Web-based surveys and sample frame bias in a choice experiment

Researchers are seeking effective, low cost means of gathering high quality data. Technological advancements offer new avenues for achieving this objective. Web based surveys are relatively common outside the economics discipline however applied non-market valuation practitioners have been slow to adopt this modernisation in survey methodology. The non-random exclusion of individuals from the sample frame is often cited as a major problem with web-based surveying. This paper presents a comparison of data from two choice experiment survey modes, traditional mail-and-return and web-based. The socio-demographic composition of the samples is significantly different for half the variables considered. Poe tests reveal that there are significant differences in wtp for ecological improvements with the web sample having higher mean wtp. There are no differences detected for all other attribute wtp estimates. Poe tests also reveal that there are no differences detected for compensating surplus estimates over 15 scenarios. A pooled model with a dummy variable equal to 1 if respondent uses web mode is specified and is found to be negative and statistically significant at a 1% level