The use of uncertainty to choose matching variables in statistical matching

Abstract

Statistical matching aims at combining information available in distinct sample surveys referred to the same target population. Thematching is usually based on a set of common variables shared by the available data sources. For matching purposes just a subset of all the common variables should be used, the so calledmatching variables. The paper presents a novel method for selecting the matching variables based on the analysis of the uncertainty characterizing the matching framework. The uncertainty is caused by unavailability of data for estimating parameters describing the association/correlation between variables not jointly observed in a single data source. The paper focuses on the case of categorical variables and presents a sequential procedure for identifying the most effective subset of common variables in reducing the overall uncertainty.