Sequential dependencies in children's probability learning.

Abstract

Two experiments were conducted to investigate the applicability of a linear operator nodel to probability learning by preschool children. In Experiment I naive Ss made a motor response under an E-S controlled procedure. 8020 and 50:50 frequency conditions, with test runs inserted into the event sequences, resulted in positive recency in both groups, but sequential statistics indicated the model was not applicable. In Experiment II 16 Ss, only 4 of whom were naive, were each given a different random 50:50 sequence of colors to predict. Response probabilities conditionalized on the run length during the preceding trials again revealed positive recency. Alternation behavior and position preference found in Experiment I did not occur in Experiment II. A 2-parameter linear learning model gave very good predictions for 18 sequential statistics.