EM Algorithm For Binary Markov Chains Of Longitudinal Data With Missing Data

In this paper,the response variables are discussed as binary variables and meet the needs of firstorder Markov;and the graph model is used to describe longitudinal data which include missing data. The missing mechanisms discussed in this paper is a kind of cannot-be-ignored-missing model,i.e. the individuals in every moment responding or not relys on its current potential response,using EM algorithm to estimates the parameters.This EM algorithm is simple in calculating when the model is based on one or two points in time.However,if there are three or more points in time,this EM algorithm is complicated and lacks a regular pattern.Therefore,this paper intends to find out a simple algorithm to give it a regular pattern and apply this algorithm to a series of medical data containing missing data to show its superiority.