Deming Regression

Deming regression fits a straight line when both variables contain measurement error, using the ratio of the two error variances to weight vertical and horizontal deviations.

Deming 회귀는 두 변수 모두에 측정오차가 있을 때 오차분산의 비를 반영해 직선을 적합하는 방법입니다.

I am comparing two measurement methods that both have error; how do I test for proportional and constant bias between them?

둘 다 오차가 있는 두 측정법을 비교할 때 비례 편향과 고정 편향을 어떻게 검정하나요?

Deming regression estimates the line relating two error-prone measurements by minimizing deviations in both directions, weighted by the ratio of their error variances. Agreement is judged by whether the slope's confidence interval contains 1 and the intercept's contains 0.

When to use Deming Regression

• Two methods measuring the same quantity are compared and both have random error
• The ratio of error variances is known or can be estimated from replicates
• Errors are roughly normal without many outliers

When to use something else

• Data contain outliers or the error variance ratio is unknown; use Passing-Bablok regression instead
• The predictor is measured without error; use simple linear regression instead

Assumptions and requirements

• Both variables have independent, normally distributed measurement errors
• The error variance ratio λ is correctly specified (λ = 1, orthogonal regression, unless set)
• The relationship between the true values is linear
• Error variances are constant across the range

What the analysis reports

• Slope and intercept with jackknife standard errors and confidence intervals
• Tests of slope = 1 and intercept = 0
• Error variance ratio used
• Ordinary least squares slope for comparison

How to report it

Deming regression (λ = [value]) of [method B] on [method A] gave a slope of [value], 95% CI [[value], [value]], and an intercept of [value], 95% CI [[value], [value]], indicating no proportional or constant bias.

Related methods

• Passing-Bablok Regression
• Simple Linear Regression

Run Deming Regression on Quorant