Signal Detection Theory (SDT) is one of the most successful and widely applied mathematical frameworks in psychology. Developed in the 1950s and formalized by Green and Swets in their landmark 1966 book Signal Detection Theory and Psychophysics, SDT provides a principled method for separating an observer's ability to discriminate between signal and noise from their willingness to report a signal's presence.
The Core Framework
SDT assumes that on every trial, the observer receives a sample from one of two distributions along an internal decision axis: the noise distribution (when no signal is present) and the signal-plus-noise distribution (when a signal is present). Both distributions are typically assumed to be Gaussian with equal variance. The observer sets a criterion along this axis; observations exceeding the criterion produce a "yes" response.
c = −0.5 × [z(HR) + z(FAR)]
β = f(z_HR) / f(z_FAR)
The key insight of SDT is that these two aspects of performance — sensitivity (d′) and criterion (c or β) — are mathematically independent. An observer can be highly sensitive but conservative, or poorly sensitive but liberal, and these properties are fully separable in the SDT framework.
ROC Analysis
The Receiver Operating Characteristic (ROC) curve plots hit rate against false alarm rate across different criterion settings. For equal-variance Gaussian SDT, the ROC curve is symmetric, and d′ determines the curve's distance from the diagonal (chance performance). Unequal-variance models produce asymmetric ROC curves, which are commonly observed in recognition memory experiments.
While the basic SDT model addresses yes/no detection, extensions handle rating-scale responses (producing multiple ROC points from a single experiment), forced-choice paradigms (where sensitivity is measured without criterion effects), and multi-dimensional signals where the decision space becomes a hypersurface rather than a single point on a line.
Applications Across Psychology
SDT has been applied far beyond its origins in psychophysics. In memory research, it distinguishes true recognition ability from response bias. In medical diagnosis, it separates a radiologist's skill from their tendency to call findings positive. In eyewitness identification, it measures how well a witness can discriminate between guilty and innocent suspects, independent of their willingness to make an identification.
The mathematical elegance and practical utility of SDT have made it one of the most enduring contributions of mathematical psychology to the broader behavioral sciences. It demonstrated that a simple mathematical model could resolve longstanding debates about the nature of sensory thresholds and laid the groundwork for modern computational approaches to perception and decision-making.