Assess Measurement Agreement Beyond Correlation

Define acceptable disagreement first

When comparing two measurement methods, ask how far apart their readings can be before the difference matters for the intended use. Write that tolerance in measurement units before seeing the results. A method can track changes closely while consistently reading higher than another method.

Bland and Altman's original agreement paper, available from the author's university page, explains why correlation alone cannot establish agreement. Correlation describes association; interchangeability requires examining differences and their practical consequences.

Preserve the pairing and direction

Build one row per independent participant with the two measurements taken under the planned conditions. Record method identifiers, units, time separation, and any repeated attempts. Define the difference consistently, such as new method minus reference method. A positive average then means that the new method reads higher on average.

Use the agreement-study planning worksheet to record the acceptable difference, sampling range, and handling of repeat measurements. Do not select the most favorable replicate after inspecting the results. A simple independent-pair summary is not a complete solution for repeated observations per person.

Interpret a fictional summary

Imagine independent paired differences with mean 2 units and sample SD 3 units. Approximate 95% limits of agreement are 2 ± 1.96 × 3, or −3.88 to 7.88 units. The Bland–Altman summary tool illustrates these calculations.

These fictional limits describe the spread of individual differences under the method's assumptions. They are not a confidence interval for the average difference, and the endpoints themselves have sampling uncertainty. A small average difference can coexist with limits too wide for the intended purpose.

Inspect the pattern before deciding

Plot each pair's difference against its average. Look for an increasing spread, a changing mean difference across the measurement range, and unusual observations that merit source checking. Examine whether approximate normality and reasonably stable variability of differences are plausible for the simple limits calculation.

A summary calculator cannot inspect these features from a mean and SD alone. If variability changes with magnitude, a transformation or another agreement model may be needed. Preserve original units and explain how any transformed results relate to practical tolerances.

Complete the review by comparing the observed limits and their uncertainty with the prespecified acceptance range. Record any data corrections, exclusions, or changes to the analysis. Avoid declaring two methods interchangeable from a high correlation or from a nonsignificant test of average bias.

Source checked on 31 October 2026; fictional arithmetic independently recalculated.