People optimistically search for favourite images

In many real-world decisions, options appear sequentially, and choosing an option ends the search. These situations are often described as “fiancé problems,” since one cannot return to a rejected potential marriage partner if a better option is not found. Mathematically optimal decision makers in such scenarios should use prior knowledge about the distribution of option values, such as their expected value and variability, to decide whether continued search is likely to reveal a better option than the one currently available. Previous work shows that people often sample too few options, relative to optimal benchmarks, when searching through numerical economic values such as prices. Recent computational modelling attributes this undersampling to mis-specified, or biased, prior beliefs, whereby participants expect future options to be lower valued than they should. Here, we tested whether biased prior beliefs might also explain the opposite bias: the tendency to sample too many options in image-based searches, such as choosing the most attractive face. In ten datasets spanning five image-based choice contexts we replicated oversampling bias in every dataset. We applied computational modelling and, across these image-based tasks, most, but not all, participants were best fit by the biased prior model, which expected future option values to be higher than warranted by optimality. Thus, optimistically and pessimistically biased priors may contribute respectively to over- and undersampling biases. These suboptimal search strategies could manifest in real-world decisions such as shopping, choosing dating partners, or evaluating trustworthy job candidates.