The dispersions of estimates of sensitivity obtained from four psychophysical procedures: Implications for experimental design

The dispersions of estimates of sensitivity obtained from the yes-no, two-alternative forced-choice (2AFC), matching-to-sample, and same-different tasks were examined to determine which task would be more appropriate to use in a given experimental context. Consideration was given to the effects of corrections for extreme sampled proportions. These corrections result in biased estimators, and hence the mean-square deviation of the sampled values about the population mean [MSD $(hat d')$ ], rather than that about the mean of the estimates [VAR16 $(hat d')$ ]> indicates more completely the extent of the error in the estimator. For barely discriminable events (d′ ? 0.5), the yes-no and 2AFC tasks had the lowest values of MSD $(hat d')$ . However, for very discriminable events (d′ > 3), the same-different and matching-to-sample tasks had lower values of MSD $(hat d')$ .