аЯрЁБс;ўџ ўџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџ§џџџGўџџџH  !"#$%&'()*+,-./0123456789:;<=>?@ABCDEFJўџџџIўџџџKLMNOPQRSTUVWXYZ[\]^_`abcdefўџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџRoot Entryџџџџџџџџ РF†ЇЇW #Л@CompObjџџџџџџџџџџџџbWordDocumentџџџџџџџџОObjectPoolџџџџ†з”D #Л†з”D #Лд 4@д ўџџџ ўџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџўџ џџџџ РFMicrosoft Word 6.0 Document MSWordDocWord.Document.6;ўџ ўџ р…ŸђљOhЋ‘+'Гй0X˜мDЈЬ №  8\€ ЄШь 4џџџџџџџџџџџџџџџџD:\WINWORD\TEMPLATE\NORMAL.DOTdiscussion paper 96-05Referendum Design and CV....>contмЅe-Р e"ƒОkfШыЎ˜Ў˜˜ВRъВ<юЕюЕюЕ$hЗhЗhЗhЗhЗ<ЄЗРЗdhЗЎМl$ИFjИLЖИЖИОИЦИE Й\gЙ0OЛQЛQЛQЛpЛММНTnНЉЎМюЕ—ЙCGОИТИ—Й—ЙЎМ—Й|Д€ДЖИ$И—Й—Й—Й—ЙоЕЖИцЕКИOЛЖ€’Жж&Г€ІГжˆД€Еж—ЙOЛ—ЙИ—ЙReferendum Design and Contingent Valuation: The NOAA Panel's No-Vote Recommendationfillin "ENTER TITLE OF PAPER (If title prints out longer than four lines, reduce font until it fits.)" fillin "ENTER NAME(S) OF AUTHOR(S) (If you reduce title font, reduce author font, but no smaller than 12pt.)"Richard T. Carson, W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert C. Mitchell, Stanley Presser, Paul A. Ruud, and V. Kerry Smith, with Michael Conaway and Kerry Martin Discussion Paper fillin "ENTER PAPER NUMBER (e.g. 94-01)"96-05 fillin "ENTER PUBLICATION DATE (e.g. March 1994)"November 1995 Resources for the Future 1616 P Street, NW Washington, DC 20036 Telephone 202-328-5000 Fax 202-939-3460 Љ 1995 Resources for the Future. All rights reserved. No portion of this paper may be reproduced without permission of the author(s). Discussion papers are research materials circulated by their authors for purposes of information and discussion. They have not undergone formal peer review or the editorial treatment accorded RFF books and other publications. Referendum Design and Contingent Valuation: The NOAA Panel's No-Vote Recommendation Richard T. Carson, W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert C. Mitchell, Stanley Presser, Paul A. Ruud, and V. Kerry Smith with Michael Conaway and Kerry Martin Abstract In 1992 the National Oceanic and Atmospheric Administration (NOAA) convened a panel of prominent social scientists to assess the reliability of natural resource damage estimates derived from contingent valuation (CV). The product of the Panel's deliberations was a report that laid out a set of recommended guidelines for CV survey design, administration, and data analysis. One of the Panel's recommendations was that CV surveys should employ a referendum approach. This method describes a choice mechanism that asks each respondent how they would vote if faced with a particular program and the prospect of paying for the program through some means, such as higher taxes. The Panel also recommended that CV referendum questions which commonly use only "for" or "against" answers should be expanded to explicitly offer an "I would-not-vote" response. The purpose of this paper is to consider the effects of such a "would-not-vote" option. In developing the test, we followed the important elements of the NOAA Panel guidelines for the design and administration of a CV survey and use what was acknowledged(by the Panel) as the most carefully developed CV questionnaire to that time, that is, the State of Alaska's study of the Exxon Valdez oil spill. Our findings suggest that when those selecting the "would-not-vote" response are treated as having voted "against" the offered program, offering the option does not alter: (a) the distribution of "for" and "against" responses, (b) the estimates of WTP derived from these choices, or (c) the construct validity of the results. Key Words: contingent valuation, natural resource damages, passive use, Exxon Valdez, reliability JEL Classification No(s).: D60, D61, K32, Q28 Table of Contents 1. Introduction 1 2. Hypotheses and Survey Procedures 4 A. Background 4 B. Survey Procedures 5 3. Results 7 4. Implications 13 References 15 List of Tables Table 1. Effects of Would-Not-Vote Option on Choices with Conservative Recoding 8 Table 2. Effects of Would-Not-Vote on WTP Estimates 9 Table 3. Selecting Would-Not-Vote and Its Effect on Construct Validity of CV 10 Referendum Design and Contingent Valuation: The NOAA Panel's No-Vote Recommendation Richard T. Carson, W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert C. Mitchell, Stanley Presser, Paul A. Ruud, and V. Kerry Smith with Michael Conaway and Kerry Martin 1. Introduction The increasing use of contingent valuation surveys to present respondents with economic tradeoffs for proposed programs has generated debate over the validity of this approach for measuring values for non-market environmental resources. As part of the efforts to develop regulations for damage assessment