Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Insurance Demand Experiments and Elicitation Methods

Abstract. Insurance demand experiments measure how individuals and organizations evaluate contingent transfers, prices, deductibles, exclusions, payment reliability, and the timing of uncertain losses. Their results depend on whether the task elicits a hypothetical statement, an incentivized lottery, a binding insurance purchase, or an administrative choice among real contracts. This monograph examines the epistemic and econometric consequences of those distinctions. It reconstructs the development of laboratory and field paradigms, formalizes the difference between preference elicitation and identified insurance demand, evaluates premium and probability manipulations, and compares multiple-price lists, certainty equivalents, contingent valuation, discrete choice experiments, and revealed contract selections. It synthesizes evidence from agricultural index insurance, automobile and homeowners deductible decisions, and health-plan choice, emphasizing the roles of basis risk, liquidity, product comprehension, trust, context, and selection. A worked two-state insurance model and a transparent synthetic price experiment identify the assumptions needed to interpret valuation and elasticity. The central conclusion is methodological: a demand estimate becomes scientifically interpretable through the joint specification of the contract, the choice environment, the incentive mechanism, and the population to which an inference applies. No original dataset, new experimental treatment, meta-analytic pooled effect, or novel named theory is claimed.

Keywords: insurance demand; stated preferences; revealed preferences; experimental economics; hypothetical bias; incentive compatibility; willingness to pay; multiple price lists; risk preferences; discrete choice experiments; behavioral insurance; basis risk; field experiments; causal identification; external validity; replication.

Introduction

Insurance demand concerns a decision made before the realization of loss: a purchaser exchanges a known or contracted premium for a state-contingent claim whose value depends on uncertainty, contractual terms, counterparties, and institutional enforcement. Insurance researchers can observe the purchase directly, ask about an imagined purchase, or infer latent preferences from choices among lotteries. These operations overlap but do not have identical estimands. The central research problem is therefore one of measurement and identification. An observable answer is produced by a task; an underlying demand relation is an interpretation of the answer conditional on the task's information, stakes, contract and context.

Jaspersen's (2016) review of insurance demand experiments provides a methodological point of departure. Its purpose is to map research designs and their respective advantages and disadvantages. That orientation matters because experimental rigor is conditional on what the experiment is designed to learn. A laboratory study can isolate how an individual responds to a premium in a stylized two-state lottery. A survey can reach a more diverse population with products not yet available. A field intervention can identify the causal effect of a particular change in a genuine purchase environment. Administrative records can reveal actual choices at scale while confronting the econometric confounding of offered menus, private information and unobserved frictions.

The principal disciplinary location is Behavioral Insurance and Risk Perception. The research also connects to Insurance Economics, Insurance Statistics and Predictive Modeling, Health Insurance, and Insurance Pricing and Reserving. These are scientific cross-references to previously published monographs; this dossier has one canonical identity. The analysis distinguishes primary experimental articles, methodological articles, journal metadata and abstracts, and the author's own mathematical and comparative synthesis. Results originally established for a particular market are interpreted within that market. International applicability is a question of transportability and institutional comparability, examined below.

Questions, terms and methodological scope

Four questions guide the investigation. First, which design features determine the interpretation of measured insurance demand? Second, when does an elicited valuation correspond to an incentive-compatible preference report or a real market transaction? Third, which experimental and observational comparisons allow price, beliefs, financial constraints, comprehension and framing to be distinguished? Fourth, which population and product features determine how far findings may be transported? The answers arise from source criticism, formal economic analysis, design comparison and careful interpretation of the reported empirical studies. No new survey was administered and no participant-level dataset was reanalyzed.

A preference is an ordering over prospects or contingent consumption allocations under a stated decision model. Willingness to pay is a maximum payment for a specified change from a defined baseline, not a universal personal trait. Take-up is an observed binary event within an available choice set, while demand is the relation between desired coverage and terms including price. Insurance literacy concerns comprehension of premiums, payment timing, limits, deductibles and claim contingencies. Trust concerns expectations that the contractual promise will be honored. Basis risk measures the discrepancy between an index trigger and the buyer's experienced loss. These distinctions are necessary to avoid treating a premium response, a survey intention and a stable preference parameter as interchangeable quantities.

Insurance Demand Experimental Paradigms

The historical transition from laboratory preferences to insurance choices

Early experimental economics developed incentivized methods for observing risky choices under controlled stakes. In Binswanger's (1980) rural India experimental measurements, interview-based certainty equivalents diverged sharply from real-payoff gambling choices. This study is important to insurance measurement because its research object was risk attitude rather than a complete insurance contract: it demonstrated that elicitation mode and stakes can change measured preference, while leaving a separate question about actual premium-funded coverage. The experiment involved a specific rural sample; it establishes the methodological issue without validating any universal correction factor for hypothetical bias.

Holt and Laury's (2002) risk-aversion and incentive study helped establish multiple-price-list lotteries as a practical instrument for eliciting risky choice. Choices between paired lotteries can be mapped to parameter intervals under expected utility and assumptions about stable preferences and understood probabilities. The resulting parameter is instrument- and model-dependent. A person who declines an even-odds gamble need not pay a correspondingly large premium for a complex insurer-issued contract. The latter includes settlement risk, service, policy language, prior experience and alternative financial protections.

