RESEARCH DOSSIER
Regret, Counterfactual Thinking, and Insurance Demand
Abstract. Regret creates a counterfactual dimension of insurance choice: a household evaluates a policy by comparing its financial and experiential consequences with those of policies it could have selected. The relevant comparisons change across an uninsured loss, an uneventful policy period, a partially insured loss, and a denied claim. This monograph investigates the formal relationship between ex ante regret anticipation, ex post regret and insurance demand, and evaluates the evidence across property, flood, health, and retirement-related insurance. Its method combines close analysis of established decision theories, insurance contract arithmetic, publisher-verified theoretical studies, open empirical articles, survey research, and an identification audit. The resulting synthesis establishes four findings. First, regret may increase or decrease coverage depending on the forgone alternative, premium, loss probability, coverage efficacy, available information and the locus of responsibility. Second, the insurance choice implications of regret theory require a specified menu and comparison rule; they cannot be inferred directly from anomalous deductibles. Third, experimental and observational studies support associations between regret, insurance attitudes, switching and demand, while their distinct designs warrant different causal interpretations. Fourth, consumer protection and product design should address verifiable claims performance and decision comprehension in addition to emotional framing. An explicitly derived binary example shows that symmetric linear regret under risk neutrality can leave the actuarially fair purchase threshold unchanged, illustrating why directional empirical claims need careful conditions. The manuscript provides testable identification strategies, policy implications and a fully traceable bibliography. No new named theory or causal effect estimate is asserted.
Keywords: insurance demand; anticipated regret; experienced regret; counterfactual thinking; regret theory; coverage choice; deductibles; probability weighting; flood insurance; health insurance; claim denials; consumer protection; causal identification; behavioral insurance economics
Introduction and Research Questions
Insurance reallocates the financial consequences of uncertain future events through contracts whose benefits are contingent on specified states. Its purchaser must evaluate monetary payments made now against potential benefits paid later. At decision time the purchaser sees an incomplete future. After the period ends, actual loss, indemnity, exclusions, the premium paid, and alternative contractual choices become visible in varying degrees. Regret arises when a person evaluates the realized outcome against a more favorable outcome that might have resulted from a different choice. This structure gives insurance an unusually transparent counterfactual: paying for protection that is unused can invite a comparison with doing nothing, while suffering an uninsured loss can invite a comparison with a policy that would have paid.
The scholarly question concerns the consequences of that comparison, rather than a general claim that emotion determines insurance. Conventional economic models already explain why risk-averse households may pay a loading above the expected claim value to reduce consumption volatility. Behavioral models address probability perception, framing, information costs and the cognitive burdens of complex menus. The Braun–Muermann insurance model (2004) shows how regret preferences can move selected coverage away from extremes. The structural deductible evidence (2013) establishes that subjective weighting of small probabilities offers an important alternative. Both results can be valid because they address different restrictions and data.
The monograph asks how the accessibility and anticipated pain of a forgone alternative affect demand; when post-outcome emotional reactions predict later renewal or changes in deductibles; when insurer nonpayment changes the relevant counterfactual; and which empirical designs can separately identify regret rather than a correlated psychological construct. The object is voluntary household insurance demand for well-defined products, with implications for some collective or compulsory arrangements considered only where their institutions are specified. Findings are organized around property and flood coverage, health insurance, and selected long-horizon contracts, each with a distinct temporal structure. Cross-jurisdictional comparisons are limited to the places actually examined in the sources.
The evidence reviewed consists of theoretical journal publications, original empirical articles, methodological reviews and verified publisher records. Studies accessible through abstracts are used only for claims supported by those abstracts. The narrative uses contextual hyperlinks connected to a claim-to-source record, and mathematical calculations are explicitly demonstrative. There is no original survey, field experiment, meta-analysis or universal effect-size estimate. This division between formal implication, observed association and causal identification is maintained throughout.
Conceptual Foundations: Regret and Counterfactual Evaluation
Counterfactual thinking is the cognitive construction of an alternative to what actually occurred. It may be upward, imagining an outcome better than reality, or downward, imagining a worse one. The counterfactual can change how the observed outcome feels and inform a causal account of why it happened. Roese’s review of counterfactual thinking identifies both affective contrast and inferential functions; it also shows why counterfactual reflection can support learning despite momentary distress. Insurance regret belongs to this broader family, but has distinctive contractual reference alternatives that can be comparatively concrete.
Regret is frequently associated with recognizing that one's own decision contributed to a worse result than an available alternative would have produced. Disappointment can instead concern a poor outcome relative to expectations without the same comparison to an unchosen action. Regret, fear, anxiety, worry, anger at an insurer, and disappointment can co-occur. Their distinct objects matter for measurement. The regret-regulation literature relates anticipated and experienced regret to actions people take to prevent or manage future regret. In insurance, a policy can function both as material protection and as an anticipated means of avoiding responsibility for an uninsured loss; the same purchase can become the object of regret if no insured event occurs.
Bell’s decision analysis of regret and Loomes and Sugden’s formal regret theory introduced comparison-based preferences in 1982. The subsequent review of regret theory emphasizes the importance of which counterfactual outcomes become known. Decision makers may reasonably anticipate a different experience if a forgone outcome is never revealed; a counterfactual can nevertheless be imagined, so observability is graded rather than binary. Anticipated regret is ex ante disutility assigned to possible later comparisons. Experienced regret is an ex post judgment or feeling. A measure of post-purchase dissatisfaction, by itself, is a broader construct and should not automatically be labeled regret.
Four operational distinctions discipline the analysis. First, actual net financial loss and subjective regret are separate quantities. Second, an unpurchased policy must have been available and legally capable of paying the relevant claim for its comparison to be meaningful. Third, negative emotions can concern the choice process, the realized event, an insurer's conduct or the purchaser's prior information. Fourth, a counterfactual observed after the event may alter perceived risk in the next period; the resulting renewal decision can combine emotion, Bayesian updating, revised budget constraints and attention. Any empirical model claiming a regret mechanism must specify which pathway it identifies.
The relevant opportunity set is historically and institutionally situated. Insurance cannot be treated as a hypothetical ideal contract without exclusions, waiting periods, deductibles, limits, fraud checks and premium affordability. An illuminating counterfactual compares realizable options available at the original decision point, with their actual terms and information. Otherwise investigators risk confusing what the insured wishes had existed with what could have been purchased.
