, unless otherwise noted. For more information, contact [email protected].

The Publisher does not warrant or guarantee the article’s accuracy, completeness, merchantability, fitnessfor a particular purpose, or non-infringement. Descriptions of, or references to, products or publications, orinclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, orsupport of claims made of that product, publication, or service.

Copyright © 2015, INFORMS

Please scroll down for article—it is on subsequent pages

INFORMS is the largest professional society in the world for professionals in the fields of operations research, managementscience, and analytics.For more information on INFORMS, its publications, membership, or meetings visit http://www.informs.org

http://pubsonline.informs.org

http://dx.doi.org/10.1287/orsc.2015.0974

http://pubsonline.informs.org/page/terms-and-conditions

http://www.informs.org

OrganizationScienceVol. 26, No. 4, July–August 2015, pp. 1059–1078ISSN 1047-7039 (print) � ISSN 1526-5455 (online) http://dx.doi.org/10.1287/orsc.2015.0974

© 2015 INFORMS

Can Informal Communication Networks DisruptCoordination in New Product Development Projects?

Manuel E. SosaTechnology and Operations Management, INSEAD, Singapore 137686, [email protected]

Martin GargiuloOrganisational Behaviour, INSEAD, Singapore 137686, [email protected]

Craig RowlesEi3 Corporation, Montvale, New Jersey 07645, [email protected]

This paper investigates how the structure of the informal communication network that results from efforts to coordinatetask interdependence between design teams in complex product development projects moderates the effect of task

interdependence on interteam communication. Drawing on theoretical mechanisms from the social network and knowledgetransfer literature, as well as on recent empirical advances in exponential random graphs models of social networks, weexamine how the presence of a common third party in the communication network affects the likelihood of technicalcommunication between interdependent teams designing the components of a large commercial aircraft engine. Althoughtask interdependence has a strong and significant effect on the likelihood of communication between teams, this effectis moderated by the presence of common third parties. The nature of this moderation depends on the position of thecommon third party within the triadic communication structure. When the common third party seats in the middle of acommunication chain between the potential source and the potential recipient of technical communication, its presenceincreases the likelihood of communication between these two teams. However, when the communication between the sourceand recipient can trigger cyclic exchanges between the three teams, the presence of the third party reduces the likelihoodof communication between the two interdependent teams, increasing the risk of coordination disruptions. We discuss theimplications of our findings on the literature of intraorganizational networks in new product development.

Keywords : interteam communication; new product development; social networks; triads; closure; exponential randomgraph models (ERGMs)

History : Published online in Articles in Advance April 16, 2015.

1. IntroductionSince the pioneering work by Allen (1977), organiza-tional scholars have highlighted the crucial role of infor-mal communication networks in new product devel-opment organizations. Because task-related (informal)interteam communication can help teams coordinateinterdependent tasks, a proper understanding of suchcommunication patterns has been recognized as a keyelement in improving the performance of new prod-uct development organizations (Tushman 1977, Tushmanand Katz 1980, Ancona and Caldwell 1992, Brown andEisenhardt 1995, Reagans and Zuckerman 2001, Reaganset al. 2004, Lomi and Pattison 2006). The importance ofunderstanding informal communication between teamsin new product development organizations became evenmore apparent once Henderson and Clark (1990, p. 15)stated that both formal and informal communicationchannels are “the relationships around which the orga-nization builds [product] architectural knowledge” andelaborated on the consequences of a poor understandingof such channels for incumbent organizations developingcomplex products while facing architectural innovation.

Researchers have documented well the effect of taskinterdependence on predicting interteam communicationpatterns in new product development organizations. Thisline of research has shown that the network of infor-mal communication between teams aligns closely withthe network of task interdependence (e.g., see Morelliet al. 1995, Terwiesch et al. 2002, Sosa et al. 2004,Cataldo et al. 2006, Sosa 2008, Gokpinar et al. 2010).Yet the alignment of task interdependence and infor-mal communication networks is not perfect, and thismight have important consequences for organizations’capability to achieve their goals in their new productdevelopment efforts (Gargiulo and Benassi 2000, Sosaet al. 2004, Gokpinar et al. 2010). Scholars have shownthat the alignment between the two networks can bemoderated by spatial arrangements (Allen 1977, 2004;Van den Bulte and Moenaert 1998) and by the formalorganization structure (Nadler and Tushman 1997, Sosaet al. 2004). Yet less attention has been paid to the pos-sibility that the alignment could be moderated by thevery structure of the informal communication network(McEvily et al. 2014). This possibility, however, is fully

1059

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

mailto:[email protected]



Sosa, Gargiulo, and Rowles: Can Informal Communication Networks Disrupt Coordination?1060 Organization Science 26(4), pp. 1059–1078, © 2015 INFORMS

consistent with the fact that the presence or absence ofa given tie in a communication network may depend onthe presence or absence of other ties in the same net-work (Contractor et al. 2006). In particular, we are inter-ested in understanding how the presence of common thirdparties in the communication network may moderate theeffect of task interdependence on predicting communica-tion between teams in new product development efforts.Studying such a moderating effect is important becausethey may prompt “mismatches” between interteam com-munication and task interdependence in ways that areboth theoretically interesting and practically relevant forthe managers involved.

Mismatches between the network of task interdepen-dencies and the network of interteam communicationoccur when teams communicate in the absence of taskinterdependence or when interdependent teams do notcommunicate (Sosa et al. 2004). Although both types ofmismatch are likely to occur, their effect on the prod-uct development process is likely to be different. Com-munication ties in the absence of task interdependencemay result in a suboptimal allocation of resources bythe teams, in the sense that members could be spend-ing time exchanging information with other teams with-out an apparent reason to do so. Yet this “excess” incommunication is unlikely to be directly harmful to thedesign process; moreover, there may be unobserved butimportant reasons for which teams feel compelled tocommunicate (Gokpinar et al. 2010). The second type ofmismatch is potentially more consequential. A lack ofcommunication between two interdependent teams mayresult in an inadequate exchange of information, whichin turn can create coordination disruptions in the designprocess. Empirical evidence supports this claim. In astudy of the vehicle development process in a majorautomaker, Gokpinar et al. (2010) found that subsystemswith abnormally high warranty claims were designed byteams that did not communicate sufficiently with teamsin charge of other interdependent subsystems. We areespecially interested in understanding how the presenceof common third parties in the interteam communica-tion network affects the probability of this latter type ofmismatch to occur.

From a theoretical viewpoint, our interest on the ef-fects of common third parties on the likelihood of com-munication between two interdependent teams puts tri-ads at the center of this study (Simmel 1950). Researchon triadic closure suggests that the presence of a com-mon third party should increase the likelihood of com-munication between two interdependent focal teams(Granovetter 1973, Coleman 1990, Raub and Weesie1990). This general prediction, however, does not takeinto consideration the specific triadic configurationsdefined by the direction of the communication betweenthe focal teams and the common third party. We arguethat a close examination of such configurations allows

for a better understanding of how the local communi-cation structure may shape the behavior of the partiesinvolved. Whereas in some cases the presence of a com-mon third party does increase the likelihood of directcommunication between the two interdependent focalteams, as predicted by the tendency toward triadic clo-sure in communication networks, in others it may actu-ally inhibit the formation of such a tie, increasing thelikelihood that the communication triad would remainopen. To develop our arguments, we analyze the dif-ferent possible directed triadic structures of two inter-dependent focal teams communicating with a commonthird-party team and specify the conditions in which thepresence of the common third party may increase ordecrease the likelihood of communication between thetwo focal teams.

We test our predictions analyzing communication tiesbetween interdependent teams responsible for designingthe components of the Pratt & Whitney PW4098 com-mercial engine that powers the Boeing 777 twin-engineaircraft (Sosa et al. 2004). Each team was in charge ofdesigning a specific component of the engine, whereastechnical interfaces between components determined thestructure of task interdependencies between the teams.To conduct our analysis, we relied on recent method-ological advances in exponential random graph models(ERGMs) that make it possible to model the endoge-nous effects of the interteam communication network(Wasserman and Pattison 1996, Snijders et al. 2006,Hunter et al. 2008, Robins et al. 2009; see Lusher et al.2013 for a review). Our findings support our core argu-ments. We find that when the common third party fallsin the middle of a communication chain between thepotential source and the potential recipient of techni-cal communication, its presence increases the likelihoodof communication between these two teams, creatinga classic transitive structure. However, the situation issignificantly different when the communication betweenthe source and recipient can trigger cyclic exchangesbetween the three teams. In such a cyclic triad, the pres-ence of the common third party actually reduces thelikelihood of the two teams communicating, despite theirtask interdependence.

By enhancing our understanding of how common thirdparties in communication networks may affect the align-ment between interteam communication and task inter-dependence in product development efforts, our paperhas implications for organization theory and practicingmanagers. From a theoretical viewpoint, our results callthe attention on the paradox that the same communica-tion network that emerges to help teams coordinate theirtask interdependence may lead some teams to neglectexchanging information about some of those interdepen-dencies. In doing so, our findings also open venues for amore systematic analysis of how the endogenous dynam-ics of informal communication networks may influ-ence the effectiveness of such networks in facilitating

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

Sosa, Gargiulo, and Rowles: Can Informal Communication Networks Disrupt Coordination?Organization Science 26(4), pp. 1059–1078, © 2015 INFORMS 1061

coordination between interdependent intraorganizationalactors. From a practical standpoint, our findings high-light the importance of attending to situations where,contrary to what is often assumed by managers and engi-neering scholars (Cataldo et al. 2006, Olson et al. 2009),task interdependence may not trigger the informal com-munication patterns necessary to coordinate these tasks.

2. Coordination Mechanisms in NewProduct Development Organizations

Formal organizations seek to structure tasks in ways thatminimize coordination requirements and put in place anumber of formal mechanisms to meet these require-ments (Thompson 1967, Galbraith 1973). In the specificcase of engineering organizations developing complexproducts, efforts to reduce coordination requirementsfocus on minimizing the number and complexity of theinterdependencies between product components withoutcompromising product functionality (Baldwin and Clark2000). Formal coordination mechanisms involve deci-sions such as assigning the responsibility for design-ing product components to design teams, grouping theseteams into subsystem groups led by a manager, andintroducing special teams to facilitate horizontal coordi-nation across teams designing related components.

