Title: Information and Willingness to Pay in a Contingent Valuation Study: The Value of S. Erasmo in the Lagoon of Venice
Abstract: Abstract This paper reports on a contingent valuation (CV) study eliciting willingness to pay (WTP) for a public program for the preservation of lagoon, beach and infrastructure in the island of S. Erasmo in the Lagoon of Venice, Italy. We use split samples to investigate the effect of providing a summary of reasons for voting in favor and against the program before the referendum valuation question. Reminding respondents of the reasons for voting for or against the program increases WTP among less highly educated respondents, and decreases WTP among more highly educated respondents. Acknowledgements This Research was funded by a grant from the Consortium for the Management and Coordination of Research on the Venice Lagoon System (CORILA) within the Research Programme 2000 – 2004: Research Area 1.1: The Economic Evaluation of the Protection of the Environment. The authors are grateful to Richard Carson, Luigi Fabbris and Andrea Galvan for their invaluable help and comments. The authors are also indebted to the participants of the workshop on 'Information and technology' held in Copenhagen, Denmark, April 2003, and the participants of the Session on Contingent Valuation III at the Twelfth Annual EAERE Conference, Bilbao, Spain, June 2003. Notes The method of Contingent Valuation (CV) is a well-established technique used to assign a monetary value to non-market goods and services, such as environmental resources (Mitchell & Carson, Citation1989). CV is a survey-based technique, in that it asks individuals to report their willingness to pay for a specified improvement in environmental quality. Willingness to pay is defined as the amount of money that can be taken away from a person's income at the higher level of environmental quality to keep his utility constant. It is, therefore, the theoretically correct measure of the welfare change—and hence the benefits—associated with the change in environmental quality. No benefit-cost analyses have been conducted for public works of limited scope, but some rudimentary benefit-cost work has been conducted for public works and engineering feats with broader, system-wide impacts. For example, efforts have been made to list the possible categories of benefits associated with the construction and operation of MOSES (Modulo Sperimentale Elettromeccanico, originally the prototype mobile barrier against high tides used for experimental and study purposes between 1988 and 1992, and later the name of the entire project for mobile barriers at the Lagoon inlets; see www.salve.it), but we are not aware of actual benefit estimates, and the results of this study are unpublished. If g( ) is different from f( ), but the expected value of q1 is the same, then WTP' and WTP will generally differ if V is a non-linear function of environmental quality. This can be shown using Jensen's inequality. The situation with non-stochastic q0 and q1 is obtained as a special case by letting f( ) and g( ) be degenerate distributions. What actually constitutes 'new' or 'different' information depends on the researcher's judgement. Fischhoff & Furby (Citation1988) study the effect of different levels of information on WTP for atmospheric visibility. They point out that different investigators choose different details in describing atmospheric visibility and conclude that the details provided in the scenario vary because investigators have different intuitive theories of their respondents' construal processes. Ajzen et al. (Citation1996) argue that the effect of information depends on the degree of respondent involvement with the good being valued. They found that if the good was highly relevant to the respondents, the latter were not sensitive to irrelevant cues such as priming procedures. In conditions of high personal relevance, willingness to pay increased depending on the quality of the cognitive arguments. In cases of low personal relevance, affective priming cues had a greater effect on WTP than did cognitive arguments. In Italy, the Regions are jurisdictions that are roughly comparable to the States in the US or the Provinces in Canada. Venice is the capital city of the Veneto Region. The referendum question was as follows: 'Suppose that the only way to finance the program is through a one-time income tax to be paid in 2002 by the residents of the Veneto Region, and that a Regional referendum were on the ballot to decide on this matter. If there were a majority of no votes, the program would not be implemented. If there were a majority of yes votes, the program would be implemented, and would be funded through the one-time tax. If the cost of the program to your household, to be incurred in 2002 only, were estimated by a team of experts to be €FILL, would you vote in favor or against the project?" Our vote question does not include a reminder of the respondent's income constraint, a practice suggested by the NOAA Panel (Arrow et al. Citation1993) and adopted by many practitioners. Studies based on split samples (Loomis et al., Citation1994b; Loomis et al., Citation1996) have found no discernible effect of such budget reminders on WTP. When budget constraint reminders have been folded into an explicit warning about hypothetical bias, they were found in some cases to exacerbate the possible positive hypothetical bias (Aadland & Caplan, Citation2003a, Citationb). For these reasons, and because our pre-test respondents found the income reminder distracting and even annoying, we chose to omit it from the vote question. Use values refer to the utility from direct consumption of the good. Non-use values are generally classified into existence, option, and bequest. In particular, existence is due to the utility an individual derives from the awareness that a good exists, even though the individual does not use it and will not do so in the future. Option value derives from the possibility to use the good in the future, as individuals cannot forecast their future preferences. Finally, bequest value is about the utility from preserving the good for future generations. See Freeman (Citation1993). Two hundred pre-test interviews confirmed that S. Erasmo is virtually unknown to people living 50 km or farther from the Lagoon of Venice, but that these persons nevertheless hold positive values for the public program. Our focus groups confirmed that people did not find a national-level program credible for such small-scale works. Champ et al. (Citation1997) report similar problems with Wisconsin residents and a program that would remove roads from the Grand Canyon, choosing to abandon the referendum format, and to replace it with a voluntary donation mechanism. SeeChamp & Bishop (Citation2001) for a discussion of incentive compatibility issues associated with voluntary donations. In the original plan we envisioned a treatment-control experiment for the entire sample. However, due to budget considerations, the experimental treatment was restricted to the groups that live farthest from the Lagoon, which we expected to be more responsive to the treatment. About one-third of the respondents do not report his or her family