Formal Insurance Choice under Expected Utility
Consider an individual with initial liquid wealth W, a monetary loss L that occurs with probability p, and a full-indemnity policy with fixed premium P. Assume the loss and price are known, the indemnity is legally payable if the loss occurs, and the policy imposes no other frictions. A risk-neutral individual compares the expected cost of remaining uninsured, pL, with P. Coverage is preferred at P < pL and rejected at P > pL; equality gives indifference. Expected utility instead compares (1-p)u(W)+pu(W-L) with u(W-P), where u is increasing and concave. Under appropriate wealth-domain assumptions, concavity produces willingness to pay above the fair premium pL because insurance reduces variation in final wealth.
This elementary result is a reference case for behavioral theories. It neither asserts that all insured risks are priced fairly nor that observed nonpurchase must be irrational. Administrative loading, insurer default, liquidity, exclusions, competing expenditures, ambiguity about contract terms and the purchaser's beliefs affect the relevant decision. Cohen and Einav’s automobile deductible estimation demonstrates how risk preferences and unobserved claims risk can be jointly relevant to insurance data. Their model does not directly estimate regret; it clarifies why demand cannot be read from premiums and losses alone.
Policy comparisons also require a temporal budget. A low-deductible plan charges a higher certain premium for a lower contingent out-of-pocket payment. A high-deductible plan preserves cash in uneventful states while shifting more expense to the purchaser in claim states. For an individual with binding liquidity constraints, the low-deductible plan can protect against an expenditure that is hard to finance even when the difference in expected financial cost is substantial. Insurance analysis therefore must distinguish ability to bear a loss from preferences for uncertainty itself.
A useful technical principle follows: establish the monetary benchmark first, and then add regret as a specific comparison term. If a model omits the monetary menu, it may mistakenly attribute standard risk aversion or contract misunderstanding to regret. If it ignores the chosen and forgone outcomes in each state, it does not contain the causal object that regret theory investigates.
Regret-Dependent Coverage Choice
A general state-contingent expression makes the added mechanism explicit. Let a be the policy chosen from a feasible set A, let s be the realized state, and let x(a,s) be monetary wealth net of premiums and indemnities. Define the realized utility of choice a in state s as u(x(a,s)). A stylized decision criterion is the expected value of monetary utility less an anticipated comparison cost: E_s[u(x(a,s)) - λ_s g(max_{b in A}u(x(b,s)) - u(x(a,s)))], where g(t) is nonnegative and increasing for t ≥ 0 and λ_s ≥ 0 describes the salience or intensity of a relevant comparison. This expression is an expository device, not a proposed new theory. Formal regret models differ in their pairwise comparisons, aggregation, axioms and treatment of rejoicing. Its value here is to state the counterfactual choice menu, state mapping and regret penalty directly.
Regret makes the attractiveness of a given policy depend on what else the purchaser could have selected. An insurance portfolio is therefore evaluated partly against forgone contracts. In the 2004 insurance-demand analysis, regret-theoretical preferences tend to move coverage away from a no-coverage or full-coverage extreme predicted by the particular expected-utility benchmark. That is a theorem within specified model assumptions; its sign is not a universal empirical effect. A risk-averse consumer who would otherwise fully insure may be induced to retain exposure because of regret over excess premiums in loss-free states. A consumer who would otherwise take almost no coverage may insure more to reduce regret after an uninsured loss.
The qualitative pattern depends on the loading of premiums, wealth, contract form and curvature of regret evaluation. A high loading makes insurance purchase more expensive in no-loss states and can raise the salience of wasted-premium regret. Severe potential losses make the uninsured-loss counterfactual especially vivid. These forces cannot be combined by simply adding two emotion scores without their probabilities and outcome-specific payoffs.
The published wealth-comparative-statics study extends the formal discussion by showing that an insurance good's apparent normal or inferior response to wealth can change when anticipated regret is modeled. The wider lesson is that observed insurance elasticity reflects preferences over future comparisons as well as current expenditure. Yet determining actual regret parameters requires more than fitting aggregate quantity demanded to price: different parameter combinations may produce the same demand schedule, and unmeasured beliefs or product quality can replicate the pattern.
Two-State Analysis and a Symmetry Result
A compact algebraic exercise shows the importance of assumptions. Retain a binary choice between no insurance and a full-indemnity policy, a fixed loss L, loss probability p, premium P, and risk-neutral monetary valuation. Suppose P < L. If the household buys insurance and no loss occurs, the forgone no-insurance action would have saved P; anticipated action regret is λ_N P. If the household remains uninsured and the loss occurs, the forgone insurance action would have saved L-P; anticipated inaction regret is λ_L(L-P). Let both λ values be nonnegative and suppose there are no other affective terms. The expected net advantage of buying rather than remaining uninsured is
The first term is the risk-neutral financial gain. The second represents avoidance of uninsured-loss regret that would otherwise be borne; the third represents anticipated premium regret under insurance. After collecting terms, buying is preferred when
For equal intensities λ_L = λ_N = λ, this reduces exactly to P* = pL. Thus symmetric linear regret in this narrowly defined, two-action risk-neutral comparison changes the emotional description of the alternatives but leaves the financial purchase threshold unchanged. When λ_L > λ_N, the threshold exceeds pL; when λ_N > λ_L, it falls below pL. The conclusion follows by algebra, with no sample data or estimated coefficients. It is not asserted as a new author theory, and it does not overturn Braun and Muermann’s results, which use a richer preference specification and contract menu.
For example, set p = 0.02, L = $100,000, λ_L = 0.50 and λ_N = 0.10. Then the purchase threshold is approximately $2,707.58: numerator 0.03 × 100,000 = 3,000, denominator 0.03 + 0.98 × 1.10 = 1.108, so P* ≈ $2,707.58. This is a calculated hypothetical threshold, not an observed willingness to pay. If λ_L = λ_N = 0.50, the threshold returns to $2,000.
The exercise also exposes identification difficulties. The data could identify a threshold, while a continuum of (λ_L, λ_N) pairs generates it. Without targeted variation that changes the comparison or provides independent regret measures, there is no unique empirical estimate of either parameter. Randomizing feedback about the forgone alternative or separately eliciting plausible action and inaction regret can generate identifying restrictions, but even then the experiment must test whether those manipulations alter beliefs or attention.
Feedback, Observability, and Responsibility
A regret comparison requires some account of what would have happened under an alternative choice. That account may come from observed claims, public damage information, known contractual formulas, an insurer's detailed explanation or an imagined narrative. Zeelenberg's expected-feedback research shows that knowledge or anticipation of feedback on foregone outcomes can change behavior and may support risk seeking as well as risk avoidance. Its general finding matters for insurance because the feedback structure differs across policies: no-loss premium regret may be easy to compute, while the hypothetical payout of an unpurchased policy can be uncertain.
