Comparison of paired ordinal data with mis-classification and covariates adjustment.

This article focuses on developing a statistical estimation and testing procedure for comparing matched-pair ordinal outcomes in studies affected by confounding factors and mis-classification, using data with partial validation. The method models the distribution of paired ordinal variables via correlated bivariate Gaussian latent variables, adjusts for confounders as covariates, and estimates mis-classification probabilities through a two-stage maximum likelihood approach. Simulation studies demonstrate the procedure's accuracy, robustness to various mis-classification patterns, sample sizes, and model misspecifications, as well as appropriate control of type I error and power. The approach is applied to malaria parasite prevalence data from the Garki Project, showing significant differences in infection densities between sexual and asexual forms of *Plasmodium falciparum* while adjusting for the number of microscope fields examined. The paper discusses potential extensions, including handling multiple fallible devices, mis-classified covariates, and more complex data structures.