Still the same in human causal reasoning: Evidence for a uniform similarity parameter in generalisation and summation.

Abstract

Wagner's replaced elements model (REM) theory implies that generalisation and summation tests depend on a common similarity parameter, but few studies have assessed generalisation and summation within the same experimental paradigm. Three experiments adapted a methodology used in non-human animal studies to investigate this question in the case of human causal reasoning. The studies combined different amounts of training on simple discriminations (A+ vs. C-), compound discriminations (AB+ vs. CD-), and irrelevant novel cue discriminations (An+ vs. Cn-) with testing on single cues, compound cues, and compounds of cues with novel cues. The results were compared with predictions of Pearce's configural model and Wagner's REM elemental model. They also were examined to determine whether generalisation and summation could be accounted for using a single value for the similarity between stimulus compounds and the separable constituent cues of which the compounds were composed. The findings indicated that each theory needed to include common cues to account for the data.