assessing perceived consequentiality: evidence from a contingent valuation survey on global climate...

Upload: maninepal

Post on 30-May-2018

214 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    1/16

    Assessing Perceived Consequentiality: Evidencefrom a Contingent Valuation Survey on Global Climate

    ChangeMani Nepal, a# Robert P. Berrens b and Alok K. Bohara c

    aCentral Department of Economics,Tribhuvan University, PO BOX 11520, Kathmandu, Nepal

    Email: [email protected].

    bDepartment of Economics,University of New Mexico, MSC05 3060, Albuquerque NM 87131, USA

    Email: [email protected] of Economics,

    University of New Mexico, MSC05 3060, Albuquerque NM 87131, USAEmail: [email protected].

    (Forthcoming in International Journal of Ecological Economics and Statistics, Vol

    14, P09, Spring 2009, pp 14-29)ABSTRACT

    It has been argued that a key element for any contingent valuation (CV) study to provide valid preference information is that the survey must be perceived as potentially consequential to the respondent. In light of sustained debate about hypothetical bias, validity and the quest for consequential CV questions, we propose that respondent-based, self-assessed indicators (taken from a series of follow-up questions) could be used to investigate the consequentiality of a CV survey. Using an available data set, this exploratory investigation uses a set of such survey questions to construct four different indices of perceived consequentiality. Across these indices there is evidence of a significant consequentiality effect on U.S. household willingness to pay for efforts to mitigate global climate change, and significant differences in the preference functions of

    sub-samples with high and low indices of consequentiality.Key words: Consequentialism, Contingent valuation, Global climate change, Kyoto Protocol,

    Willingness to pay

    Journal of Economic Literature (JEL) Classification Number: C42, C81, Q54

    Mathematics Subject Classification Number : 62F03, 62F10, 91B76

    1. INTRODUCTION

    Efforts to assess the economic value of changes in environmental goods have global applications.Methodologically, interest in the application of survey-based contingent valuation (CV) approaches toelicit monetary values for changes in non-market environmental goods remains strong, despitepersistent concerns over validity (Boyle 2003; List et al. 2004). It has been argued that valid studieswill use valuation questions that are perceived as consequential by the respondents (Carson et al.2000; Carson and Groves 2007). Thus, the focus is on eliciting individual valuation responses, whosevalidity is dependent upon the participant perceiving the questions as consequential. So, rather thanonly thinking of the analyst operating from some larger utilitarian lens (moving from consequences to

    # Address all correspondence to: Mani Nepal. The work was done while the first author was a visitingassistant professor at the Department of Economics, University of New Mexico, Albuquerque, NM.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    2/16

    individual values to aggregation for use in benefitcost analysis, or natural resource damageassessment), we also have to investigate whether the participants themselves perceive the processfrom their own consequentialist perspective. Do they think their survey answers are likely to be usedby policy makers, are a good way of collecting information, or will have an impact on policy in someway? Identifying consequential survey questions is a key area for further investigation of the validity ofstated preference approaches (List 2004).

    Given persistent concerns over potential upward hypothetical bias in CV surveys, there is a need forvalidity assessments of the performance characteristics of alternative CV formats (Boyle 2003;Carson et al. 2001). It is argued that consequentiality may be tied to particular elicitation formats. Forexample, consequentiality may be enhanced by the use of a referendum voting format, which hasbeen the source of considerable debate (e.g., Arrow et al. 1993; Carson et al. 2000; Champ et al.2002; Harrison 2006; Carson and Groves 2007). The intention here is not to investigate thetheoretical and empirical debate about the incentive compatibility of the referendum format. Rather,we argue that the assessment, as opposed to the simple assertion by any given researcher ofwhether or not a CV survey is perceived as consequential warrants scrutiny. The objective is to pushforward the investigation of how we might go about actually assessing a consequential CV question.

    Using an advisory referendum format, and sample data from previously published work (Berrens et al.2003, 2004), we investigate some initial alternative CV survey follow-up questions that could beconsidered as indicators of consequentiality. If such questions merit consideration, then in the idealthe CV discourse community (Randall 1993) might begin moving towards some consensus forcodifying or standardizing such an assessment.

    2. A CONSEQUENTIAL CV SURVEY QUESTION DEFINED

    The application of the survey-based contingent valuation (CV) method to elicit individual values forchanges in nonmarket public goods is clearly within the utilitarian/welfarist tradition in economics (Sen1993; Castle 1999). More generally, traditional economic approaches, such as benefit-cost analysis

    (BCA), to collective choice questions have their philosophical roots in utilitarianism. Further, thenormative basis of utilitarianism is considered to have three basic elements: consequentialism,welfarism (i.e., methodological individualism and value measures) and sum-ranking for comparingalternatives (Sen 1987; Hamlin 1989). Consequentialism evaluates all policy choices solely on thebasis of their consequences (outcomes) for alternative social states, 1 and combines with welfarism tofocus on the individual values attached to any given set of consequences. While generally leftunstated, these basic assumptions are a foundation for much of the applied works in environmentaleconomics (Castle 1999). For example, we can think of the analyst as operating from behind autilitarian lens as they conduct a CV survey, and then use the results in a BCA or other planningprocess.

