Abstract
The agreement between quantitative replicate measures is usually assessed by using the intraclass correlation coefficient (ICC). Estimates of this coefficient can be influenced by outlying observations (the outlying mean or outlying variance of the measures). The purpose of this paper is to provide a procedure to detect these two types of outlier by means of approximate tests. These are derived from an analytical formula of the influence of a subject i0 on the maximum likelihood estimator of the ICC. Satterthwaite's approximation is used to derive approximate probability density functions, and Bonferroni's bound is then used to obtain two approximate tests allowing us to detect a potential outlier because of location slippage and a potential outlier because of dispersion slippage. Then, a Monte Carlo study is conducted to determine the type I error probabilities and the performance of the tests in the presence of a contaminant observation, as defined by Barnett and Lewis. The tests turn out to be rather conservative and their ability to detect a contaminant observation rapidly increases with the degree of contamination.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call