Experimental evaluation of policies for sequencing the presentation of associations

Two policies for sequencing the presentation of associations are compared to the standard policy of randomly cycling through the list of associations. According to the modified-dropout policy, on each trial an association is presented that has not been presented on the two most recent trials and on which the observed number of correct responses since the last error is minimum. The second policy is based on a Markov state model of learning: on each trial, an association is presented that maximizes an arithmetic function of Bayesian estimates of residence in model states, a function that approximately indexes how unlearned associations are. Retention is improved relative to the standard policy only for the model-based policy.