Do two ways of measuring the same thing agree? A correlation won’t tell you — Bland & Altman say plot the DIFFERENCE against the MEAN, draw the average bias, and mark limits at bias ±1.96 SD. ~95% of differences should fall inside. Slide to tighten the two methods.
A Bland–Altman plot charts each pair’s difference (method A − method B) against their mean, then draws the BIAS (mean difference) and the LIMITS OF AGREEMENT = bias ± 1.96·SD of the differences. Under normal differences ~95% fall within the limits; a trend in the cloud reveals proportional bias a correlation would hide. It answers ‘can these methods be used interchangeably?’, not ‘are they correlated?’. A fail-loud self-check throws unless at least 90% of points sit inside the ±1.96·SD limits. ◆ real statistics, node-verified.
The synthetic paired data is the illustration; bias and ±1.96·SD limits are the exact construction. Whether the limits are clinically acceptable is a judgement the peer must supply — the plot shows agreement, not whether it’s good enough.