Passing-Bablok Regression

Passing-Bablok regression is a nonparametric method-comparison line whose slope is a shifted median of all pairwise slopes, robust to outliers and requiring no distributional assumption about errors.

Passing-Bablok 회귀는 모든 점 쌍 기울기의 이동 중앙값으로 직선을 정하는 비모수 측정법 비교 회귀로, 이상값에 강하고 오차 분포를 가정하지 않습니다.

How do I compare a new laboratory method with a reference method without assuming normal errors?

정규 오차를 가정하지 않고 새 검사법과 기준 검사법을 비교하려면 어떻게 하나요?

Passing-Bablok regression estimates the slope as a shifted median of all pairwise slopes and the intercept as the median of the remaining differences, so a few outliers have little influence. Confidence intervals for slope and intercept test proportional and constant bias, and a CUSUM test checks whether a straight line is adequate.

When to use Passing-Bablok Regression

• Two methods measuring the same quantity are compared on continuous data
• Errors may be non-normal or contain outliers
• The two measurements are highly and positively correlated

When to use something else

• Error variances are known and errors are normal; use Deming regression instead
• You want to describe agreement of individual differences; use a Bland-Altman agreement analysis instead

Assumptions and requirements

• The two measurements are linearly related
• Measurements are continuous and positively correlated
• Error distributions of both methods have similar shape
• Sample size is adequate, ideally 30 or more (at most 4,000 pairs of measurements)

What the analysis reports

• Slope and intercept with confidence intervals
• Whether the slope interval contains 1 and the intercept interval contains 0
• CUSUM linearity test statistic and p value
• Kendall's tau between methods

How to report it

Passing-Bablok regression of [method B] on [method A] (n = [value]) gave a slope of [value], 95% CI [[value], [value]], and an intercept of [value], 95% CI [[value], [value]]; the CUSUM test showed no deviation from linearity (p = [value]).

Related methods

• Deming Regression
• Theil-Sen Regression

Run Passing-Bablok Regression on Quorant