under the Oil Pollution Act of 1990, the National Oceanic and Atmospheric Administration (NOAA) appointed a Panel of leading social scientists, co-chaired by Kenneth Arrow and Robert Solow [see Arrow et al. 1993], to assess whether CV was capable of providing estimates of lost passive use values reliable enough for use in the assessment of natural resource damages. The Panel's report included a set of Guidelines, recommendations for future research and a general conclusion that: "...CV studies can produce estimates reliable enough to be the starting point of a judicial process of damage assessment, including lost passive use values." [Arrow et al., 1993, p. 4610] One of the Panel's recommendations was that CV surveys should employ a referendum approach. This method describes a choice mechanism that asks each respondent how they would vote if faced with a particular program and the prospect of paying for the program through some means, such as higher taxes. As Portney [1994] recounts, the Panel felt that the referendum format resembled the way people actually make choices regarding public programs. The Panel believed that adopting this format for CV would make individuals' responses more reflective of their "true" preferences. The Panel also recommended that CV referendum questions which commonly use only "for" or "against" answers (though interviewers are often instructed to accept "don't know" or "not sure" responses) should be expanded to explicitly offer an "I would-not-vote" response. The text of their report and subsequent discussion [Schuman, 1994] indicates that their objective was to mimic the practice of voting in which people can decide not to participate in the referendum. There have been many studies in the survey research literature of offering explicitly respondents a "not-sure" or "don't know" response option, but there has been no research conducted on the impact of a "would-not-vote" option in the context of a CV study. Several authors, including Schuman [1994], have conjectured, based on the results of the past studies, that offering a "would-not-vote" option would lower estimates of willingness to pay (WTP) derived from respondents' choices. The purpose of this paper is to consider the effects of such a "would-not-vote" option. In developing the test, we followed the important elements of the NOAA Panel guidelines for the design and administration of a CV survey and use what was acknowledged (by the Panel) as the most carefully developed CV questionnaire to that time, that is, the Carson et al. [1992] study of the Exxon Valdez oil spill conducted for the State of Alaska. Our findings suggest that when those selecting the "would-not-vote" response are treated as having voted "against" the offered program (Schuman's recommendation for a conservative coding), offering the option does not alter: (a) the distribution of "for" and "against" responses, (b) the estimates of WTP derived from these choices, or (c) the construct validity of the results. Section 2 outlines our basic hypothesis and the survey procedures. The third section describes our results, and the last section summarizes our conclusions and their implications. 2. Hypotheses and Survey Procedures A. Background There are at least two arguments that suggest hypotheses about the effects of implementing the NOAA Panel's recommendation. The first follows from the survey research literature dealing with "filter" questions [Schuman and Presser, 1981]. If the "would-not-vote" option is comparable to a "don't know" filter, then it should substantially increase the "would-not-vote" responses, though it may not alter the ratio of "for" and "against" votes. The Schuman-Presser analysis of attitudes for example, found that explicitly offering a "don't know" option did not affect the marginal distributions of the other response categories. Their result was generally insensitive to the subject of the questions they studied, and it likely underlies Schuman's suggestion that recoding these responses as "against" the program would lower the implied willingness to pay estimates. A second interpretation would hold that "not sure" and "would-not-vote" responses are expressing indifference. Under this view, offering the "would not vote" would not affect the number of respondents responding "for" and "against." They would be drawn from "not sure" or "don't know" categories. This argument follows from the axioms of conventional preference theory. Each individual is assumed able to evaluate every possible bundle as preferred, inferior, or equivalent (indifferent) to another bundle. Under the indifference view, recording "would-not-vote" responses as votes "against" would also lead to more conservative estimates of