emdi - The R package for estimating and mapping disaggregated indicators
11 Jul 2021The R package emdi provides a large collection of small area estimation methods. The package documentation and the vignettes serve the purpose to explain the available methods and/or to shows the functionality. This page will offer small examples for specific applications.
For the most recent technical and methodological updates, emdi is available on GitHub.
Overview
The methods implemented in emdi can be classified into following categories:
- Direct estimation
- Area-level models
- Unit-level models
While direct estimation is provided, the focus of the package are the small area estimation methods. Therefore, the following descriptions will focus on the small area estimation methods.
Area-level models
With function fh, package emdi provides a large range of area-level models basically following the idea of Fay and Herriot (1979). The function includes following modelling options:
- Basic area-level model
- Transformations: log and arcsin
- Spatial correlation
- Robust estimation
- Measurement errors
Unit-level models
Most unit-level models are based on the model proposed by Battese, Harter and Fuller (1988). So far, the package provides some variations of the empirical best prediction (EBP) approach proposed by Molina and Rao (2010):
- Census EBP approach (mentioned in Guadarrama et al. 2016)
- Data-driven transformations for the EBP (Rojas et al. 2020)
- EBP under informative sampling (Guadarrama et al. 2018)