Abstract
Our purpose is to help clinicians decide whether a measured change in a patient's performance, usually in the direction of improvement, is likely to represent a true change. We set out the calculations to derive a minimal statistically signi'cant change in certain measures commonly used in evaluating the outcome of rehabilitation. Many of these test measurements, e.g. walking time, are likely to show increased variability within patients as the mean level of performance increases in magnitude. We pay particular attention to the calculations in this case as the data should be transformed to remove the relationship by taking their natural logarithm. Con'dence interval calculations for single measurements and for differences between two measurements (using the repeatability coef'cient) are given both for data that show and for data that do not show a relation between variation and level of performance.