A Note About Classical Test Statistics in Incomplete-Data Models

In econometrical literature, various incomplete-data models have been developed to reflect economic phenomena. For incomplete-data models, it is not easy to construct test statistics because the observed log-likelihood function often involves intractable integrals which make it very difficult to evaluate. In this paper, within the well-known EM framework, the new forms of three commonly-used classical test statistics based on maximum likelihood estimators are discussed. The main idea is to replace the observed log-likelihood function with the complete-data log-likelihood function because it is relatively easier to evaluate in most situations