A flood victim without insurance may observe that neighbors with a similar dwelling received indemnities, increasing the credibility and emotional intensity of the uninsured-loss counterfactual. However, neighbor policies may differ in eligibility, excesses, limits and exclusions; superficial similarity does not establish that the household would have received the same payment. Similarly, a healthy household that paid premiums for a year can readily imagine retaining those premiums, while the avoided financial exposure was never observed because the insured event did not occur. Ex post 'waste' is a state-dependent appraisal, not proof that the ex ante protection was valueless.
Responsibility can alter the emotional meaning of a comparison. A purchaser may feel culpability when they rejected clear coverage after understanding a hazard. They may feel deceived or angry when a policy was represented as comprehensive but the claim falls outside a disclosed exclusion; if the exclusion was opaque, experienced disutility may arise from unfair conduct rather than personal regret. The regret-regulation framework links regret to decision ownership and corrective action, but real insurance markets also involve brokers, employers, family decision makers and institutional defaults. Responsibility is distributed in ways that need direct measurement.
Information policy may produce an unusual tradeoff. Revealing a better alternative can improve future decisions by clarifying a choice error, while also intensifying retrospective regret. Withholding useful information to prevent a negative feeling would impair informed decision making. The appropriate empirical question is whether information improves prospectively evaluated welfare after accounting for its effects on emotions, comprehension and trust, rather than treating a reduction in measured regret as a sufficient welfare objective.
Ex Post Regret after Insurance Outcomes
Four outcomes are analytically distinct. In a loss-free year, a household may view premium payment as a forgone saving, even though insurance successfully provided protection throughout the period. Following a loss with full payment, the purchase can be retrospectively validated, while the household still experiences distress from property damage or health impairment. Following a partially reimbursed loss, the household may regret selecting a high deductible, insufficient limit or exclusion-sensitive rider. Following a denied claim, the household can experience both the monetary loss and an evaluation of the purchase as ineffective; attribution to regret depends on whether a more suitable available contract would have paid.
These outcomes create different future decisions. An uninsured-loss experience can motivate purchase, but it may also deplete assets and lower ability to pay for premiums. A paid claim can increase trust in insurance or reveal claims friction, potentially acting in opposing directions. A loss-free year can cause cancellation, particularly if attention shifts toward short-run cost, while repeated years without claims might rationally update a genuinely changing risk estimate. The claim that 'loss history causes regret-driven switching' therefore requires distinguishing subjective beliefs, wealth shocks, updated information, and emotion.
An important temporal distinction is regret about the past purchase and anticipated regret about a future purchase. A consumer may retrospectively regret paying an expensive premium and still buy the same policy for the next year if a longer time horizon makes the protection valuable. Conversely, someone may retrospectively approve of past insurance because of a paid claim but switch insurers following unexpectedly poor service. Satisfaction and renewal are at best imperfect proxies for regret, since renewal decisions are also determined by switching costs, market pricing, availability, defaults and provider networks.
Regret can be used as a learning signal. The counterfactual cognition literature describes how comparing outcomes with plausible alternatives can improve causal inference and future action. Such learning is valuable when the alternative really was feasible and a revised choice can lower expected loss. It is hazardous when a dramatic but very rare loss induces overestimation of future frequency, or when an uneventful period prompts a mistaken conclusion that insurance is unnecessary. A welfare evaluation must distinguish information about the stochastic process from information about one realized draw.
Claims Denial, Nonpayment, and Contractual Credibility
Anticipated regret concerns more than the decision to insure. A policy that does not pay when the consumer reasonably expected payment changes both material risk transfer and the expected comparison with going uninsured. If an insurance contract contains an exclusion that applies in the realized state, the premium may have purchased protection against other states but not the one that occurred. The scientific evaluation therefore asks whether a rejected claim followed the insured peril, a deductible, a limit, late notification, lapse, misrepresentation, noncovered property, or nonperformance by the insurer. These explanations describe materially different counterfactuals.
Suppose a household pays P for a policy with a 90 percent chance of reimbursing an otherwise eligible loss, conditional on an insured event, because documentary exclusions or insurer nonpayment sometimes defeat recovery. If the loss probability is p and the loss amount L, expected indemnity in this reduced-form illustration is 0.9pL, before administrative costs or correlated claims conditions. The consumer can regret paying the premium after a denial; the intensity of that response depends on whether another available policy was more reliable. Yet even a policy that pays with high probability may be valuable as a partial hedge. Treating all denial events as evidence that insurance provides no expected financial benefit commits a state-selection error.
Contractual clarity changes the information basis for regret. Where an exclusion is plainly communicated before purchase, the consumer may knowingly accept reduced protection for a lower premium. Where a misleading product description obscures a major exclusion, retrospective disappointment can coexist with a genuine information failure or unfair treatment. Neither outcome alone identifies regret as the mechanism behind future demand. A regression using claims-denial experience should control for policy features, insurer availability, hazard severity, litigation outcomes and household solvency; even with rich controls, denial selection and unobserved contract complexity remain important confounders.
A prospective experiment can isolate one component by holding loss risk, premium, legally enforceable indemnity and expected payouts constant while varying a verifiable explanation of what an alternative policy would have done. The resulting difference in demand would measure the effect of the comparison message under experimental conditions. It would still require manipulation checks to show that the message did not alter perceived claim reliability, fairness or comprehension. A second experiment varying actual indemnity reliability would identify the product-quality effect separately. This distinction keeps consumer protection from being reduced to emotional persuasion.
The practical lesson is that a purchaser needs both an intelligible contract and accurate claims information. Counterfactual narratives about a denied claim are often determined by fine contractual boundaries. Evidence on regret becomes useful when those boundaries, rather than an idealized label such as “insured,” are represented in the observed choice.
Flood Insurance and Rare-Loss Decisions
Flood coverage is a salient test case because disaster losses can be severe, probabilities low and households' mental availability of events uneven across time and locations. The neglected-disaster analysis explains why low probabilities may fail to justify the perceived information costs of learning about protection. That explanation operates even without regret: a household may never reach the stage of comparing contracts. Once exposed to an uninsured flood, the same household can imagine the policy it declined and experience intense inaction regret. The event may also revise the subjective loss probability, reveal previously unknown hazard exposure, or exhaust funds needed to purchase future coverage. These channels have different observable consequences.
