You are on page 1of 32

J. Eng. Technol. Manage.

21 (2004) 83114

Exploring why more communication is not better: insights from a computational model of cross-functional teams
Ralitza R. Patrashkova, Sara A. McComb
Isenberg School of Management, University of Massachusetts Amherst, Amherst, MA 01003, USA

Abstract Recent evidence suggests that communication and performance in cross-functional new product development (NPD) teams are curvilinearly related, but fails to pinpoint the reasons for this relationship. We developed a computational model to study the communication activities of cross-functional new product development teams. Our simulation conrms the recent evidence and offers insights into the underlying reasons for the curvilinearity. We provide guidelines regarding when the top performance occurs, for both frequency and duration of synchronous and asynchronous communication. Further, we perform a series of post-hoc analyses to examine the reasons for the curvilinearity of the communicationperformance relationship. The work concludes with a discussion of the theoretical and practical applications of the results. 2004 Elsevier B.V. All rights reserved.
JEL classication: C63; O31 Keywords: Communication; Performance; New product development; Cross-functional teams; Simulation

1. Introduction Communication is an essential component of the new product development (NPD) process (Brown and Eisenhardt, 1995). The challenge for cross-functional teams (CFT), routinely used for NPD (Denison et al., 1996), is to ascertain the level of information exchange among team members that will allow them to optimize their performance. Communication frequency is often explored with the assumption that it is linearly related to performance (Allen, 1977; Katz and Tushman, 1981; Ancona and Caldwell, 1992;
Corresponding author. Tel.: +1-413-545-5681; fax: +1-413-545-3858. E-mail address: mccomb@som.umass.edu (S.A. McComb).

0923-4748/$ see front matter 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.jengtecman.2003.12.005

84

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

Smith et al., 1994). Recently, evidence has shown, however, that both high and low levels of communication can impede team performance, thus suggesting a curvilinear relationship between performance and team communication (Hutchins, 1995; Patrashkova et al., 2003). One possible explanation for these results is offered by the research on information processing. Team members have limits on the amount of information they can process (Boisot, 1995), as too much information overloads the capabilities of team members and inhibits their performance (Goodman et al., 1986). At the same time, infrequent communication cannot supply the necessary information, which also leads to low performance. The main objective of this research is to provide insights into how communication relates to performance and why this relationship holds. In order to obtain this goal, we develop and test a computational model of communication in CFT. Computational modeling refers to incorporating mathematical and theoretical models into computer simulations (Hulin and Ilgen, 2000; Zeigler, 1976). We use computational modeling as a primary research tool, because when using it the researcher is able to control the variables under consideration, manipulate them and examine all possible combinations and interactions (Lant, 1994). The model we derive, formalize and code, is based on theoretical assumptions stemming from extant theory and empirical results. We use the model to verify earlier research by assessing whether too much, or too little, communication among team members impedes performance. Further, after establishing the relationship, we perform a series of post-hoc analyses to gain a better understanding of communication in teams. Specically, we explore the effects of information content, team members expertise and project complexity on the communication/performance relationship. Comparing results allows us to isolate the impact of these variables. This work makes several contributions to research and practice. To our knowledge, this is the rst computational model of CFT communication processes. Designing the model required us to precisely specify many relationships among variables that are implied, but not quantied, by theory and to formalize a team interaction procedure. The existence of such a computational model allows us to move beyond conrming a relationship between variables to an explicit understanding of the nature of that relationship. Consequently, our ndings are a verication and extension of earlier work in this area. Finally, the results we obtained give clear guidance about identifying the level of communication corresponding to tops level of performance.

2. Communication in cross-functional teams Communication is the primary means through which CFTs collaborate. In her study of NPD teams, Dougherty (1992) observed how difcult, yet essential, it is for team members to effectively collaborate, and therefore, effectively communicate. Team members are typically drawn from many different functional areas within an organization and they bring their unique perspectives, or thought worlds to the team. As they exchange information through communication, the team members may have a difcult time collaborating if they do not compensate for their different perspectives regarding the teams work. Too little information exchanged will result in confusion and misunderstandings. Alternatively, too

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

85

much new information may tax the information processing capabilities of the team members (Boisot, 1995; Goodman et al., 1986). Further, if they select an inappropriate medium for these information exchanges, misinterpretations can also result (Carlson and Zmud, 1999). Effective communication, therefore, requires that team members select the most appropriate medium for the information transfer and communicate the optimal amount of information in order to achieve top performance. Media selection refers to the communication medium (e.g., telephone, email) chosen to transfer information (Daft and Lengel, 1986). When selecting media, team members must decide how best to communicate the requisite information. Two theoretically different media are available: synchronous or asynchronous. Synchronous communication media (e.g., face-to-face meetings, telephone conversations) are employed when two or more members engage at the same time in the communication act, whereas asynchronous communication (e.g., electronic mail, written communication) occurs when the members do not engage in communication at the same time (Levitt et al., 1994). Synchronous and asynchronous communication media have different capabilities to transfer information (Daft and Lengel, 1986). Specically, synchronous media are able to transfer more information per message than asynchronous media, because it utilizes more channels (e.g., facial expressions, intonation) for information transfer. For example, in face-to-face communication the tone of the voice, the context and the facial expressions are used as additional cues clarifying and supplementing the information content of the message. Asynchronous media lack these additional channels for information transfer. We quantify the amount of communication that occurs during information exchange by measuring both the frequency and duration of the interactions. The majority of research has assessed amount of communication by measuring communication frequency (number of messages exchanged) (e.g., Ancona and Caldwell, 1992; Patrashkova et al., 2003; Smith et al., 1994). Frequency, however, does not distinguish between long information intensive meetings and short emails asking for a small amount of information. Communication duration (the time in which team members are engaged in communication) has not been used as often to capture the amount of communication transpiring among individuals (e.g., Kraut et al., 1990). We include duration because it provides a more comprehensive depiction of the communication activities of the team. Much of the past research on team communication presumes that more communication among team members will lead to higher performance (Allen, 1977; Katz and Tushman, 1981). Recently, communication has been shown to be curvilinearly related to performance (Hutchins, 1995; Patrashkova et al., 2003). Hutchins (1995), using computational modeling, compared the development of cognitive maps of team members based on their frequency of communication. His results show that more communication is not always better. When members of a group exchange too much information their cognitive maps become too similar, and the group is assumed to be incapable of innovation. Too little communication, conversely, will not bring the cognitive maps close enough for a mutual understanding. These results give insights into what happens in a group when its members communicate, however, they do not explain how communication relates to performance. The study by Patrashkova et al. (2003) is a cross-sectional investigation of the relationship between communication frequency and team performance. Using a sample of 60 project

