A Comparison of Conjoint Analysis Response Formats

Critiques of contingent valuation in the early 1990s (Hausman NOAA Panel) appear to have accelerated rcscarch efforts to investigate the use of conjoint analysis and choice-based methods for eliciting nonmarket values (Adamowicz, Louviere, and Williams; Adamowicz et al.; Garrod and Willis; Johnson and Desvousges; Johnston and Swallow; Holmes, Zinkhan, and Mcrccr; Layton and Lee; Mackenzie; Opaluch et al.; Roe, Boyle, and Teisl; Stevens, Barrett, and Wills). This is a logical progression as these methodologies are all “close cousins” in the general family of stated-preference methods. In fact, they are so closely related that it would be surprising if any one approach proved to be a panacea for all of the problems that have been asserted to apply to stated-preference methods. However, cross fertilization of these literatures is likely to have a posit ive influence in the further refinement of stated-preference methods applied to the elicitation of nonmarket values . A review of these methodologies reveals an extensive literature dedicated to investigating the validity and reliability of Hicksian surplus estimates derived from contingent valuation (Hanemann Mitchell and Carson). While conjoint analysis has an extensive literature (Louviere, 1 %#a and 1988b), applications to the estimation of Hicksian surplus are rather new. Thus, it makes sense that some of the same issues that have been investigated for contingent valuation should also be investigated for other stated-preference methodologies, including conjoint analysis. For example, while we know that contingentvaluation response formats (e.g., open-ended, payment card, and dichotomous choice) influence estimates of Hicksian surplus (Welsh and Poe), we do not know how different response formats in conjoint studies (e.g., ratings, ranks, and choose one) influence estimates of Hicksian surplus. A related issue is that contingent-valuation questions ask respondents to reveal information about their Hicksian surplus directly, while conjoint studies ask people to reveal relative preference orderings. If the choices do not include a “would not buy” or “status quo” alternative, a nonzero value is implied in the estimated likelihood function for people who would not choose one of the alternatives. In general, this serves to bias estimates of Hicksian surplus upward. In this study, we investigate whether recoding ratings to ranks or choose one, and rccoding ranks to choose one, result in comparable