"Nonparametric" A' and other modern misconceptions about signal detection theory.
Many modern descriptions of signal detection theory (SDT) are, at best, distorted caricatures of the Gaussian equal-variance model of SDT (G-SDT). The distortions have sometimes led to important, but unwarranted, conclusions about the nature of cognitive processes. Some researchers reject using d' and beta because of concerns about the validity of explicit underlying assumptions (that are shared with most inferential statistics), instead using either the supposedly "nonparametric" measures of A' and B" or measures known to confound ability and bias. The origins, development, and underlying assumptions of SDT are summarized, then contrasted with modern distortions and misconceptions. The nature and interpretation of common descriptive statistics for sensitivity and bias are described along with important pragmatic considerations about use. A deeper understanding of SDT provides researchers with tools that better evaluate both their own findings and the validity of conclusions drawn by others who have utilized SDT measures and analyses.