|
Abstract:
|
The paper introduces global-local shrinkage priors for random effects in small area estimation. These priors with earlier success in Bayesian multiple testing and Bayesian variable selection can capture potential sparsity, which in the present context means lack of significant contributions by many of the random effects. In addition, these priors can quantify and assess disparity among the random effects. The basic idea is to employ two levels of parameters to express prior variances of random effects. One is the global shrinkage parameter, which is common for all the random effects and introduces an overall shrinkage effect. The others are local shrinkage parameters, which act at individual levels and prevent overshrinkage. We show via data analysis and simulations that the global-local priors work as well or even better than some of the other priors proposed earlier.
|
