POVERTY COMPARISONS WITH ABSOLUTE POVERTY LINES ESTIMATED FROM SURVEY DATA

The objective of measuring poverty is usually to make comparisons over time or between two or more groups. Common statistical inference methods are used to determine whether an apparent difference in measured poverty is statistically significant. Studies of relative poverty have long recognized that when the poverty line is calculated from sample survey data, both the variance of the poverty line and the variance of the welfare metric contribute to the variance of the poverty estimate. In contrast, studies using absolute poverty lines have ignored the poverty line variance, even when the poverty lines are estimated from sample survey data. Including the poverty line variance could either reduce or increase the precision of poverty estimates, depending on the specific characteristics of the data. This paper presents a general procedure for estimating the standard error of poverty measures when the poverty line is estimated from survey data. Based on bootstrap methods, the approach can be used for a wide range of poverty measures and methods for estimating poverty lines. The method is applied to recent household survey data from Mozambique. When the sampling variance of the poverty line is taken into account, the estimated standard errors of Foster–Greer–Thorbecke and Watts poverty measures increase by 15–30 percent at the national level, with considerable variability at lower levels of aggregation.