Background

Web-based surveys may have advantages related to the speed and cost of data collection as well as data quality. However, they may be biased by low and selective participation. We predicted that such biases would distort point-estimates such as average symptom level or prevalence but not patterns of associations with putative risk-factors.

Methods

A structured psychiatric interview was administered to parents in two successive surveys of child mental health. In 2003, parents were interviewed face-to-face, whereas in 2006 they completed the interview online. In both surveys, interviews were preceded by paper questionnaires covering child and family characteristics.

Results

The rate of parents logging onto the web site was comparable to the response rate for face-to-face interviews, but the rate of full response (completing all sections of the interview) was much lower for web-based interviews. Full response was less frequent for non-traditional families, immigrant parents, and less educated parents. Participation bias affected point estimates of psychopathology but had little effect on associations with putative risk factors. The time and cost of full web-based interviews was only a quarter of that for face-to-face interviews.

Conclusions

Web-based surveys may be performed faster and at lower cost than more traditional approaches with personal interviews. Selective participation seems a particular threat to point estimates of psychopathology, while patterns of associations are more robust.