Why are Web Survey Results Biased?

Findings from Two Experimental Web Surveys

Abstract

Organizing two experimental surveys, this article inspects causes of bias that Web surveys occasionally indicate. The first experiment compares mail respondents and Web respondents, all who are extracted from a single sampling frame, the electoral roll, and randomly assigned to either of the two response modes. The second one discriminates Web respondents from two different sampling frames: opt-in volunteer panels and those randomly extracted from residential register. The first experiment examines the effects of undercoverage and low response rate of Web surveys, while the second one seeks out measuring self-selection bias. The results are simple and straightforward; the influence of coverage and nonresponse errors is minimal and practically ignorable, whereas self-selection bias is conspicuous. Opt-in Web panels show a unique attitudinal tendency at most question items, even after controlling demographic factors and major Web activity variables.