    Recently, the concept of consequentialism, of a much more particular stripe has emerged in theenvironmental economics literature - in the debate concerning the CV method for eliciting preferencesfor non-market environmental goods. In fact, it is probably better to drop the -ism altogether, and talkabout the concept of a consequential [CV] survey question (Carson et al. 2000; Carson and Groves2007).

    In a paper originally presented as plenary address to the 1999 European Association of Resource andEnvironmental Economists (Carson et al. 2000), and more recently published in Environmental and Resource Economics (Carson and Groves 2007), it was argued that CV survey studies would be valid

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    3/16

    if the survey questions were consequential. Two criteria are proposed that would make surveyquestions consequential: (1) the agent answering a preference survey question must perceiveresponses to the survey as potentially influencing agency action; and (2) The agent needs to careabout what the outcome of the action is. More explicitly, Carson and Groves (2007, p. 183) defineconsequential and inconsequential CV survey questions in the following way:

    Consequential CV survey : If a surveys results are seen by the agents as potentiallyinfluencing an agencys actions and the agent cares about the outcome of that action, theagent should treat the survey questions as an opportunity to influence those actions. In sucha case, standard economic theory applies and the response to the question should beinterpretable using mechanism design theory concerning incentive structures.

    Inconsequential CV survey : If a surveys results are not seen as having any influence on anagencys actions or the agent is indifferent to all possible outcomes of the agencys actions,then all possible responses by the agent will be perceived as having the same influence onthe final outcome. In such a case, economic theory makes no predictions.

    Thus, if the agents think that agency would use their inputs in the policy making, and if the agents

    care about the outcomes due to the changes in the policies, then such type of survey questionsprovide incentive to the agents to reveal the truth about their preferences. Such survey valuationquestions would then be classified as consequential (Carson and Groves 2007). But would all agentsbe the same?

    Based on this concept of a consequential survey question, Carson and Groves (2007) examine anumber of key issues to, illustrate both the power and the limitations of economic theory to interpret alarge body of empirical evidence about such questions. Issues include: the performance andincentive compatibility of various binary discrete choice (DC) elicitation formats (including referenda);reasons for consistency or inconsistency of responses to repeated DC questions; comparisons of theperformance of DC and open-ended (OE) formats; and comparisons of valuing a single good versusvaluing multiple goods.

    The lack of consensus on CV survey procedures and the probable biases demand that a CV surveymust be designed, implemented, interpreted and reported carefully (Mitchel and Carson 1989; Carsonet al. 2001). Absent any better alternatives to providing information about passive use values forenvironmental goods, and persistent controversy about the validity of the CV results, some form ofconsensus methodology is needed to make the use of the CV survey outcome less disputable.

    Over the last several decades, a number of attempts to move towards some type of consensusmethodology in the CV literature have become focal points and touchstones for future research.Prominent examples include the Reference Operating Conditions (ROC) of Cummings et al. (1986),and the proposed NOAA Panel guidelines of Arrow et al. (1993), as sponsored by the U.S. NationalOceanic Atmospheric Administration. While something short of a full set of guidelines, the concept of

    consequentiality and a consequential CV survey question (Carson and Groves 2007) is likely to be atouchstone, if not a lightning rod, for future CV studies and discussions. For example, it is common tosee valuation researchers indicate that the incentive properties of CV survey questions differconsiderably based on whether or not the question is viewed as consequential. 2 While there my besome ex-ante assertions about elicitation formats (e.g., referendum), this is something different thanex-post assessments of whether a particular survey and sample is consequential.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    4/16

    3. ASSESSING A CONSEQUENTIAL CV SURVEY QUESTION

    As CV survey studies have been frequently used in the valuation of non-market goods, the question ishow we can objectively test the consequentiality of CV questions. Even assuming the use of apreferred format (i.e., public goods referendum), is it always possible to assert that a survey isperceived by the respondents as consequential? Researchers and sponsors may have an incentive

    to make this assertion, especially after the money is spent and the survey data is collected. Someassertions will stand up to scrutiny and professional judgment better than others. Various aspects ofsurvey design, and the overall study program, are typically taken to be indicators of data quality, andpossibly by inference of the perceived consequentiality of the survey. Beyond the standard structuralcharacteristics (e.g., contingent scenario, definition of the good, payment vehicle, elicitation format,and implementation rule, experimental treatment design, follow-up questions to account for responseuncertainty), the list of possible process indicators is likely to include: use of focus groups and scopingsessions, pilot tests and pre-testes of the survey instrument(s), professional editing of the text, highresponse rates, and large sample size, etc. At some level there is validity in this argument, e.g., inensuring that you have drawn a representative sample and created coherent contingent scenarios.The message for researchers is that if you do these things and check all these boxes, then you are

    likely to improve the quality of the survey data.Our concern is that such aspects of survey and study design are likely to be incomplete and imperfectindicators of consequentiality. It is possible that a research project might check all these boxes andstill end-up with some significant fraction of the respondents who simply do not perceive the exerciseas likely to be consequential. Here the issue can clearly began to be seen in relative terms.Specifically, what fraction of the respondents is necessary or required for it to meet some qualitystandard in the CV research or discourse community? How sensitive is this fraction to the type ofgood, and the characteristics of the target population (socioeconomic status, religion, culture, and soon)?

