RESEARCH DOSSIER
Caregiver Supply and Informal Care Substitution in Insurance Models
Abstract. Informal caregiving is a major input into long-term care, yet its effects on insurance expenditures cannot be represented by a single universal substitution coefficient. An unpaid relative may replace purchased domestic assistance, facilitate professional nursing, delay institutional entry, alter the date on which an insured benefit is claimed, or provide indispensable support without changing contractual payments. These channels operate through different risk sets, service technologies, financing rules, and household constraints. This monograph investigates caregiver supply as an endogenous and potentially discontinuous determinant of service utilization and insurer expenditure. It combines an original critical synthesis of causal studies, official comparative care evidence, coverage definitions, statutory arrangements, and actuarial experience-study documentation with explicitly derived utilization, transition, benefit, and valuation equations. The analysis distinguishes need, entitlement, delivered care, billed care, recognized claims, paid claims, and total social resource costs. It develops identification requirements for estimating service-specific cross-effects, quantifies illustrative cash-flow sensitivities, and specifies reproducible validation standards for a caregiver-sensitive long-term care model. The established evidence supports substitution for some household-help services, complementarity or weak substitution for more skilled services, and conditional effects on institutional entry. It provides no basis for treating all family care as an actuarially free substitute or for transferring an observed substitution parameter uncritically across countries and contracts. A practical actuarial implication is to model the joint evolution of disability severity, caregiver availability, formal care, benefit eligibility, and care setting, with transparent scenario analysis when longitudinal insurer data are unavailable.
Keywords: Informal caregiving; caregiver supply; long-term care insurance; disability insurance; formal care; substitution; complementarity; home care; nursing home entry; insurance claims; utilization; actuarial projection; care burden; causal identification; social insurance; multi-state models; benefit design; source-based valuation.
Introduction
A long-term care insurance contract promises a defined monetary or service response when a covered condition is met. Much of the day-to-day care that makes life at home feasible, however, is supplied outside the professional care market. Family members, friends, and other members of social networks may provide domestic assistance, personal care, supervision, transport, appointment coordination, or ongoing help with activities of daily living. Insurers observe some of the resulting formal care expenditures but often have limited direct observation of the informal services supplied. A model that explains paid claims solely through age, sex, functional severity, and time since claim may consequently misinterpret shifts in household support as changes in morbidity, utilization preference, service price, or benefit generosity.
The scale and heterogeneity of this activity are documented in the OECD's 2025 informal-carers indicators. Across the countries and survey populations analyzed there, about 13% of respondents aged 50 and older reported providing informal care; across 26 OECD countries, women represented 61% of identified informal carers. These figures depend on survey definitions, country coverage, reference periods, and sampling restrictions. They describe surveyed caregiving, not the probability that an insured person has sufficient help when disability develops. The OECD's 2024 affordability assessment separately demonstrates why the distinction matters: public benefits, care needs, family supply, and out-of-pocket burdens vary sharply across jurisdictions.
The scientific question is the following: under which conditions does a marginal change in available informal care change the quantity, type, timing, quality, and insured cost of formal assistance? The answer must specify the service affected, severity and setting, counterfactual source of assistance, legal payer, covered event, and observation period. An informal hour used for shopping may replace a purchased domestic-help hour; an informal hour spent recognizing clinical deterioration may increase the use of skilled nursing. Both effects can occur in the same household and during the same claim.
The empirical starting point is a set of distinguishable results. Van Houtven and Norton (2004) associate instrumented adult-child informal care with lower formal home-health use and delayed nursing-home entry in their older-adult utilization study. Charles and Sevak (2005) report that instrumental-variable estimates differ from unadjusted associations in their nursing-home entry analysis. Bonsang (2009) finds substitution for paid domestic help, attenuation with severe disability, and weak complementarity with nursing care in a European SHARE analysis. These findings concern specific populations and identification strategies; they supply disciplined hypotheses rather than universally transferable insurer rates.
The method is a structured comparative and analytical review. Original empirical estimates are represented only as reported in their cited studies; no new causal estimates or proprietary insurer experience are asserted. The mathematical material is independently derived from standard probability, regression, survival, and cash-flow reasoning, with fully labeled hypothetical inputs. The analysis covers commercial long-term care insurance, social care financing, and relevant disability-income interfaces where caregiving affects household production or the observed claim process. It maintains clear boundaries between work incapacity and dependency on care. The evidentiary limit is especially important because publicly accessible insurer studies generally lack linked information about caregivers' hours, employment, willingness, and health.
Conceptual Foundations and Historical Development
The unit of need and the unit of payment
Care need is a functional and contextual phenomenon. The World Health Organization's International Classification of Functioning describes functioning and disability in relation to activities, participation, and environmental factors. In this monograph, a person's latent care requirement is represented by a task- and intensity-specific vector N(t), conditional on health, cognition, living environment, and assistive arrangements. Its components include basic personal assistance, instrumental household tasks, skilled clinical tasks, and supervision. The entries are care requirements in defined units; they are not insurance claims.
Let I_k(t) denote informal services actually delivered for task k, F_k(t) paid formal services delivered, and U_k(t) unmet or inadequately met need. A simplified accounting identity, when task outputs are comparable and all quantities refer to the same period, is N_k(t) = I_k(t) + F_k(t) + U_k(t). This is a descriptive conservation convention rather than an equality of health outcomes: an informal hour and a professional hour may differ in task scope, skill, productivity, continuity, and safety. For clinical services or supervision, a more faithful specification is N_k(t) = g_k(I(t), F(t), equipment, context) + U_k(t), with a task-specific production function g_k. The distinction prevents the arithmetic of hours from creating a false equivalence of care.
