Minimal clinically important differences (MCID) and smallest worthwhile differences (SWD) are widely used to interpret outcomes in sport and exercise science. When applied as fixed decision rules, they conflate statistical inference with clinical judgement, promoting dichotomous thinking that discards the uncertainty essential for individual-level decisions. We argue that Bayesian inference and multilevel modelling support more coherent evaluation of meaningful change—not by eliminating thresholds, but by returning them to where they belong: the clinical and performance context, not the statistical model.
An athlete improves single-leg hop distance by 4% following a 6-week rehabilitation programme. The predefined MCID is 5%. The athlete is classified as a ‘non-responder’. But is that conclusion justified?
The threshold is the problem, not merely its value. Decrease the SWD from 5% to 4% and athletes near the boundary change category—yet nothing in the real world has changed. The same instability runs deeper for MCID: applying different derivation methods…