Modeling Group Decisions
There are many occasions in which a group, rather than an individual, has to make a decision. A good example is the board of director's decision about a major capital purchase. In these circumstances, it maybe important to model the group's decision so that it can be easily communicated to others or so that the group can have more self insight. This chapter discusses how an analyst can model a group's decision.
A model of a group of decision makers could be thought of as an average of the models of various members of the group. If so, a model is built for each individual member and the parameters of the models are averaged to represent the group. But groups are not just a collection of individuals. There is a synergy of thought and ideas that emerges from successful groups. Mathematical aggregation of individual's models into a group model loses the real advantage of having the group -- the real reason for having a group in the first place: to arrive better decisions than the best member of the group. To effectively model a group's decision, one should create a model that reflects the group's consensus, which maybe very different than the average of each individual's input.
Decision analysts have for sometime focused on mathematical methods of aggregating the judgment of various experts or decision makers (Jacobs 1995, Meyer, Booker 1991). This chapter does not review the literature on mathematical aggregation of experts' opinions. Instead, this chapter focuses on behavioral methods of getting a group to come to a consensus around a model. The goal is to design a process, where group members can, after reasonable deliberation, agree around a specific model and its parameters.
The task of structuring a decision and estimating the parameters of the model requires a great deal of input from the group. The model building effort may be seen as an artificial process to most decision makers. The questions posed by the analyst may seem a contorted way of looking at the decision at hand. The analyst's frequent interruption to structure and assess various estimates may interfere with the group's deliberations. If a group's decision has to be modeled effectively, then a process needs to be found that constructs a model without interfering with the group's interactions. This chapter provides one such group process.
The chapter starts with a short history of commonly used approaches to structured group meetings. Research on group processes started in earnest in late sixties and early seventies. It has grown in recent years to studies of group decision support systems, where computers are used to facilitate meetings. Our review is selective as the literature is vast. After the short history, we present a new approach called Integrative Group Process (IGP). This approach borrows from many of the existing methods of improving group processes.
A Short History of Group Processes
To understand what works in group processes, we start with one of the simplest group processes, the Group Communication Strategy. This approach originates from a set of normative instructions proposed by Hall and Watson (1970). Before the meeting starts, members are instructed to:
No other process change is made. Group members talk as they wish about topics they choose, with no break in how and when various components of the meeting are accomplished. During these conversations, the analyst constructs a model of the group, with each member contributing at will and commenting on what is relevant. After the start of the meeting, group members may or may not follow the recommended rules for interaction. The expectation is that most will follow these simple instructions, if reminded to do so at start of the meeting. In the decade following Watson and Hall's work, a number of studies evaluated their recommendations. Early research showed that the Group Communication Strategy leads to high quality judgments without sacrificing the group's later acceptance of the judgments (Nemiroff, Pasmore, and Ford 1976). But the study was done with student subjects, in situations where status differences may not exist. When there are large status differences, group members weigh the opinions and suggestions of high status persons more heavily (Forsyth 1998, Pagliari, Grimshaw 2002). In these groups, the Group Communication Strategy may not do well. Furthermore, little is known about the success of normative instructions in situations where group members are in conflict or have substantial stakes in the final group judgment. A process was needed that could work in these circumstances.
Another approach to helping groups talk through their differences is the Nominal Group Process (Delbecq, Van de Ven and Gustafson, 1975). This is a generic name for face-to-face group techniques in which instructions are given to group members not to interact with each other except at specific steps in the process. The following are the steps in the process:
The Nominal Group Process produces a prioritized list of ideas as well as numerical estimates in a short time frame. For example, an analyst can use this process to solicit attributes for a value model and then repeat the process to solicit weights for the attributes. This approach remains popular despite the fact that it has been more than 30 years since its inception (Moon 1999, Carney, McIntosh, Worth 1996) and is one of the key approaches used to develop treatment algorithms and consensus panels ((Cruse, Winiarek, Marshburn, Clark, Djulbegovic 2002). Research on the Nominal Group Process is extensive. A recent review (Black, Murphy, Lamping et al. 1999) shows that in numerous circumstances the process produces better results than unstructured group interactions. The performance of Nominal Group Process may depend on task structure, the selection of participants, the presentation of the scientific information available to the group, the way group interaction is structured, and the method of synthesizing individual judgments. Given the various factors that affect the performance of Nominal Group Process, analysts should rely on what makes sense. In our view, three components of the Nominal group process explain when it will be successful in managing groups:
Despite widespread use of Nominal Group Process in consensus panels (Jones and Hunter, 1995), the process is not without serious problems. In tasks that require judging the worth of several alternatives, e.g. in developing a multi-attribute value model, this technique may produce judgments inferior to the judgment of the most knowledgeable group member (Holloman and Hendrick 1972; Nemiroff, Pasmore and Ford 1976). But by far, the most serious problem with the process is that participants feel awkward about restrictions in their interactions. After the group meeting they may feel that the process and not they led them to the conclusions they arrived at. Therefore, they may not be committed to the group's consensus. A group of experts reviewing the data on group processes recommended that when acceptance of the group's decision is crucial as to whether the model is put to use, less structured group processes produce more widely accepted group decisions (Rotondi, 1999).
