Statistical output, e.g. totals or means of a target variable, are often published for subpopulations that are defined by categorical domain variables (such as categories of educational level, categories of economic activity, and so on).
It is important to check the quality of these variables. A way to do this is to perform an audit on a sample of that population, that is representative with respect to the domain variable. When a possibly non-representative sample of units has already been audited, it would be most efficient to re-use as many of these already audited units as possible. In this paper, a method is introduced that selects an audit sample which re-uses previously audited cases by considering the selection of an audit sample that is representative with respect to domain variables as a constrained minimization problem.
