Willingness-to-Pay Elicitations

Updated 4 July 2026

Willingness-to-pay elicitation is a measurement strategy that recovers the monetary value agents attribute to goods, public policies, and risk reductions using methods like direct questions, auctions, and choice experiments.
It employs various methodologies including reservation price assessment, contingent valuation, and utility-based inference to derive precise valuation metrics and support demand curve estimation.
Applications span new product demand, public goods, environmental valuation, and pricing strategies, while addressing challenges such as biases, framing effects, and information distortions.

Willingness-to-pay elicitations are procedures for recovering the monetary value that an agent assigns to a good, attribute, risk reduction, or policy change. In the papers considered here, that object appears in several formally distinct but related forms: as a reservation price at which a consumer is indifferent between buying and not buying, as compensating or equivalent variation for changes in public goods, as a ratio of utility coefficients in discrete-choice models, and as a posterior distribution over dollar-valued feature upgrades in hierarchical conjoint analysis (Sun, 14 Jun 2026, Hofstetter et al., 2020, Karney et al., 2024, Helveston, 2022, Pillai et al., 14 Sep 2025). This suggests that “WTP elicitation” is not a single instrument but a family of measurement strategies spanning direct questioning, contingent valuation, auction-like mechanisms, repeated choice experiments, and model-based recovery of implied monetary tradeoffs.

1. Conceptual foundations

At the most elementary level, WTP is a reservation value. In the new-product demand framework of "Estimating Demand for a New Product" (Sun, 14 Jun 2026), each consumer has a reservation value $W$ , and purchase occurs if and only if price is below WTP. Hofstetter, Miller, Krohmer, and Zhang define actual WTP in closely aligned terms as the maximum price at or below which a consumer will definitely buy one unit (Hofstetter et al., 2020). The same paper on new-product demand also states that if utility is $U(\chi,q)$ , where $\chi\in\{0,1\}$ indicates purchase and $q$ is a composite of all other goods, then there exists a cutoff price $\overline p$ such that the consumer buys for $p<\overline p$ and does not buy for $p>\overline p$ ; that cutoff is the consumer’s WTP (Sun, 14 Jun 2026).

In welfare-theoretic settings, WTP is often identified with compensating variation rather than with a market reservation price. "The WTP-WTA Gap for Public Goods" (Karney et al., 2024) derives exact closed-form solutions for compensating variation and equivalent variation under homothetic utility with an independently homogeneous underlying utility function. For a proportional public-good change $t>1$ , the paper gives

$CV(\mathbf{z}_1,t)=\left(1-t^{\phi}\right)m,\qquad EV(\mathbf{z}_1,t)=\left(t^{-\phi}-1\right)m,\qquad \phi\equiv -\theta/\eta.$

In that framework, WTP for a public-good increase is $CV$ , WTA for a public-good decrease is $U(\chi,q)$ 0, and the paper identifies an income effect, an absolute preference effect, and a relative preference effect as the three mechanisms governing the magnitudes of $U(\chi,q)$ 1 and $U(\chi,q)$ 2 (Karney et al., 2024).

For risk reduction, the relevant monetary object is the amount an agent would pay to reduce the probability of a loss. "Willingness to pay, surplus and Insurance policy under dual theory" (Saidi, 2022) defines WTP for reducing loss probability from $U(\chi,q)$ 3 to $U(\chi,q)$ 4 through

$U(\chi,q)$ 5

and under dual theory obtains

$U(\chi,q)$ 6

The same paper emphasizes that, for partial risk reduction, a risk-averse decision maker can have a willingness to pay smaller than that of a neutral one, while also showing that a strongly averse decision maker is willing to give more for a reduction of a high-probability portion rather than a low-probability one (Saidi, 2022). A plausible implication is that elicited WTP for safety improvements is inseparable from the decision model used to represent probability weighting.

2. Direct statement, contingent valuation, and auction-style formats

Direct elicitation formats ask respondents to state a monetary amount or select a payment band. The papers here include open-ended direct questions, Becker–DeGroot–Marschak-style elicitation, ordered payment categories, and sequential contingent valuation with multiple payment vehicles.

