Incorporating Theories of Group Dynamics in Group Decision Support System (GDSS) Design Lisa Troyer Department of Sociology, W140 Seashore Hall, University of Iowa, Iowa City, IA 52242-1401
[email protected]
Abstract GDSS design can enhance decision-making by managing information exchange patterns (e.g. ideas, facts, questions, evaluations) in groups. Groups cycle through stages involving identifiable information exchange patterns, and the optimal pattern depends on the group's stage. A GDSS that manages information exchange patterns can improve decision quality. Developing such a system requires formally modeling optimal information exchange patterns. I present exemplary models and research identifying key parameters for developing future models. The models indicate the benefits of large-scale collective decisionmaking. I discuss the principles, benefits, and challenges of developing such a GDSS that incorporates theories of group dynamics in systems design.
1. Introduction This paper builds on prior work [1, 2, 3] in which my colleague and I suggested that Group Decision Support System (GDSS) design has not recognized patterns of information exchange that affect the quality of collective decisions. We focused on groups confronted with illstructured decision tasks (i.e., decision problems for which there are no known or clearly superior solutions, and for which methods of evaluating alternative solutions are not well established). We proposed and tested a model of collective decision-making based on a group dynamics theory of information exchange. Our research showed that optimal outcomes for ill-structured decisions are achieved when members exchange ideas and evaluations in particular proportions. We suggested that GDSS design could incorporate user-based message categorization or text-parsing and language recognition routines to manage information flows and approximate these optimal proportions of evaluation and ideation. We proposed that such a system would facilitate higher quality decision outcomes in groups. Development and tests of such a system suggest that this is a feasible and beneficial direction for research on group decision-making [4].
In this paper, I further elaborate the importance of integrating theories of group dynamics in GDSS design. To accomplish this, I draw on elaborations of information exchange theory [5], research on group development [6, 7], and research on the evolution of status hierarchies in groups [8]. I suggest that GDSS design can enhance the quality of group decision making by managing two dynamics in groups: (1) information exchange patterns, and (2) the emergence of social hierarchy.
2. Decision structure & group processes Researchers have long recognized that groups offer many advantages over individuals in decision-making [9]. The diversity, expertise, commitment, and synergy that are found in groups are cited as important benefits [10]. These benefits offer particular advantages, when the decision task is less structured [11, 12, 13, 14]. Decision structuredness refers to whether solutions exist (or can be calculated in advance), methods for evaluating the relative quality of solutions are well established, and stakeholder preferences related to outcomes of the decision are known. If decisions are well structured, group decisionmaking offers few advantages and incurs high costs (e.g., wasteful investment of human resources). Increasingly, well-structured decisions are being automated. When decisions are ill-structured, however, the range and mixture of perspectives and expertise that tend to be represented in groups contributes substantially to decision quality. Ill-structured decisions are typically decisions made under uncertainty, such as decisions regarding concepts or inventions. These types of decisions benefit from diverse perspectives and expertise. Researchers have found that innovative ideation is increased when the range and number of solutions offered in a group is increased, and when unconventional combinations of common solutions are generated. Innovative ideation, in turn, is the basis for superior solutions to ill-structured decision tasks. The more diverse the actors proffering solutions are in the decision-making group, the broader the range of solutions that will be generated, and the more likely it is that synergistic combinations of solutions will arise. Generally, diversity in perspective and expertise are more
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
likely to be maximized in a group (and in larger groups), rather than by an individual. Thus groups tend to outperform individuals on ill-structured decision tasks. While groups offer many advantages over individuals when it comes to ill-structured decision-making, the picture is complex. Groups introduce challenges that threaten decision quality. These challenges fall into two categories: (1) information management, and (2) social losses. With respect to the former, bringing diverse individuals together to share information (which is how decisions are made) poses problems related to the organization, storage, recall, and synthesis of information that are not generally evident when individuals make decisions alone [15]. With respect to the latter, group decision-making is inherently a social process, and as such, introduces dynamics (e.g., social loafing, dominance processes), which can severely undermine the effectiveness of the group by biasing the decision. Efforts to address issues related to information management in recent decades have focused on the development of computer systems to support the decision process. These systems, Group Decision Support Systems (GDSSs), vary in features and implementation, but tend to rely on a server-client networks, with information stored and processed on a centralized server [16, 17]. The models for information management characterizing the systems, however, are increasing in complexity. In the face of this growing complexity, the speed at which the systems are able to manage information is being compromised. Time is an important and sensitive variable when it comes to group dynamics, because differences in milliseconds can affect individual cognitive processing and communication patterns. Yet, few researchers have explored the potential of alternative network models to improve the delivery of services under different GDSS designs. Distributive networks may offer a solution to the growing speed trap in information management that is coming to characterize GDSSs. While the rapid growth and deployment of GDSSs initially emerged as a result of their superior information management capabilities, researchers have increasingly attended to the potential of these systems for managing social processes in groups [4, 18, 19, 20]. Drawing on theories of group dynamics, researchers working on this facet of GDSS design have paid particular attention to collective decision-making as a communication process. A GDSS has the ability to track who is communicating to whom and the content of the communication. Researchers have long recognized biases in communication that undermine group performance. Biasing patterns are known as "process losses" and are captured by the Ringlemann Effect [21] documented in Figure 1.
