Small area estimation of poverty in four West African countries by integrating survey and geospatial data
Conference
64th ISI World Statistics Congress - Ottawa, Canada
Format: IPS Abstract
Keywords: big data, small area estimation
Wednesday 19 July 2 p.m. - 3:40 p.m. (Canada/Eastern)
Abstract
This session discusses a recent experimental effort to produce small area estimates of poverty in Chad, Guinea, Mali, and Niger. Due to the absence of recent census data, we integrate survey and geospatial data which are used as covariates in model-based estimation. Using geospatial data increases the precision of small area estimates compared to direct estimation. This enables reporting poverty estimates more frequently and at more disaggregated administrative levels. The paper leverages recent census-based estimates in Burkina Faso to conduct a sensitivity analysis to evaluate the geospatial-based small area estimates using the same set of candidate geospatial variables and the same survey instrument. In Burkina Faso, the geospatial estimates are highly correlated with the census-based estimates in sampled areas but moderately correlated in non-sampled areas. Estimates generated using a household or grid-level model are more accurate than estimates generated using an area-level model, due to the use of more geographically disaggregated auxiliary data. The results demonstrate that in the absence of recent census data, small area estimation with publicly available geospatial covariates is feasible. In this context, it leads to large efficiency improvements compared to direct estimation and improves the timeliness of producing small area estimates. Estimates are consistent with census-based estimates in sampled areas but should be treated with a high degree of caution in non-sampled areas where conditioning on the sample observations is not possible.