Does Overconfidence Lead to Poor Decisions? A Comparison of Decision Making and Judgment under Uncertainty

Article excerpt

Two within-subject studies of business executives indicate that overconfidence in judgment is reduced when actual decisions are made. Subjects projected quarterly earnings for 50 firms based on reported earnings from the previous 12 quarters. They stated their confidence in forecasts, were given a $10 allocation and were allowed to invest in any, all or none of their forecasts. Subjects chose to act on a relatively small portion of their forecasts but were more accurate and better calibrated when making investment decisions than when making forecasts. The relatively more accurate subjects made more investment decisions and riskier decisions than the relatively less accurate subjects. This suggests that subjects had a sense of whether their knowledge was appropriate for decision making, and acted accordingly. Improved calibration for decision making (investments) compared to judgments (forecasts) appears to have been the result of an additional evaluation stage which occurred between making a forecast and acting upon it. Most subjects tended to 'think twice' before acting, which may have lead, to more thorough information processing and improved calibration.

Findings in the literature on overconfidence are wide-ranging and robust (Fischhoff, 1982; Lichtenstein, Fischhoff, & Phillips, 1982).1 Poorly calibrated decision makers are thought to make ill-advised choices based on sub-optimal information searches. Generalizing somewhat, decision makers tend to be overconfident for difficult problems and underconfident for easy problems (Lichtenstein & Fischhoff, 1977). Findings are not unequivocal with respect to experts. Some show significant overconfidence while others show reasonably well calibrated experts. Characteristics of the decision tasks may account for the differences (Keren, 1991 ). Weather forecasters, accountants and loan officers can be trained to be calibrated due to the clarity and timeliness of feedback while physicians, psychologists and strategists are more prone to overconfidence due to the perceived uniqueness of the circumstances at hand and the difficulty of learning from feedback under these circumstances (Christensen-Szalanski & Busheyhead, 1981; Hogarth, 1981; Oskamp, 1965; Russo & Schoemaker, 1992; Tomassini, Solomon, Romney, & Krogstad, 1982).

Relatively unaddressed in this research is the question of whether there is a difference between overconfidence with respect to judgments and overconfidence in decisions. Most studies of overconfidence require subjects to evaluate information and state their confidence level in judgments they would make from the information. The findings of overconfidence in the reported judgments is taken as an indication that overconfidence would also be evident in decision making, that is, that overconfidence in judgment leads to ill-advised action. However, in many real-world circumstances, people are not required to act on their judgments. Action implies consequences for which decision makers are often held accountable. If they believe their knowledge is limited, individuals may restrict their decision making, that is, the items upon which they are prepared to act, to a subset of their judgments. (Clearly, in some instances, it is a decision to not act upon an item, particularly when non-action implies consequences. However, in general, I mean to restrict the category 'decision' to actions based upon judgments.)

For example, most of us have expressed an opinion rather strongly only to recant somewhat when asked to back it up with action, in the form of a bet or an investment. Timidity with respect to actual decisions is likely the result of loss aversion as decision makers have been shown to weigh losses at least twice as much as gains (Kahneman & Lovallo, 1993; Kahneman & Tversky, 1979). Faced with the possibility of a loss, decision makers 'think twice' before acting, resulting in somewhat cautious behavior. Overall, the result may be a "conjunction of biases" (Kahneman & Lovallo, 1993, p. …