Standard Error of Measurement (SEM) & Minimal Detectable Change (MDC)
How large must a change in a scale/imaging measurement be before it exceeds measurement error and counts as a "real" change? SEM quantifies measurement error; MDC gives the individual-level threshold above which a true change can be declared. Computed from "SD + reliability ICC" or directly from test–retest paired data. Computed locally; data never uploaded.
① Input
| Measurement SD | Reliability ICC / r (0–1) |
|---|---|
How to use & methodology
SEM vs SD?
SD is between-subject variation (people differ); SEM is the error of repeated measurement of the same subject (the measurement's own instability). The smaller the SEM, the more precise the measurement. SEM = SD·√(1−reliability).
How do I use MDC?
At follow-up, a person's change must exceed MDC before you can be statistically confident it is real rather than measurement noise. MDC95 (95% confidence) is commonly reported. For example, if MDC95=8, a change <8 cannot be confirmed as genuine improvement.
Which input should I choose?
If you already have the scale's reliability (ICC) and measurement SD, use 'SD + ICC'. If you have repeated measurements on a group of subjects, use 'test–retest' to estimate SEM directly from the difference SD/√2 — closer to your own sample.
Is MDC the same as MCID?
No. MDC is the 'minimal change exceeding measurement error' (a statistical concept); MCID is the 'minimal change patients find meaningful' (clinically anchored). Both are usually discussed together when reporting treatment effects.