The LORELIA Residual Test - A New Outlier Identification Test for Method Comparison Studies

Abstract

Method comparison studies are performed in order to prove equivalence between two measurement series resulting from two measurement instruments. The presence of outliers in method coparison studies can indicate problems and errors in the measurement process such as contaminated samples, wrong measurement or wrong callibrated instruments. Therefore, it is of great importance to identify existing outliers and to examine and report them carefully. Datasets in method comparison studies often show a very inhomogeneous sample distribution and heteroscedastic random error variances. Common outlier tests proposed in the literature however are based on the assumption of a homogeneous sample distribution and homoscedastic random error variances. In this work, a new outlier test for methods comparison studies based on robust linear regression is proposed to overcome these special problems. Outlier limits are constructed as local confidence intervals for the orthogonal residuals without making assumptions on the underlying residual variance model. The new LORELIA (LOcal RELIAbility) Residual Test is based on a local and robust residual variance estimator, given as a weighted sum of the observed residuals.