RESEARCH DOSSIER
Inclusive Insurance Take Up Retention and Renewal Behavior
Abstract. Inclusive insurance operates through a sequence of linked decisions: an eligible person obtains an offer, accepts it, satisfies activation conditions, remains insured, and either renews, leaves, or returns after an interruption. The scientific object of this monograph is the longitudinal behavior of that sequence in underserved insurance markets. Enrollment, policy persistency, contract renewal, and realized protection are defined separately, allowing commercial microinsurance, community-based health insurance, and subsidized public arrangements to be compared without treating their institutional structures as interchangeable. The investigation uses comparative institutional analysis, targeted synthesis of original experimental and longitudinal studies, and explicit denominator and exposure-time analysis. Evidence from India, Nicaragua, Ethiopia, and Ghana demonstrates materially different patterns of participation and renewal. The findings establish the importance of distinguishing affordability at purchase from affordability over time, observed service value from insured risk transfer, and voluntary exit from interruption caused by administrative, institutional, or financial constraints. The monograph develops reproducible measurement instructions, examines competing explanations of persistence, and evaluates the implications of selective renewal for risk pooling and consumer welfare. Illustrative calculations establish the consequences of cohort composition, competing exits, and missing follow-up. The analytical contribution is an integration of existing insurance economics, longitudinal measurement, and consumer protection, with no claim to a new theory or newly estimated population effect.
Keywords: Inclusive insurance; microinsurance; insurance demand; take-up; retention; renewal; lapse; adverse selection; affordability; consumer protection; longitudinal analysis; risk pooling; health insurance; rainfall index insurance; financial protection.
Introduction
An inclusive insurance market must be evaluated through the protection that its participants can obtain and maintain across time. A policy purchased once may provide useful protection for its contractual term. A household facing repeated exposure to loss requires additional opportunities to maintain appropriate coverage as circumstances change. The behavior of policyholders at these opportunities is a central question of insurance science because it connects the organization's ability to pool risk with the household's ability to maintain a contingent financial claim.
The insured population and the target population are different objects of observation. The target population comprises persons, families, or enterprises eligible for the relevant insurance arrangement. The insured population comprises covered exposure units under specified contracts at a stated time. Purchasers may represent an intermediate population when waiting periods, acceptance, eligibility validation, or failed payments postpone activation. The population of people who eventually receive a benefit is smaller or differently composed, because payment requires a covered event and satisfaction of benefit conditions. Each population answers a distinct research question.
Three questions organize the investigation. First, how do the actual offer, eligibility rules, price, delivery procedure, and prior experience influence enrollment? Second, which economic, institutional, and behavioral mechanisms shape renewal among existing insured participants? Third, which records and statistical methods distinguish a deliberate departure from an unobserved, temporary, or involuntary interruption of protection?
The empirical scope includes voluntary microinsurance and selected social health arrangements serving informal workers or low-income populations. Public programs enter as comparative evidence of participation behavior under alternative financing and eligibility designs. Any inference from social insurance to commercial insurance is conditional on the correspondence of the relevant participation mechanism. A comprehensive assessment of mandatory social protection, index construction, claims adjudication, or digital distribution is outside the central object; these domains enter only where they affect participation dynamics.
The evidence base is a purposive, comparative review of original studies and supervisory documentation rather than a systematic review or a meta-analysis. Search terms included microinsurance uptake, renewal, dropout, subsidy field experiment, and rainfall insurance repurchase. Priority was given to studies separating first enrollment from later membership, establishing temporal order, describing treatment assignment, or identifying product and population. Journal versions and working-paper versions of a single research project are treated as one family of evidence. Accessible abstracts establish only the findings actually reported in them. Institutional material is interpreted within its stated legal status. No private customer data, primary survey, or independent replication of a published regression was undertaken.
An original contribution of this study is the joint interpretation of denominators, contractual states, and causal estimands. These distinctions determine whether apparently inconsistent market outcomes actually contradict each other. They also establish the information necessary for an insurer, supervisor, or researcher to identify whether participation disappears because a product is unattractive, because payment fails, because eligibility is lost, or because an observation is missing.
Conceptual Foundations and Historical Development
The insurance relationship over time
Insurance transfers specified financial consequences of uncertain events under contractual or statutory terms. In an inclusive arrangement the contribution can be paid by an individual, a household, an employer, a distribution partner, a public authority, or a combination of these. The source of payment modifies the participation decision: a household offered a publicly funded entitlement faces an activation and use decision, whereas the purchaser of a voluntary unsubsidized contract faces an additional premium allocation decision. A renewal comparison must preserve this distinction.
