Sensitivity Analysis of Bias of Estimates from Web Surveys with Nonrandomized Panel Selection

• Activity Number: 488
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Vladislav Beresovsky*
• Companies: National Center for Health Statistics
• Keywords: web survey ; reference survey ; biased estimates ; nonrandomized experiment ; propensity score

Abstract

Rising costs, growing non-response and uncertain frame coverage stimulate growing interest to use web surveys instead of representative samples for estimation of general population characteristics. Previous research has demonstrated that weight adjustment by propensity score constructed on demographic variables may be insufficient for balancing volunteer panels of web surveys selected in nonrandomized fashion. This may result in biased estimates. Assuming systematic dependence of propensity score and target variable on unobserved variable $U$, we derive dependence of bias on the model parameters associated with $U$. Direct analytic dependencies were calculated for limiting cases of model parameters using Taylor series expansions. Better understanding the reasons for biased estimates will stimulate development of methods for bias reduction.