Insurance eligibility E(t), claim submission J(t), claim acceptance A(t), paid benefit B(t), and resource need N(t) must be separately recorded. An individual can be functionally dependent without meeting a particular insured trigger. A covered person may satisfy a trigger but receive sufficient free family assistance and submit no reimbursement claim. A cash indemnity may be payable even without purchased care. The claim process is therefore a filtered, contract-specific view of functional dependency.
In the United States, Internal Revenue Code section 7702B defines criteria for federally tax-qualified long-term care contracts. Its chronic-illness provisions include inability to perform at least two specified activities of daily living for the required period or substantial supervision arising from severe cognitive impairment, subject to practitioner certification. This statutory definition helps define a subset of qualified contracts; it is not a universal international definition of need or a statement that every contract pays the same benefit. Medicare's long-term care explanation further illustrates payer separation: Medicare generally does not finance long-term nonmedical custodial assistance, although other covered skilled services can be involved in a person's care.
From household production to insurance expenditure
The economic history of informal care research developed through household-production analysis, selection and simultaneity problems, and increasingly explicit treatment of the opportunity costs of care. A caregiver allocates finite time across employment, rest, household responsibilities, and care. The resulting service supply is constrained by work schedules, health, geography, obligations, preferences, and availability of alternatives. Carmichael and Charles (2003) analyze gender and labor-market aspects in their opportunity-cost study. Van Houtven, Coe, and Skira (2013) examine caregiving, work, and wages, emphasizing the need to address which causes which.
Commercial insurers historically price long-term care obligations around claim incidence, claim continuation, utilization patterns, contract limits, investment returns, and policyholder behavior. The NAIC's long-term care insurance overview, updated in July 2026, describes the experience-driven problems of historical policies, including underestimated incidence and claim duration and unexpectedly low lapses. Caregiver supply is a plausible additional influence on utilization and claim timing; the NAIC overview does not isolate or quantify its causal effect. This dossier does not convert such plausibility into an observed industry parameter.
Which substitution is at issue?
At least four distinct outcomes must be distinguished. Service substitution means that an increase in informal assistance reduces paid service use for a specified task, holding the relevant counterfactual fixed. Setting substitution means that support at home changes the timing or probability of residential care. Financial substitution means that a different source finances an otherwise comparable service, even if care inputs remain unchanged. Intertemporal substitution means that current home support postpones expenditure, potentially changing later severity and duration. The economic and actuarial signs of these effects can differ.
A fifth mechanism is complementarity: an informal caregiver's presence may make formal services possible or more productive, as when the caregiver coordinates clinical visits, supervises treatment, arranges professional respite, or detects escalating need. Bolin, Lindgren, and Lundborg (2008) find differences across home, physician, and hospital services in their European single-living-elderly study. The service category, not the label “formal care,” determines the empirical hypothesis.
The historical movement from viewing households as a passive background to studying family care as an economic input has direct valuation consequences. A payment-based insurer model can still be internally correct for its defined claims outcome while remaining incomplete as an explanation of the underlying need. An economic analysis that adds unpaid labor without defining whose perspective it represents may likewise double count or confuse fiscal savings with welfare gains. Both perspectives require explicit accounting boundaries.
Informal Caregiving Supply
Availability, capacity, willingness, and delivered care
The word “available” conceals several states. Potential caregiver supply is the existence of a person who might help; accessible supply requires practical proximity, communication, and the ability to be present; effective capacity requires physical and cognitive ability and task-specific competence; voluntary supply additionally requires willingness and sustained consent. Delivered hours are a realized outcome of these conditions and of the recipient's needs. For each potential caregiver j at time t, define C_j(t) as the vector (proximity, usable time, capability, willingness, continuity). A person-level supply measure C(t) should aggregate feasible task-specific support across caregivers rather than equate the number of adult children with usable hours.
A household may contain three potential carers but only one who can assist with transfers or nightly supervision. Another may deliver remote administrative support without providing on-site personal care. A spouse who is themselves frail may have limited safe capacity. Different kinds of support can be jointly supplied: care planning may be complementary to purchased home nursing, whereas domestic labor may be substitutable. Model calibration should therefore avoid a scalar caregiver-count variable where task-level information is feasible.
The OECD's informal-carer survey definitions differ across national and international sources, and its published indicators concern people aged 50 and older. Reports of weekly help, daily assistance, and co-resident caregiving should be kept distinct. A 12-month recall of “any caregiving” is informative about participation, but an insurer's weekly service projection needs hours, skills, and continuity aligned to the benefit period. The 2025 U.S. caregiving report issued by AARP and the National Alliance for Caregiving estimates 63 million adults providing ongoing care to adults or children with complex needs or disabilities. This broader survey universe cannot be used as an estimate of the proportion of long-term care insurance claimants with competent unpaid home-care substitutes.
A supply schedule with constraints
A transparent behavioral accounting framework begins with a caregiver time budget T_j = H_j + L_j + I_j + O_j, where H_j is paid work, L_j rest and personal time, I_j caregiving, and O_j other obligations over a common measurement interval. A change in I_j may displace wages, leisure, health recovery, or other unpaid work. Its shadow cost need not equal the caregiver's market wage; a wage-based replacement-value method answers a different valuation question. When several caregivers share duties, their hours may overlap, and double counting is possible unless simultaneity is recorded.
