Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Health Insurance Cost Sharing Demand Elasticities

Abstract. Health insurance cost sharing changes the price of care faced by patients, the timing of medical consumption, the distribution of financial risk, and the incentives of clinicians and insurers. Its empirical study requires precise distinctions among the elasticity of visits, treatment quantities, pharmaceutical use, medical expenditures, initiation of episodes, provider choice, and clinical outcomes. A nonlinear insurance contract generates multiple prices within the same year: the immediate marginal price, the expected price over the remaining coverage period, and the total household financial exposure produced by deductibles, coinsurance, copayments, exclusions, and spending limits. This monograph reconstructs the historical experimental evidence of the RAND Health Insurance Experiment, critically assesses its frequently cited elasticity estimate of approximately −0.2, and relates that evidence to subsequent natural experiments, insurance-choice models, prescription drug analyses, and financial-protection research. The principal findings are that utilization responds to cost sharing; that numerical elasticities depend upon the estimand, type of service, population, baseline contract, and horizon; that prospective incentives alter behavior under nonlinear coverage; and that reduced spending alone cannot determine the welfare value of reduced care. Original analytical synthesis identifies the conditions required to transport elasticity estimates across policy settings, distinguishes individual savings from systemwide resource effects, and specifies a replicable evaluation protocol for heterogeneous treatment effects. The international comparison places patient-price responsiveness within the wider institutions of health financing and effective coverage. The study draws on verified original research records and institutional sources and makes no claim to a new theory or original re-estimation of microdata.

Keywords: Health insurance; cost sharing; price elasticity; demand for medical care; deductible; coinsurance; copayment; nonlinear budget constraint; RAND Health Insurance Experiment; moral hazard; forward-looking behavior; Medicare Part D; income heterogeneity; chronic illness; value-based insurance design; financial protection; causal identification; welfare economics; universal health coverage.

Introduction

The price of medical care is simultaneously a private budget constraint, an insurance-contract provision, and an institutional signal influencing how care is sought and delivered. Under health insurance, the patient usually pays an amount different from the amount ultimately received by the provider. The insurer, social fund, employer, or public budget finances the remaining eligible expenditure according to a set of rules. A change in the patient's required payment may therefore alter utilization even when medical technology, disease incidence, and the provider's nominal charge remain unchanged. This is the empirical problem of demand elasticity under insurance cost sharing.

The canonical Health Insurance monograph locates the subject within risk pooling, entitlement, access, and financing. The present dossier isolates a narrower causal question: how the demand for medically related services and associated spending changes when the patient-facing price schedule changes. The distinction matters because the demand for medical care is derived from expected health benefits, professional advice, available time, liquidity, and uncertainty about treatment effectiveness. A deductible changes a budget constraint but also modifies future exposure to a potentially large bill. A fixed prescription copayment can change whether a treatment is initiated, continued, or abandoned. A low marginal price can increase visits without proportionately increasing the intensity of each visit.

The central empirical benchmark is the RAND Health Insurance Experiment (HIE). Manning and colleagues (1987) reported an approximate price elasticity of −0.2 and a 31 percent expenditure reduction in the catastrophic plan relative to free care in their randomized cost-sharing analysis. Aron-Dine, Einav, and Finkelstein (2013) later demonstrated why a single summary elasticity requires modeling assumptions beyond randomized assignment and should be handled cautiously when applied to complex nonlinear plans in their reassessment of RAND. These are compatible results: randomization supports specific contrasts between assigned plans, while generalizing the contrasts to an arbitrary new deductible or marginal price requires an additional model.

This dossier uses experimental and quasi-experimental evidence, source-critical bibliographic review, formal definitions, and comparative institutional interpretation. It distinguishes three classes of statement throughout. Empirical findings concern documented study populations and interventions. Mathematical statements concern explicitly defined prices and outcomes. Policy implications are reasoned applications whose validity depends on additional assumptions. An original analytical synthesis is developed through disciplined comparison rather than the invention of a proprietary theoretical doctrine.

The scope extends internationally because the economic question appears under tax-financed services with user charges, statutory sickness insurance, mandatory private coverage, employer-sponsored plans, and voluntary policies. At the same time, the detailed causal evidence discussed here comes disproportionately from the United States. The comparative discussion therefore treats international financial-protection indicators as institutional context rather than as direct international estimates of a common elasticity.

Foundations and Source-Critical History

The object of study and the meaning of price

In standard demand analysis, an elasticity measures a proportional response of quantity to a proportional change in price, conditional on the relevant variables. Medical care introduces several complications. Services are heterogeneous. Some are discrete episodes; others are repeat prescriptions or continuing care. The observed payment can reflect both a patient's choice and the contract's movement from one coverage phase to another. The price of a first physician visit, the additional patient charge for a diagnostic test, and the expected cost of a hospitalization can differ within one insurance plan.

The scientifically relevant distinctions include demand for any use, the number of encounters conditional on any use, the intensity of care per encounter, the choice among providers and treatment modalities, and annual total spending. A reduction in spending can arise from fewer services, a change in the mix of services, negotiated provider-price reductions, or changes in treatment intensity. These components require separate data and produce different elasticities.

Price is also conditional on information. A patient may know a fixed copayment but remain uncertain about deductible accumulation, eligibility for an exemption, the provider's negotiated price, or whether the annual maximum has been reached. Consequently, the legal schedule, the economically expected payment, and the price perceived at the moment of choice can diverge. That divergence is an empirical object rather than an accounting error.

Cost sharing traditionally includes deductibles, coinsurance, copayments, and explicit payment ceilings. Coverage exclusions and balance billing can create additional household liabilities, although they operate differently from the marginal patient share of a covered service. Premiums determine the cost of obtaining a plan but ordinarily do not equal the marginal price of an insured visit. The separation of enrollment price and utilization price is essential for identifying the mechanism under study.

