The R package emdi offers a methodological and computational framework for the estimation of regionally disaggregated indicators using small area estimation methods and provides tools for assessing, processing and presenting the results. A range of indicators that includes the mean of the target variable, the quantiles of its distribution and complex, non-linear indicators or customized indicators can be estimated simultaneously using direct estimation and the empirical best predictor (EBP) approach (Molina and Rao 2010). In the application presented in this paper package emdi is used for estimating inequality indicators and the median of the income distributions for small areas in Austria. Because the EBP approach relies on the normality of the mixed model error terms, the user is further assisted by an automatic selection of data-driven transformation parameters. Estimating the uncertainty of small area estimates (using a mean squared error - MSE measure) is achieved by using both parametric bootstrap and semi-parametric wild bootstrap. The additional uncertainty due to the estimation of the transformation parameter is also captured in MSE estimation. The semi-parametric wild bootstrap further protects the user against departures from the assumptions of the mixed model in particular, those of the unit-level error term. The bootstrap schemes are facilitated by computationally effcient code that uses parallel computing. The package supports the users beyond the production of small area estimates. Firstly, tools are provided for exploring the structure of the data and for diagnostic analysis of the model assumptions. Secondly, tools that allow the spatial mapping of the estimates enable the user to create high quality visualizations. Thirdly, results and model summaries can be exported to Excel spreadsheets for further reporting purposes.
offcial statistics parallel computation small area estimation visualization