Another technique widely used by consensus groups in health care settings is the Delphi process (Jones and Hunter 1995). Dalkey and Helmer (1963) designed this process as a non face-to-face procedure for aggregating group members' opinions. Group members answer three interrelated questionnaires, the analyst summarizes the responses from each survey and mails the synthesis back to the same or other groups (Ryan, Scott et al. 2001) for further comment. For example, the first questionnaire may ask the group members to describe the attributes in a value model, the second questionnaire may present the attributes and ask for revisions as well as weighting of the relative importance of the attributes and the third questionnaire may continue with the revisions of the model as well as present a set of scenarios and ask the respondents to rate them. In some applications, for example in forecasting technological changes based on insights of a large group of experts, the Delphi method has proven useful . Delphi is also useful in situations where conflict or status differences among group members are so strong as to be dysfunctional. The most insightful feature of the Delphi process is that a meeting of minds can occur without an actual meeting.
Studies show that face-to-face interaction is superior to Delphi's remote and private opinion gathering (Cho, Turoff, Hiltz, 2003, Woudenberg 1991). Two studies comparing Nominal Group Process to Delphi found the two techniques about equal (Seaver, 1977; Miner, 1976). In other studies, Delphi has been less accurate than Nominal Group Process. In one study, for example, Delphi's remote feedback was detrimental and reduced the accuracy of the group members' estimates (Gustafson et al. 1973).
Difficulty with existing group processes has led a number of investigators to design their own modeling process that can be used in group settings. These approaches often use computers to help model the decision maker in a group setting. In the 1990s, a large number of studies were published regarding what affects the performance of group decision support systems (Fjermestad and Hiltz 1997). The Social Judgment Analysis (Hammond et al., 1976) is a group process that uses computers to model each member as they actively interact with each other. It was designed to reduce the pressure for group members to comply with the group's mind set just to feel more accepted by the group. The following are the steps in the process:
Computer facilitated meetings, in general, and Social Judgment Analysis, in particular, may seem too much work for some meetings. But with growing use of computers, many meetings are occurring through computers anyway. Many decision makers are in different locations and must use the computer to collaborate. Naturally, in these settings Group Decision Support Systems can help improve the self insight of individual group members and eventually the quality of and speed of arriving at group consensus. The value of this technique is demonstrated in recent studies focused on helping clinicians understand their own judgments (Holzworth and Wills. 1999). Social Judgment Analysis works well because it focuses decision maker's attention on why they prefer an alternative as opposed to which alternative they prefer. It provides group feedback that helps decision makers focus their reasoning. Data show that people change their opinion to conform with group norms. While this behavior is healthy for keeping the peace in the group, it is counter productive if ideas are being judged based on the popularity as opposed to their merits. Rohrbaugh's 1979 study showed that when feedback focused on why an idea was preferred as opposed to which idea was more popular, the group's final judgment was more accurate. Rohrbaugh also showed that Social Judgment Analysis was more accurate than the Delphi method as well as the Nominal Group Process (Rohrbaugh 1981) .
