Question 1

What is Passing-Bablok regression?

Accepted Answer

Passing-Bablok regression is a non-parametric method for comparing two measurement methods. Unlike ordinary least squares, it does not assume that either method is the reference standard — both have measurement error. The slope and intercept are computed from the median of all pairwise slopes, making it robust to outliers. If the 95% CI for the slope includes 1, there is no proportional bias. If the CI for the intercept includes 0, there is no constant bias.

Question 2

What is Deming regression?

Accepted Answer

Deming regression is a parametric method for comparing two measurement methods when both have measurement error. Unlike ordinary least squares (which assumes X is measured without error), Deming regression accounts for error in both variables. The parameter $\lambda$ (lambda) is the ratio of error variances: $\lambda$ = σ²_x / σ²_y. When $\lambda$ = 1 (default), both methods are assumed to have equal measurement variability.

Question 3

When should I use Passing-Bablok vs Deming?

Accepted Answer

Use Passing-Bablok when: (1) you cannot assume normality of residuals, (2) outliers may be present, or (3) you want a non-parametric approach. Use Deming when: (1) residuals are approximately normal, (2) you know the error variance ratio, or (3) you need parametric CIs. In practice, report both — they should give similar results if assumptions hold.

Question 4

What does the Bland-Altman plot show?

Accepted Answer

A Bland-Altman plot visualizes agreement between two methods by plotting the difference (Y − X) against the mean ((X + Y) / 2). The horizontal line is the mean bias. The dashed lines are the 95% limits of agreement (mean $\pm$ 1.96 SD). If differences are randomly scattered within the LOA with no trend, the methods agree acceptably.

Question 5

How do I interpret slope and intercept?

Accepted Answer

If slope = 1 and intercept = 0, the methods give identical results on average. Slope ≠ 1 indicates proportional bias (one method reads systematically higher/lower by a factor). Intercept ≠ 0 indicates constant bias (one method reads consistently higher/lower by a fixed amount). Use confidence intervals — not point estimates — to judge significance.

Method Comparison Analyzer

Upload Paired Measurements

Method Comparison Fundamentals

Frequently Asked Questions

Next Steps

Laboratory Equipment

Diagnostic Test Calculator — Binary accuracy measures

General Stats Calculator — t-tests and correlation

ROC/AUC Calculator — Threshold optimization

Need Help?