A stylized feasible-hours function is I_j(t) = min{D_j(t), K_j(t)} × W_j(t), where D_j is requested task-compatible care, K_j feasible capacity, and W_j a bounded measure of willingness or participation. This is an analytical illustration, not a validated behavioral production function. Its purpose is to make explicit why the marginal hour of family care may be unavailable even when an adult child is observed in a demographic registry. For severe cognitive impairment, care intensity also includes readiness and continuous supervision that cannot be reduced reliably to additive one-hour units.
Labor-market effects give this accounting substantive importance. Van Houtven, Coe, and Skira's work-and-wage study addresses employment and pay consequences, while Carmichael and Charles's opportunity-cost analysis shows why family assistance has an economic cost even when no insurer receives an invoice. A projected reduction of insurer-paid formal hours should therefore be reported as a claim-cost effect, not as an equal social cost saving.
Caregiver health and continuity
Caregiver supply can deteriorate for reasons unrelated to the recipient's illness. Employment changes, caregiving fatigue, moving home, illness, family conflict, and bereavement can alter both the quantity and quality of care. Pinquart and Sörensen's (2003) caregiver-health meta-analysis, comprising 84 articles, documents a relevant distribution of psychological and physical-health differences between caregivers and noncaregivers. Associations of caregiver status and health do not prove that any one episode caused subsequent illness, but continuity assumptions should recognize health-related supply risk.
A transition analysis consequently benefits from a caregiver state G(t) rather than a single baseline indicator. Illustrative states include robust availability, constrained availability, temporarily absent, and permanently unavailable. Transitions can be reversible, and their intensities may depend on the caregiver's age, health, employment, and the recipient's care requirement. If data support it, the recipient's and caregiver's conditions should enter a joint transition history. Treating baseline family status as permanently fixed understates exposure to caregiver disruption over a multi-year claim.
Caregiver supply also has selection effects at insurance inception. Households with more family support might be less inclined to buy reimbursement coverage; households facing severe inherited or anticipated care needs may both purchase insurance and reorganize family arrangements. These channels can change the distribution of caregiver availability in an insured cohort without changing its population-level incidence. Selection into coverage, selection into observed claims, and selection into formal care use must be analyzed separately.
Measurement and missingness
A practical longitudinal data dictionary should include recipient functional measures; cognitive supervision needs; household composition; kinship; caregiver location; verified hands-on hours; instrumental support; work status; unmet need; paid-service claims; eligible event dates; elimination periods; benefits; care setting; mortality; and caregiver transitions. The observational unit may be an insured-month, person-week, household-quarter, or episode, but the denominator must remain consistent with the rate being estimated. Hours of care offered and hours delivered are different variables.
Informal care hours are especially prone to heaping, proxy-report error, and misclassification. Twenty-four-hour supervision may be recorded as 168 hours of labor when the economic and clinical meaning differs from continuous active assistance. Conversely, brief cognitive prompts, medication reminders, and coordination can be omitted. If the insurance claim system records only reimbursed professional services, the missingness of unpaid care is structural, not random. Neither zero submitted claims nor zero formal invoices imply the absence of need.
Recommended observations include recipient-carer linked interviews, time diaries, provider billing, benefits records, episodic functional assessment, and documented changes of residence or household composition. Each source answers a different measurement question. When data are sparse, caregiver supply must be treated as an explicit uncertainty driver, with scenario bounds justified from observed conditions rather than invented incidence rates.
Formal Service Substitution
What the empirical evidence establishes
Van Houtven and Norton (2004) use two-part models and instrumental-variable methods to study older adults in U.S. survey data. Their formal-care utilization findings support lower formal home-health use and delayed nursing-home entry associated with adult-child care in the analyzed setting. The study's observational period, instrument assumptions, and eligibility universe limit its direct transportability to currently insured policies. It also demonstrates why a positive raw association between informal and formal care can reflect greater underlying illness rather than complementarity.
Charles and Sevak (2005) study 6,855 older adults using waves of the AHEAD survey. Their instrumented institutional-care results find reduced nursing-home entry from informal care even though ordinary least squares can show the opposite sign. The methodological insight is durable: conditioning on observed medical need does not necessarily remove unobserved severity, family preferences, or selective living arrangements.
Bonsang (2009) distinguishes domestic help from nursing and separates participation from intensity using European SHARE observations. In the service-specific substitution evidence, informal care substitutes for paid domestic help, the domestic-substitution effect attenuates with higher disability, and nursing care is a weak complement. Bolin, Lindgren, and Lundborg (2008) likewise report different formal-service relationships: substitution for formal home care can coexist with complementarity for medical visits. The convergence concerns qualitative heterogeneity across tasks, not a shared causal elasticity suitable for direct actuarial import.
A synthesis must attend to population selection. Some studies focus on noninstitutionalized elders or adults living alone. Such samples exclude or underrepresent households where informal care has collapsed or where severe cases already moved to institutions. Cross-country analyses mix different provider capacities, benefit designs, and norms. The literature supports explicit interactions and sensitivity checks; it does not support a blanket assumption that every additional family-care hour reduces paid services by a constant fraction.