A separate behavioral literature asked how an insurance description changes the meaning participants assign to outcomes. Johnson and colleagues (1993) experimentally demonstrated framing and probability distortions in hypothetical insurance choices. When people respond to frames rather than solely to actuarial probabilities and state-contingent wealth, the researcher must treat frame as a measured experimental component. A scientifically meaningful comparison specifies whether the purpose is to estimate a stable response to a product or to measure the causal effect of its presentation.

A typology by scientific estimand

The laboratory risk task measures decisions within an induced and often simplified outcome space. The hypothetical insurance survey measures stated demand within a constructed description. An incentivized insurance lottery measures a payoff-motivated decision under the laboratory's settlement mechanism. A field experiment measures the effect of assigned offers or information on behavior in a real institutional setting. Administrative revealed-choice research estimates how insured individuals respond to menus or contract variations already encountered, subject to available variation and selection. This ordering is a classification by estimand and not a ranking of scientific worth.

Harrison and List (2004) identify six field-context dimensions: subject pool, information brought to the task, commodity, task or trading rules, stakes, and environment. Applied to insurance, an experiment conducted in a rural marketplace with a real premium and delayed settlement may identify a credible purchase decision yet also embed trust and seasonal cash-flow effects. A laboratory task may remove these effects and identify a psychological mechanism with greater precision. One must specify the desired inference before selecting a setting.

Risk theories as tested hypotheses

Under expected utility, decreasing marginal utility creates a reason to transfer wealth from favorable to unfavorable states. The relevant choice is evaluated over final wealth and contractual cash flows. Prospect theory instead models reference-dependent valuation and probability decision weights, as developed in Kahneman and Tversky's (1979) analysis of risky decisions. Tversky and Kahneman's (1992) cumulative prospect theory offers a formal explanation for why low-probability events can be overweighted in particular decision configurations. These frameworks generate testable implications, but any observed premium choice can reflect several mechanisms simultaneously.

A credible insurance experiment manipulates observable consequences of competing explanations. If participants overpay for low deductibles, risk aversion, probability weighting, loss aversion, confusion and perceived insurer service can each be candidate explanations. Evidence for one mechanism is strongest when another account predicts a different response under a controlled change. Experimental design therefore begins with identification rather than a descriptive list of behavioral effects.

Temporal and institutional dimensions

Insurance pays after contingencies arise, whereas many laboratory lotteries resolve immediately. Delayed settlement introduces discounting, inflation, claims-paying uncertainty and transaction costs. Andersen and colleagues (2008) show why joint elicitation of risk and time matters when monetary outcomes arrive at different dates. A premium paid today for a possible loss next year is a temporally structured promise. A laboratory lottery settled before participants leave the room measures a narrower economic object unless future timing is specifically implemented.

Experiments must also preserve the relevant legal-economic institution: an indemnity contract responds to a proved loss within policy limits; index insurance pays on a contractually defined trigger; a deductible shifts a first layer of realized loss to the insured; coinsurance shares a fraction of covered expense. Each changes the payoff distribution and the information required for a rational choice. When instruments compress these differences into an unexplained binary question, an answer can be produced without demonstrating comprehension of the policy being valued.

Hypothetical versus Incentivized Choices

The hypothetical question as a measurement event

A hypothetical insurance question usually asks whether a respondent would buy coverage at a stated premium or how much they would pay. It can elicit responses for products outside existing markets, supply broad geographic coverage and test carefully controlled variations in description. It also introduces an additional decision: what to report when answers have no binding financial consequence. Respondents may express aspirations, social norms, perceived expectations of the researcher, or a belief that a particular answer may shape future offers. Such responses can still contain information, but their interpretation depends on the scenario and the perceived consequences of answering.

Carson and Groves (2007) examine incentives in preference questions. Their theoretical analysis explains why the consequentiality and strategic structure of a survey matter for truthful revelation. Researchers should therefore document whether an insurance offer is framed as merely imaginary, a policy consultation likely to affect real decisions, or a mechanism under which some choices become binding. Changing the wording to sound realistic is insufficient to establish incentive compatibility.

Incentivized response methods exchange one validity problem for a more controlled but narrower setting. Cash-reward lotteries can make participants attend to probabilities and payoff magnitudes, while insurance policies involve third-party settlement and complex state definitions. Even a binding purchase at a randomly drawn price may fail to reproduce the household's long-run financial situation when the experiment subsidizes the premium, guarantees claims independently of an insurer or restricts the covered peril to a short artificial horizon.

Hypothetical bias, strategic reporting and uncertainty

The central comparison is between a stated intention and a choice under real costs and benefits, holding the offered object as constant as practicable. A gap between the two is a descriptive discrepancy; calling it hypothetical bias presupposes that the real-stakes benchmark and the intended valuation target are comparable. In insurance, realism can change the product itself: an actual policy may have exclusions, waiting periods, document requirements and an insurer reputation absent from a hypothetical description. A researcher must separate inconsistencies of response from differences in the choice environment.

Binswanger's (1980) comparison of interviews and stakes cautions against equating an interview risk score with an observed risk preference under payment. Dohmen and colleagues (2011) validate a general risk survey question against an incentivized lottery in a representative German sample. Together these studies support a qualified conclusion: carefully designed questions can be useful measures, and neither a particular self-report item nor a paid lottery is a universal surrogate for willingness to buy insurance. Validation must be conducted at the level of the outcome and population for which an instrument will be used.