The experimental origin of the modern evidence base

The RAND experiment conducted during the 1970s and early 1980s randomly assigned participating families to insurance plans with substantially different financial terms. Its design produced an unusually credible basis for comparing spending and utilization under different assigned cost-sharing arrangements. The RAND retrospective research summary explains both the experimental architecture and the distinction between utilization and health outcomes.

In the interim controlled-trial results, Newhouse and colleagues (1981) reported that full coverage increased expenditures relative to income-related catastrophic protection. The ensuing Manning and colleagues (1987) analysis separated effects across service types and examined the relationship between patient cost sharing and expenditures. The results established that medical care demand is responsive to its insured out-of-pocket price. The magnitude of the effect depended on the comparison and the functional approximation imposed on the observed plan differences.

The experiment is historically important for two reasons. First, exogenous variation in assigned coverage reduced the confounding produced when sick people purchase generous plans. Second, the varied contract designs revealed that demand cannot be interpreted as a single homogeneous medical good with a single continuously observed price. The original experiment estimated responses to insurance arrangements, and the elasticity summary became useful for prediction only when a model translated those differences into a price-response parameter.

Aron-Dine and colleagues' (2013) three-decade methodological reassessment provides a corrective to the tendency to cite −0.2 without specifying a denominator, the nonlinear contract, or the extrapolation. They also discuss issues of participation and outcome measurement. Their review does not erase the randomized evidence; it specifies its proper evidentiary scope. The proper archival reading is therefore to preserve both the original treatment-effect evidence and the later interpretive critique.

What the historical literature establishes

RAND also studied health outcomes, allowing the central policy distinction between using less care and becoming less healthy. Brook and colleagues (1983) examined physiological and clinical outcomes under free care compared with cost sharing. In their randomized health-outcomes trial, average effects across several measured indicators were limited, but lower-income participants with hypertension and selected other vulnerable groups benefited from free care. This finding is decisive for interpretation: an aggregate utilization reduction does not identify which foregone services were valuable.

The evidence base subsequently widened beyond randomized insurance assignments. Researchers exploited employer plan changes, institutional copayment reforms, Medicare prescription benefit design, and administrative variation in public programs. These studies supplied information about older adults, chronically ill patients, low-income households, and nonlinear contract dynamics that RAND could not completely resolve. The development of this literature shifted the scientific question from whether a price response exists to which price is behaviorally relevant, whose care changes, what consequences follow, and how the answer varies by institution.

The original empirical contributions remain distinguishable. RAND establishes causal plan contrasts. Employer-plan experiments identify responses to new deductible structures in a specified insured workforce. Drug-level studies identify variation across therapeutic categories. Studies of public program expansions identify effects of gaining coverage, an intervention broader than changing a coinsurance rate. Each speaks to the scientific object, but none alone identifies a universal response of all medically necessary care to all prices.

Elasticity Estimands and Nonlinear Schedules

Definitions, denominators, and zero prices

Let q_i(p) be expected service quantity for person i under a specified patient price p, holding the relevant environment fixed. The local Marshallian price elasticity, when both quantity and price are positive and differentiability holds, is

εq,p=∂q∂ppq=∂log⁡q∂log⁡p.\varepsilon_{q,p} = \frac{\partial q}{\partial p}\frac{p}{q} = \frac{\partial \log q}{\partial \log p}.

For total spending S, a different elasticity is εS,p=(∂S/∂p)(p/S)\varepsilon_{S,p} = (\partial S/\partial p)(p/S). If spending is the product of provider price r and quantity q, then dlog⁡S=dlog⁡r+dlog⁡qd\log S = d\log r+d\log q. Thus an observed expenditure elasticity can contain both quantity and provider-price effects. Even if r is unchanged, quantity composition can move the average amount paid per encounter.

A conditional elasticity requires an exact outcome definition: prescriptions filled, patient-days, emergency visits, total allowed claims, total reimbursed spending, or household out-of-pocket expenditure. An elasticity of log annual paid claims among users is not the elasticity of probability of any use. A service-specific response should not be silently substituted for a spending-weighted portfolio response.

At a zero patient price, the logarithmic expression is undefined. Comparisons between free care and positive coinsurance rates are better represented by mean outcomes, absolute differences, percentage differences relative to a stated baseline, or carefully justified discrete-response models. An arc elasticity between two strictly positive prices may be appropriate when the experiment permits it, but a zero endpoint prevents an ordinary logarithmic price ratio. A treatment-effect contrast between free care and cost sharing should not be presented as a direct point estimate of a local elasticity at p = 0.

For a specified intervention T, the average treatment effect on an outcome Y can be written as

ATE=E[Y(1)−Y(0)].ATE = E[Y(1)-Y(0)].

When random assignment changes the entire benefit contract, this is a causal contract effect. It becomes a price elasticity only after the investigator maps the contract difference into an economically relevant price and specifies a response function. Such mapping must account for people who do not cross the deductible, people who cross it early, and people who reach an annual cap.

A further distinction concerns extensive and intensive margins. If q=P(q>0)⋅E[q∣q>0]q = P(q>0)\cdot E[q\mid q>0], then proportional changes in mean quantity decompose into changes in the probability of using care and changes in quantity conditional on use. The first margin may dominate decisions to seek initial care; the second may depend more strongly on provider choices after entry. RAND's reported distinction between contacts and intensity of contacts in the final utilization analysis illustrates why this decomposition matters.

Contractual price functions and annual dynamics

Consider an illustrative annual plan with deductible d, post-deductible coinsurance c between zero and one, and eligible out-of-pocket maximum m. Let X be cumulative allowed covered charges within the policy year, excluding items governed by separate special provisions. In a simplified schedule, cumulative patient payment is

C(X)=min⁡{m, min⁡(X,d)+cmax⁡(X−d,0)}.C(X) = \min\{m,\ \min(X,d)+c\max(X-d,0)\}.

