COMMENTARYPreference uncertainty in contingent valuation
Introduction
The concept of ‘preference (or respondent) uncertainty’ has gained a significant amount of attention in the stated preference literature over the past fifteen years. Hanemann et al. (1995) first proposed a welfare model that incorporates an element of uncertainty about individual preference. Building upon the Hanemann et al. (1995) framework, Li and Mattsson (1995) extended the theory of preference uncertainty to define preference uncertainty as a stochastic error term which arises in a hypothetical valuation scenario as individuals do not know their true values of a good with certainty. Li and Mattsson (1995) argued that ignoring preference uncertainty in stated preference studies may result in measurement bias. Researchers have developed and applied a variety of methods for addressing preference uncertainty in contingent valuation (CV) studies. A number of empirical studies have used information on preference uncertainty to understand the disparity between hypothetical values and actual economic behaviour (Champ et al., 1997, Ethier et al., 2000, Champ and Bishop, 2001, Poe et al., 2002). In a second phase of preference uncertainty research, attempts have been made to develop calibration techniques to incorporate information about respondent uncertainty into welfare estimates. Empirical evidence indicates that different certainty measurement methods and calibration techniques1 generate different welfare estimates in terms of value, efficiency of the estimate (related to the notion of standard deviation) and model fit statistics (Shaikh et al. 2007).
Besides the fact that the effect of preference uncertainty on welfare estimates varies depending upon specific certainty measurement method and calibration technique, the question that currently arises is whether or not these measurement methods and calibration techniques produce consistent results across different studies. More importantly, how useful is it to perform this additional exercise of calibrating respondent uncertainty information? Does this additional information about respondents’ levels of confidence help to obtain improved welfare estimates relative to the conventional certainty model? The aim of this paper is to address these emerging issues in the light of two widely used preference uncertainty measurement methods. A number of empirical studies in the CV literature that applied either a numerical certainty scale (NCS) method or a polychotomous choice (PC) method or both methods of measuring preference uncertainty are summarized. The results of these studies are analyzed to address the research questions.
The next section of the paper presents a description of the NCS and PC methods followed by a discussion of different techniques for calibrating uncertain responses. Section 4 presents a discussion of the results from preference uncertainty models estimated by Champ and Bishop (2001), Loomis and Ekstrand (1998) and Samneliev et al. (2005). In Section 5, a summary of the validity test results of NCS and PC methods is presented. A summary of the results of preference uncertainty calibrated willingness to pay (WTP) estimates is delivered in Section 6. Section 7 contains discussion and concluding remarks.
Section snippets
The NCS and PC methods for measuring preference uncertainty
The NCS method and the PC method are two widely used techniques of measuring preference uncertainty in CV studies. Under the NCS method, the standard ‘Yes/No' dichotomous choice (DC) valuation question is followed up by a numerical certainty scale ranging from 1 to 10 where respondents are asked to indicate the level of certainty about their ‘Yes/No’ voting decision by selecting a certainty score within the scale. Li and Mattson (1995) first constructed a post-decisional confidence rating for
Treatment for NCS and PC responses
A second issue that surrounds the preference uncertainty debate is how to recode the uncertain responses. The most commonly used technique of incorporating the NCS measure of preference uncertainty is to recode the original ‘Yes/No’ DC responses based on different certainty scale cut-off points. The certainty scale can be applied by calibrating only ‘Yes' or only ‘No' responses with certainty 8, 9 and/or 10 and treating them as No (or Yes, respectively) or as missing, or calibrating both ‘Yes'
Preference uncertainty and economic theory
In this section the results of econometric models estimated to establish a causal relationship between the levels of preference uncertainty and one or a group of theoretically and intuitively expected independent variables are discussed. Although no explicit theoretical model to explain variations in preference uncertainty has been developed as yet, there is a general agreement about some hypotheses that has emerged after Loomis and Ekstrand (1998) estimated their econometric model. The
Preference uncertainty and hypothetical bias
It has been claimed that the calibration of preference uncertainty information in hypothetical CV responses can eliminate hypothetical bias (Champ and Bishop, 2001). Table 1 presents a summary of certainty calibration cut-off points in the NCS method and the treatment of uncertain responses using a PC method at which different empirical studies found hypothetical behaviour converging to actual behaviour. Champ et al. (1997) first compared certainty calibrated hypothetical DC responses for
NCS and PC adjusted WTP estimates
After surveying the preference uncertainty literature, seven empirical studies were identified in which the authors estimated preference uncertainty adjusted WTP and compared the results with a conventional DC CV WTP estimate. The effects of accounting for preference uncertainty on estimated WTP in these studies in terms of the value of the welfare estimate, efficiency of the estimate (measured in terms of 95% confidence interval of WTP over mean WTP) and model fit statistics are summarized in
Discussion and conclusion
The aim of the paper was to address emerging issues in the CV literature in relation to accounting for preference uncertainty. Summarizing the results from empirical studies that have applied the NCS and/or the PC method of measuring preference uncertainty, the consistency of treatment impacts has been investigated. In terms of the percentage difference between the preference uncertainty adjusted welfare estimates and the model fit statistics, we find that the results are somewhat mixed. Under
References (20)
- et al.
Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty
J. Environ. Econ. Manage.
(2003) - et al.
Hypothetical versus real willingness to pay in the health care sector: results from a field experiment
J. Health Econ.
(2001) - et al.
Using donation mechanisms to value non-use benefits from public goods
J. Environ. Econ. Manage.
(1997) - et al.
Discrete choice under preference uncertainty: an improved structural model for contingent valuation
J. Environ. Econ. Manage.
(1995) - et al.
Alternative approaches for incorporating respondent uncertainty when estimating willingness to pay: the case of the Mexican Spotted Owl
Ecol. Econ.
(1998) - et al.
Using conjoint analysis to derive estimates of compensating variation
J. Environ. Econ. Manage.
(1996) - et al.
Treating respondent uncertainty in contingent valuation: a comparison of empirical treatments
Ecol. Econ.
(2007) - et al.
Experimental results on expressed certainty and hypothetical bias in contingent valuation
South. Econ. J.
(1998) - et al.
Donation payment mechanisms and contingent valuation: an empirical study of hypothetical bias
J. Environ. Econ. Manage.
(2001) - et al.
An investigation of preference uncertainty in the contingent valuation study
Appl. Econ. Lett.
(2007)
Cited by (55)
Public support for community microgrid services
2022, Energy EconomicsCitation Excerpt :To help mitigate hypothetical bias, respondents were asked to rank their level of certainty in their valuation question response on a scale of 0 through 10. This numerical certainty scale is used widely as a method to mitigate hypothetical bias (Akter et al., 2008; Li and Mattsson, 1995; Morrison and Brown, 2009). Several recoding schemes were implemented in which Yes votes are recoded to No votes if they do not surpass certainty levels.
Willingness-to-pay for restoration of water quality services across geo-political boundaries
2021, Current Research in Environmental SustainabilityA review on health cost accounting of air pollution in China
2018, Environment InternationalCitation Excerpt :In general, values vary significantly due to differences in question design (Cao and Han, 2015), making results highly uncertain. In the absence of sufficient verification, WTP becomes a not-so-reliable health cost accounting tool (Akter et al., 2008). Benefit Transfer Method (BTM) applies original or adjusted information from a prior study, such as parameters or formulas, to a new research (Lin et al., 2006).
Public values and preference certainty for stream restoration in forested watersheds in Finland
2017, Water Resources and Economics