New imputation procedures in the measurement of inequality, growth, and poverty in Brazil

This paper develops a new imputation methodology applied to missing incomes values in PNAD. PNAD is the main Brazilian household survey, but it has no imputation. The imputation process starts by fitting regression models applied to different income sources considering the complex sampling design of the survey. Later this procedure is combined with stochastic methods. In 2015, 2.5 per cent of the sample had per capita incomes imputed, resulting in slightly higher levels of inequality. Inequality and poverty changes were not affected by imputation. We took advantage of the methodology proposed to input rents and to preserve pressure points in income distributions associated with Brazilian institutional features such as minimum wages.