Untangling Differences in Values from Internet and Mail Stated Preference Studies

The increasing use of internet surveys for stated preference studies raises questions about the effect of this choice of survey mode on sample representativeness, value estimates and data quality. A handful of studies have now been conducted that compare the use of the internet with other survey modes, but all of these comparisons apart from one have confounded testing of mode effects with sample frame effects, and no previous studies have separated these two effects when testing the consequences of using an online panel. In this study we seek to identify the effects of using an online panel, and conduct three treatments to decompose both model and sample frame effects.Using choice modelling and a case study focusing on riverine health we find that there are some differences in response rates and differences in mean values of the socio'demographics, experience of the river and knowledge. We identify moderate sample frame effects and more substantive survey mode effects. As we account for taste heterogeneity with the mixed multinomial logit model, the implicit prices decrease for most attributes. Using a newly developed generalised multinomial logit (GMNL) model, we find that scale heterogeneity is important in refining the estimates.