Contingent valuation and actual payment for voluntarily provided passive-use values: Assessing the effect of an induced truth-telling mechanism and elicitation formats

Abstract

A fundamental question about the contingent valuation (CV) method is to what degree it predicts actual payments (AP). This has particularly been an intriguing matter related to voluntary provision of public goods representing primarily passive-use values. This paper reports the results from such a CV–AP comparison. Applying a voluntary payment mechanism there exists a theoretical expectation of upward bias in CV estimates and downward bias in AP. This study applied an induced truth-telling mechanism in one treatment group to assess the hypothetical bias effect in CV. The CV estimates in this treatment group were significantly lower than in the group that did not face this mechanism. But this effect was limited to those responding/acting to dichotomous choice, not affecting those responding to open-ended questions about willingness to pay.

Acknowledgements

Financial support for this project by the Norwegian Research Council (NFR), the Environment and Development programme, is acknowledged. We have benefited from constructive comments by Richard Carson and Michael Hanemann on the incentive structures. We also thank Mattias Boman, Göran Bostedt, Hans Fredrik Hoen and Kjartan Sælensminde for valuable comments. Any remaining errors are the responsibility of the authors.

Notes

¹ Duffield and Patterson (1991) present their test as a test of criterion validity, but in their endnote number 10 they interestingly raise the following: ‘Does an actual cash trust fund provide a criteria for this construct (CV)? The problem is that given nonexcludability, there is a strong likelihood of free rider behaviour. In a sense there is no imaginable ‘appropriate market’ for goods lacking the excludability criteria. In this sense, it could be argued that our actual cash trust fund measure is not a criteria, but another measure and that this exercise is one in … ‘convergent validity’)’.

² There are several ways to validate a survey-based method like CV. Mitchell and Carson (1989) refer to tests of criterion, construct and content validity, a taxonomy proposed by the American Psychologist Association in 1974. Construct validity contains (internal) theoretical and (external) convergent validity. An assessment of the theoretical validity is accomplished by measuring consistency with standard theoretical predictions, e.g. a downward sloping curve in a discrete choice bid–accept diagram and observing the expected signs of regression model covariates. Content validity is based on a non-quantitative assessment of the CV questionnaire, i.e. the evaluation of survey design and survey administration.

³ Navrud (1992) and Schechter (1996) also report on tests of hypothetical payment against actual payment, but the hypothetical estimates in their studies were based on survey instruments that differed somewhat from a CV survey.

⁴ More fundamentally such a referendum-simulated CV must cast the preference question in terms of being able to influence a government decision, i.e. it requires a consequential survey question (Carson et al., 2000). Referendum-simulated CV is also the only type of CV that principally be tested for criterion validity, by comparison against actual voting. The few studies we are aware of all contribute to confirm CV's criterion validity (Vossler and Kerkvliet, 2003; Vossler et al., 2003; Champ and Brown, 1997; Carson et al., 1987).

⁵ ‘Seip and Strand used members of the environmental group as the interviewers in their hypothetical treatment, thus increasing pressures for compliance. They compared hypothetical phone responses with responses to an actual mail solicitation. Duffield and Patterson compared hypothetical mail solicitations from the University of Montana with actual mail solicitation from the Nature Conservancy.’ (Hanemann, 1994, p.31.)

⁶ The interest to free ride is reduced when a level of aggregated contributions must be reached for enabling provision, and when the individual gets her money back if the provision point is not reached, or receives excess benefits or a proportional rebate if the provision point is surpassed. Yet, this provision point mechanism for voluntary funding of public goods is still not an incentive compatible mechanism (Poe et al., 2002, p.108; Carson et al., 2000, pp.10–12).

⁷ It was stated that the 12 publicly protected areas amounted to 4,700 ha (from 12.5 to 1,800 ha of size) and that the 13 areas to be leased by WWF's Forest Fund amounted to 300 ha (from 4 to 60 ha of size). To communicate some common comprehension of size, these numbers were related to the size of a football field.

⁸ WTP data collected from the open-ended questions in the pilot test was used to construct an empirical distribution of the WTP, and then bids were assigned based on a balanced drawing of random numbers from a uniform distribution. For a sample of size n (n even), Boyle et al. (1988) drew n/2 random numbers P ₁, … , P_{n/ 2} from a uniform distribution on the interval (0,1). The quantiles corresponding to P ₁, … , P_{n/ 2}, … , and l − P ₁, … , 1 − P_{n/ 2} were calculated from an empirical distribution of pre-test WTP amounts and used as bids.

⁹ The use of a mail survey was necessary in this study to create equal settings for the test of CV versus AP, and it exemplifies that a combined telephone-mail survey has the potential to reduce the problem of low response rates. Arrow et al. (1993) argue that it is ‘unlikely that reliable estimates of values could be elicited with mail surveys’ and instead recommend face-to-face or telephone interviews. Other authors have argued that such a combined telephone-mail survey might have an advantage in terms of giving the respondents more time to consider their answers and avoiding the pressure for compliance (Dillman, 1983; Mitchell and Carson, 1989).