Figure 1. Ringlemann Effect Productivity 1600 1400 1200 1000
Process Loss
800 600 400 200 0 0
2
4
6
8
10 12 14
Group Size
Potential Productivity Observed Productivity As this figure indicates, groups reach their maximum performance at about a size of about 10-11 members, although their observed performance is far below their hypothetical potential. Beyond this size, their performance declines and increasingly deviates from their potential. Researchers have cited several reasons for these performance decrements, or process losses. Some reasons represent practical concerns, like scheduling. Others involve group dynamics, including social loafing (a tendency for group members to reduce their own performance under the assumption that others will pick up the slack), groupthink (a tendency for group members to prematurely arrive at a consensus without exploring the liabilities of their decision), group development processes (maturation processes, which demand more resources as the group size increases), dominance processes (a tendency for some members to constrain the communication in the group). Most GDSS models address process loss by establishing recommended size ranges for group membership and norms for using and administering the systems, rather than building the process loss recognition and management directly into the systems. Such system level capability is, however, quite feasible. Furthermore, as I will demonstrate, increasing the size of the group is one way (although not the only way) of enhancing group decision-making. Considering group decision-making as a process of information exchange, my colleague and I have developed theory-driven formal models of social process management for group decisionmaking [1, 2, 5]. These models are specified in a manner that permits them to be incorporated into GDSS design,
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
leading to a "smart" GDSS. I turn now to a discussion of our theoretical strategy for developing a "smart" GDSS.
2.1 Information exchange theory & process losses Our theoretical framework emphasizes group decisionmaking as a process of information exchange. That is, in making decisions, group members pool different types of information (e.g., ideas, evaluations, facts, questions) to arrive at a solution. The information is continuously integrated and assessed as the group moves toward a decision. The information exchange involves both a task process and social process. Conceptualizing information exchange as a task process highlights the relative importance of different types of information. While decisions clearly demand a range of information types (e.g., facts, questions, evaluations, and ideas) researchers have found that ideas and negative evaluations play critical roles in the generation of quality [22]. Ideas hold obvious importance as candidate decision solutions (or solution components). Negative evaluations are important because they are the fundamental mechanism for discriminating among alternative candidate decision solutions. While positive evaluations can also contribute to sorting, negative evaluations are argued to be more important because they prevent groupthink (premature consensus and failure to consider shortcomings of a candidate solution). These critical information types (ideas and negative evaluations) are also the ones that are central to bias in social processes. In considering information exchange as a social process, we recognize that participants are not only interested in decision quality, but also in maintaining their status, or social position, in a group. Sociologists have long recognized the effects of status processes on individuals [23]. Status organization emerges in all collectives and tends to be based on differences in observable or known personal attributes (e.g., gender, ethnicity, age, organizational rank, education), whether or not these attributes are relevant to the decision task. Rewards (material and social) are allocated on the basis of status. Consequently, it is not surprising that members seek to preserve their status in a group. This begs the question, "How is status lost in groups?" My colleague and I have demonstrated that status loss arises through the receipt of negative evaluation in groups [3, 5]. Furthermore, certain types of information may have a greater propensity to elicit a negative evaluation than other types. In particular, ideas and negative evaluations are the two types of information that are most likely to elicit a negative evaluation directed to the source of the ideas and negative evaluation. Thus, if group members are motivated to avoid status-threatening negative evaluations, then we expect that they will undersend information types that are likely to elicit negative
evaluations (i.e., ideas and negative evaluations), relative to other information types (such as positive evaluations, facts, and questions). This is problematic for ill-structured decision-making, since innovative ideas increase with the number of ideas exchanged within groups. In an experimental study of individuals working on a decision task, we found that this was the case [20]. Groups that generated more ideas also generated more innovative ideas. Furthermore, we found a significant effect of negative evaluation on innovative ideation. Figure 2 illustrates this effect. Figure 2. Innovation & Negative Evaluation Idea Innovativeness 0.25