Inclusive insurance has a broader market-development meaning than any single low-premium product class. In its 2012 application paper, the IAIS connects improved market access with supervisory safeguards and credible delivery of the promised benefits. For present purposes, inclusion denotes an opportunity for an underserved target group to obtain suitable risk transfer on understandable, financially and operationally accessible terms. The regulatory definition in a particular jurisdiction may differ from this analytical description.
The historical policy significance of participation follows from the transition between informal assistance and formal risk pooling. Community provision, mutual arrangements, commercial products, and publicly sponsored insurance can coexist. They may protect distinct risks, compete for scarce resources, or jointly support an insured household. An established community association may reduce the expense of learning about a formal product, but its assistance may also substitute for formal insurance. An evaluation must observe alternatives before attributing low demand solely to unfamiliarity.
An insurance contract with a fixed end date creates a different participation problem from a continuously renewable contract funded by regular installments. In the former, a qualifying new agreement is commonly required at the end of the term. In the latter, premium failure, cancellation, or noncompliance may interrupt coverage within an intended continuing relationship. Seasonal crop insurance offers repeated coverage opportunities tied to an agricultural calendar. Life insurance may reach its scheduled maturity, and a funeral contract may terminate after the insured death. Each event has its own legal and economic significance.
Participation states and the unit of analysis
Take-up measures acceptance or completed enrollment among an explicitly defined eligible or offered population over an observation window. Acceptance can be measured from a signed application or recorded affirmative consent; effective take-up requires successful activation. Retention describes continuing membership or coverage over an interval for a preexisting cohort. Renewal refers to a transition at a defined contractual opportunity. Repurchase refers to acquisition of cover in a subsequent period, potentially after an interruption or under altered terms. Persistency is the duration of uninterrupted contractual status under a specified rule. These endpoints need not coincide.
The unit of analysis must be specified before computing any rate. A single group policy may cover one hundred people; a household may contain several insured members; one person may hold multiple products. Counting policies rather than persons measures contract volume. Counting covered people measures breadth of protection. Counting insured household-months or person-days measures exposure actually covered. Aggregating units without deduplication may generate a misleading measure of inclusion.
An administrative record labeled active does not by itself establish effective protection. Eligibility under a policy, exclusions, waiting periods, and premium status may vary between risks. A person can have active accident cover while still in a waiting period for a sickness benefit. The relevant status should therefore be defined at the level of peril and observation date when the question concerns realized financial protection.
Consider a seasonal insurance program that offers drought protection only for crops planted by a specific deadline. A farmer may remain in the insurer's database between seasons, but coverage of the prior season has expired. Retention in the database, renewal of the insurance contract, and insured acreage in the new season refer to different quantities. A similar problem arises when an annual health membership card expires before the system updates the member's payment status.
Participation, risk transfer, and welfare
Risk-averse households may value insurance despite paying a premium above expected indemnity. The benefit derives from moving financial resources into adverse states of the world. Evaluation therefore requires a distinction between expected utility, expected claim payment, and realized net cash flow in a short observation period. The absence of a claim during a year can coexist with valuable ex ante protection. Conversely, recurrent routine benefits can be valued by households even when their insurance function is limited.
Let W denote initial resources, P the premium, L the loss, I the indemnity, and u a concave utility function. Under a one-period expected-utility benchmark, the household compares E[u(W-P-L+I)] with E[u(W-L)]. This conditional comparison is a familiar analytical benchmark; the present argument does not rely on it as a complete empirical description. It assumes that the relevant contract is understood, the premium can be paid, and the household expects the commitment to be honored. Transaction costs, borrowing constraints, uncertain claim payment, and alternative sources of assistance modify the feasible choice.
A household may prefer the insured allocation yet lack liquid funds on the date payment is due. A loss occurring before purchase or during an uncovered gap cannot ordinarily be transferred by buying coverage retrospectively. The sequencing of income, payment, and exposure is therefore a substantive component of an insurance decision. Low take-up is compatible with risk aversion, and high take-up is compatible with weak comprehension or low marginal cost under a subsidy.
A renewal is informative about behavior conditional on prior participation and an offered opportunity. It does not establish that the original terms were understood, that the policy remains suitable, or that continued participation has improved welfare. Those propositions require additional evidence about consent, covered events, claim performance, household costs, and counterfactual outcomes. Likewise, a voluntary departure from an unsuitable product may reflect a welfare-improving reallocation rather than a failure of inclusion.
Primary Institutional and Policy Documentation
Supervision and the contractual evidence base
The IAIS application guidance on inclusive markets (2012, particularly sections 1 and 2) places expansion of insurance access within a supervised market and emphasizes the reliability of promised protection. The publication is international supervisory guidance. Its application depends on the domestic legal and institutional framework. The IAIS Financial Inclusion Forum page identifies work revising the 2012 paper. The available evidence establishes the historical approval and the stated revision activity; it does not establish the content or effective date of a final successor.