Yet formal mechanisms are seldom sufficient to ade-quately coordinate interdependent tasks. This insuffi-ciency prompts the emergence of informal communica-tion between interdependent teams, which exchange task-related information that helps them coordinate their work(Nadler and Tushman 1997, McEvily et al. 2014). Thiscommunication is particularly important in the develop-ment of new and complex products. The very novelty andcomplexity of such products makes it harder to antici-pate the nature and the intensity of task interdependen-cies between the teams, which limits the effectivenessof formal mechanisms in coordinating interdependenciesand strengthens the role of informal interteam commu-nication (Henderson and Clark 1990, Sosa et al. 2004,Gokpinar et al. 2010).

Despite the importance of informal communicationin helping teams coordinate task interdependencies, ourknowledge of the factors that shape the emergence ofsuch communication remains vague. The basic tenet isthat the informal communication network between teamsshould mirror the structure of task interdependenciesbetween these teams (Henderson and Clark 1990). Thismirroring results from a process through which a teamwhose task is affected by the task of another team typ-ically seeks to acquire the relevant information fromthe second team, which normally provides such infor-mation to the first (Sosa et al. 2004, Gokpinar et al.2010, McCord and Eppinger 1993, Morelli et al. 1995,

Terwiesch et al. 2002). The result is a network ofdirected communication ties between teams that largelymatches the network of task interdependencies betweenthose teams.

Although the relationship between the task interde-pendence and the communication networks has beendocumented in a number of occasions (see Colfer andBaldwin 2010 for a review), research has also shownthat this relationship is not perfect (Gargiulo and Benassi2000). In other words, not all interdependence ties are“matched” with communication ties. The mismatchescan result from communication between teams in theabsence of task interdependence or from lack of com-munication between interdependent teams (Sosa et al.2004). These two types of mismatches are not equallyimportant. Communication in the absence of interde-pendence may result in inefficient allocation of teamresources, but it is unlikely to be detrimental in its ownright. Failure to exchange information between inter-dependent teams may be also inconsequential, but themissing exchanges can have disruptive effects on thenew product development effort (Gokpinar et al. 2010).From an analytical viewpoint, if the missing communi-cation ties were randomly distributed, their occurrencewould not be a fruitful place for theorizing, althoughthey would still be a source of concern for managers. Yetexisting organizational and network theories suggest thatthe absence of communication between interdependentteams may not be random.

Previous research has shown that spatial arrange-ments, the formal organizational structure, and thestrength of the task interdependence affect the likelihoodof communication between interdependent teams (Sosaet al. 2004). Physically distant teams are significantlyless likely to exchange information than proximate teamsare, even in the presence of task interdependence (Allen2004, Sosa et al. 2002). Members of a formal unit maycommunicate well with one another, but they are alsolikely to neglect establishing communication ties acrossunits, generating a “silo effect” that undermines cross-unit coordination (Gulati 2010). The strength of taskinterdependence also plays an important role in explain-ing missing communication ties. These missing ties aremore likely to correspond to weak task interdependences(Sosa et al. 2004). Yet scholars have not examined thepossibility that the very structure of the informal com-munication network that is supposed to help organi-zations coordinate interdependence between teams mayitself induce teams to neglect establishing communica-tion ties with other interdependent teams. In particu-lar, if the presence of common third parties preventsthe creation of communication ties between interdepen-dent teams, we would be confronted with an interestingparadox: the same communication network that emergesto help teams coordinate their technical interdependen-cies may be leading some teams to neglect exchanging

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


information about some of those interdependencies. Tounderstand where and how this may happen, we turnour attention to how the common third-party teams inthe communication network may help or hinder thealignment between task interdependence and interteamcommunication.

3. Mismatched Interdependencies andCommon Third Parties

The possibility that the communication network thatemerges to coordinate task interdependence betweenteams could itself affect the likelihood of communicationbetween those teams is consistent with our knowledgeof the endogenous mechanisms that shape communica-tion networks—in particular, the effect of common thirdparties in such networks (Contractor et al. 2006, Robinset al. 2009). The central idea behind these mechanisms isthat the communication between the third party and eachof the two interdependent teams can affect the propen-sity of the latter ones to exchange information regardingtheir own task interdependence. To clarify how thesemechanisms operate, we need to examine the differenttriadic configurations involving the three actors.

Figure 1 depicts the four possible triadic structureswith directed communication ties between two interde-pendent teams i and j and a common third party k.The dashed arrows represent exogenous task interde-pendence, and the solid lines represent informal (task-related) communication between teams. The directionof the arrows indicates task-dependence flows (e.g., thetask of team j affects the task of team i) or flows oftask-related information between the respective teams.We focus solely on directed ties to simplify the discus-sion and to remove the potentially confounding effectsof reciprocal ties, although these are controlled for inour empirical analyses. In addition, in our analysis wecontrol for the possibility that the communication tieswith the common third party k may (or may not) beaccompanied by task interdependence flow. Yet, for thepurpose of our argument, we assume that if communi-cation flow occurs it is because the recipient used suchinformation to carry out its design tasks.

Figure 1 (Color online) Triadic Structures with Two Interdependent Teams and One Common Communication Third Party

A: Transitiveclosure

i j

k

B: Common recipientclosure

i j

k

C: Common sourceclosure

i j

k

D: Cyclicclosure

i j

k

Communication flow

Task interdependence flow

Insofar as communication ties help coordinate exist-ing task interdependence, the presence of the interde-pendence xij should trigger an informal communicationtie yij through which team j sends relevant informationto team i. This communication tie would “match” theexogenous interdependence between these two teams.Yet the presence of the common third party k in the localcommunication structure may affect the likelihood of thecommunication tie yij . In principle, the mechanisms typ-ically associated with triadic closure should increase thebaseline tendency to form a communication tie yij thatresults from the underlying interdependence tie xij , irre-spective of the direction of the communication ties withthe common third party. In other words, the commonrelationship with k should make it easier for i to seekinformation from j and compel j to respond favorably tosuch a request, or even proactively sending this informa-tion in some cases. The mechanisms typically invokedinclude the role of the common third party k in facilitat-ing mutual awareness and trust between i and j (Simmel1950; Granovetter 1973, 1985; Burt 2005), as well asj’s reputational concerns in the event that k becomesaware of its noncooperative behavior toward i (Coleman1990, Raub and Weesie 1990). These considerations aresupported by empirical evidence. Research on knowl-edge transfer confirms that the presence of commonthird parties facilitates the source transferring knowledgeto the recipient (Reagans and McEvily 2003) and therecipient acquiring knowledge, even when the knowl-edge exchange occurs across organizational boundaries(Tortoriello et al. 2012).

Yet the tendency toward triadic closure, and hence theeffect of the common third party on the likelihood ofa communication tie between the two focal teams, mayvary depending on the position this party occupies in thelocal communication structure, as defined by the spe-cific information flows between the focal teams and thecommon third party. Specifically, we argue that in somecases the position that the common third party occupiesin the local structure may create incentives to discour-age the communication between the two focal teams,despite their interdependence. This could prevent theclosure of the triad in the communication network, leav-ing the interdependence xij without the corresponding

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


communication tie yij and generating a “mismatched”interdependence, which may disrupt interteam coordina-tion. To understand how this may be the case, we needto focus on how the structural differences across the fourtriads shown in Figure 1 may affect the role that the thirdparty plays in inducing or inhibiting the communicationbetween the two other interdependent parties. We arguethat such differences are rooted in the directionality ofthe communication flows between the third party k andthe two other members of the triad (Robins et al. 2009).

Triad A is a classic transitive triad (Robins et al.2009). In this triad, the third party k receives informationfrom j and sends information to i. The transitivity logicthat operates in most communication networks suggeststhat this local structure should facilitate the emergenceof trust and awareness between i and j , making j morelikely to send information to i. That j provides informa-tion to k and k provides to i should also make k moreaware of i’s need for information from j . In knowledgetransfer terms, a common third party that understandsthe information provided by the source (team j) and alsounderstands the needs of the recipient of information(team i) is likely to facilitate the transfer of informationbetween i and j because it can help the source framingits outputs and the recipient acquiring the knowledge itreceives (Reagans and McEvily 2003, Tortoriello et al.2012). Thus, this local structure allows k to reduce anyfriction that might exist against information being sentfrom j to i, which should encourage i and j to commu-nicate, further increasing the trust-enhancing propertiesof the triad. The mechanism operating here is similar toBurt’s (2001) argument on the effect of common thirdparties on trust and reputation, whereby the commonthird party “amplifies” the strength of the relationshipbetween the other two members of the triad (see alsoBurt and Knez 1995). Although it is possible that kmight also act as intermediary between i and j , this indi-rect information is unlikely to completely substitute forfirsthand information obtained directly from the sourcein complex product design projects such as the one stud-ied in this paper. In sum, both the trust-enhancing prop-erties of transitive triads and the ability of the third party,based on its exchanges of information with both i and j ,to facilitate conditions for the ease transfer of informa-tion from j to i increase the likelihood that j would sendtechnical information to i, closing the communicationtriad.

Whereas triad A is a classic transitive triad, triads Band C represent other types of transitive triads in whichthe common third party either receives or sends infor-mation to the other two interdependent teams in the triad(Robins et al. 2009). In triad B, the common third partyk receives information from both i and j , but it does notsend information to either. In triad C, the common thirdparty k sends information to both i and j , but it doesnot receive information from either. This local structuredoes not create incentives for k to either encourage or

discourage the communication between the two inter-dependent teams i and j because the communicationbetween i and j is less likely to affect k’s work in anydiscernible way. In triad B, team k is downstream of thedecisions of both i and j , so once k receives informationfrom them, it can carry on with its task without worryingabout what i and j do. Analogously, in triad C, k likelyoccupies an upstream position in the design chain, sothe decisions k makes cascade down to i and j , whichwould need to adjust their work accordingly, but suchadjustments should not impact the work of k. Hence, thepresence of a common third party in the communicationnetwork is less likely to significantly modify the base-line tendency for the two interdependent teams i and jto communicate in triads B and C.