Incentives, transaction realism and payment architecture

Three features should be reported separately. Salience means participants notice and understand the contingent cost and benefit. Consequentiality means their choice changes some real outcome. Incentive compatibility means truthful revelation is an optimal strategy under specified behavioral and mechanism assumptions. A real monetary prize satisfies consequentiality; it does not automatically satisfy incentive compatibility for every elicitation method. This is the essential analytic lesson of comparing utility measurement with insurance purchasing.

The Becker-DeGroot-Marschak family of procedures descends from Becker, DeGroot, and Marschak's (1964) single-response utility measurement. In a stylized insurance willingness-to-pay implementation, a participant states a maximum premium for an exactly specified policy, a price is drawn independently from a known distribution, and purchase is executed when the stated valuation meets or exceeds that price. Under quasi-linear or more general assumptions supporting threshold choice, independent price draws and correct understanding, the participant has a reason to reveal a genuine reservation threshold. A welfare interpretation nevertheless requires consistency between the insured contingency, the participant's financial baseline and the real policy's performance.

Randomly paying one of several elicitation decisions reduces experimental cost, but it introduces assumptions about the separability of decisions and the way participants aggregate compound risks. Cubitt, Starmer, and Sugden (1998) test random-lottery incentive validity by comparing multi-task randomly paid designs with one-real-choice treatments, finding no statistically significant contamination in their reported experiments. That finding supports the examined mechanism under the reported conditions. It does not show that random-lottery selection is neutral for every task, wealth state, probability perception or insurance instrument.

An interpretable validation strategy

A rigorous two-arm or multi-arm test can assign the same precisely written policy to hypothetical, paid-lottery and binding-offer conditions while keeping premiums, base wealth information, and comprehension checks comparable. The estimands are the differences in take-up and price-response patterns between assigned mechanisms, rather than an unobserved universal correction factor. Some implementation features cannot be held constant: actual policy enforcement and claims adjudication exist only with binding coverage. Those differences belong in the design and report rather than being treated as noise.

The strongest research interpretation is triangulated: specify an expected-utility or behavioral benchmark, test it against the instrument response, and compare the instrument with a closer-to-market outcome. When a discrepancy remains, estimate its magnitude and uncertainty for the particular product and sample. Report heterogeneity by financial capacity and comprehension without inferring that a group possesses a fixed psychological trait from a single purchase response.

Insurance Lottery Design

Contract state space and final-wealth model

Let initial wealth be W, a binary insurable loss be L greater than zero, and the probability of loss be p in the open interval (0,1). A policy pays indemnity x, with 0 less than or equal to x less than or equal to L, when the loss occurs. Its prepaid premium is P(x). Terminal wealth is W-P(x) without loss and W-L+x-P(x) with loss. Expected utility for a strictly increasing concave Bernoulli utility function u is

EU(x)=(1−p)u(W−P(x))+p u(W−L+x−P(x)).EU(x)=(1-p)u(W-P(x))+p\,u(W-L+x-P(x)).

This representation displays the experimental primitives: baseline resources, physical or financial loss, believed loss probability, coverage, and premium schedule. Each must be stated to participants or deliberately estimated. A task that supplies only a verbal label such as 'insurance' leaves open participants' own assumptions about claim eligibility, payment reliability and incidental service.

For an actuarially fair premium P(x)=px, differentiable expected utility has marginal value

dEUdx=p(1−p){u′(W−L+(1−p)x)−u′(W−px)}.\frac{dEU}{dx}=p(1-p)\{u'(W-L+(1-p)x)-u'(W-px)\}.

For x less than L, the wealth in the loss state is lower; concavity implies larger marginal utility there. The derivative is nonnegative and strictly positive when u is strictly concave. Thus full coverage x=L maximizes expected utility in this frictionless two-state model. This result is conditional on fair pricing, an enforceable contract, known probabilities, and absence of separate transaction benefits or costs. It is a formal benchmark and not a prediction of universal full-insurance take-up.

With proportional loading lambda at least zero and P(x)=(1+lambda)px, the derivative becomes

dEUdx=−(1−p)(1+λ)p u′(W−(1+λ)px)+p[1−(1+λ)p]u′(W−L+x−(1+λ)px).\frac{dEU}{dx}=-(1-p)(1+\lambda)p\,u'(W-(1+\lambda)px)+p[1-(1+\lambda)p]u'(W-L+x-(1+\lambda)px).

Loading reduces the attraction of coverage and may yield an interior optimum or a zero-coverage corner. The exact solution depends on utility curvature, affordability and contract limits. A research design that varies premiums while keeping benefits and losses fixed can estimate a demand schedule; it cannot recover a unique structural risk-aversion coefficient without imposing and evaluating a model.

Certainty equivalent and reservation premium

For the all-or-nothing full-insurance offer under constant absolute risk aversion, u(w)=-exp(-rw) for r greater than zero, the maximum premium P* that equates insured certain wealth and uninsured expected utility is

P∗=1rlog⁡[(1−p)+pexp⁡(rL)].P^*=\frac{1}{r}\log[(1-p)+p\exp(rL)].

A useful verification is the limit as r approaches zero: by l'Hopital's rule it equals pL, the expected indemnity. For strictly positive r and nondegenerate risk, Jensen's inequality implies P* exceeds pL, subject to the assumed wealth domain and utility specification. This closed-form expression converts an experimentally measured reservation premium into an implied curvature only under the maintained CARA model and the specified loss distribution.

