Sampling weights for analyses of couple data in demographic and health surveys

Stan Becker, Johns Hopkins University
Bryan Sayer, Social & Scientific Systems, Inc.

In demographic and health surveys sometimes both women and men are interviewed in selected households allowing matching of partner information and analyses of couples. Typically individual data from such sample surveys have sampling weights which incorporate factors for both the sampling probability and non-response. For analyses of couple data neither the weights for females nor the weights for males are appropriate. We present formula for estimating the appropriate weights for couples. To see how results vary when using male or female weights instead of couple weights, we analyze couple data from one DHS survey–-The Dominican Republic survey of 1996. Utilizing women’s weights, results for means, standard errors, and regression coefficients and their standard errors were within 2% of the values using couple weights in 43 of 52 comparisons, and also 43 of 52 were within 2% using men’s weights. Calculation and use of couple weights is straightforward and desirable.