Criterion validity tests of the contingent valuation methods with market /non-market comparisons of hypothetical bias and willingness to pay estimates compared with paired comparison estimates of willingness to accept

Abstract

This dissertation investigates the sources of hypothetical bias in contingent valuation methods and differences between willingness to pay (WTP) and willingness to accept (WTA) measures. It involves two separate experiments including six independent treatments all of which utilize the same private good (art print). Hypothetical and actual cash WTP are elicited with both open-ended and dichotomous choice (DC) question formats, which are compared for differences in hypothetical bias. The effectiveness of a reminder statement in eliminating or reducing hypothetical bias in both question formats is tested. The reminder requested that respondents not report what they thought a fair market price was for the good and to act as if they were in a real market. The DC data was tested for starting point bias. WTP measures elicited with the DC question format were compared with WTA measures for the same art print elicited with the method of paired comparison, which involves having individuals make binary choices between receiving a particular good or a sum of money and inferring WTA from the ranking of dollar amounts and the good of interest. WTP and hypothetical bias are compared for those in the market and those not in the market for art prints. With the open-ended question format, equality of hypothetical and actual WTP was rejected by the Multi-Response Permutation Procedures at the 0.05 level of significance when the reminder was included, but it was not rejected by the Kolmogorov- Smimov test of distributions, and the Mann-Whitney test of medians rejected equality at the 0.10 level of significance but not at the 0.05 level. Hypothetical WTP was about two times larger than actual cash WTP. With the DC question format, neither the non-asymptotic approach to the chi-square statistic developed by Mielke and Berry or Fisher's exact test reject equality of hypothetical and actual cash WTP based upon the raw data. However, the Mann-Whitney test of medians rejects equality of the hypothetical and actual cash WTP based upon logit estimated data according to Hanemann's utility difference approach; but a likelihood ratio test and the method of convolutions (both based upon the logit estimated data) reject equality of hypothetical and actual DC responses at the 0.05 but not the 0.01 level. The logit estimated data produced hypothetical WTP measures about two times the estimated actual cash WTP. No starting point bias was found to exist in the estimated data. Paired comparison estimates of mean (median) WTA for the art print is estimated at $59($52) compared with the DC contingent valuation estimates of mean (median) WTP of $28(28). The difference persisted in spite of the chooser reference point employed by the method of paired comparison, so the findings do not support Kahnemann et al.'s proposal that endowment effects and loss aversion are responsible for differences between WTP and WTA. The chapter concludes that the method of paired comparison does not produce valid measures of WTA because it does not offer respondents an option of indifference between goods, which is required by the microeconomic axiom of completeness. The market/non-market analysis of the independent samples indicate that hypothetical WTP, both with and with no reminder, and actual WTP are all significantly greater for those in the market for art prints than for those not in the market. The data indicate that those in the market for art prints and those not in the market come from two separate and distinct populations with separate and distinct WTP behaviors, and it is statistically inappropriate to combine them in one sample. Market/non-market comparisons indicate that the reminder was effective in eliminating any significant hypothetical bias for those in the market for art prints and those not in the market for art prints, but it was not effective in eliminating hypothetical bias for those who neither agreed nor disagreed that they were in the market for art prints. Those not in the market caused skewness in our samples, which were non-normally distributed. Their bids consisted of many zero and low bids. Thus non-parametric statistics are analyzed in depth, and unique applications are employed.