Where differentiable, the instantaneous patient share is 1 while eligible spending is below the deductible, c after the deductible before the ceiling, and 0 after the out-of-pocket maximum. Actual policies may apply copayments, separate drug deductibles, tiered networks, exclusions, or service-specific caps, so this equation is an analytical example, not a universal legal contract.

The current marginal price can therefore be pt=C′(Xt)p_t=C'(X_t), but this price often fails to summarize the incentive to consume a service today. A forward-looking person considers the probability that future illness will move spending into a lower-cost phase. If a patient expects to reach the annual maximum anyway, the present economic cost of an additional covered service may be substantially lower than its immediately billed amount. Conversely, a person who expects to remain below the deductible may experience a price near the full allowed charge.

A compact dynamic description is

Vt(s)=max⁡a∈A(s){ut(a,s)−Ct(a,s)+βE[Vt+1(s′)∣s,a]},V_t(s) = \max_{a\in A(s)} \left\{u_t(a,s)-C_t(a,s)+\beta E[V_{t+1}(s')\mid s,a]\right\},

where s includes health, remaining deductible, cumulative spending, liquid resources, and remaining coverage time; a is a care choice; u is the expected health and consumption benefit; and the transition law determines future states. The equation formalizes the importance of continuation value. It does not establish that real patients solve a fully rational dynamic program.

Aron-Dine, Einav, Finkelstein, and Cullen (2015) exploited variation in enrollment timing and the year-end reset to distinguish otherwise similar initial prices with different expected future prices. Their dynamic-incentive evidence found responses to expected future prices, rejecting a purely spot-price interpretation in the populations examined. This finding supports the relevance of anticipated deductible and coverage transitions while leaving room for imperfect foresight, liquidity constraints, and institutional variation.

The timing of the observation window is consequently an identification decision. A January utilization response to a new deductible may differ from a November response when deductible accumulation has occurred. A cross-sectional estimate that assigns the current realized coinsurance rate to all prior utilization is vulnerable to reverse causality: high use helps determine the attained coverage phase.

Identification of causal price responses

Randomized plan assignment provides the cleanest route to a causal average effect when randomization remains valid, interference is limited, and outcomes are measured comparably. Even then, assignment may change the complete contract, including risk exposure, annual cap, and patient expectations. A separate elasticity requires assumptions about the mapping from assigned plans to prices.

Natural experiments can provide credible local variation through employer benefit reforms, eligibility thresholds, administrative copayment changes, or timing rules. The identification assumptions differ. Difference-in-differences requires credible counterfactual trends and assessment of anticipation. Regression discontinuity depends on continuity of potential outcomes around a threshold and absence of manipulation. Instrumental variables require a justified exclusion restriction and careful interpretation of a local treatment effect. A structural model adds preference and information assumptions to recover demand parameters that may be used for counterfactual simulation.

A minimal causal diagram contains the following relationships: underlying health status influences both desired care and prospective spending; insurance generosity influences financial exposure and selection into coverage; provider prices and availability influence actual utilization; prior utilization moves the deductible state; and the resulting state determines the next observed patient price. Treating the realized patient price as exogenous while omitting prior consumption can produce a mechanically distorted elasticity.

A strong empirical report states the intervention, units of randomization or policy variation, eligible population, price schedule, outcome denominator, follow-up interval, compliance, missing-data treatment, and whether effects are intention-to-treat, treatment-on-the-treated, or model-based demand responses. The original RAND record and its modern reanalysis jointly show the importance of preserving these distinctions.

Structural and reduced-form interpretations

Reduced-form evidence describes what happened under an observed change. It can be highly credible for the intervention studied but less transferable to a different nonlinear plan. Structural approaches specify preferences, expectations, health shocks, and contract rules to predict behavior under unobserved combinations. Their advantage is explicit counterfactual simulation; their price is dependence on functional form, information assumptions, and distributional restrictions.

Brot-Goldberg, Chandra, Handel, and Kolstad (2017) used a large employer's move from free care to a high-deductible plan to study spending, quantity, substitution, and price shopping. Their deductible natural experiment reported a reduction of approximately 11.8–13.8 percent in total spending, attributed in the study to reductions in the quantity of care rather than meaningful provider-price shopping. The result is a contract-reform effect in a particular employed population and is not an estimate that all deductibles lower national expenditure by the same fraction.

The interpretive implication is concrete. A deductible can reduce expenditure without improving the efficiency of provider selection; a spending effect does not identify whether low-value or high-value care was curtailed. The study also reports reductions across service categories. That observation connects the causal design to the welfare question, but the expenditure estimate alone cannot calculate the clinical value of forgone services.

Heterogeneity by Illness and Income

Severity, clinical value, and service categories

Medical care includes acute emergency episodes, predictable follow-up, chronic therapies, discretionary diagnostic services, preventive interventions, and treatments with uncertain expected benefit. Each category has its own opportunity for substitution, delay, and provider influence. An observed overall elasticity is an expenditure- or quantity-weighted combination of heterogeneous responses, with weights that themselves change under policy reform.

A patient with severe symptoms may have little practical ability to postpone hospitalization. A patient considering a routine appointment may have more discretion over timing. Chronic medication creates a different choice: reducing a dose or failing to fill a prescription can lower current expenditure while affecting longer-run health and later resource needs. Clinical necessity is not simply identical to low price responsiveness, because cash constraints can prevent high-value care even when its expected health benefit is large.

The evidence for heterogeneity is direct. Chandra, Gruber, and McKnight (2014) used exogenous variation in copayments among lower-income enrollees in Massachusetts Commonwealth Care. Their low-income cost-sharing analysis reported an overall elasticity near −0.16, with smaller responses for some chronically ill groups, including patients with asthma, diabetes, and high cholesterol. This is evidence about that program and its service mix, not a universal ordering of all chronically ill and healthy persons.

