Chapter 3. Non-Market Valuation Techniques
3.3 The Contingent Valuation Method (CVM)
3.3.4 Issues Surrounding the Use of the CVM
The CV technique has been in use for thirty five years or so to estimate passive-use values. Arrow et al. (1993) observe a dramatic increase in the number of academic papers and presentations related to the contingent valuation technique resulting from the growing interest both nationally and internationally in environmental problems and policies due to global warming, controversy surrounding the validity of the technique, and the growing use of the contingent valuation technique in estimating lost passive-use values in litigation arising from American state and federal statutes designed to protect natural resources. Arrow et al. (1993) further assert that the appeal of the CVM lies in its potential to estimate lost passive-use values in damage assessments where there appears to be no behavioural trails to be followed.
The value estimates obtained using the CVM do not only depend on the scenario presented to the respondents but also on how the respondents perceive it, the payment vehicle used, the method of conducting the survey (mail, telephone, in-person interview), and the bidding (elicitation) method (Carson, Groves & Machina, 2000). Different elicitation formats yield different results. For example, results from independent studies by Gibbard, (1973) and Satterthwaite, (1975) suggest that it is impossible to formulate a continuous response question that has the same incentive and informational properties as an incentive-compatible binary discrete-choice question. However, Bateman et al. (1995) contend that differences in value estimates from
different elicitation formats can be explained by economic theory and psychology.
The CV technique has been criticized on the basis of unreliable results (Diamond & Hausman, 1994; Cummings et al., 1995; Arrow et al., 1993) which arise from:
i. Its reliance on answers to hypothetical questions rather than observed economic choices, resulting in hypothetical bias. “Ask a hypothetical question and you get a hypothetical answer” (Seller, Stoll, & Chavas, 1985, p. 158). Taylor and Douglas (1999) define hypothetical bias as observed difference between people’s responses to hypothetical referenda and real referenda. Stated values tend to over estimate the “true” values they are supposed to measure (positive hypothetical bias). Harrison et al. (2008) state that hypothetical bias arises as a result of strategic over-bidding when respondents do not expect any actual payments to be made for the provision of the good.
ii. Respondents’ answers that may be inconsistent with the tenets of rational choice. For example a respondent may express a zero value (zero WTP) for required environmental goods and services because they do not agree with some aspect of the proposed program for religious or cultural reasons or they just cannot be bothered to formulate a true or realistic value for the service or good. For example, in New Zealand, a Maori respondent may refuse to place a value on a wetland because he believes that it is wrong, according to their cultural norms, to put a monetary value on any natural resource.
iii. Respondents’ lack of understanding regarding what it is they are being asked to value due to inexperience in valuing non-use benefits and public goods. Respondents may, as a result, value something different from what the researcher is trying to measure (Mitchell & Carson, 1989; Carson, Groves & Machina, 2000).
iv. Poorly designed survey (Mitchell & Carson, 1989; Haab & McConnell, 2003). Responses to a poorly designed survey questionnaire may not be consistent with expectation and the results from data analysis may not have economic interpretation.
v. Respondents’ failure to take contingent valuation questions seriously because of the hypothetical nature of the whole exercise and the non-binding nature of the results of the surveys. Where there is no incentive to state the true preference,
respondents are tempted to just state any large amount since their response is not binding.
vi. Strategic Bias / Over and Under Bidding. Respondents may either overstate or understate their WTP in an attempt to influence the outcome or results in their favour. A simple test for strategic bias is the analysis of the distribution of WTP responses. A bimodal distribution or the existence of fat tails may be an indication of significant strategic bias. Schultze, d’Arge and Brookshire (1981) conclude, from a review of six contingent valuation studies, that strategic bias in revealing consumer preferences is not likely to be a major problem. This conclusion is also supported by Smith (1977); Grether and Plott (1979); and Bishop, Heberlein and Kealy (1983).
vii. Anchoring and Starting Point Bias Starting point bias is mainly associated with the use of iterative bidding in CV
surveys. Mitchell and Carson (1989) contend that respondents may interpret the initial bid amount as being indicative of the value which the interviewer or questionnaire expects from them. If the starting bid is too high or too low in relation to the respondent’s true WTP value, respondents may change their responses prematurely to get over the question quickly and save their time. For single bounded DC formats respondents may exhibit a tendency to agree regardless of the bid level (O’Conor et al., 1999). This is an extreme form of anchoring (100%) which Mitchell and Carson (1989) refer to as “yea-saying”.
viii. Truncation and Endogenous Stratification effects Where onsite samples are taken, non-users of the sites are excluded from the
sample so that the observations are truncated at positive site demand. The likelihood of an individual being sampled depends on the frequency of visiting the site so that those who frequent the site most (who have similar characteristics or interests) are most likely to be selected. This is referred to as endogenous stratification (Shaw, 1988; Englin & Shonkwiler, 1995).
ix. Embedding effects
Respondents may fail to separate aspects of the programme from the overall national or regional environmental policy and state maximum WTP reflecting the overall valuation of the environmental issues instead of the proposed programme. Where a resource is valued as part of a larger programme or
package, WTP is lower than when it is valued alone (Loomis, Lockwood, & DeLacy, 1993; Brown, Barro, Manfredo, & Peterson, 1995).
x. Warm Glow / Moral satisfaction
When respondents are asked to state their maximum WTP for a given programme, the stated values may not be based entirely on the demand for the goods and services delivered by the program as assumed in the survey. Respondents instead may feel good about contributing to a worthy cause; i.e. respondents may derive satisfaction from giving per se (Andreoni, 1989; Arrow et al., 1993).
