6. Willingness to Pay for Pro-Environmental Farming
6.3. Mean and Median Willingness to Pay for Pro-Environmental Farming
In Tables 6.1 and 6.2, mean and median WTPs are reported in three categories, for the whole sample, for those who have not received additional information, and for those who have received additional information. Mean and median WTPs have been estimated by using two elicitation methods, the combined bidding game-payment card format and the dichotomous choice format. The mean and median WTP elicitation based on the combined bidding game-pay-ment card technique is carried out in two slightly divergent versions. The first version covers ali the 663 observations, and leaving out observations in which the expressed WTP was equal to or higher than RIVI 2,500 produces the second one. The reasoning behind this practice is to eliminate responses that can be considered outliers. However, an elimination process of this kind has not its roots in any unambiguous postulate of economic theory. It merely illuminates the fact that mean WTP can be quite sensitive to a small number of deviating individual WTPs.
It is also possible to utilize the dichotomous choice format because the starting bids used in the combined bidding game-payment card (BG-PC) tech-nique form a bidding vector consisting of offers FIM 100, 500 and 1,300.
Concerning the dichotomous choice, two different, although closely related, model specifications of parametric nature are applied. The starting point is a random utility model with the logit specification acting as the cumulative distri-bution function. The standard model is based on the approach introduced by Hanemann (1984). In the application of the standard model, the effect of differ-ent truncations is tested. The simple spike model in turn follows the guidelines set by Hanemann and Kriström (1995). In addition, a non-parametric estimation technique that is based on dichotomous choice data is also used (Kriström 1990b). Ali these approaches were reviewed in detail in Chapter 6.2.
There is some variation in the average figures depending on the chosen elicitation method or model specification. Mean WTP for the whole sample ranges from FIM 290 to FIM 615 and median WTP for the whole data from FIM 150 to FIM 379. This supports the common notion that median is a more robust estimate than mean. It is a matter of taste if the observed variation in mean and median WTPs is considered significant from the policy-making perspective.
The highest mean WTP estimate is approximately four times bigger than the lowest median WTP estimate. If the monetary estimate is the only criterion for decision-making, the differences in average figures can be regarded as worry-ing. However, if the monetary estimate is taken as auxiliary information the purpose of which is to facifitate the decision-making process and not to dictate its outcome, the situation becomes less restricting.
Table 6.1. Mean WTPs Estimated by Using Different Elicitation Methods and Statistical Models, FIM/Person/Year.
BG-PC N=663 Standard model (BG-PC w/o (Standard model outliers N=647) [Standard model
Table 6.2. Median WTPs Estimated by Using Different Elicitation Methods and Statistical Models.
Non-param. model 0..A Simple spike (Non-param. model 0..A+500) model [Non-param. model 0.A-500]
Consider first the apparent sensitiveness of the combined bidding game-payment card format in relation to the highest bids. The removal of 16 highest individual WTPs (2.4% of the sample) reduces the mean WTP by 16.9% (from FIM 402 to FIM 334). If the same removal of observations is done concerning the standard model [_cx0..00] and the non-parametric model [0.A], the mean WTP changes somewhat less, 11.4% (from FIM 297 to FIM 263), and 8.2% (from FIM 562 to FIM 516), respectively. A corresponding comparison of changes in median WTP produces exactly the opposite order of magnitude changes. There is no alteration when the combined bidding game-payment card format is in question (FIM 150 - FIM 150), 11.4% (from FIM 297 to FIM 263) for the standard model Foo....] and 7.5% (from FIM 281 to FIM 260) for the non-parametric model [0.A].
The estimation of the standard error of the mean WTP is another way to assess the sensitivity related to the exclusion or inclusion of the highest bids. In the BG-PC model, the standard error of the mean WTP diminishes from FIM 24.7 to FIM 17.8 (28%) when the 16 highest WTP responses are eliminated. In the standard model the approximate estimate for the standard error of the mean WTP decreases from FIM 66.5 to FIM 63.5 (4.4%). Thus, based purely on the statistical properties of different models, the dichotomous choice format seems to be more reliable in terms of the mean WTP estimation, al-though the BG-PC approach seems to be in this context more robust in terms of the median WTP estimation.