Robinson and Botzen's Dutch homeowner threshold study (2018) reports a sample of 1,041 Dutch homeowners and associates anticipated regret and worry with a lower threshold of concern about flood risk. The Netherlands lacked ordinary household flood insurance in the setting described by the researchers; their demand outcomes came from an experimental policy scenario, with 624 of 1,041 respondents assigned a performance-related payment condition, rather than actual take-up of commercially offered flood insurance. A lower concern threshold may mean that the consumer attends to a very small perceived probability at which insurance becomes relevant. Their subsequent demand analysis (2020) uses closely related Dutch homeowner evidence to study regret, worry, probability weighting, perceived control and flood-insurance demand. These publications should not be counted as two wholly independent population replications. Their findings support the importance of measured psychological constructs in a particular institutional setting; they do not establish that increasing regret would cause a proportionate rise in coverage everywhere.
Another line of evidence examines revealed flood-contract choices. Collier and coauthors' large-sample flood analysis uses more than 100,000 flood-insurance contracts to study small- and large-stakes risk preferences. Actual contracts have stronger behavioral realism than hypothetical survey questions, but they ordinarily lack an independent measurement of anticipated regret or the purchaser's salient counterfactual. Many consumers choose low deductibles and relatively high limits; the study's contribution is evidence about apparent risk preferences, not an estimate of regret causality.
Country-specific market arrangements matter. In a setting where public compensation is expected after a catastrophe, the forgone-insurance option may involve anticipated government relief. Where banks require coverage as a loan condition, purchase may not reflect voluntary willingness to insure at all. Where coverage is scarce or excluded for high-risk properties, the allegedly forgone insurance contract may never have been available. An international synthesis must therefore specify land-use exposure, public relief, lender mandates, risk communication and insurability before comparing regret coefficients.
The flood literature suggests two research priorities. One is a randomized comparison of communications presenting identical hazard frequencies but differing in whether they describe the financial consequences of an uncovered loss, the forgone indemnity, or merely numerical risk. The second is an administrative-data follow-up observing actual applications, renewals and claim outcomes after the communication. Together they can separate emotion-report changes from durable protection choices.
Health Insurance: Repeated Choices and Renewal
Health insurance provides repeated coverage decisions in which premiums are salient, claims and care needs may recur, and plan attributes reach far beyond the simple cash payout. Deductibles, provider networks, formularies, copayments, annual maximum exposure and the possibility of changing health status determine the actual counterfactual. A healthy year can elicit regret about paying for low-deductible coverage; a costly year can elicit regret about selecting a high deductible. These responses can induce switching even when the underlying statistical risk is unchanged.
Pauly and Kunreuther's repeated-choice health-insurance experiment is unusually useful for separating experience from physical risk. Their internet survey of 1,567 adults uses a sequence of ten hypothetical annual choices and randomly drawn losses, with underlying loss conditions fixed across rounds. Participants can change plan choices after outcomes and report reactions. The experiment detects substantial switching and emotional responses to prior loss experience even though realized events do not change the model's next-period objective probability. Because incentives are hypothetical and the sample and elicitation conditions matter, it demonstrates a plausible laboratory mechanism, not the magnitude of real-market switching caused by regret.
Born and Sirmans' post-purchase health-insurance study uses survey data on American private-market insurance decisions and connects financially unfavorable selections with a lower reported probability of renewal. This is informative about perceived bad choices and prospective retention, while the estimated association can be explained by missing knowledge about plan benefits, premiums, insurer behavior, household health, preference heterogeneity or retrospective judgment. A consumer who rejects renewal after discovering a worse-than-expected network has learned about product quality; that decision need not be regret-driven in the technical sense.
Real-market research sharpens that caution. Bhargava, Loewenstein and Sydnor's dominated health-plan study examines selections among employee health plans and finds mistakes associated with misunderstanding the financial implications of plan attributes. A seemingly irrational health-insurance choice may therefore reflect an incorrect price calculation rather than counterfactual preferences. If an intervention increases comprehension and changes enrollment, the result cannot automatically be attributed to diminished regret. The intervention may have corrected the underlying representation of the available contracts.
For causal research, the strongest feasible design would record the full feasible menu, price quotes, previous claims, known household health needs, objective option dominance, comprehension tests, separately elicited regret, and eventual actual renewal. One randomization could present accurate future-state comparisons; another could present ordinary contract information without emotional narratives. Researchers would need to rule out changed perceived hazard probabilities and nonfinancial provider benefits. The central analytic question is whether a specific counterfactual channel predicts behavior after these alternative mechanisms are measured.
Long-Horizon Protection and Retirement Regret
Insurance choices with long horizons generate a different time structure of feedback. Life-contingent annuities, long-term care insurance and certain disability protections may not produce observable benefits for many years. Purchasers must finance premiums or forego liquidity now, while future care needs, mortality, family support and product availability remain uncertain. Retrospective regret is especially sensitive to life events and to what respondents believe the forgone product would have delivered.
Hurwitz and Mitchell's older-age financial-regret experiment (2025) uses 1,612 older respondents connected to the U.S. Health and Retirement Study and examines regret about several financial decisions, including failure to purchase long-term care protection and annuities. They report substantial retrospective regret: about one third expressed regret about not purchasing long-term care insurance and roughly one quarter about not having purchased annuities, with wording, populations and timing defined in the article. Their longevity-information treatment also changed stated annuity regret: 32 percent under information compared with 21 percent in the control condition in a reported comparison. These are survey responses after an information intervention, not measured annuity purchases or identified welfare gains.
Long-horizon regret may reflect genuine ex post information. A respondent who experiences a costly care need can observe the relevant uninsured expense and may plausibly prefer to have bought protection. Another respondent may know only a generic product description, while a real policy would have included underwriting, benefit limits, inflation features, waiting periods and possible exclusions. A valid counterfactual must reconstruct the contract that was actually available at the earlier date, including insurability and premiums at the purchaser's then age and health status. Otherwise the hypothetical comparison can overstate the achievable outcome of buying insurance.
Care costs and retirement wealth also alter the causal path. Experiencing a long-term care shock can affect current regret, wealth and demand through separate channels, while older individuals may no longer be eligible to buy the same product. An observed association between regret and current purchase intention can therefore exaggerate future purchasing responsiveness. The meaning of regret here is often retrospective evaluation of a closed opportunity set. Policy implications concern timely comprehension, transparent product suitability and realistic early-life comparisons rather than emotional prompting after eligibility has ended.
A comprehensive empirical design would combine panel observations recorded before purchase eligibility ends, contemporaneous product quotations, later care or mortality outcomes, and repeated regret questions. Historical underwriting access and the household's actual feasible options are indispensable for distinguishing regret about a realistic forgone choice from a hypothetical ideal policy that was never offered.
