ABSTRACT
There are many missing value or incomplete data problems in the statistical literature. The EM ( expectation-maximization) algorithm, formalized by Dempster et al. (1977), has become a popular tool for handling such problems. Surprisingly, many important inference problems in statistics, such as latent variable and random parameter models, turn out to be solvable by EM when they are formulated as missing value problems. In these applications, it is not the data collected from the field that are in real sense m1ssmg. it is the hidden parameters or variables which are systematically not observed. Here, we deal mainly with this kind of structural missing data.
