WTADJX Is Coming: Calibration Weighting in SUDAAN When Unit Nonrespondents Are Not Missing at Random and Other Applications

• Activity Number: 563
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Phil Kott*+
• Companies: RTI International
• Keywords: Response model ; Nonresponse bias ; Generalized raking ; Variance estimation ; Optimal calibration

Abstract

SUDAAN 10 (RTI 2008) introduced the WTADJUST procedure, which produces calibrated survey weights. By forcing the weighted totals for a set of "calibration variables" to equal benchmark totals computed from either the whole sample, a larger sample, or the entire population, calibration weights can reduce or eliminate the potential for bias from unit nonresponse under a reasonable response model. WTADJX, which will be available in SUDAAN 11 (RTI 2012) allows the set of model variables governing the response model to differ from the set of calibration variables. As a result, the new procedure may be used to assess and perhaps even correct for unit nonresponse that is not missing at random, that is to say, when unit response is a function of variables known only for the respondents. There are other potential uses of WTADJX. These include producing calibrated weights asymptotically identical to so-called "optimal" ones and simplifying the computation of replicate weights when using a nonlinear calibration routine.