A sequential design for psychophysical experiments: an application to estimating timing of sensory events.

An experimental subject sequentially receives different levels of a stimulus, and data are recorded on response or non-response to the stimulus. To ensure that the subject cannot predict the next stimulus level based on previous stimulus levels, a randomized design, based on a generalized Pólya urn model, is used to allocate the stimulus levels. The goal of the experiment is to elicit information efficiently about the relationship between stimulus level and response (either for an individual subject or a group of independent subjects), by estimating quantiles of the stimulus-response curve. Our design allocates stimulus levels unimodally and symmetrically around the unknown median of the stimulus-response curve. We discuss estimation under a broad family of distributions and also fully discuss design issues and options. This design was used for an experiment in neurophysiology in humans to estimate the timing of onset of kinesthetic stimuli. Such psychophysical studies can increase our understanding of normal and pathological function. We present data from that experiment.