Income Mis-Measurement and the Estimation of Poverty Rates An Analysis of Income Poverty in Albania

We investigate the reliability of estimated income poverty profiles for Albanian survey data. We find evidence that a significant number of households with low reported incomes have relatively high living standards and are consequently misclassified as poor. We extend the theory of contaminated distributions to incorporate direct measures of well-being as indicators of data contamination, and develop a new nonparametric approach for constructing bounds on conditional poverty rates. We find very large upward biases in measured income poverty under the assumption of independence between living standards and the misreporting propensity, but a wide range of uncertainty under more general conditions. NON-TECHNICAL SUMMARY Empirical analyses of income poverty are often used to guide the design of welfare policies such as social assistance programmes and employment subsidies. The typical poverty study is based on survey data covering a large number of randomly-selected households, whose members are interviewed in detail about their incomes and other circumstances. This information is first used to classify households as ‘poor’ or ‘non-poor’ in relation to a poverty line, then statistical methods are used to analyse the way that poverty varies between different social groups. Groups which are identified as particularly vulnerable to poverty then become very important for anti-poverty policy. An accepted problem with this type of analysis is the accurate survey measurement of household income. Researchers have sometimes reported significant numbers of survey respondents with implausibly low measured incomes, in relation to other indicators of the household’s standard of living. We investigate this issue, using data from recent cross-section and panel surveys of Albanian households. We find very strong evidence that measured income greatly understates true living standards for a substantial group of households below the poverty line in the 2002 Albanian Living Standards Measurement Survey. These problem cases are identified in two ways: through a discrepancy between the position of the household in the measured income distribution and its position in the distribution of measured consumption expenditure; and through a discrepancy between its income and other nonfinancial indicators of living standards (such as the number of durables owned or the household’s own subjective assessment of well-being). Using a smaller panel dataset for 2002 and the later years 2003-4 which have no consumption information, we find that the income measurement problem is most serious in 2002 and that there is only a weak tendency for households who ‘misreported’ their incomes in 2002 also to show signs of mis-measurement in later years. Thus, these difficulties with income measurement do not appear to be very persistent over time, at the level of individual households. The paper also develops a new method of taking account of this measurement problem in making poverty analyses; this is applied to the 2002 cross-section survey. We find that, under the (strong) assumption that income mis-measurement is unrelated to living standards, ‘true’ poverty rates are clearly much lower than measured rates. However, if this assumption is relaxed, to allow the possibility that understatement of income is mainly confined to households with high living standards, then we can only draw a much weaker conclusion that there is a very wide range of uncertainty associated with measured poverty rates, and measured poverty may therefore be much too high.