A study of actual renewal in a jurisdiction should collect the licensing and product framework, the prevailing policy form, amendments, the applicable cancellation and reinstatement conditions, rules on group eligibility, and any statutory notice obligations. Research on a public scheme should additionally identify the legislation or program regulation, means-tested or categorical eligibility, contribution rules, subsidy funding, and the institutions responsible for providing services. These materials define the legal transition events that the administrative record is intended to represent.
The evidence necessary to classify interruption includes the agreed end date, grace period, contractual notice, the date on which premium payment was due, attempts at collection, and the effective date of termination. The relevant item is legally effective status, supported by an audit trail, rather than the day a database flag was changed. When regulations provide retroactive continuity after correction of an administrative failure, the exposure calculation must apply that rule consistently.
The present review does not establish current domestic legal provisions across all sampled countries. The empirical reports describe the studied programs as they operated during specified observation periods. Those descriptions are suitable for interpreting historical evidence of behavior. They do not constitute a legal opinion about present-day access, mandatory enrollment, or renewal rights.
Contract design and the information required at renewal
A renewal offer should be distinguishable from a mere solicitation. The record should identify the new premium, covered risks, benefit limits, exclusions, start date, and materially changed terms. A renewal may involve repricing or a different benefit schedule. In a product-replacement program the researcher must determine whether the new contract offers substantially continuous protection or constitutes a switch between materially different products.
At enrollment and renewal, a disclosure process may reduce uncertainty about payment obligations and the conditions of a claim. Its effectiveness must be measured through comprehension rather than document delivery alone. The participant should be able to identify the covered event, the major exclusions, the amount or calculation of benefits, the period of cover, the cost, and the means of requesting assistance or terminating recurring payment. Measuring recall independently of the sales interaction can distinguish demonstrated understanding from acknowledgment.
Consent has an economic and ethical role. An automatic continuation mechanism might preserve coverage after a temporary attention failure; without accessible notice and cancellation it might also charge for an unwanted product. A service provider should distinguish active renewal by choice, automatic renewal pursuant to valid authorization, sponsor-paid continuation, and enrollment imposed by a statutory requirement. The proportion of each mechanism is a relevant contextual variable in a persistency study.
The institutional meaning of participation indicators
A provider's policy count is principally an operational statistic. A supervisor may require a different measure of active coverage across regulated undertakings. A household survey estimates persons' reported access or ownership, which may diverge from legal activation. Program coverage data often reflect an administrative target population whose definition changes as eligibility records are updated.
A reproducible publication should specify for each measure its unit, numerator, denominator, treatment of multiple policies, observation dates, and the source of its status information. Insurer and survey estimates can be reconciled at an aggregate level without attempting to identify individual respondents where privacy rules prevent linkage. Any reconciliation should account for the difference between a respondent reporting insurance and the actual contractual status of each reported benefit.
Payment systems provide another distinct record. A successful transfer can fund one premium, multiple missed installments, fees, or an unrelated service. The payment reference must be matched to the correct policy and contractual period. A failed transfer is a payment event, not necessarily termination. Claims accounting similarly distinguishes claims submitted, assessed, accepted, rejected, withdrawn, and paid. These distinctions allow meaningful analysis of the relationship between customer experience and continued participation.
The most useful evaluation of inclusion consequently combines the outreach denominator, the coverage record, and evidence of whether covered events generate accessible protection. A measure of account ownership or access to a digital channel identifies one possible means of distribution, not the number of insured people. The conclusion follows from the difference between an access instrument and an insurance contract, without requiring an assumption about the current level of either indicator.
Initial Enrollment
Opportunity, price, and affordability
The first purchase can occur only when an eligible exposure unit is offered an obtainable contract. Let N be the eligible target population at a specified baseline, O the number receiving an actual offer, and A the number legally activated within the window. The ratios O/N and A/O describe outreach and offer conversion, while A/N describes initial coverage penetration. If every activated unit belongs to the offered and eligible populations and each is counted once, A/N equals (O/N)(A/O). This arithmetic shows why the same aggregate penetration can result from different bottlenecks.
For a hypothetical target of 10,000 households, an offer reaching 4,000 with 2,000 activations yields 50% conversion but 20% penetration. A second program reaching 8,000 with the same 2,000 activations yields 25% conversion and 20% penetration. The first merits investigation of outreach constraints; the second merits investigation of acceptance and activation constraints. Neither rate alone identifies a causal explanation. If the eligible population is underestimated by 20%, the apparent penetration overstates coverage: 2,000 divided by 8,000 is 25%, rather than 20%.
