INDIVIDUAL This chapter describes and evaluates ways of combining individual judgments to produce improved judgments. There are essentially two approaches to the problem: mathematical and behavioral aggregation. Mathematical aggregation, involves techniques such as the calculation of a simple average of the judgments of the individual group members. In behavioral aggregation a group judgment is reached by members of the group communicating with each other either in open discussion or via a more structured communication process. Mathematical Aggregation There are a number of advantages to be gained by using mathematical aggregation to combine the judgments of the individual members of a group. In particular, the methods involved are relatively straightforward. For example, we might ask each member of a group to estimate the probability that the sales of a product will exceed 10000 units next year and then calculate a simple average of their estimates. This means that the more complex and time-consuming procedures of behavioral aggregation are avoided. Moreover, the group members do not have to meet. Their judgments can be elicited by telephone, post or computer and therefore the influence of dominant group members is avoided. Aggregation Judgment in General Single-value estimates of factors such as costs, sales or times to complete a project are often used in decision analysis models when the use of a probability distribution for every unknown quantity would lead to a model which was too complex to be useful. Two methods of combining individual estimates of unknown quantities are considered below. a. Taking a simple average of the individual judgments If the situation where the individual group judgments can be regarded as being unbiased, with each persons estimate being equal to the true value plus a random error which is independent of the errors of the other estimates. In these circumstances it can be shown that taking the simple average of the individual estimates is the best way of aggregating the judgments. The reliability of this group average will improve as the group size increases because the random error inherent in each judgment will be averaged out. However, each additional member of the group will bring progressively smaller improvements in reliability, so that a point will be reached where it will not be worth the effort or cost of extending the group because a
MILA MUJAADILAH 29115053 DECISION MAKING
sufficiently reliable estimate can be achieved with the existing
membership. b. Taking a weighted average of the individual judgments When some members of the group are considered to be better judges than others then it may be worth attaching a higher weight to their estimates and using a weighted average to represent the group judgment. But using weighted averages is that the judgmental skills of the group members need to be assessed in order to obtain the weights. Methods which have been proposed fall into three categories: self-rating, rating of each individual by the whole group (see, for example, De Groot3) and rating based on past performance. Aggregating Probability Judgments The most pragmatic approach to aggregating probabilities would appear to be the most straightforward, namely, to take a simple average of individual probabilities. This method may not be ideal, but as von Winterfeldt and Edwards put it: The odds seem excellent that, if you do anything more complex, you will simply be wasting your effort. Aggregating Preference Judgments Theres two option when a group of individuals have to choose between a number of alternative courses of action is it possible which is Aggregating preference orderings and aggregate value utilities. a. Aggregating preference orderings There is a satisfactory method for determining group preferences when the preferences of individual members are expressed as orderings. There four conditions which can considered that a satisfactory procedure should meet: (1) The method must produce a transitive group preference order for the options being considered. (2) If every member of the group prefers one option to another then so must the group. (You will recall that this condition was not fulfilled in the production manager/accountants problem which we considered earlier.) (3) The group choice between two options, A and B, depends only upon the preferences of members between these options and not on preferences for any other option. (If this is not the case then, as we saw above, an individual can influence the group ordering by lying about his preferences.) (4) There is no dictator. No individual is able to impose his or her preferences on the group.
MILA MUJAADILAH 29115053 DECISION MAKING
In Ferrel well-known Impossibility Theorem Arrow proved that no
aggregation procedure can guarantee to satisfy all four conditions. It suggests that it is impossible to derive a truly democratic system for resolving differences of opinion. Any method which is tried will have some shortcoming. b. Aggregate Value Utilities It is important to note that Arrows Impossibility Theorem refers only to situations where individuals have stated the order of their preferences. A statement giving an individuals preference order does not tell you about that persons intensity of preference for the alternatives. The derivation of individual values and utilities can help each group member to clarify his or her personal understanding of the problem and also to achieve a greater appreciation of the views of other members. Sensitivity analysis can then be used to test the effect of using individual values and utilities. This may reveal, for example, that certain options are to be preferred to others, irrespective of which individuals utility function is used. Unstructured Group Processes One of the major conclusions of research work on descriptions of group decision making is that of well-documented shortcomings. The presence of powerful individuals can inhibit the contribution of those who are lower down the hierarchy. Groupthink is essentially the suppression of ideas that are critical of the direction in which a group is moving. It is reflected in a tendency to concur with the position or views that are perceived to be favored by the group. Structure Group Processes Awareness of the factors that can degrade group decision making combined with the implicit belief that group judgment can potentially enhance decision making has led to a number of structured methods to enhance group decision making by removing or restricting interpersonal interaction and controlling information flow. One such major method has been Delphi Essentially, Delphi consists of an iterative process for making quantitative judgments. The phases of Delphi are: 1) Panelists provide opinions about the likelihood of future events, or when those events will occur, or what the impact of such event(s) will be. These opinions are often given as responses to questionnaires which are completed individually by members of the panel. 2) The results of this polling of panelists are then tallied and statistical feedback of the whole panels opinions (e.g. range or medians) are
MILA MUJAADILAH 29115053 DECISION MAKING
provided to individual panelists before a repolling takes place. At
this stage, anonymous discussion (often in written form) may occur so that dissenting opinion is aired. 3) The output of the Delphi technique is a quantified group consensus which is usually expressed as the median response of the group of panelists. Decision Conferencing Decision conferencing was invented in the late 1970s by Cameron Peterson at a US-based consulting firm called Decision and Designs Incorporated. Essentially, decision conferencing brings together decision analysis, group processes and information technology over an intensive twoor three-day session attended by people who wish to resolve a complex issue or decision. In the background another decision analyst uses interactive decision-aiding technology to model individual and group views on such issues as multi-attribute option evaluation and resource allocation.