    If possible, making CV surveys consequential would be the first best solution. But even following

    state-of-the-art methods and protocols, researchers may have limited control over how a respondentperceives the surveys potential consequentiality. Use of follow-up questions in CV surveys iscommonly recommended for a variety of reasons (e.g., Arrow et al. 1993; Boyle 2003), andresearchers can include standardized questions to help assess perceived consequentiality. Further,they can also report the value of the goods in questions for the entire sample as well as for the split-samples based on the degree of consequentiality. Such a portfolio approach of presenting varyingvalues by varying perceptions of consequentiality may be more informative than the existing singlevalue approach.

    Here we propose that respondent-based self-assessed indicators could be used to help answer someof these questions. Thus we argue that self-assessed measures or scores could be used asindicators of perceived consequentiality. We argue that it is possible, and even desirable, to constructa set of one or more fairly standardized questions that could be used across different studies andresearch programs for making comparisons. Consistent with Randall (1998), this would be consistentwith efforts to map the performance characteristics (and in particular, the perceived consequentiality)of different CV instruments. This could include comparisons from studies examining hypotheticalversus real behavior in split samples for public good provision. Various standards might beestablished for assessing consequentiality. Reaching such a point is likely to require discussion anddialogue in the CV discourse community.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    5/16

    4. EXAMPLES OF RESPONDENT-BASED ASSESSMENT QUESTIONS

    How might we design a survey instrument for assessing respondents perceptions of potentialconsequentiality? To answer this question we use a survey instrument (see Berrens et al. 2004) thatwas used in a matching telephone (TEL) sample survey and three web-based Internet samplesurveys. These instruments were developed to conduct an advisory referendum to evaluate the

    perception of US citizens towards the provisions of the Kyoto Protocol (KP) or mitigating globalclimate change (GCC). The contingent scenario and the elicitation format were based on a nationalreferendum to advise US Senators on ratifying the KP. The proposed payment vehicle was theincreased market prices of gasoline and energy. In all four surveys, immediately after the valuationsection, respondents were asked the following self-assessed questions, which we label as Q1-Q3(with corresponding variable name below).

    Q1. ( TREATY): Government officials in the U.S. are currently considering a proposedinternational treaty that concerns global climate change, called the Kyoto Protocol. In 1997,representatives from U.S. and approximately 150 other nations developed and signed theKyoto Protocol, which calls for reducing the production of greenhouse gasses. The U.S. hasnegotiated similar treaties with other nations to try to deal with other environmental problems,such as acid rain and ozone depletion. Using a scale where zero means it is a very bad idea,and ten means it is a very good idea, how do you view international treaties as a way to dealwith environmental problems?

    Q2 . (SENATE): On a scale where 0 means "it is certain that the Senate would not seriouslyconsider the results of a national vote or referendum", and 10 means "it is certain that theywould", how certain is it that the Senate would give the results serious consideration indeciding how to vote on the Modified Kyoto Protocol?

    Q3. (KYOTO-EFFECT ): Suppose that the Kyoto Protocol were ratified by the Senate, and theU.S. worked with other industrialized countries to reduce greenhouse gas emissions. Using ascale where 0 means "you are certain that the Kyoto Protocol will have no effect on global

    warming and 10 means "you are certain it will reduce global warming", how likely is it that thetreaty would reduce global warming?

    In addition to these three self-assessed questions, the following two good way questions were alsoasked to respondents in order to get information about their perception towards the nationalreferendum, and about willingness to pay questions as the means of collecting information. Thesequestions are labeled as Q4 and Q5 (with corresponding variable names below).

    Q4. (GW-REF): In your view, would a national referendum be a good way for citizens toexpress their preferences regarding the Kyoto Protocol to the U.S.Senate? [Options: Yes/ No ].

    Q5 . ( GW-WTP): Frequently, U.S. residents are asked about their willingness to pay indonations or higher prices to address environmental problems such as global climate change.The answers are often provided to government officials to assist in developing policies toaddress these problems. Is this a good way for government officials to make policy choicesabout global climate change? [Options: Yes/ No, & Not Sure ].

