Features of statistical analysis of quantitative data obtained from fellow eyes, nonparametric tests

Abstract

Purpose. To compare various approaches to statistical analysis of fellow eyes and to describe correct analysis using nonparametric tests with R software. Material and methods. Various approaches to statistical analysis of fellow eyes are analyzed. Three general strategies are stated: 1) inclusion of both eyes in the same group using standard methods of statistical analysis; 2) inclusion in the same group only one eye of each subject using standard statistical methods; 3) inclusion of the both eyes in the same group using advanced statistical methods accounting correlation between fellow eyes. Results. The first approach leads to a significant underestimation of p-values when comparing groups and increases the risk of rejecting the correct null hypothesis. The second approach does not allow taking into account all available data and decreases the statistical power of a study. The third approach uses all available data and allows making valid inferences. Conclusion. Unreasonable use of standard statistical approaches for analyses quantitative data of fellow eyes leads to a significant distortion of p-values, does not allow taking into account all the material. Best practices for such situations are advanced statistical techniques accounting correlations between fellow eyes, such as the RGL and DS methods of package clusrank for R language. Key words: clustered data, fellow eyes, Mann–Whitney U test, Wilcoxon test, R software environment, clusrank package