The best guess hypothesis in multistage inference

Abstract

Intuitive multistage inferences are typically excessive when compared with the optimal model, a modified form of Bayes' theorem. One explanation for this excessiveness is that the S primarily attends to the implications of the probable event described by the first-stage inference, neglecting the implications of less likely events. If a S follows this strategy, called a "best guess" strategy, then a testable implication is that his probability revision at the upper level should be insensitive to variations in the distribution of probabilities across all but the most likely event described by the first-stage inference. The results of the present experiment support this hypothesis.