Persistence of Causal Illusions After Extensive Training

Abstract

We carried out an experiment using a conventional causal learning task but extending the number of learning trials participants were exposed to. Participants in the standard training group were exposed to 48 learning trials before being asked about the potential causal relationship under examination, whereas for participants in the long training group the length of training was extended to 288 trials. In both groups, the event acting as the potential cause had zero correlation with the occurrence of the outcome, but both the outcome density and the cause density were high, therefore providing a breeding ground for the emergence of a causal illusion. In contradiction to the predictions of associative models such the Rescorla-Wagner model, we found moderate evidence against the hypothesis that extending the learning phase alters the causal illusion. However, assessing causal impressions recurrently did weaken participants' causal illusions.

Keywords: Rescorla-Wagner model; causal illusion; causal learning; contingency learning; extensive training; illusion of causality.

FIGURE 1

Four simulations of the Rescorla-Wagner model. The figure legend summarizes the number of a, b, c, and d trials (see Table 1) included in each simulation. Learning rate parameters α_cue, α_context, and β were set to 0.4, 0.2, and 0.6, respectively. The value of λ was set up to 1 for trials in which the outcome was present and to 0 for trials in which the outcome was absent. The figure shows the average results of 2,000 iterations with random trial orders.

FIGURE 2

Mean causal ratings (A) and conditional probability ratings (B) after 48 trials in the Standard group and after all 288 trials in the Long group. Error bars denote 95% confidence intervals.

FIGURE 3

Mean causal and conditional probability ratings after each block of 48 trials in the Standard group. Error bars denote 95% confidence intervals.