In triad D, the directionality of the information flowswith the common third party is the exact opposite of theflows in triad A. Team j receives information from k,which in turns receives information from i. If team jsends information to i, they would then form a com-munication pattern characterized by cyclic triadic clo-sure, which is fundamentally different from the otherthree triads shown in Figure 1. If i makes changes inits design to align it with j ′s design, these changes mayend up forcing modifications to k’s design, which couldin turn cause j to further adjust its design. Such a cyclicpattern may cause the three teams to become involvedin an endless chain of information exchanges such asthe one depicted in triad D. Although cyclic interdepen-dence patterns are not infrequent in new product devel-opment efforts (Smith and Eppinger 1997, Mihm et al.2003), they are difficult to manage and increase the like-lihood of errors (Sosa et al. 2013). Given the difficultiesassociated with cyclic communication in triads, the thirdparty k is unlikely to induce j to send information to i,because this information may end up forcing k to makefurther adjustments to its own design. On the contrary,once k has received information from i and understandshow i’s design can affects its own, k has an incentive toencourage i and j to “freeze” (or even neglect) their taskinterdependence to prevent design changes that may leadto subsequent rework in k’s design. Because the cyclewould also affect i and j , these teams may also preferto freeze their interface to avoid triggering the cycliccommunication involving k. This approach to managingiterative problem solving has been recognized in the newproduct development literature (Eppinger et al. 1994,Mihm et al. 2003). When a set of tasks are related toeach other in a cyclic manner, the teams involved maydecide to “cut” the communication cycle by removing,freezing, or making assumptions about one of the inter-dependencies in the cycle, which allows them to carryout their activities in a sequential rather than iterativeway (Mihm et al. 2003). In this structure, the presence ofthe common third party k should reduce the likelihoodof communication between teams j and i.

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


The previous discussion suggests that we shouldexpect the influence of a common third party on thebaseline tendency for two interdependent focal teams tocommunicate to vary with the position it occupies in itstriadic structure. When the third party falls in the middleof a communication chain whose closure would result ina transitive triad, the third party’s presence is likely toamplify the awareness of the interdependence betweenthe two focal teams and reduce any friction that may pre-vent the information exchange to take place, thus induc-ing the exchange of technical information about theirinterdependence. Conversely, when the third party fallsin the middle of a communication cycle whose closureis likely to be detrimental to this third party, the thirdparty may have an incentive to encourage “freezing” theinterdependence between the two focal teams to avoidchanges that may cycle back and affect the third party’swork. Hence, the presence of a common third party mayend up hindering the communication between the twofocal teams when this communication would result in acyclic triad.

Finding empirical evidence to test the predicted effectsof common third parties on the likelihood of commu-nication between interdependent teams is not straight-forward, however. The very nature of the endogenousmechanisms discussed in this paper makes standard sta-tistical techniques inappropriate for testing for the pres-ence of these effects. Fortunately, recent advances in theuse of ERGMs allows for testing whether the observedfrequency of a given structural configuration in a net-work is significantly different from what one wouldexpect had this configuration occurred randomly, aftercontrolling for appropriate endogenous and exogenouseffects. In the context of this paper, the configurationsof interest are the four triadic structures depicted in Fig-ure 1. If our reasoning is correct, we should observethat the frequency of triad A is significantly higher thanexpected, indicating a tendency for communication clo-sure in this triad. We should also observe that the fre-quency of triad D is significantly lower than expected,indicating a tendency for cyclic triads to remain “open.”Finally, the observed frequencies for triads B and Cshould not depart significantly from randomness, as theeffect of the common third party is not clearly dis-cernible in these triads. We now elaborate on how wetest these predictions.

4. Data and MethodsWe test our predictions by studying the detailed designphase in the development of a large commercial air-craft engine. Figure 2 shows the product architecture andformal organizational structure of the PW4098 engine.Panel (a) is a cross-sectional diagram of the engine thatshows the eight subsystems into which the 54 enginecomponents were configured. Panel (b) shows the (flat)

formal organizational structure by which design teamswere grouped into subsystems of teams, quasi-mirroringthe engine architecture.

We also captured network data associated with boththe organization and the product. We captured theinterteam communication network of the 60 teams (54design teams plus 6 functional teams) that designed theengine components and evaluated the overall engine per-formance as well as the technical interface network ofthe 54 components comprising the engine (Sosa et al.2004, 2007). We collected data from multiple sources.Data on interteam communications were gathered byinterviewing and surveying key members of the teams(Marsden 1990). Product network data were constructedby interviewing several experienced engine architects.

4.1. Interteam Communication Network DataThe informal communication network between the teamsis defined by the presence or absence of task-related tech-nical communication between any 2 of the 60 teams (i.e.,54 core design teams plus 6 functional teams) involved inthe 10-month detailed design phase of the developmentprocess. Our focus on technical task-related interactionsis akin to what Allen (1977) and Morelli et al. (1995)defined as “coordination-type communication.” The factthat the design of each engine component depends incomplex ways on the design of some other componentsas well as the stringent functional requirements of a prod-uct that needs to operate faultlessly in extreme condi-tions highlights the importance of direct communica-tion between teams to adequately understand the effectsthat design choices might have on product performance.Although we acknowledge that various types of interteamtechnical information flows are likely to exist during thedesign of complex systems, we did not try to distin-guish between specific types of technical exchanges andfocused instead on an overarching measure that capturesall these possible technical exchanges in a single indica-tor. Doing so was both consistent with previous researchstudying technical communication in engineering organi-zations (Allen 1977, McCord and Eppinger 1993, Morelliet al. 1995) and important to avoid posing insurmount-able cognitive hurdles to our respondents that could jeop-ardize the quality of the data.

We measured the presence of technical communica-tion between teams through a questionnaire addressedto team leaders and validated (or revised, when neces-sary) their responses by interviewing at least one otherkey member of each team. Presentations describing theterminology and overall objective of the data collectionwere made in two sessions to more than two-thirds of therespondents; the others were later briefed individually.Respondents were presented with the roster of the 54design and 6 functional teams and asked to report if theyhad received nontrivial technical information from eachof the other teams during the design phase of the project.

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Figure 2 (Color online) The (a) Aircraft Engine Studied and (b) Formal Organizational Structure of the Project

Fan

Low-pressure compressor (LPC)

High-pressure compressor (HPC)

Combustion chamber (CC)

High-pressure turbine (HPT)

Low-pressure turbine (LPT)

Mechanical components (MC)

Externals and controls (EC)

(a) (b)

PW4098Fan

7 design teams

7 design teams

7 design teams

5 design teams

5 design teams

6 design teams

7 design teams

10 design teams

LPC

HPC

CC

HPT

LPT

MC

EC

6 functional teams

Respondents were encouraged to focus on technicalinformation exchanges that actually took place (i.e.,“how it was,” not “how it should have been”) and thatcontributed to “fulfil[ling] the functional requirementof their component design or program goals” (Rowles1999, p. 51). We were able to obtain responses from 57of the 60 teams. The three teams whose direct responseswere missing were the interface team at Boeing (one ofthe six functional teams), one team in the high-pressurecompressor subsystem group, and one team in the exter-nals and controls subsystem group. Yet because theseteams were identified as information providers by otherteams in the development process, we did not excludethem from the network. We estimated all our modelsexcluding the three missing teams, and the results wereconsistent with the ones reported here.

Consistent with previous research on technical com-munication (e.g., Allen 1977, Morelli et al. 1995, Lomiand Pattison 2006, Rank et al. 2010), we used adichotomous definition of interteam communication andassumed that relevant communication did take placeinsofar as team i (the recipient) acknowledged receivingsome technical information from team j (the source).Based on these data, we reconstructed the informaltechnical communication network 8y9 among the 60teams involved in the project, where each element yijof 8y9 captures the transfer of technical communicationfrom team j (the source) to team i (the recipient). Theresulting communication network contains 680 directednonzero interteam communication ties (yij = 1) amongthe 60 teams—that is, 19% of all possible interteamcommunication ties.

Figure 3, panels (a) and (b) exhibit the team com-munication network and its corresponding adjacencymatrix. In panel (a), each node of the network is shadedaccording to subsystem membership, and each link indi-cates technical communication between two teams. In

panel (b), a nonzero cell of the adjacency matrix cap-tures a directed interteam technical communication.

4.2. Product Network DataOur product data capture the breakdown of the enginestructure into 8 subsystems and 54 components, as wellas the five types of design dependencies among thosecomponents: spatial constraint, structural constraint asa result of transfer of loads, exchange of material,exchange of energy, and exchange of information, asidentified by the system engineers. Even though such adetailed map of the product architecture was not avail-able at the project’s onset, design teams shared a commonunderstanding of (i) the engine’s division into subsys-tems and components and (ii) the relevant technical inter-faces between their own components and the componentsdesigned by other teams. Each technical interface xijis a function of a vector of the five design dependen-cies that capture the various types of technical linkagesgoing from component j to component i (see Sosa et al.2007 for details). In our analysis, we are concerned withthe presence or absence of a technical interface; there-fore, xij = 1 if the design of component i depends onthe design of component j; otherwise, xij = 0. Therewere 569 technical interfaces between the 54 componentsidentified by system architects. This means that 20% ofthe theoretically possible interfaces between componentswere actually present in the project.

Figure 4, panels (a) and (b) exhibit the componentinterface network and its corresponding technical inter-face matrix. In panel (a), each node of the 54-nodenetwork is shaded according to subsystem membership,with each link representing a technical interface betweentwo components. In panel (b), a nonzero cell in thetechnical interface matrix represent a directed technicalinterface between two engine components. The techni-cal interface matrix shown in panel (b) quasi-mirrors theadjacency matrix shown in Figure 3, panel (b) because

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Figure 3 (Color online) Interteam Technical Communication Network

their elements are identically sequenced so that the first54 elements of the adjacency matrix (the 54 designteams in the organization) mirror the 54 elements in thetechnical interface matrix (the 54 engine componentsdesigned by each of the 54 design teams).