income. The average household income of €21 000 is calculated for those respondents who did report their household income. Official statistics show that women account for 51.5% of the population in the region where the survey was conducted, which implies that our sample overrepresents women. We believe that this is due to the legal restrictions on calling times our survey firm was confronted with, which made it difficult to reach working men. Self-selection into the sample on the part of women would bias our estimates of WTP if the likelihood of participating in the survey is correlated with WTP for the program, and both depend on gender. We do not believe this to be the case. First, when the interviewer first spoke to the person who answered the telephone, the topic of the survey was not explicitly mentioned to the respondent. Second, our regressions indicate that WTP does not vary systematically with gender. (We note that a formal test of the hypothesis of self-selection of women into the sample would be possible if we also had information about the people that were contacted but declined to participate in the study. If so, we would estimate a system of two equations. The first would be a probit explaining that the participant is a female, and the second would be an equation for WTP, with a gender dummy included among the regressors. The error terms of the two equations would be allowed to be correlated. One would conclude that there is no evidence of self-selection of women into the sample if the correlation between the two error terms was found to be insignificantly different from zero. Unfortunately, we do not have the data required to carry out this test.) An analysis of the motivations for the responses to the payment questions (see Alberini et al., Citation2004) suggests that most of our respondents' answers were consistent with economic behavior, and that very few of our subjects protest the public works or the valuation exercise. It is important that the comparison be done across control subjects (respondents who live in Zone C, D and E, but were not slated for the treatment) and treatment subjects. Were we to compare all subjects who did not receive the caveat treatment with those who did receive it, we would be making inappropriate inference, since respondents who live in the city of Venice and in zones A and B are more likely to visit S. Erasmo and hence are more likely to have greater willingness to pay for the program. We work with the Weibull distribution because Weibull variates are defined on the positive semi-axis and have a flexible shape. We compared the fit of the Weibull log likelihood with normal, log normal and exponential log likelihoods, and found that the Weibull outperformed the log normal and exponential, and was comparable to the normal. The standard errors around the estimates were computed with a simulation-based approach (see Alberini & Cooper, Citation2000). Mean WTP is greater than median WTP when the distribution is positively skewed, as is the case here, because the shape parameter of the Weibull, θ, is less than 3.6. Because the estimate of mean WTP is entirely driven by the upper tail of the distribution, median WTP is used as a robust lower bound for mean WTP. In conventional cost-benefit analyses, the total benefits of the project are obtained as mean WTP times the total number of beneficiaries. We present below total benefits computed in this fashion, as well as a more conservative, but statistically robust, benefit figure based on median WTP. Median WTP can also be interpreted as the largest tax amount that would still result in a majority approval of the project. Since only about 8% of our respondents had visited S. Erasmo in the last 12 months, we included both those persons that had visited the island in the previous year and those persons who had taken trips to the Lagoon, without necessarily visiting S. Erasmo, in this group. The Wald statistics for comparison of users and non-users are 10.26 for mean WTP and 46.09 for median WTP. The Wald statistics for comparison of potential users and non-users are 11.03 for mean WTP, 101.51 for median WTP. Under the null hypothesis of no difference, each Wald statistic is distributed as a chi square with one degree of freedom. The critical limit at the 5% significance level is 3.84. We created a dummy variable, PCAPINCMISS that takes on a value of 1 when the respondent did not answer the income question. Missing values in the income variable were then replaced with zeros, and both the income variable and PCAPINCMISS were included in the right-hand side of the WTP equation. The coefficient on PCAPINCMISS, therefore, captures any systematic differences in WTP between those respondents who did and did not report income. The coefficient on PCAPINC, the income variable, tells us how WTP varies with income, conditional on information on income being available. A similar procedure was followed for MALE and AGE. Based on a simple model with just the intercept and the CAVEAT dummy, we calculate that if the CAVEAT treatment changes mean/median WTP by 18%, the power of the t statistic for its coefficient is 0.15. The power of the t statistics increases to 0.43 if the CAVEAT treatment changes mean/median WTP by one-third, and to 0.93 if it changes mean/median WTP by one-half. We checked for a quadratic effect of age by adding aged squared in the right-hand side of the WTP regression, but the coefficient on this variable was insignificant. Table 8 assumes that the effects of CAVEAT, if any, are limited to the right-hand side of log WTP, while the error term ε in equation (3) is homoskedastic, its scale, θ, being constant and unaffected by CAVEAT. To check for a more complex effect of CAVEAT, we re-estimated the log WTP equation allowing for CAVEAT to enter in both λi and the scale parameter of ε: θ = exp (α 0 + α 1 · CAVEAT). Parsimonious specifications of this model were well behaved, but failed to detect any significant effect of CAVEAT on the scale of ε. α 1 was negative, implying that providing subjects with a reminder of the reasons for voting in favor or against the program tends to reduce the variance of log WTP, but this effect was very small (α 1 = − 0.05) and statistically insignificant. With broader specifications, the maximum likelihood routine often failed to converge. We also tried log normal specification for ε in lieu of the type I extreme value, but not even in this case did CAVEAT have a significant coefficient. To illustrate, if 20% of the households in zone A in our sample were users or potential users, and Nj is the number of households residing in this zone as per the most recent Census (2001), the total number of households with users or potential users in zone A is computed to be Nj U = Nj × (0.20). The total number of households in zone A without users would be Nj NU = Nj × (0.80). An anonymous reviewer suggests that an even more conservative calculation attaches a value of zero for those persons who declined to participate in the survey. Since these people account for about 37% of those contacted, this conservative recalculation results in total benefits equal to €25, 669, 207 (based on median WTP figures for the other 63% of users and non-users).