Consider a transparent illustrative scenario with p=0.10, L=$1,000, and r=0.0005 per dollar. The expected loss is $100. The expression yields a maximum premium of approximately $126; an offer of $120 is accepted by the model, while an offer of $140 is rejected. These figures are calculations from the stipulated example, not observations from a published sample. With a paid premium of $120 and reliable full indemnity, terminal wealth becomes W-$120 in either state; with no policy, it remains W or W-$1,000. The mechanics should be visible in an experimental display rather than concealed behind a label such as 'comprehensive coverage.'

What changes with nonpayment and index triggers

A realistic contract can fail to compensate a genuine loss because of exclusions, claims denial, insufficient documentation, counterparty nonperformance or an imperfect index. Let T be a trigger and Y be actual loss. The contract pays x when T=1, while the purchaser's loss depends on Y. The joint distribution of Y and T, and the buyer's beliefs about it, determine final-wealth risk. Premium and loss probability alone cannot identify index-insurance valuation. Clarke's (2016) analysis of index demand demonstrates theoretically that basis risk can make optimal index coverage nonmonotonic in wealth, price and risk aversion. A monotonic comparative-static claim derived from complete indemnity therefore cannot be transferred mechanically to a weather index contract.

Gine, Townsend, and Vickery (2008) document rainfall insurance participation patterns associated with basis risk, wealth, credit constraints and vendor familiarity. These findings show the value of measuring contract mismatch and institutional context rather than interpreting all low take-up as weak aversion to risk. A valid experiment should present joint loss-trigger states, including 'loss but no payout' and 'payout without loss,' with probabilities and amounts consistent with the actual contract.

Design dimensions that alter estimates

Probability varies through numeric frequency, natural-frequency examples, visual grids, verbal descriptors and experienced draws. The probability manipulation may simultaneously alter understanding or salience. Payoff framing varies between losses from a stated endowment and gains relative to a reference account. Lottery scale changes the curvature region being examined. Payment timing changes liquidity and discounting. The number of repeated decisions can create learning and fatigue. Insurance labels invoke beliefs about institutions. Each variation is a treatment dimension requiring explicit documentation.

Gneezy and Potters (1997) demonstrate evaluation-period effects on risk taking in experimental investment decisions. Their result is not itself an insurance purchase estimate; it illustrates why evaluation horizon is a substantive element of an elicitation instrument. For repeated-loss insurance, yearly premium framing versus monthly installment framing similarly changes the unit of comparison and perceived affordability. A well-designed study computes equivalent present-value amounts where possible and treats cash-flow framing as a separately randomized intervention.

Synthetic illustration: a randomized premium experiment

Suppose a study hypothetically randomizes two otherwise identical and fully explained offers. At premium $80, 60 of 100 participants purchase; at premium $140, 40 of 100 purchase. The estimated take-up difference is 0.40-0.60=-0.20, or a 20-percentage-point decrease. Using the symmetric midpoint formula, the arc elasticity is [(-0.20)/0.50] / [($140-$80)/$110] = -0.7333. These are deliberately invented numbers for exposition, not empirical evidence, and there is no confidence interval because they are not sampled observations.

This design identifies the causal effect of premium assignment on take-up in the experimental population when random assignment, exclusion of interference and consistent measurement hold. The elasticity measures the particular price interval and contract. If a higher price also signals better product quality, or if subsidies vary with financial status, a structural interpretation requires additional design and analysis. Changing prices independently of the insurer's perceived credibility is especially important in low-trust markets.

Preference Elicitation and Incentive Compatibility

Multiple-price lists and interval measurement

In a multiple-price list, participants choose repeatedly between safer and riskier lotteries while probabilities or payoffs change systematically. A switching point translates into an interval for a structural parameter only after the experimenter specifies a utility function, probability perception and error process. Multiple switching can indicate misunderstanding, noisy attention, state-dependent preference, reference effects or a menu that does not satisfy monotonicity assumptions. Researchers should retain raw choices and report the distribution of switching patterns before fitting one parameter per participant.

Charness, Gneezy, and Imas (2013) compare risk elicitation task families and explain that the preferred method depends on the research question and sample. Crosetto and Filippin (2016) show in their comparison of elicitation methods that task mechanics create part of the heterogeneity in estimated risk attitudes, with further differences remaining beyond simulated measurement error. The implication for insurance scholarship is concrete: an MPL-derived curvature parameter should not be substituted into a premium-demand model as though it were an invariant observed fact.

Certainty equivalents, reservation prices and choice sequences

A certainty-equivalent instrument asks for a guaranteed amount regarded as equivalent to a specified risky payoff. A reservation-price instrument asks for the maximum premium a participant would pay for a defined coverage improvement. The first describes a lottery-to-certain-money comparison; the second describes a contingent contract-to-no-contract comparison. Their economic equivalence requires matching reference wealth, state probabilities, insurance payoff, timing, taxes and mechanisms for binding implementation. An insurance experiment often has a multidimensional object even when respondents are asked to supply one number.

The BDM-style mechanism can align a truthful report with a purchase threshold under its assumptions, but random-price support must span plausible valuations and participants must understand the draw and implementation rules. Open-ended values may encourage rounding or anchoring; a restricted price grid censors values at its endpoints. Repeated binary choices identify intervals rather than exact valuations and can be affected by the order of prices. The design report should therefore distinguish an actual observed maximum from an interval inferred by assumption.

