Generalized adaptive procedure for psychometric measurement

A new Bayesian adaptive psychometric method based on the theory of optimal experiments is introduced and evaluated in simulations over a set of possible observers. The method is flexible enough to be adapted to a wide range of psychophysical experiments (blocks of constant stimuli, n-AFC, yes/no, and discrimination paradigms) and allows specification of many types of assumptions. The four parameters of standard psychometric functions can be estimated in real-time (scales easily to more parameters). An important novelty is the possibility to adjust the desired accuracy of the parameters of interest by weights. Dynamic termination criteria can significantly improve efficiency compared to a fixed number of trials. Yes/no designs turned out to be more efficient than n-AFC in most circumstances. When using blocks of constant stimuli the performance was comparable to single stimuli placement while being more robust to certain experimental phenomena. Simulations showed that the method is at least as efficient and much more flexible than established methods