Tutorial 'Small Area Estimation with R'
Virgilio Gómez Rubio (Universidad de Castilla-La Mancha, Spain)
Background
Small Area Estimation deals with the problem of providing estimates of some parameters of interest for sub-populations. These techniques are often required when data from a larger survey are available but parameter estimates are required at a smaller level. Typical problems include the estimation of income per household or unemployment in sub-populations from survey data.
This is an important topic that has not been widely covered in previous useR! conferences. Furthermore, there is a shortage of training materials on Small Area Estimation with R. This tutorial aims at providing an updated version of the tutorial presented at useR! 2008 by including new R packages in the 'Official Statistics and Survey Methodology' Task View.
Tutorial Overview
The tutorial will introduce different types of statistical methods for the analysis of survey data to produce estimates for small domains (sometimes termed 'small areas'). This will include design-based estimators, that are only based on the study design and observed data, and model-based estimators, that rely on an underlying model to provide estimates. The tutorial will cover frequentist and Bayesian inference for Small Area Estimation. All methods will be accompanied by several examples that attendants will be able to reproduce.
This tutorial will be roughly based on the tutorial presented at useR! 2008 but will include updated materials. In particular, it will cover new R packages that have appeared since then.
All tutorial materials will be available from a Github repository.
Detailed Outline
The tutorial will use some case studies to present the main statistical methods in Small Area Estimation. This includes:
- Short introduction to survey sampling strategies with R: simple random sampling, systematic sampling, clustered sampling, two-stage sampling
- Design based-estimators: Horvitz-Thompson, generalized regression (GREG) and calibration estimators
- Model-based estimators: Fay-Herriott estimator, linear regression for area and unit level models
- Synthetic and composite estimators
- Mixed-effects models: area and unit level models with random effects, EBLUP estimators
- Models with spatial random effects: spatial EBLUP for area and unit level models
- Bayesian inference for Small Area Estimation: area and unit level models
- Non-linear models: disease mapping, estimation of unemployment
Background Knowledge
General knowledge on survey data analysis and mixed-effects models would be useful. Attendants are assumed to be R users.
Potential Attendees
This tutorial may be of interest to practitioners and researchers at national offices for statistics, universities and research institutions working on Small Area Estimation.
##Required software
R packages
You can install the required R packages by running file install.R.
###WinBUGS
WinBUGS is required to run some of the models presented in the tutorial. You can download from here (read very carefully the instructions to install it).
###INLA
R-INLA is also required for some of the models. You can download it from here. You can also install it by running the following:
install.packages("INLA", repos="https://www.math.ntnu.no/inla/R/stable")###SAE2 package
The SAE2 package contains some functions to fit some of the models in the tutorial. You can install it from github:
devtools::install_github("becarioprecario/SAE2")