Diagnostics › Method comparison

Bland–Altman repeatability

Bland-Altman repeatability analysis uses two or more replicate measurements per subject to yield the within-subject SD, the repeatability coefficient and the CV.

What is Bland–Altman repeatability?

Single-measurement Bland-Altman compares two methods on the same subjects. Repeatability extends the framework to multiple replicates of the same method — quantifying measurement noise (within-subject SD) without confusing it with between-subject biological variation.

Mechanically a one-way ANOVA with subject as the random factor. Within-subject SD = √MS_within; repeatability coefficient RC = 1.96·√2·SD_within is the within-subject 95% interval for the difference between two replicates.

When should I use Bland–Altman repeatability?

Test-retest reliability assessment.
Establishing the measurement noise floor before a method-comparison study.
Calibrating an assay's analytical precision.

What data does it need?

Long format: subject ID + measurement.

What does it report?

Repeatability coefficient RC = 1.96·√2·SD_w, CV = SD_w / grand mean × 100%, between-subject SD, ICC(1, 1).

What does it assume?

Independent subjects.
Replicates within a subject are exchangeable (no order effect).
Within-subject variance is constant across subjects — check with a within-vs-mean plot.

How do I interpret the result?

If two future replicates of the method on one subject differ by more than RC, you can be 95% confident the difference exceeds measurement noise.

ICC(1, 1) > 0.9 ⇒ excellent reliability; 0.75–0.9 = good; 0.5–0.75 = moderate; < 0.5 = poor.

References

Bland & Altman (1986); Bland (2015) — newer SE formulas for the LoA.