Using a Reference Survey and Propensity Scores to Adjust BRFSS Survey Weights to Compensate for Non-Cell Phone Coverage

• Activity Number: 526
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Julia Zhu*+
• Companies: CDC
• Keywords: Survey ; propensity score ; selection bias ; BRFSS ; Surveity weight

Abstract

As the percentage of cell-phone-only households continues to grow, the validity of the traditional random digit dialing (RDD) landline sampling model has become open to question. We use propensity scores to adjust weights to compensate for non-cell phone coverage. We use demographic and additional lifestyle variables to develop a "cellphographic" model which is then used to calculate propensity score adjusted weights. We apply the adjusted weights to the Behavioral Risk Factor Surveillance System (BRFSS) survey-a traditional landline-based RDD survey - to the 4 most underestimated important health outcomes for which underestimation is due to non-cell phone coverage (Hu, Balluz, Battaglia & Frankel, 2010): not receiving health care due to cost (Relative Bias, -9.17); current cigarette smoker (Relative Bias, -10.25); binge drinker (relative Bias, -14.81); and ever being tested for human immunodeficiency virus (Relative Bias, -5.10). For these four outcomes, we compare nation-level propensity score adjusted estimated prevalence from (landline-only) BRFSS with estimates from the 2008 dual-frame estimate. when adjustments are made, the differences decreased.