Length of the postfeedback interval and learning of probabilistic inference tasks

Learning of negative linear and J-shaped cue-criterion relations in single-cue learning tasks was studied in an experiment which varied the length of the postfeedback interval. Contrary to the predictions from a hypothesis construction model assuming that subjects construct hypotheses from the data provided by the task, there was no effect of the postfeedback interval. The results were interpreted to mean that the subjects' constructions are based on elements stored in memory, rather than on data from the task.