Representativeness of Top Expenditures in Arab Region Household Surveys

Abstract

This study investigates evidence of measurement errors among households with the highest expenditures in eleven harmonized national household-expenditure surveys from five Arab countries: Egypt 2008, 2010 and 2012, Jordan 2006 and 2010, Palestine 2007, 2010 and 2011, Sudan 2009, and Tunisia 2005 and 2010. Distribution of expenditures and the corresponding Gini index of inequality are corrected by replacing top expenditures with values predicted under smooth parametric distributions. Pareto distribution among expenditures classified as top expenditures, and generalized beta distribution of type II on the entire national distribution of household expenditures are fitted and used as alternative specifications for replacing top expenditures. Across the eleven surveys, inequality of expenditures is found to be modest, and neither parametric correction performed debunks this conclusion. Gini coefficient is consistently between 29 and 31 in Egyptian surveys, and between 36 and 41 in the rest of surveys. Jordanian and Palestinian 2010 data are found to include clear outliers that influence estimates of inequality upward relative to statistics expected under parametric distributions. Other surveys exhibit better representativeness for the expected distribution of expenditures that may be approximated by parametric distributions. Pareto law appears to hold well among top expenditures in Egypt and Palestine, justifying Pareto specification. In Jordan, Sudan and Tunisia, however, a four-parameter generalized beta distribution appears more appropriate to model top expenditures. Ginis estimated under generalized beta distribution are somewhat higher than nonparametric or Pareto-distribution Ginis in Jordan and Palestine, similar in Egypt, and lower in Sudan and Tunisia. These patterns are consistent across alternative sample delineations and across multiple waves of national surveys. Nevertheless, the alternative estimates of Ginis are within one another’s confidence intervals, implying that neither set of estimates is clearly preferred. Whether nonparametric, or Pareto or generalized-beta parametric Ginis are closest to true statistics remains a question for future research.