INTUITIVE STATISTICAL INFERENCES ABOUT DIFFUSE HYPOTHESES

When data are sampled from a population and subjects revise probability estimates about which population is being sampled, their revisions are less than the optimal amount calculated by using Bayes's theorem; they are conservative. The experiments reported here used binomial populations with proportions that were either defined precisely by a display or defined diffusely by a sample of data. The experimenter randomly selected one of two populations and then sampled data from the selected population. The subjects made very nearly Bayesian revisions on the basis of the first datum sampled, but became markedly conservative when the task required aggregating evidence across a sequence of data. This result was independent of whether population proportions were defined precisely or diffusely. © 1968.