
492 Current Trends in Bayesian Methodology with Applications
exp onentia l distr ibution for W , good performance of the model fitting on the
simula ted data with the gamma and GIG distributions on W are indications
of another level of flexibility of our model. We generate the measurement
error ǫ from a Gaussian dis tribution with variance 0.05, with α = 10% con-
tamination, where the contaminated variance is 0.75. Finally, we calculate
the simulated data by Y = Xβ + SA
1
2
U + ǫ. Quantile plots for all three
simula tions showed significant non-Gaussian tail behaviors.
We randomly select 5, 000 observed locations and keep them aside a s a
test case. The