Title: On the compatibility of experts' Lorenz curves with plausible assumptions on unit nonresponse in household surveys
Authors: Sarah Dahmann - The University of Melbourne (Australia)
Brendan Beare - The University of Sydney (Australia)
Rami Tabri - The University of Sydney (Australia) [presenting]
Abstract: Many datasets from household surveys used in studies on income inequality of a reference household population suffer from unit nonresponse. A widely shared view on such surveys is that the missing incomes and households tend to belong to the top of the income distribution, which contrasts with the standard practice by survey designers who implement so-called corrections that assume unit nonresponse as missing completely at random. Consequently, a variety of approaches that adjust observed income distributions have been proposed by distributional experts to mitigate the bias in inequality estimates from such surveys. However, as with this practice by survey designers, such expert adjustments may be incompatible with plausible assumptions on unit nonresponse, which raises the question of how to justify using a particular expert's adjustment method in practice. A statistical procedure is proposed that helps practitioners answer this question. We derive bounds on the population Lorenz curve under different plausible unit nonresponse mechanisms and propose a design-based inference procedure for assessing the compatibility of an expert's adjustment of the observed Lorenz curve with those mechanisms. Finally, these bounds and the inference procedure are illustrated using the Household, Income and Labour Dynamics in Australia Survey.