Analysis of time trends in adaptive designs with application to a neurophysiology experiment

Time trends are present in many sequential experiments. Adaptive designs use accruing data to select future design points. It has been observed that the presence of time trends in adaptive designs can bias results of the study. We propose one method of dealing with time trends in analysing adaptive designs. The method, relevance weighted likelihood, weights individual components of the likelihood differently. Consequently, one can downweight earlier data if there is a clear time trend that converges at some point in the study. We apply this methodology to a data set from an adaptive design in neurophysiology. We find that the method is robust and useful in getting more precise estimates of an individual subject's median response.