Integrate weighting in household-based survey

Muthetho Solomon Nkwinika, South Africa


Most Household Surveys use two-stage complex sample design to measure person and household characteristics over time. Over the years there has been no reliable way to measure household growth over time the data collected. Demographers measure population growth taking into consideration migration, mortality, and fertility or using the crude mathematical model. Whatever method demographers use, at least there is a model to measure person population dynamics. Generalized Raking is a family of optimization procedures that has been successfully used in many fields of study including Physical and Engineering Sciences. In sample surveys these procedures have been used to post-stratify or benchmark person population estimates stratified by demographic and geographic variables. This is done to produce unbiased person estimates consistent with population estimates of the statistical organization conducting the sample survey. This paper seeks to demonstrate the use of Generalized Raking model to generate household estimates that are consistent with person population estimates from the Household survey data.  Generalized Raking model used in this way has been termed Integrated Weighting because it integrates the sample design weights and the population projections to produce one set of weights. The resulting set of weights can then be used to estimate both the person and the household characteristics without biasing the sample results. The SAS macro - CALMAR - developed in France by I.N.S.E.E is used to implement Integrated Weighting.


Keywords: weighting