A chronically ill patient may have more predictable utilization, a higher probability of satisfying a deductible, and stronger provider recommendations. Those features can reduce some immediate responses. At the same time, recurring expenditure can compound household financial strain. Inference about chronic-disease welfare requires outcome-specific evidence rather than a single elasticity classification.

Income, liquidity, and financial exposure

Income enters the cost-sharing mechanism through at least four channels: ability to pay at the point of service, the value of foregone consumption, access to credit or family transfers, and the financial risk of an unexpected large bill. Higher-income patients may respond to price changes because they perceive low-value care as avoidable. Lower-income patients may respond because the marginal payment competes with basic consumption. Equal measured utilization elasticities across income groups do not imply equal welfare consequences.

The randomization in RAND permits more credible analysis of income-related differences than ordinary comparison of generous and less generous insurance purchasers. Newhouse and colleagues' 1981 interim results report broadly similar expenditure responses by income group under the experiment's income-related cost-sharing arrangements, while the later health-outcome evidence identifies clinically important benefits for selected poor and sick participants. These findings should be read together: response magnitudes and welfare effects are different estimands.

A simple household constraint clarifies the mechanism. Let disposable resources be y, nonmedical consumption be z, and patient payment for medical service q be P(q). Then feasibility requires z+P(q)≤yz+P(q)\leq y, abstracting from borrowing and future income. A fall in y can raise the shadow value of liquidity and alter the set of affordable care choices. If a health shock simultaneously lowers earnings and increases required services, the joint constraint intensifies. A constant coinsurance rate across households does not equalize the marginal social cost of reduced care.

Low-income cost-sharing studies must record subsidies, exemptions, household-level caps, payment arrears, transportation costs, and whether patients can obtain needed medications on credit. Administrative claims may miss treatment abandoned before billing, deferred visits, or care obtained outside the insurer's data. Financial-protection measures complement utilization measures in capturing these dimensions.

Drug-specific and individual-specific responses

Prescription drugs offer comparatively precise service definitions and repeated demand observations, but they also illustrate heterogeneity. A branded therapeutic substitute, a generic medication, a preventive drug, and an essential maintenance therapy may respond differently to the same dollar copayment. The measured price can vary by formulary tier, deductible phase, pharmacy network, and anticipated coverage gap.

Einav, Finkelstein, and Polyakova (2018) studied cost-sharing patterns in Medicare Part D and estimated price elasticities across more than 150 drugs and 100 therapeutic classes. Their drug-specific elasticity study found systematic differences across treatments and reported that private plans tended to set higher cost sharing for drugs or classes with more elastic demand. This evidence helps explain why average medical elasticity is an inadequate design statistic for drug benefits.

Such correlations between demand elasticity and plan cost-sharing policy do not by themselves establish welfare optimality. A higher patient payment can reflect expected substitutability, bargaining arrangements, risk selection, competitive incentives, or benefit regulation. Clinical benefit and equity need independent evaluation. An elasticity measures behavior under price incentives, not therapeutic value.

Individual behavior is heterogeneous even within the same plan and service. Einav, Finkelstein, Ryan, Schrimpf, and Cullen (2013) used employer panel data to examine heterogeneity in behavioral responses and the possibility that people select plans partly according to anticipated responses. Their selection-on-moral-hazard study demonstrates why policy simulation should jointly model plan choice and the induced utilization response.

Insurance selection and market equilibrium

Changing a deductible can alter who enrolls as well as how existing enrollees behave. A policy that makes low-cost coverage attractive to relatively healthy people may change the remaining risk pool. Conversely, an individual who anticipates substantial consumption may select comprehensive coverage. The observed spending difference between generous and lean plans can therefore combine selection with treatment effects.

Suppose plans A and B differ in premium π\pi, cost sharing C, and covered services. An individual's expected generalized financial cost is Gj=πj+E[Cj(X)∣Ii]G_j=\pi_j+E[C_j(X)\mid \mathcal I_i], while the utility of plan j also depends on financial-risk protection and care access. Selection into j depends on expected health needs and preferences, which also affect outcomes. A cross-sectional regression of annual spending on plan generosity will generally fail to isolate an effect of price on utilization.

The independent literature on plan selection and behavioral heterogeneity establishes that a new high-deductible option may attract people with systematically different price responses. The selection-on-moral-hazard evidence explicitly warns against ignoring this joint process. An insurer's forecast must consequently distinguish the treatment effect for an unchanged enrolled population, the risk-composition effect of enrollment shifts, and a possible provider-market equilibrium response.

Welfare and Behavioral Implications

Price response, health outcomes, and welfare

The reduction of utilization after higher patient payments is a descriptive effect. Welfare analysis asks whether the forgone services were worth their resource cost, how risk changed, who bore the losses and gains, and whether delayed care produced later harms. A larger absolute utilization response does not necessarily imply greater inefficiency; an unresponsive patient may be consuming high-value care, or may lack alternatives.

A stylized social evaluation can express the incremental net benefit of a policy change as

ΔW=ΔH∗−ΔR−ΔA−ΔD,\Delta W = \Delta H^* - \Delta R - \Delta A - \Delta D,

where ΔH∗\Delta H^* represents monetized or otherwise consistently valued changes in health and well-being, ΔR\Delta R the real resource cost of care, ΔA\Delta A administrative costs, and ΔD\Delta D changes in financial-risk and distributional burdens under a stated social valuation. The symbols are bookkeeping categories rather than estimated quantities, and their signs must be defined consistently in applications. Financial transfers between insurer and household should not automatically be counted as net social resource savings.