86

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

teams, they found that performance decreases after some peak communication frequency is reached. The empirically established curvilinear relationship holds for several types of performance (goal achievement, project efciency, and cohesion) and two communication media (face-to-face and e-mail). The relationship is particularly pronounced for e-mail communication. These results empirically establish the curvilinear relationship between communication and performance, but, because of the research design, no possible causes could be explored. Based on the results of Hutchins (1995) and Patrashkova et al. (2003), we suggest that the relationship between communication frequency and performance will be curvilinear. Further, because Patrashkova et al. (2003) found unique curvilinear relationships for synchronous and asynchronous communication media, we examine them separately. Thus, we hypothesize: Hypothesis 1. Team performance will have a curvilinear relationship with team synchronous communication frequency, such that low and high communication frequencies will be associated with lower levels of team performance while moderate levels of communication will be associated with high levels of team performance. Hypothesis 2. Team performance will have a curvilinear relationship with team asynchronous communication frequency, such that low and high communication frequencies will be associated with lower levels of team performance while moderate levels of communication will be associated with high levels of team performance. We also extend previous research by including communication duration as an alternative measure of the communication activities of the team. To our knowledge, no other work has compared the measures of communication frequency and duration, nor have they attempted to study the relationship between duration and performance. We, therefore, rely on logic to propose that communication duration will behave similarly to communication frequency with respect to team performance. Thus, we proffer the following hypotheses: Hypothesis 3. Team performance will have a curvilinear relationship with team synchronous communication duration, such that low and high communication duration will be associated with lower levels of team performance while moderate levels of communication will be associated with high levels of team performance. Hypothesis 4. Team performance will have a curvilinear relationship with team asynchronous communication duration, such that low and high communication duration will be associated with lower levels of team performance while moderate levels of communication will be associated with high levels of team performance. Our methodology provides us with the opportunity to further extend our knowledge about the relationship between communication and performance by examining why the results occur. We accomplish this task by conducting post-hoc analyses where certain variables of interest are held constant to determine their effects. Additional experiments, such as these, are easily conducted using computational modeling and simulation.

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

87

3. Research approach To extend our understanding of the team communication process, we utilize a computational modeling approach. Computational modeling and simulation refer to an approach where the researcher develops a theoretical model of the system of interest, formalizes this model by developing an algorithm of system behavior, codes the model in a computer programming language and subsequently executes the code so data about the behavior of the system are obtained (Law and Kelton, 2000; Zeigler, 1976). Computational modeling has been successfully used in research to represent many organizational constructs, including withdrawal (Hanish, 2000), training and turnover (Glance et al., 1997), organizational learning (Lant and Mezias, 1992), cultural transmission (Harrison and Carroll, 1991), and team decision making (Kang et al., 1998). The computational modeling process begins when the researcher develops a theoretical model as a series of decision rules that represent theories of human behavior. The theoretical model is used as a basis upon which a computer algorithm is generated. This computer algorithm formalizes the behaviors of the elements of the system and determines the output (Hulin and Ilgen, 2000). The computer algorithm needs to be coded in a high level programming language, such as C, C++, Pascal or the like. The modeled system is simulated when the resulting code is executed, or run. Each run gives a single observation of the system behavior. The development of a computational model requires both an explicit quantication of the variables included and a detailed specication of the relationships among the variables. Thus, the design process of a computational model forces the researcher to be systematic and specic in the behavior description (Kang et al., 1998). The resulting model is a simplied reality, allowing reliable causal relationships to be established. Computational models are developed to address the functioning of complex systems and the behavior of individuals in such systems by focusing on what if? questions. Although it is a simplied reality, the model is complex enough to adequately test theoretical assumptions (Bendor and Moe, 1992). Computational modeling and simulation are an especially appropriate methodology for studying teamwork and quite suitable for representation of human information processing activities necessary for effective team communication (Kang et al., 1998). Further, the combination of modeling and simulation gives the researcher much exibility. Specically, modeling provides the ability to investigate the effects of individual variables by keeping them constant. Simulation also provides large samples through the possibility to run the code as many times as required (Taber and Timpone, 1996). In sum, computational modeling can, and should, be used as a research tool (Bendor and Moe, 1992). 4. Computational model description The melding of perspectives required in a CFT is achieved through exchanging and processing the information the team members possess (Hinsz et al., 1997). We, therefore, base our computational model on information processing theory (see Schroder et al., 1967; Streufert and Streufert, 1978; Streufert and Swezey, 1986). NPD is a process through

88

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

which information, in the form of requirements, is converted into information that describes the nal product. This transformation is an information processing activity (Safoutin and Thurston, 1993). In our model the specic information each team member possesses is transferred to the team through communication. This received information is subsequently processed. This process is repeated until all the necessary information is exchanged. Fig. 1 highlights the variables and decision rules in our model. The simulation begins by generating a project. The project is used to determine the information content requirements, which quantify the task and establish the project schedule. Next, a team is generated based on the information content requirements of the project. In the third phase of the computational model, the team members communicate with each other until the information content

Fig. 1. Structure of the model.

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

89

requirements are fullled. At the end of the simulation, communication frequency, communication duration and on-schedule performance are calculated. Each of these phases is discussed in the subsequent paragraphs. Details regarding specic variable generation and decision rules are provided in Appendix A. 4.1. The project The project is the assignment for which a CFT is created and about which the team members must communicate. In order to develop a high level formalization of the project, we model it as information units. This approach allows us to use information as both a knowledge objective and a unit of communication. Specically, knowledge intensive tasks can be expressed as a number of information units (Streufert and Streufert, 1978). Moreover, information can also represent the tacit knowledge exchanged among team members to accomplish their assignment (Boisot, 1995). Taken together, information provides us with a means of initializing the scope of the project and assessing when enough information has been exchanged to consider the project complete. This representation allows us to cover a wide range of projects, thereby achieving greater generality, greater realism and greater explanatory power (Boisot, 1995). To generate a project, we begin with the establishment of project parameters, particularly cost and technical objectives, as is typical in NPD projects (Bowen et al., 1994). In our model, the cost parameter is randomly generated. We then use cost to determine the technical objectives. Organizational capability is also important to determine the amount of work required to complete any project, therefore we include this parameter and randomly generate a value for it. Organizational capability and the technical objectives are used to establish the information content requirements. Information content requirements refer to the overall number of information units a specic project needs to be completed. Each project is also represented as a structured sequence of events, shown in Table 1 (Jones, 1997), that does not change from one simulation run to the next. We use this structure and the information content requirements to determine the project schedule (i.e. the amount of time allocated to each sub-phase) as well as the information content requirements for each sub-phase by functional area. In sum, through this portion of the model, the project parameters are transferred from a high level objective (cost and technical objectives) to specic information content requirements and timeframes associated with each sub-phase by functional area. 4.2. The team The team is created to communicate and process the information needed to complete the initialized project. The process of establishing the team begins with the determination of the team size, based on the information content requirements. Table 1 shows the functional composition of the team, delineating which functional areas participate in each of the seventeen sub-phases. We include eleven functional areas previous research has identied as critical for effective NPD (Brown and and Eisenhardt, 1995; Sethi et al., 2001) and distinguish between the dominant and participating functional areas for each sub-phase (Jones, 1997). The specic information content requirements are used to determine how

90

Table 1 Seventeen project phases and the corresponding functional representation Customer Predevelopment (1) New product opportunity examined (2) Need identied (3) Ideas generated (4) Ideas assessed (5) Project planned Development (6) Concept dened (7) Design established (8) Ideas developed (9) Ideas modeled Execution/implementation (10) Technical requirements detailed (11) Work schedules executed (12) Prototype tested (13) Product developed Termination (14) Product nalized (15) Product reviewed and accepted (16) Manufacturing started (17) Project evaluated ( ): dominant functional area, ( ): participating functional area. Marketing R&D Engineering Manufacturing Sales Quality Finance Purchasing Project management Product support R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

