Short article Inferring causality assessments from predictive responses: Cue interaction without cue competition Anders Winman and Gustaf Gredeba ¨ck Uppsala University, Uppsala, Sweden The authors criticize the use of participants’ predictive responses during a learning phase as a measure of causal assessments (J. M. Tangen & L. G. Allan, 2003, 2004). Simulations demonstrate that a general aim to minimize error as required in the prediction task will produce results in accordance with those observed, in particular apparent cue interaction effects. It is argued that measures estimated from prediction responses do not reflect causal ratings, but are side-effects of learning explained by adaptation in strategy to changes in presented cell frequencies. The results are reinterpreted to be in line with a causal model view, not requiring dual processes or cue competition of lower level associ- ative processes. The question of how humans come to conclude that a certain event is the cause of an effect has been the subject of a lively debate over the past years. One basic approach postulates that causality is learned in a manner that is similar, if not iden- tical, to how Pavlovian associations are acquired in animal learning models. A contrasting approach describes humans as intuitive statisticians who make causal inference by applying rules that are guided by top-down knowledge about causal relations. (see, for example, Shanks, 1993; Spellman, 1996a; Cheng, 1997; Waldmann, 2000). The former approach will be referred to below as associative view and the latter as causal model view. In two recent studies, Tangen and Allan (2003, 2004) proposed that a measure based on predictive responses can be used to infer causality assessments. In this comment we argue that this measure is unsuitable for that purpose. Studies using multiple cues have shown cue interaction effects similar to those found in animal learning. For example, when there are several potential causes of an outcome the presence of a strong causal factor reduces judgements of the causal status of a weaker factor. Cue inter- action effects are straightforward consequences of the Rescorla – Wagner model (Rescorla & Wagner, 1972), which often is incorporated with the associative view. According to the model an outcome holds a finite amount of associative Correspondence should be addressed to Anders Winman, Department of Psychology, Uppsala University, SE – 751 42, Uppsala, Sweden. Email: anders.winman@psyk.uu.se This research was supported by the Swedish Council for Research in the Humanities and Social Sciences. The authors are grateful to Peter Juslin, Henrik Olsson, Marc Buehner, and Jan De Houwer for valuable comments. 28 # 2006 The Experimental Psychology Society http://www.psypress.com/qjep DOI:10.1080/17470210500242953 THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY 2006, 59 (1), 28–37