|
Abstract:
|
Misclassification of observational data is a common problem with adverse implications on the validity of results if not properly handled. Much research has been conducted when the response variable is misclassified, however less work has been done on the scenario in which the response variable is correlated over time and differentially misclassified. We extend previous frequentist work by investigating a Bayesian approach to four models with varying assumptions: a model that assumes differential misclassification is correlated within subjects and with the response model; a model that assumes independent response models but with differential misclassification correlated within the subject; a model with independent response processes and differential misclassification that is uncorrelated within the subject; and, lastly, a model that assumes independent response processes and that the nondifferential misclassification is uncorrelated within the subject. We compare the two approaches via estimation bias, precision, and the ability of each approach to select the “correct” model (assuming differential misclassification process is correlated with the response model and within subject).
|