Interest in use of computers in group processes has led to many innovations. Eden, Jone and Sims (1983) were one of the first to develop a process for modeling how a group arrives at its judgments through "Cognitive Mapping." This process starts with constructing two parallel statements of the problem. One showing the factors leading to the problem, the other showing the factors leading to a satisfactory solution. For example, the problem may be stated as "high labor costs," the solution may be stated as "lowering the labor cost." The causes of high labor cost and the factors leading to lowering labor costs are also organized. For example, a cause of the problem may be "shortage of qualified worker." A solution may be "more availability of qualified workers." Clearly, causes of the problem and factors leading to the solutions are related, usually a change of adjectives produces the other. Through linguistic manipulation of the statement of the problem, Cognitive Mapping hopes to stimulate new ideas.
Eden and colleagues, when they use Cognitive Mapping, often collect the group member's ideas about the problem and its solutions separately and then revise these ideas in a face-to-face meeting. Occasionally they quantify the influence of causes and effects through a Round Robin process, where group members write down their estimates and share them afterwards. They, then, simulate how changing one factor may affect the problem. These simulations may lead to new insights into the problem. Cognitive Mapping of groups of decision makers continues to be an active area of research (Vennix 2000).
Social Judgment Analysis and Cognitive Mapping Group Process were the start of many innovative methods of computer facilitated collaboration. McLeod (1992) summarized the group decision support literature and found that computer-facilitated group meetings increased decision quality, led to more equal participation by group members and more focus on task and less focus on social networking and support. At the same time, McLeod identified that computer facilitated group interaction decreased consensus and member satisfaction with the group meeting. The reduction in member satisfaction with group process might be due to the central role of a facilitator in controlling who speaks when in group processes (Austin, Liker, McLeod 1993). Strauss (1997) showed that computer facilitated groups had lower cohesiveness and group satisfaction than face to face groups, primarily because of the rate with which the group members interacted. Even a simple technology such as teleconferencing has been shown to have detrimental effects on group discussion and processes (Alemi, et al. 1997). Clearly face to face meetings, when it is possible and when it is run well, is more efficient in getting ideas across to group members and as a consequence of this improved efficiency, group members feel that others have heard their point of view.
Summary of Lessons Learned & Principles Recommended
What has been learned?
There are many group processes that can be used to model a group's decision. A review of some of these processes creates bewildering methods of conducting team work. This section reviews key lessons learned from research on effective team processes. Three and half decades of research on effective group work point to following lessons:
The Integrative Group Process (IGP) is based on the lessons learned in the last 30 years of research on group processes:
The process we recommend for your use in modeling a group's decision is called Integrative Group Process. It is an eclectic group process based on more than 30 years of research on teamwork and group interaction literature. This process has 6 steps:
Step 1: Select the Best
What is the idea?
The composition of the group is an important and generally controllable aspect of the group. Occasionally, analyst may avoid putting too many high status members in the group -- fearing that they would not be able to work together. This is a mistake. The best experts and decision makers must be invited to participate in the meeting, without them crucial information will be missing from the meeting and quality of the decision may be affected. Instead of avoiding conflict, the IGP process helps manage conflict among group member so that productive quality work can be accomplished despite status differences and presence of a history of conflict.
The analyst needs to think about what portion of the group should come from inside versus outside of the organization. If employees closest to the process are invited to the decision making group, then the group's decisions are more likely to be implemented. If people removed from the process, perhaps experts outside the organization, are engaged, then more radical solutions may be proposed (Stumpf, 1978). In the end a balance needs to be struck between the percent of the group selected from inside or outside of the organization.
The size of the group should depend on its purpose. Experiments with groups of various size have shown that if the quality of the group's solution is of considerable importance, it is useful to include a large number of members (e.g., seven to nine) so that many inputs are available to the group in making its decision. If the degree of consensus is of primary importance, it is useful to choose a smaller group (e.g., five to seven) so that members can have their opinions considered and discussed (Cummings, Huber, and Arendt, 1974, and Manners, 1975). It is a general rule of thumb that the group size should not be smaller than three to five. Groups that meet face to face should not have more than nine members as each member may not be able to participate adequately.
Heterogeneity of the group's background is closely related to the size of the group and is another important aspect of design of successful groups. A necessary, though not sufficient, requirement for accurate group judgments is to have an appropriate knowledge pool in the group. Since no one person is an expert in all aspects of a problem, diverse backgrounds and expertise are imperative for achieving this heterogeneity. Involving people from different functional units of the organizations helps bring different expertise to the problem. Difference in background and knowledge could, however, accentuate the conflict between the group members and, if neither originality nor quality are criteria for evaluating the team's work, the analyst should select group members to minimize differences in their backgrounds.