Format	Response object	Representative paper
Direct single question	Open-ended maximum WTP	(Hofstetter et al., 2020)
BDM-based elicitation	Stated reservation price under incentive alignment	(Clithero et al., 2019)
Ordered payment categories	Seven ordered payment bands	(Sahebi et al., 2024)
Sequential multi-vehicle CV	Cash contribution plus labor-days contribution	(Kassahun et al., 2020)

"A De-biased Direct Question Approach to Measuring Consumers' Willingness to Pay" (Hofstetter et al., 2020) formalizes two common direct-question formats. For the open-ended format, stated WTP is modeled as

$U(\chi,q)$ 7

where $U(\chi,q)$ 8 is a product-category-level inflator and $U(\chi,q)$ 9 is an individual-specific bias component. For dichotomous choice, the paper models anchoring as

$\chi\in\{0,1\}$ 0

Combining the two yields the correction rule

$\chi\in\{0,1\}$ 1

The paper then implements BASIC, EPSILON, and FULL de-biasing procedures and validates them against BDM-based actual WTP in two product studies (Hofstetter et al., 2020).

"Supervised Machine Learning for Eliciting Individual Demand" (Clithero et al., 2019) re-examines BDM itself as an elicitation benchmark. In that experiment, subjects stated WTP for 20 food items using a BDM task, and purchase behavior was later observed in posted-price buy decisions. The paper reports that when the posted price equals stated WTP, the observed purchase probability is 62%, which it interprets as evidence that raw BDM WTP is noisy and systematically biased by understating valuations. It then shows that supervised machine learning applied to elicited WTP or even to simple two-alternative forced-choice data substantially improves out-of-sample purchase prediction, and that prices set by SML would increase revenue by 28% over using the stated WTP, with the same data (Clithero et al., 2019).

Direct elicitation need not be point-valued. "Assessing Public Perception of Car Automation in Iran" (Sahebi et al., 2024) elicits WTP for adaptive cruise control through seven ordered payment categories, from “not willing to pay any amount” to “above 800 million IRR,” and estimates an ordered logit model for the payment amount. "Revisiting money and labor for valuing environmental goods and services in developing countries" (Kassahun et al., 2020) instead uses a sequential contingent valuation design in which respondents first face a dichotomous-choice cash contribution question and then a labor-days contribution question. In that irrigation application, cash contribution comprises only 24.41% of total average annual WTP, implying that cash-only valuation would materially understate welfare in that setting (Kassahun et al., 2020).

3. Choice-based inference and utility-space parameterizations

A large part of the modern literature elicits WTP indirectly from repeated choices over bundles that include both price and nonprice attributes. In these designs, WTP is recovered from the tradeoff between the utility coefficient on an attribute and the utility coefficient on money.

"Willingness to Pay for an Electricity Connection" (Janghorban et al., 2024) is a canonical discrete choice experiment. Respondents in rural and peri-urban Nigeria choose between two hypothetical electricity service bundles differing in guaranteed daytime hours, guaranteed nighttime hours, upfront connection fee, capacity, source, and carrier. The analysis uses a conditional logit model under random utility theory,

$\chi\in\{0,1\}$ 2

and computes marginal willingness to pay as

$\chi\in\{0,1\}$ 3

The paper reports, for example, that households value each extra nighttime hour at ₦2,025 and each extra daytime hour at ₦1,709, while SMEs place a higher value on daytime electricity than nighttime electricity (Janghorban et al., 2024).

"logitr: Fast Estimation of Multinomial and Mixed Logit Models with Preference Space and Willingness to Pay Space Utility Parameterizations" (Helveston, 2022) sharpens the econometric distinction between two ways of extracting WTP from discrete-choice models. In preference space,

$\chi\in\{0,1\}$ 4

In WTP space,

$\chi\in\{0,1\}$ 5

so the nonprice coefficients are marginal WTPs directly. The paper argues that post-estimation WTP in mixed logit can become a ratio of random variables with unreasonable distributions of WTP across the population, whereas WTP-space models place distributional assumptions directly on $\chi\in\{0,1\}$ 6. It also stresses that WTP-space and mixed logit likelihoods are non-convex and that multi-start optimization is therefore methodologically important (Helveston, 2022).

The same coefficient-ratio logic can be enriched with latent attitudes and nonlinear reference dependence. "Willingness to Pay and Attitudinal Preferences of Indian Consumers for Electric Vehicles" (Bansal et al., 2021) estimates an integrated choice and latent variable model in which EV attributes are evaluated relative to an ICEV reference point. WTP is defined as the ratio of the marginal utility of an attribute to the marginal utility of purchase price, but the derivative is state-dependent because utility depends on $\chi\in\{0,1\}$ 7 rather than on a linear difference. The paper argues that accounting for reference dependence provides more realistic WTP estimates than the standard utility estimation approach and reports, among other results, additional willingness to pay of USD 10–34 to reduce fast charging time by 1 minute and USD 7–40 to add a kilometre to the driving range of EVs at 200 kilometres (Bansal et al., 2021).

