Efficient targeting of aid relies on detailed information. The complication is that even large scale national sample surveys are usually not accurate enough to provide this detail. Small area estimation is a statistical technique developed over the last twenty years or so which can improve accuracy of surveys by using statistical modelling. A particular type of small area estimation technique, which is generically called poverty mapping and links a survey with a census at unit record level, is now often used for estimating and mapping poverty estimates at a fine level. This method has now been applied in more than 70 countries. Where there is no recent census, alternative methods exist. Small area estimation can also be extended, for example to estimating malnutrition including stunting, underweight and wasting in children under five years of age.
Three unit-level SAE techniques: the method of Elbers, Lanjouw, and Lanjouw (2003) also known as ELL or World Bank method, the Empirical Best Prediction (EBP) method of Molina and Rao (2010), and the M-Quantile (MQ) method of Tzavidis et al. (2008), have been widely used to estimate the micro-level FGT poverty indicators: poverty incidence, gap and severity (Foster et al., 1984). The three methods have in common that they use both unit level survey data and (possibly model-based) unit level census data. However, they differ in their applicability because real data sets do not always follow their underlying assumptions. The performance of these three methods is compared in terms of small area poverty estimates and their standard errors. The effects of using a model-based unit record census data reconstructed from available cross-tabulations are discussed, as are the effects of small area-heterogeneity and cluster-heterogeneity in the over-arching superpopulation model. The three methods also have variants. A three-level nested-error regression model-based ELL method is applied for comparison with the standard two-level model-based ELL method which does not contain a random component at small area level, and with EBP and MQ. A comparison study uses a simulation based on 2003 data from Bangladesh. An important finding as outlined in Das and Haslett (2019) is that the number of small areas for which a method is able to produce sufficiently accurate estimates is more often driven by the type of data available than by the model per se.
This talk covers broad context, what small area estimation is and the level at which it can work, discussion of alternative techniques when survey and census unit record data is available, and why small area estimation is useful.
Stephen Haslett is Professor Emeritus of Statistics at Massey University in Aotearoa New Zealand, Visiting Research Fellow in the Research School of Finance, Actuarial Studies and Statistics at the ANU, Professorial Fellow at the National Institute for Applied Statistics Research Australia at the University of Wollongong and Director of Systemetrics Research Associates Ltd.