A Self Documenting Programming Environment for Weighting.pdf

1A Self Documenting Programming Environment for Weighting Wilfried Grossmann and Petra Ofner Department of Statistics and Decision Support Systems, University Vienna, A-1010 Vienna, Universit?tsstrasse 5, Austria Abstract. The paper presents a proposal for a statistical computing environment for weighting, which processes statistical data and description of data simultaneously. The developed model is of interest also for other types of computing in official statistics. Keywords. Programming environments, Weighting, Metadata 1 Introduction Weighting plays an important role in survey computing, mainly for improving quality of estimates with respect to bias and variance but also for adjusting of estimates in such way that they match controls. Various weighting algorithms have been developed, which need besides the actual data additional information about the data, in particular information about underlying populations and information about the sampling design, stored frequently separate from survey data. Hence, only data with computed weights are delivered outside the production environment. Consequently, in case of reuse of the data it is difficult to evaluate the adequacy of existing weights, in some cases even impossible. The paper presents a proposal for an environment, which supports weighting as well as structured documentation of the weighting process. Besides weighting the proposed approach seems to be useful for integration of processing and documentation in other areas of statistical computing. Section 2 discusses the requirements on the environment from the statistical point of view. In section 3 we consider a data model fulfilling the requirements outlined in section 2 and section 4 describes processing inside the environment in case of weighting. 2 Statistical requirements for a weighting environment Let be the vector of the values of a variable Y of interest in a finite population and denote by )y,,y,y