Diagnosticity and pseudodiagnosticity

Five experiments with 596 undergraduates contrasted Ss' intuitive evaluation of data for hypothesis testing with the Bayesian concept of diagnosticity. According to that normative model, the impact of a datum, D relative to a pair of hypotheses, H and H, is captured by its likelihood ratio, equal to P(D/H)/P(D/H). Results show that when Ss were asked to test the validity of H, only half expressed an interest in P(D/H). That proportion increased when they were asked to determine whether H or H was true. That proportion decreased when the instructions more forcefully encouraged Ss to solicit only pertinent information. Thus Ss generally had a strong interest only in the conditional probability that mentioned the hypothesis (or hypotheses) that they were explicitly asked to test. When, however, they were presented with both components of the likelihood ratio, most Ss revealed a qualitative understanding of their meaning vis-à-vis hypothesis testing. Results are discussed in terms of the kinds of understanding that people might have for statistical principles. (22 ref) (PsycINFO Database Record (c) 2006 APA, all rights reserved).