WTP. B. Survey Procedures Our test of the effects of offering the "would-not-vote" option was conducted as part of a larger experimental design. The analysis used the questionnaire developed to evaluate the Exxon Valdez oil spill and described at length in Carson et al. [1992]. The only modifications made to the questionnaire resulted from the change in the dates of the two surveys (1991 for Carson et al. and 1993 for our analysis). The National Opinion Research Center (NORC) of the University of Chicago completed 1,182 nation-wide face-to-face interviews of adult respondents in English speaking households in 34 counties throughout the United States. Of the 1,182 total surveys completed, 622 households were randomly assigned to either the original Alaska questionnaire or the version that included an explicit "would-not-vote" option. The two versions of the questionnaire used as the basis for the split sample analysis described the condition of Prince William Sound before the Exxon Valdez spill, as well as the damage caused by the spill. Respondents were then told that a plan had been proposed to prevent a future oil spill causing the same magnitude of damage as the Exxon Valdez to Prince William Sound. The plan involved setting up a fleet of escort ships that would guide oil tankers into and out of the sound and could prevent the spread of oil if some were spilled. Respondents were told that implementing the plan would cost their households a specified dollar amount, to be paid as a one-time addition to their federal income tax (either $10, $30, $60 or $120, randomly assigned to each respondent). In the standard version, respondents were asked to answer the referendum vote question and were not given the explicit "would-not-vote" option. Three hundred respondents received the standard version. Another 322 people were asked the same question with the addition of an explicit "would not vote" option. 3. Results Our analysis considers the effects of the "would-not-vote" option on: (a) the fraction of respondents selecting answers other than "for" or "against" the proposal; (b) differences in WTP estimates across the two samples; and (c) tests of construct validity. When the "would-not-vote" option was offered, the percentage of the sample selecting other than a "for" or "against" response significantly increased (p - value < 0.01) from 6.7% (20 of 300 respondents) to 17.7% (57 of 322 respondents). This confirms the Schuman conjecture that this type of filter would lead to a significant increase in the proportion of respondents selecting these types of answers. To evaluate the effect of a "would-not-vote" option on estimates of WTP, we first consider the distribution of "for" and "against" responses by tax amount, conservatively recoding the "not-sure" and "would-not-vote" responses as "against." If the distribution of "for" and "against" votes is not affected by the inclusion of the "would-not-vote" option, then the estimates of WTP should be similarly unaffected. Table 1 reports the results of this test with the relevant chi-square tests. In contrast to the hypothesis suggested by both frameworks described above, offering the "would-not-vote" option did not significantly affect the distribution of respondent votes at any of the tax amounts. Two different approaches were used to test for split sample differences in estimates of WTP for the prevention program, a median WTP based on a Weibull hazard model specification and a lower-bound estimate of mean WTP based on a Turnbull model specification. Table 1: Effects of "Would-Not-Vote" Option on Choices with Conservative Recoding ChoiceaStandard Version (%)Would-Not-Vote Offered (%)Tax Amount = $10 Vote For Vote Against n(2 = 0.58; p-value = 0.45 67.8% 32.2% 87 73.2% 26.8% 82Tax Amount = $30 Vote For Vote Against n(2 = 0.66; p-value = 0.42 56.1% 43.9% 66 49.4% 50.6% 87Tax Amount = $60 Vote For Vote Against n(2 = 0.27; p-value = 0.60 49.4% 50.6% 81 45.2% 54.8% 73Tax Amount = $120 Vote For Vote Against n(2 = 0.46; p-value = 0.50 33.3% 66.7% 66 38.8%b 61.3% 80a n = number of observations b Percentages do not add to one due to rounding error (e.g., percentages are 38.75 and 61.25 respectively. A Weibull hazard model, comparable to that discussed in Carson et al. [1992], was applied to the responses from the first discrete choice question. As shown in Table 2, there was no significant difference in the estimated parameters of the model across the split samples or in the implied estimates for the median. The table also reports the Turnbull [1976] lower bound mean. The Turnbull model estimates the fraction of the WTP distribution lying in each of the intervals defined by the tax amounts. The estimated lower bound mean uses these fractions to weight