The RAND outcomes illustrate the limitation of spending-only evaluation. In the Brook et al. randomized assessment, average effects across many indicators were limited, while improvements from free care were reported for particular vulnerable groups. The substantive lesson is that aggregate null findings need not imply absence of clinically important subgroup effects. Statistical power and outcome choice are integral to the welfare conclusion.

The Oregon Medicaid lottery provides a complementary contrast, although it estimates the effect of access to insurance rather than a continuous variation in cost sharing. Finkelstein and colleagues (2012) found higher care use and reduced financial strain among lottery winners with greater Medicaid coverage in the first-year Oregon evidence. Baicker and colleagues (2013) reported increased health care utilization and financial protection, with mixed effects across measured clinical outcomes after approximately two years in their clinical-outcomes study. Comparing these outcomes with cost-sharing trials requires explicit recognition that gaining insurance changes premium exposure, coverage status, network possibilities, and patient prices together.

A well-grounded welfare assessment therefore separates four questions: Did care use change? Was the changed care clinically valuable? How did the policy redistribute financial risk and resources? Were effects concentrated among groups for whom access or liquidity was already constrained? An observed expenditure decline answers only part of the first question.

Hospitalization offsets and inter-payer effects

Higher copayments for outpatient care or drugs may reduce immediate claims, but the saving can be partly offset by subsequent inpatient or emergency costs. Whether such offsets occur depends on disease mix, treatment effectiveness, substitute services, and the payer bearing each cost.

Chandra, Gruber, and McKnight (2010) analyzed increased patient cost sharing for supplemental insurance among retired California public employees. Their hospitalization-offset analysis found reductions in physician visits and drug use and increases in hospital utilization; costs were shifted in part from the supplemental insurer toward Medicare. This is a direct warning against treating the savings of one payer as systemwide savings.

The mechanism does not imply that every outpatient copayment leads to higher hospital spending. RAND's original analyses did not support a universal claim that reducing outpatient coverage raises total expenditure, and the older-adult supplemental-plan context differs from RAND in age and institutions. The appropriate scientific response is conditional: hospitalization offsets are a credible risk requiring direct measurement, especially in populations with complex chronic disease and fragmented payer responsibility.

Emergency department copayment policies raise related concerns about access to urgent care. Hsu and colleagues (2006) examined the safety and clinical-event implications of emergency care copayments in the emergency-cost-sharing study. An evaluation should include emergency visit volume, severity mix, potentially avoidable events, urgent admissions, and follow-up consequences rather than infer clinical safety from a utilization reduction alone.

Inter-payer effects also have a public-finance dimension. If a private or employer-sponsored benefit reduces medication use and increases costs borne later by a statutory program, the accounting incentive of the first payer differs from the system's objective. Coordinated benefit design and long-horizon data linkage can reduce this analytical blind spot.

Behavioral responses and prospective incentives

The conventional interpretation of insured demand as moral hazard describes a price-induced increase in covered service use; it does not mean that patients are acting dishonestly. More importantly, it is an incomplete model when contracts are nonlinear, prices opaque, and care clinically complex.

The dynamic incentive study by Aron-Dine and colleagues (2015) shows that patients can respond to expected future prices, not only the current share. Conversely, informational barriers can cause some patients to react to a simplified or inaccurate representation of their financial liability. The relevant elasticity may be defined with respect to the perceived price or to a randomized change in the contract, not the realized ex post share.

Brot-Goldberg and colleagues' (2017) deductible reform evidence is particularly informative because the reduction in spending was attributed to less quantity rather than successful price shopping. This result rules out an automatic inference that giving patients greater financial responsibility turns them into more effective purchasers of competitively priced care. The mechanisms of reduced consumption and lower negotiated prices are empirically distinct.

Three behavioral channels deserve separate measurement. Attention determines whether price information enters the choice at all. Expectations determine how the patient values future deductible and maximum transitions. Liquidity determines whether the contemplated care can be purchased at the present time. A perfectly informed patient without adequate liquid resources can still defer valuable care; an affluent patient without accurate price information can fail to shop despite financial incentives.

Benefit design and distributional consequences

A uniform deductible applies the same monetary threshold to services with widely differing clinical values. Cost-sharing design can therefore encourage the reduction of beneficial and discretionary care alike. Chernew, Rosen, and Fendrick (2007) argued for differentiating patient payments according to clinical value and patient heterogeneity in their value-based insurance design framework. This is a benefit-design approach developed in prior literature, not a new framework proposed by this dossier.

Subsequent evaluation by Chernew and colleagues (2010) found that lowering selected drug copayments in a value-based program was associated with increased drug use and reduced nondrug medical service use in the program evaluation. The study is informative but cannot be interpreted as proof that all such programs are cost-saving under every implementation. Evidence about adherence, health, downstream spending, and administrative costs remains necessary.

The design problem can be stated as choosing patient payments conditional on expected clinical value, treatment alternatives, income, illness, and the administrative ability to identify eligible patients. Lower charges for high-value care may improve adherence, but value is often patient-specific and uncertain. The administrative burden of classifying care can itself impair access. Selection into plans and provider responses can change the expected effects of the benefit schedule.

Distributional analysis requires a further distinction between average cost sharing and realized financial vulnerability. A healthy household may pay little over the year even under an expensive deductible; a chronically ill household may encounter the deductible repeatedly or reach the annual cap early. Out-of-pocket maxima protect against some covered spending but do not necessarily eliminate uncovered care, time costs, or income loss. Insurance finance must therefore be analyzed jointly with the adequacy of the benefit package.

Data, Measurement, and Validation

Data architecture and reproducible estimands

A high-quality empirical analysis requires a longitudinal unit of observation at the person, insurance contract, provider, service, and time level. Enrollment files establish coverage and eligibility. Contract records establish deductible levels, coinsurance rates, copayments, service exemptions, network rules, and out-of-pocket maximums. Claims record allowed amounts, plan-paid amounts, patient liabilities, service categories, diagnostic codes, and service dates. Clinical records, mortality data, and patient surveys can measure outcomes that claims alone cannot establish.

