Robust Small Area Estimation and Oversampling in the Estimation of Poverty Indicators

Abstract

There has been rising interest in research on poverty mapping over the last decade, with the European Union proposing a core of statistical indicators on poverty commonly known as Laeken Indicators. They include the incidence and the intensity of poverty for a set of domains (e.g. young people, unemployed people). The EU-SILC (European Union - Statistics on Income and Living Conditions) survey represents the most important source of information to estimate these poverty indicators at national or regional level (NUTS 1-2 level). However, local policy makers also require statistics on poverty and living conditions at lower geographical/domain levels, but estimating poverty indicators directly from EU-SILC for these domains often leads to inaccurate estimates. To overcome this problem there are two main strategies: i. increasing the sample size of EU-SILC so that direct estimates become reliable and ii. resort to small area estimation techniques. In this paper we compare these two alternatives: with the availability of an oversampling of the EU-SILC survey for the province of Pisa, obtained as a side result of the SAMPLE project (Small Area Methods for Poverty and Living Conditions, http://www.sample-project.eu/), we can compute reliable direct estimates that can be compared to small area estimates computed under the M-quantile approach. Results show that the M-quantile small area estimates are comparable in terms of efficiency and precision to direct estimates using oversample data. Moreover, considering the oversample estimates as a benchmark, we show how direct estimates computed without the oversample have larger errors as well as larger estimated mean squared errors than corresponding M-quantile estimates.