One of the questions when discussing the usefulness of Web surveys is whether they gain the same response rates compared to other survey modes. Anecdotal literature reviews suggest that in general, the Web survey response rates are considerably lower. However, such unsystematically synthesized evidence might be highly misleading. As an alternative, systematic meta-analytic procedures synthesizing controlled experimental mode comparisons could provide accurate answers, but to the best of our knowledge, these studies have not been conducted so far.

To overcome this gap, we conducted a meta-analysis of 36 published experimental comparisons between Web and other survey modes. Due to the fact that response rates proportion differences tend to overestimate the heterogeneity of effects, we took the natural log of the odds-ratio as our effect size measure. To estimate the study population effect under a random effects model assumption, we firstly computed the mean effect size weighted for the inverse effect size variance as suggested by Lipsey and Wilson (2001). Subsequently, a homogeneity analysis and a meta-regression were performed. To predict the response rate variance on the effect size measure, the following study descriptors were included as independent variables: type of alternative mode (e.g., eMail, CATI, mail, fax, IVR), year of study, type of the target population (e.g., students, general population, ...), sponsorship (academic, government, commercial), and the implementation procedures (incentives, No of contacts, contact mode).

While the weighted mean effect size indicated a slight advantage for other than Web modes, the 95% confidence interval around this mean effect size estimate included zero. Thus, this results suggests that there is no systematic response rate differences between Web surveys and other modes. A homogeneity test revealed that the 36 effect sizes are homogeneous, indicating that the study descriptors are not systematically related to the effect size variability. This result was further supported by a meta-regression with the study descriptors included as independent variables, none of them exerting a significant predictive effect on the effect size variability.
Taken together, our meta-analysis highlights that the common assumption of lower response rates for Web surveys compared to other modes does not hold true if scrutinized with the aid of meta-analytic research synthesis procedures, taking into account experimentally controlled primary studies. From a practical point of view, these results might contribute to improve the reputation of Web surveys as one survey mode of comparable data quality with respect to response rates.