The central database must preserve the exact order of events. At service time t, the investigator should reconstruct the patient's deductible accumulator just before the service; the contract parameters then define the prospective marginal price. Using the end-of-year deductible balance to assign an earlier price can condition on utilization that occurred after the earlier decision. When insurers revise claims, the analyst must specify whether accumulation follows service dates, adjudication dates, or the legally relevant benefit accounting rule.

At a minimum, an analysis should report the following estimands in parallel:

  • The effect of assignment to a given insurance design on total allowed spending per enrolled person-year.
  • The effect on the probability of any outpatient, inpatient, emergency, preventive, or drug use.
  • Conditional service quantity among users and its sensitivity to selection into the user group.
  • The effect on household out-of-pocket liability, debt, and catastrophic financial exposure.
  • The effect on clinical outcomes relevant to the services whose use changed.
  • Heterogeneous effects by baseline illness, income, prior use, age, and expected deductible exposure.

The population denominator is decisive. Per-enrollee and per-user spending can move in different directions when people delay initiating care. Studies restricting the sample to people with positive expenditure risk conditioning on a post-treatment variable. Where feasible, zero-heavy expenditure distributions should be modeled with a method appropriate to the causal estimand, not by silently excluding zeros.

Researchers should make a complete record of study dates, treatment implementation, follow-up, inflation adjustment, standardized pricing, confidence intervals, statistical multiplicity, missingness, attrition, and revisions to benefit rules. The exact reproducible price function belongs in the research archive, alongside assumptions about information available to patients.

Bias, uncertainty, and sensitivity analysis

The most important threats to credible elasticity estimation are selection into coverage, endogenous attainment of deductible thresholds, changing provider prices, secular trends, patient anticipation, measurement error in out-of-pocket liability, and outcome substitution outside the observed insurance system.

For randomized designs, the research record should report baseline balance, compliance, differential attrition, treatment contamination, and intention-to-treat contrasts. For observational designs, investigators should graph event-study leads and lags, specify control-group eligibility, test sensitivity to alternative windows, and assess plausible confounding from simultaneous network or benefit changes. For instrumental-variable studies, the instrument's first stage, monotonicity assumptions, and exclusion concerns should be explicit.

A useful sensitivity exercise compares at least three prices: contractual marginal price immediately before an encounter, modeled expected end-of-year price, and measured or elicited perceived price. Divergence among elasticities is itself evidence about the mechanism. A second exercise separates standardized-price quantities from observed expenditure so that provider charges do not obscure utilization behavior. A third estimates effects within prespecified morbidity and income strata, reporting uncertainty rather than ranking noisy point estimates.

The causal effect of a plan change should be reported with both absolute and relative magnitudes. If expenditure falls from 1,000 to 900 in a stated currency and base-year price level, the absolute change is −100 and the change relative to the initial level is −10 percent. These quantities are not an elasticity without an identified price ratio and a suitable demand specification. The illustrative numbers in this sentence are algebraic examples, not estimates from an empirical study.

The literature contains a further source-critical limitation. Some original articles are available only as abstracts or bibliographic landing pages through publisher interfaces; secondary accessible manuscripts and research summaries do not always contain all supplementary methods. The evidentiary status of each reference must therefore be recorded in the evidence ledger. A publisher metadata check establishes the identity and stated finding of a paper; it does not substitute for independent replication of restricted microdata.

A practical protocol for scholarly validation

This monograph adopts a reproducible research-readiness checklist within established scientific methods. First, define whether the intervention changes the marginal price, the whole plan, or coverage itself. Second, reproduce the actual policy schedule, including caps, exemptions, and timing. Third, state the outcome and denominator. Fourth, identify the experimental or quasi-experimental source of price variation. Fifth, distinguish the reduced-form contract effect from any elasticity inferred through a model. Sixth, report precision, heterogeneous effects, and downstream consequences. Seventh, describe the jurisdiction and institutional conditions that govern transferability.

For each material numerical claim, the source should be identified at the level of the original study rather than a citation chain. The RAND −0.2 figure is associated with the original Manning et al. results, while its limits for modern counterfactuals are articulated in the Aron-Dine et al. review. The 11.8–13.8 percent spending change refers to the Brot-Goldberg et al. employer study, not an economy-wide estimate. The near −0.16 estimate pertains to the Massachusetts low-income study. This attribution practice prevents the mistaken production of an apparently precise but context-free average.

An evidence ledger should separately identify historical randomized evidence, current or recent institutional data, clinical outcome studies, and the models used for extrapolation. Null findings require confidence intervals and endpoint definitions. Positive findings require attention to multiple outcomes and selective reporting. A study may be internally valid for its intervention and have limited relevance for another country's benefit design; that is a transportability judgment, not a contradiction.

Technical validation must match the reviewed source bytes and full bibliography. Scientific validation must also assess whether each research source supports the adjacent claim, whether short inline citations are semantically informative, and whether a document name or DOI actually identifies the work described. This dossier's evidence ledger records source-specific checked claims and verification boundaries; the final release must remain tied to the unchanged reviewed manuscript.

International Synthesis and Scientific Conclusions

Comparison across health-financing systems

Insurance cost sharing operates in different institutional settings. Tax-financed health services can impose user charges at selected points of delivery. Social insurance may combine statutory copayments with income-related caps and exemptions. Mandatory private systems may use regulated deductibles, standardized coverage, and subsidies. Voluntary supplementary insurance can reimburse statutory patient charges. These arrangements affect both the amount paid by the patient and the population for which a price response can be identified.