91

many team members from each functional area will participate in the team. Thus, the team size varies from project to project. After the team size, and corresponding functional composition, has been determined, each team member is given a set of personal characteristics identied as important to effective team functioning. The characteristics are functional expertise, communication potential, cohesion and ability to process information (Gladstein, 1984). Together, these characteristics are used to determine how effectively the team members exchange and process information. 4.3. Team collaboration process During the team process, team members communicate with each other. The communication process, as modeled in this work, consists of: identifying a need for information, generating a message, communicating the message, and processing the received information. Specically, each team member is able to detect a need for information based on the difference between the information content requirements and the information s/he possesses. Further, the team member can nd the appropriate source for getting this information (within or outside the team, as boundary spanning is important for team performance (Ancona and Caldwell, 1992)) and communicate his/her need to the source. The source generates a message containing information in a response. The message is characterized by duration, information content, relevance, complexity and ambiguity (Boisot, 1995). The receiver of the message then processes the information and determines if more information is needed. This cycle continues until each individual team member has satised his/her information content requirements. The generated message is sent to the requester either synchronously or asynchronously. The selection of the appropriate medium is done based on the ability of the medium to transfer information (Daft and Lengel, 1986). Therefore, team members choose a communication medium depending upon the ambiguity of the information that needs to be transferred. Highly ambiguous information is transmitted via synchronous medium, while asynchronous communication is used when the information communicated has lower ambiguity. 4.4. Team output: team performance In this research, we focus on team-level on-schedule performance. One of the main reasons cross-functional teams are assembled is the ability of a CFT to reduce product development time (Cardinal and Lei, 2000). Further, as these reductions are the result of a participative (i.e. collective) process, instead of a linear, sequential, individualized process (Jassawalla and Sashittal, 2000), we consider on-schedule performance a valuable indicator of team performance. Team performance is evaluated at the end of the simulation using the earned value procedure, which allows for comparisons across simulations by calculating a ratio of budgeted time (project schedule) to actual time (cumulative time required for message request, message generation, information exchange, and processing). For a detailed description of the earned value procedure, see Kerzner (2001).

92

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

4.5. Team output: team communication At the end of the simulation, communication frequency and duration are calculated. Communication frequency is a count of the number of messages exchanged throughout the whole project. Communication duration is the sum of the duration of all messages exchanged throughout the project. We calculate frequency and duration separately for synchronous and asynchronous media. 4.6. Simulation experiments The computational model is coded in Visual C++ computer programming language. A total of 10,000 projects, employing the decision rules described, were simulated. This number of simulations provides results with an absolute error of 0.01 for on-schedule performance with 95% precision (Law and Kelton, 2000). 5. Results Graphical summaries of the simulation results are shown in Figs. 2 and 3. Each data point represents the relationship between on-schedule performance (on the Y-axis) and communication (on the X-axis) for a single simulation run. Thus, each point represents the relationship between performance and the communication (synchronous or asynchronous, frequency or duration) for a unique project with a random set of parameters. As can be seen from the gures, our hypotheses are supported for all conditions. 5.1. Communication frequency and on-schedule performance We examine the relationships between communication frequency and performance for synchronous and asynchronous communication. The relationship between synchronous communication frequency and on-schedule performance is given in Fig. 2a. The resulting curve conrms the curvilinear form of the communication/performance relationship. When a small number of synchronous communications occur, the performance varies from very low to very high and does not show any systematic relationship. Fig. 2b presents a view of the relationship between synchronous communication frequency and on-schedule performance in order to better observe the optimal peak. After analyzing this gure and reviewing the raw data, we determined that the best performance is achieved for frequencies between about 10 and 75 communications. The performance rapidly decreases when more than 75 communications occur. We utilized the same procedure to ascertain the levels of communication associated with peak performance reported herein. The relationship between frequency of asynchronous communication and performance shows a similar trend (see Fig. 2c). Communication frequencies below 5 do not affect performance in any systematic way. Peak performance occurs at about 40 communications and decreases after about 140 communications, though not as steeply as for synchronous communication.

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

93

5.2. Communication duration and on-schedule performance Communication duration also conrmed the curvilinear relationship expected. The relationships between duration of synchronous communication and performance (Fig. 3a) and between asynchronous communication and performance (Fig. 3b) exhibit the same uctuations for very small values, as do communication frequency. Thus, there is no systematic relationship for communication with short duration. Still, the relationship with performance is curvilinear, as there is a drop in performance when communication duration increases. The

Fig. 2. (a) Relationship between synchronous communication frequency and performance;(b) truncated graph relationship between synchronous communication frequency and performance for values up till 200 communications; (c) relationship between asynchronous communication frequency and performance.

94

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

Fig. 2. (Continued ).

peak occurs at approximately 120 h for synchronous communication and at approximately 70 h for asynchronous. 5.3. Post-hoc analyses One of the primary benets of computational modeling and simulation as research tools is that they provide the researcher with the ability to conduct post-hoc analyses that conrm or reject theoretical explanations of the observed behavior. The theoretical model exists as a set of decision and variable generation rules. These rules can be easily modied to pinpoint their effect on the system. We performed several post-hoc analyses to deepen our understanding of the communicationperformance relationship and conrm the theoretical explanations we are offering. For each of these post-hoc analyses a variable or set of variables were held constant as a control. In each case, we performed an additional 2000 simulation runs. The specic analyses and the results obtained are described in the following paragraphs. 5.4. Effects of project complexity We conducted a post-hoc analysis in which we controlled for project complexity. This analysis was necessary to ensure that the results from the original computational model were not unduly inuenced by the project. In other words, we wanted to examine if all projects resulting in low levels of communication were also of low complexity, and vice versa. To this end, we set project complexity at a medium level, which results in the same project being conducted in all simulation runs. All other parameters were allowed to vary as per the original design. Controlling for project complexity in this manner ensures that performance is dependent only upon communication and the respective characteristics of

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

95

Fig. 3. Relationship between (a) synchronous and (b) asynchronous communication duration and performance.

the team members. The results are presented in Fig. 4a and b. The peak performance is achieved for about 2040 synchronous communications, while peak performance for asynchronous communication is achieved for about 7090 communications. Both relationships were curvilinear as expected and, therefore, provide further conrmation of our hypotheses. The results also validate that our original results were not a function of the complexity of the projects being undertaken in each simulation run.

96

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

Fig. 4. Relationship between (a) synchronous and (b) asynchronous communication frequency/performance for single project.

5.5. Efciency of communication media In order to test whether asynchronous communication is less efcient than synchronous media, we performed the second post-hoc analysis. Specically, we reduced the information content being transferred and increased the duration required for the exchange when generating asynchronous messages. For example, if in the original model e-mail carried 5 units of information and took 20 min to be understood, in the post-hoc analysis the same e-mail carried 3 units of information and took 25 min to be understood. The relationships are given in Fig. 5a and b. The resulting curve conrms the curvilinear form of the communication/performance relationship. After analyzing the data, we determined that the best performance for synchronous communication appears to be achieved for frequencies

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

97

Fig. 5. Relationship between (a) synchronous and (b) asynchronous communication frequency/performance for low information content.

between about 130 and 300 communications. For asynchronous communication the best performance is achieved for about 7590 communications. 5.6. Effects of team member skills In the third post-hoc analysis, we simulated team performance where all team members were generated with lower levels of expertise than in the original model, to see whether the skills of team members affect the communication/performance relationship. Our results (Fig. 6a and b) show a curvilinear relationship with best performance achieved for

98

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

Fig. 6. Relationship between (a) synchronous and (b) asynchronous communication frequency/performance for low team member skills.

synchronous communication between about 90 and 140 communications and for asynchronous about 1300 communications. An interesting phenomenon that can be observed in these gures is the almost complete lack of an unsystematic relationship at low levels of communication. 5.7. Amount of information exchanged In our nal post-hoc analysis, we explored the relationship between the amount of information exchanged and team performance. To accomplish this post-hoc analysis, we

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

99

Fig. 7. Relationship between amount of information exchanged and performance.

used the earned value procedure (Kerzner, 2001) to calculate the ratio of effective information exchange. Specically, we divided what the team members actually communicated (quantity of information) by what they needed to communicate (information content requirements). A ratio of one indicates that exactly the right amount of information was exchanged. If the ratio is greater than one, excessive amounts of information were communicated, and vice versa. Using the data from the original simulation runs, we plotted effective information exchange versus the performance achieved for each project (Fig. 7).