the lower end-point of the relevant interval. The unobserved mean is bounded from below by the estimated lower-bound mean. Table 2. Effects of Would-Not-Vote on WTP Estimates Weibull Hazard ModelaSampleNLocationScalelog (L)MedianbTurnbull Lower Bound MeancStandard3004.73 (15.28).90 (3.74)-198.3546.03 [30.40-69.68]52.81 (4.08)Would-Not-Vote3224.71 (16.53).90 (3.92)-212.8945.54 [30.64-67.70]54.01 (3.89)Combined6224.73 (22.51).90 (5.42)-411.2545.77 [34.38-60.95]53.50 (2.82)LRd0.02a The numbers in parentheses below the estimated parameters are asymptotic Z statistics for the null hypothesis that the relevant parameter was zero. b The numbers in brackets below the estimated median correspond to the 95% confidence interval. c The numbers in parentheses are estimates of the asymptotic standard errors. d The Likelihood Ratio (LR) statistic is distributed as an asymptotic chi square with 2 degrees of freedom. The test cannot reject the null hypothesis of equal parameters in the model underlying the observed choices with and without the would-not-vote option. Table 3: Selecting Would-Not-Vote and Its Effect on Construct Validity of CVa ProbitMultinomial Logit Independent VariablesSelect Would-Not-VoteAgainst PlanWould-Not VoteTax Amount.003 (0.99).017 (4.72).015 (2.66)Log (income)-.222 (-1.40)-.205 (-0.97)-.595 (-1.83)Self Identifies as Strong Environmentalist (=1)-.269 (-0.94).084 (0.26)-.431 (-0.71)Would Like to Visit Alaska (=1)-.685 (-2.31)-.511 (-1.62)-1.565 (-2.42)Protecting coastal areas from oil extremely important (=1)-.169 (-0.66)-.333 (-0.93)-.407 (-0.78)Government should set aside a "very large amount" or a large amount of wilderness (=1)-.485 (-2.04)-.678 (-2.28)-1.308 (-2.69)Caucasian (=1).086 (0.28)-.649 (-1.67)-.148 (-0.24)At least College Education (=1)-.525 (-1.37).486 (1.39)-.666 (-0.79)Below High School (=1)-.079 (-0.24).335 (0.69)-.095 (-0.14)Names Exxon Valdez as major environmental accident caused by humans (=1)-.157 (-0.60).110 (0.36)-.401 (-0.75)Respondent thinks less damage in absence of escort plan (=1).496 (1.92)1.355 (3.75)1.539 (2.84)Respondent thinks will be more damage in absence of escort plan (=1)-.130 (-0.34)-.969 (-2.20)-.856 (-1.45)Respondent thinks will be great deal more damage (=1)--1.223 (-2.24)-34.164b (-0.00)Respondent thinks proposal will prevent some damage (=1).318 (1.13)1.471 (3.71)1.546 (2.56)Respondent spontaneously protests payment, thinks Exxon should pay all the cost (=1)-.150 (-0.41).984 (2.43).418 (0.56)Respondent thinks Valdez spill more serious prior to interview (=1)-.287 (-1.08).141 (0.46)-.359 (-0.65)Intercept1.366 (0.88)1.550 (0.72)5.060 (1.60)Pseudo R2.189--Test of Equality (( 2, df = 20)-18.34 p-value =.304n322322 Table 3 (continued) a The numbers in parentheses below the estimated coefficients are asymptotic Z statistics for the null hypothesis that the relevant parameter was zero. b This large estimate for the coefficient for the effect of believing a new spill would create a great deal more damage on the likelihood of selecting the would-not-vote option. This arises because of the split of respondents between "for," "against," and "would-not-vote" choices in the two sub groups defined by this variable as detailed below: Respondent thinks a great deal more damage Choice No Yes For 142 25 Against 119 6 Would-not-vote 30 0 Total 291 31 An absence of respondents selecting the "would-not-vote" option who also felt there would be a great deal more damage is consistent with a priori expectations and the reasons why this measured coefficient is large and insignificant. There is no basis for discrimination on this dimension.  Both the test of the parameters of the Weibull hazard model and the Turnbull lower-bound mean are consistent with the conclusions drawn from the contingency table results. That is, there is no significant difference between the estimates of willingness to pay between the two samples when a conservative coding scheme is used to interpret respondents' "would-not-vote" choices as "against." The last "would-not-vote" issue concerns who selects this response and whether a test of construct validity would support the recoding of these responses to "against." Table 3 summarizes these findings. The first column is a probit model describing those who select the "would-not-vote" option (coded as 1 if a respondent indicated "would-not-vote" or reported "not sure" and 0 if a respondent voted "for" or "against"). The definitions for the independent variables correspond to those used in Carson et al.'s [1992] evaluation of construct validity for the original Alaska survey. We added three variables associated with the respondent's education and dropped two from the original set because there was no discrimination in those