Discrete choice experiments and attribute trade-offs

A stated-choice experiment can vary premium, deductible, limit, claim delay, index basis risk, insurer reputation, exclusions and payment frequency in a factorial or efficient design. Respondents choose among policy profiles and a no-purchase option. A random-utility model represents utility of profile j for respondent i as a systematic component V_ij plus an unobserved term epsilon_ij. Under an extreme-value error assumption, multinomial logit assigns a choice probability proportional to exp(V_ij). Attribute coefficients yield marginal willingness-to-pay ratios only when the price coefficient and scale are properly identified and model restrictions are defensible.

A 'no insurance' alternative is essential if the scientific question concerns market take-up rather than relative preferences among forced insurance options. Policy labels and choice-set complexity can alter the error process, creating confounding between attribute taste and understanding. Repeated observations from one respondent require within-person dependence to be modeled. Estimation should report uncertainty, heterogeneity and predictive performance, and should test whether willingness-to-pay ratios are stable under plausible alternative specifications. Stated-choice coefficients do not automatically forecast real-market sales.

Revealed choice and structural inference

Administrative contract data record premiums and deductibles chosen with real money, often across large populations. Cohen and Einav (2007) estimate preferences from auto deductibles using a structural framework with unobserved heterogeneity in both risk and risk aversion. Barseghyan and colleagues (2013) show the importance of probability distortions in deductible choice. Both studies illustrate how a revealed insurance decision can inform preference estimation while requiring a model linking claim probabilities, contract terms and unobserved heterogeneity.

Observed coverage can also reflect misunderstanding. Sydnor (2010) studies home insurance deductible overpayment and finds patterns difficult to reconcile with conventional modest-stakes curvature. Handel and Kolstad (2015) combine health-plan claims and survey information to examine how information frictions and hassle costs change estimated preferences and welfare. Bhargava, Loewenstein, and Sydnor (2017) analyze dominated employee health-plan choices and use hypothetical choice experiments to examine possible causes; their documented dominance pattern cannot be explained by the ordinary risk-protection motive within the financial plan menu they study.

The correct inferential rule is joint identification. Estimate preferences conditional on the offered menu, private loss risk and nonprice frictions; test sensitivity to plausible choice errors; and avoid treating the chosen contract as a transparent readout of risk aversion. A risk-preference parameter may be separately identified from administrative choice only because researchers impose and validate restrictions on the structural system.

Mechanisms as rival explanations

Three kinds of experiments should be distinguished. A descriptive experiment asks how take-up varies with a manipulable feature. A mechanism experiment manipulates channels predicted to differ across models. A welfare experiment establishes whether a change improves consequences according to a specified welfare criterion. A price discount can increase purchase without revealing whether underlying beliefs changed. An information intervention can increase understanding and change adoption in either direction. Default enrollment can increase take-up while reducing welfare for consumers whose financial needs differ from the default. These distinctions prevent the research design from smuggling normative conclusions into a descriptive dependent variable.

Sampling and External Validity

Population, season and institution

Sample selection shapes both the exposure distribution and the meaning of insurance. Students may understand lottery tasks well but have limited experience with real claims. Farmers facing seasonal income constraints may treat a small premium as a substantial liquidity commitment. Formal employees can have employer-curated health menus and payroll deductions. Homeowners face institutional differences in regulation and claims practice. A valid international account reports the setting of each study and asks which mechanisms plausibly persist under a changed population, contract and regulatory environment.

Harrison and List's (2004) field-experiment framework indicates why 'realistic' is multidimensional. A field study can use real goods and familiar buyers but still provide an unusual promotional subsidy or temporary information session. External validity is an argument about the invariance of causal mechanisms when treatment implementation, product and institutional context change. Replication in multiple contexts is one source of evidence; explicit transport assumptions are needed even with replication.

Agricultural insurance: price, liquidity, trust and learning

Cole and colleagues (2013) conduct randomized rainfall-insurance interventions in rural India and find that price sensitivity coexists with constraints related to liquidity, trust and salience. A low observed adoption rate therefore does not identify a single underlying preference. Their interventions make separate predictions testable, but the evidence remains situated in the contract and delivery channels of the investigated market. A design that shifts a premium without measuring seasonal liquidity may conflate willingness to insure with immediate ability to pay.

Cai, de Janvry, and Sadoulet (2015) use randomized network information exposure in rural China. The authors report that an additional treated friend produced a take-up effect equivalent to a 13-percent premium reduction in their empirical setting, and their design distinguishes knowledge diffusion from imitation of purchase decisions. This is a study-specific equivalence, not a policy-invariant global elasticity. Network experiments also require attention to interference: one participant's treatment can affect another participant's outcome. When spillovers are the estimand, designs must randomize or otherwise identify the relevant network exposure.

Karlan and colleagues (2014) examine risk and credit interventions in Ghana, randomly allocating grants and rainfall insurance opportunities. The reported investment responses highlight a distinction between insurance demand as an immediate binary outcome and insurance as a protection technology changing subsequent productive decisions. A project that records take-up alone can miss welfare-relevant consequences or wrongly attribute productive changes to preferences that were never measured. Trust and observed payouts evolve over time; one-season experiments may lack the horizon necessary to estimate learning.