6. Discussion We constructed a computational model to formalize and increase our understanding of the relationship between communication and on-schedule performance. The results conform to our expectations that a curvilinear relationship exists between communication and on-schedule performance. Regardless of the nature of communication (synchronous/ asynchronous) and the measurement system used (frequency/duration), the results follow a similar pattern. Furthermore, as demonstrated in our rst post-hoc analysis, these results do not appear to be a function of the project about which the team is communicating. Overall, low levels of communication result in sporadic performance, the relationship peaks at a mid-level of communication and tapers off as communication increases. In addition to conrming our hypotheses, our results show a striking similarity between the behavior of communication frequency and duration. Both measures exhibit nearly the same curvilinear relationships. We conclude, therefore, that communication frequency, despite its decien-

100

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

cies as a measure of communication, is indeed a good approximation of team communication activity. The most interesting difference among these results is the relative location of the maximum points for synchronous and asynchronous communication. Asynchronous communication (Figs. 2c and 3b) peaks earlier than synchronous communication (Figs. 2a and 3a). We did not anticipate that asynchronous communication would create information overload before synchronous media, because in our model asynchronous media are used to exchange small, easily processed units of information. A possible explanation for this result is that because asynchronous messages carry less information content, they require relatively more time to complete the entire exchange. As performance depends on the information acquired, a medium unable to supply the requisite information fast enough may hinder performance. For example, if two individuals are attempting to set a meeting time, several e-mail exchanges may occur as they negotiate the time. A telephone conversation or a face-to-face interaction may accomplish the same purpose with one communication exchange. The results of our second post-hoc analysis conrm our explanation. By reducing the information content being transferred and increasing the duration required for an exchange when generating asynchronous messages, we caused the information overload to occur much more quickly than when messages were generated according to the original set of rules (see Appendix A). As can be seen in Fig. 5a and b, asynchronous communication impacts performance at a lower level (about 90 communications) than in the original simulation (about 140 communications) (Fig. 2c). In sum, our results show that the exchange of a high number of asynchronous messages, no matter how short or easy to process, will hinder performance. A similarity among all the results is the sporadic performance demonstrated at low levels of communication (see Figs. 2 and 3). These unsystematic results are less dramatic as communication increases. One would anticipate that low levels of communication would not provide enough information to achieve high performance, which is the result in approximately half of the simulated runs where low levels of communication occur. In the cases where low levels of communication result in high performance, team member skills may play a signicant role. If highly competent team members are collaborating on a reasonably easy project, high performance should result. To see whether the skills of team members affect the communication/performance relationship in such way, we simulated team performance where all team members were generated with lower levels of expertise than in the original model in our second post-hoc analysis. As can be seen in Fig. 4, the sporadic performance for low frequencies does not exist when skills and abilities of all team members are minimal. These results demonstrate that highly competent team members can compensate for low levels of communication. In other words, a team comprised of highly qualied members may require less overall communication. This nding underscores the importance of team member knowledge, skills and abilities on teams capacity to achieve high performance. In our nal post-hoc analysis, we explored how effectively the teams exchanged information (Fig. 7). In our computational model, team members can communicate more information than required, so we tested if higher performance occurs when more information is exchanged. Contradicting one of the main assumptions regarding team communication (e.g., Allen, 1977; Katz and Tushman, 1981), our results show that more content is not

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

101

always benecial. Our ndings suggest that instead of exchanging every available piece of information, performance improves when teams focus on exchanging only the requisite pieces of information. We can see that very high ratios, which represent too much information being transferred, result in extremely low performance. When too much information is transmitted, individuals must sift through everything in order to nd the relevant pieces, thereby using their limited time to process unnecessary information. In practice, team leaders should monitor not only the time team members are engaged in communication, but also the quantity of transferred information. Finding the proper balance is not easy, but is worth the effort. Our results provide proof that there is an optimal level of information exchange. In sum, to achieve the best performance not all of the available information needs to be shared, only the requisite units. Like all research, the present study has some limitations that need to be considered. We begin with two general limitations of computational modeling that we could not avoid. First, although computational models are not simple, they are a simplied representation of human-to-human interactions. In realistic cases, the phenomenon under investigation is too complicated to be adequately modeled and simplication is required (Zeigler, 1976). Such simplications produce valid models, but decrease the generalizability of the results obtained. Our computational model, therefore, is valid only for cross-functional teams with similar project structures. A second limitation of the model stems from a general weakness of computational models: the results we obtained are not based on observed behavior, but on inputs of a mathematical model. We put signicant effort into formalizing the relationships according to the existing theory and generating representative parameters, but the results we obtained should be considered as guidelines. Two other limitations stem from the rules we used while developing the computational model. First, we assumed that the skills and communication potential of the team members were constant throughout the project. These rules helped us to clearly depict the relationship between communication and performance, but prevented us from a full analysis of CFT interaction. Second, we intentionally did not include product quality as a performance measure, because CFT are assembled, primarily, to decrease development time (Cardinal and Lei, 2000). The relative obstacles in representing mathematically the quality of the nished product also played a role in our decision, because it would be challenging to mathematically represent quality with acceptable accuracy. Future research can extend the present computational model in two possible ways. The rst type of extension is to perform additional post-hoc analyses using the current model. In this work, our main goals were to establish the relationship between communication and performance and to prove that the amount of information communicated and processed is the main cause of the curvilinearity. As information exchange and processing are the main focus of this work, we excluded some potentially insightful, but unrelated, post-hoc analyses. For example, the role of functional team composition and the boundary spanning of team members can be examined. In this model, we assumed that the organization would have enough capacity to always provide the required number of team members. Without changing the model, it is possible to impose a limit on the team members available and to explore whether the curvilinearity between communication and performance will be maintained. A second feature included in the model is the possibility for team members to seek outside information when needed. Evidence suggests that teams who seek information through

102

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

boundary spanning achieve higher team performance (Ancona and Caldwell, 1992). Future research could explore how boundary spanning impacts the communicationperformance relationship in a number of ways. The second type of extension is through renements of the theoretical model. Our computational model can be further developed by incorporating additional aspects of team processes to it. We propose three possible extensions. Future models could include team member learning related both to communication potential and skills. Learning may alter the performance curves slightly, because team members will be able to transfer more information via fewer messages if they become more skilled communicators. Under these conditions, team members may reach their overload potential more quickly, but if they are more skillful, with respect to their domain, they can process the information faster. Another possible extension is the creation of an explicit mechanism through which team members can control the level of information they are transferring. In the current model, the members are required to exchange information until they satisfy the information content requirements. They cannot stop this exchange when they start falling behind schedule. If they stop the communication process before fullling the requirements, a lower quality outcome may be obtained, but the on-schedule performance should be better. Inclusion of a set of rules allowing tradeoffs between quality and time will give us insights into how to improve control of the information exchange process. Finally, the inclusion of other performance measures, such as attaining technical objectives and product quality standards, should be examined. In conclusion, we present a computational model that explores the relationship between communication and on-schedule performance. Our results contribute to the extant literature on project team communication in a number of ways. First, we conrm the curvilinear relationship between communication and performance. Too much, as well as too little, communication causes low performance for both synchronous and asynchronous communication. Second, we provide a theoretical explanation for the curvilinear relationship by showing that the quantity of information exchanged reaches a point of diminishing returns, contrary to earlier research. Third, we show that communication frequency is a viable approximation for measurement of communication activities of the team, because it behaves similarly to communication duration. Fourth, our computational model augments our understanding of CFT communication. In particular, we explicitly include information processing time, in addition to the time required to conduct the communication exchange. This expands on previous work in this area and provides us with a deeper understanding of the information transfer process. Finally, the model can serve as a foundation for future computational models that explore other team processes and performance measures. Our results also have implications for practitioners. First, the identication of the optimum levels of synchronous and asynchronous communication, associated with effective information exchange among team members, provides guidance for team leaders attempting to achieve top performance. Second, training programs can be developed based on our results. Potential team members need to understand that not all information they possess has to be communicated to the team, but only the requisite units. Lastly, we provide insights into how the communication processes of teams should be managed. Our results underscore that more communication is not better and that quantity, as well as quality, of communication should be monitored.

