A0901
Title: Estimation of area-wise income distributions based on household-level grouped data
Authors: Yuki Kawakubo - Chiba University (Japan) [presenting]
Genya Kobayashi - Meiji University (Japan)
Abstract: From the household-level grouped income data, various characteristics in the area-wise income distributions are estimated. We observe which of the mutually exclusive intervals, separated by pre-specified thresholds, the sampled households' incomes belong. When estimating area-wise income distribution from such data, we face two problems: one is how to recover the underlying continuous variable (income) from the grouped data, and the other is that the estimation efficiency becomes poor when the sample size in each area is not sufficiently large. To address these issues, we treat the underlying household income as a latent variable, which is assumed to follow a mixed-effects model that incorporates household-level and area-level auxiliary variables and random effects as area effects. The effectiveness of using mixed-effects models has been actively studied on small area estimation. By predicting the latent variable for each household based on the observations as grouped data and auxiliary variables, we estimate various characteristics in the area-wise income distribution. This method is applied to Japanese income data to estimate not only the mean or median income, but also the Gini coefficient and several poverty indices for each area.