Cross-effects and functional specification
Let F_k(t) denote paid care hours or episodes of service k. The local causal cross-effect of a feasible change in informal input for task l is β_kl(t,x) = ∂E[F_k(t) | do(I_l(t)=i), X=x] / ∂i, where X represents a clearly defined pre-treatment history and eligibility universe. A negative β denotes substitution; a positive value denotes complementarity. This is a conceptual causal derivative, estimable only under an identification design. A regression coefficient on observed I_l does not automatically estimate it.
To reflect the frequent zero use of particular services, a two-part specification decomposes E[F_k | X,I] into P(F_k>0 | X,I) × E[F_k | F_k>0,X,I]. The first part measures service initiation; the second measures conditional quantity. A total-effect change may arise from both factors. For institutional use, a time-to-event or competing-risk formulation is preferable when information about eligibility, death, and follow-up time is available.
A task-specific interaction might be β_kl = β_kl,0 + β_kl,s × severity + β_kl,p × provider-access + β_kl,b × benefit-design, evaluated only within the support of observed data. The parameters belong to the estimation design and have no assumed empirical values here. A safety constraint is also required: a negative fitted cross-effect should never be interpreted as feasible replacement of skilled services where the caregiver lacks clinical qualifications.
The production perspective adds context. For routine household assistance, tasks may be substitutable over a relevant range, making β potentially negative. For skilled nursing, household labor and professional tasks may be jointly required, allowing β positive. When informal care facilitates timely assessment, measured professional use may increase while preventable deterioration decreases. Whether insurer expenditure rises depends on which services are covered, their timing, and the downstream path of disability.
Setting transitions and competing events
Define recipient states H = independently supported at home, D = dependent at home with formal or informal assistance, R = residential care, and X = death. If warranted by data, D can be stratified by disability severity and caregiver availability. For an individual in state a, the transition intensity to b is q_ab(t | Z(t),G(t)), where Z includes clinical and contract-relevant characteristics and G is caregiver state. The probability of entering residential care before death is not the same as the conditional admission rate among survivors. A proper cumulative incidence for transition D to R incorporates competing events, including death, under the specified risk set.
A time-homogeneous Markov approximation might use a generator Q(G), but the Markov assumption is strained when residential admission depends on duration of need, prior hospital use, or caregiver fatigue. A semi-Markov or flexible duration-dependent model lets q_ab depend on time since entry to state a and time since caregiver disruption. That extension is methodologically justified if observed event histories display duration effects. It should not be imposed merely for sophistication when a simpler model fits and calibrates adequately.
A caregiver-loss event can generate several different trajectories: additional home-care hours, temporary respite, assisted-living transition, immediate institutional entry, or an episode of unmet need. The insurer may observe only some of these. Death also terminates many streams of future payments but may occur in the same risk intervals as provider use. Transition accounting should retain dates and common time conventions to avoid look-ahead errors.
Identification and falsification
The main causal problem is endogeneity. Greater impairment tends to increase both informal and formal care; preferences and resources affect both; formal service supply can itself reduce the relative's delivered hours. As a result, the descriptive correlation Corr(I,F) is not a reliable substitution parameter. Studies using instruments, fixed effects, difference-in-differences, or policy variation still require defensible assumptions.
An instrumental-variable analysis needs an instrument that changes caregiver supply and affects the formal-care outcome only through that supply, conditional on the specified controls. Child distance, employment opportunity, or family size can violate exclusion through housing, wealth, or health-monitoring channels. Bonsang's European estimation discussion illustrates the importance of such concerns. Researchers should report first-stage strength, alternative channels, balance, reduced-form effects, instrument sensitivity, and the local population to whom the estimate applies.
A policy change that pays family carers can simultaneously alter willingness, labeling, claims visibility, and the price of paid formal work. A design interpreting any subsequent change in billed professional hours as pure physical substitution confuses service production with financial reclassification. Similarly, a new eligibility rule may increase observed claims while underlying care remains constant. The relevant estimand must be declared before a preferred statistical method is chosen.
Useful falsification tests include examining tasks that informal carers could not plausibly perform, testing pre-trends where a natural experiment is claimed, studying anticipatory changes preceding caregiver loss, and comparing outcomes affected by the same severity confounders but not the hypothesized substitution pathway. Null or sign-changing results should be retained. Evidence that is valid for one category of domestic assistance may not validate a nursing-home transition effect.
Claims and Access Sensitivity
Translating service use into paid claims
Insurer expenditure is a mapping from eligible health states and covered services to contractual payments. For a reimbursement benefit, an idealized period payment can be represented as B_t = 1{eligible and accepted} × min{L_t, α_t × max(0, E_t - d_t)}, where E_t is eligible billed expense, d_t a deductible or client share expressed in money, α_t the reimbursable proportion, and L_t the period limit. Actual contracts may involve elimination periods, daily or monthly limits, aggregate pools, exclusions, provider conditions, coordination, and escalation clauses. The formula is an illustrative reduced form, not a statement of any particular policy.
For an indemnity or fixed cash benefit, payments can instead resemble B_t = 1{covered status and waiting conditions satisfied} × K_t. In this arrangement a decline in purchased formal care may have no contemporaneous effect on the insurer's fixed cash obligation. If cash is usable to compensate relatives, payment may continue while the household reallocates who provides care. Equating fewer professional invoices with insurer savings would be an accounting error.
