Overconfidence refers to the phenomenon that people’s confidence in their judgments and knowledge is higher than the accuracy of these judgments. To investigate this effect, the subjective judgment of confidence in the correctness of a set of answers is compared with the objective accuracy of these answers. In a typical study on overconfidence, participants solve a number of two-choice questions, such as “Which of these cities has more inhabitants: (a) Islamabad or (b) Hyderabad?” Participants answer each question and then indicate on a scale from 50% to 100% how confident they are that their answer is correct. The overconfidence effect occurs when the confidence ratings are larger than the percentage of correct responses. For example, typically only 75% of the answers, for which a participant indicates a level of confidence of 90%, are correct. Normatively, however, nine out of ten answers should be correct. Thus, the judge is overly confident because the subjective confidence exceeds the actual accuracy.
Theoretical Explanations of Overconfidence
The overconfidence effect has been explained by two classes of explanations: biases in information processing and effects of judgmental error. The first class of explanations considers the overconfidence effect as a result of biases in information processing. According to this line of research, a judgmental process starts with a tentative answer to a given question. Then, when estimating the confidence range, test persons selectively search for evidence that the chosen answer is correct but neglect to search for disconfirming pieces of information. Moreover, because of the associative network structure of the brain, confirming pieces of information come to mind more easily than do disconfirming pieces of information. In addition, people often have reasons why they want a particular answer to be true. For example, if they want to appear knowledgeable, this can also contribute to their biased search for confirming information. All these processes combined often lead to an overrepresentation of confirming information. The judges are not aware that the search for information was biased, so they regard the result of this information search process as support for their initial answer and thus express a high level of confidence.
The second class of explanations purports that judgmental errors can occur even if information processing is unbiased. According to this line of research, overconfidence may, for example, be the result of selected item sampling. When confronted with questions such as “Which city is larger, A or B?” participants in an experiment look for a cue that distinguishes these two cities (e.g., “Only city A has an airport. Normally, only large cities have airports.”), and decide accordingly (“City A is larger.”). When asked for the level of confidence, they estimate the validity of the cue (“In 90% of the cases, the city with an airport is larger than the city without an airport.”) and report this value as confidence judgment. If the questions are harder than normal, the cue leads in the wrong direction more frequently than it normally does (e.g., in an experimental sample of questions, only 60% of the cities with an airport are actually larger). This means that for the sample at hand, the validity of the cue is lower than normal. However, the participants have no reason to assume that the sample is not representative (i.e., that the set of questions is harder than normal), and are therefore entitled to report their initial estimation of cue validity as confidence judgment. They thus appear to be overconfident but the apparent overconfidence is the result of selected item sampling on behalf of the experimenter.
Boundary Conditions of Overconfidence
Research has shown that the overconfidence effect does not always occur but is subject to boundary conditions.
First, the size of the overconfidence effect depends on the type of question that has been asked. For two-choice questions, the effect is weaker than for confidence range questions, where participants are asked to estimate a number (e.g., the number of inhabitants of Hyderabad). Instead of estimating the exact number, they have to give a range such that there is a 90% chance that the correct number lies somewhere in the range. Confidence range questions are more prone to effects of biased information processing because there are no explicit alternatives as in the case of two-choice questions. Participants start with guessing a number (e.g., “50,000 inhabitants”) that might be far from right and then search for confirming pieces of information.
Second, the degree of overconfidence depends on the domain of questions. For some domains, the effect is stronger than for others. This effect can be attributed to the difficulty of the set of questions. For hard sets of questions, most answers are wrong, whereas for easy sets, most answers are correct. The overconfidence effect is stronger for harder sets of questions, whereas easy sets tend to produce an underconfidence effect.
Third, individual differences contribute to differences in the degree of overconfidence. Some people express more confidence than others do regardless of the domain of questions. Overconfidence is correlated positively with confidence, but negatively with accuracy of judgment. This means that people who are most overconfident are more confident and less accurate in their judgment then are other people.
Fourth, the degree of overconfidence depends on the level of expertise. People who frequently give judgments of the same type display little or no over-confidence effect. They are well calibrated. This effect is restricted to the area of expertise: When confronted with questions from other domains, their calibration is the same as everybody else’s.
Implications of Overconfidence
The overconfidence effect is not limited to laboratory situations but has been demonstrated in many areas of professional life such as investment banking, clinical psychology, medicine, and others. Unwarranted confidence in one’s own knowledge and competence can yield reckless behavior and lack of openness for disconfirming information, and thus lead to poor performance and severe mistakes. On the other hand, displaying high levels of confidence can also be beneficial for two reasons. First, competence cannot always be measured. Therefore, others might not find out that a confident person is actually overconfident. Second, overconfidence in one’s competence encourages actions that one wouldn’t undertake if one were less confident, but which may nevertheless be successful.
- Hoffrage, U. (2004). Overconfidence. In R. Pohl (Ed.), Cognitive illusions: A handbook on fallacies and biases in thinking, judgment, and memory (pp. 235-254). Hove, UK: Psychology Press.