Demographic and Health Surveys (DHS) using standardized questionnaires have been conducted in over 90 developing countries with many countries having 3, 4 or more such cross-sectional surveys since 1987 when the program started, giving a total of 255 surveys. 125 of these surveys include separate questionnaires for both men and women in households so data from married and in-union couples can later be matched. The surveys only give nationally representative results after weighting to account for different sampling probabilities and different response rates in different sampling domains. The response rates for men are lower than those for women so the DHS couples'data files include weights for both men and women. Investigators have been unsure which weight to use for analyses of couples'outcomes. Neither weight is correct nor will this project provide calculations of a new couple weight which will be used for the 11 DHS where the couple weight can be obtained because all couples can be identified in the household questionnaire. Two proxy or alternative couples weights will also be calculated as these are available for all surveys with couple data. The percentage bias and absolute percentage bias for means, proportions, regression coefficients and their standard errors will be calculated using each of the 2 alternative weights as well as the female and male weights by comparing to results for the same indicators obtained using the actual couple weight. Estimates of Mean Square Error for the indicators with the five weights will also be calculated and compared. One question that will be answered is: Do proxy couple weights need calculation or does use of male or female weights give results that are sufficiently close to estimates using couple weights so one or the other of these is acceptable to use? This study will document the level of bias that can be expected from analyses of couple data using female or male weights and will indicate the best weight to employ for researchers who use the couple data in DHS and other similar surveys.
This research will utilize a new sampling weight for couples in 11 Demographic and Health Surveys where it can be calculated. The level of bias--in estimates of proportions, means, regression coefficients and their standard errors--from using two alternative weights that can be calculated for all 125 DHS surveys with couple data will be estimated as well as the bias from using men's or women's weights, instead of the true couple weight. Results will aid all researchers who utilize couple data from DHS and other national surveys.
