A number of approaches (e.g. Delphi) have been developed to structure information exchange in nominal groups to aid judgment under uncertainty. The authors address the rational underlying such techniques and point to shortcomings in research on the validity of the techniques per se. They advocate a new direction of research, moving from the holistic appraisal and comparison of techniques, towards consideration of the processes responsible for inducing individual judgment change during such interventions, and the identification of key personal and situational factors that may predict such change. The authors describe one repeated-measures design in which groups of subjects are asked to forecast economic and political events using Delphi-like procedures that differ according to the presence/absence and nature of feedback. Characteristics elicited from subjects include measures of ‘confidence’, ‘desirability of forecasts’, and ‘self-rated expertise’. the authors analyse how individual characteristics on the above dimensions are related to (a) accuracy and (b) willingness to change estimates, for each of the Delphi-like procedures. Results indicate an increase in accuracy over each of the three conditions, including a no-feedback control, although similar outcomes appear to derive from different processes. Measurement of panellists’ ‘self-rated expertise’, ‘objective expertise’, ‘confidence’, and ‘desirability of outcomes’ show differential utility for a priori panellist selection for structured groups. The authors discuss results in terms of the interaction of technique and panellist specifications. In general, the results support the ‘theory of errors’ as an explanation of the effectiveness of the Delphi technique. Implications of the present findings for the creation of judgment-aiding techniques, panellist selection, and future research are discussed.
