Two-sample estimation of poverty rates for disabled people : an application to Tanzania

Estimating poverty measures for disabled people in developing countries is difficult, partly because relevant data are not available. We develop two methods to estimate poverty by the disability status of the household head. We extend the small-area estimation proposed by Elbers, Lanjouw and Lanjouw (2002, 2003) so that we can run a regression on head's disability status even when such information is unavailable in the survey. We do so by aggregation and by moment adjusted two sample instrumental variable estimation. Our results from Tanzania show that both methods work well, and that disability is indeed associated with poverty.