Contingent Valuation

Contingent Valuation in Health Care Evaluation

Contingent valuation puts a money value on a change that has no market price by describing the change in a survey and asking people whether, or how much, they would pay for it. In health economics it supplies the benefit side of a cost-benefit analysis and can value things a QALY leaves out, such as reassurance, convenience or the way a service is delivered. The answers are only as good as the hypothetical market built around the question, and the way yes or no answers are turned into a mean or a median can decide whether a programme passes a cost-benefit test. The welfare theory is on the willingness to pay page.

From oil spill damages to health care programmes

The method grew up in environmental economics. Johnston and colleagues, in their guidance for stated preference studies, record that the 1989 Exxon Valdez oil spill changed the field permanently. After consultants for Exxon argued that contingent valuation could not reliably measure non-use values, the US National Oceanic and Atmospheric Administration (NOAA) convened an expert panel co-chaired by Kenneth Arrow and Robert Solow. Its 1993 report concluded that studies following its guidelines could produce estimates reliable enough to be the starting point of a judicial process of damage assessment.

In health care, Diener, O'Brien and Gafni found 48 studies published between 1984 and 1996, 42 of them carried out for cost-benefit analysis, and only three of those 42 stated whether they were measuring compensating or equivalent variation. Reviewing 111 health studies published since 1985, Smith judged that they had performed poorly in constructing and presenting the contingent market, with little if any improvement over 15 years, and Smith and Sach later described the method's influence on national health decision making as non-existent.

Building the hypothetical market

The survey has to tell respondents what they are buying and how they would pay. Johnston and colleagues recommend presenting the baseline, the mechanism of change and the change itself, with its timing, uncertainty and risk, and checking that respondents understand and believe them. Smith's review assessed five parts of the market in health studies, namely the scenario, the payment vehicle, the expression of risk, the time period of valuation and the way the survey was administered.

The payment vehicle is the means of payment, such as a tax, an insurance premium or a charge at the point of use. Johnston and colleagues recommend one that is realistic, credible, familiar and binding, and report clear evidence that the choice of vehicle can change the values obtained. In a tax-funded health system a user charge can look implausible or provoke objections to charging.

Who is asked, and in what role, is also part of the design. O'Brien and Gafni argued that health studies should distinguish compensating variation from equivalent variation and recognise that respondents can gain or lose from a programme, asking willingness to pay or willingness to accept questions as appropriate. Values given ex ante, before people know whether they will need a service, and ex post, once they know, answer different questions, and Klose found the influence of this state of valuation had rarely been investigated.

Asking the valuation question

An open-ended question asks for a maximum amount, a bidding game moves an amount up or down until a maximum is reached, a payment card shows a range, and a dichotomous choice question offers one amount, varied across respondents, for a yes or no answer. The US Environmental Protection Agency guidelines for economic analyses note starting point bias in iterative bidding, possible effects of the range and intervals on a payment card, the larger samples needed for dichotomous choice and, in double-bounded designs that follow up with a higher or lower amount, second answers that may not be independent of the first. In health studies, Klose found that first bids in bidding games significantly influenced valuations and saw indications of range bias from payment cards, but found no strategic bias.

The NOAA panel recommended posing the question as a vote on a referendum and asking willingness to pay instead of compensation required, as the conservative choice. Johnston and colleagues add that a single binary choice question per respondent is the most straightforward way to make the question incentive compatible for public goods, provided respondents see the choice as consequential and the payment as binding.

