Estimating the Negative Binomial Dispersion Parameter

Abstract

We proposed several estimators for the negative binomial dispersion parameter. The proposed estimators are combinations of existing ones using appropriate weights. We then compare, by simulation, the biases and efficiencies of the proposed estimators with those of the method of moments estimators and the maximum quasi-likelihood estimators. The simulation results indicate that the proposed estimators perform well in terms of biases and efficiencies in many instances. We conclude from this study that the combined estimators significantly reduce the mean bias of the estimators and more efficient than existing estimators. The relative efficiencies of all the combined estimators increases as the sample size mcreases.