    The summary statistics of the responses to these self-assessed questions are presented in Table 2.Our conjecture is that responses to these self-assessed questions in the survey may provide somebasis for measuring or indicating the individuals perceived consequentiality of survey instruments.Question Q1 captures the respondents perception towards international treaties as a way of dealing

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    6/16

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    7/16

    an equal weighing scheme. Another adjustment in the data is the treatment of Not Sure option alongwith Yes and No options in the case of the responses to the question Q5 . We treat this Not Sure option equivalent to No response. 4

    6. MODELING APPROACH AND HYPOTHESES

    The dependent variable in this study is the VOTE by the respondents for or against ratification of the

    KP by the U.S. Senate. More specifically, the advisory referendum question reads as:

    The US Senate has not yet voted on whether to ratify the Kyoto Protocol. If the US does not ratify thetreaty, it is very unlikely that the Protocol can be successfully implemented. Suppose that a nationalvote or referendum were held today in which the US residents could vote to advise their senatorswhether to support or oppose ratifying the Kyoto Protocol. If US compliance with the treaty would costyour household [t i]5 dollars per year in increased energy and gasoline prices, would you vote for oragainst having your senators support ratification of the Kyoto Protocol? Keep in mind that the [t i] dollarsspent on increased energy and gasoline prices could not be spent on other things, such as otherhousehold expenses, charities, groceries or car payments [For / Against].

    Consistent with Cameron and James (1987), we assume that the respondents true value for the good

    in question (WTP) is y i , and that 'i i i y Z u where u i is normally distributed with 0 mean andstandard deviation , and Z i is a vector of explanatory variables. As the true value of WTP i isunobservable, the threshold value [ t i ] proxies its lower bound if respondent votes for the proposal. So,the i th individual makes a decision to vote Yes for the KP if the WTP i is greater than the paymentamount [ t i ], the proposed annual cost that is an addition to the respondents current householdexpenditure in terms of the higher energy and gasoline prices. We denote the Yes response to VOTE by y i =1, and y i = 0, otherwise. Therefore, the probability that the i th individual votes for the KP isgiven by:

    Prob ( 1i y ) = Pr[ ( )]i iu t Z = 1 - [( ' ) / ]i it Z ,

    6 (1)

    It is expected that the higher the payment amount, the lower the probability of voting Yes for the KP,

    ceteris paribus . So, we have the following likelihood function:

    1

    1

    [1 (.)] [ (.)]i in

    y y

    i

    L , (2)

    Log likelihood function is given by:

    1

    ln { ln[1 (.)] (1 ) ln[ (.)]}n

    i ii

    L y y , (3)

    Here (.) denotes [( ' ) / ]i it Z , a standard normal cumulative density function. The loglikelihood function is maximized using Newton-Raphson algorithm. 7

    Turning to hypothesis testing, for the empirical analysis we propose that if the value of the INDEX i isequal or less than some threshold level, or if the responses to the Q4 and Q5 are No, then for thoserespondents the survey is considered as inconsequential; these respondents either do not believe thattheir input would be used for policy-making, or that they do not care much about the outcomes of thepolicy (two conditions necessary for a consequential CV survey question), or that they view themethod of collecting information for policy making as inappropriate. The basic idea is that if the CVsurvey were consequential, then the response to the survey questions reflects the respondents true

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    8/16

    preference/valuation of the good in question. As suggested by Carson and Groves (2007), in such acase, the standard economic theory applies and the response to the question should be interpretableusing mechanism design theory concerning incentive structures. Basically, respondents responsesare the reflection of the preference functions of the respondents. It might be the case that those whovote Yes to Q4 and Q5 and/or choose higher value for Q1, Q2, and Q3, would have differentpreference functions from those who says No or assigns lower value for the respective questions.To test for differences in the willingness to pay, and differences in the preference functions of therespondents with higher index of consequentiality versus a lower index, we propose the hypotheses:

    H1 : Household Willingness to Pay to reduce Global Climate Change depends upon the perceived degree of consequentiality.

    H2 : The underlying preferences are not the same under the two perception regimes (those who perceive the survey as consequential vs. those who dont).

    The null hypothesis for H1 is that the WTP to reduce GCC does not depend on the perceived degreeof consequentiality (i.e., that two sub-samples with different degrees of consequentiality would havethe same WTP for the good in question). The null hypothesis for H2 is that the underlying

    preferences are the same under the two regimes. To test these hypotheses, we divide the sampleinto two sub-samples based on the value of the index of consequentiality. For example, in the case ofQ4 (GW-REF) and Q5 (GW-WTP) questions, we put all respondents who said Yes into one sub-sample, and those who said No into another sub-sample. In the case of INDEX i , we use

    {0.5,0.6,0.7} as cut-off points to see how robust the results will be depending upon the value ofthe perceived consequentiality. 8 We use difference in WTP for two sub-samples to test H1 and thelikelihood ratio test for evaluating H2.