The NAIC's coverage categories include home health, respite, personal assistance, assisted living, and community services, among others, while benefit and underwriting terms differ. Contract-level mapping must therefore precede cross-product aggregation. A standard spreadsheet grouping all formal long-term care under one expenditure code will often hide the interactions that determine the sign of caregiver-related effects.
For a prospective cash-flow valuation, write V_t = E_t[∫ from t to T of D(t,u) B_u du + D(t,T) H_T], where D is the appropriate discount factor and H a covered terminal payment if any. A discrete monthly implementation replaces the integral with a sum over mutually exclusive state probabilities, benefit amounts, and survival/termination conditions. The relevant actuarial effects of caregiver supply can operate through claim incidence, commencement delays, transitions between settings, utilization intensity, recovery, care duration, and claim reporting. Parameter changes in those components must be jointly simulated to avoid counting the same effect twice.
A reproducible expense-shift illustration
Consider an entirely hypothetical set of 100 accepted reimbursement claimants, each receiving 52 weeks of covered home services. Baseline formal service use per claimant is 12 domestic-assistance hours weekly at $30 per hour and three nursing hours at $80 per hour. Gross weekly billed expense is 12 × 30 + 3 × 80 = $600. Annual gross eligible expense is 52 × $600 = $31,200 per claimant, or $3,120,000 for 100 claimants. Assume, solely for illustration, that a contract pays 80% of eligible expense with no further limit, deductible, or waiting period in the analyzed window. Baseline insurer payment is $2,496,000.
Scenario A supposes that a loss of caregiver help causes the purchase of four additional domestic-assistance hours per week, with nursing use unchanged. Added gross expenditure per claimant is 4 × $30 × 52 = $6,240. Across 100 claimants, gross expenditure rises by $624,000 and insurer payment increases by $499,200. This is an exact arithmetic sensitivity under stated inputs; it does not imply that four hours is an empirically estimated effect of caregiver absence.
Scenario B illustrates complementarity rather than a uniform offset. Suppose increased family support replaces eight domestic-help hours weekly but permits four additional skilled-nursing hours. Gross spending changes by -8 × $30 + 4 × $80 = +$80 per claimant per week. Across 100 claimants for 52 weeks, gross spending increases $416,000 and the assumed insurer share rises $332,800. In this scenario, total formal hours fall by four per week even though paid expenditure increases. The example establishes that the sign of the expenditure effect cannot be inferred from the sign of an aggregate formal-hours effect.
Scenario C uses the same shift in care but assumes the contract pays a fixed $150 each eligible day independent of invoices. For unchanged eligibility and covered days, the insurer's modeled payout is unchanged by either substitution pattern. If additional informal support instead delays first eligibility or changes covered setting, the cash stream may change through that separate channel. The contractual trigger must be assessed independently; “care at home” alone is not a synonym for ineligibility.
Sensitivity of reserves and capital analysis
Caregiver disruption should be considered alongside mortality, functional deterioration, inflation in care wages, provider shortages, policyholder behavior, and benefit utilization in reserve stress analysis. A useful two-way stress varies both caregiver hours and formal provider availability because a family shock does not guarantee that replacement professionals can be hired. One scenario increases insurer-paid claims; another generates unmet care with little immediate change in claims; a third shifts to institutional services with different coverage or expense limits.
The first derivative of expected claims with respect to caregiver supply, where it exists, decomposes as ∂E[B]/∂I = (∂E[B]/∂F) × (∂E[F]/∂I) plus terms from eligibility, claim reporting, service prices, and setting transitions. This is a chain-rule representation under differentiability and consistent conditional expectations, not a claim that these mechanisms are independent. In practice, nonlinearity at deductibles, daily limits, benefit pools, and elimination periods produces threshold effects that require simulation rather than a single elasticity.
A multi-year insurer cohort must also account for attrition, insurance selection, and investment discounting. For example, an increase in informal care may postpone covered payments, but a later caregiver collapse may lead to intense expenditure when the insured is older or more severely disabled. The present value depends on the full hazard path. An apparently favorable one-year claim ratio may mask redistribution of costs across durations rather than a reduction in lifetime claim obligation.
The SOA/LIMRA 2015–2024 group LTD termination study, published September 30, 2026, covers over 43 million life-months and reports large numbers of recoveries and deaths. Its study population is group long-term disability income claims; it does not directly estimate informal-care substitution in long-term care insurance. This illustrates a general limitation of experience tables: the event and denominator must match the modeled mechanism. Work-disability recovery hazards should not be relabeled as care-dependency improvement or used as a proxy for family care capacity.
Access, unmet need, and distributional incidence
The welfare interpretation of reduced paid claims turns on what happens to unmet need. A household unable to replace lost family support may ration food preparation, bathing, supervision, transport, or treatment coordination. An insurer may experience stable or lower covered expenditure precisely because access worsens. A scientifically responsible model therefore reports delivered assistance, unmet need, and access constraints alongside paid claims whenever data permit.
The OECD's 2024 affordability analysis identifies substantial residual out-of-pocket burdens despite public support in many settings. Its comparison uses stylized need cases, specific public financing rules, and survey data; these estimates are not direct projections of private insurer losses. They do demonstrate why low observed formal utilization can reflect financial or geographical barriers as well as substitution by relatives.