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

103

Appendix A Appendix A provides details regarding variable generation and decision rules. We begin with a brief overview of the theoretical background on parameter distribution and parameter scales. The remainder of Appendix A is devoted to a complete, phase-by-phase list of the variables used in the model. For each phase of the model (Fig. 1), we describe the purpose of the phase and the variables included. Next, we give the theoretical rationale for each variable and the generation rules. We conclude each section with a description of the programming sequence. If further detail is required, the simulation code is available from the corresponding author. Computational modeling of social processes requires that some parameters in the model be determined by random variables. If such random parameters are used, they can be generated from either the normal or the uniform distribution (Taber and Timpone, 1996). When the underlying distribution is unknown, as it is for our parameters, the uniform distribution should be used because it assigns equal probability to all possible outcomes (Whicker and Sigelman, 1991). The ranges from which the random numbers will be drawn are inductively determined and intended to be reasonable (Koput, 1997). As we are not concerned with absolute numerical values, but with the overall behavioral pattern (Dutta, 2001), we standardized the majority of variables used in the model on an integer scale from 0 to 5, where 0 indicates low and 5 indicates high. For the variables that are not measured on a 0 to 5 scale, we provide the rationale behind our decision. To ensure that the results are not an artifact of a particular combination of initial values and parameter settings, the variables are randomly varied over the ranges described from simulation to simulation (Koput, 1997). A.1. The project The project phase of the model is used to transform the project parameters into the specic information content requirements and a project schedule. The variables included in this phase of the model are: project parameters (cost and technical objectives), organizational capability, information content requirements, information content requirements for each sub-phase, information content requirements for each functional area, project structure, project complexity, project schedule per sub-phase, and overall project duration. A.2. Project parameters The project is the assignment for which the CFT is created and about which the team members must communicate. Initially the project is represented as a vector with the following parameters (Srinivasan et al., 1997): Project = (cost, technical objectives). The cost is the rst parameter to be generated from an discrete uniform distribution. It ranges between 0 (low cost) and 25 (high cost). Our goal was to generate a large variety of projects in order to avoid results that stem from a particular combination of parameters. Thus, instead of generating the project cost on a scale of 05, we chose a larger scale

104

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

(025), with the assumption that it will provide a wider range of projects for the simulation runs. As the initial two parameters are highly interrelated (Pinto and Kharbanda, 1995), they should be correlated. Because we must use the uniform distribution, we are not able to generate correlates of the cost variable. Therefore, to obtain interrelated values, we use cost to establish the ranges in which technical objectives will be generated. In this way, the resulting variable depends on cost. Technical objectives are calculated as follows: to a discrete uniform number from one to the value of cost is added the value of cost. For example, if the generated value of cost is 15, technical objectives equals 15 + a random number between 1 and 15. If the next number generated is 8, technical objectives will be set at 23. Technical objectives are used to determine the information content required to complete the project. A.3. Organizational capability Organizational capability represents the knowledge and resources an organization possesses that impact the amount of work required to complete any project (Bowen et al., 1994). High organizational capability indicates that the organization members assigned to the team possess high levels of tacit knowledge about the project and are able to proceed faster and easier while performing the actual work. The variable is generated as a uniform random number between 0 and 5, with 0 representing minimal organizational capability relating to the project and 5 representing maximal organizational capability. We use organizational capability to help determine information content requirements. A.4. Information content requirements The amount of information the simulated team members need to communicate and process in order to meet the projects objectives is given by the information content requirements. The information content requirements are determined from the technical objectives and the organizational capabilities. One of the decisions a computational modeler has to make is how to achieve the right level of detail (Zeigler, 1976). If there is too much detail, a lot of time and effort will be lost during the simulation. If there is not enough detail, the model will not be a reliable representation of a true phenomenon. Thus, we developed the scheme outlined here to achieve a reasonable level of detail for this work. We apply the following rule: the range for technical objectives is divided into ve mutually exclusive intervals. The range of values is 050 because the maximum value technical objectives can take is 50, (the maximum value of cost is 25 that is added to a random number between 0 and 25). The intervals, therefore, are: 010, 1120, 2130, 3140, 4150. Organizational capability is divided into two mutually exclusive intervals: 02, representing low capability and 35, representing high capability. The cross-product of these intervals gives us 10 mutually exclusive intervals (e.g. technical objectives 110 and organizational capability 02; technical objectives 110 and organizational capability 35, etc.). All intervals are given in Table A.1. To each one of these intervals is assigned a range of values for information content requirements per project. These ranges are also mutually exclusive. Each range is 3500 units, as the maximum value for information content is set to 35,000 (this number was chosen, as it is close to the maximum integer that a standard C++ compiler can generate

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114 Table A.1 Information content requirements and project complexity Organization capabilities 2 Technical objectives [0,10] Technical objectives [11,20] Technical objectives [21,30] Technical objectives [31,40] Technical objectives [41,50] Information content [1000, 3500] Complexity = 0 Information content [7000, 10500] Complexity = 2 Information content [14000, 17500] Complexity = 4 Information content [21000, 24500] Complexity = 6 Information content [28000, 31500] Complexity = 8 Organization capabilities 3

105

Information content [3500, 7000] Complexity = 1 Information content [10500, 14000] Complexity = 3 Information content [17500, 21000] Complexity = 5 Information content [24500, 28000] Complexity = 7 Information content [31500, 35000] Complexity = 9

and provides an acceptable level of detail for the information requirements per project). Depending on the technical objectives and organizational capabilities, the information content is generated in the corresponding interval. For example, if technical objectives are 23 and organizational capability is 4, a random number between 17,500 and 21,000 represents information content requirements of the project. Information content requirements are subsequently used to determine the information content per sub-phase and the schedules for the sub-phases. A.5. Information content per sub-phase The information content per sub-phase represents the portion of the information content requirements that needs to be communicated and processed during a given sub-phase. Thus, the information content requirements for the project must be divided into information content per sub-phase. When developing the model we set the information content per sub-phase to vary from sub-phase to sub-phase (i.e. all seventeen sub-phases require different amounts of information to be communicated and processed) to create a realistic representation of the events in a NPD project. We use a binary search procedure to emulate the negotiation process that occurs among team members as they establish the requirements for each phase of the project. The binary search procedure ensures that each sub-phase will have different requirements and the sum of these requirements will not exceed the total information content requirements. We proceed in the following manner. First, the total information content requirements are divided by 17 (the number of sub-phases), so an average requirement per project is obtained. Next, the specic requirements for each sub-phase are negotiated. The process begins when eleven (one for each functional area represented on the team) random estimations between one and the average requirement are generated. The highest and the lowest of these estimations determine the new range in which the next random estimations of the information requirements per sub-phase will be generated. Repeating this procedure gradually decreases the range of generation until only one discrete number can be generated. This number is the information content requirements for the given sub-phase. The binary search is repeated for all 17 sub-phases.

