Bayesian geoadditive seemingly unrelated regression (Q1887221)

The authors demonstrated the power and flexibility of the Bayesian approach to fix complex regression models. They presented Bayesian semiparametric geoadditive seemingly unrelated regression (SUR) models where predictors incorporate linear parametric components, additive components for nonlinear effects of metric covariates and spatial components -- for geographical effects. This approach extends previous work by \textit{L. Fahrmeir} and \textit{S. Lang} [J. R. Stat. Soc., Ser. C 50, No. 1, 201--220 (2001); Ann. Inst. Stat. Math. 53, No. 1, 11--30 (2001; Zbl 0995.62098)]. The proposed methodology is based on appropriate smoothness priors which allow different forms and degrees of smoothness in a general framework. Inference is fully Bayesian and uses recent Markov chain Monte Carlo techniques for estimation. Advantages of P-splines are also exploited. Applications to real data are discussed.