Notice: the WebSM website has not been updated since the beginning of 2018.

Web Survey Bibliography

Title Bayesian Combining of Web Survey Data from Probability- and Non-Probability Samples for Survey Estimation
Year 2017
Access date 09.04.2017
Abstract Sample surveys are frequently used in the social sciences to measure and describe large populations. While probability-based sample surveys are considered the standard by which valid population-based inferences can be made, there has been increased interest in the use of non-probability samples to study public opinion and human behavior, particularly through web surveys. This increased interest is driven by multiple factors such as costs which can be significant when recruiting a probability-based sample. A second factor is the popularity of the web as a survey platform which has led to increased adoption of online access panels that can deliver cheaper and timelier survey results compared to traditional probability-based surveys. However, online access panels are heavily criticized because they do not employ probability sampling methods to recruit panel members, and therefore the mathematical probability theories that underlie valid statistical inference cannot be applied. While non-probability-based surveys are not ideal for making population-based inferences, their attractive cost properties make them potentially useful as a supplement to traditional probability-based data collection. In this paper, we examine this notion by combining probability and non-probability Web survey samples under a Bayesian framework. The Bayesian paradigm is well-suited for this situation as it permits the integration of multiple data sources, and a potential for increased precision in estimation. On the other hand, combining probability samples with non-probability samples that could be biased may offset gains in efficiency. Thus, there is likely to be a bias-precision tradeoff when combining probability- and non-probability samples. We examine this tradeoff using the German Internet Panel (GIP), a nationally-representative, probability-based web survey in combination with a set of non-probability-based web surveys that fielded a subset of the GIP questionnaire during the same time period. We apply the Bayesian combining framework to produce estimates of survey items and compare them to the probability-based estimates alone. We examine the accuracy and precision of the resulting survey estimates to determine whether combining the probability and non-probability samples yields valid inferences (and a likely cost savings) relative to the probability survey alone.
Year of publication2017
Bibliographic typeConferences, workshops, tutorials, presentations
Print

Web survey bibliography (4086)

Page:
Page: