The annual Social Climate Survey of Tobacco Control utilizes random samples of the US population. From 2001-2008, annual samples were obtained using Random Digit Dialing methods. This study examines the utility of a mixed-mode approach, used in 2009-10, and examines the validity of overall and subpopulation findings.The RDD frame included US households with landline telephones. Knowledge Networks provided the internet panel. Members were randomly recruited by telephone and mail surveys, and households were provided with access to the Internet and hardware if needed. Weights were computed in two steps. First, both frames were weighted based upon 2009 US Census estimates. Second, adjustments to these initial weights were computed to account for the overlap in the two samples. Comparisons to the Census demonstrated that the KN panel provided better representation of adults with low education and young adults. Age and education are correlated with smoking. Predictably, KN panel estimates for tobacco control indicators were marginally lower than RDD frame estimates.

Estimates from the National Health Interview Survey, administered via household interviews, were used to examine the validity of the mixed-mode approach. Both surveys use the same protocol to assess smoking status. The RDD frame produced the lowest estimate (14.8%) and the the KN frame produced the highest estimate (20.3%) for smoking, whereas the mixed-mode dataset produced an estimate of 18.3%, which is within the sampling error of the NHIS estimate (19.9%). Subpopulation comparisons demonstrated that estimates for current smoking from the mixed-mode dataset closely matched those from the NHIS for young adults, adults with low education, and Hispanic adults.

These findings suggest that the mixed-mode approach used significantly reduced our

undercoverage sample bias. Mixed mode samples and weighting may help researchers select methods that will provide more representative samples and increase validity for tobacco use and other health behaviors.