Widely-Used Measures of Overconfidence Are Confounded With Ability

The concept of “overconfidence” is one of the great success stories of psychological research, influencing discourse in the popular press, business, and public policy. Relative to underconfidence, overconfidence at various tasks is purportedly associated with greater narcissism, lower anxiety regarding those tasks, higher status, greater savings, more planning, and numerous other differences. Yet much of this evidence could merely indicate that there are associations with ability or skill rather than associations with overconfidence. This results from two underappreciated properties of the typical measures of overconfidence. First, performance is an imperfect measure of ability; analyses that account for performance do not sufficiently account for ability. Second, self-evaluations of performance ought to reflect ability in addition to performance; because performance assessments are ambiguous, people should rely upon their prior beliefs about their own ability. I show that these uncontroversial principles imply that commonly-used measures of overconfidence are necessarily confounded with ability. I support these analytical results by reanalyzing two open datasets from prior research. In the first, overconfidence predicts subsequent performance, consistent with overconfidence as a signal of ability but inconsistent with overconfidence as a bias. In the second, the purported association between overconfidence and financial planning can be readily explained by modeling financial knowledge as a common cause of both. Statistical techniques that explicitly address the role of measurement error provide a potential solution if researchers are willing to accept strong and untestable assumptions. This research serves as a stark reminder: when researchers propose that differences in overconfidence are associated with other behaviors, beliefs, or evaluations, they must carefully rule out differences in ability as an alternative explanation