How to weight survey data with a dyadic multi-actor design?

Abstract

This paper deals with adjustment for nonresponse in dyadic multi-actor survey designs. It presents a multi-dimensional approach to weighting that addresses the various analytical units represented in such data, so that sampling design weights are correctly accounted for and so that consistency between weights is achieved. This approach is demonstrated by using the primary respondents in the Divorce in Flanders study, which is a typical example of a dyadic multi-actor design. Five sets of weighting coefficients are made available whereby different subsets of data, according to different analytical units, are selected: the subset of the dyads, the subset of men and women respectively, and two subsets of marriages. Post-stratification – with the year of marriage, status of the reference marriage at the sampling date and five-year divorce cohort as auxiliary variables – was chosen as the weighting adjustment technique.