Benefit Transfer

How benefit transfer reuses an existing value estimate

Benefit transfer applies a monetary or preference-based value estimated in one study context to a different policy context. It can provide timely evidence when a new primary valuation study is infeasible, but its validity depends on similarity in the valued change, affected population, decision setting, and valuation method. This page explains the main transfer approaches, the adjustments they require, and how uncertainty and transfer error should be assessed.

The study site and policy site have different roles

The study site is the context in which the original value estimate was generated. The policy site is the new context in which the estimate will be used. Transferability asks whether the relationship measured at the study site remains meaningful after differences between the two sites are considered.

Component	Study site	Policy site
Population	People whose preferences generated the source estimate	People affected by the new policy
Good or outcome	The change described and valued in the original study	The change expected from the policy
Baseline	Initial risk, access, quality, or environmental condition	Starting condition in the target setting
Valuation method	Stated or revealed preference design and model	Methodological basis being transferred
Currency and date	Original price level and monetary unit	Required decision-year values
Decision use	Original research or policy purpose	Appraisal, cost-benefit analysis, or model input

Benefit transfer is not evidence transfer in general

The term usually refers to transferring economic values, such as willingness to pay, willingness to accept, or the monetary value of a health or service change. It is different from transferring a clinical treatment effect, a utility score, or a cost estimate, although those activities face related generalisability problems. The transferred quantity and its welfare interpretation should be stated precisely.

Using an estimate because it is available is not sufficient. The source must value the same underlying good or a defensibly comparable change.

Unit-value transfer uses a summary estimate

Unit-value transfer applies a mean, median, or other summary value from one study to the policy population. It is simple and transparent but assumes that differences between settings do not materially change value, or that a limited adjustment can address them. It is most defensible when the source and target contexts are closely matched.

If the adjusted value per affected person is (V_P) and (N_P) people experience the defined change, aggregate benefit is:

$$ B_P=N_P\times V_P $$

Aggregation requires evidence that (N_P) represents the population holding the valued change and that the value is not already expressed per household, episode, year, or other unit.

Adjusted unit-value transfer changes selected factors

An adjusted transfer modifies the source value for price level, currency, income, or another defensible difference. Adjustment should follow an economic rationale and preserve the original unit. Adding many ad hoc multipliers does not make a weak source transferable.

An income adjustment can be represented as:

$$ V_P=V_S\left(\frac{Y_P}{Y_S}\right)^{\eta} $$

where (V_S) is the study-site value, (Y_P/Y_S) is the target-to-source income ratio, and (\eta) is the income elasticity of the value. The elasticity should be evidence-based and tested because results can be sensitive to it.

Value-function transfer uses estimated relationships

Function transfer applies an estimated valuation equation from the source study to policy-site characteristics. It can adjust simultaneously for income, demographics, baseline risk, scope, and other predictors. It requires compatible variable definitions and confidence that the estimated relationship transports beyond the original sample.

A simplified function is:

$$ V_i=\beta_0+\boldsymbol{\beta}^{\top}\mathbf{X}_i+\varepsilon_i $$

Policy-site values are predicted by inserting target characteristics into (\mathbf{X}_i). Extrapolation outside the source data range can be unreliable even when the equation produces a number.

Meta-analytic transfer pools several valuation studies

Meta-analytic benefit transfer estimates a function across multiple source studies, allowing value to vary with study, population, good, and methodological characteristics. It can broaden the evidence base and quantify heterogeneity. It also inherits publication bias, inconsistent definitions, and correlation among estimates from the same study.

The meta-analysis should distinguish sampling error, between-study heterogeneity, and methodological differences. Study-level associations are not automatically valid for individual-level prediction, and a good statistical fit does not guarantee policy-site similarity.

The valued change must match in direction and magnitude

Values depend on what changes, by how much, for how long, and from which baseline. A willingness to pay for avoiding one event cannot be transferred automatically to avoiding several events or to reducing severity. Marginal values can change with the size and sequence of the improvement.

The transfer should compare:

• Outcome definition and severity.
• Baseline probability or condition.
• Absolute and relative change.
• Duration, delay, and certainty.
• Reversibility and who receives the benefit.
• Whether the value represents prevention, treatment, access, reassurance, or information.

Scope and scale affect value

Scope refers to the quantity or extent of the good being valued. A valid valuation should often increase with a larger beneficial change, but not necessarily in direct proportion. Multiplying a value for a small change by the number of units can overstate benefits when marginal value declines.

Scope tests in the source study and scenario analyses at the policy site can reveal nonlinearity. The aggregation rule should not assume constant marginal value unless evidence supports it.

Baseline risk changes absolute value

Health-risk valuations depend on the size of the risk reduction and sometimes on the starting risk, dread, voluntariness, and affected population. A relative risk reduction does not define a welfare change without baseline risk. Transfers should use absolute changes over a stated time horizon.

