4. In Example 9.6 (spatially varying predictor effects) try instead a model with spatially fixed predictor effects, but a bivariate spatial error, as in (9.17.2) or (9.17.3), combined with spatially unstructured effects ui1 for males and ui2 for females, as in (9.17.1). The latter may be independent between the two outcomes or also follow a multivariate prior. How does this compare with the spatially varying predictor model in predictive compatibility with the data (replicate data reproducing the actual data) and in terms of fit as measured by the DIC?