Using ex ante approaches to obtain credible signals for value in contingent markets: evidence from the field.

Next, we examine the proportion of "Yes" votes for each price level, across pricing sequences A and B. We use the [chi square] statistic to compare identical prices when offered first and second, for each of the four treatments. In contrast to the results of ALP, we find no evidence of significant sequencing effects in the voting proportions for either price level, suggesting that anchoring may not be an important phenomenon in the marketplace. (14) Thus, for efficiency purposes, we pool the data within price cells. The subsequent results use the pooled data, which are summarized in table 2. (15)

Treatment Effects

Figure 1 (Figure 2) provides a graphical depiction of the pooled data contained in table 2 by presenting the proportion of "Yes" votes across treatments for the $5 ($10) price level. The data paint an interesting picture: 32.8% of subjects voted to fund the public good in the real $5 treatment, while in the $5 consequential treatment, 32.2% voted "Yes." These proportions are notably similar. On the other hand, the proportion of "Yes" votes in the $5 cheap talk treatment was considerably greater, at 46.4%, and the $5 hypothetical treatment exhibited a much larger proportion of "Yes" votes: 84.4%. Similar trends are evident in the $10 data. (16) Overall, perusal of table 2 and figures 1 and 2 suggests that voting behavior across the four referenda is considerably different. This insight is documented statistically, as the [chi square] (df = 3) statistic for the test for equality of the four proportions allows us to reject the homogeneity null at the p < 0.01 level for both price levels ($5: [chi square] = 45.5980 and $10: [chi square] = 58.3138).

[FIGURES 1-2 OMITTED]

Turning to a comparison of the individual treatment effects, we present table 3, which summarizes statistics of pair-wise [chi square] tests. The upper right (lower left) triangular elements present the statistics for the $5 ($10) price level. A first important question is whether voting behavior in the hypothetical treatments is different from voting patterns in the real treatments. The raw data suggest large differences between the hypothetical referendum and the actual referendum: whereas 84% (75%) voted "Yes" to the proposition in the hypothetical treatments at the $5 ($10) offer price, only 33% (19%) voted "Yes" in the real treatment. Indeed, as is presented in table 3, the proportion of affirmative votes in the hypothetical referendum is statistically different from the percentage of affirmative votes in the real treatment (as well as the other two treatments) at the p < 0.01 level. (17) Thus, our evidence suggests that subjects' respond differently in hypothetical referenda than they respond in our three other types of referenda.

Turning to comparisons of data from other treatments, we find that voters in the cheap talk treatment tend to vote "Yes" more often in both the $5 and $10 treatments compared to voters in the real treatment. While this pattern is stark, these observed differences are not statistically significant at conventional levels: for the $5 offer price, the [chi square] (df = 1) statistic for real versus cheap talk is 2.5486 (p-value = 0.1104), and for the $10 price level, the [chi square] (df = 1) statistic for real versus cheap talk is 1.9038 (p-value = 0.1677). Yet, it should be noted that using a one-sided alternative, these differences are significant at the p < 0.10 level.

Since valuation experiments are typically utilized to estimate willingness to pay (WTP) we are interested in whether data from these different treatments produce comparable measures in this regard. We used the nonparametric Turnbull to estimate the lower bound of mean WTP (Haab and McConnell 2002). In doing so, we utilized data on the response to only the first price offered in each price sequence (since the Turnbull requires independence of responses to randomly assigned prices). We find that we cannot reject the null hypothesis [WTP.sub.cheap talk] = [WTP.sub.real] at the p = 0.1921 level (t = 1.3091, df = 131) for a two-tailed test. But, again, if we use a one-sided alternative, we reject the equality of lower bound WTP estimates at p < 0.10.

