A0285
Title: Semiparametric inference on inequality measures with nonignorable nonresponse
Authors: Chunlin Wang - Xiamen University (China) [presenting]
Abstract: Measuring inequality of economic variables, such as income, is vital in economics, social science and statistics. Reliable estimation and inference of inequality measures can provide insights and have crucial implications in policy-making procedures. However, income survey data inevitably suffer from nonignorable nonresponse, in the sense that the response probabilities depend on the missing income values. This creates challenges for the estimation of inequality measures, in particular, the model identifiability and selection bias issues. To address these issues, we exploit the commonly available callback data in income surveys and propose a semiparametric modeling strategy. We develop a semi-parametric full-likelihood approach for making inference on inequality measures with nonignorable nonresponse. We establish large-sample properties of the proposed estimators of the inequality measures, including the quantile, Gini index, Theil index and the generalized entropy class. Additionally, we devise a stable expectation-maximization algorithm for efficient computation. The simulation results and a real income data example demonstrate that proposed method corrects the bias of the estimated inequality measures and leads to reliable inference results.
