Clinical significance: False positives in the estimation of individual change

In applied research and in clinical practice we often need to assess the change experienced by patients as a result of the treatment they have received. This paper assesses the performance of several statistical methods designed to estimate such change. This study focuses on one aspect that still has not received attention: the rate of false positives. We have simulated a situation of no-change (pre-post design with no change between pre and post) in which the behavior of nine different statistics have been evaluated. Three different sample sizes (25, 50 and 100) were used and 1000 samples of each size were simulated. To evaluate the behavior of the chosen statistics we have calculated the percentage of times that each statistic has detected change. Since no-change is the simulated situation, any occurrence of change should be considered a false positive. Results are quite striking: none of the nine statistics evaluated offers an acceptable behavior. Good performance is achieved only when the standard deviation of pre-post differences and the traditional criteria are used and not when those proposed by the literature related to the clinical significance are used.