[Back]

A. Kreuzmann, S. Pannier, N. Rojas, T. Schmid, M. Templ, N. Tzavidis:
"Small Area Estimation in R with Application to Mexican Income Data";
Talk: NTTS 2017, Brussels (invited); 2017-03-15; in: "NTTS 2017 Proceedings", (2017), 1 - 5.

In the last decades policy decisions are often based on statistical measures. The more detailed this information is, the better is the basis for targeting policies and evaluating policy programs. For instance, the United Nations suggest more disaggregation of statistical indicators for monitoring their Sustainable Development Goals and also the number of National Statistical Institutes (NSIs) that notice the need of more disaggregated statistics is increasing. Dimensions for disaggregation can be characteristics of the individuals or households like sex, age or ethnicity, economic activity or spatial dimensions like metropolitan areas or districts. Primary data sources for variables that are used to estimate statistical indicators are national household surveys. However, sample sizes are usually small or even zero at disaggregated levels. Therefore, direct estimators based only on survey data can be unreliable or not available for small domains. While the option of more specific surveys is costly, model-based methodologies for dealing with small sample sizes can help to obtain reliable estimates for small domains. The so-called Small Area Estimation (SAE) methods [1,2] link survey data that is only available for a proportion of households with administrative or census data available for all households in the area of interest. Even though a wide range of SAE methods is proposed by academic researchers, these are, so far, applied only by a small number of NSIs or other practitioners like the World Bank. This gap between theoretical possibilities and practical application can have several reasons. One reason can be the lack of suitable statistical software. The free software environment R helps to counteract this issue since researchers can make their codes available to the public via packages. Thus, new methods can reach the practitioner faster than with non-free software. The next two sections summarize which packages are already available and what could be improved in the future.

Small Area Estimation, R

http://publik.tuwien.ac.at/files/publik_258456.docx