Willingness to Pay

Willingness to Pay in Health Economics

Willingness to pay (WTP) converts a person's preference for a change in health or health care into a sum of money, which is what allows cost-benefit analysis to set health benefits and resource costs against each other on a single scale. Its theoretical standing comes from welfare economics, where it is a Hicksian measure of a change in wellbeing, but its practical value rests on survey and statistical methods with well-documented weaknesses. For health economists the concept matters in two places, as the benefit measure in cost-benefit analysis and as the evidence behind attempts to attach a monetary value to a QALY.

The concept here is the individual, demand-side one. WTP is defined through compensating and equivalent variation, it is elicited through contingent valuation and discrete choice experiments or inferred from behaviour, and a credible study has to pass recognised validity tests. It is usually smaller than willingness to accept, it is open to the objection that it reflects ability to pay, and it links to the value of a QALY. The value per QALY built up from such preferences is the subject of the willingness-to-pay threshold, and the benchmark a health system applies to cost-effectiveness results is the cost-effectiveness threshold.

Hicksian foundations in compensating and equivalent variation

WTP is defined against a utility level that the individual is held to, not against a market price. The simplest case is a person with income Y who moves from health state H0 to a better state H1. WTP for that improvement is the payment that leaves them exactly as well off with the improvement as they were without it, which can be written as follows.

U(Y − WTP, H1) = U(Y, H0)

where U is the individual's utility function, Y is income before the change, H0 is the initial health state and H1 is the improved health state. Because the payment returns the individual to their original utility level, WTP for a gain is the compensating variation of that gain.

The same framework defines the other Hicksian measures, and the four combinations are easy to confuse. For a gain, the equivalent variation is the smallest sum the person would accept in place of the improvement, which is a willingness-to-accept measure. For a loss the roles reverse. The most a person would pay to avoid a deterioration is its equivalent variation, and the least compensation they would accept to bear it is its compensating variation.

Which measure is appropriate therefore depends on the reference point, that is, on whether the individual is taken to be entitled to the status quo or to the changed state. Two further features follow from the definition. WTP is measured against the individual's own budget, so it is bounded by income and wealth, which is the root of the equity concern discussed below. It is also not confined to health, since a person can value reassurance, convenience or the process of care, none of which a QALY captures. Where a good is bought at a price, the difference between WTP and the price paid is the consumer surplus from that purchase.

From individual values to cost-benefit analysis

In cost-benefit analysis the monetary benefit of a programme is the sum of the compensating variations of everyone affected. If that sum exceeds the opportunity cost of the resources used, the gainers could in principle compensate the losers and still be better off, which is the Kaldor-Hicks potential Pareto improvement test. The compensation is hypothetical and is not actually paid, which is one reason the distribution of WTP across income groups matters.

Aggregation uses mean WTP multiplied by the relevant population, because that product equals the sum of individual values. Median WTP describes the typical respondent and indicates whether a majority would support the programme at a given cost per head, but it is not a measure of total benefit. Reporting both is good practice, since a large gap between them signals a skewed distribution driven by a small number of high values.

Olsen and Smith, reviewing 71 WTP surveys of health and health care published in English between 1985 and 1998, identified two main arguments for preferring WTP to QALYs, namely that it allows a more comprehensive valuation of benefits and that it can improve social efficiency because benefits can be set against social costs. Their review asked how far the surveys lived up to these arguments, both in how fully their scenarios described the benefits being valued and in whether they compared WTP with social costs at all.

Contingent valuation and the hypothetical market

Contingent valuation asks respondents to value a described change within a constructed market. The credibility of the answer depends on how that market is built, and the health care record here is weak. Smith's critical review of 111 contingent valuation studies in health care published since 1985 assessed five aspects of market construction, covering the scenario and how it is presented, the payment vehicle, the expression of risk, the time period of valuation and the way the survey is administered. It concluded that studies had performed poorly in constructing, specifying and presenting the contingent market, with little, if any, improvement over the preceding 15 years.