Estimating willingness to pay from yes or no answers

A dichotomous choice answer reveals only whether a respondent's maximum payment lies above or below the amount offered. The share answering yes at each amount traces out the share whose willingness to pay exceeds it, and a binary response model such as the logit smooths that curve.

$$P(\text{yes} \mid B) = \frac{1}{1 + e^{-(\alpha - \beta B)}}$$

where $P(\text{yes} \mid B)$ is the probability of accepting bid amount $B$, $\alpha$ is the estimated intercept and $\beta$ is the estimated bid coefficient, positive so that acceptance falls as the bid rises. The median is the bid that sets the index $\alpha - \beta B$ to zero.

$$\text{Median WTP} = \frac{\alpha}{\beta}$$

where $\alpha$ and $\beta$ are the coefficients defined above. Because the fitted logistic distribution is symmetric its mean equals the median, but it includes negative values, implying that some respondents would need paying to accept the programme. If willingness to pay is held at zero or above, the mean is the area under the acceptance curve from zero, which the analyst may also stop at the highest bid offered.

$$E[\text{WTP}] = \int_{0}^{B_{\max}} P(\text{yes} \mid b) \, db = \frac{1}{\beta} \left[ \ln\left(1 + e^{\alpha}\right) - \ln\left(1 + e^{\alpha - \beta B_{\max}}\right) \right]$$

where $E[\text{WTP}]$ is mean willingness to pay, $b$ is the bid over which the integral runs and $B_{\max}$ is the upper limit; letting $B_{\max}$ grow without limit drops the second term and gives the mean truncated at zero. Johnston and colleagues recommend reporting central tendency and dispersion, computed by methods that are transparent and theoretically and statistically well defined.

Worked example of a referendum survey for a community nursing service

All figures are illustrative. A health authority serving 120,000 households is considering a rapid-response community nursing service that treats frail older people at home instead of in hospital, at a cost of GBP 7,500,000 a year. A survey asks whether respondents would vote for it if it added a stated amount to household taxes each year. Each respondent sees one of five amounts from GBP 10 to GBP 200, and a logit model fitted to the answers gives $\alpha = 1.2$ and $\beta = 0.02$.

1. Median. The median is 1.2 / 0.02 = 60, so half of households would vote for the service at GBP 60 a year.

2. Mean with willingness to pay held at zero or above. Since $e^{1.2} = 3.3201$ and ln(4.3201) = 1.46328, the mean truncated at zero is 1.46328 / 0.02 = GBP 73.16.

3. Mean restricted to the bid range. With $B_{\max} = 200$, the second term uses $e^{-2.8} = 0.06081$ and ln(1.06081) = 0.05903, so the mean is (1.46328 − 0.05903) / 0.02 = GBP 70.21.

4. Aggregate and net benefit. The service costs 7,500,000 / 120,000 = GBP 62.50 per household a year. Each mean is multiplied by 120,000 households (annual totals in GBP).

5. Interpretation. The sign of the net benefit turns on an assumption about the tails, not on the data, so the choice should be fixed in advance and the alternatives reported. With willingness to pay held at zero or above, the mean shows the service passing the cost-benefit test, yet at a tax of GBP 62.50 the model predicts that only 48.75% of households would vote yes, because the median lies below the cost per household. A minority who value a service highly can make it efficient in the Kaldor-Hicks sense while it would still lose a vote.

Testing whether the values are credible

Johnston and colleagues group validity checks into content, construct and criterion validity, the last rarely possible because it needs a presumed true value, and note that economists looking at the same scope test evidence have reached opposite conclusions. The NOAA panel put the burden of proof on survey designers and would judge a study unreliable if it showed, among other faults, a high non-response rate or inadequate responsiveness to the scope of the change.

The NOAA panel read the experimental evidence then available as showing that the technique was likely to overstate real willingness to pay. Johnston and colleagues point out that this conclusion of hypothetical bias rests largely on comparisons with cash payments for private goods, and that studies using actual votes as the benchmark suggest the bias is not universal. For protest responses, where a respondent rejects the scenario or payment vehicle instead of the good, they find no agreed best practice and call for debriefing questions, transparency and sensitivity analysis. In health care, Klose found theoretical validity shown but no results on criterion validity and few on reliability.