selecting the "would-not-vote" answer. Overall, few variables appear to be associated with the decision to select a "would-not-vote" response. The respondent's evaluation of the plan, interest in visiting Alaska, and belief that the government should protect wilderness areas all had plausible effects on the selection. None of the remaining variables were significant determinants of this decision. Thus, there was little basis for determining the characteristics of those selecting the "would-not-vote" (or "not sure") responses. One reason for this limited ability to explain these responses follows from the model itself. Our coding of the responses includes the "for" or "against" choices in the same category--as selecting the voting option. This is not likely to be correct. Indeed both the original Carson et al. analysis and our analysis suggest that several variables are important determinants of these different decisions. Often their effects arise in opposite directions for the "for" and "against" choices. This confounds attempts to explain respondents selecting "would not vote" over the choices associated with voting--either "for" or "against" decisions. The next two columns of Table 3 relaxes the "for" and "against" comparability assumption by using a three outcome, multinomial logit framework, distinguishing "for", "against", and "would-not-vote" (with the "for" category serving as the base or reference outcome). This model confirms the earlier Carson et al. analysis. The economic, attitudinal, and program related variables are significant determinants of respondents' choices. Moreover, this formulation allows the proposed recoding of the "would-not-vote" (and "not sure") responses as an "against" vote to be tested. This is accomplished by testing the restriction that the coefficients of "against" and "would not vote" are equal for each variable. As the chi-square statistic at the bottom of Table 3 indicates, this hypothesis cannot be rejected. Thus, tests of the construct validity of CV responses for this survey would not be influenced by the proposed conservative recoding of "would-not vote" and "not sure" responses. 4. Implications Offering a "would-not vote" option in referendum format CV surveys significantly increased the fraction of respondents who chose not to vote "for" or "against" a proposal offered to them in a CV choice. However, it also seems that these respondents would have voted "against" the proposal if the "would-not-vote" option had not been offered. This conclusion follows from the results of three separate tests. The first considered the distribution of "for" and "against" responses with the standard form of the question, recoding "would-not-vote" and "not sure" responses as "against." The marginal distributions by tax amount were not significantly different. The second type of test considered how these responses would influence a simple Weibull hazard model used to estimate willingness to pay. The estimated model's parameters were not significantly different across the split samples. Median and lower bound mean estimates for WTP were also found to be comparable in the two formulations for the referendum question. Finally, we considered the effects of the recoding on judgments about the construct validity of CV responses. Using the original construct validity specification proposed by Carson et al. [1992] as the basis for an expanded model within a multinomial logit framework, we cannot reject the null hypothesis that the coefficients for determinants of voting "against" equal those of selecting the "would-not vote" (or "not sure") option. Although these results relate to only one CV survey, taken together with Schuman and Presser's findings that filter questions are generally insensitive to the subject of the question, they suggest that offering a "would-not-vote" option in CV surveys (as proposed by the NOAA Panel's guidelines) is not likely to alter the conclusions derived from such surveys in relation to what they would be without them. References Arrow, K., R. Solow, P. R. Portney, E. E. Leamer, R. Radner and H. Schuman. 1993. "Report of the NOAA Panel on Contingent Valuation," Federal Register, vol. 58, no. 10 (January 15), pp. 4601-4614.Bergstrom, Theodore C., Daniel L. Rubinfeld, and Perry Shapiro. 1982. "Micro-Based Estimates of Demand Functions for Local School Expenditures," Econometrica, vol. 50 (September), pp. 1183-1206.Carson, Richard T., Nicholas E. Flores, Kerry M. Martin, and Jennifer L. Wright. Forthcoming. "Contingent Valuation and Revealed Preference Methodologies: Comparing the Estimates for Quasi Public Goods," Land Economics.Carson, Richard T., W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert C. Mitchell, Stanley Presser, Paul A. Ruud, and V. Kerry Smith. 