Clarke's (2016) basis-risk demand analysis explains theoretically why extremely risk-averse individuals can rationally avoid imperfect index coverage. Comparative work must distinguish a failure of an insurance market to reach clients from rational rejection of poorly matched protection. Researchers should measure perceived versus objective basis risk, contract familiarity, premium financing and product reliability before describing demand as irrational.

Health and property insurance: menus and comprehension

Employer health-plan data offer repeated real choices, but the employer's selected menu, tax treatment and employee characteristics determine the comparison. Handel and Kolstad's (2015) information-frictions analysis uses supplementary survey information to differentiate financial protection from costly plan navigation. Bhargava and colleagues' (2017) dominated-plan evidence demonstrates that observed selection can diverge substantially from the standard informed-choice benchmark. Researchers designing health-insurance choice experiments should present overall expected spending, tail exposure, provider networks and any dominance relationships clearly, then check understanding before inferring preference curvature.

Homeowners or automobile deductible studies permit relatively sharp variation in expected out-of-pocket losses, but they still depend on actuarial expectations and consumer perceptions. An apparent willingness to pay for lower deductibles may be a price for simplified budgeting, an expectation of service or a misperception of loss probability. Distinct explanations must be associated with observables or targeted treatments. These conclusions also apply to commercially insured small enterprises, where bargaining, delegated purchasing and firm liquidity further complicate the assumption of a single household decision maker.

Representativeness, selection and response

Random assignment gives internal comparability between experimental arms within the recruited population. It does not establish that the recruited population represents all policyholders. Survey weighting may adjust measured population margins when the sampling mechanism and relevant auxiliary variables are known, but it does not repair unmeasured selection automatically. Attrition and incomplete policy comprehension can introduce post-treatment selection bias if experiments condition analysis on participants who complete a quiz or purchase stage. Report recruitment, consent, exclusions, baseline balance, compliance, follow-up and missingness by assigned arm.

Heterogeneous treatment effects should be defined around pre-specified and substantively relevant characteristics: wealth or short-term liquidity, product experience, exposure to loss, comprehension, trust, digital access, and institutional claims experience. Comparisons across countries should hold currency, purchasing power, coverage quality and contract enforceability sufficiently comparable. Country labels alone are weak explanatory variables when actual exposures and institutions differ substantially.

Transportability and limits of aggregating results

A pooled 'insurance demand elasticity' across different products is often underdefined. Some studies report an extensive-margin semi-elasticity of take-up with respect to quoted premium; others estimate conditional coverage, stated willingness to pay, or a structural preference distribution. These estimands have different denominators and response margins. A defensible synthesis can compare sign, magnitude and uncertainty within a defined class of contracts, but a statistical meta-analysis requires comparable outcomes, designs, and sampling errors. This monograph therefore performs structured methodological synthesis rather than reporting a pooled numerical effect unsupported by harmonized data.

The international conclusion is conditional and substantive: price, trust, liquidity, comprehension, payment reliability and basis risk repeatedly emerge as candidate mechanisms in different insurance settings, but their relative importance is a property of the population-product-institution combination. Transfer to a new market requires renewed measurement and often a fresh experiment.

Methodological Standards and Replication

A reproducible question-to-estimand protocol

A research program should begin by stating the decision and causal contrast in operational terms. For a randomized premium experiment, one may define the average intention-to-treat effect as E[Y(140)-Y(80)] where Y(premium) is an indicator of actual policy purchase under that quoted premium. Random assignment identifies the sample-population mean difference if the assignment mechanism is sound and outcomes are measured consistently. With noncompliance, assignment identifies an offer effect; identifying a treatment-on-the-treated effect requires further assumptions and an explicit instrument or compliance design. For information treatments, a change in comprehension may mediate adoption, but conditioning on post-treatment knowledge without an identification strategy can bias the mechanism estimate.

For clustered interventions, such as village-level information sessions, standard errors and randomization inference must respect the cluster-level assignment. Individual-level precision that ignores a limited number of randomized clusters exaggerates the information supplied by the design. Network spillovers require exposure mappings or designated saturation designs. Pre-registered primary outcomes and treatment contrasts help distinguish confirmatory results from exploratory heterogeneity analysis. Where many outcomes are tested, adjusted inference or transparently reported multiplicity is necessary for calibrated interpretation.

Registration and transparent reporting

The American Economic Association's RCT Registry policy documents registration and version history for field experiments and its journals' policy of trial registration before submission. A protocol can identify the sampling frame, choice task, product terms, premium randomization, power calculations, primary outcome, follow-up horizon, data exclusions and analysis plan. An amendment trail explains how design changes occurred. The record should clearly separate planned hypotheses from later exploration.

The CONSORT 2025 reporting statement is primarily a guideline for randomized clinical trials. Its ideas about participant flow, allocation, missing outcomes and transparent reporting are informative analogues for insurance-demand experiments, while its clinical scope and reporting context must remain clear. An economic insurance experiment should adopt field-appropriate standards and journal requirements, and state the exact source of each methodological rule rather than implying a clinical guideline governs all insurance research.

Materials, data and reproducibility

A replicable insurance study should archive the exact task screens and question wording, instructions, translations, comprehension checks, payment protocols, randomization scripts, contractual documents, data dictionaries, permissible analysis code and a reproducible description of financial calculations. For real policies, anonymized transaction timestamps, quote menus, and contract versions can be decisive. Privacy, commercial confidentiality and insurance consumer protections constrain public release of microdata; controlled access, synthetic teaching files and reproducible aggregate outputs are alternative forms of transparency. Reproducibility means another qualified researcher can inspect the actual analytical choices without needing unauthorized access to personal information.

