Preprint Glautier (2017) Non-local influences on associative learning: new data and further model evaluation

Abstract

Previous work (Glautier, 2013) showed that the responses made by humans on trial n in simple associative learning tasks were influenced by events that took place on trial n − 1 and a simple extension of the Rescorla-Wagner Model (RWM Rescorla & Wagner, 1972), the Memory Environment Cue Array (MECA) model, was presented to account for those results. In the current work further evidence of non-local influences on responding during associative learning tasks is presented. The Rescorla-Wagner model and the MECA model are evaluated as models for the observed data using qualitative, Chapman & Robbins, 1990;Dickinson, Shanks, & Evenden, 1984;Lachnit, 1988).Equation 1 Two experiments are reported. These differ from standard associative learning experiments in that their designs are such that cues that were present on trial n − 1were actually informative about the outcome events on trial n. There is no mechanism in the standard Rescorla-Wagner model to take account of this non-local information.Therefore, if participants actually make use of non-local information, the the MECA model should provide a better model of participant behaviour than the Rescorla-Wagner model in these designs. Experiment 1In Glautier (2013) noisy learning curves were better described by the MECA model than by the Rescorla-Wagner model. The noise in those learning curves was attributed to the influence of trial n − 1 events on trial n responding. In the current experiment a more direct attempt was made to show how the events of trial n − 1 could exert an associative influence on the response observed on trial n. Experimental sequences were set-up so that whenever cue A appeared on trial n − 1 then the outcome occurred on trial n. It was predicted that likelihood of response would be greater on trials involving cue B, which was partially reinforced, if the previous trial had involved cue A. Method Procedures were approved by the University of Southampton ResearchGovernance Office and the School of Psychology's Ethics Committee.Participants. One hundred and seven participants took part. One hundred and six were high school students and one was a teacher who accompanied the students. Design and procedure. Participants were given a brief introduction to PEGS and the experiment before being asked to complete a consent form and navigate to the PEGS website. Participants were randomly allocated to one of the four game implementations and then experienced a series of 60 trials following the design summarised in Table 1. The trial sequence contained 15 B+ trials (B+ means cue B followed by reinforcement (+)). The B+ trials always occurred after an A-trial (A-