0.2
0.15
0.1
0.05
0 0
0.1 0.2 0.3 Ratio of Negative Evaluation to Ideas
0.4
This figure illustrates that innovative ideation is a quadratic function of the ratio of negative evaluations to ideas. On the basis of these insights, we can write a quality function for group decision-making in terms of the exchange of ideas and negative evaluations as follows: n
n
[
QG* = ∑ ∑ I i + I j − α ( I j − RN ij ) − α ( I i − RN ji ) i−1 j =1
2
2
]
(1)
Where n = number of group members; Ii = number of ideas sent by ith member; Ij = number of ideas sent by jth member; Nij = number of negative evaluations sent by ith member to jth member; Nji = number of negative evaluations sent by jth member to ith member; R = the ideal ratio of negative evaluations to ideas (0.10 < 1/R < 0.25).
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
In our research, we also found that status affects the messages sent in groups, as well as the costs of negative evaluations to their targets. More specifically, higher status actors tend to send more messages (including ideas and negative evaluations), and the cost of a negative evaluation increases as the status of its source increases. This result, corresponding to prospect theory [24] and supported in our empirical research, is interesting in its convex form. Individuals overvalue evaluations from higher vs. lower status actors. This suggests that if individuals change their reference point in assessing negative evaluations, then the expected costs of the evaluation would be substantially reduced, leading to a higher tolerance for negative evaluation (and hence, continued ideation). On the basis of these insights, we have shown mathematically that a status-equal group should generate higher quality decision solutions than a status heterogeneous group [5]. This hypothesis is also supported in our empirical research [5, 20]. Using our formal model of decision quality, a smart GDSS could be designed to analyze information exchange patterns in groups and assess whether the ratio of negative evaluation to ideation is within the optimal range. For full automation, language analysis routines are required. Such routines are rapidly emerging and being refined, as new algorithms for classifying and analyzing text are being developed and tested. Until adequately accurate routines are in place, however, users of the system could classify their input into relevant categories (e.g., ideas, facts, questions, positive evaluations, negative evaluations). The implications of eq.(1), however, are not entirely unproblematic. As noted above, innovation is only partly dependent on the number of ideas that a group generates. If idea volume was the only concern, then, to increase quality, based on eq. (1), we would simply increase the group size, while maintaining status equality. Increasing group size exacerbates the developmental challenges that plague groups, which I will elaborate later. It is partly because of these challenges that decision-making groups today are vastly undersized. Imposing status equality (particularly in the real-world context of organizations) is unrealistic. Organizations are inherently structured along dimensions like organizational authority, occupation, education, wage, tenure. Each of these dimensions is a source of status [25]. Moreover, although these dimensions are the source of status hierarchies, they also are sources of diversity. As I noted earlier, one of the benefits of group decision-making arises from the diversity of group members' backgrounds; it is neither desirable nor practical to consider engineering decisionmaking groups so that they are homogenous. Group heterogeneity is desirable in ill-structured decisionmaking, even though the status hierarchies it engenders are not. We can model the heterogeneity of a group as:
k
m
∑1− ∑ p
2 c
h=
a =1
c =1
(2)
k
Where k = number of attributes present in group; m = number of categories of attribute a; pc = proportion membership in category c. If heterogeneity contributes to quality solutions for illstructured decision tasks, it should be incorporated in the decision quality function. Through preliminary analyses of experiments involving groups with varying degrees of heterogeneity, I found that an exponential contribution seemed to generate the best fit. Thus, I modified eq. (1) as: n
n
[
QG* = ∑ ∑ I i + I j − α ( I j − RN ij ) − α ( I i − RN ji ) i−1 j =1
2
]
2 h +1
(3)