The reporting record should include failures and null findings. An experiment that shifts a framing variable without changing purchase can constrain the proposed mechanism, especially when confidence intervals exclude substantively meaningful effects. Trials with weak power may leave major uncertainty, and a large p-value cannot by itself establish absence of an effect. Readers need effect sizes and uncertainty for the primary outcome, and a clear distinction between statistical identification and practical policy relevance.

Measurement diagnostics

At minimum, every instrument should be checked for monotonicity of response in premium where theory warrants it, internal consistency of repeated choices, comprehension of loss and payout states, randomized-order sensitivity, attention and fatigue, systematic rounding, and response to incentives. Trial evidence should report whether the participant understood the policy, and whether the understanding measure was itself changed by treatment. It should test mechanisms with direct measures rather than residual interpretations of take-up.

Holt and Laury's (2002) incentive-effects evidence and Crosetto and Filippin's (2016) task comparison findings jointly imply that calibration exercises should vary stake size and elicitation format where feasible. Field validation can compare experimental responses with subsequently observed enrollment, renewal, claim experience and cancellation, while preserving the distinction between predictive association and causally identified effects. The point of validation is to make clear which observable behavior an instrument predicts, not to obtain a single universal accuracy score.

Causal evidence, structural models and welfare

Experiments answer questions about treatment effects in the implemented decision environment. Structural models answer questions about latent primitives and counterfactual choices under their maintained assumptions. Linking the two requires disciplined estimation of beliefs, contract payoffs, choice errors and constraints. Einav, Finkelstein, and Cullen (2010) demonstrate a price-variation approach to insurance welfare using demand and cost curves, with selection assessed in their empirical application. Their framework clarifies that purchase benefits alone do not establish social welfare: insurer costs, adverse selection and consumer surplus matter as well. It also shows that even a convincingly identified demand curve is one component of a welfare calculation.

The welfare effect of a default, subsidy, simplified explanation or new index product depends on who obtains coverage, how well it matches their exposure, what it costs to supply and whether the choice is understood. A policy that raises take-up can improve protection for one population and generate expensive, weakly protective contracts for another. The scientifically relevant endpoint may include expected consumption, catastrophic shortfall, realized claims, household investment or financial distress, in addition to enrollment. A multi-outcome evaluation must identify the primary normative criterion, record trade-offs and report distributional impacts.

Cross-study synthesis and editorial inference

The literature supports six integrated methodological conclusions. First, reported insurance demand is a function of contract parameters and elicitation rules, making exact task disclosure foundational. Second, an incentive-compatible report is conditional on the rules and preferences assumed by the elicitation mechanism. Third, lottery risk attitudes provide evidence about preferences under the task but do not uniquely determine purchase of an insurer's contingent promise. Fourth, randomization identifies causal effects of the assigned intervention, while explanations involving beliefs or frictions require separate measurement and identification. Fifth, product mismatch, liquidity and institutional trust are substantive demand determinants rather than afterthoughts. Sixth, external validity requires a transport argument specifying which conditions are shared across populations and contracts.

These conclusions constitute an original analytical synthesis of established research rather than a novel proprietary theory. The conceptual gain is precision of inference: the researcher can explain exactly when an insurance experiment informs an individual's valuation, a market's take-up response, a causal mechanism or a welfare claim. The literature permits all four questions, but each requires an independently justified link between observed choice and asserted meaning.

Conclusion

Insurance demand experiments are most informative when designed around the economic content of the insured promise. Hypothetical surveys enable systematic elicitation over rich policy attributes, incentivized tasks introduce real consequences, field experiments identify effects of specified interventions, and administrative datasets show actual contract selection. Their outputs become comparable only after the analyst specifies the contract's state space, probabilities and beliefs, premium schedule, information environment, claim reliability, sample and decision horizon.

The reviewed record establishes recurring interpretive challenges. Risk elicitation varies by task and stakes. Framing and probability representation alter insurance choices. Index coverage carries basis risk, while households may face liquidity constraints and uncertain trust. Administrative deductible and health-plan choices can reflect information frictions as well as risk preferences. Randomized information and premium interventions can illuminate causal behavior in specific settings, and their international significance must be justified through contextual comparison and replication.

The decisive research practice is to make the inferential chain explicit from offered contract to observed response, from response to identified estimand, and from estimand to any claimed theoretical or welfare conclusion. Clear in-text links to verified sources, full bibliographic records, reproducible financial derivations and documented sampling restrictions make that chain inspectable. This monograph supplies a source-critical methodological synthesis and formal examples within the published insurance-science architecture. It makes no assertion of a new named theory or universal demand coefficient, claims no original participant data, and reserves first-publication metadata for an authorized release after production verification.

References

  • American Economic Association. 2026. "AEA RCT Registry: Registration Policy and Procedures." American Economic Association policy webpage. DOI or original record.

  • Andersen, Steffen, Glenn W. Harrison, Morten I. Lau, and E. Elisabet Rutstrom. 2008. "Eliciting Risk and Time Preferences." Econometrica 76 (3): 583-618. DOI or original record.

  • Barseghyan, Levon, Francesca Molinari, Ted O'Donoghue, and Joshua C. Teitelbaum. 2013. "The Nature of Risk Preferences: Evidence from Insurance Choices." American Economic Review 103 (6): 2499-2529. DOI or original record.