The WHO and World Bank's 2025 universal coverage monitoring report evaluates financial protection alongside access to services. Its measurements address financial hardship arising from out-of-pocket payments and provide context for the distributive significance of cost sharing. The report does not provide a single internationally standardized causal price elasticity. Reading it as a direct estimate of demand responsiveness would conflate household exposure with behavior following a price change.

The OECD's 2025 financing-system comparison reports that out-of-pocket payments accounted for just under one-fifth of health spending across OECD countries in 2023. The financial-hardship chapter further distinguishes medical goods, outpatient services, dental care, and long-term care and describes different levels of exposure across countries. These descriptive statistics establish the variability of institutional context. They cannot identify the causal effect of a specified copayment without an intervention or identification strategy.

International comparison is particularly vulnerable to denominator differences. Out-of-pocket spending as a share of total health expenditure differs from out-of-pocket spending as a share of household consumption; both differ from the probability of catastrophic spending. Informal payments, excluded services, private supplementary coverage, and price differences can alter each indicator. Comparing a United States deductible estimate with an international household-burden statistic requires an explicit account of these measurement differences.

The empirical evidence supporting detailed elasticities is also geographically concentrated. The RAND experiment, the employer reform studied by Brot-Goldberg and colleagues, Massachusetts cost sharing, and Medicare Part D are American institutional settings. Transporting their results to another health system requires evidence about provider organization, reimbursement prices, income support, waiting times, benefit exemptions, and whether patients can choose among substitute providers.

Transferability and decision criteria

The decisive question for transferability is whether the mechanism generating an estimate remains sufficiently similar in the destination setting. Five conditions deserve attention.

First, the relevant price schedule must be comparable. A fixed copayment on a publicly subsidized medicine is not the same intervention as a deductible on all outpatient services. Second, the patient population must be comparable in morbidity, age, liquidity, and prior coverage. Third, provider availability and clinical alternatives must be similar enough that a reduction in visits means a comparable behavior change. Fourth, the horizon must capture delayed treatment and future medical consequences. Fifth, the payer boundary must be clear so that savings to a supplemental insurer are not confused with net resource savings to the health system.

The empirical literature itself illustrates these boundaries. RAND's overall treatment contrasts coexist with the elderly hospitalization-offset evidence. A broad deductible can reduce total quantity while drug-specific studies show very different responses among therapeutic classes. A lower-income program can display an elasticity near −0.16 while clinically vulnerable households face different financial and health consequences. The results answer different questions about different policies.

For actuarial forecasting, the proper input is therefore a service- and population-specific response specification with uncertainty intervals and explicit behavioral assumptions. For public policy, the proper output includes health outcomes, financial protection, distributional impacts, and programwide spending. For regulation, the relevant questions include whether exemptions, caps, and accessible information prevent harmful deterrence of necessary care. None of these purposes is served well by treating −0.2 as a universal constant.

Integrated findings and research agenda

The synthesis supports six substantive conclusions. The first is that the elasticity of utilization must be defined at a precise margin and for a precise price; several distinct estimands commonly described as demand elasticity are not interchangeable. The second is that the nonlinear contract makes realized marginal price endogenous to prior use, while expected future price can influence behavior before the current deductible threshold changes. The third is that average elasticities conceal major differences across illnesses, service categories, income, liquidity, and anticipated utilization.

The fourth conclusion concerns welfare. Reduced expenditure alone cannot reveal whether the reduction represents avoidance of low-value care, loss of high-value treatment, deferred episodes, improved provider substitution, or financial transfer between payers. This judgment depends on observed mechanisms and clinical outcomes. The fifth conclusion concerns institutions: private-plan and public-budget incentives can diverge through hospitalization offsets, benefit exclusions, risk selection, and payment fragmentation. The sixth is methodological: external validity is a set of testable conditions rather than a presumption conferred by the reputation of a landmark experiment.

These findings are an original analytical synthesis of established empirical results and definitions. They do not constitute a new named theory, an original causal estimate, a systematic meta-analysis, or an independently validated predictive model.

The next research priorities follow directly. Linked claims and clinical records should test whether different copayment schedules affect adherence and health across narrowly specified diseases. Randomized price-transparency interventions could separate increased price salience from increased actual liability. Multi-payer longitudinal data could measure the net systemwide effect of cost sharing, including costs shifted to other programs. Comparative quasi-experiments outside the United States could establish whether estimated elasticities remain stable across different public purchasing and provider-access institutions. Finally, patient-level models should jointly estimate plan selection, expected deductible transitions, and liquidity constraints while retaining transparent reduced-form benchmarks.

Conclusion

Health insurance cost-sharing demand elasticities are causal or model-based measurements of how care use and related expenditure respond to a change in the financial terms faced by patients. Their interpretation depends upon the service, population, contract, time horizon, and empirical design. The RAND Health Insurance Experiment established the responsiveness of utilization to cost sharing and provided a historically important approximate elasticity of −0.2. Later research shows why that summary cannot stand in for the many distinct prices and margins generated by contemporary nonlinear insurance contracts.

Evidence from employer reforms, Medicare drug coverage, lower-income insurance programs, and older-adult supplemental plans demonstrates heterogeneous behavioral responses, dynamic incentives, and potential downstream cost shifts. Clinical-outcome studies and the Oregon insurance experiment show why the value of coverage cannot be inferred from expenditure alone. International evidence on out-of-pocket financing and financial hardship further establishes that the meaning of a patient payment depends on the financing institution and its capacity to protect households against health shocks.

A scientifically adequate cost-sharing evaluation therefore reports an identified intervention, a reconstructed contract price schedule, exact utilization and expenditure estimands, heterogeneous and dynamic effects, health and financial-protection outcomes, and explicit limits on transferability. The durable implication is that actuarial and policy decisions should use appropriately contextualized evidence rather than a decontextualized elasticity coefficient. This conclusion follows from the reviewed evidence and the formal structure of insured demand, without requiring a new theoretical system.

