Adaptive Methods for Estimating Empirical Thresholds

Adaptive procedures estimate the stimulus level needed for a fixed level of performance. The stimulus that yields some proportion of correct responding in a forced-choice task (or a specified hit rate in yes-no) is found by systematically varying the stimulus difference during an experimental run. Procedures differ in the rules by which they change or select the next stimulus level, target performance accuracy, criterion for ending a run, method of computing a threshold from the data. Rules that are more complex make greater use of the history of the run, the prior judgments of the experimenter, and the expected form of the psychometric function, whereas simpler rules tend to use this information less. Computer simulations show the more complex rules to be more efficient, that is, to produce less variable and less biased estimates, if their assumptions are correct. The most popular discrimination paradigm, 2AFC, is inferior to m AFC and yes-no, especially at low performance levels.

Econometrics · Computer Science · Mathematics · Statistical Methods and Inference