xi. Scope
Responses may not always be sensitive to the quantity and quality of the proposed change (Diamond & Hausman, 1994; Ready, Berger, & Blomquist, 1997). In a study by Boyle et al. (1994), respondents’ mean WTP to avoid the loss of 2000, 20000 and 200000 migratory birds was found not to differ significantly. Arrow et al. (1993) suggest that studies finding such insensitivity to scope may be defective in that the choices may not have been clearly presented to the respondents so that they view the alternatives as being essentially the same. Scope sensitivity may be tested for, ex-post, by estimating a value function from the CV data and assuring that it has a positive slope – indicating sensitivity to the scope of the change (Ready, Berger, & Blomquist, 1997).
xii. Free Ridding
Respondents may find it in their best interest to state zero values if they suspect that the commitment is binding and expect others to commit so that the programme is implemented (Mitchell & Carson, 1989). Once the service or good is available they will not be excluded from consuming it.
xiii. Zero protest bids
Zero protest bids have been treated differently in contingent valuation studies as the majority of researchers discard them (Bennett et al., 1998; Bowker & Didychuk, 1994; Colombo, Calatrava-Requena, & Hanley, 2006) while others include these as genuine zeros in the analysis (Jorgensen & Syme, 2000; Bateman et al., 2006a). Those who discard protest bids specifically test for these in their surveys through a follow up question on zero bids to determine if the individual is in fact protesting or not.
xiv. Outliers or extreme tails
Carson, Wilks, and Imber (1994) state that, in the binary discrete choice and double bounded discrete choice methods, extreme values are avoided. Where an open ended question is posed, respondents may state high values where they know that their responses are not binding
xv. Mean or median estimates. Where sample data is normally distributed, the mean and median may coincide or be very close to each other so that selecting either does not influence the end result. However where the sample is skewed to one side or the distribution has fat tail(s), the mean and median will take on significantly different values. In these cases the median is the preferred measure of central tendency (Hanemann, 1984) as it gives the value that applies to at least 50% of the sample. In a study by Streever et al. (1998) the median was selected as the more appropriate measure of central tendency since the sample WTP distribution was skewed to the right. The median gave more conservative results than the mean as the mean was higher than the median.
xvi. Inequivalance between WTA and WTP for the same environmental change. A detailed discussion of this issue was presented in Chapter 2.
xvii. Self-selection bias or non-response bias. The decision to respond to a survey questionnaire depends on the importance, or lack of it, the respondent attaches to the service, good or resource being valued. Individuals who value the service, good or resource highly are most likely to respond as opposed to those individuals who value the good less. The majority of the responses may therefore reflect a high valuation from a particular stratum of the population. Applying estimates from such survey responses to obtain aggregate values over the entire population may result in serious overestimates of value (Bateman et al., 2000a).
3.3.4.1 Suggested Solutions to Some of the Problems Associated with the CVM
Table 3.1 provides a summary of suggested solutions to some of the problems associated with the CVM. The objective of these solutions was to make the results of CVM more acceptable buy resolving some of the major areas of contention.
Table 3.1 Summary of problems and suggested solutions
Study Problem Suggested Solution or Action
Bishop and Heberlein (1979)
Strategic bias Non-response bias Extreme values
Use of referendum contingent valuation format. This method is recommended by NOAA (1993); Mitchell & Carson (1989). Schulze et al.
(1981)
Strategic bias A review of six studies revealed that strategic bias is not likely to be a major problem. Suggest an ex-ante visual inspection of the frequency distribution of the WTP responses to find out if a bi-modal clustering of values at abnormally high and/or low levels exists.
Mitchell and Carson (1981, 1984 as cited in Mitchell & Carson, 1989)
Starting point bias Used anchored payment cards and tested their results for the effects of anchoring on final bids and found no evidence of any influence.9
Walsh et al. (1984) Self-selection or non- response bias
Urging all to respond. Financial incentive may be used to motivate responses.
Stevens et al. (1991, p. 393)
Free rider effects The use of a payment vehicle “such as taxes, which exact payment from everyone” reduces free riding. Carson, Wilks &
Imber (1994)
Extreme values (outliers) In the binary discrete choice and double bounded discrete choice methods, extreme values are avoided. NOAA (1994,
1996)
Hypothetical bias Blue ribbon panel recommended that hypothetical bids be deflated using a ‘divide by 2’ rule unless these bids can be calibrated using actual market data.
9
Boyle and Bishop (1988) cite a study by Randall, Hoehn and Tolley (1981) which found that anchored payment cards generated fewer protest bids than did the other questioning formats. They also found evidence of strong influence of anchors on the final bids.
Carson, Wilks, and Imber (1994)
Strategic bias Used double bounded, discrete-choice elicitation format and argued that it gives respondents little opportunity to bias their answers.
Ready, Berger, and Blomquist (1997)
Scope and non-zero intercept
Estimated a flexible value function from contingent valuation data that allowed for statistical test for positive slope and for non-zero intercept. Streever et al.
(1998)
Starting point bias Used a range of randomly generated starting bids (seed values) to prevent any systematic bias upwards or downwards.
List and Shogren (1998)
Hypothetical bias Estimated a calibration function to correct for the exaggeration in the WTP (calibrated the welfare estimates ex post).
Cummings and Taylor (1999); List (2001);
Hypothetical bias Advocated the use of “cheap talk” to mitigate ex ante the effects of hypothetical bias in CVM. Harrison and Rutström. (1999) Bateman et al. (2000a) Self-selection or non- response bias.
Aggregate benefits should only be obtained by scaling the sample estimates over the proportion of the population that reflects this selection bias. For example statistics may be available on household usage or visitation to a site.