Abstract
In visual search tasks, the relative proportions of target-present and -absent trials have important effects on behavior. Miss error rates rise as target prevalence decreases (Wolfe, Horowitz, & Kenner, Nature 435, 439-440, 2005). At the same time, search termination times on target-absent trials become shorter (Wolfe & Van Wert, Current Biology 20, 121-124, 2010). These effects must depend on some implicit or explicit knowledge of the current prevalence. What is the nature of that knowledge? In Experiment 1, we conducted visual search tasks at three levels of prevalence (6%, 50%, and 94%) and analyzed performance as a function of "local prevalence," the prevalence over the last n trials. The results replicated the usual effects of overall prevalence but revealed only weak or absent effects of local prevalence. In Experiment 2, the overall prevalence in a block of trials was 20%, 50%, or 80%. However, a 100%-valid cue informed observers of the prevalence on the next trial. These explicit cues had a modest effect on target-absent RTs, but explicit expectation could not explain the full prevalence effect. We conclude that observers predict prevalence on the basis of an assessment of a relatively long prior history. Each trial contributes a small amount to that assessment, and this can be modulated but not overruled by explicit instruction.