Market Equilibrium and Product Design
A purchaser's preference for avoiding regret can feed back into the menu supplied by insurers. If buyers compare purchased and unpurchased contracts, the presence of a low-premium/high-deductible policy may influence the attractiveness of a full-coverage product even among customers who never select the cheaper option. Conversely, a prominently displayed comprehensive policy may become the reference against which a budget policy is later regretted following a claim. The menu then has consequences beyond the direct expected payout of each contract.
Huang, Muermann and Tzeng's hidden-regret market model formalizes insurance contracting with private information and regret preferences and examines pooling, semipooling and separating equilibria. Their distinctive insight is that the characteristics of an option that remains unpurchased can still matter to equilibrium through regret comparisons. This is a formal implication of a specified economic environment; it does not show that all observed insurance marketing uses the mechanism or that a particular menu is socially optimal.
Product design must account for contract selection and adverse selection. If low-risk and high-risk customers evaluate the same forgone policy differently, a menu optimized under ordinary willingness-to-pay estimates may miss how consumers rank alternatives. Still, menu complexity can increase errors and search costs. A firm that offers many nearly overlapping options may alter consumer counterfactual comparisons while making substantive comparison harder. The welfare effect depends jointly on actual risk transfer, administrative load, comprehension, sorting and the regret experience.
A productive design comparison holds available actuarial value and coverage outcomes fixed while altering the presentation or discoverability of foregone options. A second comparison changes underlying contract quality while keeping display stable. The former identifies menu-framing consequences; the latter identifies coverage economics. Only with both contrasts can researchers determine whether a reported improvement is due to better risk protection, more informative disclosure or changed anticipated regret.
Insurers and public regulators face different objectives. A firm may value immediate purchase conversion or lower lapse rates. Consumer-welfare analysis values feasible financial protection, understandable obligations, fair claims performance and informed decisions over time. Inducing a consumer to buy an expensive product by maximizing imagined uninsured-loss regret can increase sales while reducing net consumer welfare. Conversely, disclosing the actual loss scenarios under which adequate coverage matters can improve decisions when consumers otherwise misunderstand the risk. Empirical design therefore needs outcome measures that extend beyond take-up.
Alternative Explanations and Identification
Regret is only one potential explanation for insurance-demand anomalies. Prospect theory emphasizes reference-dependent gains and losses and nonlinear probability weighting, as summarized in Kahneman and Tversky's foundational article. Low-probability overweighting can generate high willingness to pay for modest-risk protection even without anticipated counterfactual regret. The insurance-preference structural estimates provide empirical motivation for that alternative. Risk aversion, ambiguity aversion, loss aversion, mental accounting, limited attention, affordability, insurer trust and inaccurate perceptions of risk may likewise generate insurance patterns that resemble regret effects.
A plausible causal diagram includes baseline wealth W, subjective hazard beliefs B, preference parameters U, product menu M, anticipated regret R, coverage choice C, realized loss S, observed indemnity I, experienced regret E and next-period choice C'. Wealth, beliefs, preferences and menu characteristics affect C, while realized loss S and insurer performance I jointly affect E and the resources and information available for C'. Regressing C' on E without modeling S and I is vulnerable to confounding because insured and uninsured households observe different sets of events and payments. Conditioning on a realized claim can also select a nonrandom post-choice subset.
The ideal regret manipulation changes the expected emotional comparison while preserving the state probabilities, financial payoffs, contract feasibility, comprehension and attention. In practice the emotional framing itself may increase risk salience or alter probability beliefs. A study should therefore pre-register the expected pathway, separately test the probability-belief channel, and report the total treatment effect on actual choice alongside any mediation analysis. A significant treatment effect on regret ratings is a manipulation check, not proof that the observed behavior was mediated exclusively by regret.
Randomized experiments can identify effects of particular messages, menu presentations or feedback arrangements. They rarely identify a stable psychological parameter that generalizes across countries, insurance lines and time horizons without a structural model. Observational panel studies can detect patterns of renewal after events but need assumptions about unobserved market responses. Structural estimation can use rich menus and revealed choices, yet different preference representations can be observationally equivalent. The field-risk-preferences review provides an important framework for interpreting the degree of identification that insurance choice data can sustain.
A practical evidence hierarchy follows from these distinctions. A formal theorem demonstrates an implication of its axioms; it does not establish prevalence. A survey identifies stated associations under its sampling and measurement assumptions. A hypothetical incentive experiment demonstrates responses in a controlled choice setting. A randomized field intervention can identify a treatment effect on real decisions in the treated setting. A general causal claim about regret requires triangulation across these designs and direct evidence about the counterfactual comparison itself. The distinction is epistemic, not a ranking of intellectual importance.
Measurement and Experimental Research Design
Regret measurement starts with a defined decision and a feasible forgone alternative. Asking simply “Do you regret your insurance?” conflates emotional regret with disappointment, anger, satisfaction, hindsight knowledge and perceived financial disadvantage. A more specific instrument identifies the choice date, the eligible alternatives, the realized state, and the question of whether the respondent believes another selectable policy would have produced a better outcome. The instrument should distinguish anticipated regret before the decision from experienced regret afterward.
Useful prospective items measure the expected intensity of regret if an uninsured loss occurs, if an insured year passes without a claim, if a claim falls under a deductible, and if a claim is denied because of contract conditions. The same person can anticipate all four and assign distinct weights. Respondents should also report perceived loss frequencies, conditional indemnity probabilities, understanding of exclusions, confidence in the insurer and expected ability to finance a loss. These measures help discriminate regret from incorrect information about the policy.
Experimental choice tasks should use realistic contracts, including premiums, deductibles, loss frequencies, claim limits and exclusions in consistent units. Random assignment may vary whether the participant is shown the outcomes of forgone alternatives after each period. A separate randomization can vary the description of a verified, feasible alternative without changing numerical payoffs. Following the design concerns surveyed in Jaspersen's methodological review, researchers should distinguish hypothetical from financially incentivized choices and disclose whether incentives reflect the full scale of insured losses.
The outcome structure should be preregistered: primary real-policy application or renewal, secondary premium willingness to pay, selection of deductible, comprehension, probability beliefs and regret reports. The design can compare immediate treatment responses with decisions made weeks or months later. If administrative data are available, actual payment and claims performance should be linked using privacy-preserving procedures and appropriate consent or lawful authorization. Risky experimental manipulations that falsely represent coverage or conceal material exclusions would be unsuitable for consumer-protection research.