106

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

The information content requirements per sub-phase represent the number of information units that must be exchanged during a given sub-phase, regardless of functional area. In order to identify where the information can be acquired, the information content requirements per sub-phase are further divided into the requirements for each functional area. The requirements for each functional area are allocated based on estimates of cross-functional participation determined by Dooley et al. (2000). They performed a study using survey questions and constructed an event history le to determine the role each functional area is expected to play in a cross-functional team. Their results show that the dominant team members contribute between 30 and 60% of the work. Thus, in our model, the dominant functional area (specied in Table 1) is randomly assigned 3060% of the information content requirements. The remaining information content requirements are distributed among the participating functional areas, with each area responsible for a randomly assigned 1020%. These percentages represent a small, but reasonable, number of information units. The algorithm allows for overlap (the percentages can sum to over 100). This overlap ensures that no information content is lost because of the binary search and the allocation process. The procedure just described establishes the number of information units required for each sub-phase by each functional area. The next step is to assign each team member this identical set of information content requirements. These requirements represent the information units that the team member must receive and process from each functional area (except his/her own) before the team can proceed to the next sub-phase. The information is received from his/her team members, or an outside source. Team members can only provide information regarding their functional area, and the amount of information they can provide is limited by the amount of time they have available to work on the project (the establishment of which is described in a subsequent section). Creating separate requirements for each team member in this manner allows us to ensure that all team members participate in the information exchange process. A.6. Project complexity We use project complexity to describe the level of innovativeness and creativity required for the project. We assign one value of project complexity to each of the 10 intervals in Table A.1, thus generating the variable as a uniform discrete number from 0 to 9. This way projects with high information content will have higher complexity. Project complexity is used to determine the time required for the project, as we assume that highly complex projects will require more time to complete than projects with low complexity. A.7. Project schedule The schedule for each sub-phase is determined based on the information content requirements and the project complexity. The information content per project is divided into seven intervals, each with an increment of 5000 (35,000 is the maximum possible information content requirement). Project complexity is divided into two intervals (04 as low and 59 as high). The cross-product of these intervals results in 14 intervals, as shown in Table A.2. To each of these intervals we assigned a range (see Table A.2) with which a random number is generated to represent the timeframe per sub-phase. We assume that the timeframe is in

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114 Table A.2 Project schedule Project complexity 4 Information content [1000, 5000] Information content [5000, 10000] Information content [10000, 15000] Information content [15000, 20000] Information content [20000, 25000] Information content [25000, 30000] Information content [30000, 35000] Time per sub-phase = 1 Time per sub-phase [1,2] Time per sub-phase [2,3] Time per sub-phase [2,4] Time per sub-phase [3,4] Time per sub-phase [3,5] Time per sub-phase [4,5]

107

Project complexity 5 Time per sub-phase [1,2] Time per sub-phase [1,3] Time per sub-phase [2,5] Time per sub-phase [3,4] Time per sub-phase [3,5] Time per sub-phase [3,6] Time per sub-phase [4,6]

whole weeks. For example, if project complexity is 6, and information content per project is 18,500, timeframe per sub-phase will be randomly generated as 3 or 4 weeks. The sum of the resulting timeframes is the total time for the project. The total time for the project is used to calculate the on-schedule performance at the end of the simulation. A.8. Programming sequence The simulation proceeds as follows. First, the project parameters are generated. Second, organizational capabilities are initialized. Both project parameters and organization capabilities are used to determine the information content requirements. The information content requirements and the project structure (see Table 1) are used to determine the information content requirements for each sub-phase. Next, the information content requirements for each sub-phase are divided into information content requirements for each functional area. Finally, the project schedule is established. The project schedule is created in the following manner. First, project complexity is determined from project parameters and organization capabilities. Next, the information content requirements, project complexity and project structure are used to determine the project timeframes per sub-phase. These timeframes then are summed to determine the expected overall project duration. A.9. The team The team is the entity responsible for completing the project through collaboration. The team consists of the team members determined by the information content requirements. In addition to the information content requirements previously discussed, each team member is assigned a functional area of expertise, time requirements for each sub-phase and personal characteristics. A.10. Team composition Team composition describes the number of team members and the functional area each team member represents. Each team member in this model can belong to exactly one functional area. In Table 1, we identify eleven functional areas considered to be important for effective collaboration on NPD projects (Brown and and Eisenhardt, 1995; Sethi

108

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

et al., 2001), distinguishing between the dominant and participating functional areas for each sub-phase (Jones, 1997). First, one team member from each area is initialized. If, after the information content requirements are allocated, more team members are needed, they are initialized in the requisite area. The procedure for this allocation is discussed in Section A.11. A.11. Time per team member Team members have work-hours assigned for the sub-phases of the project in which they participate. They can participate in the communication and information processing activities of the team only during their work-hours. The work-hours are determined by the information content requirements per sub-phase and the project complexity. For complex projects (complexity > 2), the work-hours per unit of information required are randomly set between 4 and 6, and for less complex projects (complexity 2) between 1 and 3. Thus, to calculate the work-hours, we multiply the work-hours by the number of information units. For example, if a team member has an information content requirement of 50 and project complexity of 5, the time will be a random number (4, 5 or 6) multiplied by 50. If the resulting work-hours are more than 40 h per person per simulated week, we generate additional team members in the affected functional areas by dividing the total work-hours for a given week by 40 h per week. Any fractional values represent part-time worker requirements. If, for example, there is a 60 h requirement for a given functional area, there will be two members from this area: one will work 40 h and the second one 20 h. A.12. Team member characteristics The next four parameters represent the personal knowledge, skills and abilities of the team members. The members must possess several personal characteristics that have been identied to be important for effective team performance. In this research, we focus on cohesion, functional expertise, communication potential and information processing ability (Gladstein, 1984). Cohesion assesses whether team members feel they make a contribution to the team, because employees need to feel proactive towards their work situations so that they can exhibit their skills and abilities (Dubinsky et al., 1986, p. 196). Cohesion is set initially to 0.01, because we assume all team members are initially strangers. As they interact, however, their cohesiveness increases linearly according to the expression: Cohesiont = cohesiont1 + 0.3personal communication frequency In setting cohesion to increase with the level of personal communication, we modeled the way a more cohesive team will emerge. The rationale for this decision stems from the results of Stewart and Barricks (2000) study. They reported that teams with high communication frequencies are invariably more cohesive than teams with low communication frequencies. As one of our reviewers pointed out, however, this may not always be the case and conict can emerge that will prevent a team from being cohesive. To explore this alternative, we ran 2000 simulations in which cohesion increased, stayed the same and decreased with

