Abstract: It is commonly known that the benefit of decision-making group could pool me mbers’ partial information. The groups could compose an unbiased recognition of the decision alternatives and then make the best decision. However, the biased sampling model of group discussion suggested that group members often fail to ef fectively pool their information because of two kinds of tendencies: (1) Informa tion that members held in common before discussion and (2) information that supp orted members’existing preferences. This study tested the above hypotheses of t he model by a simulation in laboratory experiments of selection of candidates fo r student body president and then met in 4-person groups to decide which candida te was best suited for the position. 180 Zhejiang University undergraduates part icipated in these experiments under 3 conditions. Assignment to experimental con dition and to decision-making groups within conditions was random. There were 60 subjects of 15 groups in each of the shared, unshared／consensus and unshared／ conflict conditions. Their preference and information about each candidate they could remember in pre-and post-discussion were measured on questionnaires, which were compared by several methods. The results showed: most of the groups decide d in favor of the candidate initially preferred by most of the members rather th an the most favorable candidates. The data of group members’ pre-and post-discu ssion recall of candidate attributes indicated that discussion tended to perpetu ate, not to correct members’ distorted pictures of the candidates. The results confirmed the hypotheses of the biased sampling model of group discussion propos ed by Stasser and Larson. In order to improve the efficiency of pooling informat ion in group decision-making, some measures, such as giving group decision train ing in advance, were also suggested by this study.

Key words: decision-making group, information sampling, bias