Analysis should estimate intention-to-treat effects first. Mediation through reported regret requires stronger sequential ignorability assumptions and should not be presented as established merely because regret reports and behavior both move. Researchers should publish treatment wording, balance checks, attrition by condition, multiplicity adjustments and robust uncertainty intervals. The insurance-framing experiments offer historical motivation for studying presentation effects, but framing effects need not be regret effects. The most informative experiment separates the two directly through orthogonal manipulations and observed belief checks.
For longitudinal models, a panel can observe individual fixed effects, before-and-after beliefs, claimed losses, premiums, contract changes and regret. Event-study graphs can reveal whether changes precede or follow a flood or claim denial. Yet post-event insurer repricing and changes in access complicate causal interpretation. Where a natural shock jointly changes risk beliefs, wealth and emotions, an event-study coefficient estimates their combined effect. Decomposition into regret and nonregret pathways requires independent variation or credible restrictions.
Cross-Jurisdictional Interpretation and External Validity
The scientific evidence reviewed here spans formal models, United States homeowner and health choices, Dutch flood-risk surveys, long-horizon decisions among older Americans and international theoretical frameworks. It is therefore geographically heterogeneous but not a representative global sample. Country-level insurance institutions determine which losses can be insured, whether coverage is mandatory, whether public relief is expected, what claims procedures apply and how much the consumer actually pays.
A Dutch homeowner reporting anticipated flood-loss regret was evaluating a study scenario: the 2018 Dutch survey explained that household flood insurance was then unavailable in that market, even though 624 respondents received performance-related experimental incentives. This differs from a purchaser of an actual American flood policy. A United States health-plan participant faces employer menus and medical provider constraints distinct from health systems financed by general taxation or social health insurance contributions. A retired respondent's annuity regret concerns a savings-to-income conversion problem whose opportunity set differs sharply from annual property coverage. Generalizing one regression coefficient across these domains would confuse institutional context with stable psychological preferences.
Even within one country, credit constraints, income, education, disability status, age, previous claims, exposure to disaster news and access to trustworthy intermediaries shape the relationship between regret and purchase. A household unable to afford a premium might report intense anticipated uninsured-loss regret while remaining uninsured. Another household may report little regret yet buy coverage because of a mortgage covenant. Observed demand is a joint result of preferences, constraints and contracts.
Comparative research should document four classes of institutions before interpreting regret: the compulsory or voluntary status of cover, the contract's legal coverage and claims reliability, the financial role of public or employer protection, and the actual distribution and price of offered policies. Only then do cross-country response comparisons have an interpretable denominator. Translation and scale invariance of regret questions also require testing. An English item about “regret” may have no exact emotional equivalent in another language, while more detailed counterfactual vignettes may travel better when translated and validated locally.
The appropriate international conclusion is conditional. Regret is a plausible and evidenced component of insurance decision making in several settings. Its magnitude, sign and welfare meaning depend on the risk and institution. The next generation of cross-national studies should collect matched prospectively defined counterfactual questions and real coverage actions under locally accurate contract scenarios, rather than treating a pooled cross-section of incomparable insurance products as one universal market.
Welfare, Consumer Protection, and Disclosure
Regret reduction is an incomplete welfare criterion. A household can avoid regretting an uninsured loss by purchasing every available policy and still suffer severe financial harm through excessive premiums. It can avoid regret over unused premiums by remaining uninsured and expose itself to losses that it cannot finance. Sound policy analysis evaluates risk transfer and affordability over the full probability distribution, along with service quality, rights and consumer understanding. Emotional outcomes may be valuable in their own right, but their treatment requires explicit normative assumptions.
A first consumer-protection priority is transparent presentation of the exact covered events, material exclusions, deductibles and aggregate limits. This enables the consumer to construct accurate comparisons rather than imagined indemnities. A second priority is plain-language explanation of what happens in both an uneventful year and a costly claim year. Insurance is a contingent service: the premium purchases the right to a conditional payment and risk transfer during the insured period; it is not necessarily wasted when no claim occurs. This point can be taught without suppressing the legitimate opportunity cost of paying premiums.
Third, claim-handling quality matters for the meaning of insurance. Unexplained denial, inconsistent communications or prolonged uncertainty may produce a retrospective sense of having chosen badly, but an improvement in redress should be assessed through actual contractual compliance and fairness as well as satisfaction. Fourth, consumers benefit from comparable quotes that distinguish protection from price and make the true forgone options visible. A simple state-by-state comparison can reduce misunderstanding even when it makes some past choices appear worse ex post.
The dominated-plan evidence illustrates why consumer financial comprehension deserves direct attention. An intervention that repairs misunderstandings may increase objective choice quality and change the emotional appraisal of the outcome. Consumer-protection evaluation should measure comprehension, affordability and coverage fit alongside emotional responses; emotional language used to increase insurance conversion should be scrutinized for misleading impressions or exaggerated expected claims.
A particularly challenging disclosure question concerns regret-based messages. A vivid description of what a consumer may feel after an uninsured disaster can be accurate yet amplify one state while ignoring the state in which premiums are paid and the event does not happen. Balanced communication should show the loss frequency, the size of potential loss, premium cost, legally payable benefit and material limitations. Where comparative claims are made, the alternative contract must have been genuinely purchasable by the target consumer. Informed financial resilience is a defensible objective for insurance policy; maximizing fear of regret is not a reliable surrogate.
Robustness Exercises and Worked Examples
The earlier binary formula permits transparent comparative statics. With p = 0.02, L = $100,000, λ_L = 0.50 and λ_N = 0.10, the threshold premium is $3,000 divided by 1.108, or approximately $2,707.58. Under risk neutrality and zero regret, the actuarially fair benchmark is $2,000. If λ_L and λ_N are both 0.50, the threshold remains $2,000. At p = 0.02, L = $100,000, λ_L = 0.10 and λ_N = 0.50, the threshold becomes $2,200 divided by 1.492, or about $1,474.53. These examples demonstrate directionality under an expressly defined linear-regret specification, not empirical measurements of households.
The price comparative static is instructive: P* is a maximum premium under the illustrative decision rule. For P below P*, insurance is preferred; for P above P*, remaining uninsured is preferred. The criterion includes a feasible fully indemnifying policy and ignores insurer default, taxes, nonmonetary claims costs and utility curvature. If the loss probability is subjective rather than objective, p must be interpreted as the decision maker's belief, and a change in probability estimates can reproduce the same price threshold without any change in regret. The examples therefore cannot identify psychological parameters from a single observed choice.