We find that affirmative responses in the consequential and real treatments are roughly equivalent: 32.2% (20.3%) in the $5 ($10) offer price versus 32.8% (18.8%), respectively. For the $5 offer price, the [chi square] (df = 1) statistic for real versus consequential is 0.2594 (p-value = 0.6105). For the $10 offer price, the [chi square] (df = 1) statistic for real versus consequential is 0.04934 (p-value = 0.8215). We therefore cannot reject the hypothesis that these data are the same at conventional significance levels. In addition, we cannot reject the null hypothesis that the Turnbull estimates of the lower bound of mean WTP are equal (t = 0.2941, df = 121;p-value = 0.7692). The evidence is in favor of the consequential design's ability to provide reliable signals of value.

As a final test, we compare data from the cheap talk and consequential treatments. For the $5 offer price, the [chi square] (df = 1) is 2.6656 (p-value = 0.1025), and for the $10 price it is 1.2682 (p-value = 0.2601). We therefore cannot reject the null hypothesis that the data are derived from the same underlying parent population for either offer price at conventional significance levels. Moreover, turning to the Turnbull estimate mean WTP, we find evidence that suggests we should not reject the null hypothesis [WTP.sub.cheaptalk] = [WTP.sub.consequential] (t = 0.9814, df = 122; p-value = 0.3283). (18)

Discussion and Conclusions

Whether contingent markets can produce credible value estimates remains of utmost policy importance. Indeed, for public regulators and damage assessors, contingent surveys remain the only method that can potentially obtain estimates of total economic value for nonmarketed commodities. Using data gathered from more than 250 subjects, we find experimental evidence that suggests responses in hypothetical referenda are significantly different from responses in real referenda. This result is in accordance with many of the studies that have examined hypothetical and real statements of value. Yet, we do find evidence that when decisions potentially have financial consequences, subjects behave in a fashion that is consistent with behavior when they have consequences with certainty. Our results furthermore suggest that estimates of the lower bound of mean WTP derived from "consequential" referenda are statistically indistinguishable from estimates of the actual lower bound of WTP. (19)

Such insights represent good news for stated preference surveys, as a necessary condition for their efficiency is that they are able to provide accurate estimates of value. Yet, this news should be tempered in that such results represent only the beginning of the research process. Even if our results are found to hold across different experimental designs and other types of manipulations the necessary next step is ensuring that survey respondents view the instrument as consequential. In our experiment and other related laboratory exercises (i.e., Cummings and Taylor 1998), the probabilities utilized are clearly objective, being defined by the experimental monitor in a transparent way (the appropriate mix of different colored bingo balls or specific outcomes associated with the roll of a 10-sided die or coin flip). In the field, beliefs about a contingent referendum vote actually affecting policy are subjective, largely out of the control of researchers.

Utilizing postsurvey questionnaires, previous research suggests that survey respondents' believe that the money generated would actually be spent on the proposed project (Powe, Garrod, and McMahon 2005) and that the majority of respondents regard the CV results as something that is likely to be of use to policy makers (Brouwer et al. 1999). However, we are unaware of any results in the stated preference literature that offer an explicit assessment of perceived consequences of survey respondents. Thus, it strikes us that another important focus of future research should be to assess perceived consequences of survey respondents subsequent to value elicitation and learn about the factors that influence such perceptions. While we are unaware of how various procedures increase the likelihood of consequentialism, stated preference researchers generally realize the importance of providing background information on the public good of interest and policy options available for addressing its provision, which might heighten consequentialism. We cannot emphasis enough the importance of pretesting surveys in order to improve the perception of realism on the part of respondents.

In addition, practitioners of stated preference should continue to focus on the realism associated with payment vehicles (the hypothetical method by which payment for the public good would be made). For example, higher overall price levels may not seem tied to public good provision in a realistic way, but on the other hand, higher electricity prices, taxes, or the institution of user fees probably will. As suggested above, debriefing questions can help to improve the understanding of respondents' perceptions of the survey questions. A simple Likert-scale assessment of perceived consequences (i.e., level of agreement/disagreement with some statement regarding the likelihood that survey responses will influence the eventual policy decision) could be quite informative and not likely to be onerous or costly to collect.