Familiarity Bias and Belief Reversal in Relative Likelihood Judgment

People are often called on to make an assessment of the relative likelihood of events (e.g., which of two investments is more likely to outperform the market?) and their complements (which of the two investments is more likely to perform no better than the market?). Probability theory assumes that belief orderings over events and their complements should mirror each other (i.e., P(A) ≥ P(B) iff P (not-A) ≤ P(not-B)). This principle is violated in several surveys in which we asked people to assess the relative likelihood of familiar versus unfamiliar events. In particular, respondents are biased to view more familiar events (and their complements) as more likely than less familiar events (and their complements). Similarly, we observe that subjects are biased to view less familiar events (and their complements) as less likely than more familiar events (and their complements). Further studies demonstrate that the familiarity bias is less pronounced among subjects who are asked to judge the probability of each event rather than which event is more likely. Moreover, a greater proportion of subjects rate the more familiar event as more likely than assign a higher probability to that event. These patterns can be construed as belief reversals, analogous to the preference reversal phenomenon in decision making. The data are consistent with a contingent weighting model in which the process of judging relative likelihood biases attention toward evidence supporting the target hypothesis (and away from evidence supporting its complement). Because it is easier to recruit evidence supporting familiar events than unfamiliar events, this skewed attention causes both familiar events and their complements to be judged more likely, on average, than unfamiliar events and their complements.