Distribution also matters. The availability of family labor varies by household structure, place of residence, gender, and economic circumstances. Caregiver burden can fall disproportionately on women, as shown descriptively by the OECD's gender distribution of carers. Pricing and benefit strategies that assume a constant free household contribution may shift risk to people with weaker support networks and create inequitable effective protection among insureds with similar functional need.
Social cost and insurer cost are different accounts
An insurer's expected payment is a contractual cash-flow quantity. A family's cost includes forgone earnings, reduced leisure, health consequences, additional household expenses, and potentially improved continuity or preferences. The public payer's cost includes funded services and transfers; society's opportunity cost considers real resource use and displaced production, with careful treatment of transfers. A reduction in the insurer's cost is consequently compatible with an increase in household opportunity cost or unmet need.
A consistent perspective matrix can be maintained in valuation work: insurer = covered benefits and expenses; household = out-of-pocket services, time, and related losses; public sector = care and cash transfers; combined resource perspective = formal resources, caregiver time, and relevant health effects, avoiding double counting of transfers. The analysis need not assign a dollar value to every dimension. It must state which perspective its headline result answers.
Data, Legal, and Model Validation
Jurisdictional arrangements change observed substitution
An international model must distinguish three financing architectures. In an expense-reimbursement design, unpaid help can reduce covered invoices without changing underlying entitlement. In a fixed cash-benefit design, service choice may not change the payment amount. In social systems with budgets, caregiver allowances, in-kind packages, or mixed benefits, paying a family caregiver can change how labor is classified and reported even when the same person performs the same tasks. These distinctions are legal and accounting determinants of measured substitution.
Germany provides a particularly informative contrast. Its federal health ministry's long-term care insurance explanation describes a mandatory system introduced in 1995, care grades, and choices involving professional services or monetary support that can assist family caregiving. The program is explicitly a system of partial benefits. A comparison of professional service claims under different household arrangements must separately account for cash allowances and the remaining costs borne by households.
In the Netherlands, the Long-term Care Act guidance describes a needs assessment by CIZ and possible provision of qualifying long-term care at home where safe, practicable, and cost-efficient. The government's personal-budget explanation further describes arrangements with care providers and payment administration. Household choice, cost-efficiency safeguards, and specific budget rules alter the observed formal/informal boundary.
The U.S. Medicaid self-directed services guidance permits eligible program participants to direct approved service arrangements under state-plan or waiver authorities, with relevant requirements for individualized planning, budgets, staffing, backup care, and financial management. Programs and state rules must be checked before classifying a paid relative as an “unpaid” caregiver. A policy shift from agency provision to participant-directed provision can change provider category and invoice flows without causing an equivalent change in total care delivered.
In England, Care Act 2014 section 10 establishes a carer's assessment and expressly considers the carer's ability and willingness to continue providing care. This provides a legal example of an important modeling distinction: a person's caregiving capacity and desire to continue are relevant observations, not inexhaustible inputs silently assumed in a forecast. The cited statutory page is the originally enacted text; any application to a specific year or legal dispute requires the operative version and subordinate rules.
Contract and coding audit
A source-to-claim legal audit should begin with the precise applicable contract and its effective date. Enumerate covered events, functional or cognitive triggers, certification, waiting periods, benefit types, approved service categories, qualifying providers, informal or familial provider restrictions, caps, offsets, coordination with public programs, and claims-review rules. Terms such as “home health care,” “custodial care,” “personal assistance,” and “skilled nursing” are not interchangeable coding labels.
The section 7702B qualified-care definition is relevant to U.S. tax-qualified policies, whereas the Medicare coverage guidance concerns a different public payer. German care grades, Dutch needs assessment, English carer assessment, and U.S. policy ADLs represent different legal operations. A multinational experience study should preserve original categories and create explicit crosswalks for analytic comparison rather than relabel them as equivalent definitions.
For every expense or service category, audit whether the source data indicate need, delivered service, submitted claim, adjudicated eligible claim, or settled payment. Mark replacements and family-employed paid services separately. Check that claims paid after service date are assigned correctly, that reimbursement limits are applied at the correct level, and that noncovered professional services are not mistaken for informal help. The same episode may have two payers and multiple service invoices; deduplicate at episode and payer levels as required.
Empirical design and uncertainty protocol
A suitable observational risk set is insured persons alive and at risk of specified dependency during a defined period, with benefit terms known. A care-utilization subsample conditions additionally on measured need; an accepted-claims subsample answers a different question and is subject to collider selection. Caregiver supply should be time-indexed at or before the relevant utilization period. If caregiver hours are reported after care decisions, a lagged design and a clear timeline are required.
The causal target must specify the population and intervention. For example: among community-dwelling insured persons aged 65 or older with moderate ADL limitations at the start of the month, what is the change in expected paid domestic-help hours over the following quarter if feasible family assistance increases by five hours per week, holding the defined service environment? This estimand does not automatically translate to nursing care, residential entry, or policy-level total claims. Other targets may be more relevant for severe cognitive impairment or cash-indemnity policies.
A credible analysis should compare raw, severity-adjusted, and identification-driven estimates; include service-specific and severity-interaction effects; report confidence or uncertainty intervals; consider clustered household dependence; and distinguish participation and conditional intensity. It should also document reasons for unavailable caregiver information. When instrumental variables are used, the exclusion and relevance assumptions must be exposed, not hidden in a generic robustness label. Where causal designs are weak, report associations and scenario ranges rather than causal substitution rates.