4.3. VariablesWe model the observed network of interteam techni-cal communication 8y9 between the 54 design teams(which is the dependent variable in the analysis) asa function of both exogenous and endogenous vari-ables. The key exogenous variables comprise the tech-nical interface network, which captures the presence ofa directed technical interface xij between each pair ofcomponents in the product and therefore indicates theexistence of task interdependence between two teams.In addition, there are a number of exogenous controlsdescribed below. The endogenous variables are the pat-terns of interteam communications related to any dyad inthe interteam communication network. Because the phe-nomenon of interest—the moderating role that commonthird parties may have on the effect of task interdepen-dence in predicting interteam communication—involvesboth the exogenous technical interface network and theendogenous interteam communication network, we mustdescribe the variables that test our predictions along withthe statistical network modeling approach described inthe next section.

4.3.1. Exogenous Nodal Variables.• Component redesign: An important factor that af-

fects both a team’s workload and the need to communi-cate with other interdependent teams is the novelty of theteam’s component. The more novel a component withrespect to the prior generation, the more likely it willinvolve extra work for the team, and the more likely itwill affect interfaces with adjacent components. By con-trast, components that carry over a significant portionof design content from previous engine models decrease

the demand for attention to associated interfaces. Wecapture this factor by measuring component redesign asthe percentage of novel design content in a componentrelative to its design in the product’s previous version.Because it is impossible to determine the exact amountof redesign in a component, we relied on estimates pro-vided by the team leader responsible for the design ofeach component.

• Component complexity: Another component attri-bute that can affect the team’s workload is its inter-nal complexity. We measure a component’s complexityin terms of its estimated number of distinct parts. Forthis estimate we relied on the experience of one of theauthors, who is a design expert with substantial experi-ence in similar engine programs and who also reviewedthe design work for this particular project.

• Component connectivity: The component connec-tivity is simply the number of incoming and outgo-ing technical interfaces of each component in a dyad,irrespective of their strength. For each component, thismeasure counts the number of components with whichit shares at least one design dependency, and it is thusequivalent to the “degree” of the component in the tech-nical interface network.

• Team size: The team’s capacity to communicatewith others might well be affected by the resourcesavailable to the team. Because managers, when allocat-ing resources to design teams, may take into accountthe workload entailed by component characteristics andby its interfaces with other engine components, theeffects the communication network structures may beconfounded by the effects of the resources available tothe team. To control for this possibility, we include afour-point ordinal variable that accounts for the man-power resources allocated to teams. Although we wereunable to collect precise data on team size (which, inany case, varied throughout different stages of the pro-cess), we did obtain a qualitative assessment of team sizebased on one author’s direct experience in the project.

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Figure 4 (Color online) Technical Interface Network of 54 Engine Components

• Subsystem group membership: Because each designteam is formally assigned to a subsystem-level group,teams in the same group might share some characteris-tics that affect their ability to attend to technical inter-faces. These teams will be also part of the same formalcoordination structure, with their leaders reporting to themanager responsible for the subsystem. For instance,the seven teams that formed the fan group share tech-nical expertise and experience relevant to the design offan components, which is different from that of designteams responsible for components of the low-pressurecompressor or high-pressure turbine subsystems. On theorganizational side, teams in the same subsystem groupshare a subsystem manager who is responsible for plan-ning and control of resources. All these factors cre-ate “silo” effects that foster communications betweenteams within a subsystem while hindering communica-tions between design teams that belong to different sub-systems (Allen 1977, Sosa et al. 2004). To account forthis, we created a categorical nodal variable that whenentered in our statistical model allows us to control forwhether or not any pair of design teams belong to thesame subsystem group.

Table 1 shows descriptive statistics and correlationcoefficients of the continuous nodal variables used in ouranalysis.

Table 1 Descriptive Statistics and Pairwise Correlations ofExogenous Nodal Variables

Variable Mean S.D. 1 2 3

1. Component 00487 00333 10000redesign

2. Component 410593 660189 −00334∗ 10000complexity

3. Component 130074 60434 −00217∗ 00577∗ 10000connectivity

4. Team size 20444 00925 00142 00308∗ 00533∗

Note. N = 54.∗p < 0005.

4.3.2. Exogenous Dyadic Variables.• Technical interface strength: Our statistical model-

ing approach takes directly into account the effect of taskinterdependence as captured by the presence or absenceof a technical interface going from component i to com-ponent j . In addition, we include a dyadic control forthe strength of the technical interface that is equal tothe sum of the nonzero directional dependencies fromcomponent j to i.

• Functional third parties: As mentioned previously,the project studied included the participation of six func-tional teams that were not responsible for the design ofany engine component. Hence, for the purpose of ouranalysis, we consider these teams to be exogenous tothe core network of 54 design teams. Nonetheless, toaccount for the possible influence of these functionalteams on the interteam communication patterns of the 54design teams, we count the number of functional teamsthat act as common third-party teams between any pairof design teams i and j .

5. Statistical Network AnalysisA standard tenet in network analysis is that the pres-ence or absence of a relationship between two actorsin a network is not independent of the structure of thatsame network. This endogenous nature of the statisti-cal relationship between the network structure and thepresence or absence of specific ties within this samenetwork makes traditional statistical techniques inap-propriate to test our hypotheses. Hence, we resort toERGMs, also known as p∗ models (refer to Lusheret al. 2013 for a comprehensive review). In essence,ERGMs allow for modeling an observed network—inour case, the binary network of technical communica-tion among the 54 design teams—as a function of bothexogenous parameters (i.e., the attributes of both teamsand components) and endogenous parameters (e.g., thenumber of ties, the number of reciprocated ties, the num-ber of transitive triads in the communication network).

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


This approach allows us to obtain reliable estimates ofthe effect of task interdependence on the probability ofobserving an interteam communication tie as well as toexamine how the presence of common third parties mod-erates such an effect, net of the effect of other endoge-nous and exogenous factors that might also affect thatprobability (Contractor et al. 2006).

To estimate the effects of interest associated witheach triad shown in Figure 1, we estimate a proba-bilistic model of the bivariate network formed by theinterteam communication network of design teams andthe exogenous technical interface network of enginecomponents. As in any bivariate ERGM, our modelspecification can be divided into within-network effects,based on ties from one network, and cross-networkeffects, based on ties from both networks (Lusher et al.2013, Chapter 10). In our case, the within-networkeffects correspond to configurations of the communica-tion network. The cross-network effects correspond tothe “entrainment” (or co-occurrence in the same direc-tion) of an interteam communication and a technicalinterface (which captures the effect of task interdepen-dence on the probability of observing an interteam com-munication), as well as triadic configurations such asthe ones shown in Figure 1. Because the technical inter-face network is exogenous (or fixed), its “within-networkparameters” are not estimated. For estimation purposes,we use XPNet (Wang et al. 2009).

Our model considers interteam communication as arandom variable. Hence, for i and j (which are distinctteams of the network of 54 design teams), we considera random variable Yij where Yij = 1 if design team i (therecipient) receives technical information from team j(the source) and where Yij = 0 otherwise. We specifyyij as the observed value of Yij , where Y is the matrixof all Yij variables and y is the matrix of observed ties(i.e., the binary adjacency matrix corresponding to theinterteam technical communication network of 54 designteams). Intuitively, our model predicts the probability ofteam i receiving technical information from team j asa function of (i) within-communication-network config-urations; (ii) cross-network configurations that capturecertain types of associations between the communica-tion network and the technical interface network, includ-ing the triadic cross-network configurations of interest(shown in Figure 1); (iii) certain attributes of the designteams (nodal effects); and (iv) and certain attributes ofthe dyads (dyadic effects). Formally, the model we esti-mate takes the following form:

Pr4Y =y �X=x1W =w1D=d5

=1�

exp4WithinCommEffects+CrossNetworkEffects

+NodalEffects+DyadicEffects5

=1�

exp(

∑

k

�kZk4y1x1w1d5

)

1 (1)

where (i) Y is a random network of size 54 with possi-ble ties Yij and y being a realization of Y ; (ii) X is the54 × 54 binary matrix of technical interfaces with real-ization x; (iii) W is a 54 × p array of nodal attributes(e.g., component redesign, component complexity, teamsize) with realization w; (iv) D is a 54 × 54 × p arrayof dyadic attributes (i.e., strength of technical interfaces,number of common functional third parties) with realiza-tion d; (v) Zk4y1 x1w1d5 is a network statistic that canbe computed for a particular y that may also depend onthe matrix x of technical interfaces, the array w of nodalattributes, and the array d of dyadic attributes; and (vi) �is a normalizing quantity to ensure that Equation (1) isa proper probability distribution.

As mentioned above, our model includes four types ofparameters, which define the family of probability distri-butions that can generate the observed interteam commu-nication network. The coefficient parameters (�k5 in themodel are estimated to maximize the fit between the ran-dom networks and the observed data. Table 2 describesqualitatively each parameter included in the model andthe tendency that it accounts for. Given the novelty ofthe cross-network triadic effects included in our models,we describe the network statistics of such parameters inthe table.

Table 2 describes 11 interteam within-communication-network parameters that capture network configurationsin a typical intraorganizational network (Contractor et al.2006, Rank et al. 2010, Lomi et al. 2014). To this end,we follow Robins et al. (2009) in specifying the variousconfigurations in which both closure and two-path com-munications can occur because they capture the variousways in which common third-party teams may influencethe communication between a focal pair of design teams.

Table 2 also describes the cross-network effectsincluded in our model. First, we include the effect ofentrainment of interteam communication and technicalinterface between any pair of teams to test for the effectof task interdependence. Then, we describe two setsof cross-network triadic effects. The first set capturesthe overall tendency of a technical interface to be sur-rounded by common third parties in the communica-tion network. Because our network data are directed,there are four possible ways in which a common thirdparty can communicate with interdependent teams i andj whose components share a technical interface (Robinset al. 2009). As specified by the parameters’ networkstatistics, this set of cross-network triads consists merelyof controls in our models because they count the numberof occurrences of such configurations in our data with-out capturing whether a communication between i and jtakes place.

