Scale implications for confidence elicitation

Abstract

Confidence judgments are important in applications such as medical diagnosis and eyewitness identification, where we are interested in a person's certainty of making the correct choice. A common finding in confidence research is that people are overconfident: they overestimate the probability that their choices are correct. One problem with confidence research, however, is that the scale on which judges report confidence can have an adverse impact on the results. The purpose of the current study was to solve this problem through the use of a new confidence scale. WSU undergraduates completed a 30-question multiple choice test on general financial knowledge. For each question, participants chose a correct answer and reported confidence in their choice. Experimental groups differed in the confidence scale that they used throughout the test. One group reported confidence on a 0 - 100% scale, which is typical of previous research. The second group reported confidence using any number they wished, with smaller numbers indicating less confidence. Results indicate that the latter scale leads to confidence judgments that are more predictive of correct answers. These results have implications for the use of confidence in the real world.