An empirical analysis of five descriptive models for cascaded inference

Abstract Subjective cascaded (multistage) inference is typically less conservative than noncascaded (single stage) inference. Several explanations including a “best guess” strategy have been hypothesized to describe the intuitive process for cascaded inference. The present experiment examined the ability of five models to predict subjects' cascaded odds in a probabilistic inference task based upon unreliable reports of binomial events. Three different groups of 30 subjects each served under various conditions. The results suggest that a subject's cascaded odds can best be predicted from the odds he estimates in a single stage inference task in which the likelihood ratio of a datum is equivalent to the formally appropriate likelihood ratio of a datum in the cascaded condition. This finding suggests that subjects may be using the same process to aggregate evidence in both single stage and multistage inference tasks.