To account for the moderating effect of common thirdparties on the effect of task interdependence, we mustdefine a second set of cross-network triadic statisticsthat accounts precisely for the four triadic configurations

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 2 Summary of Within-Network and Cross-Network Effects Included in Our ERGMs

Parameter Visual description What it accounts for

Within-communication-network effects

Arc Baseline tendency for interteam communication to occur

Reciprocity Tendency to reciprocate communication

Source spread Tendency for variation in the degree to which a team sendstechnical information to others

Recipient spread Tendency for variation in the degree to which a team receivestechnical information from others

Transitive path closure incommunication network (AT-T)

Tendency for transitive path closure to occur in the interteamcommunication network

Cyclic closure in communicationnetwork (AT-C)

Tendency for multiple cyclic structures to occur in the interteamcommunication network

Activity closure incommunication network (AT-D)

Tendency for shared activity closure (teams i and j providinginformation to a common recipient) to occur in the interteamcommunication network

Popularity closure incommunication network (AT-U)

Tendency for shared popularity closure (teams i and j receivinginformation from a common source) to occur in the interteamcommunication network

Multiple two-path incommunication network(A2P-T)

Tendency for multiple two-path communications to occur in theinterteam communication network

Two-path activity incommunication network(A2P-D)

Tendency for shared activity (with a common recipient) to occur inthe interteam communication network

Two-path popularity incommunication network(A2P-U)

Tendency for shared popularity (with a common source) to occurin the interteam communication network

Cross-network effects

(Interteam communication (Network A) Technical interface (Network B) 5

Entrainment of interteamcommunication and technicalinterface (Arc AB)

Tendency for interteam communication and technical interface toco-occur in the same direction; this tests for the baseline effectof task interdependence

Cross-network triadic effects between a focal interface and communications with a common third party

Cross-network transitive triad(T-ABA)

Overall tendency for cross-network transitivity to occur; thenetwork statistic for this network configuration is∑54

i

∑54j 4∑54

k yikykj 5xij

Cross-network common recipienttriad (T-AAB)

Overall tendency for a cross-network common recipient triad tooccur; the network statistic for this network configuration is∑54

i

∑54j 4∑54

k y ′

ikykj 5xij

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 2 (cont’d)

Parameter Visual description What it accounts for

Cross-network common-sourcetriad (T-BAA)

Overall tendency for a cross-network common source triad tooccur; the network statistic for this network configuration is∑54

i

∑54j 4∑54

k yiky′

kj 5xij

Cross-network cyclic triad(C-AAB)

Overall tendency for a cross-network cycle to occur; the networkstatistic for this network configuration is

∑54i

∑54j 4∑54

k y ′

iky′

kj 5xij

Moderating effects of a common third party on the effect of task interdependence

Cross-network transitive closure(triad A in Figure 1)

Tendency for triad A (shown in Figure 1) to occur; the networkstatistic for this network configuration is

∑54i

∑54j 4∑N

k yikykj 5xijyij

Cross-network common recipientclosure (triad B in Figure 1)

Tendency for triad B (shown in Figure 1) to occur; the networkstatistic for this network configuration is

∑54i

∑54j 4∑N

k y′

ikykj 5xijyij

Cross-network common sourceclosure (triad C in Figure 1)

Tendency for triad C (shown in Figure 1) to occur; the networkstatistic for this network configuration is

∑54i

∑54j 4∑N

k yiky′

kj 5xijyij

Cross-network cyclic closure(triad D in Figure 1)

Tendency for triad D (shown in Figure 1) to occur; the networkstatistic for this network configuration is

∑54i

∑54j 4∑N

k y′

iky′

kj 5xijyij

Categorical nodal effects

Communication withinsubsystem group

Tendency for communication to occur between teams in the samesubsystem group

Entrainment within subsystemgroup

Tendency for co-occurrence to occur between teams in the samesubsystem group

Continuous nodal effects

Communication with sum ofattributes

Tendency for communication to occur between teams with highervalues of a specific continuous attribute

Entrainment with sum ofattributes

Tendency for co-occurrence to occur between teams with highervalues of a specific continuous attribute

Dyadic effects (Dyadic covariate = )

Communication with dyadicattributes

Tendency for communication to occur between teams with highervalues of a specific dyadic attribute

Entrainment with dyadicattributes

Tendency for co-occurrence to occur between teams with highervalues of a specific dyadic attribute

shown in Figure 1. The network statistics of these triadicconfigurations are also included in Table 2. Contraryto the network statistics of the cross-network triadicparameters described above, this second set of net-work statistics only accounts for the technical interface

between i and j when matched by the correspondinginterteam communication between i and j .1 Estimatingthe coefficients for this latter set of triadic parametersallows us to test our key predictions.2 Specifically, weexpect to observe a positive coefficient for the effect

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


of cross-network transitive closure (triad A in Figure 1)and a negative coefficient for the effect of cross-networkcyclic closure (triad D in Figure 1).

Finally, Table 2 includes categorical and continuousnode- and dyadic-level effects to capture whether exoge-nous factors such as the membership of a design teamin one of the subsystem groups (a categorical nodalattribute), the complexity of the component a team isdesigning (a continuous nodal attribute), and the num-ber of common functional third parties between a pair ofdesign teams (a dyadic attribute) influence the likelihoodof a communication tie and of a entrainment tie.

6. ResultsTable 3 presents ERGMs estimating the probability ofobserving a directed technical communication tie be-tween two design teams in the project studied. The esti-mates were obtained using the Markov chain MonteCarlo maximum likelihood estimation (MCMCMLE)procedure implemented in XPNet (Wang et al. 2009).Although the model of interest is Model 4 (the fullmodel), we also present partial models to introducesequentially the various effects that our models accountfor. After describing the results, we also discuss how weassessed their robustness and the models’ goodness of fitto the data.

Model 1 in Table 3 includes within-network effectsthat capture the basic endogenous tendencies (withinthe communication network) to form ties. To that end,this model suggests that the interteam communicationnetwork exhibits a positive tendency to form recipro-cated ties, to form (short-path) transitive closure, and toform two-path communications resulting from commonthird parties that receive information from two teamsthat do not communicate with each other. Other effectsare either nonsignificant or become nonsignificant insubsequent models that include cross-network triadiceffects. Most importantly, Model 1 includes a dyadiccross-network effect that captures the baseline effect oftask interdependence. The positive and significant coef-ficient for the entrainment of interteam communicationand technical interface provides empirical evidence forthe effect of task interdependence on communication:teams designing interdependent components are morelikely to communicate. Holding other variables constant(in Model 1), the baseline probability of observing adirected communication tie between two design teamsis 12 times higher if their components are connected bya technical interface (e20505 = 1202); this probability is17 times higher in the full model.

Model 2 includes various exogenous nodal and dyadiceffects. As described in Table 2, for each of these exoge-nous effects, two parameters are estimated. One param-eter captures how the exogenous attribute influences the

tendency of team i receiving information from team j ,whereas the other parameter captures how the exoge-nous attribute moderates the tendency for entrainment ofinterteam communication and the technical interface tooccur along interface ij. For instance, the effects of for-mal organizational subsystem group boundaries (a cat-egorical nodal effect) are captured by first showing apositive and significant (1.667, p < 00001) greater ten-dency for teams that were part of the same subsys-tem organizational group to communicate than for thosewho belonged to different subsystem groups, indepen-dent of whether or not their components shared a tech-nical interface.

The moderating effect that organizational boundarieshave on the entrainment of the technical interface andinterteam communication (Sosa et al. 2004) is cap-tured by the negative and significant coefficient forthe communication/interface entrainment (−10106, p <00001), which suggests that the predictive power oftechnical interfaces diminishes within subsystem groupboundaries, where other forces associated with the for-mal organizational structure play an important role intriggering interteam communication. This is consistentwith the fact that teams that belong to the same subsys-tem group are much more likely to communicate in theabsence of a technical interface than those from differ-ent subsystems do (Lessard and Zaheer 1996). Withinsubsystems, 20% of the potential communication tiesbetween teams whose components do not share an inter-face are actually realized; the corresponding figure forteams across subsystems is only 2%.

Model 2 also includes various (continuous) nodaleffects to control for the possibility that both interteamcommunication and communication/interface entrain-ment may be explained by the properties of the enginecomponents designed by teams i and j or by the proper-ties of the teams. These controls, however, do not havemuch influence on the patterns of interteam communica-tion, nor do they moderate the effect of task interdepen-dence. Only team size seems to have a significant effecton interteam communication: bigger teams are morelikely to communicate. Of the two dyadic effects, thenumber of functional teams that act as a common thirdparty between teams i and j shows a positive effect onthe communication between these teams. The more twoteams communicate with any of the six functional teams,the more likely they are to communicate with each other.However, communication with functional teams does notseem to influence the effect of task interdependence oninterteam communication.

