Item order matters in a function learning task

In a function learning task, participants are taught the relationship between 2 variables, a predictor (e.g., the dosage of a drug) and a criterion (e.g., its effect on mood). Of particular interest in this article is the question of what information does a participant use to generate a response for test examples that fall outside the training region-so-called, extrapolation items. In this article, we test whether the presentation of training items has an impact on the pattern of responses for items requiring participants to extrapolate, and examine, whether the 2 dominant accounts of function learning (Population of Linear Experts [POLE]: Kalish, Lewandowsky, & Kruschke, 2004; and Extrapolation Association Model [EXAM]: DeLosh, Busemeyer, & McDaniel, 1997) can account for this effect. The results show that a manipulation of trial-to-trial changes in the relative magnitudes of the predictor and criterion does influence subsequent extrapolation, and neither POLE, nor EXAM, was able to account for this effect in their current forms. We demonstrate that a model that encodes information about the trial-to-trial changes in the predictor and criterion, and which subsequently uses this information to adjust the retrieved value of the criterion, can account for the effect.