Pattern-mixture models for multivariate incomplete data

Pattern-mixture models are specified for multivariate data with missing values. The pattern-mixture models lead to maximum likelihood estimates with explicit forms, and Bayesian small-sample analyses are used. Then, identifying restrictions are applied to the maximum likelihood estimates and the asymptotic variance of the resulting function is computed directly. Lastly, the pattern-mixture models require prior information on the missing data mechanism in order to identify the parameters.