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

109

equal probability. The results indicate that the maximum achieved on-schedule performance is signicantly lower than in simulations without conict, but the curvilinear form of the relationship between communication and performance is preserved. We, therefore, maintain the rationale that the team member cohesion increases with communication for our purposes in this research, but acknowledge that this area may be a fruitful domain for future research. Functional expertise represents the skills that team members bring to the CFT. Evidence has shown that skills possessed by team members can affect team performance (Rulke and Galaskiewicz, 2000). We randomly generated each team members functional expertise level on a scale of 05, with 0 meaning low expertise and 5 meaning high expertise. Functional expertise is held constant through each simulation run, as we assume that members do not learn throughout the project. Communication potential refers to the ability of a team member to effectively exchange information. The ability of each team member to adequately communicate with others is important, especially in a CFT domain (Safoutin and Thurston, 1993). Similarly to skills, we initialized team members communication potentials on a scale of 05 and held them constant. Information processing time is the amount of time a team member will need to adequately process information. Information processing refers to the ability of a person to code, or classify the incoming information into their already possessed knowledge scheme (Boisot, 1995). A team members coding abilities can be overwhelmed easily by complex, new information and thus, classication of the incoming new information can take a signicant amount of time. Hicks (1952) demonstrated that performing a task with several choices, the time decisions require increases as the complexity of incoming information increases. He constructed the following equation representing the time necessary to process incoming information: Information processing time = k log (N) where k is a scaling constant based on the skills of the team member and N the distinct number of information units a team member must process. In our model, the constant k is set to 1.5 for a team member with high skills and to 3 for a team member with low skills. These values for k were derived based Hicks results. He established this relationship with an experiment where the unit of information was a Morse code. Depending on the skill of the subjects, he obtained values for k of 0.5 and 0.9. As we assumed that the project in our model is more complex than deciphering Morse code, we used values for k approximately three times higher than Hicks. A.13. Programming sequence The team is generated in the following sequence. First, one team member from each functional area is generated. Next, work-hour requirements for each area are calculated from the information content requirements. Based on the work-hour requirements, additional team members are initialized (if necessary) and the time available for collaboration is assigned to each team member. Third, the personal characteristics of each team member are generated.

110

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

A.14. Team communication process The communication process represents the way in which the team members collaborate. Collaboration is achieved through message exchange. The variables used in this section are: message generation and media choice. A.15. Message generation Each time a team member receives a request for information, a message is generated. A request for information is generated when a team member determines that s/he has a lower number of information units than his/her information content requirements in the current sub-phase for a given functional area. When a deciency is established, the team member sends a request to the individual who possesses the needed information, i.e. a team member from the respective functional area. If no team member can provide the information due to a lack of availability, it is sought from an external source. As a response to the request for information, the initiating team member receives a message from the individual possessing the requisite information with the following characteristics: information content, complexity, relevance, ambiguity, and duration (Boisot, 1995). These characteristics depict how well the message transfers information. Information content is the amount of information a message contains. It is determined by the functional expertise and cohesion of the sender. Content is randomly set between 0 and 2 if functional expertise and/or cohesion are low (between 0 and 2) and between 3 and 5 if functional expertise and cohesion are high (between 3 and 5). The logic behind the numbers assigned is the following: when team members are highly competent and feel devoted to the team, they will give as much information as possible. If they lack knowledge and/or do not feel comfortable with the team, the maximum information will not be supplied. Complexity is an assessment of how dense the information included in the message is. It is based on the communication potential and the functional expertise of the sender. Complexity is randomly set between 0 and 2 if functional expertise and/or communication potential are low (between 0 and 2) and between 3 and 5 if functional expertise and communication potential are high (between 3 and 5). In this case, team members lacking knowledge and communication abilities will not be able to generate highly complex messages that are difcult to understand. Relevance is an indicator of the senders ability to provide the exact information requested. It is based on the functional expertise and cohesiveness of the sender. Relevance is randomly set between 0 and 2 if functional expertise and/or cohesiveness are low (between 0 and 2) and between 3 and 5 if functional expertise and cohesiveness are high (between 3 and 5). We assumed that team members who do not feel as if they make a contribution to the team and who do not possess enough knowledge in their respective area will be less likely to give information that is pertinent. Ambiguity refers to the clarity with which the message is sent. It is dependent upon the senders communication potential. If the senders potential is high, then the ambiguity is low (randomly generated between 0 and 2). If the senders potential is low, then the ambiguity is high (between 3 and 5).

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

111

Duration characterizes the time required to send and receive the message, so both sender and receiver are involved. Duration depends upon the complexity of the message. For messages with low complexity, the duration is randomly set from 15 to 45 min. The time required for highly complex messages ranges from 46 to 120 min. These values were derived based on the ndings of Kraut et al. (1990). Their results suggest that scheduled conversations have durations between 15 and 60 min. We extended the time range to 120 min, because when collecting their data, Kraut et al. (1990) collapsed all conversations that lasted longer than 60 min into the 60 min category to remove outliers. As it is not reasonable to assume that all meetings last <1 h, we allowed interactions in which highly complex information is transferred to last up to 120 min. A.16. Media selection Research shows that messages communicated on media that are inappropriate for the amount of the information sought to be transmitted may be misinterpreted by recipients or may be otherwise ineffective with regard to their intended purpose (Carlson and Zmud, 1999). In addition to engendering a risk of misinterpretation, medium-message mismatches can require communication participants to engage in compensating activities, like sending several emails to clarify message content, leading to possible communication inefciencies (Hollingshead et al., 1993). As a result, team members unable to select the most appropriate communication medium will be less effective in accomplishing their work. Thus, we controlled for media selection to be sure the appropriate medium was chosen. In our model, information is exchanged either synchronously (e.g., face-to-face interactions and telephone calls) or asynchronously (e.g., e-mail and written communications). The mode selected is dependant upon message ambiguity. If the ambiguity is high (between 3 and 5) synchronous communication will be used, otherwise, asynchronous media will be chosen. To simulate reality, we also included a small amount of time for weekly team meetings and personal communication, which occur synchronously. Weekly meetings are attended by all team members and last 1 h. Personal communications are determined via random number generation with duration of the messages from one minute to 30 min, based on the ndings of Kraut et al. (1990) that informal meetings have a median duration of 15 min. A.17. Programming sequence The program proceeds as follows. First, each team member checks whether the information s/he possesses is less than the information requirements for each functional area during the current sub-phase. If the possessed information is less than required, new information is requested from the appropriate functional representative or a source external to the team. Next, a message is generated by the recipient of the request. The message is received and processed by the team member initiating the communication, and its information content stored. This cycle repeats until the requirements for each functional area are fullled. Movement through the phases and sub-phases (see Table 1) is regulated by the fulllment of the information content requirements. The whole team can proceed to the next phase only when

