Adaptive methods dynamically select stimulus levels during an experiment based on the observer's previous responses, concentrating observations near the threshold where they are most informative. These procedures dramatically reduce the number of trials needed for precise threshold estimation compared to classical methods like constant stimuli.
Staircase Procedures
The simplest adaptive method is the up-down staircase: increase intensity after a miss, decrease after a hit. The transformed up-down method (Levitt, 1971) uses rules like "2-down, 1-up" (two correct before decreasing, one incorrect before increasing) to target specific points on the psychometric function — the 2-down/1-up rule converges on the 70.7% correct point.
2-up/1-down → targets 29.3% point
1-up/2-down → targets 70.7% point
1-up/3-down → targets 79.4% point
Bayesian Adaptive Methods
QUEST (Watson & Pelli, 1983) and its successor QUEST+ maintain a posterior distribution over threshold (and optionally slope) parameters. After each trial, the posterior is updated via Bayes' theorem, and the next stimulus level is chosen to maximize expected information gain. This approach provides the most efficient threshold estimation available, often requiring only 20–40 trials for precise estimates.
The PEST procedure (Parameter Estimation by Sequential Testing) uses a modified staircase with step-size rules designed to converge quickly. Modern implementations like Psi methods and entropy-based approaches extend these ideas to simultaneously estimate multiple parameters of the psychometric function.