Reasons for confidence (Koriat, Lichtenstein & Fischhoff, 1980)

108 A. KORIAT, S. LICHTENSTEIN, AND B. FISCHHOFF Somewhat different conclusions have been recently emphasized by investigators in the area of decision making. In these studies (reviewed by Lichtenstein, Fisch- hoff, & Phillips, 1977), confidence judg- ments were elicited as assessments of the probability that a statement is true. Ap- propriateness of confidence was measured by comparing these assessed probabilities with the observed relative frequencies of being correct (hit rates}. An individual *is well calibrated if, over the long run, for all answers assigned a given probability, the proportion correct equals the probability assigned. The general conclusion from these studies is that people are rather poorly calibrated. Although higher probabilities are typically associated with larger hit rates, in an absolute sense the probabilities and hit rates diverge considerably. The major systematic deviation from perfect calibration is overconfidence, an unwar- ranted belief in the correctness of one's answers (Lichtenstein et al., 1977). Typical results have shown, for example, an ob- served hit rate of .6 associated with assessed probabilities of .70, whereas for answers assigned probabilities of .90, only about 75% are correct. Recent studies using a variety of question and response formats found that answers endorsed with absolute certainty (i.e., p = 1.00) were wrong about 20% of the time (Fischhoff, Slovic, & Lichtenstein, 1977). Subjects had enough faith in these expressions of cer- tainty that they were willing to risk money on them. Although overconfidence has been a pervasive bias in a variety of tasks, existing experimental data leave the underlying psychological process (es) quite unclear. Building on some speculations by Fischhoff et al. (1977) and Koriat and Lieblich (1977), the present work looks at the origins of unwarranted certainty. To assess one's confidence in the truth of a statement, one fi*st arrives at a confidence judgment based on internal cues or "feelings of doubt" (Adams & Adams, 1961). The judgment is then transformed into a quantitative expression, such as the prob- ability that the statement is correct. Un- warranted certainty might be linked either to the confidence judgment or to the trans- lation of that judgment into a number. The monotonic relationship between as- sessed probability and percentage correct indicates that people can monitor the cor- rectness of their answers with some success. Thus, miscalibration may be due primarily to inappropriate translation of those feel- ings into probabilistic terms. Such transla- tion difficulties might be correctable by proper training. In fact, a calibration training program that provided calibration feedback after each of 11 sessions of 200 items was found to be effective by Lichten- stein and Fischhoff (in press). Their finding that almost all the improvement was ac- complished after the first round of feed- back suggests that what subjects learned was to make simple adjustments in the magnitude of the subjective probabilities they reported. These results, however, do not preclude the possibility that miscalibration arises from the way in which information is used and evaluated when making confidence judgments rather than from the transla- tion of these judgments into probabilities. The fact that the improvement in Lichten- stein and Fischhoff's (in press) subjects failed to generalize to probability assess- ment tasks with the same response mode but different content suggests that transla- tion problems are not the whole story. One information-processing mechanism that would produce overconfidence is to rely more heavily on considerations con- sistent with a chosen answer than on con- siderations contradicting it. Such a pre- disposition could express itself either during memory search and retrieval, by gradually biasing the search toward evidence sup- porting a tentatively preferred answer, or in a postdecisional stage in which the evidence is reviewed and confidence is assessed. Whichever is the case, forcing subjects to write down as much pertinent evidence as possible before choosing an answer or evaluating its validity should reduce the selective bias in the utilization of evidence and result in an improved calibration.