For baseline risk (p_0) and policy risk (p_1):

$$ \Delta p=p_0-p_1 $$

If a source estimate is based on a different (\Delta p), proportional scaling should be justified rather than assumed.

Currency conversion and inflation are separate adjustments

Values should be moved to a common price year using an appropriate domestic price index and then converted to the target currency using a stated exchange-rate method. Market exchange rates reflect financial conversion, while purchasing-power parity can better reflect differences in domestic purchasing power for some applications. The choice should match the welfare and decision context.

A general conversion is:

$$ V_{P,t}=V_{S,s}\times\frac{PriceIndex_{S,t}}{PriceIndex_{S,s}}\times FX_{S\rightarrow P,t} $$

Income adjustment, if required, is conceptually separate from inflation and currency conversion. Applying all adjustments without checking overlap can double count economic differences.

Population preferences may not transport

Values can differ with income, age, health, experience, culture, trust, insurance, payment responsibility, and social context. Sample representativeness at the source does not guarantee representativeness at the policy site. The transfer should identify effect modifiers rather than relying on broad country similarity.

Equity concerns arise because willingness to pay is constrained by ability to pay. Transferred values can reproduce income inequalities and should not be interpreted automatically as equal social importance.

Payment vehicle and elicitation method matter

A tax, insurance premium, copayment, charitable contribution, or direct payment can produce different responses because credibility, obligation, and budget constraints differ. Open-ended questions, dichotomous choice, payment cards, discrete choice experiments, and revealed-preference methods also estimate different quantities under different assumptions.

Methodological differences should not be corrected with a simple multiplier unless supported by evidence. A source using an implausible or unacceptable payment vehicle may be unsuitable regardless of statistical quality.

Willingness to pay and willingness to accept are not interchangeable

Willingness to pay measures the maximum payment for a gain or to avoid a loss, while willingness to accept measures the minimum compensation required to give up an entitlement or accept harm. They can differ because of income constraints, reference points, substitution, and loss aversion. The correct measure depends on the implied property rights and policy question.

Transferring one as though it were the other can change welfare interpretation substantially. The report should preserve the original elicitation and reference point.

Use and non-use values should remain distinct

Some goods produce value through direct use, while others generate option, altruistic, existence, or bequest value without current use. Health applications can include reassurance, preparedness, or value placed on access for others. Aggregating across populations can double count overlapping motives.

The source study should define whose values are elicited and what respondents are asked to consider. Policy aggregation should use a population consistent with that framing.

Source-study validity comes before transferability

A perfectly matched study is not useful if its estimate is biased or uninterpretable. Source review should examine sampling, scenario credibility, comprehension, consequentiality, protest responses, payment design, modelling, missingness, and uncertainty. Transfer adjustment cannot repair fundamental valuation defects.

Useful source criteria include:

• A clearly defined and policy-relevant valued change.
• A population and sampling frame appropriate to the stated value.
• Valid and transparent elicitation and analysis.
• Sufficient reporting to reconstruct units and uncertainty.
• A price year, currency, payment frequency, and time horizon.
• Evidence on scope, protest responses, and sensitivity where relevant.

Transferability requires structured comparison

The study and policy sites should be compared before selecting an estimate. A transferability table can score or describe differences, but the final judgment should not be a mechanical total. Some mismatches, such as a different outcome or payment unit, can be fatal even when other characteristics align.

The review should compare:

• Good, outcome, baseline, change, duration, and certainty.
• Population characteristics, income, experience, and preferences.
• Institutions, coverage, prices, and available substitutes.
• Elicitation method, payment vehicle, and survey mode.
• Geographic, cultural, temporal, and policy setting.
• Intended welfare concept and aggregation unit.

Transfer error should be measured when possible

Transfer error compares a transferred prediction with a value estimated directly at the policy site. It can be expressed as an absolute or percentage difference, though percentage error is unstable when the target value is near zero. External validation across held-out sites provides stronger evidence than fit to source studies.

One percentage measure is:

$$ Transfer\ error=\frac{|\widehat{V}_P-V_P|}{|V_P|}\times100% $$

where (\widehat{V}_P) is the transferred value and (V_P) is a credible local estimate. Validation results should match the transfer method and policy contexts of interest.

Uncertainty includes more than the source standard error

A transferred estimate contains source sampling uncertainty, parameter uncertainty, between-study heterogeneity, adjustment uncertainty, policy-site input uncertainty, and structural uncertainty about transferability. Reporting only the confidence interval from the source study substantially understates uncertainty.

Probabilistic analysis can sample:

• Source value or function coefficients.
• Income elasticity.
• Price indices and exchange assumptions where material.
• Policy-site population and outcome change.
• Heterogeneity or transfer-error terms.
• Alternative source-selection and functional-form scenarios.