Where n = number of group members; Ii = number of ideas sent by ith member; Ij = number of ideas sent by jth member; Nij = number of negative evaluations sent by ith member to jth member; Nji = number of negative evaluations sent by jth member to ith member; R = the ideal ratio of negative evaluations to ideas (0.10 < 1/R < 0.25); h = heterogeneity index for the group, given by eq. (2). Eq. (3) accounts for the fact that heterogeneous groups generate more innovative decisions than homogeneous groups. Also, innovativeness arises earlier in heterogeneous than homogeneous groups. Note, however, that I have introduced a paradox. On the one hand, heterogeneity benefits groups charged with ill-structured decision tasks. On the other hand, heterogeneity represents a source of status, which imposes risks on members leading to biases (i.e., under-sending of ideas and negative evaluations, especially by lower status members). Unlike face-to-face environments, however, a GDSS environment may allow the innovation-enhancing benefits of heterogeneity to enter the group decisionmaking process, while managing the biasing status costs of heterogeneity. The strategy on which most researchers have relied to accomplish this is to require anonymity during group decision-making. Experimental studies demonstrate that there is less conflict and more ideation in groups whose members interact anonymously over a GDSS, compared to groups whose members are identified [26, 27]. Despite these encouraging results, a non-trivial concern has surfaced: Anonymous groups take up to four times longer to generate the same number of ideas as
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
groups whose members are identifiable. This undesirable outcome cannot be ignored when it comes to extending theoretical models of GDSS design and implementation to real-world organizations. Time is money. One reason why anonymity impedes group efficiency is that anonymity makes it difficult for groups to organize. Groups, like individuals, mature by progressing through developmental stages. Maturity, in turn, corresponds to efficiency. Anonymity interferes with reaching maturity, in part, because it removes status markers in groups. In other words, although status generates biasing effects that undermine achieving optimal solutions by decisionmaking groups, it is also a key mechanism that facilitates the organization of the group. By understanding the developmental cycles that characterize groups, however, we are in a better position to understand when allowing status markers to be salient and when blocking them may be most effective in the group. I turn now to a discussion of developmental cycles in groups, the patterns associated with them, and how they affect group decision-making.
3. Developmental cycles in groups Researchers [6, 7] have documented the stages that typify groups. The stages are "forming" (identifying who is a group member and the positions in the group that members will occupy), "norming" (establishing behavior expectations), "storming" (resolving challenges to the established positions and expectations), and "performing" (focused work on the group's task). Early research suggested that the stages were sequential and unidirectional (i.e., each stage was completed before the next, with no reversions to earlier stages). It was argued that a group reached equilibrium only when it reached the final stage, and that equilibrium was critical to performance (i.e., the more quickly a group reached this final stage, the more rapidly it could complete its work). More recently, social scientists have challenged these assumptions. Researchers [28, 29] have demonstrated that groups in actual work settings cycle back to prior stages (e.g., changes in membership or redefinitions of the group's task catalyze forming, storming, and norming). Furthermore, under certain circumstances, equilibrium may undermine performance. For instance, the identification of who is a member of the group and the assignment of group members to particular positions (as occurs during forming) is likely to lead to routinization, and effectively determines the type and amount of participation allocated to each individual throughout the group's life [5, 20]. To the extent that the group's decision task is ill-structured (e.g., is not readily addressed by known solutions), this can result in a "garbage can" solution [30] with devastating effects. Garbage can solutions represent recycled solutions that reflect outcomes that key decision-makers have implemented in
the past. In this respect, they are not innovative and may be poorly suited to the current problem. They are adopted because they are familiar. The emergence of garbage can solutions often corresponds to the crystallization of robust status orders in groups. Higher status actors propose solutions with which they are most familiar and re-define the decision task in terms of known problems that the solutions fit. Lower status actors (who are managing status) are disinclined to negatively evaluate the solution, which is then rapidly accepted in the group. In contrast, challenges to group norms and positions (as tend to occur in the storming stage) can be important catalysts for innovation (i.e., leading new ideas and processes to emerge and resulting in new alternative decision solutions. A smart GDSS might incorporate routines that (1) identify a group's developmental stage, (2) assess whether the group would best be at a different stage, and if so, (3) manage the identification of members to move the group in or out of anonymous interaction, and (4) re-direct the flow of information in the group to optimize decision outcomes. Such routines are based on analysis of the information exchange patterns.
