Robust methods for survey research

Abstract

Classical robust statistics provides methods for dealing with outliers and these methods are now implemented in many software packages such as SAS and R. Complex surveys pose special problems that cannot be dealt with by the classical approach. Progress has been made in adapting robust methods to the context of survey research, and the methods can be used by experts in these methods and with sufficient programming skills. However, the methods and programmes have not yet reached a wider community. The prototypes have been programmed in the open source statistical environment and language R. The aim of the project is to give a wider community of researchers access to these specialised robust methods for survey data. The prototypes need to be tested, documented, packaged and published, and some new methods should be added. The result of the project will be an integrated suite of packages in R with documentation to enable non-specialists to use them. A vignette explaining the use of the packages to novice users will also be published. The survey research community would benefit from well-documented access to these new methods.