Model 3 includes a set of controls that capture cross-network triadic effects. These effects account for thetendency for a technical interface to be surrounded bycommon third-party teams in the communication net-work (see Table 2 for a visual description of theseeffects). Model 3 shows that there is a negative tendency

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Table 3 Maximum Likelihood Estimates of ERGMs for Interteam Communication

Variable Model 1 Model 2 Model 3 Model 4

Within-communication-network effectsArc −20905∗∗ 4102135 −40054∗∗ 4108605 −30740∗ 4109795 −40089∗∗ 4200365Reciprocity 20505∗∗∗ 4002845 20229∗∗∗ 4003015 20505∗∗∗ 4003195 20629∗∗∗ 4003385Source spread 00001 4006445 −00176 4007705 −00455 4009535 −00349 4009545Recipient spread −00944∗ 4005295 −00916 4006645 −00784 4005735 −00775 400595AT-T 00728∗∗∗ 4002275 00620∗∗ 4002465 00472∗∗ 4002385 00560∗∗ 4002425AT-C −00422∗∗∗ 4000935 −00424∗∗∗ 4001135 −00116 4001395 −00125 4001495AT-D 00099 4001435 00048 4001655 −00088 4001545 −00137 4001585AT-U 00318∗ 4001745 00289 4001985 00194 4001735 00216 4001765A2P-T −00086∗∗∗ 4000265 −00077∗∗∗ 4000295 −00020 4000325 −00012 4000315A2P-D 00117∗∗∗ 4000145 00136∗∗∗ 4000205 00092∗∗∗ 4000215 00105∗∗∗ 4000255A2P-U −00039 4000535 −00006 4000535 −00012 4000565 −00015 4000585

Task interdependence effectArc AB 20505∗∗∗ 4001525 20901∗∗∗ 4008445 20823∗∗∗ 4008855 20842∗∗∗ 4009245

Nodal categorical effectsCommunication within subsystem group 10697∗∗∗ 4003435 10727∗∗∗ 4003505 10708∗∗∗ 4003275Entrainment within subsystem group −10106∗∗∗ 4003875 −10136∗∗∗ 4003915 −10231∗∗∗ 4004135

Nodal continuous effectsCommunication with sum of component redesign −00208 4002955 −00177 4002845 −00180 4002855Entrainment with sum of component redesign 00190 4003745 00176 4003785 00200 4003825Communication with sum of component complexity −00002 4000025 −00001 4000025 −00001 4000025Entrainment with sum of component complexity 00000 4000025 00000 4000025 00001 4000025Communication with sum of component complexity −00015 4000195 −00028 4000225 −00028 4000225Entrainment with sum of component complexity 00024 4000235 00026 4000255 00016 4000245Communication with sum of team size 00329∗∗∗ 4001265 00331∗∗∗ 4001275 00331∗∗∗ 4001285Entrainment with sum of team size −00219 4001595 −00260∗ 4001565 −00272∗ 4001625

Dyadic effectsCommunication with functional third parties 00224∗∗ 4001085 00162 4001115 00157 4001245Entrainment with functional third parties 00021 4001435 00042 4000415 −00011 4001625Communication with technical interface strength 00023 4000405 00085 4001545 00036 4000435

Cross-network triadic effects between a focal interface and communications with a common third partyT-ABA −00075 4000675 −00090 4000835T-AAB 00182∗∗∗ 4000335 00151∗∗∗ 4000415T-BAA 00069 4000445 00037 4000565C-AAB −00144∗∗∗ 4000535 −00087 4000615

Moderating effects of a common third party on the effect of task interdependenceCross-network transitive closure (triad A) 00223∗∗ 4001025Cross-network common recipient closure (triad B) 00097 4000915Cross-network common source closure (triad C) 00085 4000955Cross-network cyclic closure (triad D) −00235∗∗ 4001005

Note. Standard errors are given in parentheses.∗p < 0010; ∗∗p < 0005; ∗∗∗p < 0001 (two-tailed).

for third-party teams to surround technical interfacesin a cyclic triad (C-AAB). Yet this negative tendencybecomes nonsignificant in the full model (Model 4).Model 3 also shows that there is a positive tendencyfor third-party teams to act as a common receiver ofinformation from the other two interdependent teams inthe triad (T-AAB). These basic tendencies are importantcontrols because they account for the overall tenden-cies for these four cross-network triadic configurationsto form, but they do not test our predictions becausethey do not account for whether the focal interface ismatched by its corresponding interteam communication.

Model 4 tests how the local structure in which thecommon third-party team is embedded influences theeffect of task interdependence on interteam communica-tion. To do so, the model includes four triadic effectsthat account for the four possible ways in which a com-mon third-party team may interact with a pair of inter-dependent teams that exchange information in the samedirection as their task interdependence (as shown in Fig-ure 1). The coefficients associated with the tendency fortriads B and C to form are not significant. This suggeststhat the presence of a common third party does not sig-nificantly moderate the effect of task interdependence on

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


interteam communication when this party is a commonrecipient or a common source of information for theinterdependent teams, which is consistent with our the-orizing. As predicted, there is a positive and significanttendency for cross-network transitive closure (triad A inFigure 1) to form (0.223, p < 0005). When the infor-mation flowing through the common third-party teamsflows in the direction of the focal interface, the thirdparty can facilitate the knowledge transfer between theinterdependent teams, which reinforces the effect of taskinterdependence on the likelihood of communication.Also, as predicted, Model 4 shows that there is a neg-ative and significant tendency for cross-network cyclicclosure (triad D in Figure 1) to form (−00235, p <0005). This suggests that when a common third partyis involved in a potentially cyclic triadic communica-tion structure with two interdependent teams, the pres-ence of this common third party may actually hinderthe exchange of information between the interdependentteams, effectively freezing their interdependence.

6.1. RobustnessWe test the robustness of our results in several ways.First, our results are robust to the exclusion of any ofthe exogenous nodal and dyadic effects, as well as tothe exclusion of the four cross-network triadic effectsthat act as controls. Second, we estimated alternativemodels in which our key variables of interest (i.e., thecross-network closure variables) are calculated takinginto account the communications with the six func-tional teams, and the results are consistent with theones presented in Table 3. Third, we also estimated thecross-network closure variables considering only com-munications with the third-party team that were matchedby the corresponding technical interface with the com-mon third-party team. The results obtained with thisalternative specification were also consistent with theones presented here. Fourth, we also tested the robust-ness of our results with alternative specifications of thewithin-network effects. Our results are robust to theinclusion of Markov parameters for transitive and cyclictriads, as well as to the inclusion of social circuit effectswith higher values of the � parameter (we use �= 2 inthe models reported here). Finally, our results are alsofully robust to the inclusion of a cross-network reci-procity effect (which is not significant in our models), aswell as to the inclusion of four additional cross-networktriadic effects that account for a focal pair of communi-cating teams and common third parties in the technicalinterface network.

6.2. Goodness-of-Fit EvaluationWe examined the goodness of fit of our ERGMs by com-paring the structural statistics of the observed networkwith the corresponding statistics on a sample of net-works simulated from the fitted model (Hunter et al. 2008,

Robins et al. 2009). We built our sample out of 10 mil-lion simulated networks. Using t-ratios3 from such asample, we estimated whether the observed graph fea-ture is extreme compared with the simulated distribution.As indicated by Robins et al. (2009, p. 112), “For effectsexplicitly [included] in the model, good convergence ofthe estimation algorithm is represented by [t-ratio] val-ues close to zero (less than 0.1 is desirable). For anobserved graph feature not included in the model, wedecided that a t-ratio less than two in absolute valueindicates that the observed feature is not unusual in theestimated graph distribution.” Overall, all our modelsseem to provide a good fit to our data as indicated notonly by t-ratios below 0.1 for all the estimated param-eters included in the models but also by t-ratios below1.0 for all additional structural parameters available inXPNet that were not included in our models. We alsoverified that the t-ratio of additional nodal and dyadicparameters that were not included in our models butwere available in XPNet were not unusual in the esti-mated graph distribution. For Model 4, all such t-ratioswere below 2.0.

Our models also seem to replicate well features thatare not typically modeled, such as the degree distri-bution and clustering coefficient (Hunter et al. 2008,Robins et al. 2009). For instance, in the full model(Model 4), the magnitudes of t-ratios for the standarddeviations of both in- and out-degree distributions wereless than 0.4, whereas the magnitudes of the t-ratiosfor the skewedness of both in- and out-degree distribu-tions were less than 1.0. For global and alternating clus-tering coefficients, the magnitudes of the t-ratios wereless than 0.3. Overall, our models (and, in particular,Model 4) seem to be consistent with the observed datafor the graph properties examined. All goodness-of-fitdiagnostics carried out after extensive simulations areavailable from the authors upon request.

7. Discussion and ConclusionsThis paper considers whether the local structure ofthe informal interteam communication networks in newproduct development organizations might affect thecommunication patterns between interdependent teams.Although our findings show that task interdependenceis a strong exogenous determinant of the communica-tion network among design teams, we show that thiseffect can be moderated by the presence of commonthird parties in this network. Our analysis of task inter-dependence and informal communication among theteams in charge of designing a major aircraft engineshows that the presence of a common third party in theinformal communication network can affect the likeli-hood of communication between interdependent teamsin different ways. Whereas task interdependence createsincentives for two teams to exchange information, their

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


communication with a common third party affects thelikelihood that they will actually do so. The directionof the effect, however, is contingent on the position thecommon third party occupies in the local communica-tion structure. When the common third party is posi-tioned in the middle of a transitive triad so that it getsinformation from the potential source and also sendsinformation to the potential recipient, then the localstructure enables the third party to be a facilitator ofthe information exchanged between the source and thereceiver in the focal dyad. In such triads, the com-mon third party may alleviate any friction associatedwith the exchange of knowledge between the source andthe recipient (Reagans and McEvily 2003, Tortorielloet al. 2012). In these local structures, the presence ofthe common third party increases the likelihood of com-munication between the two focal interdependent teams.However, when the common third party team sits inthe middle of a triad that can become cyclic if the twointerdependent teams exchanged information, then thepresence of the common third party hinders the commu-nication between the two interdependent teams, prevent-ing the formation of the cyclic communication structure.In this case, the presence of the common third partydecreases the likelihood of communication between thetwo interdependent teams, which can increase the riskof coordination disruptions in the product developmentprocess.

Our findings highlight an interesting paradox of infor-mal communication networks between interdependentactors. This communication typically results from theactor’s efforts to coordinate their interdependence, butwe show that, in some circumstances, the emerging localstructure of the communication network may affect thelikelihood of actors actually engaging in such efforts. Inexploring the mechanisms behind these effects, we arguethat the position occupied by the common third party inthe local structure can make this party a catalyst or aninhibitor of communication between two focal interde-pendent teams. When the communication between thetwo interdependent teams results in a typical transitivetriad, the local structure is more likely to turn the thirdparty into a catalyst of communication between the inter-dependent teams. Conversely, when the communicationbetween the two interdependent teams would put thethird party in the middle of a communication cycle thatcan result in an endless iteration in the design chain, thepresence of the common third party may actually inhibitthe communication between the interdependent teams.It is important, however, to highlight that these mecha-nisms do not necessarily assume intentionality or evenactive agency from the parties involved; rather, they arerooted in how the structure of the local communicationnetwork shapes the role of the common third party inways that makes it more or less likely to promote or

inhibit the communication between the two interdepen-dent focal teams.