A Bayesian extension appears in "What is in a Price? Estimating Willingness-to-Pay with Bayesian Hierarchical Models" (Pillai et al., 14 Sep 2025). There, a choice-based conjoint survey is analyzed with a Bayesian Hierarchical Logit model in which respondent-level coefficients are drawn from population distributions. WTP is computed from posterior coefficient draws using

$\chi\in\{0,1\}$ 8

with an additional rescaling adjustment because predictors were standardized. The paper’s methodological emphasis is that the output is not only a point estimate but a full posterior probability distribution for the dollar value of each feature (Pillai et al., 14 Sep 2025).

4. From elicited WTP to demand and pricing

One important line of work treats elicited WTP not merely as a welfare statistic but as the primitive object from which market demand can be constructed. In "Estimating Demand for a New Product" (Sun, 14 Jun 2026), the demand curve for a new product is derived directly from the conditional distribution of reservation values. Under the assumptions that each consumer buys at most one unit, that consumers have WTP $\chi\in\{0,1\}$ 9 distributed according to $q$ 0, and that purchase occurs if and only if price is below WTP, the paper writes conditional demand as

$q$ 1

for continuous $q$ 2, with an analogous discrete- $q$ 3 expression. The paper then proposes a two-step plug-in estimator: collect individual WTP and covariates, estimate the conditional CDF of WTP and the distribution of $q$ 4, construct pseudo-demand observations $q$ 5, and fit a parametric or local-polynomial demand specification by OLS or GMM. Under convergence of the empirical CDF and density estimates, the second-stage estimator is consistent (Sun, 14 Jun 2026).

The same paper shows that familiar empirical demand forms arise as special cases of specific WTP distributions. If conditional WTP is uniform, demand is linear; if WTP is exponential, demand becomes semi-log linear; if WTP is Pareto, demand becomes log-log. When the WTP distribution is unknown, the paper proposes local Taylor approximations such as

$q$ 6

or a second-order expansion with powers and interactions (Sun, 14 Jun 2026). This suggests a general workflow in which elicited valuation data become the first-stage input for a demand system rather than an endpoint.

Pricing applications use the same monetary tradeoff logic at a more granular level. In the Bayesian hierarchical conjoint framework, feature-level posterior WTPs support deconstruction of a premium product’s price into constituent parts, while the machine-learning paper on individual demand shows how elicited valuations can be mapped into stochastic demand curves and revenue-maximizing personalized prices (Pillai et al., 14 Sep 2025, Clithero et al., 2019). In both cases, WTP elicitation becomes part of a pricing pipeline rather than an isolated survey exercise.

A boundary case is "Pay-Per-Crawl Pricing for AI: The LM-Tree Agent" (Archer et al., 1 Apr 2026). That paper is explicit that it is not a willingness-to-pay elicitation paper in the classical sense. Instead, it calibrates latent article-level valuations from observed AI crawler traffic using

$q$ 7

draws query-level valuations around that center, and lets a pricing agent learn only from binary purchase outcomes. The result is closer to behavior-based revealed-preference-style approximation and pricing under latent valuations than to direct elicitation proper (Archer et al., 1 Apr 2026).

5. Bias, framing, beliefs, and the relation between stated and revealed values

The literature repeatedly shows that elicited WTP is sensitive to reporting error, framing, information, and institutional context. In the demand-from-WTP paper, reported WTP is modeled as

$q$ 8

with the reported-WTP CDF differing from the true-WTP CDF by a distortion term $q$ 9. The paper then shows two robustness results: if $\overline p$ 0 is uniformly distributed, misreporting shifts demand by a constant while preserving the slope with respect to price; and if reporting errors are small, symmetric, independent of true WTP, and the WTP density is locally flat, the distortion is small. This is the basis for its practical recommendation that local estimation via Taylor approximation may be more robust when misreporting is a concern (Sun, 14 Jun 2026).

Framing effects can be much larger than small-noise perturbations. "Eliciting the Endowment Effect under Assigned Ownership" (Zhong, 2018) randomly varies whether airline-seat attributes are framed in WTP or WTA terms and finds median WTA/WTP ratios of 15:1 for recline and 20:1 for legroom. The paper interprets these gaps as evidence that assigned ownership alone can generate endowment-effect-like responses, implying that WTP and WTA are not interchangeable and that entitlement cues materially change elicited values (Zhong, 2018).