The payment vehicle is the means by which the respondent is told they would pay, such as a tax, an insurance premium or a charge at the point of use. It has to be believable in the setting, because respondents who reject the vehicle may reject the whole question. In a tax-funded health system, an out-of-pocket charge can itself provoke refusals that say more about attitudes to charging than about the value of the service.

Elicitation formats and why they give different answers

The valuation question can be put in several ways, and the choice of format changes the numbers obtained. The four formats most used in health care differ in how much work they ask of the respondent and in the biases they invite.

The benchmark guidance comes from a panel of the US National Oceanic and Atmospheric Administration (NOAA), whose members included Kenneth Arrow and Robert Solow and whose report was published in 1993, on the use of contingent valuation in natural resource damage assessment under the Oil Pollution Act of 1990. The panel judged open-ended questions unlikely to provide the most reliable valuations, because the task lacks realism and invites strategic overstatement. Its guidelines recommended posing the valuation question as a vote on a referendum and using the WTP format instead of compensation required, because WTP is the conservative choice. The panel also thought reliable values unlikely from mail surveys, regarded face-to-face interviews as usually preferable, and called for a conservative design wherever choices are ambiguous, reminders of substitutes and of other things the money could be spent on, and an explicit "no-answer" option, with yes and no votes followed by a question asking why.

Whether guidance written for environmental damage transfers to health care has been debated in the health economics literature, and evidence from health care confirms that format matters. In a UK study by Frew and colleagues of alternative methods of colorectal cancer screening, a closed-ended format produced significantly higher WTP than open-ended and payment scale formats, which gave broadly similar values; the authors proposed anchoring and yea-saying as explanations. A follow-up study that elicited values for the same tests through a bidding game in interviews found considerably higher valuations than either questionnaire format, and significant differences between final values reached from different initial bids.

Discrete choice experiments with a cost attribute

A discrete choice experiment asks respondents to choose repeatedly between alternatives described by attributes, such as waiting time, place of treatment or chance of success, one of which is a cost. Choices are analysed with a random utility model. Because cost enters the utility function, the rate at which respondents trade an attribute against money gives its marginal WTP.

marginal WTP for attribute k = −βk ÷ βc

where βk is the estimated utility coefficient on attribute k and βc is the estimated coefficient on the cost attribute, which is negative when a higher cost reduces the utility of an alternative. The ratio is the amount of money that leaves utility unchanged when attribute k improves by one unit.

Discrete choice experiments value the components of a service separately and can include attributes that lie outside health. Lancsar and Louviere's user's guide notes that they could contribute to outcome measurement for both cost-benefit and cost-utility analysis. They share the central weakness of stated preference, because the cost attribute is hypothetical. In an experiment by Ozdemir and colleagues on rheumatoid arthritis treatments, respondents shown a "cheap talk" script, which warns about the tendency to overstate in hypothetical surveys, gave generally lower WTP estimates; the script changed the coefficient on cost and also shifted preferences for other attributes.

Revealed preference and values inferred from behaviour

Revealed preference methods infer WTP from choices that people actually make, which avoids the hypothetical market altogether. Examples include wage premiums accepted for riskier jobs, spending on safety or preventive goods, and the time and travel costs people incur to reach a service. Values for small changes in mortality risk from such studies are commonly expressed as a value of a statistical life, calculated as WTP for a small reduction in the risk of death divided by the size of that reduction.

The approach is harder to apply to health care itself. Insurance and public funding mean that patients rarely face prices that reflect cost, clinicians act as agents in many choices, and patients often lack information about outcomes, so observed behaviour is constrained in ways that obscure preferences. Revealed preference can also value only changes that already exist in a market, not a new treatment or a proposed service.

Validity tests and known biases in stated WTP

Stated WTP is only useful if it responds to the things theory says it should and is not driven by features of the survey that should not matter. Four problems recur in the health literature, and each has a design response and a test that a credible study reports.

