Enhancement and reduction of associative retroactive cue interference by training in multiple contexts.
search is that if a conditioned stimulus (CS) is a very good predictor of an unconditioned stimulus (US), the conditioned response elicited by other CSs that are trained in compound with the good predictor will be impaired. This effect is often called stimulus competition, or cue selection. A typical example is the forward-blocking effect (Kamin, 1968), in which a CS, A, is paired to the US during Phase 1 (i.e., A ), and then, in Phase 2, A is presented in compound with a novel CS, X, and followed by the US (i.e., AX ). This results in weaker responding to the target cue X in a subsequent test phase relative to responding by control subjects that were not exposed to the A pairings in Phase 1. There are many other designs through which stimulus competition effects have been observed (e.g., Wagner, Logan, Haberlandt, & Price, 1968), but all of them have included compound training of the good predictor and the target cue. Consequently, despite many important discrepancies among the theories that attempt to explain these effects, all of them depend on the AX compound training as essential for A influencing responding to X (e.g., Mackintosh, 1975; Miller & Matzel, 1988; Pearce & Hall, 1980; Rescorla & Wagner, 1972). That is, according to most theories, if X were not trained in compound with A, responding to X would not be impaired. The purpose of this research was to test this assumption. Our basic paradigm consisted of a simplification of the typical forward blocking and backward blocking treatments, so that cue A would not be presented in compound with X. (Backward blocking consists of reversing the order of the two training phases of forward blocking; Shanks, 1985.) Thus, in our paradigm, Group Forward Competition was exposed to A in Phase 1 and to X (rather than AX ) in Phase 2 and was then tested on X. Group Backward Competition received the same treatment except that the two training phases were conducted in the reverse order. The predictions of various theories for those two groups are quite clear. According to traditional associative theories, no competition between A and X should be observed in either case. Nevertheless, because the context is trained in compound with both A and X, if the context acquired strong associative strength, these theories would predict either exclusively forward blocking by context (e.g., Mackintosh, 1975; Pearce & Hall, 1980; Rescorla & Wagner, 1972) or similar degrees of forward and backward blocking by context (e.g., Miller & Matzel, 1988; Shanks & Dickinson, 1987). Thus, if competition were stronger in the forward case, or if similar degrees of responding were observed regardless of trial order, the experiment would not speak to the issue of competition between individually trained CSs. Moreover, several nonassociative (rule-based) models would also predict impaired responding to X in the two groups, regardless of trial order, not because of associative interference but because of the identical statistical contingency to which the two groups are exposed (e.g., Allan, 1980; Busemeyer, Myung, &