S3Ri seminar series - next session Ordered small area estimation using a perturbation random effects model Thurs 31st Oct, 2.15pm, 54/10031 Prof. Li-Chun Zhang (S3RI , Division of Social Statistics & Demography) We develop a perturbation random-effects model for estimation of the order statistics of small area means (or proportions). The approach provides a general way to accommodate the first 3 (or 4) moments of the random effects no matter what the true underlying distribution may be. It is more appropriate than the constrained empirical Bayes (CEB) approach in cases of skewed random effects, while more flexible than the typical fully parametric SAE models such as Gaussian-Gaussian, Gaussian-log-normal, beta-binomial, etc.. Utilizing well-known skewed parametric (e.g. normal) distributions does not seem straight-forward because the observations (or sample statistics) consist of two random components, both of which may have skewed distributions, such that in general one would need to handle the compound of, say, two skewed normal distributions. In comparison, it will be shown that applying the proposed perturbation model to two- or multiple-level random components is quite manageable.
Posted on: Tue, 29 Oct 2013 16:18:05 +0000