3.1 Group development & social hierarchy As noted above, the emergence and stabilization of social hierarchies in groups corresponds to forming and norming, stages of group development. Thus, research on the dynamics that characterize social hierarchies provides a foundation for identifying a group's developmental stage. My work with colleagues articulates how hierarchy develops in heterogeneous and homogenous groups [8, 20, 31]. As noted previously, in heterogeneous groups members are initially differentiated along social dimensions (e.g., gender, race) and task dimensions (e.g., occupation, skill). In heterogeneous groups hierarchy emerges rapidly and stabilizes quickly. In homogeneous groups, members are initially undifferentiated in terms of identifiable or known social or task dimensions. Although there is no initial basis for differentiation among members of homogenous groups, differentiation does occur as the result of early interactions between group members [32]. In these groups, hierarchy emerges more slowly compared to heterogeneous groups, but often, rapidly in absolute terms (e.g., several seconds, a few minutes). Stabilization, however, takes notably longer. This is because, in both types of groups, a stabilized hierarchy arises from the resolution of pairwise status contests between members. In heterogeneous groups, the contests are more quickly resolved (because contestants can rely on established cultural expectations related to differentiating social and task attributes that dictate who has the right to dominate and obligation to defer). In homogenous groups (in which scripts for allocating dominance rights are non-existent), these contests tend to be more extended.
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
This discussion reveals that status processes are sometimes beneficial to groups. Moreover, these benefits increase as the group size increases. By recognizing these and managing these benefits, it is possible to consider substantially increasing the size of decision-making groups, far beyond the scope of the 10- to 12-person size that has become normative as a result of studies that document the complexity of group development and the increase in process losses that occur beyond this size. According to our models of quality in group decisionmaking, if the information exchange patterns among group members can be managed to maintain optimal ratios of negative evaluations to ideas, then substantial increases in group size would offer an advantage. As I have noted in this section, increasing group size also requires managing group development, which is facilitated by status processes. The awkwardness and uncertainty of interaction in the absence of status markers engenders an extended and painful maturation period for the group, which undermines task focus and efficient decision-making. Consequently, these insights suggest that it may be beneficial to allow status markers to remain salient in early stages of interaction (or when status contests re-emerge) in order to move the group rapidly through these less task-focused stages.
3.2 Information exchange & status contests While the salience of status markers may be beneficial in early stages of group interaction, as I have already demonstrated, this salience may be problematic in the performing stage (i.e., when groups are in task-relevant stages of work). This begs the question, "How do we know when a group is at a particular stage of development?" Groups have resolved forming, norming, and storming stages and are at a performing stage when status contests become rare in the interaction of a group. Thus, recognizing status contests, prolonging them, limiting them, or re-initiating them represent strategies for engineering group processes that may enhance collective decision-making. (In this paper, I focus on recognizing and limiting status contests. The principles I offer, however, can be generalized to cases when it may be beneficial to re-initiate or prolong them; e.g., when a group prematurely exits a forming or storming stage.) Returning to the conceptualization of group decisionmaking as information exchange between members, we can identify patterns of information exchange that may mark status contests. I found an interesting pattern in a secondary analysis of information exchange in experimental groups: Rates exchanges of negative evaluations are significantly higher earlier in group interaction than later in group interaction in both homogeneous and heterogeneous groups, but even more so in homogeneous groups. Furthermore, overall
rates of negative evaluation are higher in homogeneous than heterogeneous groups. These observations suggest that rates of negative evaluations may be a key indicator of status contests early in a group's career. Additionally, to the extent that normative patterns of information exchange are established early in a group's career, it appears that the status contests marked by negative evaluation establish a norm of negative evaluation in groups. Given the insights above on the deleterious effects of excessive negative evaluation on innovative ideation, this suggests that homogeneity early in a group's career is problematic (because it generates robust hierarchy and increases negative evaluation). On the basis of these insights, it appears that dense clusters of negative evaluation are markers of early stages of interaction (i.e., forming and norming stages). They may also be characteristic of storming (i.e., the reemergence of status contests later in the group), although I have not empirically investigated this. As the clusters become less common, the group appears to move into a performing stage. Although I have only taken initial steps in investigating the correspondence between information exchange patterns and stages of group development, an interesting adjunct pattern seems to be emerging. In the information exchange of heterogeneous groups, I found that not only are there distinct patterns of negative evaluation. There are also distinct patterns of silence, which are not replicated in homogeneous groups. More specifically, early in the interaction, periods characterized by more dense negative evaluation exchange are nearly always followed by an uncharacteristic period of silence (e.g., five to eight seconds). An inspection of later periods of interaction in heterogeneous groups (when they appear to be task-focused in performing stages, as opposed to forming, norming, or storming stages) indicates that silences are relatively brief (e.g., one to three seconds). Researchers have paid very little fine-grained attention to temporal issues in studies of group interaction. Most attention related to time has been focused on grosser correlates of time, such as how long it takes a group to complete a task. My observations are consistent with anecdotal reports of how silence is experienced in group members. Silence is often experienced with distrust. In fact, group members often rapidly move to fill silences in interaction to avoid them and dissipate the anxiety they generate. Consequently, a higher tolerance for silence is indicative of high levels of trust and a high degree of confidence in the group's organization. A smart GDSS design is one that capitalizes on these patterns to manage group development in a facilitative manner. For instance, the GDSS design could allow identification of group members early in the interaction, tracking the density of negative evaluation clusters and silence. As the clusters taper off and extended silences emerge, the system could shift members to an anonymous
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
interaction mode. This shift would encourage ideation. If negative clusters begin to re-emerge (indicating the emergence of a storming phase, suggestive that the group needs further (re)organization), then the interaction mode could be shifted back to one that identifies members in their information exchanges. To summarize, it is critical to follow up on these initial findings. Further analyses of how negative evaluation patterns (vis-à-vis other types of information exchange) correspond to a group's developmental stage are central to the formulation of models that can be integrated with a smart GDSS design. These models will be enhanced if further analyses of the timing of events (including silence) are incorporated. Note, however, that a smart GDSS design like the one that I am proposing is not without challenges. In particular, the design and implementation of a smart GDSS is computationally intensive.
4. Challenges for GDSS design posed by incorporating theories of group dynamics While my past experience suggests that incorporating formal models of group process theories in GDSS design is feasible [4], it is clear that these models introduce layers of complexity that may challenge the performance of such systems. In particular, these models (and their implementation) require computational intensiveness that adds a burden to the more routine tasks of simply relaying data between members. A smart GDSS not only relays data; it must also analyze it and manage it. Additionally, as evident from the theories and models that I have presented, groups may be vastly undersized because of process losses and logistical issues (like scheduling, finding a location that can seat all the group members at once), which are exacerbated in face-to-face meetings. To the extent that a GDSS can effectively manage process losses and address logistical issues, larger groups would be desirable, since they would theoretically generate higher quality solutions. In fact, it may be useful to investigate a contingency model of group size in which group size becomes a function of the structuredness of the decision task. Early in this paper, I treated structuredness as a dichotomy (ill- vs. well-structured). Yet, in reality, structuredness varies along a continuum. At the lowest end of the continuum (i.e., completely unstructured decision tasks), extremely large-scale groups (in the order of thousands of participants) may be optimal. Because of the information management demands and process loss issues I have discussed, it has not been practical to even consider groups of such scale in traditional settings (e.g., face-to-face meetings or existing GDSSs). I have suggested strategies (involving the incorporation of formal models of group dynamics in GDSS design) for managing process losses. In addition, interaction over a
GDSS may make asynchronous meetings and/or meetings that take place in distributed locations feasible, thereby substantially reducing logistical problems related to scheduling and space. It is important to note, however, that increasing group size further increases the computational demands on the system. These demands are likely to lead to pauses in the system that members will inaccurately experience as silence. As I have suggested in the prior section, this may generate artificial process losses (for instance, by proliferating distrust in the group or otherwise biasing cognitive processes). A solution to these challenges may lie in moving the smart GDSS from a client-server model to a distributed network model. Two features suggest this solution. First, the computational tasks I have proposed that the formal models of group dynamics involve are inherently divisible. The calculations required could be divided into subsets, carried out by separate processors, and later integrated. Second, the natural flow of information exchange in groups is such that all participants are rarely simultaneously participating. This is especially likely in asynchronous meetings and for larger collectives. Thus, the idle processing power corresponding to inactive nodes could be put to use for the purpose of more efficiently carrying out the computational tasks of a smart GDSS.
