Modeling Frame Deficiencies for Improved Calibrations

• Activity Number: 138
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Eric Slud*+
• Companies: U.S. Census Bureau
• Keywords: Frame Errors ; Calibration ; Survey Frame ; Master Address File ; Markov model ; Time-dependent covariates

Abstract

One of the important reasons advanced for calibrating household surveys to population totals is the imperfect coverage of target frame populations by operational frame lists. Calibration is an implicitly model-based operation, the theoretical justification for which depends on relatively simple assumptions about the correctness of the frame list. Essentially, one assumes that this list is nearly perfect except possibly for random omissions similar to those in the pseudo-randomization model for unit-level nonresponse. The author is not aware of previous studies extending the design consistency of calibration to more complicated models of frame deficiencies, although such models are formulated and studied at the Census Bureau for other purposes. In this research, we formulate a simple set of models allowing dynamic transitions of units on an ideal frame list into and out of the operational frame and describe the provable design-based properties of calibrated survey-weighted estimation methods under these models. The theoretical results will be illustrated via simulation and on CPS 2012 first-month-in-sample data.