Differences in General Health of Internet Users and Non-users and Implications for the Use of Web Surveys

Web surveys have become popular in many fields of research. To compensate persisting undercoverage and nonresponse problems of web surveys, weighting strategies are used. However, the underlying assumptions of weighting are rarely tested. If the probability of missing data depends on the missing data itself (missing not at random, MNAR), no standard weighting method will correct for nonresponse or undercoverage bias. We postulate a MNAR selection effect due to health conditions. Using real data from large scale non-internet surveys in different countries (European Social Survey (ESS), n 55; 000, Behavioral Risk Factor Surveillance System (BRFSS), n 492; 000), large differences in general subjective health between Internet users and non-users can be observed. Weighting by calibration on age, gender, ethnic background, urban residence, education and household income does not eliminate the observed health differences. Therefore, the underlying missing data mechanism might be considered as an example of MNAR. If this holds, no weighting strategy will be able to eliminate health bias in web surveys.