5. Conclusions My purpose in this paper has been to discuss how theories of group dynamics can be integrated into GDSS design to take these systems to the next level. A smart GDSS is one that goes beyond relaying data between members. It analyzes the data in real-time and employs the analysis to inform the execution of routines that steer the group toward the information exchange patterns that optimize decision outcomes. This is a revolutionary direction in GDSS design. Common systems today attempt to engineer optimal group processes by establishing norms and rules for the individuals who use them. This social engineering, however, can be carried out by the system itself, through the active integration of formal and empirically substantiated theories of group dynamics in the system design. The proposal that group size should be considered in light of the structuredness of the decision is also novel. Researchers have long recognized the benefits of bringing multiple perspectives to bear on decisions, as the structuredness of the decision decreases. Yet, expanding groups to sizes beyond 10-12 members has not been deemed feasible because of logistical difficulties and the high risk of costly process losses that become exacerbated in large groups. As I have suggested, however, a GDSS developed and implemented on a foundation of theory and research in group dynamics may reduce these risks, making large-scale collective decision making beneficial.
0-7695-1926-1/03/$17.00 (C) 2003 IEEE
In summary, a smart GDSS design that recognizes group dynamics can substantially improve information management and the information exchange patterns that contribute to quality in group decision-making. I have outlined examples and principles for developing a smart GDSS. There are still many challenges and additional avenues of investigation (some of which I have identified here) that must be pursued to advance this research agenda. Nonetheless, this paper has demonstrated the general value of integrating theories of group dynamics with systems design.
References [1] S. D. Silver & L. Troyer. Managing information exchange to increase the quality of group decisions: Conceptual foundations for an event-sensitive GDSS. Proceedings of the 23rd Annual Meeting of the Western Decision Sciences Institute, 1994. [2] S. D. Silver & L. Troyer. Social risk in the process of collective decision making: Cognitive biases & the assessment of participation costs. Proceedings of the 26th Annual Meeting of the Decision Sciences Institute, 1995. [3] L. Troyer & S. D. Silver. Negative evaluation & innovativeness in group decision making: Insights for GDSS development from a social risk framework. Proceedings of the 27th Annual Meeting of the Decision Sciences Institute, 1996. [4] S. D. Silver, L. Troyer, & B. Marks. System for management & analysis of real-time teams (SMART), GDSS software, 1998. [5] S. D. Silver & L. Troyer. Judging the consequences of evaluation by others in status heterogeneous groups: Biases in the microlevel heuristics of group information exchange. Advances in Group Processes, Vol. 15, E. J. Lawler, J. Skvoretz, & J. Szmatka (Eds.). JAI, Greenwich, CT, 1998. [6] B. W. Tuckman. Developmental sequences in small groups. Psychological Bulletin, 63:384-99, 1965. [7] B. W. Tuckman & M.A. Jensen. Stages of small group development revisited. Group & Organization Studies, 2:41927, 1977. [8] R. K. Shelly & L. Troyer. Speech duration & dependencies in initially structured & unstructured task groups. Sociological Perspectives, 44:419-44, 2001. [9] K. R. MacCrimmon & R. N. Taylor. Decision-making & problem solving. Handbook of Industrial & Organizational Psychology, M. D. Dunnette (Ed.). Rand McNally, Chicago, IL, 1976. [10] S. A. Stumpf, D. E. Zand, & R. D. Freedman. Designing groups for judgmental decisions. Academy of Management Review, 4:589-600, 1979. [11] D. G. Ancona. Groups in organizations: Extending laboratory models. Annual Review of Personality & Social Psychology: Group & Intergroup Processes, C. Hendrick (Ed.). Sage, Beverly Hills, CA, 1987. [12] B. E. Collins & H. Guetzkow. A Social Psychology of Group Processes for Decision-Making. Wiley, NY, 1964. [13] H. D. Mintzberg, D. Raisinghani, & A. Theoret. The structure of unstructured decisions. Administrative Science Quarterly, 21:246-75, 1976. [14] A. Newell & H.A. Simon. Human Problem Solving. Prentice-Hall, Englewood Cliffs, NJ, 1972.