1994. Prospective Interim Lost Use Value Due to DDT and PCB Contamination in the Southern California Bight. Report to National Oceanic and Atmospheric Administration, Natural Resource Damage Assessment, Inc., La Jolla, Calif., September.Carson, Richard T., W. Michael Hanemann, Raymond J. Kopp, Jon A. Krosnick, Robert C. Mitchell, Stanley Presser, Paul A. Ruud, and V. Kerry Smith, with M. Conaway and K. Martin. 1995. "Temporal Reliability of Estimates From Contingent Valuation," Discussion Paper 95-37, Resources for the Future, Washington, D.C., August.Carson, R. T., R. C. Mitchell, W. M. Hanemann, R. J. Kopp, S. Presser and P. A. Ruud. 1992. A Contingent Valuation Study of Lost Passive Use Values Resulting from the Exxon Valdez Oil Spill. Report to the Attorney General of the State of Alaska, Natural Resource Damage Assessment Inc., La Jolla, Calif., November 10.Diamond, Peter A., and Jerry A. Hausman. 1994. "Contingent Valuation: Is Some Number Better Than No Number?" Journal of Economic Perspectives, vol. 8 (Fall), pp. 45-64.Gilljam, Mikael, and Donald Granberg. 1993. "Should We Take Don't Know for an Answer?" Public Opinion Quarterly, vol. 57 (Fall), pp. 348-357.Hanemann, W. Michael. 1994. "Valuing the Environment through Contingent Valuation," Journal of Economics Perspectives, vol. 8 (Fall), pp. 19-44.Krosnick, Jon A. 1991. "Response Strategies for Coping With the Cognitive Demands of Attitude Measures in Surveys," Applied Cognitive Psychology, vol. 5 (May/June), pp. 213-236.Mitchell, Robert Cameron, and Richard T. Carson. 1989. Using Surveys to Value Public Goods: The Contingent Valuation Method (Washington, D.C., Resources for the Future).Portney, Paul R. 1994. "The Contingent Valuation Debate: Why Economists Should Care," Journal of Economic Perspectives, vol. 8 (Fall), pp. 3-17.Ready, Richard C., John Whitehead, and Glenn C. Blomquist. 1991. "Polychotomous Choice Valuation Questions," paper presented at the U.S.D.A. W-133 Meeting, Monterey, Calif., February.Schuman, Howard. 1994. "The Sensitivity of CV Outcomes to CV Survey Methods," paper presented at DOE/EPA Workshop on Using Contingent Valuation to Measure Non-Market Values, May 19-20.Schuman, Howard, and Stanley Presser. 1981. Questions and Answers in Attitude Surveys: Experiments on Question Form, Wording and Context (New York, N.Y., Academic Press).Turnbull, Bruce W. 1976. "The Empirical Distribution Function with Arbitrarily Grouped, Censored and Truncated Data," Journal of the Royal Statistical Society, B38 (no. 3), pp. 290-295.  The authors are, respectively: Associate Professor of Economics, University of California (San Diego); Associate Professor of Agricultural and Natural Resource Economics, University of California (Berkeley); Senior Fellow, Resources for the Future; Associate Professor of Psychology and Political Science, Ohio State University; Professor of Geography, Clark University; Professor of Sociology, University of Maryland (College Park); Professor of Economics, University of California (Berkeley); and Arts and Sciences Professor, Duke University and University Fellow, Resources for the Future. Conaway and Martin are members of Natural Resource Damage Assessment, Inc. and made extensive contributions throughout the effort. The work described in this paper was funded by the Damage Assessment Center of the National Oceanic and Atmospheric Administration as part of a natural resource damage assessment under contract number 50-DGNC-1-00007. Additional support to aid in the preparation of this paper was provided to Kopp by the Alfred P. Sloan Foundation through its support of the Welfare Economics Program at Resources for the Future, and to Smith by the UNC Sea Grant Program under Grant No. R/MRD-25. Thanks to Richard Bishop, Trudy Cameron, Nicholas Flores, and Alan Randall for comments on aspects of this work. All opinions expressed in this paper are those of the authors and should not be attributed to the National Oceanic and Atmospheric Administration, the Alfred P. Sloan Foundation, or any persons or organizations acknowledged above.  As used by the NOAA Panel, the reliability of a measure is the degree to which it measures the theoretical construct under investigation. However, in the empirical social sciences, this preceding definition pertains to validity, whereas reliability is defined as the extent to which the variance of the measure is not due to random sources and systematic sources of error. To avoid confusion and to maintain consistency with the NOAA Panel we are using the term reliability in this paper as meaning validity.  