The base-rate fallacy in the context of sequential categories

Three experiments assessed base-rate utilization in a paradigm in which the relevant categories of the population were trials in sequences of variable length. In one version of stimulus materials, for example, a trial was represented by a toss of a coin, and a sequence was a series of coin tosses. Since each described sequence was to end upon the scoring of a “hit” in any trial, the likelihood of a trial’s being attempted was a decreasing function of its position in the sequence. Subjects who estimated the particular trial upon which the hit in a sequence was most likely to occur showed a significant tendency to ignore the probable higher prevalence—the greater base rate— of early trials in the overall population of atttempts, typically estimating that the hit in the sequence was equally likely on any trial. This tendency, however, was modified by the specific problem context. In all experiments, significantly more normatively correct estimates were made in the coin context, in which the disparity between base rates of early trials and later trials was most pronounced. Although subjects in the third experiment who had previously taken a statistics course displayed superior performance on the coin problem, there was no evidence of transfer of this superiority to another context. These results are interpreted within the theoretical framework of the base-rate fallacy. In particular, the article discusses implications of the finding that general base rates may be ignored even in the absence of specific diagnostic indicators.