One response to such concerns is to compare stated and revealed measures within a common structural framework. "Reconciling revealed and stated measures for willingness to pay in recreation by building a probability model" (Amiran et al., 2021) uses an exponential travel-cost model to estimate the same parameter $\overline p$ 1 from actual visitation behavior and from stated willingness to continue visiting under a higher travel cost. In the Rio Vista Falls application, the resulting aggregate stated and revealed WTP measures are 17.41 million dollars and 17.68 million dollars respectively, a ratio of 98.5%. The paper treats that close agreement as evidence that, under some conditions, stated and revealed WTP can converge (Amiran et al., 2021).

Other literatures document systematic heterogeneity in stated WTP. "Valuing insurance against small probability risks: A meta-analysis" (Mankaï et al., 2024) standardizes insurance WTP as relative willingness to pay,

$\overline p$ 2

and reports a weighted mean $\overline p$ 3 of 0.875 in the trimmed sample, summarized as average stated WTP equal to 87% of expected losses. The meta-regression finds that information about loss probability and probability levels positively influence relative willingness to pay, whereas respondents’ average income and age have a negative effect; methodological factors related to sampling and data collection also significantly influence stated WTP (Mankaï et al., 2024).

The most direct paper on information-friction in labor-market tradeoffs is "Pay Beliefs and the Amenity-Pay Tradeoff" (Andresen et al., 1 Jun 2026). It distinguishes clean full-information stated preferences from no-pay job-ad choices filtered through wage beliefs. Baseline beliefs under-predict starting salaries by 18% and imply a much steeper positive association between perceived pay and advertised amenities than is present in actual pay data. Exposure to pay information raises mean pay beliefs for similar jobs by 4% and reduces belief dispersion by 15%, but leaves the strong positive association between perceived pay and advertised amenities essentially unchanged. The paper’s central methodological point is that stated amenity-pay tradeoffs in no-pay settings are belief-conditional and therefore should not be read as pure preferences (Andresen et al., 1 Jun 2026).

6. Applications, scope conditions, and frontier domains

The applications in this corpus span new products, nonmarket valuation, transport, electricity access, workplace amenities, insurance, and AI systems. The new-product paper presents its method as applicable within and outside academia, including teaching economics, product launch decisions, and non-market valuation (Sun, 14 Jun 2026). The Nigerian electricity study uses elicited mWTP to compare households and SMEs on timing of electricity supply, capacity, source, and connection fees (Janghorban et al., 2024). The ACC study in Iran embeds WTP inside a Technology Acceptance Model and uses ordered payment bands to study monetized adoption propensity under affordability constraints (Sahebi et al., 2024).

Meta-analytic work shows that elicited WTP can also be used as an input to benefit transfer and long-run valuation. "Global evidence on the income elasticity of willingness to pay, relative price changes and public natural capital values" (Drupp et al., 2023) estimates

$\overline p$ 4

with an income elasticity of marginal WTP of around 0.6 across 735 income-WTP pairs from 396 contingent valuation studies. Combined with good-specific growth rates, the paper estimates relative price change of ecosystem services of around 1.7 percent per year and argues that natural capital values should be uplifted accordingly (Drupp et al., 2023). This suggests that elicited WTP is not only a cross-sectional welfare object but also an input into intertemporal price adjustment.

Frontier domains extend WTP inference to machine agents and subjective decision support. "Would a LLM Pay Extra for a View?" (Reusens et al., 10 Feb 2026) infers LLM WTP from repeated hotel-room choices using multinomial logit models and compares the implied valuations with human benchmark values from the hospitality-economics literature. The paper reports that meaningful WTP values can be derived for larger models, but that prompt wording, option order, persona framing, and prior examples materially change the inferred valuations; business-oriented personas and expensive examples push WTP upward, while cheap-choice conditioning can bring valuations closer to human benchmarks but can also induce negative WTP for some attributes (Reusens et al., 10 Feb 2026). A plausible implication is that elicitation logic based on choice tradeoffs generalizes beyond human respondents, but validity then depends on the decision process being modeled.

Scope conditions remain substantial. The new-product demand approach is static, assumes at most one unit per consumer in the baseline model, and does not provide a specialized estimator for interval-censored or binary-choice contingent valuation data (Sun, 14 Jun 2026). WTP-space mixed logit improves interpretability of WTP heterogeneity but introduces non-convex optimization, sensitivity to starting values, and difficulties when the scale parameter is random (Helveston, 2022). Proxy-based pricing systems, such as pay-per-crawl calibrated from traffic rather than from direct valuation reports, sit at the edge of the elicitation literature rather than at its center (Archer et al., 1 Apr 2026). Overall, the papers considered here imply that willingness-to-pay elicitation is best understood as a technically heterogeneous measurement problem in which the central challenge is not only to obtain a number, but to clarify which monetary object that number is intended to represent.