
482 Current Trends in Bayesian Methodology with Applications
and analyzed with Gaussian RRSM, arguing that estimates ar e still best lin-
ear unbiased predictor (BLUP), and co nditioning on the estimates of V and
ν
2
obtained from EM algorithm. When distributional ass umptions are wrong,
then likelihood based approaches, such as EM algorithm, suffer and do not
provide accurate estimates of model parameters and the predictions become
unreliable. One can use dis tribution free estimation techniques, such as binned
method of moments (MOM) estimation (see [6] and for more robust version
[15]). But binned estimatio n does not provide a ready positive-definite ...