Although it is possible to suggest that the highest bids in the BG-PC format are outliers and should be left out because of this, the elimination should always be based on sound theoretical arguments. In other words, the question is about the validity of WTP responses. If the highest bids are eliminated, the argument must be that the highest bidders do not reveal their "true" WTP. This may be the case, for instance, because of some sort of goal-oriented response motives. A much-used strategy is to compare the expressed individual WTP to individual income. If the individual WTP is implausibly high in relation to income, we can conclude that the respondent behaves strategically. However, it is very difficult to define what is an "implausibly high" proportion of income, because income may be an inadequate indicator of ali the resources that the respondent is ready to sacrifice to match his expressed willingness to pay. Especially when environ-mental values are at stake, many people are willing to make considerable efforts in order to promote their views. Thus, a comparison between expressed WTPs and income levels may appear to be a too simplifying choice as a screening criterion.
Table 6.3 presents an analysis of whether the 16 highest bids can be deemed the results of strategic behavior. Of course, it is not possible to derive an assessment criterion that would unambiguously show if a certain response was a misstatement of the "true" WTP. Consider first the ratio between WTP and pre-tax income. As already argued in the previous paragraph, there is no clear-cut way to define when this ratio becomes "implausibly high". If one-percent level is chosen, then 14 observations are dubious of their nature. If five-percent level is chosen instead, then only one observation looks suspicious. Ali the responses are also consistent in the sense that ali but one respondent are willing to convert at least 25% percent of total agricultural land under pro-environmental cultiva-tion. In addition, ali but one respondents have a clearly positive attitudel5 towards sustainable development, which can be perceived to be a strong indica-
15 The methods that were applied to identify attitudinal groups are explained in Chapters 7.1, 7.2, and in Appendix F.
Table 6.3. Some Characteristics of Possible "Outlier" Observations.
Attitude towards Proportion of pro- sustainable environmentally
*) The more `+' signs, the more positive attitude towards sustainable development. The more `—' signs, the more negative attitude towards sustainable development.
tor that the respondents truly prefer pro-environmental farming to conventional agriculture.
Observation number '592 and especially observation number 435 qualify as potential strategic responses because at the first glance there appears to be some inconsistency. Observation number 435 expresses high WTP but is not in favor of sustainable development, which is a central concept closely related to pro-environmental farming. However, a closer look reveals that respondent number 435 has a very positive attitude towards conventional farming. He is also a farmer and his father has been a farmer. Because he also wants to convert only 10% of the total agricultural land to pro-environmental farming, the conclusion is obvious. Respondent number 435 considers pro-environmental farming a means to guarantee that there will also be conventional agriculture left in the future. Thus, the answer is strategic in the sense that the respondent values something else than what is meant to be valued in the valuation scenario.
However, it does not mean that the respondent misstates his WTP because he would probably be willing to pay the amount he has mentioned if it really secured the future of conventional agriculture. It is likely that respondent number 592 represents rather similar views. He does not have a farming background, but he has a very positive attitude towards conventional agriculture and a positive attitude towards pro-environmental farming, too. The result is that there is not
enough substantiated evidence to prove that the highest bidders conceal their
"true" WTP and express exaggerated values.
This example clearly shows that, although the dichotomous choice format can be more desirable in terms of statistical properties, it lacks the wider analytic power of the BG-PC format. The motivations behind expressed WTPs and some other not so obvious relationships in the data are more easily detect-able when an elicitation format which directly gives the individual WTP is applied. In this respect this study supports the view presented by Boyle et al.
(1996), who argue that the prevailing endorsement of the dichotomous choice format should not lead to a complete rejection of other questioning formats.
Although the estimation results of mean and median WTPs achieved through different methods and specifications are consistent with certain theoretical and statistical properties that they are supposed to fulfill, one anomaly deserves a closer inspection. When mean and median WTPs are compared, ali but two pairs of estimates follow the hypothesis that mean or median WTP for those who have received additional information is higher than for those who have not.
The exceptions occur when the mean WTP is estimated by using the standard models
[0....]
and[0..A.].
The reason for this phenomenon is that the logit model behind the standard models is statistically somewhat inappropriate when the subsample of those who have not received additional information is in question. The estimation results can be found inAppendix D,
and they show that the t-value for the parameter ci (INTERCEPT) is statistically questionable (0.1081). In addition, the result of the likelihood ratio test (X2 = 0.39) indicates that it is not possible to reject the hypothesis that a = =0,
meaning that the applied logit model severely lacks explanatory power.As an overall conclusion, we can state that any of the applied elicitation techniques and model specifications seems to be reliable enough when mean or median WTPs are estimated for the whole sample. The observed variation in the average figures is due to different presumptions behind alternative models, not inadequate reliability of the CV method as such. However, reliability cannot be separated from validity. It is clear that different valuation questions and other details of the valuation scenario considerably influence the outcome. This will be illuminated in more detail in Chapter 6.5, where possible starting point bias and information effects are examined.
6.4. Demand Function for the Pro-Environmentally Cultivated