The conventional form of statistical simulation proceeds by selecting a few\nmodels and generating hundreds or thousands of data sets from each model.\nThis article investigates a different approach, called BayesSim, that generates\nhundreds or thousands of models from a prior distribution, but only one (or a\nfew) data sets from each model. Suppose that the performance of estimators in\na parametric model is of interest. Smoothing methods can be applied to\nBayesSim output to investigate how estimation error varies as a function of\nthe parameters. In this way inferences about the relative merits of the estimators\ncan be made over essentially the entire parameter space , as opposed to a\nfew parameter configurations as in the conventional approach. Two examples\nillustrate the methodology: One involving the skew-normal distribution and\nthe other nonparametric goodness-of-fit tests.
Loading....