Data Portal

In microeconomics, people and households are key units of analysis making it important that they sensibly relate to each other in the data with which they are studied. Although most Statistics South Africa household surveys are designed to be simultaneously representative of people and households, weighting practise is to split the calibration of person and household weights, rendering the data only representative of either the household or the person universe at a time. This process introduces noise to inference and needlessly curtails the number of questions the data can be used to answer in a conceptually coherent way. The motivation for splitting the person and household weights was that policymakers in particular needed an accurate series o ftotal household counts and it was unknown whether constraining the existing calibration onboth person and household information would yield the desired outcome. This paper uses cross-entropy estimation to reweight a stacked series of cross-sections from the October Household Survey (1994-1999) and the General Household Survey (2002-2015) to restore representativenes sand bring weighting practise back in line with sampling practise, whilst focusing on achieving accurate total counts of both people and households. The new weight not only achieves accurate total counts but outperforms the separate person and household weights over