Combining several transferred values needs rules

When several eligible sources exist, analysts can select the closest study, pool estimates, use a meta-regression, or present a range. Choosing the estimate that supports the preferred conclusion is not defensible. Selection and synthesis rules should be pre-specified.

Weights based only on statistical precision can favour large but poorly matched studies. Relevance, quality, and independence should inform the approach without being hidden inside an arbitrary score.

Aggregation can create double counting

Values may overlap when several outcomes, services, household members, or motives are summed. A respondent's value for a complete programme can already include several component benefits. Adding separate component values can exceed the welfare change actually elicited.

Aggregation should identify:

• The unit of value: person, household, case, episode, year, or programme.
• Whether values are marginal or total.
• Whether health, convenience, access, and reassurance overlap.
• Whether the same population appears in several components.
• Whether values already include future or spillover effects.

Equity and distribution should not disappear in aggregation

An average transferred value can conceal differences in who benefits, who pays, and whose preferences were measured. Income adjustment can improve prediction while embedding ability-to-pay differences in policy weights. Decision makers should separate positive prediction of willingness to pay from normative judgments about social value.

Distributional reporting can show benefits, payments, and transferred values by income, health need, geography, or other relevant group. Alternative equity weights or non-monetary outcomes may be needed when the policy objective is not fully represented by willingness to pay.

Benefit transfer can inform cost-benefit analysis

Cost-benefit analysis compares monetised benefits with costs. A transferred value can make analysis possible when primary valuation is unavailable, but the benefit-cost ratio can appear precise despite weak transferability. Results should show transferred and non-transferred components separately.

Net benefit is:

$$ NB=B-C $$

and the benefit-cost ratio is:

$$ BCR=\frac{B}{C} $$

Both depend on compatible price years, discounting, population, time horizon, and avoidance of double counting. A ratio above one does not resolve distribution, affordability, or evidence-quality concerns.

A new primary study may be worth the cost

Transfer is attractive because it is faster and cheaper, but a high-stakes, irreversible, or highly context-specific decision may justify new valuation. The choice depends on the likely transfer error, sensitivity of the decision, time available, and cost of research. Pilot local validation can sometimes be more valuable than a full new study.

Primary research is more strongly indicated when:

• No source values the same outcome or policy change.
• Context differences are likely to modify value materially.
• The decision changes under plausible transferred values.
• The affected population or expenditure is large.
• Equity or cultural acceptability is central.
• Existing source studies have serious validity limitations.

Worked adjusted transfer

Suppose a source study estimates annual willingness to pay of $120 at average income $40,000. The policy-site average income is $50,000 and the chosen evidence-based income elasticity is 0.6. Ignoring currency and price-year changes for this illustration, the adjusted value is:

$$ V_P=120\left(\frac{50{,}000}{40{,}000}\right)^{0.6}\approx137.2 $$

If 100,000 people receive the same defined annual benefit, the mechanically aggregated value is about $13.72 million. The result remains conditional on population representativeness, scope, income elasticity, payment unit, and whether aggregation across all 100,000 people is valid.

Common mistakes

Benefit transfer can create a plausible number even when the source and policy questions are materially different. The following errors commonly overstate validity. Each should be checked before the estimate enters an appraisal.

• Selecting a source because its value is convenient rather than transferable.
• Adjusting currency and inflation while ignoring differences in the valued good.
• Treating relative and absolute risk changes as equivalent.
• Assuming values scale linearly with scope, duration, or population.
• Mixing willingness to pay and willingness to accept.
• Applying household values to individuals or annual values to one-time benefits.
• Double counting components already included in a total programme value.
• Reporting only source-study sampling uncertainty.
• Treating willingness to pay as an equity-neutral measure of social importance.
• Using a statistically complex function outside the range of its source data.

Reporting a benefit transfer

Transparent reporting should let readers reconstruct the source value, every adjustment, and the final aggregation. It should explain why the source is credible and comparable rather than merely cite it. Uncertainty and alternative sources should remain visible.

• Define the policy good, baseline, change, population, duration, and decision use.
• Identify every source study, estimate, unit, currency, price year, and method.
• Compare study and policy sites across outcomes, populations, institutions, and methods.
• Report inflation, currency, income, scope, and function adjustments with equations and sources.
• Report source, parameter, heterogeneity, transfer, and aggregation uncertainty.
• Show alternative eligible sources and explain selection or pooling.
• State the aggregation population and safeguards against double counting.
• Report equity implications, validation evidence, limitations, and update date.

The decision standard

A benefit transfer is credible when the source estimate represents the same underlying welfare change, the study and policy contexts are sufficiently comparable, and every adjustment has a defensible economic basis. The transferred number should remain visibly more uncertain than a directly observed local value when context differences are material. Transfer is a practical evidence strategy, not permission to replace an unavailable value with the nearest published number.