Monte Carlo Benchmarks for Discrete Response Valuation Methods: Comment

Abstract

In a recent article, Huang and Smith (1998, hereafter HS) use Monte Carlo simulation methods to suggest that the procedural variance observed between open-ended (OE) and dichotomous choice (DC) contingent valuation (CV) responses can be attributed to specification error in modeling the DC responses. In particular, HS argue that employing alternative specifications of the error term for DC responses can provide estimates of mean willingness to pay (WTP) that are identical to the mean of the raw data derived with open-ended CV and not significantly different from the mean . .. for actual purchases (p. 191). Hence, they claim that the evidence from a large body of laboratory and field research that DC-CV question formats yield substantially larger estimates of mean WTP than OE response methods is unfounded (p. 187). While we applaud the Monte Carlo methods used by HS to demonstrate that error specification is important in providing unbiased and accurate estimates of WTP and conditional WTP, we wish to caution the reader that the gap between OE and DC WTP estimates is not likely to be bridged by simply imposing alternative specifications. When real humans make real or stated decisions, observed procedural variance across elicitation methods and the degree of hypothetical bias are more fundamental than relying on alternative econometric specifications. To demonstrate this point we roughly follow the organization of the HS paper. In the following section we provide a brief review of the Balistreri et al. (2001) data showcased by HS.' This data is used to demonstrate that, in contrast to HS, the mean WTP estimate from DC-CV data is significantly different from ctual contributions and that the DC an OE distributions are significantly different from each other. In the third section, we rais concerns about the functional forms, error distributions, and welfare estimates used in the Monte Carlo analysis of HS. Using a broader range of utility-theoretic specifications than the linear logistic and probit models employed by HS, we demonstrate that employing alternative error specifications is not likely to overcome the observed disparity in mean WTP values across elicitation methods. The fourth section addresses concerns about the Turnbull lower bound estimator used by HS in support of their not significantly different and virtually identical claims, and the increased application of this method to provide conservative estimates of mean WTP from DC-CV responses. We conclude with some final thoughts on relying on alternative error specifications, rather than seeking a better understanding of human