[15] M. J. Driver & S. Steufert. Integrative complexity, an approach to individuals & groups as information processing systems. Administrative Science Quarterly, 14:272-85, 1969. [16] G. W. Dickson & G. DeSanctis. Information Technology & the Future Enterprise: New Models for Managers. PrenticeHall, Upper Saddle River, NJ, 2001. [17] P. Gray, S. L. Alter, G. DeSanctis, G. W. Dickson, R. Johansen, K. L. Kraemer, L. Olfman, & D. R. Vogel. Group decision support systems. Information Systems & Decision Processes, E. A. Stohr, & B. R. Konsynski (Eds.). IEEE Computer Society Press, Los Alamitos, CA, 1992. [18] G. DeSanctis & J. Fulk. Shaping Organization Form: Communication, Connection, & Community, Sage, Newbury Park, CA, 1999. [19] J. F. Nunamaker, A. R. Dennis, J. S. Valacich, & D. R. Vogel. 1991. Information technology for negotiating groups: Generating options for mutual gain. Management Science, 37:1326-46, 1991. [20] S. D. Silver, B. P. Cohen, & L. Troyer. Effects of experimenter-inserted negative evaluations on idea generation & information exchange in computer-mediated groups. Proceedings of the 29th Annual Meeting of the Decision Sciences Institute, 1998. [21] M. Ringlemann. Research on animate sources of power: The work of man. Annales de l'Institute National Agronomique, 12:1-40, 1913. [22] H. A. Simon. The New Science of Management Decision. Prentice-Hall, Englewood Cliffs, NJ, 1977. [23] J. Berger, B. P. Cohen, & M. Zelditch. Status characteristics & social interaction. American Sociological Review, 37:241-55, 1972. [24] A. Tversky & D. Kahneman. Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk & Uncertainty, 5:297-323, 1992. [25] S. D. Silver, B. P. Cohen, & L. Troyer. Effects of member status on the exchange of information in team decision-making: When team building isn't enough. Advances in the Interdisciplinary Studies of Work Teams, Vol. 7, M. M. Beyerlein, D. A. Johnson, & S. T. Beyerlein (Eds.), Elsevier Science, New York, NY, 2000. [26] T. Connolly, L. M. Jessup, & J. S. Valacich. Effects of anonymity & evaluative tone on idea generation in computermediated groups. Management Science, 44:589-609, 1990. [27] M. S. Poole, M. Holmes, & G. DeSanctis. Conflict management in a computer-supported meeting environment. Management Science, 27:926-53, 1991. [28] C. J. Gersick. Time & transition in work teams: Toward a new model of group development. Academy of Management Journal, 31:9-41, 1988. [29] C. J. Gersick. Revolutionary change theories: A multi-level exploration of the punctuated equilibrium paradigm. Academy of Management Review, 16:10-36, 1991. [30] M. D. Cohen, J. G. March, & J. P. Olsen. A garbage can model of organizational choice. Administrative Science Quarterly, 17:1-25, 1972. [31] R. K. Shelly & L. Troyer. Emergence & completion of structure in initially structured & unstructured task groups. Social Psychology Quarterly, 64:318-22, 2001. [32] M. H. Fisek, J. Berger, & R. Z. Norman. Participation in heterogeneous & homogeneous groups: A theoretical integration. American Journal of Sociology, 97:114-42, 1991.
0-7695-1926-1/03/$17.00 (C) 2003 IEEE