A Contamination Model for Selective Editing

The aim of selective editing is to identify observations affected by influential errors. A score function based on the impact of the potential error on target estimates is useful to prioritize observations for accurate reviewing. We assume a Gaussian model for true data and an “intermittent” error mechanism such that a proportion of data is contaminated by an additive Gaussian error. In this setting, scores can be related to the expected value of errors affecting data. Consequently, a set of units can be selected such that the expected residual error in data is below a prefixed threshold. In the context of economic surveys when positive variables are analyzed, the method is more realistically applied to logarithms of data instead of data in their original scale. The method is illustrated through an experimental study on real business survey data where contamination is simulated according to error mechanisms frequently encountered in the practical context of economic surveys.