112

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

all the information content requirements for the current phase are met. If they are not met, the team must continue working until they are, thereby lengthening the project duration. If, however, some information content requirements are outstanding at the end of a sub-phase, they can be carried over into the next sub-phase in an effort to keep the project on-schedule. For example, if a team has not completed the predevelopment process (a project phase), they cannot begin development. But, if the team has not fully completed idea generation (a project sub-phase), they can begin to assess the various ideas without jeopardizing the integrity of the project. References
Allen, T.J., 1977. Managing the Flow of Technology. MIT Press, Cambridge, MA. Ancona, D.G., Caldwell, D.F., 1992. Bridging the boundary: external activity and performance in organizational teams. Administrative Science Quarterly 37, 634665. Bendor, J., Moe, T., 1992. Bureaucracy and subgovernmentsa simulation model. In: Masuch M., Warglien, M. (Eds.), Articial Intelligence in Organization and Management Theory. Elsevier, North Holland, pp. 119143. Boisot, M., 1995. Information Space: A Framework for Learning in Organizations, Institutions, and Culture. Routlege, London. Bowen, H.K., Clark, K.B., Holloway, C.A., Wheelwright, S.C., 1994. The Perpetual Enterprise Machine. Oxford University Press, New York. Brown, S.L., Eisenhardt, K.M., 1995. Product development: past research, present ndings, and future directions. Academy of Management Review 20 (2), 343378. Cardinal, L.B., Lei, D., 2000. Structuring research and development teams in the technological conversion process. In: Beyerlein, M.M., Johnson, D.A., Beyerlein, S.T. (Eds.), Product Development Teams. JAI Press, Standford, CT, 31 pp. Carlson, J.R., Zmud, R.W., 1999. Channel expansion theory and the experiential nature of media richness perceptions. Academy of Management Journal 42 (2), 153170. Daft, R.L., Lengel, R.H., 1986. Organizational information requirements, media richness and structural design. Management Science 32 (5), 554571. Denison, D.R., Hart, S.L., Kahn, J.A., 1996. From chimneys to cross-functional teams: developing and validating a diagnostic model. Academy of Management Journal 39, 10051023. Dooley, K., Durfee, W., Shinde, M., Anderson, J., 2000. A river runs between us: legitimate roles and enacted practices in cross-functional product development teams. In: Beyerlein M.M., Johnson D.A., Beyerlein S.T. (Eds.), Product Development Teams. JAI Press, Stamford, CT, pp. 283302. Dougherty, D., 1992. Interpretive barriers to successful product innovation in large rms. Organization Science 3 (2), 179202. Dubinsky, A.J., Howell, R.D., Ingram, T.N., Bellenger, D.N., 1986. Salesforce socialization. Journal of Marketing 50, 192207. Dutta, A., 2001. Business planning for network services: a systems thinking approach. Information Systems Research 12 (3), 260283. Glance, N.S., Hogg, T., Huberman, B.A., 1997. Training and turnover in the evolution of organizations. Organization Science 8 (1), 8496. Gladstein, D.L., 1984. Groups in context: a model of task group effectiveness. Administrative Science Quarterly 20, 499517. Goodman, P.S., Ravlin, E.C., Argote, L., 1986. Current thinking about groups: setting the stage for new ideas. In: Goodman, P.S. (Ed.), Designing Effective Work Groups. Jossey-Bass, San Francisco, CA, 1 pp. Hanish, K.A., 2000. The impact of organizational interventions on behaviors: an examination of models of withdrawal. In: Ilgen, D.R., Hulin, C.L. (Eds.), Computational Modeling of Behavior in Organizations. American Psychological Organization, Washington, DC, pp. 3360. Harrison, J.R., Carroll, G., 1991. Keeping the faith: a model of cultural transmission in formal organizations. Administrative Science Quarterly 36 (4), 552583.

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

113

Hicks, E., 1952. On the rate of gain of information. The Quarterly Journal of Experimental Psychology 4, 1126. Hinsz, V.B., Tindale, R.S., Vollrath, D.A., 1997. The emerging conceptualization of groups as information processors. Psychological Bulletin 121 (1), 4364. Hollingshead, A.B., McGrath, J.E., OConnor, K.M., 1993. Group Task-Performance and Communication TechnologyA Longitudinal-Study of Computer-Mediated Versus Face-To-Face Work Groups. Small Group Research, 24 (3), 307333. Hulin, C.L., Ilgen, D.R., 2000. Introduction to computational modeling in organizations: the good that modeling does. In: Ilgen, D.R., Hulin, C.L. (Eds.), Computational Modeling of Behavior in Organizations. American Psychological Organization, Washington, DC, pp. 318. Hutchins, E., 1995. Cognition in the Wild. MIT Press, Cambridge, MA. Jassawalla, A.R., Sashittal, H.C., 2000. Strategies of effective new product team leaders. California Management Review 42, 3451. Jones, T., 1997. New Product Development: An Introduction to a Multifunctional Process. ButterworthHeinemann, Linacre House, Jordan Hill, Oxford. Kang, M., Waisel, L.B., Wallace, W.A., 1998. Team-Soar: a model for team decision making. In: Prietula, M.J., Carley, K.M., Gasser, L. (Eds.), Simulating Organizations: Computational Models of Institutions and Groups. MIT Press, Cambridge, MA, pp. 2346. Katz, R., Tushman, M.L., 1981. An investigation into the managerial roles and career paths of gatekeepers and project supervisors in a major R&D facility. R&D Management 11, 103112. Kerzner, H., 2001. Project Management: A Systems Approach to Planning, Scheduling, and Controlling. Wiley, New York. Koput, K.W., 1997. A chaotic model of innovative search: some answers, many questions. Organization Science 8, 528542. Kraut, R.E., Fish, R.S., Root, R.W., Chalfonte, B.L., 1990. Informal communication in organizations: form, function, and technology. In: Spacapan, S., Oskamp, S. (Eds.), Peoples Reactions to Technology in Factories, Ofces and Aerospace. Sage Publications, pp. 145194. Lant, T.K., Mezias, S.J., 1992. An organizational learning model of convergence and reorientation. Organization Science 3 (1), 4771. Lant, T.K., 1994. Computer simulations of organizations as experiential learning systems: implications for organization theory. In: Carley, K.M., Prietula, M.J. (Eds.), Computational Organization Theory. Lawrence Erlbaum Associates, Hillsdale, New Jersey, pp. 195217. Law, A.M., Kelton, W.D., 2000. Simulation Modeling and Analysis. McGraw-Hill. Levitt, R.E., Cohen, G.P., Kunz, J.C., Nass, C.I., Christiansen, T., Jin, Y., 1994. The Virtual Design Team: simulating how organization structure and information processing tools affect team performance. In: Carley, K.M., Prietula, M.J. (Eds.), Computational Organization Theory. Lawrence Erlbaum Associates, Hillsdale, New Jersey, pp. 119. Patrashkova, R., McComb, S.A., Green, S.G., Compton, W.D., 2003. Examining a curvilinear relationship between communication frequency and team performance in cross-functional project teams. IEEE Transactions on Engineering Management 50(3), 262269. Pinto, J.K., Kharbanda, O.P., 1995. Successful Project Managers. Van Nostrand Reinhold, New York. Rulke, D.L., Galaskiewicz, J., 2000. Distribution of knowledge, group network structure, and group performance. Management Science 46, 612625. Safoutin, M.J., Thurston, D.L., 1993. A communications-based technique for interdisciplinary design team management. IEEE Transactions on Engineering Management 40, 360372. Sethi, R., Smith, D.C., Park, W., 2001. Cross-functional product development teams, creativity, and the innovativeness of new consumer products. Journal of Marketing Research 38, 7385. Smith, K.G., Smith, K.A., Olian, J.D., Sims Jr., H.P., OBannon, D.P., Scully, J.A., 1994. Top management team demography and process: the role of social integration and communication. Administrative Science Quarterly 39, 412438. Srinivasan, V., Lovejoy, W., Beach, D., 1997. Integrated product design for marketability and manufacturing. Journal of Marketing Research 34, 154163. Stewart, G.L., Barrick, M.R., 2000. Team structure and performance: assessing the mediating role of intrateam process and the moderating role of task type. Academy of Management Journal 43 (2), 135148.

114

R.R. Patrashkova, S.A. McComb / J. Eng. Technol. Manage. 21 (2004) 83114

Schroder, H.M., Driver, M.J., Streufert, S., 1967. Human Information Processing. Holt, Rinehart and Winston, New York. Streufert, S., Streufert, S.C., 1978. Behavior in the Complex Environment. Wiley, New York. Streufert, S., Swezey, R.W., 1986. Complexity, Managers, and Organizations. Academic Press, Orlando, FL. Taber, C.S., Timpone, R.J., 1996. Computational Modeling. Sage Publications, Thousand Oaks. Whicker, M.L., Sigelman, L., 1991. Computer Simulation Applications: An Introduction. Sage Publications, Thousand Oaks. Zeigler, B. P., 1976. Theory of Modelling and Simulation. Wiley, New York.

You might also like