The statistical manipulation of Delphi statements

• Main Author: Nelson, Bradley William
• Format: Others
• Published: PDXScholar 1977
• Subjects: Delphi method
• Online Access: https://pdxscholar.library.pdx.edu/open_access_etds/835; https://pdxscholar.library.pdx.edu/cgi/viewcontent.cgi?article=1834&context=open_access_etds

Since its invention twenty years ago, the Delphi technique has been gaining wider and wider acceptance as a tool for forecasting technology; gathering expert opinion from a local to world wide "advice community" upon which government, industry, and other policy making bodies must so frequently rely; and providing judgmental input for studies (e. g., social sciences) where hard data are unavailable or too difficult to obtain. Accompanying this increased acceptance is an increased danger of manipulation of a Delphi to produce the results desired by one certain individual or group of individuals. Manipulation is increasingly being mentioned in the literature as a danger but little has been done to study the problem. Two groups of thirty United States Air Force Officers enrolled in a Masters of Business Administration program participated in a fact probing Delphi containing thirty statements. The participants of one group were given falsified statistical feedback on fifteen of the statements, while the participants of the other group were given falsified statistical feedback on the other fifteen statements. A similar study was done with another group of officers using a value probing Delphi. The results of these studies showed a high degree of success in obtaining a desired value through the use of manipulated statistical feedback. This success was enhanced by running additional rounds. It was also found that the statistical manipulation had a significant effect on the convergence and stability of the Delphi statements. The effects of statistical manipulation on confidence as measured by self-rating was also studied. It was found that there was a significant tendency for Delphi participants to shift their selfrating during later rounds toward the middle. The effect of the distance between a participant's original estimate and the median reported back to him on the amount the participant changed his self-rating was investigated. The results were inconsistent. Statistically manipulated participants showed an overall decrease in confidence, regardless of their original self-rating. Suggestions for extending the research in the area of manipulation of Delphi statements plus a taxonomy of the variables that comprise the problem of manipulation are discussed.