3.3. Contingent Valuation Method
The questionnaire consisted of four different sections. In brief, the questionnaire first asked the respondents about previous visit experience to national parks and temples as well as their awareness of inholdings and perceptions of the importance of the conservation of temples’ cultural and religious value. The second part of the questionnaire was the CVM section and the following part asked the respondents’ environmentally responsible behaviors in national parks and their trust in the government’s environmental policies. Finally, socio-demographic questions were asked in the last part of the questionnaire.
As indicated above, the questionnaire also contained the CVM section to assess individuals’ economic value derived from the conservation of temples and temple forests in the national parks. A dichotomous choice format was used that requires respondents to answer YES or NO. Compared to the open-ended format which asks the respondents to reveal the actual the amounts they would be willing to pay, a closed-ended is simpler because respondents are asked to reveal their price taking decision by checking “YES” or “NO” to a CVM question. Nonetheless, this format has been also criticized in that limited information was generated of whether a respondent’s true WTP is above the proposed bid amount threshold when she answers YES or her WTP is below the proposed bid amount threshold when she answers NO. Thus, due to its inefficiency nature, the dichotomous choice CVM requires a large sample to increase the precise estimation of the WTP values [
34].
As an alternative means, the double bounded dichotomous choice (DB) CVM has been used to improve statistical efficiency with a follow-up dichotomous choice question for a second answer [
34]. If a respondent’s answer is YES to the first dichotomous choice question for a price of
$X, the respondent is asked if they are willing to pay a higher bid of
$X +
$Y in the follow-up question. If a respondent answers NO, the follow-up question is provided to determine whether she is willing to pay a lower bid of
$X −
$Y. The sequence of questions may increase the complexity of the method because the second question is additionally used. However, the insertion of the second question helps to produce more precise WTP estimates by obtaining more information from each respondent [
35,
36]. To estimate individuals’ WTP value, this study made use of this DB-CVM.
The study instrument including the DB-CVM questions was carefully designed with a group of economists and social scientists specializing in outdoor recreation management and was also pretested through Embrain.com to ensure readability and improve validity and reliability. Among different payment options considered, household income tax was selected as a feasible payment vehicle because the government has considered monetary compensation in case a daily entry is abolished. Furthermore, the hypothetical nature of CVM is often a source of criticism and thus, estimated WTP is known to be commonly inflated. This is called hypothetical bias and several different treatments have been used to reduce the difference between hypothetical and real valuations [
37]. Among those, cheap talk has been gaining popularity recently as an effective means to reduce the degree of hypothetical bias [
38,
39]. The basic idea of cheap talk is to provide a detailed explanation of the hypothetical bias problem and encourage respondents to make more realistic decisions in CVM questions. As a result, a short script of cheap talk was inserted:
According to a number of studies on the valuation of public goods (such as parks or protected areas), individuals often indicate that they are willing to pay more for the protection of these goods on surveys than what they are actually willing to pay in real life. We believe this is due to the fact that individuals do not really consider how big an impact of the additional spending would actually have on their real-life budget. Individuals are often more likely to be generous and contribute to causes when the scenario is hypothetical than when their money is actually at stake, a problem known as “hypothetical bias”. Please be aware of this bias and try to answer the following question in accordance with what you would choose to do in a real scenario.
Then, the exact wording of the first DB-CVM question was as follows:
National parks are protected areas directly managed by the national government to preserve the ecosystems as well as natural and cultural resources that represent Korea. Out of a total of 22 national parks, 17 famous mountains are designated as national parks. The national parks provide vital ecosystem services such as the provision of ecological habitats, securing water resources, air purification through the absorption of carbon dioxide, water purification, and prevention of soil erosion. The costs of managing and conserving the national parks have been progressively increasing as there is more demand for national park visits derived from growing household income and leisure and health interests.
Almost all of the 17 national parks designated around famous mountains have temples and their forests (called Sachallim), which covers 7.0 percent (280 km2) of the total area of the national parks. Despite the fact that visitors’ admission to national parks is free of charge, the temples independently collect an admission fee between ₩1000–₩5000 per person) approved by the national government in order to conserve cultural and natural resources in the temples and their forests. The government is considering abolishing the private admission fee due to the resistance of visitors who do not want to visit the temple(s) and the temples, consequently, want to develop the forest land in case they lose the financial source. In this scenario, ecosystem services provided by the national parks are expected to decrease by at least the forest land developed. Therefore, for the conservation of the temple forests, the government plans to compensate the temples with the tax as much of the value as the temple forests. Would you be willing to pay an annual household income tax for the compensation of the temples? Please answer the following question carefully after taking into account that your household income is limited and needs to be spent for a variety of purposes.
If the tax is only used for the compensation of the temples for the value of temples and temple forests, would you be willing to pay additional household income tax of ₩_____ per annum?
