Valuing risk reductions: Testing for range biases in payment card and random card sorting methods

Abstract

Ongoing concerns with regard to the appropriate approach to elicitation of willingness-to-pay responses in contingent valuation studies have led to the development of a number of alternative techniques. One of the most recent, and on the surface, most promising of these is the random card sorting approach (RCS) which has been used for policy purposes to value risk reductions. This paper provides the first formal test of this procedure, comparing it against the widely used payment card (PC) format from which it is developed and whose recognised problems, such as range bias, it claims to address. However, the findings suggest that the RCS procedure is no less vulnerable to range bias than the PC method for eliciting both monetary values of health risk reductions and non-monetary estimates of death rates. Conclusions for future research initiatives are drawn.

Acknowledgements

The authors wish to thank Sven Baaz for collecting the survey data reported in this study. We are grateful to the Economics for the Environment Consultancy (EFTEC) for support for this research.

Notes

1 The OE method is generally discounted by researchers because of its high non-response rate and tendency to exhibit free-riding and strategic overbidding tendencies (e.g. Bateman et al., 1995) whereas the IB method is particularly susceptible to starting point bias (e.g. Roberts et al., 1985; Stalhammar, 1996).

2 This replicates the approach of Carthy et al. who introduced the RCS procedure. Note that, as in the latter study, while RCS respondents were not told in advance about the range of values on the RCS, they were allowed to revise their responses once the initial sorting of value cards was completed.

3 Copies of the stimulus materials and interview protocols used by the investigators are available from the lead author.

4 Although these range effects in the PC format are lower than those recorded by the Dubourg et al. study, they are still substantial. This contrasts with the results of Rowe et al. (1996) who report no significant PC range effects unless “the upper end of the value distribution is inappropriately curtailed” (p. 184). Given this, it may be that the ranges in the current study did not extend to sufficiently high values to avoid range effects. However, while this may have affected responses, given that both the PC and RCS treatments employ the same set of ranges, any ‘curtailment effect’ will not invalidate those comparisons across these elicitation methods that are the principle focus of this study.

5 With 14 cards in each pack, the median might have been expected to be nearer £50 – 75 for the LOW range and £150 – 200 for the HIGH range.

6 It is worth noting that the correlations found in this study between the first card value and the best estimates were of a similar magnitude to the correlation of 0.142 reported in Chilton et al. (2004).