Addressing Incomplete Data in Two-Way Contingency Tables with Three-Level Variables

Handling incomplete, imprecise, or missing observations in two-way contingency tables presents a challenge in accurately assigning uncertain data points to appropriate groups. Various approaches have been proposed to address this issue. Traditionally, the Expectation and Maximization (EM) algorithm has been utilized to estimate the group of incomplete observations. Subsequently, two alternative methods, derived from modifications to the formula used in the E-step of the algorithm, were introduced. However, these methods have mainly been evaluated in scenarios where categorical variables have two levels. This study aims to broaden the comparison of these methods to contingency tables with categorical variables featuring three levels. Through a simulation study, our findings indicate that while the classical EM algorithm generally performs well, there are specific parameter configurations where alternative methods demonstrate superior performance.