A procedure for detecting variability of psychophysical thresholds.

In adaptive psychophysical testing, it is desirable to identify experimental runs during which the subject's threshold changes. This is not easy to do, because the testing level changes during the course of the run. The following screening procedure can assist in detecting threshold variability. At the conclusion of the run, determine for each trial the discrepancy between observed and expected response. Next, determine the linear regression of this discrepancy upon trial number. If the subject's response behavior is stable and well described by the psychometric function, then the expected value of slope of the linear regression is equal to zero. If the observed slope differs greatly from zero, then it is unlikely that the subject's response behavior is stable. Results from computer simulations and psychophysical experiments are presented.