Multidimensional Control Totals for Post-stratified Weights
Creel, D. V., & Fahimi, M. (2005, August). Multidimensional Control Totals for Post-stratified Weights. Presented at Joint Statistical Meetings, Minneapolis, MN.
Calculating analysis weights for surveys with complex designs can be a major undertaking, oftentimes requiring numerous adjustment steps. While the basic steps for weighting data for simple surveys include calculation of base weights followed by adjustments for eligibility, nonresponse, and poststratification, for complex surveys one often encounters issues, known or unknown, that require some modifications to the basic adjustment steps. This paper discusses a number of such issues related to the use of control for poststratification and the solutions that were implemented while calculating the analysis weights and replicate weights for the 2004 National Study of Postsecondary Faculty (NSOPF:04) survey. Some examples are balancing the weights with respect to a large number of control distributions without generating extreme weights and using information from other sources to create subgroup control totals for which no published estimates are available.