Consider next a two-deductible illustration that holds claim coverage otherwise equal. Plan H charges an annual premium of $800 and has a $5,000 deductible; Plan L charges $1,300 and has a $1,000 deductible. For one covered $10,000 loss with probability 0.10, a risk-neutral purchaser's expected annual outlay is $800 + 0.10($5,000) = $1,300 under H and $1,300 + 0.10($1,000) = $1,400 under L. The extra protection of L lowers the conditional out-of-pocket expense by $4,000 at an additional certain premium of $500, implying an incremental actuarial expected benefit of $400. Its incremental markup relative to that expected benefit is $100. This illustration does not establish that choosing L is an error: financial strain, risk aversion, beliefs and nonfinancial benefits can rationally affect the selection.
Under those simplified terms, a consumer choosing H after a loss could regret not paying the additional $500 premium to reduce the deductible by $4,000. In a no-loss year a consumer choosing L could regret paying $500 extra. Those are asymmetric state-contingent counterfactuals, but they also require verifying that both policies were actually available to the same applicant. A menu with different exclusions would change the contingent difference. The worked example clarifies the comparison object; it supplies no estimate of its psychological valuation.
A further sensitivity check concerns claim reliability. Let an otherwise full-indemnity policy pay a covered loss only with probability q independently of the loss occurrence. A risk-neutral monetary premium threshold becomes pqL, assuming no secondary costs and no recovery if it fails. At p = 0.02, L = $100,000 and q = 0.90, that expected indemnity is $1,800, compared with $2,000 when q = 1. Changes in actual contract reliability can therefore shift demand even without regret. If reliability is correlated with loss severity, replace pqL with the appropriate conditional expectation of eligible indemnity; assuming independence would then be erroneous.
These checks bound the quantitative claims of this monograph. The equations are illustrative calculations from stated assumptions, while the cited studies supply independent theoretical or empirical support for mechanisms and real market patterns. Their coefficients, samples and effects are never treated as inputs that calibrate the numerical exercises.
Evidence Assessment, Limitations, and Research Priorities
The reviewed material supports different grades of inference. Bell, Loomes and Sugden, and Braun and Muermann supply foundational formal analyses of comparison-dependent choice and insurance coverage under their model assumptions. Bleichrodt and Wakker review how such theories relate to broader decision science. Huang, Muermann and Tzeng examine the role of unchosen contracts in insurance-market equilibrium. These are genuine conceptual foundations. Their mathematics and internal consistency do not themselves identify the prevalence of regret-driven insurance demand among actual policyholders.
The strongest direct empirical insight comes from studies that ask about regret or manipulate relevant decision experiences. Robinson and Botzen examine reported anticipated regret and flood-insurance responses in a Dutch sample, with two papers involving overlapping evidence. Pauly and Kunreuther observe switching after loss realizations in a ten-period hypothetical health-insurance task with unchanged stochastic conditions. Hurwitz and Mitchell experimentally change longevity information and record older respondents' stated financial regret. Born and Sirmans associate post-purchase choice quality with renewal intentions. Each supports a different empirical proposition; none licenses a single universal quantitative causal effect on global insurance take-up.
Revealed-preference studies by Sydnor, Barseghyan and coauthors, Cohen and Einav, Collier and coauthors, and Bhargava and coauthors are crucial counterweights. Their observed deductible, flood and health-plan patterns show why field data cannot by themselves identify anticipated regret. In particular, probability weighting, contract misunderstanding, selection on hazard and the distribution of feasible products can generate patterns that resemble those predicted by regret theories. An evidence synthesis that omits these alternatives would overstate the strength of the regret hypothesis.
The present review was a targeted, theory-driven documentary synthesis rather than a preregistered systematic review or a meta-analysis. Source access varied: selected publisher-hosted full-text articles and method sections were inspected, while other peer-reviewed works were verified through publisher abstracts, official bibliographic records or publicly available excerpts. No unpublished datasets were acquired; no independent replication of study estimates was performed; no original participants were recruited. Bibliographic and scope checks protect against misidentification of articles, while reported effect sizes and assumptions remain tied to the accessible portions of their original publications.
A stronger research program would pre-register a family of orthogonal interventions: regret-comparison framing, numerical probability information, claims-reliability information and plain-language contract comprehension. It would follow real insurance purchases and renewals, report actual premiums and available menus, measure wealth and liquidity, and use longitudinal instruments for prospective and retrospective regret. Cross-jurisdictional comparisons should predefine institutional strata and translate counterfactual questions with measurement-invariance checks. Independent replications would then support sharper estimates of when regret raises demand, lowers demand or leaves it unchanged.
The present analytical contribution is integration of the counterfactual-choice object with explicit insurance payment states, careful negative evidence on identification, and worked calculations that expose the sign conditions of anticipated regret. It is original synthesis and reasoning within established literatures, with no claim to a new named theory, doctrine, model family or independently discovered empirical law.
Conclusion
Regret contributes to insurance demand by making alternative choices part of how purchasers evaluate protection. The uninsured person facing a loss can imagine the policy that would have paid; the insured person facing no loss can imagine keeping the premium; a partially paid or denied claim can reveal a different policy that might have performed better. These counterfactuals are financially and psychologically distinct, and their relative importance depends on loss probability, premium, available coverage, information, claim reliability and the responsibility assigned to the choice.
Formal regret theory establishes why insurance decisions can depend on foregone contracts as well as on realized financial payoffs. The cited empirical studies show meaningful associations, experimentally induced responses and rich real-market patterns, while requiring care about the difference between regret, risk perception, probability weighting, plan misunderstanding and changing household constraints. The binary example further demonstrates that equal linear anticipated regret across the two states need not move the risk-neutral purchase threshold. Directional claims about regret are therefore conditional theoretical propositions or carefully identified empirical results, not universal slogans.
This synthesis yields a practical scholarly and institutional conclusion. The most reliable way to understand regret in insurance is to reconstruct what the household could actually have purchased, describe what each feasible contract would pay in each relevant state, measure both anticipated and experienced comparisons, and test whether those measures explain real decisions after controlling for monetary payoffs and competing behavioral mechanisms. A clear contract and an intelligible comparison of outcomes support informed protection. The scientific study of insurance regret advances by treating counterfactuals as verifiable components of decisions and by reporting the limits of every inference.