Model validation needs at least five separable tasks: reproducibility of data cleaning and legal crosswalks; historical fit to service-specific utilization; calibration of transition and exposure estimates; out-of-sample prediction where feasible; and decision sensitivity under caregiver shocks. Inspect predictions by age, sex, care severity, household structure, location, product design, and observation year, subject to adequate cell sizes and confidentiality. A model can forecast total claims well while materially misrepresenting households with no available care; aggregate accuracy alone is insufficient.
If the data cannot distinguish caregiver supply from severity or provider access, partial-identification and stress testing are preferable to overprecision. Let the effect on eligible formal hours lie in an explicitly justified interval [β_L, β_U] for each service. The induced insurer-cost range should be computed through the actual benefit schedule and state transitions, not simply by multiplying that interval by a single average service price. Show both the narrow interval under chosen assumptions and broader sensitivity to plausible violations.
Temporal comparability, ethics, and audit trail
Long-term care service prices, labor shortages, demographic composition, policy designs, and regulatory rules change over time. A calibration based on older SHARE or AHEAD waves must not be presented as a measured 2026 private insurance elasticity. Vintage effects belong in the data and claims assumptions. The 2026 SOA/LIMRA disability termination study demonstrates that large, recent standardized exposure datasets are possible, but its accessible summary contains no linked caregiver-supply covariate suitable for this dossier's causal question.
Private household data are sensitive. Caregiver interviews may reveal illness, employment, financial constraints, family relationships, and willingness to provide assistance. Privacy-preserving linkage, explicit permission, purpose limitation, secure retention, and audit trails are prerequisites for deployment with personal records. The analytical need to understand caregiver supply does not justify treating unpaid assistance as a resource available for insurer exploitation or substituting assumptions for consent.
Every adopted quantitative claim should retain its population, period, measure, source version, and uncertainty. Every scenario input should be marked synthetic or empirically estimated. Every substantial claim should be traceable through a source register to an accessible research report, statute, official technical page, or publisher record. For paywalled journal studies, verification limited to official abstracts or metadata must be stated explicitly. For contracts, the applicable form and jurisdiction must be supplied before any estimate is described as insurer-specific.
Integrated Findings and Research Agenda
What the evidence supports
The reviewed literature establishes that the caregiver–formal-service relationship is heterogeneous. Several published instrumental-variable studies support substitution in domestic or home-care settings and possible delay of nursing-home entry. Evidence on skilled nursing and physician or hospital services allows complementarity, and the relationship varies with severity and care setting. The most consequential analytical correction is to estimate formal-service categories separately before translating their quantity effects into insurer payments. A model that forces a common negative coefficient across all formal services may understate some covered spending while overstating the ability of households to meet complex care needs.
The second finding concerns the distinction between underlying care need and observed insurance utilization. A household can exhibit unchanged dependency with lower insurer-paid claims because a relative supplies work, because services are unaffordable, or because the policy pays a fixed benefit. These mechanisms imply different consumer and actuarial outcomes. The observable claim count therefore cannot identify caregiver supply without complementary data or a credible design.
The third finding concerns intertemporal dynamics. The caregiver's employment, health, preferences, and proximity can change during the insured's period of dependency. The timing of caregiver loss may change the distribution of expenses across service categories and care settings. The correct object for prospective insurance valuation is consequently a joint trajectory of need, caregiver capacity, service access, and contract payments. A fixed baseline family indicator is a coarse approximation whose limitations should be stress-tested.
The fourth finding is about perspective. A lower claim cost can coexist with higher caregiver opportunity cost, health strain, or unmet need. The model must declare whether its objective is insurer solvency, affordability, access, or total social resource use. The OECD's care affordability comparison and the caregiver-health evidence give substantive reasons to keep these outcomes visible rather than compressing them into a single savings percentage.
Research questions that can be tested
First, within comparable severity and policy strata, how much of the association between caregiver capacity and paid service use survives adjustment for health, household resources, provider access, and insurance selection? The necessary data include linked care reports and claims. Second, which care tasks exhibit negative or positive causal cross-effects, and at what severity do substitutions cease to be feasible? This requires service classification and credible variation in caregiver supply.
Third, does the timing of family-care disruptions change residential transition hazards, or merely reallocate paid home-care hours? A duration-sensitive design with competing death risk can test this. Fourth, when family care is remunerated through benefits or participant-directed budgets, what share of the apparent formal/informal shift represents reclassification instead of real output substitution? The evaluation requires person-level service records and rules for eligibility and payment.
Fifth, does a caregiver-sensitive claims model improve calibration and stability out of sample compared with a severity-only model, after accounting for the increased information and privacy demands? The answer requires preregistered comparisons and insurer or administrative data. Sixth, how do projected insurer savings compare with household time costs and unmet care under common scenarios? This requires explicit, nonoverlapping accounting perspectives.
Implementation priorities for a reproducible insurer study
A defensible implementation begins with a small number of care-service categories and a documented cohort. Match each category to contractual reimbursement, fixed-cash, or public-payment treatment. Link functional assessments, relevant household characteristics, and a repeated caregiver-state measure. Estimate the participation and intensity of professional service use separately and evaluate the appropriateness of causal identification. Fit transitions between home and residential settings conditional on severity and caregiver conditions, with death as a competing event.
