The effect of group interaction processes on performance in time series extrapolation

Abstract

This study explores the ability of groups to forecast and establish judgmental confidence intervals in time series extrapolation. Thirty-six three-person groups were used to evaluate four different group interaction processes. In addition to staticized, nominal group technique and consensus processes, the study utilizes a modified consensus process, where a selected group member completes the task prior to group discussion and interaction. Using real life time series, subjects produced forecasts and related confidence intervals for six periods. Groups in the modified-consensus structuring process exhibited significantly greater forecast accuracy than all other experimental conditions ( p < 0.001). The superiority was most pronounced for series of high forecast difficulty. These results are discussed in relation to the contribution of the initial estimates as an anchor on which the modified-consensus group can focus.