Inhaltszusammenfassung

In a Bayesian setting prior distributions on the variance components have to be specified for random effect models. Due to the non-availability of this in- formation, the variance distributions are frequently modeled using improper priors. Therefore, the question of integrability of the posterior distribution arises. Natarajan and McCulloch (1995) stated a theorem which ensures integrability (propriety) of the posterior of the variance component for a class of mixed models for binomial respo...In a Bayesian setting prior distributions on the variance components have to be specified for random effect models. Due to the non-availability of this in- formation, the variance distributions are frequently modeled using improper priors. Therefore, the question of integrability of the posterior distribution arises. Natarajan and McCulloch (1995) stated a theorem which ensures integrability (propriety) of the posterior of the variance component for a class of mixed models for binomial responses with independent random effects. The independency of random effects is generally not valid in applications to economics, social or geographical sciences where typically spatial correlations are observed. This additional information is included into the model through conditional autoregressive dependent random effects. The current paper extends the Natarajan and McCulloch (1995) theorem to spatial random effects. An application to unemployment data in Germany demonstrates the potential benefits of this extension.» weiterlesen» einklappen

Autoren

Klassifikation

DDC Sachgruppe:
Statistik