Covariation, base-rate and causal power in human contingency judgment

This dissertation begins with a review of competing theories of human contingency judgment, and then describes a series of four original experiments designed to systematically investigate the strengths and weaknesses of current models. Among these, Cheng's (1997) Power PC theory of human contingency judgment has risen to prominence. This theory is said to address the inadequacies of both earlier associative (Rescorla & Wagner, 1972) and computational (Jenkins & Ward, 1965) approaches. Several prior tests of the Power PC theory have returned mixed or inconclusive results (Lober & Shanks, 2000; Vallee-Tourangeau, Murphy & Drew, 1997). The four experiments presented here were designed to assess predictions of the Power PC theory which had yet to be fully empirically tested, as well as to expand our current understanding of causal reasoning. Experiment 1 results were consistent with the Power PC predictions in terms of the pattern of participants' judgments, but not with regards to the levels of those judgments. Through a replication and elaboration upon conditions from Experiment 1, Experiment 2 determined that the level of the Experiment 1 judgments was not due to a ceiling effect, as could have been argued by Power PC proponents. Experiment 3 served to investigate whether the concept of reliability could possibly explain the observed deviations from Power PC, and indeed demonstrated a significant interaction of power and reliability. Despite this finding, the conjunction of Power PC and reliability is shown to be problematic. Experiment 4 confirmed that the results from experiments 1 through 3 can be expected to hold true in negative contingency space as well. Following the empirical results, further discussion of the roles of causal power, DeltaP, and reliability raises new questions and reveals a number of viable options for future research in this area. Taken as a whole, the results from these four experiments do not support the original Power PC theory, but do provide insight toward alternatives involving both confidence and reliability, which will provide a more comprehensive account of human contingency judgment