Sensitivity to scope

Theory requires WTP not to fall as the quantity of a desired good rises, so a larger health gain or risk reduction should attract at least as high a valuation. It does not require proportionality, because diminishing marginal utility and income effects mean a doubled benefit can reasonably attract less than double the WTP. The NOAA panel listed inadequate responsiveness to scope among the problems that would make a study's findings unreliable. Diamond and Hausman went further, arguing that the "embedding effect", in which WTP responses are similar for goods that theory says should be valued differently, reflects an absence of preferences rather than poor survey design, and concluding that current contingent valuation methods were not suitable for benefit-cost analysis. In health care, Olsen, Donaldson and Pereira reported two studies in Norway and Portugal that varied the number of patients treated by a programme and the size of a reduction in the risk of heart attack found no statistically significant differences in WTP between the different-sized health effects.

Hypothetical bias

Respondents do not pay the amounts they state, and stated values tend to exceed what people would pay in a real transaction. The NOAA panel was persuaded that hypothetical markets tend to overstate WTP for private as well as public goods, and it presented conservative design choices as a way of offsetting the bias. Cheap talk scripts and calibration against real payments, where these can be observed, are the main design responses.

Starting-point and anchoring bias

Any amount shown to a respondent can act as a cue to the "right" answer. Stålhammar's Swedish survey of patients taking antisecretory drugs gave a clear example. In it, 82 patients willing to pay extra for a drug that could be taken with meals completed a bidding game that started at either SEK 20 or SEK 1,000. The average WTP was SEK 70 among the 42 patients offered the low starting bid and SEK 289 among the 40 offered the high one, against SEK 138 overall. The range printed on a payment card and the bid amounts in a dichotomous choice design can anchor answers in the same way, which is why the bid design is piloted and reported.

Protest zeros

Some zero responses reflect rejection of the scenario, the payment vehicle or the principle of paying, not a zero valuation. A respondent who believes a service should be funded through general taxation may refuse to state an out-of-pocket amount while valuing the service highly. Follow-up questions separate protest zeros from genuine zeros. Excluding protesters assumes their values resemble those of other respondents, while counting them as zeros understates mean WTP, so analyses commonly report results both ways. The NOAA panel warned that eliminating protest zeros and other problematic responses can leave an effective response rate so low that the remaining sample no longer represents the population.

The gap between willingness to pay and willingness to accept

Willingness to accept (WTA) is the compensation-based counterpart of WTP, and the choice between them follows the entitlement question set out above. For a good with close substitutes, economic theory predicts the two should be close. Hanemann showed that, for a change in the quantity of a non-market good, the gap depends on the ratio of the income elasticity of demand for the good to its elasticity of substitution with other goods, so it can be large when the good has few substitutes even if income effects are modest. A person's own health, or access to the only effective treatment, is a clear case of a good with few substitutes.

Empirically, WTA usually exceeds WTP by a wide margin. Horowitz and McConnell's review of WTA/WTP studies found the ratio highest for public and non-market goods, next highest for ordinary private goods and lowest for experiments involving forms of money. Behavioural explanations such as loss aversion and the endowment effect offer a competing account in which the gap reflects the reference point, not income or substitution effects.

The practical consequence is that the choice of measure can decide the result. The NOAA panel recommended WTP partly because WTA should exceed it, making WTP the conservative choice. Where a service that people already have is to be withdrawn, however, and its users are taken to be entitled to it, WTA is the appropriate measure of the loss, and valuing it by WTP understates it.

Ability to pay and the equity objection

Because WTP is bounded by income, people with higher incomes will typically state more for the same health gain, and an unweighted sum of WTP gives their preferences more weight. Where programmes benefit different income groups, cost-benefit analysis based on WTP can favour programmes that benefit the better-off, which conflicts with the principle, common in publicly funded systems, that access should depend on need. Studies respond by reporting WTP by income group, estimating how WTP varies with income, applying distributional weights, or asking respondents to value programmes from a societal perspective, for example through taxation for services that others will use.