Getting people to devote their time to a meeting is difficult. Many remember wasted efforts in other meetings and avoid new meetings. The analyst can take several steps to increase participation. First, examine the purpose of the meeting. If it is difficult to obtain participation, perhaps the the problem being addressed is not important. People who are close to a problem, invariably care about it and are willing to address the problem. But if they feel the problem is not real, or the search for solution is a formality, they are less likely to participate.
Second, the analyst can improve meeting participation by clear communications concerning the meeting logistics and expectations. The communication should clarify why is the meeting important, why is it useful to model the task, and what can be expected at the end of the meeting. It should clarify the logistics of the meeting (i.e., when, where and how) and emphasize that the meeting is an ad hoc group.
Third, it maybe useful to remind the invited group members about who else is being invited. People are more likely to come to meetings where people they admire are present. It also helps to emphasize who nominated the potential group member and that there are very few people asked to participate. It should be clear that the group member's contribution is unique and valued. Finally, it helps if group members are reimbursed for their time by providing them with an honorarium.
Step 2: Meet Before the Meeting
What is the idea?
Before face-to-face meeting, group members are individually interviewed and modeled. If group members live far apart the interviews are done by phone, a series of emails, or through computer connections. Whether done remotely or face-to-face, the interview is scheduled ahead of time. During the interview, which takes roughly one hour, the analyst explains the group's task, elicits the participant's model (attributes or clues), asks the participant to estimate utilities or probabilities, and walks the participant through the steps in the IGP process, so there will be no surprises in the actual meeting. The bulk of the interview time, however, is spent listening to the group member and trying to model the reasoning behind his or her choices.
For example, if the group is to create a decision tree, the analyst asks the member to do so and listens to why specific steps are important in the decision. If the group's task is to suggest alternative solutions, the analyst obtains a list of viable alternatives and tries to understand what evaluation criteria are important for evaluating these alternative. If the group member's opinions depend on many rules, the analyst solicits these rules and makes sure that the mathematical model reflects them. The point is that no matter what the task is, the analyst tries to not only accomplish the task but also understand the reasoning behind it.
It is important to keep in mind while the bulk of interview time and interactions are spent on structuring the decision, the analyst must also assess utilities and probabilities so that the participant understands what is going to happen at the meeting. A brief training in probability concepts might also be useful before assessing probabilities.
After interviewing each group member individually, the analyst collates the responses of all participants and creates a "straw model" of the decision. A straw model refers to a decision model designed to be redone. The face-to-face meeting starts with a presentation of the "straw model" and proceeds with a request for improvements. Constructing the model before the group meeting has three advantages:
Step 3: Redo the Straw Model
Better the Next Time Around
What is the idea?
The analyst lists the basic structure of the "straw model" (the attributes used in evaluation task, the decision events used in a tree structure, or the clues used in a forecast) on flip charts; one component per chart. The flip charts are spread around the room so any member can see what has been said to date. The group convenes to revise the "straw model".
The analyst introduces himself/herself, explains the group's task and agenda, restates the importance of the task, and asks members to introduce themselves. These introductions as an important part of the group process. If members are not explicitly introduced, they will do so implicitly through out their deliberations.
The analyst presents the "straw model," asks the group to revise the model, and focuses group's attention on one of the components in the "straw model" as a starting point. The focus on one component at a time is an important way of managing group's time and conflicts. As group members suggest new ideas or modifications, the analyst records them on the appropriate pages in front of the group. Thus, the analyst serves as a secretary to the group making sure that ideas are not lost. Recording the comments reassures the group members that their ideas are being put in front of the group for further consideration. The process continues until the group identifies, discusses and revises all relevant components in the decision structure.
Through active listening (e.g., nodding, asking for clarification) and recording of group member ideas on the flip charts, the analyst plays the important role of directing the discussion, preventing premature closure of ideas (Van de Ven and Delbecq, 1971), returning the group to task related activities, distributing the group's time over different aspects of the task, restraining expression of critical comments during the generation of ideas, and always separating people from ideas so that ideas are judged on their own merits. The analyst uses the instructions developed for Group Communication Strategy as guidelines for this phase of interaction. The analyst should not participate in the content of the discussions; and should not reword what has been said in his/her own terms.
