We consider the problem of estimating θ = (θ1,...,θp) under the weighted squared error loss when the observations xi, i = 1, 2, ..., p are independently from negative binomial distributions NB(ri, θi). There has been considerable interest in providing the Stein type estimators for negative binomial parameter θ. Hwang has shown that the UMVUE is inadmissible when p ≥ 3 under certain weighted squared error loss. It is therefore natural to ask about the admissibility of the MLE estimator. In this paper we address this problem and prove that, under the weighted squared error loss, where the weights are positive and bounded, the MLE is admissible for all p through a stepwise Bayes argument.
All Science Journal Classification (ASJC) codes
- Statistics and Probability
- Numerical Analysis
- Statistics, Probability and Uncertainty