    Along with household per capita income ( INCOME ), we use age ( AGE ), gender ( SEX ), and ameasure of the education ( EDUCATION ) of respondents as control variables. The respondentpreference function may depend on these variables. Other control variables of interest are related tothe attitude of the respondent towards the severity of the environmental crisis ( CRISIS ), therespondents perceived level of attention to the issue of GCC ( ATTENTION ), and the perception ofrespondent towards the importance of the acts of industrial nations to reduce production ofgreenhouse gasses ( IMPORTANCE ). We use categorical variables to differentiate the effects ofdifferent budget treatments (mental accounting vs. standard remainder) on the voting decision (see Liet al. 2004). As we are using the combined data set of four different matching sample surveys, weuse three dummy variables ( HI1, HI2 , and KN ) to control for the potential effect of survey modes in thevoting decision, where telephone sample survey ( TEL) is the base category.

    We use two simple ( GW-REF and GW-WTP ) and two composite indices ( INDEX 1 and INDEX 2 ) as themeasures of consequential CV survey question. First , using one index at a time, we estimate fourWTP-probit models (Model I-IV) and calculate the median (mean) WTP from each model. The

    goodness of fit of these models is determined by the McFadden R 2

    or likelihood ratio index (Greene2000). Second , to test the hypotheses ( H1 and H2 ) that samples with lower value and higher value ofthe indices of consequentiality have different amount of WTP and have different preference functions,we split the sample into two groups as described earlier. Here we consider the CV survey

    consequential if INDEX i > {0.5,0.6,0.7} , for i = 1, 2; and it is less consequential otherwise. Three

    different cut-off values of the index of consequentiality are used to check robustness of the results.The significance of such a split is measured by t -ratio and likelihood ratio tests for the WTP-probitmodels (Greene 2000).

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    9/16

    7. RESULTS

    Table 1 provides definitions and descriptive statistics for the variables used in the analysis. Table 2presents the distribution of the responses for the five different consequential survey questions. Onthe scale of 0-10, the average scores to the first three questions (Q1-Q3) are 6.83, 7.49, and 5.60,respectively. The voting pattern for the next two questions (Q4-Q5) is 78% Yes to Q4, and 33% Yes

    to Q5, suggesting that respondents view the REFERENDUM as a good-way to collect informationregarding the GCC, but view asking WTP questions in a less positive way. We caution that resultsfrom this exploratory research are not likely to be generalizable, because respondents perceptionsregarding the methods of collecting information may vary depending on the survey mode and issueconsidered. 9

    Results from the maximum likelihood estimation for the log-normal distribution are presented in Table3. All coefficients of the index of consequentiality, which include different measures ofconsequentiality indices ( GW-WTP , GW-REF , INDEX 1, and INDEX 2 ), have positive sign and arehighly significant in all Models (I-IV); thus, WTP is positively associated with the degree ofconsequentiality. Estimated results are consistent with the standard economic theory as thecoefficient of INCOME is positive significant in all cases, implying that the environmental quality is anormal good and respondents would demand it more when their income increases. 10 The estimatedcoefficient on the budget treatment is highly significant with expected (negative) sign. 11 The resultsalso show that if the respondents were fully informed about the KP, if they perceived that theenvironmental problem was very important, or if they perceived that it is appropriate to reduce theproduction of greenhouse gasses by industrial nations, then the probability of voting Yes would go up.The estimated coefficients of those three variables ( CRISIS, ATTENTION and IMPORTANCE ) arepositive and highly significant. The effects of the survey mode have mixed results.

    We present the median (mean) WTP values in Table 3, with corresponding standard errors asestimated using the Delta method (Greene 2000). The point estimate of WTP ranges from $500 to$520 depending on the model. The goodness of fit is measured by the McFadden R 2 , which is

    around 0.36. This index is slightly higher in Model-III and Model-IV. Therefore, we choose these twomodels for further analysis.

    We split the entire sample based on the index of consequentiality, starting with the theoretical mean(INDEX i = 0.5) to divide the sample. Once we split the sample based on the value of the indices ofconsequentiality, we re-estimate Model-III and Model-IV for split samples. Table 4 presents the mean(median) the WTP differences between the groups of respondents that view the survey asconsequential and inconsequential, and the likelihood ratio statistics for the split samples. Themedian (mean) WTP for these two groups differ significantly ($0.50 vs. $1240 in Model-III, and $9 vs.$1122 in Model-IV) that for the respondents who view the survey as inconsequential have insignificantmean (median) WTP ($0.50 - $8.97) and for the respondents who view the survey as consequentialhave their predicted WTP in the range of $1,122-$1,240.

    Given that predicted WTPs are different for these two samples, we also test for differences in thepreference functions between two sub-samples. Likelihood ratio tests indicate that preferencefunctions for the two groups differ significantly. We also re-estimate the same models with samplemean ( INDEX i = 0.6), and an even higher value (0.7) of the index of consequentiality as the cut-offpoints for two sub-samples. As shown in Table 4, the difference in the WTP values and the preferencefunctions appear to be robust in these cases as well. The mean WTP for the sub-sample ofrespondents who view the CV survey as inconsequential is very low ($0.50 to $159.13) and that for

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    10/16

    the sub-sample of respondents who view the CV survey as consequential is significantly higher($1,121 to $3,093) depending upon the cut-off points (0.5 to 0.7) and the type of INDEX i .