References
David E. Bell (1982). “Regret in Decision Making under Uncertainty.” Operations Research 30(5): 961–981. DOI: 10.1287/opre.30.5.961. https://doi.org/10.1287/opre.30.5.961
Graham Loomes and Robert Sugden (1982). “Regret Theory: An Alternative Theory of Rational Choice Under Uncertainty.” The Economic Journal 92(368): 805–824. DOI: 10.2307/2232669. https://doi.org/10.2307/2232669
Michael Braun and Alexander Muermann (2004). “The Impact of Regret on the Demand for Insurance.” The Journal of Risk and Insurance 71(4): 737–767. DOI: 10.1111/j.0022-4367.2004.00110.x. https://doi.org/10.1111/j.0022-4367.2004.00110.x
Han Bleichrodt and Peter P. Wakker (2015). “Regret Theory: A Bold Alternative to the Alternatives.” The Economic Journal 125(583): 493–532. DOI: 10.1111/ecoj.12200. https://doi.org/10.1111/ecoj.12200
Neal J. Roese (1997). “Counterfactual Thinking.” Psychological Bulletin 121(1): 133–148. DOI: 10.1037/0033-2909.121.1.133. https://doi.org/10.1037/0033-2909.121.1.133
Marcel Zeelenberg and Rik Pieters (2007). “A Theory of Regret Regulation 1.0.” Journal of Consumer Psychology 17(1): 3–18. DOI: 10.1207/s15327663jcp1701_3. https://doi.org/10.1207/s15327663jcp1701_3
Marcel Zeelenberg (1999). “Anticipated Regret, Expected Feedback and Behavioral Decision Making.” Journal of Behavioral Decision Making 12(2): 93–106. DOI: 10.1002/(SICI)1099-0771(199906)12:2<93::AID-BDM311>3.0.CO;2-S. https://onlinelibrary.wiley.com/doi/10.1002/%28SICI%291099-0771%28199906%2912%3A2%3C93%3A%3AAID-BDM311%3E3.0.CO%3B2-S
Mark V. Pauly and Howard Kunreuther (2019). “Responses to Losses in High-Deductible Health Insurance: Persistence, Emotions, and Rationality.” Behavioural Public Policy 3(1): 72–86. DOI: 10.1017/bpp.2018.42. Published online February 6, 2019. https://www.cambridge.org/core/journals/behavioural-public-policy/article/responses-to-losses-in-highdeductible-health-insurance-persistence-emotions-and-rationality/DE14D0241890F67FD47C1CC0EB8D317B
Patricia H. Born and E. Tice Sirmans (2019). “Regret in Health Insurance Post-Purchase Behavior.” Risk Management and Insurance Review 22(2): 207–219. DOI: 10.1111/rmir.12120. https://doi.org/10.1111/rmir.12120
Peter John Robinson and W. J. Wouter Botzen (2018). “The Impact of Regret and Worry on the Threshold Level of Concern for Flood Insurance Demand: Evidence from Dutch Homeowners.” Judgment and Decision Making 13(3): 237–245. DOI: 10.1017/S1930297500007671. https://doi.org/10.1017/S1930297500007671
Peter John Robinson and W. J. Wouter Botzen (2020). “Flood Insurance Demand and Probability Weighting: The Influences of Regret, Worry, Locus of Control and the Threshold of Concern Heuristic.” Water Resources and Economics 30: 100144. DOI: 10.1016/j.wre.2019.100144. https://doi.org/10.1016/j.wre.2019.100144
Rachel J. Huang, Alexander Muermann, and Larry Y. Tzeng (2016). “Hidden Regret in Insurance Markets.” The Journal of Risk and Insurance 83(1): 181–216. DOI: 10.1111/jori.12096. https://doi.org/10.1111/jori.12096
Abigail Hurwitz and Olivia S. Mitchell (2025). “Financial Regret at Older Ages and Longevity Awareness.” The Journal of Risk and Insurance 92(3): 719–739. DOI: 10.1111/jori.70008. https://doi.org/10.1111/jori.70008
Justin Sydnor (2010). “(Over)insuring Modest Risks.” American Economic Journal: Applied Economics 2(4): 177–199. DOI: 10.1257/app.2.4.177. https://doi.org/10.1257/app.2.4.177
Levon Barseghyan, 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: 10.1257/aer.103.6.2499. https://doi.org/10.1257/aer.103.6.2499
Levon Barseghyan, Francesca Molinari, Ted O’Donoghue, and Joshua C. Teitelbaum (2018). “Estimating Risk Preferences in the Field.” Journal of Economic Literature 56(2): 501–564. DOI: 10.1257/jel.20161148. https://doi.org/10.1257/jel.20161148
Jörg G. Jaspersen (2016). “Hypothetical Surveys and Experimental Studies of Insurance Demand: A Review.” The Journal of Risk and Insurance 83(1): 217–255. DOI: 10.1111/jori.12100. https://doi.org/10.1111/jori.12100
Eric J. Johnson, John Hershey, Jacqueline Meszaros, and Howard Kunreuther (1993). “Framing, Probability Distortions, and Insurance Decisions.” Journal of Risk and Uncertainty 7(1): 35–51. https://business.columbia.edu/faculty/research/framing-probability-distortions-and-insurance-decisions
Howard Kunreuther and Mark Pauly (2004). “Neglecting Disaster: Why Don’t People Insure Against Large Losses?” Journal of Risk and Uncertainty 28(1): 5–21. DOI: 10.1023/B:RISK.0000009433.25126.87. https://doi.org/10.1023/B:RISK.0000009433.25126.87
Benjamin L. Collier, Daniel Schwartz, Howard C. Kunreuther, and Erwann O. Michel-Kerjan (2017, revised 2018). “Risk Preferences in Small and Large Stakes: Evidence from Insurance Contract Decisions.” NBER Working Paper No. 23579. DOI: 10.3386/w23579. https://www.nber.org/papers/w23579
Daniel Kahneman and Amos Tversky (1979). “Prospect Theory: An Analysis of Decision under Risk.” Econometrica 47(2): 263–291. https://ideas.repec.org/a/ecm/emetrp/v47y1979i2p263-91.html
Yoichiro Fujii, Mahito Okura, and Yusuke Osaki (2021). “Is Insurance Normal or Inferior? A Regret Theoretical Approach.” The North American Journal of Economics and Finance 58: 101559. DOI: 10.1016/j.najef.2021.101559. https://doi.org/10.1016/j.najef.2021.101559
Saurabh Bhargava, George Loewenstein, and Justin Sydnor (2017). “Choose to Lose: Health Plan Choices from a Menu with Dominated Option.” The Quarterly Journal of Economics 132(3): 1319–1372. DOI: 10.1093/qje/qjx011. https://doi.org/10.1093/qje/qjx011
Alma Cohen and Liran Einav (2007). “Estimating Risk Preferences from Deductible Choice.” American Economic Review 97(3): 745–788. DOI: 10.1257/aer.97.3.745. https://doi.org/10.1257/aer.97.3.745