Where contingent valuation results are used

In a cost-benefit analysis aggregate willingness to pay is set against the opportunity cost of a programme to give its net benefit. The NICE technology appraisal manual (PMG36, section 4.3.1) asks for health effects to be expressed in QALYs for cost-effectiveness analyses, so a money value from a survey does not enter the NICE reference case.

The Green Book (2026) defines stated preference valuation as asking people directly what they would be willing to pay, cites the value of a QALY used by the Department of Health and Social Care as an example, and advises that it should only be used if robust revealed preference data are not available. Published values reach new settings through benefit transfer, and values for a QALY or for small reductions in the risk of death inform the willingness-to-pay threshold and the value of statistical life.

Contingent valuation and neighbouring methods

Willingness to pay is the quantity, and contingent valuation is one way of measuring it. Neighbouring methods produce different outputs.

Contingent valuation values the whole package and cannot say which feature of a service drives the value, which a choice experiment can. Revealed preference, which the Green Book describes as inferring prices from people's actions and decisions, avoids the hypothetical market but cannot value a service that does not yet exist.

Limitations and common misreadings

Nobody actually pays, and every estimate is conditional on the scenario, payment vehicle, format, bid design and treatment of protests. Four misreadings recur. The first is aggregating the median, which describes the typical respondent and a referendum outcome, when the welfare total needs the mean. The second is mismatching units, such as multiplying a household value by the number of individuals. The third is reading a stated amount as a proposal to charge patients. The fourth is adding willingness to pay to a QALY-based benefit for the same health gain, which counts it twice. Because willingness to pay is bounded by income, unweighted totals also favour the better-off.

Sources

• Arrow K, Solow R, Portney PR, Leamer EE, Radner R, Schuman H. Report of the NOAA Panel on Contingent Valuation. Federal Register. 1993;58(10):4601-4614. https://www.govinfo.gov/content/pkg/FR-1993-01-15/pdf/FR-1993-01-15.pdf
• Diener A, O'Brien B, Gafni A. Health care contingent valuation studies: a review and classification of the literature. Health Economics. 1998;7(4):313-326. https://doi.org/10.1002/(SICI)1099-1050(199806)7:4%3C313::AID-HEC350%3E3.0.CO;2-B
• HM Treasury. The Green Book (2026). London: HM Treasury; 2026. Chapter 8. https://www.gov.uk/government/publications/the-green-book--2/the-green-book-2026
• Johnston RJ, Boyle KJ, Adamowicz W, Bennett J, Brouwer R, Cameron TA, et al. Contemporary guidance for stated preference studies. Journal of the Association of Environmental and Resource Economists. 2017;4(2):319-405. https://doi.org/10.1086/691697
• Klose T. The contingent valuation method in health care. Health Policy. 1999;47(2):97-123. https://doi.org/10.1016/S0168-8510(99)00010-X
• National Institute for Health and Care Excellence. NICE technology appraisal and highly specialised technologies guidance: the manual (PMG36). London: NICE; 2022, last updated 13 October 2026. Section 4.3.1. https://www.nice.org.uk/process/pmg36/chapter/economic-evaluation-2
• O'Brien B, Gafni A. When do the "dollars" make sense? Toward a conceptual framework for contingent valuation studies in health care. Medical Decision Making. 1996;16(3):288-299. https://doi.org/10.1177/0272989X9601600314
• Smith RD. Construction of the contingent valuation market in health care: a critical assessment. Health Economics. 2003;12(8):609-628. https://doi.org/10.1002/hec.755
• Smith RD, Sach TH. Contingent valuation: what needs to be done? Health Economics, Policy and Law. 2010;5(1):91-111. https://doi.org/10.1017/S1744133109990016
• US Environmental Protection Agency. Guidelines for Preparing Economic Analyses. Washington, DC: US EPA; 2010, updated May 2014. Chapter 7. https://www.epa.gov/sites/default/files/2017-08/documents/ee-0568-50.pdf