Imputation of Missing Data for the Behavioral Risk Factor Surveillance System (BRFSS): A Comprehensive Approach
Fahimi, M. (2007, May). Imputation of Missing Data for the Behavioral Risk Factor Surveillance System (BRFSS): A Comprehensive Approach. Presented at American Association for Public Opinion Research Conference, Anaheim, CA.
Oftentimes, missing data encountered in survey research are imputed by relying on statistical methods. Coordinated by the Centers for Disease Control and Prevention, Behavioral Risk Factors Surveillance System (BRFSS) is the world’s largest random digit dialing survey. However, since its inception some 20 years ago, missing data for BRFSS have not been imputed – a practice that has caused inefficiencies with respect to both analysis and weighting of the resulting data. Analogous to other large surveys, BRFSS encounters nonresponse at both unit and item levels. While a basic weighting methodology has been developed to compensate for some of the bias resulting from differential nonresponse at the unit level, currently, there are no procedures in place to impute missing data for BRFSS. For the first time, a comprehensive methodology has been developed to impute all missing data for the core part of the BRFSS questionnaire, which consists of over 80 different items. Given the complex branching structure embedded in the BRFSS survey instrument, this methodology relies on a sophisticated imputation protocol to adhere to all skip patterns and questionnaire logic.