An examination of factors contributing to delphi accuracy

Three experiments examined the accuracy in the Delphi method and found that predictions derived from the group were more accurate than those of 95 per cent of the individual panelists, but did not exceed in accuracy the best panelists.

Abstract

Abstract Three experiments examined the accuracy in the Delphi method. The first experiment assessed the accuracy of group predictions over 1‐, 2‐ and 3‐ month time spans. Results indicated that predictions derived from the group were more accurate than those of 95 per cent of the individual panelists, but did not exceed in accuracy the best panelists. Experiment 2 evaluated the gross contributions of polling and feedback to Delphi accuracy. The manipulations did not improve the group's ability to forecast the probabilities of the occurrence of events, but did decrease the error in predicting when the events would occur. Experiment 3 separated the effects of polling and feedback as determinants of accuracy. Neither manipulation improved the accuracy of the group's predictions of whether an event would occur. The effect of iterated polling was to reduce the group's error in predicting the time course for those scenarios that did occur.