A Bland-Altman plot assesses agreement between two measurement methods by plotting each pair's difference against its mean, with 95% limits of agreement.
Bland-Altman dethroned correlation as the method-comparison standard in the 1980s. The argument: correlation tells you whether two methods rank subjects the same way (concordance), but two methods can be perfectly correlated and yet differ by a constant or proportional bias that matters clinically. The mean-difference plot makes both visible.
Bias is the mean of the differences (B − A). LoA is bias ± 1.96·SD_diff — the range within which 95% of differences should fall under approximate normality. If the LoA is narrow enough that you'd be willing to use the methods interchangeably for clinical decisions, the methods agree; if not, they don't, regardless of correlation.
Look for two patterns in the plot: a non-zero mean line (constant bias) and a sloping trend or fanning (proportional bias / heteroscedasticity). The first means one method is biased high or low; the second means the bias depends on the magnitude.
Two numeric columns, row-paired (one measurement per subject per method).
Bias, SD of differences, 95% LoA + CI on each LoA. Optional percentage-difference mode for proportional-bias data. Proportional-bias regression of difference on mean with CI band.
The LoA is the *predictive* range — where you expect a new pair of measurements to differ. The CI on the LoA itself shows how precisely the LoA is estimated.
A proportional-bias slope significantly different from 0 ⇒ the methods don't agree the same way at low vs high magnitudes ⇒ report percentage difference or model the bias.