Bland-Altman

When you compare two measurement methods, a high correlation doesn't mean they agree - two instruments can correlate perfectly while one reads consistently high. Bland-Altman answers the right question: how far apart do the two methods tend to be?

What it measures

• Bias. The mean difference between the two methods - the systematic offset.
• Limits of agreement. The range within which 95% of differences fall, telling you how far apart the methods can be for an individual sample.
• Proportional error. Whether the disagreement grows with the magnitude of the measurement.

Data you upload

• Two numeric columns - the same samples measured by method A and method B. One row per sample.

What you get back

• The Bland-Altman plot: difference between methods (y) against their mean (x), with the bias line and upper/lower limits of agreement drawn in.
• Bias with its 95% CI, and the limits of agreement with theirs.
• A trend test for proportional bias across the measurement range.

Method note

The bias is the mean of the paired differences; the limits of agreement are bias ± 1.96 × SD of those differences, with confidence intervals on all three. A regression of the differences on the means checks for proportional error - a non-zero slope means the methods disagree more at high values than low, which the plot makes immediately visible. Correlation is deliberately not the headline number here, because it answers a different question.

Related

For the strength of association between two variables (not their agreement), see Correlation & Regression. For replicate precision and method validation metrics, see Assay QC & Validation.