Decision-making groups can potentially benefit from pooling members' information, particularly when members individually have partial and biased information but collectively can compose an unbiased characterization of the decision alternatives. The proposed biased sampling model of group discussion, however, suggests that group members often fail to effectively pool their information because discussion tends to be dominated by (a) information that members hold in common before discussion and (b) information that supports members' existent preferences. In a political caucus simulation, group members individually read candidate descriptions that contained partial information biased against the most favorable candidate and then discussed the candidates as a group. Even though groups could have produced unbiased composites of the candidates through discussion, they decided in favor of the candidate initially preferred by a plurality rather than the most favorable candidate. Group members' pre-and postdiscussion recall of candidate attributes indicated that discussion tended to perpetuate, not to correct, members' distorted pictures of the candidates.

An information-sampling model proposed by Titus (1985,1987) and observations of discussion content (Stasser, Taylor, & Hanna, 1989) suggest that face-to-face discussions often fail to disseminate unshared information. However, groups may be less prone to overlooking unshared information if they believe that their task has a demonstrably correct answer (Laughlin, 1980). University students read a murder mystery and then met in groups to discuss the case. Groups believed they had either sufficient (solve set) or insufficient (judge set) evidence to determine the guilty suspect. When critical clues were unshared before discussion, 67% of solve, but only 35% of judge, groups identified the guilty suspect. Discussion content analyses showed that solve groups focused more on the critical clues. Stasser, Taylor, and Hanna (1989) found that decision-making groups were much more likely to discuss information that members shared before discussion than to discuss information that was held by members individually. This finding was anticipated by an information-sampling model of discussion that was proposed by Titus (1985,1987). The central idea in the model is that the probability of discussing an item of information increases as the number of members who can recall and mention the item increases. This is a fairly simple idea, but it has some disconcerting implications. For example, it suggests that group discussion may be an ineffective way of disseminating information; information that is known to only one or a few members will often be omitted from discussion. Moreover, Stasser, Taylor, and Hanna (1989) found that groups were not only more likely to mention information if it was distributed to all before discussion, but they were also more likely to bring it up repeatedly throughout discussion. These findings suggest that group decisions will often reflect the common knowledge shared by members before discussion and not the diverse knowledge emanating from their unique perspectives and experiences.Of particular interest in this article is the possibility that the failure of groups to consider fully unshared information may be due in part to how members construe their decision-making task. Members may view the task as a "matter of judgment" and let their discussion be. guided by the goal of reaching a consensus. Conversely, members may view the task as a problem to be solved and presume that there is a critical set of information that would allow them not only to identify the correct answer

The information-sampling model (Stasser & Titus, 1985) suggests that group discussion tends to focus on information that members already share before discussion and information that supports, rather than opposes, the predominant sentiment within the group. The model further predicts that the tendency to reiterate already-shared information will be reduced when little information is available for discussion (low load) and when most information is unshared before discussion (low percentage shared). In a political caucus simulation, university students read partial descriptions of three candidates for student body president and then met in 4-person groups to discuss and decide which candidate was best Depending on condition, candidate profiles contained either 12 (low load) or 24 (high load) items of information, distributed such that either 66% or 33% of the information was shared (i.e., given to all members) before discussion. Members' pre-and postdiscussion recall suggested that discussion disseminated sizable amounts of unshared information only under low percentage shared, most notably in the low-load/33%-shared condition. Moreover, discussion biased recall in favor of the group's choice.

Predictions from an information sampling model of group discussion were examined (Stasser & Titus, 1985: (a) Groups are more likely to discuss information if it is held by all members than if it is held by 1 member, and (b) this focus on already shared information increases as group size increases. University students read descriptions of candidates for student body president. These descriptions were constructed so that some information (unshared) was read by 1 member before discussion, whereas other information (shared) was read by all members. Three-and 6-person groups discussed the candidates and decided which was best suited for the position. As predicted, discussions contained, on the average, 46% of the shared but only 18% of the unshared information; this difference was greater for 6-person than for 3-person groups. Moreover, structuring discussions increased the amount of information discussed, but this increase was predominately due to discussion of already shared information.

No abstract