Step 4: Estimate Model Parameters
Let Numbers Do the Talking
What is the idea?
In this step of IGP, the analyst helps the group put numbers to the decision components (e.g. assess weights for the utility models, assess likelihood ratios associated with various clues). This task is done individually and without discussion until a major difference between the group members is identified. While working individually, the group remains in the presence of one another. Seeing each other working helps the group members exert more effort on the task at hand. As the group proceeds, the analyst collects the group's responses and puts the answers on a flip chart. The scores are not listed in the flip chart in a manner that identifies who has said it. One purpose of this step is to encourage the group to consider the merit of ideas and not who has expressed them. When there are major differences among the estimates, the analyst asks the group (not any particular person) to stop working individually and explain their reasoning to the group.
Step 5: Discuss Major Differences
Discus Issues not Each Other
What is the idea?
IGP focuses the group's discussion on the group's logic. Instead of discussing differences in numerical ratings, the group discusses the reasons behind their ratings. This approach to discussion has been shown to reduce conflict among group members (Hammond et al., 1976; Rohrbaugh, 1979). Disagreements among group members have many different sources. Some disagreements are due to fatigue and unclear problem specification. These disagreements are reduced through clarifying the assumptions behind group member's perspectives. Still other disagreements are due to differences in knowledge and experience. Discussion may reduce these conflicts if group member's succeed in communicating the rationale for their judgments. Some group members may raise considerations others have missed. Group members may see flaws in the logic. Other disagreements are due to value differences. Better communication may not reduce this type of conflict. IGP reduces conflicts that are due to misconceptions and miscommunications and accepts other conflicts as inherent in the task.
In order to save group time, the analyst should identify major differences among members, and focus the group's attention on them. Small differences are probably due to errors in estimating numbers and not due to substantive issues. When there are major differences in numerical ratings, the analyst stops the group members from working individually and asks for a discussion. This is done without identifying what is the group average (norm) or who are the people whose ratings differed from the average. Disclosures of the members involved could have ill effects by polarizing the group. For example, Castore and Muringham (1978) showed that disclosure of individual member preferences lowers the later support for the group's decision. The person who disagrees is not as important as the existence of the disagreement and the need to resolve the disagreement through discussion.
After the group has discussed their differences, the analyst asks group members to individually re-estimate the model parameters. This process of estimating, discussing, and re-estimating leads to more accurate results than the use of other processes that eliminate anyone of the three steps (Gustafson et al., 1973).
Step 6: Ignore Small Differences
What is the idea?
Although the analyst encourages group members to resolve their differences through discussion, at some point, it is necessary to stop the interaction and mathematically resolve minor group differences, such as by averaging the estimates from various group members. If major differences remain, it is necessary to report these differences. In a few days after the meeting has ended: a report is written about the meeting. This report contains several different topics including the following:
A document about the group deliberation is important not only to people who were in the meeting but also to people who were not. Cinokur and Burnstein (1974) had individual subjects list the persuasiveness of pro and con arguments. The net balance of persuasiveness of the arguments correlated with attitude change produced by group discussion. But, more important, other individuals not present in the group discussion, who were exposed to the same arguments, changed their attitudes in the same way. The work of Cinokur and Burnstein shows that the information content of group discussion is important in convincing people outside the group. A well documented decision model can help convince others.
Advanced learners like you, often need different ways of understanding a topic. Reading is just one way of understanding. Another way is through writing. When you write you not only recall what you have written but also may need to make inferences about what you have read.
Send your response by email to your instructor. Include both the question and the answers in your response. Include your contact information.
To assist you in reviewing the material in this lecture, please see slides. In addition, you can listen to the following three-part lecture. Part one reviews the literature:
Part two continues with the review. See part two of the lecture:
Part three of the lecture focuses on the Integrative Group Process:
See also slides on empowerment and cooperation.
Copyright © 1996 Farrokh Alemi, Ph.D. Created on Saturday, September 21, 1996. Most recent revision done on 10/21/2011. This page is part of the course on Decision Analysis, the lecture on Modeling Group's Decision.