    Results indicate that predicted WTP and the underlying preference functions are different betweenthose respondents who view the survey as highly consequential relative to those who dont. Theconcern is that if respondents view the survey as inconsequential, then economic theory may not be

    applicable to their responses and the inferences that we draw may be inappropriate for policy (Carsonand Groves 2007).

    It may be tempting to assume that those people who view the survey as highly inconsequential (theindex value less than sample average of 0.5, for example) might have different levels of income (e.g.,may be low income people) and they may be from different age groups. To see if such differencesexist, we compute separate summary statistics for two sub-samples that are divided based on thesample average value of INDEX 1. Table 5 compares the summary statistics of the relevant variablesfor these two sub-samples of respondents who view the survey as either inconsequential orconsequential. There does not appear to be large differences between these two groups in terms ofpersonal characteristics (age, education), household income and the bid amounts offered to them.Average household income is higher in the sub-sample that views the survey as inconsequential($55,890 vs. $52,180) and majority of them are male respondents (59%).

    Carson and Groves (2007) argue that economic theory cannot be applicable in the case of aninconsequential CV survey. If responses are not consistent with economic theory, then any predictionor inference that could be drawn from such CV survey outcome may not be valid for policy purposes.In our case, the average WTP is about $510 for the entire sample. But when the sample is splitbased on the index of consequentiality, as high as 57% of the respondents have WTP less than $68,and about 20% of respondents those who view the survey as relatively inconsequential (for the cut-off value of INDEX 1 = 0.5) - have a WTP of $0.50.

    8. CONCLUDING REMARKS

    In the light of continuing debate about the validity of CV study results, and the quest for consequentialvaluation questions, we propose that respondent-based, self-assessed indicators could be used toinvestigate the consequentiality of a CV survey. This research has presented several examples ofsuch respondent-based, self-assessed survey questions. To review, in any truly consequential CVsurvey, it would be expected that respondents would reveal the truth while answering a referendumsurvey question. This is because the respondents have incentive to tell the truth in the case of CVsurvey where (Carson and Groves 2007): (1) the respondents believe that their inputs would be usedin the policy making: (2) the respondents care about the outcome; and (3) they cannot avoid theconsequences of the policy change. Such a referendum CV format is termed as consequential. Butthe issue remains on how the analyst knows whether the given CV survey is consequential. There

    could certainly be different ways to assess this, but one approach would be to ask the respondentsregarding their views on the degree of consequentiality of the CV survey.

    Using data from four national US survey samples (TEL, HI1, HI2 and KN with combined n= 29,193),and an advisory referendum format for an environmental public good, we constructed two differentindices of consequentiality based on the respondent-based, self-assessed survey questions. Resultsindicate a positive significant estimated coefficient for those constructed indices of consequentiality inWTP modeling.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    11/16

    This approach, however, might have some inherent problems. For example, even if the respondentsperceive the CV survey questions as highly consequential, it may not be the case that this wouldalways have a positive effect on voting responses and WTP. It is unclear that this directional effectwould always hold, and further other design aspects might drive the voting decision. Thus, we alsoanalyze the possibility of different preference functions by respondents with different degrees ofperceived consequentiality. Specifically, we divide the sample between those with a high and lowdegree of consequentiality, and test for the differences in the underlying preference functions betweenthe two sub-samples. Results indicate significant differences in the underlying preferences of the twogroups of respondents, as measured by a likelihood ratio test.

    Respondents perceived consequentiality clearly matters, and can vary significantly across a sampledpopulation. While it would be inappropriate to draw definitive conclusions (e.g., direction or magnitudeof effect) from any single CV study, at a minimum, researchers should begin to systematically assessthe perceived degree of consequentiality. Our objectives are modest, but at least relatively unique inpushing for some empirical evidence of perceived consequentiality. We hope that it helps to spurfurther debate regarding how to measure consequentiality. As our understanding of the concept ofconsequentiality in a CV survey continues to evolve, further discussion is needed for finding common

    approaches. We close with tentative recommendations to the CV practitioner. First, researchersshould include respondent-based self-assessed standard questions in the CV survey. Responses tosuch questions help to collect information regarding respondents perception towards: (a) the methodsof collecting information, (b) the expected success of the program, (c) the possible use of theirresponses in affecting policy-making and (d) the extent that respondents care about policy outcomes.Second, researchers should analyze respondents perceptions of the consequentiality of the CVsurvey, and potentially split the sample based on the self-reported degree of consequentiality, andreport WTP values for the split-samples. In such a split-sample case, one may use the proportion ofthe respondents that view the survey as consequential (inconsequential) relative to the entire sampleas an indicator of the quality of the information generated from the CV survey.

    REFERENCES

    Arrow, K., Solow, R. , Portney, P. R., Leamer, E.E., Randner, R., and Schuman, H., 1993. Report ofthe NOAA Panel on Contingent Valuation. Federal Register , 58 (10), 4601-14.