References

  • Aron-Dine, A., Einav, L., & Finkelstein, A. (2013). The RAND Health Insurance Experiment, three decades later. Journal of Economic Perspectives, 27(1), 197–222. https://doi.org/10.1257/jep.27.1.197

  • Aron-Dine, A., Einav, L., Finkelstein, A., & Cullen, M. (2015). Moral hazard in health insurance: Do dynamic incentives matter? Review of Economics and Statistics, 97(4), 725–741. https://doi.org/10.1162/REST_a_00518

  • Baicker, K., Taubman, S. L., Allen, H. L., Bernstein, M., Gruber, J. H., Newhouse, J. P., Schneider, E. C., Wright, B. J., Zaslavsky, A. M., & Finkelstein, A. N. (2013). The Oregon experiment—Effects of Medicaid on clinical outcomes. New England Journal of Medicine, 368(18), 1713–1722. https://doi.org/10.1056/NEJMsa1212321

  • Brook, R. H., Ware, J. E., Jr., Rogers, W. H., Keeler, E. B., Davies, A. R., Donald, C. A., Goldberg, G. A., Lohr, K. N., Masthay, P. C., & Newhouse, J. P. (1983). Does free care improve adults' health? Results from a randomized controlled trial. New England Journal of Medicine, 309(23), 1426–1434. https://doi.org/10.1056/NEJM198312083092305

  • Brot-Goldberg, Z. C., Chandra, A., Handel, B. R., & Kolstad, J. T. (2017). What does a deductible do? The impact of cost-sharing on health care prices, quantities, and spending dynamics. The Quarterly Journal of Economics, 132(3), 1261–1318. https://doi.org/10.1093/qje/qjx013

  • Chandra, A., Gruber, J., & McKnight, R. (2010). Patient cost-sharing and hospitalization offsets in the elderly. American Economic Review, 100(1), 193–213. https://doi.org/10.1257/aer.100.1.193

  • Chandra, A., Gruber, J., & McKnight, R. (2014). The impact of patient cost-sharing on low-income populations: Evidence from Massachusetts. Journal of Health Economics, 33, 57–66. https://doi.org/10.1016/j.jhealeco.2013.10.008

  • Chernew, M. E., Juster, I. A., Shah, M., Wegh, A., Rosenberg, S., Rosen, A. B., Sokol, M. C., Yu-Isenberg, K., & Fendrick, A. M. (2010). Evidence that value-based insurance can be effective. Health Affairs, 29(3), 530–536. https://doi.org/10.1377/hlthaff.2009.0119

  • Chernew, M. E., Rosen, A. B., & Fendrick, A. M. (2007). Value-based insurance design. Health Affairs, 26(2), w195–w203. https://doi.org/10.1377/hlthaff.26.2.w195

  • Einav, L., Finkelstein, A., & Polyakova, M. (2018). Private provision of social insurance: Drug-specific price elasticities and cost sharing in Medicare Part D. American Economic Journal: Economic Policy, 10(3), 122–153. https://doi.org/10.1257/pol.20160355

  • Einav, L., Finkelstein, A., Ryan, S. P., Schrimpf, P., & Cullen, M. R. (2013). Selection on moral hazard in health insurance. American Economic Review, 103(1), 178–219. https://doi.org/10.1257/aer.103.1.178

  • Finkelstein, A., Taubman, S., Wright, B., Bernstein, M., Gruber, J., Newhouse, J. P., Allen, H., Baicker, K., & Oregon Health Study Group. (2012). The Oregon Health Insurance Experiment: Evidence from the first year. The Quarterly Journal of Economics, 127(3), 1057–1106. https://doi.org/10.1093/qje/qjs020

  • Hsu, J., Price, M., Brand, R., Ray, G. T., Fireman, B., Newhouse, J. P., & Selby, J. V. (2006). Cost-sharing for emergency care and unfavorable clinical events: Findings from the Safety and Financial Ramifications of ED Copayments Study. Health Services Research, 41(5), 1801–1820. https://doi.org/10.1111/j.1475-6773.2006.00562.x

  • Manning, W. G., Newhouse, J. P., Duan, N., Keeler, E. B., Leibowitz, A., & Marquis, M. S. (1987). Health insurance and the demand for medical care: Evidence from a randomized experiment. American Economic Review, 77(3), 251–277. https://pubmed.ncbi.nlm.nih.gov/10284091/

  • Newhouse, J. P., Manning, W. G., Morris, C. N., Orr, L. L., Duan, N., Keeler, E. B., Leibowitz, A., Marquis, K. H., Marquis, M. S., Phelps, C. E., & Brook, R. H. (1981). Some interim results from a controlled trial of cost sharing in health insurance. New England Journal of Medicine, 305(25), 1501–1507. https://doi.org/10.1056/NEJM198112173052504

  • Organisation for Economic Co-operation and Development. (2025a). Financial hardship and out-of-pocket expenditure. In Health at a Glance 2025. OECD Publishing. https://www.oecd.org/en/publications/health-at-a-glance-2025_8f9e3f98-en/full-report/financial-hardship-and-out-of-pocket-expenditure_4160e6d2.html

  • Organisation for Economic Co-operation and Development. (2025b). Health expenditure by financing scheme. In Health at a Glance 2025. OECD Publishing. https://www.oecd.org/en/publications/health-at-a-glance-2025_8f9e3f98-en/full-report/health-expenditure-by-financing-scheme_074bead0.html

  • RAND Corporation. (2006). The Health Insurance Experiment: A classic RAND study speaks to the current health care reform debate (Research brief RB-9174). RAND. https://doi.org/10.7249/RB9174

  • World Health Organization & World Bank. (2025). Tracking universal health coverage: 2025 global monitoring report. World Health Organization. https://www.who.int/publications/i/item/9789240117808