2. In Example 4.3 (Hald data), compare the predictive least squares criterion (e.g. Gelfand and Ghosh, 1998)

i (yi − ynew,i )2 under models (x1, x2, x3) and (x1, x2, x5) when ynew are sampled under separate estimations of each model. Also obtain a pseudomarginal likelihood for each model from single case omission. How do these approaches compare in terms of model choice?