    Balistrei, E., McClelland, G., Poe, G., and Shulze, W., 2001. Can Hypothetical Questions Reveal TrueValues? A Laboratory Comparison of Dichotomous Choice and Open-Ended Contingent Values withAuction Values. Environmental and Resource Economics , 18(3), 275-292.

    Berrens, R. P, Bohara, A. K., Jenkin-Smith, H., Silva, C. and Weimer, D., 2003. The Advent ofInternet Surveys for Political Research: A Comparison of Telephone and Internet Samples. Political Analysis , 11(1), 1-23.

    Berrens, R. P, Bohara, A. K., Jenkin-Smith, H., Silva, C. and Weimer, D., 2004. Information and Effortin Contingent Valuation Surveys: Application to Global Climate Change Using National Internet

    Samples. Journal of Environmental Economics and Management , 47(2), 331-363.Bohara, A. K., Berrens, R. P., Jenkins-Smith, H., and Silva, C., 1999. Public Valuation of Basic HealthCare Services in New Mexico. Final Report , UNM Institute for Public Policy, The University of NewMexico, Albuquerque, NM.

    Boyle, K. J., 2003. Contingent valuation in practice. In Champ PA, Boyle KJ, Brown TC (eds.), APrimer on Nonmarket Valuation . Boston: Kluwer (pp 110169).

    Cameron, T. A., James, M. D., 1987. Efficient Estimation Methods for "Closed-Ended" ContingentValuation Surveys. The Review of Economics and Statistics , 69 (2), 269-276.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    12/16

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    13/16

    NOTES1. For different types of consequentialism, see Stanford Encyclopedia of Philosophy http://plato.stanford.edu/entries/consequentialism/. Cited January 29, 2007.

    2. Examples include: Carson et al. (2006); Whitehead (2002); Ethier et al (2001); Balistrei et al (2001)Hanemann (2000); Johnston (2006); Veisten (2007). As of January 2008, Google Scholar listednearly 200 citations, combined, to Carson et al. (2000) and Carson and Groves (2007).

    3. Q1 Q3 summarize these issues as mentioned above.

    4. Alternatively, we dropped dont know option altogether and compared that results with the resultsfrom combined no and dont know scenario. We did not find significant differences between thoseresults and thus do not report them in detail here.

    5. The payment amount [t i] is randomly assigned amount selected from the list of 13 pre-specifieddifferent options where t = {6, 12, 25, 75, 150, 225, 300, 500, 700, 900, 1200, 1800, 2400}.

    6. Pr(y=1) = Pr[ )(1

    Z t u

    ii ]

    = Pr[(t i Z)/ < i ]

    =it

    dk f )(

    = 1-it

    dk f )(

    = 1- ) / )(( Z t i

    7. In estimating the likelihood function, we take the log of the payment amount to address scaleissues, and this makes the distribution log-normal.

    8. Out of these three cut-off points, 0.5 is the theoretical mean, 0.6 is the approximate sample mean,and 0.7 is the 60 th percentile (approximately).

    9. Previous work explored pilot applications of similar follow-up questions, and showed considerablevariation across settings. As an example, in a study of public valuation of basic health care servicesin New Mexico, Bohara et al. (1999) asked a similar good way question in their CV study aboutwhether asking questions about peoples willingness to pay is a good way to inform policy makersabout choices concerning health care coverage in New Mexico. Using a telephone survey mode, theyfound that over 81% respondents viewed asking WTP questions as a good way for making policychoices. This is more than double the 33% Yes responses to the Q5 GW-WTP question in the GCCCV survey. Further, we also see large variation across sub-samples in mean responses to thisquestion. In the sub-sample of telephone surveys, the Yes response is 64% in the GCC survey,significantly higher than the average of all four modes (33%).

    10. We also estimate all models with log of the payment amount as a right-hand variable, and find thatthe sign of the coefficient of LPAY is negative and significant in all models. We exclude those resultsas that approach requires separate calculation of WTP with transforming the estimated coefficients.

    11. As in Li et al (2004), it is expected that the mental accounting treatment will have a negative effecton WTP, in comparison to simple budget remainder statement, as respondents are asked moredirectly to consider their discretionary budget constraints.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    14/16

    Table 1: Variable Definition and Summary Statistics

    Variables Definition Obs Mean Std.Dev.

