by Benda, N.; Bürkner, P.-C.; Freise, F.; Holling, H.; Schwabe, R..
Series: 2016-01, Preprints
In psychophysical experiments with quantal dose-response a common prob-
lem is the occurrence of lapses due to inattention or, as in forced choice ex-
periments, the occurrence of incorrect guesses - or both. For these situations
an optimized sequential design is proposed based on the Fisher information
evaluated at the maximum likelihood estimate. This sequential design is compared to the classical Robbins-Monro method of stochastic approximation and modi cations thereof in the original non-parametric approach as well as adapted to the current model with false answers and to Wu's (1985) original method based on the maximum likelihood estimate by means of a
simulation study. In these simulations the optimized sequential design turns
out to perform substantially better than its non-parametric competitors, in
particular, in the situation of starting points (initial guesses of the parameter of interest) which are far from the true value. Overall the optimized version also outperforms the original maximum likelihood based method by Wu (1985).
psychophysics, quantal response, false answers, forced choice experiments, sequential design, stochastic approximation