See Schuman and Presser [1981] for a detailed overview of the early literature and specific tests of offering "not sure" or "don't know" responses, and Gilljam and Granberg [1993] for a recent discussion in the context of another set of survey results concerning nuclear power in Sweden.  A common practice in applied CV research has been to drop the "don't know" from the analysis. Under this approach, offering a "would-not-vote" response and treating them as comparable to "don't know" responses could increase the estimated WTP measure. The proposal to include "would-not-vote" as against responses was not the Panel's recommendation. It was a suggestion made later by Schuman [1994].  Construct validity refers to the degree to which a measure relates to other measures predicted by theory. As a rule, two forms of construct validity are considered: convergent validity and theoretical validity. The former refers to whether the measure of interest is correlated with other measures of the same theoretical construct and is not applicable to the research proposed here. For an example of a test of convergent validity comparing CV estimates to indirect estimates, where it is possible to obtain both in each study see Carson et al. [forthcoming].  In what follows, we have treated the voluntary reporting of "not sure" or "would-not-vote" as equivalent. We did undertake separate analyses of their effects in our evaluation of the "would-not-vote" responses and our primary conclusions are unaffected by separating them. There are a small number of responses in the two categories so it is difficult to distinguish separate factors influencing respondents' decisions to indicate "don't know" versus "would- not-vote."  This argument is comparable to the approach followed in the literature reporting micro estimates of the demand for public education. The format of these survey based estimates is comparable to referendum CV surveys except the questions are posed as "more," "less," or "the same" as the current amount of expenditures. Not sure could be interpreted as preferring no change in the status quo. See Bergstrom, Rubinfeld, and Shapiro [1982] for an example of these applications of CV to public education.  Other studies have experimented with offering more than "for" and "against" responses. See Ready, Whitehead and Blomquist [1991] as an example. As a rule these have been designed to parallel the framing used in attitude questions and not the decision process associated with referenda. There is no basis for relating their findings to those with a format designed to more closely correspond to voting choices.  The entire design is described in Carson et al. [1994] and [1995].  See Carson et al. [1992] for a discussion of the questionnaire development.  Although the counties are both urban and rural from all regions of the US, they were chosen on the basis of interviewer availability from the NORC national master frame.  The Turnbull [1976] non-parametric, maximum likelihood estimator for interval-censored data uses respondents' choices to the voting questions to estimate the latent willingness-to-pay implied by each respondent's choice (i.e., vote). As noted above, an individual's answer to a single question will distinguish either a lower or upper bound for his or her WTP. By combining respondents' choices, we obtain estimates for the relative frequency of responses at different WTP intervals, (0, W1AMTi) and (W1AMTi, (), where W1AMTi is one of the four tax amounts administered to the different sub-samples. The first pair, (0, W1AMTi), defines the interval identified by W1AMTi as an upper bound and, the second pair, (W1AMTi, (), with W1AMTi as a lower bound. The five intervals defined by W1AMT are: (1) $0 to $10, (2) $10 to $30, (3) $30 to $60, (4) $60 to $120, and (5) above $120.  The two variables dropped correspond to qualitative variables (0,1) identifying those respondents who think there would be no damage from another oil spill and those indicating the proposal will not reduce the damage at all. The education variables were added to take account of Krosnick's [1991] arguments that "not sure" responses may be related to satisficing behavior in answering attitude questions.     - PAGE ii- - PAGE iii- - PAGE 8- Carson, Hanemann, Kopp, Krosnick, Mitchell, Presser, Ruud, and Smith Referendum Design and Contingent Valuation: The NOAA Panel's No-Vote Recommendation - PAGE 7- аЯрЁБс;ўџ ўџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџђ“Ё–—™?@І0Ї0ЈрЋ…д€€€€€€€€€а....()()))()()ђŽЁ–—™@І0Ї0ЈрЋ…д€€€€€€€€€а....()()))()()аЯрЁБсSummaryInformation(џџџџџџџџџџџџˆџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџingent valuation, would-not-vote option, discussion paperMarjorie B. McElroyMarjorie B. McElroy@†Ÿј' #Л@@†Ÿј' #ЛаrT@Microsoft Word 6.02аЯрЁБс;ўџ аЯрЁБс;ўџ ўџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџSTЛМНОП23ТУьэўџ()./01378rs€‚я]^ВktvИшQ ЋД(@Rю§м0чш_`€MN Л!