Abstract:
|
Due to cost savings, web-based surveys are sometimes considered as alternatives to the traditional surveys based on probability sampling methods. Web surveys often have multiple stages of selection, with some stages structured by design-based sampling, but with other stages (possibly all) sampled non-randomly. A task for survey planners is to determine a set of web-survey weights for the respondents that lead to unbiased population estimation. One such method assigns web-survey weights based on propensity score estimation computed by combining a traditional "representative" survey with a web-survey with the latter taking the role of "treatment". For this study, a web-survey having questions in common with those on the National Health Interview Survey (NHIS) was fielded in 2015. Several variations of propensity score methodologies appearing in the literature are implemented. The results of this study are discussed.
|