Calculating Comparable Statistics from Incomparable Surveys, with an Application to Poverty in India

We develop an intuitive and easily implemented procedure to recover comparability over time of statistics computed using databases made incomparable by changes in survey design. Our methodology can be adopted whenever the statistic of interest satisfies a certain simple moment condition. The moment condition is satisfied by many interesting economic indicators, including a broad range of poverty and inequality measures. The procedure we propose requires the existence of a set of auxiliary variables whose reports are not affected by the different survey design, and whose relation with the main variable of interest is stable across the surveys. The adjusted estimates can be recovered by using a two-step method of moments framework. Root-n consistency follows easily under regularity conditions. Because most household surveys adopt a multi-stage design, we provide expressions for the asymptotic variance which are robust to the presence of clustering and stratification. We use our adjustment procedure to estimate poverty counts from the 55th Round of the Indian National Sample Survey, a large household survey carried out in 1999-2000. Due to important changes in the adopted questionnaire the unadjusted figures are likely to understate poverty relative to the previous rounds. We provide evidence supporting the plausibility of the identifying assumptions and we conclude that most of the very large reduction in poverty implied by the unadjusted figures is real