    VOTE Respondents voting decision on the advisory referendum havingtheir Senators support KP (0 = against, 1 = for)

    27840 0.56 0.50

    GW-REF Response to Q4 : would a national referendum be a good way forcitizen to express their preferences regarding KP to the US

    Senate? (0 = no & not sure, 1= yes)

    27908 0.78 0.41

    GW-WTP Response to Q5 : Is willingness to pay is a good way forgovernmental officials to make policy choices about global climatechance ? (0 = no & not sure, 1= yes)

    29193 0.33 0.47

    INDEX 1 Constructed index of Consequentiality using questions: Q1, Q2 and Q3

    29193 0.64 0.22

    INDEX 2 Constructed index of Consequentiality using INDEX 1, & twoadditional good way questions: Q4 , and Q5

    29193 0.57 0.27

    LPAY Log of randomly selected payment amount in USD wherePAY = { 6, 12, 25, .,1800, 2400}

    28121 5.38 1.86

    EDUCTION Education level of respondents (1-7 scale; 1: < High School, 7: >MS)

    29169 4.48 1.28

    AGE Respondents Age 29168 42.42 13.52SEX Gender of respondents (0= female, 1= male) 29186 0.50 0.50INCOME Income of respondents in 1000 USD 22539 53.24 36.56BUDGET Budget treatment (1 = mental account, 0= standard remainder) 29193 0.47 0.50CRISIS Respondents perception towards the intensity of environmental

    Problem in 1-10 scale (0= not a problem, 10=severe problem)28285 5.76 2.23

    ATTENTION Respondents self evaluation to the familiarity with KP in 1-10scale (0= not at all, 10= completely familiar)

    28286 5.73 2.48

    IMPORTANCE Respondents perception towards the importance of reducinggreenhouse gasses in 0-10 scale (0= not important, 10= extremelyimportant)

    28153 7.57 2.66

    HI 1 Indicator variable of first HI sample (1 if HI1, else 0) 29193 0.45 0.50HI 2 Indicator variable of second HI sample (1 if HI2, else 0) 29193 0.38 0.49KN Indicator variable of KN sample (1 if KN, else 0) 29193 0.10 0.30

    Table 2: Descriptive statistics of Consequential Responses by Survey Mode Variable Tel HI HII KN Mean Min Max

    Q1. TREATY 7.12(2.79)

    6.90(2.94)

    6.69(3.03)

    6.87(2.71)

    6.83(2.95) 0 10

    Q2. SENATE 7.84(2.43)

    7.46(2.86)

    7.54(2.84)

    7.08(2.72)

    7.49(2.82) 0 10

    Q3. KYOTO-EFFECT 6.02(2.51)

    5.70(2.69)

    5.48(2.77)

    5.36(2.34)

    5.60(2.69) 0 10

    Q4. GW-REF 0.83(0.37)

    0.78(041)

    0.76(0.42)

    0.79(0.41)

    0.78(0.41) 0 1

    Q5. GW-WTP 0.64(0.48)

    0.33(0.47)

    0.33(0.47)

    0.29(0.45)

    0.35(0.47) 0 1

    Note: Standard deviation is given within parentheses. The corresponding sample sizes for Tel, HI,HII, and KN are 1806, 13034, 11160 and 2125, respectively.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    15/16

    Table 3: Maximum Likelihood Estimates of Log-Normal Distribution (Dep. Var.: VOTE)

    Model I Model II Model III Model IVVariables Coefficient Robust-

    SECoefficient Robust-

    SECoefficient Robust-

    SECoefficient Robust-

    SEGW-WTP 0.82*** 0.090 - - - - - -GW-REF - - 0.94*** 0.108 - - - -

    INDEX1 - - - - 7.69*** 0.335 - -INDEX2 - - - - - - 5.37*** 0.268EDUCATION 0.48*** 0.037 0.53*** 0.038 0.49*** 0.037 0.55*** 0.038AGE -0.03*** 0.003 -0.03*** 0.003 -0.03*** 0.003 -0.03*** 0.003SEX 0.00 0.086 0.08 0.087 0.21** 0.085 0.14 0.086INCOME 0.01*** 0.001 0.01*** 0.001 0.01*** 0.001 0.01*** 0.001BUDGET -0.48*** 0.084 -0.47*** 0.084 -0.39*** 0.082 -0.45*** 0.083CRISIS 0.61*** 0.029 0.59*** 0.029 0.52*** 0.028 0.57*** 0.028ATTENTION 0.12*** 0.021 0.13*** 0.020 0.09*** 0.020 0.12*** 0.028IMPORTANCE 0.77*** 0.027 0.76*** 0.027 0.54*** 0.025 0.65*** 0.026HI 1 -0.46*** 0.193 -0.65*** 0.188 -0.46** 0.183 -0.27 0.185HI 2 -0.49*** 0.194 -0.69*** 0.190 -0.51*** 0.184 -0.30 0.187

    KN -0.16** 0.249 -0.35 0.246 -0.180 0.239 0.064 0.242CONSTANT -4.92*** 0.387 -5.41*** 0.405 -7.73*** 0.436 -7.49*** 0.440 4.37 0.11 4.37 0.11 4.25 0.10 4.33 0.10Median WTP($) 516.32 23.41 520.16 23.61 500.57 22.28 513.67 23.26McFaddens R 2 0.351 0.352 0.368 0.362

    2

    (12) (Slopes=0)1557 1560 1628 1583

    n 22224 22259 22307 22307Note: *, **, and *** indicate significant at 10%, 50%, and 1% level, respectively.

  • 8/14/2019 Assessing Perceived Consequentiality: Evidence from a Contingent Valuation Survey on Global Climate Change

    16/16