Reviews of Important Papers on Forecasting,
1985-1995 Reviews
Review of:

Arkes, Hal R., C. Christensen, C. Lal, and C. Blumer (1987), "Two methods of reducing overconfidence," Organizational Behavior and Human Decision Processes, 39, 133-144.

[Review written with J. Scott Armstrong]

Judgmental overconfidence was reduced by the use of two strategies. The first strategy, tested on 58 undergraduates, was to "humble" the experimental group by asking them to make predictions of five seemingly easy questions (e.g., 'The Panama Canal has two: entrances one on the Atlantic side and one on the Pacific side. Which entrance is farthest west? A. Pacific entrance. B. Atlantic entrance."

These questions were difficult (the correct answer to the preceding question is "B"). The subjects also expressed their confidence about each of their answers (from 50 to 100%). Feedback was then provided on the first five questions. On 30 subsequent questions these subjects were underconfident. The amount of their underconfidence was about the same in magnitude as the overconfidence expressed by another experimental group that had been given feedback on an initial group of five difficult questions that looked difficult (e.g., "The highest mountain in South America is: A. Aconcagua. B. Tormando." The correct answer is "A'.")

What is the practical implication of this exercise? Arkes suggests that the humbling strategy maybe useful as a way to introduce learning into a situation. That is, it helps to prepare people for new information. (I agree, and have tried it with students. It seems to be effective, although it is also a bit upsetting to some of the learners. They feel better when they receive positive feedback so that they can maintain. their beliefs.) This strategy might be useful in trying to implement the findings of a study that has yielded surprising forecasts.

The second experiment used the same questions. It examined the strategy of asking subjects to justify their answers in anticipation of a group discussion. The hypothesis was that this would lead people to consider arguments as to why their estimates may be wrong. In this experiment with 113 undergraduates, the procedure reduced overconfidence by about one-third. Neither strategy led to significant gains in accuracy.