Predictions and causal estimations are not supported by the same associative structure

Abstract

Studies performed by different researchers have shown that judgements about cue–outcome relationships are systematically influenced by the type of question used to request those judgements. It is now recognized that judgements about the strength of the causal link between a cue and an outcome are mostly determined by the cue–outcome contingency, whereas predictions of the outcome are more influenced by the probability of the outcome given the cue. Although these results make clear that those different types of judgement are mediated by some knowledge of the normative differences between causal estimations and outcome predictions, they do not speak to the underlying processes of these effects. The experiment presented here reveals an interaction between the type of question and the order of trials that challenges standard models of causal and predictive learning that are framed exclusively in associative terms or exclusively in higher order reasoning terms. However, this evidence could be easily explained by assuming the combined intervention of both types of process.

Acknowledgments

Support for this research was provided by Grant PI–2000–12 from Departamento de Educación, Universidades, e Investigación of the Basque Government to H.M. M.A.V. was supported by an FPI fellowship from the Basque Government (Ref. BFI01.31). We would like to thank Jan De Houwer, Francisco López, and Pedro Cobos for their insightful discussions concerning the experiments presented here.

Notes

¹In the prediction question used by Vadillo et al. (2005), participants had to estimate the likelihood of the outcome in the presence of the cue. Another possible prediction question would ask participants to predict the likelihood of the outcome in the absence of the cue (for an example of questions regarding what happens in the absence of the cue, see De Houwer et al., 2007). For the sake of simplicity, however, we always use the term prediction judgement as referring to what happens when the cue is present. Our assumption is that the results should, in general, be symmetrical for the cue-absent question.

² In fact, the predicted value of the asymptotic associative strength is exactly equal to Δp, if the learning rate parameter β is assumed to have the same value on outcome-present trials and on outcome-absent trials. When this constraint is met, the resulting algorithm is known as the restricted Rescorla–Wagner model (Lober & Shanks, 2000). If β has different values in outcome-present and outcome-absent trials, then the asymptotic value of the associative strength is no longer equal to Δp, but it is still dependent on this statistical index (higher levels of Δp lead to higher associative strengths).

³ Although it is well known that a high frequency of judgements can induce a recency effect (Catena, Maldonado, & Cándido, 1998), previous studies have shown that frequent yes/no responses do not affect participant's ratings given at the end of the experiment in a numerical scale (see Matute et al., 2002). For instance, both Collins and Shanks (2002) and Matute et al. (2002) observed an absence of recency in a numerical judgement given at the end of the experiment, in spite of participants being requested to give this yes/no responses during training. The recency effect only appeared when the numerical judgement itself was requested with a high frequency during training. Thus, it is unlikely that our yes/no responses are affecting ratings given at the end of the experiment.