¹⁰ The invoice was a postal giro form that represented the most common means of payment in Norway at the time of the survey. The giro form could be used for payments directly to/from bank accounts.

¹¹ This non-linear logistic, or log-logistic model, was applied by Bishop and Heberlein (1979). Hanemann and Kanninen (1999) shows that this ‘BH’ model implies a nonlinear, additive utility model with an exponential error term: v ₀ = m + δ and v ₁ = m + δ + e ^{( α + ε )/ β} , for some arbitrary δ, where v_i (·), i = 0,1, is the indirect utility function, 0 referring to ‘no’ answer and 1 referring to ‘yes’, m is income, and α and β represent unknown regression parameters. Applying logit modelling implies that the error term ε, that appears only as an exponential term in the indirect utility function for ‘yes’ answer, is assumed to have a logistic distribution.

¹² None of those accepting the amount in the CV questionnaire returned the invoice with a lower amount. Some few of those receiving questionnaires and invoices with the lowest amount, $7.88 (NOK50), increased the amount on the invoice.

¹³ For all the zero reason answers, representing approximately half of the respondents, the proportions of such inconsistent answers were 2.4% for CV-oe, 0.8% for CV-dc, 1.4% for AP-oe and 0.5% for AP-dc. For non-protest zeros these proportions were 1.7% for CV-oe, 0.8% for CV-dc, 0.9% for AP-oe and 0.2% for AP-dc. Alternative indicators of indifference/zero WTP based on attitude questions (interest in environmental issues, concern about the issue of endangered species, spending money on more protection in the Oslomarka forest, and the joy for others use of protected areas in this forest) were also explored. But the indicator resulting in consistent zero for AP comprised only 4.3% of total sample.

¹⁴ For example, random values equal to or below 0.172 were taken to imply a $7.88 bid, and thus a stated oe WTP higher than $7.88 would imply a ‘yes’ to this bid. In the dc sub-samples fairly equal sub-groups were assigned to each bid, except that twice as many received the $15.8 (NOK100) bid.

¹⁵ This test is similar to the one performed by Kealy et al. (1990), and they judged results of 0.51 and 0.53 as ‘fairly high’ correlations.

¹⁶ The likelihood ratio test (q) applied to dc responses is: , where L _{sub  x} and L _{sub  y} refer to the log likelihood values of different sub-samples (x ≠ y), and L _total is the log likelihood value based on a combined sample.

¹⁷ The logit model estimates of the quasi-dc oe data were further examined by re-testing the hypotheses (i), (ii), (iii) and (v). The results of these tests were exactly the same as for the t-test and Kruskal–Wallis test.

¹⁸ Assuming (implicitly) that the upward hypothetical bias in the CV data overwhelmingly dominated the downward free-rider bias in the AP data, Champ et al. (1997) and Champ and Bishop (2001) used the AP as benchmark for calibration, or actually a validity criterion, in order to remove the hypothetical bias from the CV data reported by Brown et al. (1996). Based on this, Champ and Bishop (2001) found that 80% of the respondents to a contingent donation question provided a response consistent with how they were predicted to respond in an actual donation situation. In cases with private goods Blumenschein et al. (1998) and Johannesson et al. (1998, 1999) also found that ‘certain’ hypothetical ‘yes’ to dichotomous choice would calibrate well to actual choices (see also Carlson, 2000 and Frykblom, 2000). Blumenschein et al. point out the theoretical shortcoming of such calibration; there is no standard theory rationale for a ‘certain’ hypothetical ‘yes’ (ticking an ‘absolutely sure’ or a 10 at a ten-point Likert scale) predicting an actual choice while an ‘uncertain’ hypothetical ‘yes’ implies that a real payment will not take place. Kim (2000) proposes an explicit theoretical model for ‘hypothetical commitments’, where Likert-scale stated certainty of hypothetical choice is scaled to a probability measure for commitment. Applying the data from Blumenschein et al., Kim found that the hypothetical-commitment model for CV was able to predict real commitment (AP). A strength of the ten-point scale follow-up question used by Champ et al. is that it enables a consistent calibration model (Kim, 2000).

¹⁹ Due to principles and laws about protection of privacy, the Norwegian Gallup Institute (NGI) – with WWF and the Agricultural University of Norway (now renamed Norwegian University of Life Sciences) – applied to the Data Inspectorate for making an intermediary connection between the registers for the CV questionnaires and the WWF money collection (AP). To obtain this permit, NGI and WWF had to make it a condition that all respondents would be informed and explained the linkage between registers in the aftermath of the study, and those who had paid would be explicitly allowed a possibility (at no cost) to have their money returned. But one respondent reacted, in the course of the surveying, against the perceived connection between the survey and the fund raising and contacted a large newspaper (Arbeiderbladet, now renamed Dagsavisen) that presented the issue with a rather critical view. Due to this negative publicity WWF decided to cancel further collection of AP.