Our study is based on comprehensive fieldwork thatenabled us to collect rich quantitative and qualitativedata on the technical interfaces between componentsof an extremely complex product as well as on thetechnical communication network among teams respon-sible for designing the components of that product.The independent measurement of task interdependencebetween teams—based on the observed technical inter-faces between their respective components—provides aunique setting to investigate how the structure of thecommunication network that emerges to address thosetechnical interfaces may itself moderate the effect of taskinterdependence in predicting interteam communication.Despite these advantages, the study has three limitationsthat result from the nature of our data.

First, our data on communication among teams cap-ture only those exchanges that were explicitly relatedto technical matters. Hence, our communication networkmay fail to capture informal social interactions betweenmembers of different teams—that is, a social networkthat might exist in addition to the observed technicalcommunication network. However, we note that unob-served social exchanges could pose a problem for ourstudy only if they are largely unrelated to the observedexchanges of technical communication. Such a possi-bility is not consistent with our qualitative observationsof the communication patterns between teams, whichsuggest that it was rare for social exchanges to occurbetween teams that did not also communicate for techni-cal reasons. This dynamic is consistent with the idea thatteams are preoccupied with their work and hence thatsocial interactions in organized settings will usually stemfrom encounters triggered by exogenous factors (in thiscase, interdependencies caused by shared interfaces). Insum, these considerations suggest that the lack of infor-mation on purely social exchanges between teams isunlikely to have biased our results. Also related to ourmeasure of technical communication, our data do notallow us to make fine-grained distinctions among var-ious types of technical communication. Unfortunately,given the complexity of the project studied, collectingthe interteam communication data at a more granularlevel was impossible at our research site. Nonetheless, itis important to acknowledge that distinguishing techni-cal communication at a more granular level could haveallowed us to test further whether there are some typesof technical flows more likely to be responsible for themoderating effects of local structures with a commonthird party observed in our study.

Second, our measure of interteam technical com-munication is based on a team’s acknowledgment ofhaving received technical information from the sourceteam. Strictly speaking, failure to observe communi-cation between the teams sharing a technical interface

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


might have resulted from failure of the recipient teamto request information or of the source team to supplyinformation, or both. By the same token, observed com-munication might have resulted from proactive behaviorby the source that is acknowledged by the recipient, evenin the absence of a specific request. Although our datacannot conclusively discriminate among these situations,our theory does not depend on this discrimination. Thisis particularly important in the case of the cyclic localstructure. Although we argue that the presence of thecommon third party in this structure hinders the commu-nication between the interdependent focal teams to pre-vent the emergence of cyclic problem-solving dynamics,we do not stipulate whether this outcome results from aless proactive recipient or from a less compliant source,or both. More generally, our reasoning focuses on howthe local communication structure shapes the behaviorsof the involved parties in ways that facilitate or hindertheir communication, without making assumptions aboutthe specific roles of the source and the recipient in gen-erating these patterns.

Another limitation associated with the way we mea-sure interteam communication is its directionality. Inreality, interteam communications are not purely direc-tional. For instance, if A talks to B, then B listens to A;this means they are communicating even if they playtwo different roles. Fortunately, our empirical settingallows us to capture in a reliable and meaningful way thedirectionality implicit in interteam communications (andintercomponent technical interfaces). The directionalityof interteam communication is determined by the impactthat the information exchanged had “on the design tasks”of the recipient team. Although this way of establish-ing directionality in communication is meaningful inour context (and probably meaningful in most contextswhere the value of the knowledge transferred is largelyvalued by the recipient), it is also true that such a direc-tionality does not capture whether the communicatingparties were engaged in more or less bidirectional infor-mation exchanges. Having said this, and given the strongtendencies for reciprocation in the communication net-work featured in this study, it is possible that many ofthe communications were indeed two-way communica-tions. Fortunately, we were able to control for the under-lying tendency to reciprocate communication ties in ouranalysis.

Third, our findings show how the structure of thecommunication network might affect the behaviors ofdesign teams involved in a complex new product devel-opment project, but we cannot say anything regard-ing the desirability or undesirability of such behav-iors. A possible interpretation of our findings is that,given the complexity of the technical interfaces in thePW4098 engine, some teams might have consciouslyneglected to attend to some interdependencies (in cyclictriads) to focus on other, more critical ones as a way to

allocate team resources more efficiently. Two elementsraise doubts about this interpretation, however. First,an additional post hoc analysis (not reported here) didnot reveal a significant association between mismatchedinterfaces in the presence of a common third party andthe strength of the mismatched technical interface. Sec-ond, anecdotal evidence from our research site suggeststhat some mismatched interfaces did cause significantrework in subsequent stages of the process. Althoughwe lack systematic data in this respect, these qualita-tive observations are fully consistent with the results ofGokpinar et al. (2010), who found a significant associ-ation between a lack of (or insufficient) communicationabout technical interfaces and quality issues in the auto-mobile industry.

Despite this evidence, we cannot say that teams were“talking too little” with other teams, as it is indeed pos-sible that omissions to establish a direct communicationbetween two interdependent teams could have some-times been an efficient way to use scarce team resources.Nor can we say that they were “talking too much.”Although our data do show instances of communicationin the absence of technical interfaces, which are dispro-portionally present within subsystem group boundariesin the formal organizational structure, such communica-tion cannot be dismissed as superfluous because it mighthave addressed unobserved technical interfaces that wereuncovered during the design process or other typesof task-related interdependencies not directly traceableto physical interfaces between components. What ourresults do show is that although the technical interfacenetwork largely determines the presence or absence ofcommunication between design teams, deviations fromthis overall tendency are systematically associated withspecific triadic configurations in the structure of thecommunication network among the teams.

Despite their limitations, our findings have importantpractical implications for managers seeking to improvethe coordination between design teams in complex prod-uct development projects. Whereas our study confirmsthe salience of task interdependence in prompting infor-mal exchanges of technical communication betweenteams, we also identify situations that can significantlyalter this association. This is important because man-agers (and engineering scholars) often assume that tech-nical interdependence should automatically trigger thecommunication necessary to coordinate design tasks(Cataldo et al. 2006, Olson et al. 2009). Managers areaware of the unintended effect of formal organizationalboundaries in hindering communication between actorsthat sit across such boundaries. However, the possibilitythat informal communication networks—often touted asa remedy to the failure of formal organization—wouldalso have unintended consequences that may increase therisk of coordination disruptions in product development

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


projects has not been considered. In particular, the pos-sibility that the presence of a common third party thatparticipates in potentially cyclic communication patternscan reduce the likelihood of other teams communicatingto coordinate their interdependence should prompt man-agers to pay special attention to identify and manageactors that are likely involved in cycles in the commu-nication network (Sosa et al. 2013). The consequencesof not doing so may be significant. Even minor disrup-tions in coordination may lead to design inadequaciesthat, although not critical, could affect the performanceor durability of the affected components and subsystems,causing significant warranty or service expenses over thelife of the product. For example, if a critical compo-nent of an aircraft engine (e.g., a turbine airfoil) failsto reach its life expectancy, this will cause additionalengine removals for maintenance. For an engine such asthe PW4098, a single such removal could cost the cus-tomer as much as $150,000, in addition to the loss ofrevenues associated with a grounded plane.

In conclusion, we refer to Simon (1996), who sug-gested that a complex system is difficult to understandbecause the behavior of the whole depends in nontriv-ial ways on how its elements interact. In studying thedeterminants of interteam communications when design-ing a complex system, we have learned that commonthird-party teams have a significant moderating influenceon the predictive power of the patterns of task interde-pendence. We have shown that our results have impor-tant implications for both theory and practice. Theyalso raise a number of interesting questions. How doesthe presence of common third parties to a focal tech-nical interface influence the ability of interdependentteams to discover and attend to other relevant interfacesthat may also require interteam communication? Howdo component connectivity and team network structurerelate to such important design decisions as componentoutsourcing and component redesign? With the uprisein open innovation projects, how does the relationshipbetween task interdependence and interteam communi-cation change when development projects are carried outacross firm boundaries? In this paper we have describeda path that can lead organizational scholars in answeringthese important questions.

AcknowledgmentsThe authors appreciate the assistance of the engineers at Pratt &Whitney Aircraft for their collaboration during the data collec-tion of this study. The authors are grateful to Steven Eppingerfor his supervision of and contributions to the research pro-gram associated with the data featured in this paper. They thankChristoph Loch for his encouraging feedback on an early ver-sion of this work. They appreciate the comments from GokhanErtug, Bruce Kogut, Luk Van Wassenhove, and participants inseminars at INSEAD, University of Michigan’s Ross Schoolof Business, The Wharton School, and MIT Sloan School ofManagement. The authors are also grateful to the senior editor,

Bill McEvily, and three anonymous referees for their commentsduring the review process.

Endnotes1The internal summation of these network statistics is definedup to N , where N = 54 so that the triadic configurations con-siders all 54 design teams as possible common third partiesbetween i and j . However, we can also set N = 60 to accountfor the six functional teams as common third parties. Ourresults are robust to both specifications.2To estimate these effects, we consider the two-path statisticassociated with these triadic configurations (i.e., the internalsummation of these network statistics) as an exogenous dyadicpredictor. We then allow XPNet to estimate the interaction ofsuch a dyadic predictor with the term “ArcAB.” Consideringthe two-path statistic to be exogenous is a necessary approxi-mation due to the limitations of all existing software packagesto estimate our full model. This is a good approximation, how-ever, to the extent that all lower-order effects are included inour model.3“These t-ratios are calculated in the traditional way as thedifference between the observed value of a particular graphstatistic and the mean from the sample of simulated graphs, asa ratio of the standard deviation from the simulated sample”(Robins et al. 2009, p. 112).

References

Allen TJ (1977) Managing the Flow of Technology (MIT Press,Cambridge, MA).

