Nonparametric indexes for sensitivity and bias: Computing formulas.

摘要: Computing formulas are derived for two nonparametric indexes of sensitivity and bias that have been suggested signal detectability studies. A relationship is shown between the index P(I), a statistic whose sampling variability known. An additional proposed, which free certain inconveniences, yet yields identical isobias contours. Use new illustrated with several sets data. Development theory has lead to renewed interest in possible processes involved perception, psychophysics, recognition memory. To considerable extent both research these areas rested on specific assumptions about underlying distributions (as various threshold theories versus normality). But even without explicit assumption distributions, data often judged by how close they lie operating characteristic curves from normal different experimental conditions characterized their value d'. Recently there growing "nonparametric" analyses detection/ experiments, where not assumed. Following one line development, Green (1964) experiments using yes-no procedure, area under (theoretical) curve can be interpreted as percentage correct an equivalent unbiased forced-choice test, this true any continuous distributions. The measure determined Pollack Hsieh (1969). similar proof ratingscale due Moses (1966). However, estimate unknown theoretical shape presents difficulties. One expedient connect points use trapezoidal rule (Green & Moses, 1966; Pollack, Norman, Galanter, 1964). If func