The objection is not unique to WTP. Donaldson, Birch and Gafni's analysis of the distribution problem in economic evaluation showed that the same distributional concerns apply to non-monetary valuations of health, to the measurement of costs and to the decision rules of cost-effectiveness analysis. On that argument, preferring cost-effectiveness analysis to cost-benefit analysis cannot be justified on the grounds that it avoids distributional questions.

Willingness to pay, the value of a QALY and the threshold

Dividing WTP for a health gain by the QALYs it produces gives a WTP per QALY, and a sizeable literature has used this route to estimate the monetary value individuals place on health. Ryen and Svensson identified 24 studies containing 383 unique estimates, with a trimmed mean of 74,159 euros and a median of 24,226 euros at 2010 prices. The WTP for a QALY was significantly higher when the gain came from life extension than from improved quality of life, and it depended on the size of the QALY gain valued, which is the scope problem reappearing in a different form.

How the question is framed matters here too. Gyrd-Hansen's Danish study, which derived WTP per QALY from preferences over health states, found a value that differed considerably from those implied by contingent valuation studies of reductions in the risk of death, suggesting that ex ante valuation of risk and ex post valuation of quality of life rest on different preference structures.

Estimates of this kind describe the consumption value that individuals place on health. They are one possible basis for a cost-effectiveness threshold, but a threshold used by a health system can instead be grounded in the health that its budget would otherwise buy. How thresholds are set, and why individual WTP for one's own health is not the same as the value a health system should place on a QALY, is covered on the willingness-to-pay threshold and cost-effectiveness threshold pages.

Worked example of a payment card survey for a screening programme

The figures in this example are illustrative. A health authority is considering a community screening programme for an adult population of 100,000, with a present value cost of £5,500,000. A pilot contingent valuation survey describes the programme and asks each respondent to mark, on a payment card listing £0, £10, £20, £50, £100, £200 and £500, the highest one-off payment through local taxation they would make for it. Ten respondents answer, and follow-up questions identify two of the zero responses as protests against paying through local taxation.

The total of the amounts marked, in pounds, is calculated as follows.

0 + 0 + 0 + 10 + 20 + 20 + 50 + 100 + 100 + 200 = 500

Mean WTP depends on how the protest zeros are treated. Counting them as zeros, mean WTP = 500 / 10 = 50. Excluding them, mean WTP = 500 / 8 = 62.5. The medians are lower, as expected with a skewed distribution. Across all ten responses the median is (20 + 20) / 2 = 20, and with protests excluded it is (20 + 50) / 2 = 35.

Aggregate benefit is mean WTP multiplied by the population. With protests counted as zeros, 50 × 100,000 = 5,000,000. With protests excluded, 62.5 × 100,000 = 6,250,000. Net benefit is aggregate benefit minus the programme cost.

With protests counted as zeros, net benefit is 5,000,000 − 5,500,000 = −500,000 in pounds.

With protests excluded, net benefit is 6,250,000 − 5,500,000 = 750,000 in pounds.

The sign of the result, and so the recommendation, turns on a single analytical choice about two respondents. Excluding the protesters assumes their valuations resemble those of the other respondents, which the follow-up answers cannot confirm; counting them as zeros treats an objection to the payment vehicle as a zero valuation. The payment card adds a second, conservative element, since each marked amount is a lower bound and the true maximum lies somewhere below the next amount on the card. Median-based totals, 20 × 100,000 = 2,000,000 or 35 × 100,000 = 3,500,000, would suggest the programme fails by a wide margin, but the median is not a welfare total and would be the wrong basis for the comparison. A real study would use a much larger sample, report confidence intervals, model the interval nature of payment card data, and present the protest treatment as a sensitivity analysis rather than choosing the more favourable result.

Common misreadings of willingness to pay

A common error is to treat an individual's WTP for a health gain as interchangeable with a cost-effectiveness threshold. The first is a demand-side valuation by the person affected; the second is a decision rule set by or for a health system, and the two can differ widely.

