Misspecification and sample size: Some effects on posterior predictive distributions in parametric nonparametric approaches

The paper compares parametric and nonparametric posterior predictive distributions. It investigates the robustness aspect of the parametric method in case of specification error and analyses the efficiency aspect of the nonparametric method, in relation to various sample sizes. Simulation experiments analyse cases where the specification errors turn on thickness of the tails, asymmetry and plurimodality of the distribution. The generated and observed samples are of sizes between 100 and 1000.