Next, apply the policy's benefit logic to simulated services and events. Calibrate against observed paid claims at multiple durations and by service category. Carry sensitivity ranges for caregiver-loss frequency, substitutability by task, provider scarcity, and compliance with limits. Present results as changes in expected paid cost, duration, setting, out-of-pocket burden, and unmet need, not a single universal “caregiving discount.”
Source governance is part of the model. Retain versioned contracts, statistical definitions, publication dates, legal citations, and code or data lineage. In this dossier, the evidence register identifies verified public records and the scope of verification. The hypothetical expense examples make no claim about an actual carrier, country, or market rate. No novel named actuarial theory or general new doctrine is proposed.
Conclusion
Caregiver supply is an economically and actuarially material condition governing how care needs become formal services and how formal services become insurance payments. Its measurable influence is mediated by task type, disability severity, family capacity, professional service availability, care setting, public financing, and contractual benefit design. The strongest available empirical basis supports service-specific substitution for some forms of domestic assistance and potential delay of institutional use, alongside complementarity in professional nursing and medical care. These patterns require models that can accommodate both directions of response.
The substantive implication is precise. An insurer should not translate the existence of relatives into a uniform negative claim-cost factor. It should distinguish potential and effective caregiver availability, model care supply as time-varying, estimate or stress-test category-specific service cross-effects, and apply the actual contract to forecast paid benefits. The valuation must preserve competing events, unmet care, household costs, and reporting distortions. An observed decline in paid expenditure establishes a financial outcome for the payer; understanding its meaning requires a coherent account of the underlying service and welfare changes.
The evidence establishes a rigorous agenda and a transparent set of modeling requirements. It does not establish a universal numerical elasticity, an insurer-wide reserve adjustment, or a novel proprietary theory. The original contribution of this monograph is the integrated analytical interpretation of caregiver supply across causal identification, care production, benefit mechanics, longitudinal state transitions, and multiperspective cost accounting, with reproducible mathematical illustrations and explicitly bounded claims. Further empirical estimation should use linked longitudinal care and claims records and jurisdiction-specific contracts. The present manuscript is a completed research synthesis intended for scientific editorial assessment and subsequent controlled release, not a public publication or a statement of insurer-specific experience.
References
AARP and National Alliance for Caregiving. (2025, July 24). New Report Reveals Crisis Point for America's 63 Million Family Caregivers (release concerning Caregiving in the U.S. 2025). AARP. AARP research release.
Bolin, K., Lindgren, B., & Lundborg, P. (2008). Informal and formal care among single-living elderly in Europe. Health Economics, 17(3), 393–409. Journal article.
Bonsang, E. (2009). Does informal care from children to their elderly parents substitute for formal care in Europe? Journal of Health Economics, 28(1), 143–154. Journal article.
Carmichael, F., & Charles, S. (2003). The opportunity costs of informal care: Does gender matter? Journal of Health Economics, 22(5), 781–803. Journal article.
Centers for Medicare & Medicaid Services. (n.d.). Long-term care. Medicare.gov. Medicare coverage guidance.
Centers for Medicare & Medicaid Services. (n.d.). Self-Directed Services. Medicaid.gov. Medicaid program guidance.
Charles, K. K., & Sevak, P. (2005). Can family caregiving substitute for nursing home care? Journal of Health Economics, 24(6), 1174–1190. Journal article.
Federal Ministry of Health (Germany). (n.d.). Long-term care insurance. Bundesministerium für Gesundheit. Federal program explanation.
Government of the Netherlands. (n.d.-a). Long-term care act (WLZ). Government.nl. Statutory program guidance.
Government of the Netherlands. (n.d.-b). I have a WLZ care needs assessment for residential care. Can I be cared for at home instead? Government.nl. Home-care and budget guidance.
National Association of Insurance Commissioners. (2026, July 23). Long-Term Care Insurance. NAIC Insurance Topics. Regulatory topic overview.
OECD. (2024). Is Care Affordable for Older People? OECD Health Policy Studies. OECD Publishing. DOI: 10.1787/450ea778-en. Full OECD report.
OECD. (2025). Health at a Glance 2025: OECD Indicators, section “Informal carers.” OECD Publishing. DOI: 10.1787/8f9e3f98-en. Informal-carers indicators.
Pinquart, M., & Sörensen, S. (2003). Differences between caregivers and noncaregivers in psychological health and physical health: A meta-analysis. Psychology and Aging, 18(2), 250–267. Meta-analysis.
Society of Actuaries Research Institute & LIMRA. (2026, September 30). 2015–2024 Group Long-Term Disability Termination Experience Study. Society of Actuaries. Experience study overview.
United Kingdom. (2014). Care Act 2014, c. 23, section 10, as originally enacted. legislation.gov.uk. Carer's assessment provision.
United States. (2026 version accessed). Internal Revenue Code, 26 U.S.C. § 7702B, text through Public Law 119-120 (October 7, 2026). Office of the Law Revision Counsel, U.S. House of Representatives. Qualified LTC insurance provision.
Van Houtven, C. H., Coe, N. B., & Skira, M. M. (2013). The effect of informal care on work and wages. Journal of Health Economics, 32(1), 240–252. Journal article.
Van Houtven, C. H., & Norton, E. C. (2004). Informal care and health care use of older adults. Journal of Health Economics, 23(6), 1159–1180. Journal article.
World Health Organization. (2001; official web resource consulted 2026). International Classification of Functioning, Disability and Health (ICF). WHO. ICF classification resource.