Allen TJ (2004) Communication networks in R&D laboratories. KatzR, ed. The Human Side of Managing Technological Innovation:A Collection of Readings, Chapter 26 (Oxford University Press,New York).

Ancona DG, Caldwell DF (1992) Bridging the boundary: Externalactivity and performance in organizational teams. Admin. Sci.Quart. 37(4):634–665.

Baldwin CY, Clark KB (2000) Design Rules: Volume 1: The Powerof Modularity (MIT Press, Cambridge MA).

Brown SL, Eisenhardt KM (1995) Product development: Pastresearch, present findings, and future directions. Acad. Manage-ment Rev. 20(2):343–378.

Burt RS (2001) Bandwidth and echo: Trust, information, and gossip insocial networks. Rauch JE, Casella A, eds. Networks and Markets(Russell Sage Foundation, New York), 30–74.

Burt RS (2005) Brokerage and Closure: An Introduction to SocialCapital (Oxford University Press, Oxford, UK).

Burt RS, Knez M (1995) Kinds of third-party effects on trust. Ratio-nality Soc. 7(3):255–292.

Cataldo M, Wagstrom P, Herbsleb JD, Carley KM (2006) Identifica-tion of coordination requirements: Implications for the design ofcollaboration and awareness tools. Proc. ACM Conf. Comput.-Supported Cooperative Work (ACM, New York), 353–362.

Coleman JS (1990) Foundations of Social Theory (Belknap Press,Cambridge, MA).

Colfer L, Baldwin C (2010) The mirroring hypothesis: Theory, evi-dence, and exceptions. HBS Working Paper 10-058, HarvardBusiness School, Boston.

Contractor NS, Wasserman S, Faust K (2006) Testing multitheoret-ical, multilevel hypotheses about organizational networks: Ananalytic framework and empirical example. Acad. Management.Rev. 31(3):681–703.

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.


Eppinger SD, Whitney DE, Smith RP, Gebala DA (1994) A model-based method for organizing tasks in product development. Res.Engrg. Design 6(1):1–13.

Galbraith JR (1973) Designing Complex Organizations (Addison-Wesley, Reading, MA).

Gargiulo M, Benassi M (2000) Trapped in your own net? Networkcohesion, structural holes, and the adaptation of social capital.Organ. Sci. 11(2):183–196.

Gokpinar B, Hopp WJ, Iravani SMR (2010) The impact of mis-alignment of organizational structure and product architectureon quality of complex product development. Management. Sci.56(3):468–484.

Granovetter M (1973) The strength of weak ties. Amer. J. Sociol.78(6):1360–1380.

Granovetter MS (1985) Economic action and social structure: Theproblem of embeddedness. Amer. J. Sociol. 91(3):481–510.

Gulati R (2010) Reorganize for Resilience: Putting Customers atthe Center of Your Business (Harvard Business School Press,Boston).

Henderson R, Clark K (1990) Architectural innovation: The reconfig-uration of existing product technologies and the failure of estab-lished firms. Admin. Sci. Quart. 35(1):9–30.

Hunter DR, Goodreau SM, Handcock MS (2008) Goodness of fitof social network models. J. Amer. Statist. Assoc. 103(481):248–258.

Lessard DR, Zaheer S (1996) Breaking the silos: Distributed knowl-edge and strategic responses to volatile exchange rates. StrategicManagement J. 17(7):513–533.

Lomi A, Pattison P (2006) Manufacturing relations: An empiricalstudy of the organization of production across multiple networks.Organ. Sci. 17(3):313–332.

Lomi A, Lusher D, Pattison PE, Robins G (2014) The focused organi-zation of advice relations: A study in boundary crossing. Organ.Sci. 25(2):438–457.

Lusher D, Koskinen J, Robins G (2013) Exponential Random GraphModels for Social Networks: Theory, Methods, and Applications(Cambridge University Press, New York).

Marsden PV (1990) Network data and measurement. Annual Rev.Sociol. 16:435–463.

McCord KR, Eppinger SD (1993) Managing the integration prob-lem in concurrent engineering. Sloan School of ManagementWorking Paper 3594, Massachusetts Institute of Technology,Cambridge.

McEvily B, Soda G, Tortoriello M (2014) More formally: Rediscov-ering the missing link between formal organization and informalsocial structure. Acad. Management Ann. 8(1):299–345.

Mihm J, Loch C, Huchzermeier A (2003) Problem-solving oscilla-tions in complex engineering projects. Management Sci. 46(6):733–750.

Morelli MD, Eppinger SD, Gulati RK (1995) Predicting techni-cal communication in product development organizations. IEEETrans. Engrg. Management 42(3):215–222.

Nadler DA, Tushman M (1997) Competing by Design: ThePower of Organizational Architecture (Oxford University Press,New York).

Olson J, Cagan J, Kotovsky K (2009) Unlocking organizationalpotential: A computational platform for investigating structuralinterdependence in design. ASME J. Mech. Design 131(3):031001–031001-13.

Rank ON, Robins G, Pattison P (2010) Structural logic of intraorga-nizational networks. Organ. Sci. 21(3):745–764.

Raub W, Weesie J (1990) Reputation and efficiency in socialinteractions: An example of network effects. Amer. J. Sociol.96(3):626–654.

Reagans R, McEvily B (2003) Network structure and knowledgetransfer: The effects of cohesion and range. Admin. Sci. Quart.48(2):240–267.

Reagans R, Zuckerman EW (2001) Networks, diversity, and perfor-mance: The social capital of corporate R&D teams. Organ. Sci.12(4):502–517.

Reagans R, Zuckerman E, McEvily B (2004) How to make the team:Social networks: Social networks vs. demography as criteria fordesigning effective teams. Admin. Sci. Quart. 49(1):101–133.

Robins GL, Pattison PE, Wang P (2009) Closure, connectivity anddegree distributions: Exponential random graph (p∗) models fordirected social networks. Soc. Networks 31(2):105–117.

Rowles CM (1999) System integration analysis of a large commercialaircraft engine. Master’s dissertation, Massachusetts Institute ofTechnology, Cambridge.

Simmel G (1950) The triad. Wolff KH, ed., trans. The Sociology ofGeorg Simmel (Simon & Schuster, New York), 145–169.

Simon HA (1996) The Sciences of the Artificial, 3rd ed. (MIT Press,Cambridge, MA).

Smith RP, Eppinger SD (1997) Identifying controlling features ofengineering design iteration. Management Sci. 43(3):276–293.

Snijders TAB, Pattison PE, Robins GL, Handcock MS (2006) Newspecifications for exponential random graph models. Sociol.Methodol. 36(1):99–153.

Sosa ME (2008) A structured approach to predicting and manag-ing technical interactions in software development. Res. Engrg.Design 19(1):47–70.

Sosa ME, Eppinger SD, Rowles CM (2004) The misalignment ofproduct and organizational structures in complex product devel-opment. Management Sci. 50(12):1674–1689.

Sosa ME, Eppinger SD, Rowles CM (2007) A network approach todefine modularity of components in complex products. ASME J.Mech. Design 129(11):1118–1129.

Sosa ME, Mihm J, Browning TR (2013) Linking cyclicality andproduct quality. Manufacturing Service Oper. Management15(3):473–491.

Sosa ME, Eppinger SD, Pich M, McKendrick D, Stout S (2002) Fac-tors that influence technical communication in distributed prod-uct development: An empirical study in the telecommunicationsindustry. IEEE Trans. Engrg. Management 49(1):45–58.

Terwiesch C, Loch CH, De Meyer A (2002) Exchanging preliminaryinformation in concurrent engineering: Alternative coordinationstrategies. Organ. Sci. 13(4):402–419.

Thompson JD (1967) Organizations in Action (McGraw-Hill,New York).

Tortoriello M, Reagans R, McEvily B (2012) Bridging the knowledgegap: The influence of strong ties, network cohesion, and networkrange on the transfer of knowledge between organizational units.Organ. Sci. 23(4):1024–1039.

Tushman ML (1977) Special boundary roles in the innovation process.Admin. Sci. Quart. 22(4):587–605.

Tushman M, Katz R (1980) External communication and project per-formance: An investigation into the role of gatekeeper. Manage-ment Sci. 26(11):1071–1085.

Van den Bulte C, Moenaert RK (1998) The effects of R&D teamco-location on communication patterns among R&D, marketing,and manufacturing. Management Sci. 44(11, Part 2):S1–S19.

Wang P, Robins G, Pattison P (2009) PNet: A program for the simu-lation and estimation of exponential random graph (p∗) models.(Software.) University of Melbourne, Melbourne, VIC, Australia.http://sna.unimelb.edu.au/PNet.

Wasserman S, Pattison P (1996) Logit models and logistic regressionfor social networks: I. an introduction to Markov graphs and p∗.Psychometrika 61(3):401–425.

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

http://sna.unimelb.edu.au/PNet.


Manuel E. Sosa is associate professor of Technology andOperations Management at INSEAD. He received his Ph.D.from the Massachusetts Institute of Technology. His researchinvestigates coordination and innovation networks in new prod-uct development organizations.

Martin Gargiulo is professor of Organizational Behaviorat the INSEAD Asia Campus in Singapore. He received hisPh.D. in sociology from Columbia University, New York. Hisresearch focuses on the dynamics and effects of informal net-work structures, with emphasis on their role in facilitating or

hindering coordination between interdependent organizationalactors.

Craig Rowles is currently providing strategic and opera-tional leadership as executive vice president of Ei3 Corpora-tion, a leading-edge provider of industrial Internet services.As an engineering manager at Pratt & Whitney Aircraft, hejoined the inaugural class of the Systems Design and Manage-ment program and received an M.S. degree in engineering andmanagement from the Massachusetts Institute of Technology’sSchool of Engineering and Sloan School of Management.

Dow

nloa

ded

from

info

rms.

org

by [

175.

176.

173.

30]

on 0

2 A

ugus

t 201

5, a

t 23:

47 .

For

pers

onal

use

onl

y, a

ll ri

ghts

res

erve

d.

can informal communication networks disrupt … · sosa, gargiulo, and rowles: can informal...

Documents