The Analysis of Misclassified Ordinal Data from Designed Experiments

Standard analyses of ordinal data from designed experiments assume that the data are not misclassified. This article considers the impact of ignoring misclassification and presents a Bayesian approach to account for it. Misclassification depends on the probabilities of misclassifying an item with a given true category to the other categories. Both the cases of known and estimated misclassification probabilities are considered. The analysis methodology is illustrated with data from a real experiment and is assessed using a simulation study.