  • Becker, Gordon M., Morris H. DeGroot, and Jacob Marschak. 1964. "Measuring Utility by a Single-Response Sequential Method." Behavioral Science 9 (3): 226-232. DOI or original record.

  • Bhargava, Saurabh, George Loewenstein, and Justin Sydnor. 2017. "Choose to Lose: Health Plan Choices from a Menu with Dominated Option." Quarterly Journal of Economics 132 (3): 1319-1372. DOI or original record.

  • Binswanger, Hans P.. 1980. "Attitudes Toward Risk: Experimental Measurement in Rural India." American Journal of Agricultural Economics 62 (3): 395-407. DOI or original record.

  • Cai, Jing, Alain de Janvry, and Elisabeth Sadoulet. 2015. "Social Networks and the Decision to Insure." American Economic Journal: Applied Economics 7 (2): 81-108. DOI or original record.

  • Carson, Richard T., and Theodore Groves. 2007. "Incentive and Informational Properties of Preference Questions." Environmental and Resource Economics 37: 181-210. DOI or original record.

  • Charness, Gary, Uri Gneezy, and Alex Imas. 2013. "Experimental Methods: Eliciting Risk Preferences." Journal of Economic Behavior & Organization 87: 43-51. DOI or original record.

  • Clarke, Daniel J.. 2016. "A Theory of Rational Demand for Index Insurance." American Economic Journal: Microeconomics 8 (1): 283-306. DOI or original record.

  • Cohen, Alma, and Liran Einav. 2007. "Estimating Risk Preferences from Deductible Choice." American Economic Review 97 (3): 745-788. DOI or original record.

  • Cole, Shawn, Xavier Gine, Jeremy Tobacman, Petia Topalova, Robert Townsend, and James Vickery. 2013. "Barriers to Household Risk Management: Evidence from India." American Economic Journal: Applied Economics 5 (1): 104-135. DOI or original record.

  • Crosetto, Paolo, and Antonio Filippin. 2016. "A Theoretical and Experimental Appraisal of Four Risk Elicitation Methods." Experimental Economics 19 (3): 613-641. DOI or original record.

  • Cubitt, Robin P., Chris Starmer, and Robert Sugden. 1998. "On the Validity of the Random Lottery Incentive System." Experimental Economics 1 (2): 115-131. DOI or original record.

  • Dohmen, Thomas, Armin Falk, David Huffman, Uwe Sunde, Jurgen Schupp, and Gert G. Wagner. 2011. "Individual Risk Attitudes: Measurement, Determinants, and Behavioral Consequences." Journal of the European Economic Association 9 (3): 522-550. DOI or original record.

  • Einav, Liran, Amy Finkelstein, and Mark R. Cullen. 2010. "Estimating Welfare in Insurance Markets Using Variation in Prices." Quarterly Journal of Economics 125 (3): 877-921. DOI or original record.

  • Gine, Xavier, Robert Townsend, and James Vickery. 2008. "Patterns of Rainfall Insurance Participation in Rural India." World Bank Economic Review 22 (3): 539-566. DOI or original record.

  • Gneezy, Uri, and Jan Potters. 1997. "An Experiment on Risk Taking and Evaluation Periods." Quarterly Journal of Economics 112 (2): 631-645. DOI or original record.

  • Handel, Benjamin R., and Jonathan T. Kolstad. 2015. "Health Insurance for Humans: Information Frictions, Plan Choice, and Consumer Welfare." American Economic Review 105 (8): 2449-2500. DOI or original record.

  • Harrison, Glenn W., and John A. List. 2004. "Field Experiments." Journal of Economic Literature 42 (4): 1009-1055. DOI or original record.

  • Holt, Charles A., and Susan K. Laury. 2002. "Risk Aversion and Incentive Effects." American Economic Review 92 (5): 1644-1655. DOI or original record.

  • Hopewell, Sally, et al.. 2025. "CONSORT 2025 Statement: Updated Guideline for Reporting Randomised Trials." BMJ 389: e081123. DOI or original record.

  • Jaspersen, Johannes G.. 2016. "Hypothetical Surveys and Experimental Studies of Insurance Demand: A Review." Journal of Risk and Insurance 83 (1): 217-255. DOI or original record.

  • Johnson, Eric J., John Hershey, Jacqueline Meszaros, and Howard Kunreuther. 1993. "Framing, Probability Distortions, and Insurance Decisions." Journal of Risk and Uncertainty 7 (1): 35-51. DOI or original record.

  • Kahneman, Daniel, and Amos Tversky. 1979. "Prospect Theory: An Analysis of Decision under Risk." Econometrica 47 (2): 263-291. DOI or original record.

  • Karlan, Dean, Robert Osei, Isaac Osei-Akoto, and Christopher Udry. 2014. "Agricultural Decisions after Relaxing Credit and Risk Constraints." Quarterly Journal of Economics 129 (2): 597-652. DOI or original record.

  • Sydnor, Justin. 2010. "(Over)insuring Modest Risks." American Economic Journal: Applied Economics 2 (4): 177-199. DOI or original record.

  • Tversky, Amos, and Daniel Kahneman. 1992. "Advances in Prospect Theory: Cumulative Representation of Uncertainty." Journal of Risk and Uncertainty 5 (4): 297-323. DOI or original record.