Imputation for Longitudinal Study of Effectiveness of an Anti-Smoking Campaign

• Activity Number: 530 - SPEED: Survey Research Methods
• Type: Contributed
• Date/Time: Wednesday, August 2, 2017 : 10:30 AM to 11:15 AM
• Sponsor: Survey Research Methods Section
• Author(s): Qiao Ma* and Josiane Bechara and Edward Mulrow and Zachary Haskell Seeskin and Morgane Bennett and Jennifer Cantrell and Elizabeth Hair and Donna Vallone
• Companies: NORC at the University of Chicago and NORC at the University of Chicago and NORC and NORC at the University of Chicago and Truth Initiative and Truth Initiative and Truth Initiative and Truth Initiative
• Keywords: Missing Data ; Hot Deck Imputation ; Multiple Imputation ; Evaluation ; Nonresponse Bias

Abstract

The Truth Initiative Longitudinal Cohort Study is designed to evaluate the impact of a television and digital campaign on youths' smoking-related knowledge, attitudes and beliefs, perceived social norms, and behaviors over time. The study administers surveys to participants over six waves between 2014 and 2017 and uses multivariate statistical models to evaluate the effectiveness of the media campaign. The survey is subject to nonresponse, which can bias estimates for the evaluation. We describe and examine different methods of imputing missing data in the context of a longitudinal study. Hot deck and model-based approaches are compared for both their performance and practicality. Examining income, the variable with the highest item nonresponse rate, we find that using either hot deck or model-based estimation helps correct for nonresponse bias in estimates from complete case analysis, and we demonstrate how multiple imputation can help account for the uncertainty in estimates due to imputation.