М!е#ж#ѓ%є%‚&ƒ& ' '‰(Š(Я1ж1Ÿ3Х3Ц3ќѕќѕќѓэыэыэыѓщущущущсэыэыэыѓолййзззггоЫЫЫЫЫЫЫЫЫЫЫЩЧ` UuDPcU^c^]^cUcc uDccc uDcc uDUcUcNЦ3э3ѓ3G4H4T4U4‡4—4Б4В4Г4э4§4555S5c5}5~55Й5Ъ5ф5х5ц566!6"6@6A6Ў6Q7X7s7в7+8Љ8Њ8g99ž9Д9Е9п9р9њ9ћ9ђ:ѓ:џ:;–;—;ї;ј;F<G<M=O=œ==ž=Ÿ=PBQBоCпCќC§CўC§јєёєьєцсєьєцсєьєцсєьєцсєёєёєёєпнлгјЯёЯЪЯЪЯЪЯХЯХЯХЯХЯјХјУЯХЯОЯИЯ Jc]c]chc]ch]ch]cuDPc``U]ch Jc]c]^c]h]cU]c` HўCџC,D.D/D6DDDEDFDоDпDjFpFqFsFtFwFEGMGпGсGуGJJuLvLЩ\й\œ]Ј]ž^Ќ^I_А_жa7b*cJcСcйcPdqde e|eСeJfjf+hˆh$iLijikili€oo‚o^pfpq‚qЃrЄr8t9tpvqvJxKxCzDzт{у{'|(|u|v|!}"}ћїѕ№ьчїчїтїтїтїнїьїлгбббббббббббббѕггЯбггггггггггЭeњџcVuDPcUV]c]^c]ch]cU]cc]c]chQ #$23˜™ФХєѕјљ€€™€š€›€3‚4‚6‚7‚9‚:‚<‚=‚@‚A‚G‚H‚J‚K‚L‚O‚P‚V‚W‚Z‚[‚\‚g‚h‚n‚o‚p‚q‚s‚З‚Й‚ ƒƒƒƒƒƒƒƒƒƒ ƒ!ƒ"ƒш…§њ§§§§њ§ђ№ыыыыхухухухухухухухухупмхухухукиucVcP"cP" uDP"uDcuDPcJЅeњџ@НОьэ01234567‚ƒ„…†ŸБЧоя№ё(xy]ВГCDјp#3ђp#Жьp#цp#Ъьp#оp#гp#*гp#*ьp#ьp#ьp#ьp#ьp#ьp#ьp#ьp#ьp#Эp#BЭp#BЭp#BЭp#BЭp#BЭp#BЧp#Чp#Чp#Чp#Чp#ООp#Еp#UОp#Ќў0§№0§xќўxРў x№ар"x№xx8џx№xLџx№  DIjktu№ а ЊЋ>?RSTgŽ ЙЧлъыьэю§їp#їp#яp#6хp#мp#бp#бp#бp#Чp#Мp#Мp#Чp#мp#хмp#мp#Ћp#Ћp#Ћp#Ћp#Ћp#Ћp#Ћp#p#p#p#p#хp#K №,аА`р" №,аА`р" ,ў№ар" №ар" №,ар"№,0§ №,0§,0§№0§§ўџR‰клм01СТЧщъыћгдяиЧШя§f#ѕ%їp#їp#цp#цp#цp#цїнp#hнp#дp#.нp#нp#нp#KЩp#Оp#Мp#рКџp#Кp#6Аp#6Љp#6Кp#6Кp#6Кџp#v Кџp#JКp#6Кp#6Мp#рЇp#рКџp#PКџp#,0§ аа,0§ №x,0§ №x,0§дў0§№0§№,а0§„р" №,0§ѕ%і% &F)‹-Œ-—-00ы2ё3ѓ3F4G4H4V4k4†4‡4Б4л4ь4ўp#6ќp#рўџp#Аў p#6ўp#6њp#рўp#6ўџp#Иѓp#6юшp#№оp#№юp#е@ Ч@ це@ цІ@ цš@ ц‰џ@ ~}@ ц №0§№xx рў№xxар" АPў№xx ИlЛО(”џЌ Фм#   №№Шар"№№Ш ИlО”џм#№№”џр"ь4э45A5R5S5}5Ї5И5Й5ф56 6!6@6­6Ў6Џ6А6,8п@ цг@ цЧџ@ ~Л@ цЉ@ цг@ цЧџ@ ~Л@ цЉ@ цг@ цЧџ@ ~Лџ@ ЅЉџ@ Ѕ˜p#‰џp#ztџp#zop#op#kџp#о:џ№ИlЛО”џм#h(џ№xар"h(џ№Дxар"ИlЛО ”џЌ Фм# №0§№xx  рў№xx АPў№xx ИlЛО(”џЌ Фм#џџџџџџ,8f9g9œ99ž9Ÿ9 9Ж9З9И9Й9Р9Т9Ы9б9й9с9ќ9ўљp#ѓn!№щn!№љp#тcгтТгкfтгт’гЗ’гЎcгЎТгЄcгЄcгЄcгššџ’ №x<  №x<№x<"ИOЛО,бг3 9й $ №x№x ИOОбл"№№ќ9§9: ::":*:>:K:L:[:_:l:w::“: :Чџ’ОcгОТгДcгДcгДcгЊгЊ’гq’гОcгОТгДcгДcгДcгЊгЊ’г8ИOЛОXбг3 579й $џџџџџџџџџџџџџџ №PP  №PP№PP8ИOЛОXбг3 579й $џџ џџ џџ џџ џџ џџ џџ  :Ё:Њ:Ў:Л:Ц:Ю:т:я:№:є:ѕ:і:ї:ќ:§:ў:џ:–;ї;F<L=M=ъ’гсcгсТгзcгзcгзcгЭгЭ’гъ’гсcсТгзcгзcгзcгЭгЭ’гъ’гМџ˜"zЋ˜".Ћ˜".™џ˜"‹„џ˜"‹ИOЛОб $ рў№<x ар"  рў№< ар"  рў№x ар"  №PP  №PP№PPИOЛОбг3 579й $M=N=ž=Ÿ= =Ї=Й=К=б=ч=ѕ=>>>>(>4>њp#ёшp#№л‚гЭ’гЭЂ гЛЂ гЎ‚г ’г г гyгl‚г^’г^г^г №x<,0§ №x<,0§&ИOЛО4$DtД$џџџџџџџџ №Д<,0§ №Д<,0§ИOЛО $DtД$ №P<,0§ №P<,0§№,0§№,И№x4>5>B>P>^>l>m>>Ћ>З>Х>Ц>ц>є>???M?[?i?w?x?Я?н?ы?њ?ћ? @@$@2@3@S@a@m@{@|@“@Ё@эгр‚гг’гггггэгр‚гг’гггггэгр‚гг’гггггэгр‚гг’гггггэгр‚гг’гггггэгр‚гг’гггггэгр‚гг’гггггэгр‚гг’г №<,0§ №(<,0§ИOЛО )IyЙ$&Ё@­@Л@М@AAA-A.AkAwA„A‘A’AзAхAѓABB8B:BIBZB[B”B B­BКBЛBCC*C6C7C{C‰Cѓгѓгргг‚гѓ’гѓгѓгргг‚гѓ’гѓгѓгргг‚гѓ’гѓгѓгргг‚гХ’гѓгѓџ@рџ@г‚гѓ’гѓгѓгргг‚гѓ’гѓгѓгргг‚гѓ’г №Z<,0§ №(<,0§ИOЛО )IyЙ$ №<,0§#‰C•CЃCЄCЎCЛCШCеCжCрCхCчCщCъC D D D!D#D'D+D,D-D.DDDEDоDѓгѓгргг‚гѓ’гѓгѓгргг‚гХ’гХгХгрггџ‚Х’гѓЂ гГЂ гг‚гХ’гХЂ гГЂ гЊp#№Њp#№ЄЄp#’џ$ЫАрў№xP ар"Аџ№№,0§ИOЛО )IyЙ$ №Z<,0§ №(<,0§ИOЛО )IyЙ$ №<,0§оD;F‚?‚M‚N‚]‚^‚_‚`‚a‚b‚љp#рїџp#шїџp#ˆїџp#Јїџp#˜їџp#ˆїџp#ˆїџp#ˆїџp#˜їp#Kїp#Kїџp#Иїџp#[ їџp#˜ѕяѕчѕмѕдѕЧѕЧp#ѕЛp#ѕЛp#ѕ #№ арР!р" !№ арР!р"P№№ №Шар"P№x№x,0§b‚c‚d‚e‚f‚И‚Й‚ƒƒƒƒƒƒ ƒ!ƒ"ƒѕѓчѓѕp#ѓѕp#ѓѕp#ѓѓѓѓѓрp#р,0§ #№ арР!р" !№ арР!&K0@ёџ0Normalр3 ар"a c.@. Heading 1 А U[]c,@, Heading 2 АU]c,@, Heading 3 А]^c"A@ђџЁ"Default Paragraph Font.@.TOC 1аа0§р,(#$,@,TOC 2а 0§,(#$,@,TOC 3аp0§,(#$,@,TOC 4а@ 0§,(#$,@,TOC 5а0§,(#$(@(TOC 6а0§,(#$@TOC 7 а0§,(@(TOC 8а0§,(#$(@(TOC 9а0§,(#$. @.Index 1а `њ,(#$. @.Index 2а 0§,(#$(.@( TOA Heading,(#$"@CaptionўOђџС_Equation Caption4@в4 Footnote Text№<<Аc,&@Ђс,Footnote Reference]`ceh +@ђ Endnote Textc *@Ђ Endnote Referenceh @ Header !рР!)@Ђ! Page Number @2 Footer #рР!"ўOB" Right Par$а]ўORPages%]cч_M Ле ѓ"‚# $‰%Љ5uI"€ 8 Э  пиwМ Ж.ЧЪџџџџџџџџџџџџџџџџџџџџџџџџџџ џџ џџ џџ џџ џџџџџџџџЭ п ЖЖЖ...ЧЧЧЧЧЪ "€џџџџџџџџџџџџџџџџџџџџџџџџџџ џџ џџ џџ џџ џџџџџџџџ м "€„џџџџє„џџџџ !џџ!џџ!џџAџџ Aџџ@џџ@џџ@џџ@џџ@џџ@џџ@џџAџџ Aџџ Aџџ @џџ @џџ @џџAџџAџџAџџ^@ м м &Рf o$Œ*ђ0-5O:/A›H№N-V5YК_Ќe"€T В/ Є  9 O мрm ЌН +-/13†шъэЦ3ўCш…CDEFD§ѕ%ь4,8ќ9 :M=4>Ё@‰CоDбTjib‚"ƒGHIJKLMNOPQRSTUVSЛО2Тў(.7r€"€''џ„'џ„'џ„ #'4;=лтфэ!џ€!џ€!џ€!џ€"€эEabstractpg_1sec_2sec_3tbl_1tbl_2marktbl_3sec_4endbibliogtч е*г.U48GжQ3Y7Y#€ tч е*г.U48GжQ3Y7Y#€ИJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCJoy Hall!M:\DPS\QE96\05_CARSN\9605_TXT.DOCSchool of the Environment!C:\DOC1\MANUSCRI\NOVOTE\022796.NVMarjorie B. McElroy B:\NOVOTE.GAMџ@HP LaserJet IIILPT1:HPPCL5MSHP LaserJet III  DLg–WЧ,,№ЯО;W0d}w;WВ}dГ Яœ€#џџџџџџџџџџџџџџHP LaserJet III  DLg–WЧ,,№ЯО;W0d}w;WВ}dГ Яœ€#џџџџџџџџџџџџџџ€jfjf€jfjflTimes New Roman Symbol &ArialTimesTimes New Roman CG TimesCG Times (W1)n!ˆЇаЖ,І,ІаrT+ДФЉdiscussion paper 96-05Referendum Design and CV....=contingent valuation, would-not-vote option, discussion paperMarjorie B. McElroyMarjorie B. McElroyаЯрЁБс;ўџ ўџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџџ