Asymmetries in cue competition in forward and backward blocking designs: Further evidence for causal model theory

Abstract

A hallmark feature of elemental associative learning theories is that multiple cues compete for associative strength when presented with an outcome. Cue competition effects have been observed in humans, both in forward and in backward blocking procedures (e.g., Shanks, 1985) and are often interpreted as evidence for an associative account of human causal learning (e.g., Shanks & Dickinson, 1987). Waldmann and Holyoak (1992), however, demonstrated that cue competition only occurs in predictive, and not diagnostic, learning paradigms. While unexplainable from an associative perspective, this asymmetry readily follows from structural considerations of causal model theory. In this paper, we show that causal models determine the extent of cue competition not only in forward but also in backward blocking designs. Implications for associative and inferential accounts of causal learning are discussed.

This paper is based on a final year undergraduate research project conducted by the first author and supervised by the second author.

Acknowledgments

We thank Duncan Brumby for assistance in collecting data and Mark Haselgrove for running RWM simulations.

Notes

This paper is based on a final year undergraduate research project conducted by the first author and supervised by the second author.

¹ Note that according to this account, full blocking (i.e., a zero rating of the redundant cue) would not be expected to occur; uncertainty about the causal status of R is reflected by a reduced rating near the midpoint of the scale (i.e., “weak blocking”). Recent research has shown that full blocking is the result of additional inferential processes (e.g., Beckers et al., 2005; Lovibond, Been, Mitchell, Bouton, & Frohardt, 2003) and occurs only when participants make additional assumptions about cue-additivity or outcome-maximality.

² In order to make our analyses comparable to those of Waldmann and Holyoak (1992), we averaged ratings during Phase 2. Contrary to Waldmann and Holyoak's experiments, however, entering time of rating into the analysis of our data revealed a significant main effect of time of rating, F(1, 46) = 6.86, MSE = 2.47, in addition to the main effects of cues, causal condition, and the Cues × Causal Condition interaction. Inspection of the data revealed that ratings generally increased from the first to the second time. Conducting the ANOVA on the ratings from the second time only produces an even stronger pattern of results: a main effect of cues, F(1, 46) = 9.98, and a Cues × Causal Condition interaction, F(1, 46) = 44.0, MSE = 7.19, with the main effect of causal condition failing to reach significance, F(1, 46) = 3.57, MSE = 10.81.

³ Strictly speaking, our (and Waldmann & Holyoak's, 1992) design lacks the adequate control group to demonstrate the absence of cue competition in the diagnostic condition (e.g., see Shanks & López, 1996). This is entirely inconsequential for our purposes, however. What matters is the asymmetry in cue competition between predictive and diagnostic learning, not its absolute size.

⁴ Comparable to Experiment 1, ratings increased between the first and second time that they were prompted, F(1, 43) = 9.32, MSE = 2.37.

⁵ Conducting the ANOVA with the terminal ratings from Phase 1, rather than the averaged ratings, produces exactly the same pattern of results: main effects of cues, F(1, 41) = 42.05, MSE = 7.14, and phase, F(1, 41) = 14.92, MSE = 7.48, as well as a Phase × Cues, F(1, 43) = 72.61, MSE = 5.34, and a Phase × Cues × Causal Condition interaction, F(1, 43) = 10.35, MSE = 5.34. The degrees of freedom from this analysis are lower, because for 2 participants the computer failed to record terminal ratings in Phase 1.