Estimating Poverty and Inequality from Grouped Data: How Well Do Parametric Methods Perform?

Abstract

Poverty and inequality are often estimated from grouped data as complete household surveys are neither always available to researchers nor easy to analyze. In this study we assess the performance of functional forms proposed by Kakwani (1980a) and Villasenor and Arnold (1989) to estimate the Lorenz curve from grouped data. The methods are implemented using the computational tools POVCAL and SimSIP, developed and distributed by the World Bank. To identify biases associated with these methods, we use unit data from several household surveys and theoretical distributions. We find that poverty and inequality are better estimated when the true distribution is unimodal than multimodal. For unimodal distributions, biases associated with poverty measures are rarely larger than one percentage point. For data from multi-peaked or heavily skewed distributions, the biases are likely to be higher and of unknown sign.

Keywords: grouped data, Lorenz curve, poverty, inequality, income distribution, POVCAL, SimSIP

JEL Classification: C13, C14, C15, C16, D31, D63, I32