24  Improving group decision making

One effective way we could improve group decisions is by replacing groups with models. A second approach is to improve the deliberative process itself.

Below are four example options, although the evidence for their efficacy is mixed.

Role assignment involves giving people discrete roles, such as labelling someone as an “expert”. The purpose is to bring out unshared information by making it clear that the individual expert has a role to play. Stasser et al. (1995) found this to be the case.

Devil’s advocacy is the appointment some group members to deliberately advocate against the group’s inclinations. The research behind devil’s advocates is mixed (Katzenstein (1996)). There is some evidence that devil’s advocacy can be helpful and can enhance group performance but this appears heavily dependent on the particular context. Further, there is less research in organisational contexts outside the lab, which leaves open the question of how the organisational context affects effectiveness. For example, if the dissenter does not genuinely hold the view they are espousing (which is often the case if the role is assigned), can they be effective? Will people discount the dissent accordingly? It may be better for groups to encourage real dissent.

Red teaming is the creation of a team tasked with criticising or defeating the preferred solution or plan. There is some evidence of their value, in areas such as in mock trials and war games (Sunstein and Hastie (2015)), but there is little general evidence that they boost group decision making performance.

The Delphi method was invented at the RAND Corporation to assist analysis during the cold war. It is a process by which you ask people to state their opinions anonymously and independently before deliberation. These opinions are then made available to others. It is effectively a secret ballot plus reasons, and provides a basis for hidden information to emerge without reputational or informational cascades. Several rounds of this process can be held as the group converges on a solution. It’s a great way to flush out doubts and dissent.

Rowe and Wright (2001) reviewed tests of the Delphi method against statistical groups. They found the Delphi groups were superior in 12 of 14 studies.

24.1 Comparing the methods

Newell et al. (2022) describe an experiment by Sniezek and Henry (1989):

Sniezek presented sales forecasting problems to four groups of five undergraduate students. The task involved predicting sales volumes for a general store on campus. The groups received time-series data for the preceding 14 months and were asked to predict sales for the following month. Sniezek was interested not only in the comparison of individual and group performance but also in different methods of group interaction.

First, all members of the group provided an independent individual sales estimate which were then collated to provide a ‘collective mean’ judgement for the group. Second, one of four different group decision techniques was imposed on the group: dictator, consensus, dialectic, or Delphi. The dictator technique required group members to decide, through face-to-face discussion, who was the best member of the group and then to submit his or her estimate as the group estimate. The consensus technique was a straightforward discussion aimed at coming to group agreement on the estimate. For the dialectic technique members were provided with the collective mean estimate and then asked to think of all possible reasons why the actual sales volume might be higher or lower than the estimate, and following this discussion a revised group estimate was decided upon. Finally, the Delphi technique required group members to provide estimates anonymously in a series of rounds, with no face to face discussion, until a consensus was reached. (This technique is supposed to maximize the benefits of group decision making and minimize possible adverse effects such as one person monopolizing discussion).

To measure accuracy Sniezek looked at the Absolute Percent Error (APE) between the collective mean estimate and the group estimates. All of the group interaction techniques led to slightly lower forecast error than the simple aggregation of individual estimates. The greatest improvement was shown by the dictator group (reduction of 7.5%) followed by the Delphi (−2.3%), dialectic (−1.3%), and consensus (−0.8%) groups.

However, the APE reduction achieved by the best members was 11.6%, indicating that the best members considerably outperformed all of the group decision techniques. It is also worth noting that although groups seemed to be successful in identifying their best member – hence the relatively good performance of the dictator group – the dictators tended to change their judgement following group discussion, and these changes were all in the direction of the collective mean and hence more error. On average the final dictator estimates had 8.5% higher APE than the initial ones.

Sniezek took care to point out that the generality of these results is not known. The groups were small, the participants were undergraduates, and the techniques were only tested in a single context (sales forecasting); nevertheless, the results suggest that group interaction and discussion can sometimes lead to improvements in judgement accuracy – at least to a level that is better than the collective mean judgement …