Pattern mixture models for multivariate normal data: a simulation study

Little&Wand (1996) introduced the pattern mixture model for
wave nonresponse as a special case of multivariate normal
longitudinal data with fixed covariate matrix. This paper was the
theoretical foundation and the induce to investigate the pattern
mixture model compared with complete case analysis by means of
simulations. The main point of interest was the mean square error
of the estimated model parameters and the efficiency of the
estimations. To estimate the variance of the model parameters we
examine the Jackknife method. Parameter estimates by the pattern
mixture model are very satisfying under ignorable mechanism but
they have to be scanned carefully under nonignorable mechanism. The
Jackknife method seems to be, with restrictions, a good estimator
for the variance of the model parameters.