The 'smallest real difference' as a measure of sensitivity to change: a critical analysis.

If assessment instruments are used to measure efficacy or effectiveness, for example of rehabilitation programmes, they have to be sensitive (or responsive) to change. However, up to now there has been no consensus on what coefficient should be used for this end. Various effect sizes and criterion measures are widely acknowledged; tests of inferential statistics are also still used. The different coefficients may well provide different rank ordering of competitive instruments. Recently, the so-called 'smallest real difference' (SRD) was proposed as a measure of sensitivity to change. In the original formulation, the SRD was defined as the 95% confidence limit of the standard error of measurement (SEM) of the difference scores. Conceptually this is equivalent to what is known as the 'reliable change index' in psychotherapy research. The absolute values of the SEM/SRD indicate measurement error. In our view, this is merely complementary to the reliability concept and not a measure of sensitivity to change. Instead, we suggest using the percentage of patients reaching the SRD criterion to compare the sensitivity to change of competitive instruments. In contrast to other sensitivity-to-change indices, such an approach takes the different reliabilities of competitive assessment instruments explicitly into account.