For the first DB-CVM question, ten bid values ranging from ₩500 (about US $0.46 as of 17 December 2017) and ₩30,000 ($27.53) were methodically chosen after adjustments based on discussions with the park officials and managers and a couple of pretests of the questionnaire with park visitors. These bid values were randomly assigned to the questionnaires asked to the respondents. For the follow-up DB-CVM question, the bid value presented was doubled, which ranged from ₩1000 ($0.92) to ₩60,000 ($55.07), if the response to the first question was “YES.” Likewise, a half price of the first bid between ₩300 ($0.28) and ₩15,000 ($13.77) was presented if the response to the first one was “NO.” For respondents who placed zero (i.e., chose NOs) in both DB-CVM questions, a question was further asked to examine protest behavior (i.e., ‘we do not trust the government to implement this properly’).
3.4. Empirical Analysis
In the case of DB-CVM which provides two bid values (M1: 1st bids, M2: 2nd bids) to the respondent i a researcher can observe four different WTP ranges depending on the respondent’s answer to the two sequential bid values. Four possible WTP ranges can be expressed as follows.
(1) a respondent says yes for both bid values: WTPi ≥ M2
(2) a respondent says yes for 1st bids but no for 2nd bids: M1 ≤ WTPi ≤ M2
(3) a respondent says no for 1st bids but yes for 2nd bids: M2 ≤ WTPi < M1
(4) a respondent says no for both bid values: WTPi < M2
Taking the structure of WTP intervals into account, this study employs the interval data probit analysis developed by Hanemann et al. [
34]. An econometric model based on DB-CVM data formulated by Hanemann et al. [
34] consists of two different parts and can be shown by:
where
i = 1, 2, …, N for each respondent,
k = 1, 2 for 1st and 2nd bid responses, respectively,
τk represents the means for the
kth responses, and
denotes random components of
ith respondent associated with
kth responses, which is assumed to be normally distributed.
If the assumption that means for each response are equal (i.e.,
τ1 =
τ2) holds, Equation (1) can be simplified by removing
k expression and rewritten as follows:
Based on Equation (2) the probability (Prob) that respondents will choose for four possible responses can be derived as follows.
(1) a respondent says yes for both bid questions (YES-YES):
(2) a respondent says yes for 1st bids but no for 2nd bids (YES-NO):
(3) a respondent says no for 1st bids but yes for 2nd bids (NO-YES):
(4) a respondent says no for both bid values (NO-NO):
Finally, a likelihood function of the
ith respondent (
) is expressed as the product of four different probabilities defined above.
In the DB-CVM, it is generally assumed that a respondent’s true WTP is not different over the sequential payment questions implying the respondent’s preference remains unchanged [
34,
40]. This assumption, however, is likely to be violated due to the possibility of preference anomalies [
40,
41,
42]. Preference anomalies indicate that the initial payments may have an influence on the response of follow-up bid questions. If preference anomalies exist, a researcher may encounter difficulty with eliciting respondent’s correct WTP, which in turn leads to inaccurate welfare measures. Therefore, in order to reach conclusive results, it is crucial for research to examine the existence of preference anomalies in the DB-CVM studies.
The representative preference anomalies among others are known as anchoring bias and shift effects [
43]. The anchoring bias arises if the respondents anchor their WTP in the initial bid. This case can happen when the respondents believe the 1st bid amounts signify the true value of the environmental goods/services [
41,
44]. In this case, the 2nd WTP would be adjusted by the sum of a weighted average of the 1st WTP (true WTP) and the 1st bid amounts such as
where
is the respondent’s anchored preference in which level of
lies between zero and one. If
is equal to zero (i.e., without anchoring bias) 2nd WTP collapses into the 1st WTP. A higher value of
indicates more anchoring bias in WTP estimates. On the other hand, shift effects take place when the respondents perceive the first bids as the actual costs of government policy. In this case, the 2nd bids may be regarded as additional charges for those who say yes for the 1st bid values while the follow-up question given to the respondents who say no for the 1st question can be considered as an alternative provision for a lower level of environmental quality [
43,
45]. This implies if the respondent’s true WTP is based on the 1st bid question, the response to the 2nd bid question can be shown by the respondent’s true WTP plus an additional parameter that captures shift effects such as
where
represents a parameter for shift effects with negative sign (If the shift effect exists, the parameter for shift effect will be negative since the provision of the 2nd bid question is more likely to make the respondents reject 2nd bid amounts [
40].
To test for possible existence of anchoring bias and shift effects in the DB-CVM survey, we adopt analytical approaches developed by Whitehead [
44], which transforms DB-CVM data into iterative valuation forms in a panel structure where the dependent variable of individual
i = 1, …, n at time
k = 1, …, k,
is defined as upper and lower bounds of WTP. More specifically, the econometric model can be set up as
where
indicates dummy variable for shift effect and
represents anchoring bias which is an interaction variable between
and the first bid value
(i.e.,
).
is the vectors of socio-demographic variables. The term
,
,
, and
are individual or vectors of coefficients to be estimated.
and
represent the individual error term for random effect and the random error term, respectively.