A second error is to read a stated WTP as a price or a proposal to charge. The payment vehicle is a device for eliciting a valuation, and a WTP study does not imply that patients should or would pay the amount stated.

A third is to treat a WTP estimate as a fixed property of the good. Values depend on the elicitation format, the starting bid or card range, the payment vehicle, the handling of protest responses and whether the question is framed ex ante or ex post. An estimate reported without these details cannot be compared with others, and transferring a value from one population or setting to another without allowing for differences in income and context carries the same risk.

A fourth is to dismiss the WTP and WTA gap as survey error. Part of it can be explained by substitution and income effects, and part by reference-dependent preferences; either way, choosing between the two measures is a judgement about entitlement, not a technical correction.

Sources

• Arrow K, Solow R, Portney PR, Leamer EE, Radner R, Schuman H. Report of the NOAA Panel on Contingent Valuation. Federal Register. 1993;58:4601-4614.
• Diamond PA, Hausman JA. Contingent valuation: is some number better than no number? Journal of Economic Perspectives. 1994;8(4):45-64.
• Donaldson C, Birch S, Gafni A. The distribution problem in economic evaluation: income and the valuation of costs and consequences of health care programmes. Health Economics. 2002;11(1):55-70.
• Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. 4th ed. Oxford: Oxford University Press; 2015.
• Frew EJ. Benefit assessment for cost-benefit analysis studies in health care using contingent valuation methods. In: McIntosh E, Clarke PM, Frew EJ, Louviere JJ, editors. Applied Methods of Cost-Benefit Analysis in Health Care. Oxford: Oxford University Press; 2010.
• Frew EJ, Whynes DK, Wolstenholme JL. Eliciting willingness to pay: comparing closed-ended with open-ended and payment scale formats. Medical Decision Making. 2003;23(2):150-159.
• Frew EJ, Wolstenholme JL, Whynes DK. Comparing willingness-to-pay: bidding game format versus open-ended and payment scale formats. Health Policy. 2004;68(3):289-298.
• Gyrd-Hansen D. Willingness to pay for a QALY. Health Economics. 2003;12(12):1049-1060.
• Hanemann WM. Willingness to pay and willingness to accept: how much can they differ? American Economic Review. 1991;81(3):635-647.
• Horowitz JK, McConnell KE. A review of WTA/WTP studies. Journal of Environmental Economics and Management. 2002;44(3):426-447.
• Johannesson M. Theory and Methods of Economic Evaluation of Health Care. Developments in Health Economics and Public Policy. Boston: Springer; 1996.
• Lancsar E, Louviere J. Conducting discrete choice experiments to inform healthcare decision making: a user's guide. PharmacoEconomics. 2008;26(8):661-677.
• Olsen JA, Donaldson C, Pereira J. The insensitivity of 'willingness-to-pay' to the size of the good: new evidence for health care. Journal of Economic Psychology. 2004;25(4):445-460.
• Olsen JA, Smith RD. Theory versus practice: a review of 'willingness-to-pay' in health and health care. Health Economics. 2001;10(1):39-52.
• Ozdemir S, Johnson FR, Hauber AB. Hypothetical bias, cheap talk, and stated willingness to pay for health care. Journal of Health Economics. 2009;28(4):894-901.
• Ryan M, Gerard K, Amaya-Amaya M, editors. Using Discrete Choice Experiments to Value Health and Health Care. Dordrecht: Springer; 2008.
• Ryen L, Svensson M. The willingness to pay for a quality adjusted life year: a review of the empirical literature. Health Economics. 2015;24(10):1289-1301.
• Smith RD. The discrete-choice willingness-to-pay question format in health economics: should we adopt environmental guidelines? Medical Decision Making. 2000;20(2):194-206.
• Smith RD. Construction of the contingent valuation market in health care: a critical assessment. Health Economics. 2003;12(8):609-628.
• Stålhammar NO. An empirical note on willingness to pay and starting-point bias. Medical Decision Making. 1996;16(3):242-247.