Statistics Glossary → Proportional bias
Proportional biasconcept
A situation in which the difference between two measurement methods is not constant but grows or shrinks with the size of the quantity being measured.
In a Bland-Altman analysis the difference between the two methods is plotted against the mean of the two measurements. With fixed bias the mean difference is non-zero but stays the same across the measurement range. With proportional bias the difference systematically increases or decreases as the measured value grows, which appears in the plot as a sloping band of points. A linear regression of difference on mean shows whether the slope differs from zero.
When proportional bias is present, a single mean difference with fixed-width limits of agreement is misleading: it makes agreement look worse than it is at small values and better than it is at large values. The remedies are to express the difference as a percentage or on the log scale (as a ratio), or to model the limits of agreement as a function of the mean. If the two methods have very different measurement-error variances, an artificial slope between difference and mean can appear even without true proportional bias, so interpretation should be cautious.
An everyday example
A new finger-prick device is compared with the laboratory method for blood glucose. Around 80 mg/dL the methods differ by about 2 mg/dL, but around 300 mg/dL the difference rises to 20 mg/dL. Because the difference grows by roughly 7% with the level, a single 'mean difference' does not describe the device correctly; expressing the difference as a percentage is more meaningful.
In MerQur
In Bland-Altman agreement analysis, the 'Proportional-bias line' adds the difference ~ mean OLS line to the plot, and the slope test is always given in the text. The 'Difference scale (extra)' option (Raw difference, Percentage difference (% of mean), Ratio (log scale)) offers an additional analysis for differences that grow in proportion to the mean; on the log scale the limits of agreement are back-transformed to ratios. 'Additional output: Regression-based LoA' gives limits of agreement that depend on the measurement level when the mean or spread of the differences changes with it.
Analyses where this concept appears
Bland-Altman Agreement Analysis