Premium affordability has several dimensions. The premium relative to annual income is a useful but incomplete statistic where earnings are irregular. A cash payment may be infeasible in a lean season although the premium is a small share of estimated annual resources. Instalment frequency, seasonal alignment, collection fees, transport expenses, and time costs determine the purchase burden. A subsidy relaxes the price constraint but may leave these other costs intact.
Price experiments should distinguish the intention-to-treat effect of an assigned offer from an estimated elasticity. A change in take-up from 20% to 30% under a subsidy is an increase of 10 percentage points, or 50% relative to baseline. It is not by itself an elasticity because the premium change and the definition of the relevant price are also required. Where the control price is zero or the subsidy creates zero net cost, log-price elasticity is undefined at that endpoint. Reporting the treatment contrast is then more appropriate.
Evidence from actual offers
The Indian rainfall-insurance experiments summarized by Cole and colleagues (2012) in Barriers to Household Risk Management demonstrate responsiveness to price alongside evidence consistent with liquidity, trust, and salience constraints. The source examined here is the original IMF working-paper summary. Its findings justify investigation of nonprice barriers even when a contract is made less expensive; they do not supply an elasticity for all microinsurance lines.
Thornton and colleagues (2010) randomized costs and enrollment arrangements for informal-sector health insurance in Managua. Their Nicaraguan randomized evaluation reports approximately 20% overall enrollment, with acquisition affected by financial cost and administrative convenience. Microfinance involvement was not uniformly beneficial in that setting. This original evidence challenges the assumption that an established distribution relationship automatically resolves the barriers to a new insurance purchase.
The two examples concern distinct economic goods. Rainfall-index protection triggers payment according to an external index, while participation in a health scheme can provide access to a specified network of services. A rainfall-index contract may impose basis risk, meaning that realized loss and index-triggered indemnity diverge. A health arrangement may offer nominal coverage whose usefulness depends on the availability and quality of covered providers. Comparative analysis must trace the relevant mechanism instead of transferring numerical treatment effects between products.
Trust, value, and household constraints
An insurance purchase exchanges current resources for a future contingent performance. Trust therefore has specific components: confidence that the insurer will remain capable of payment, that claim assessment will apply the contract fairly, that the intermediary accurately explains it, and that submitted money will reach the intended recipient. A household can trust a local intermediary but doubt the insurer, or trust the insurer while lacking confidence in an unfamiliar collection channel.
Trust should be studied with questions referring to identifiable stages of the transaction. A single general satisfaction score cannot isolate confidence in financial soundness from views of the claims process. A randomized intervention may credibly alter information about a provider; the exclusion restriction needed for a causal claim about trust itself requires that the intervention have no other influence on purchasing. An endorsement could change attention, social expectations, or effort as well as trust.
Risk preferences and product suitability introduce further heterogeneity. Households facing different losses or income opportunities may assign different value to the same contract. Access to savings and transfers can affect willingness to buy. The direction of the effect is theoretically ambiguous: savings may substitute for insurance of small losses, yet facilitate financing insurance against large losses. The interpretation depends on the covered risk, obligations, and the distribution of resources over the relevant period.
The household is not always a unitary decision maker. An individual bearing the loss may lack control over funds, identification documents, or a mobile account used for premium payment. A purchaser may choose coverage for another person. Survey data should therefore identify the covered person, payer, decision maker, and beneficiary where relevant. Subgroup analysis by income, gender, geography, disability, and occupation should start from explicit eligibility and decision-making roles, while observing privacy requirements and adequate sample size.
Activation and selection
Applications can fail before activation because of missing information, rejected identity verification, insufficient funds, ineligibility, or withdrawal of consent. Where possible, these events should be recorded separately. Reporting completed sales alone erases the stages where prospective customers are lost. It also prevents an assessment of whether some groups experience more obstacles than others.
Selection operates at multiple stages. The offered population may differ from eligible nonrecipients because distribution channels target certain communities. Those accepting may differ from those declining in their risk, wealth, and information. Those successfully activating may differ again because they can satisfy payment and documentation requirements. A sample of active customers therefore cannot be treated as representative of the entire underserved population.
A randomized offer provides an opportunity to estimate the causal effect of that offer on activation. Observational comparisons of those purchasing and those declining require additional assumptions about the sources of selection. Baseline covariates can improve description and precision, but adjustment alone cannot guarantee removal of unobserved differences. Where applicants self-select after random assignment, the original experimental groups should be retained in the principal analysis.
Renewal Persistence
Renewal as a conditional decision
At the expiration of a policy the household observes its own financial position, the product's prior performance, and the offered terms for the next period. A renewal can depend on an experience that was unavailable at first purchase. The household may have learned how the claim mechanism works, seen a neighboring participant receive assistance, experienced a delay, or discovered an exclusion. The decision also incorporates changes in risk exposure, alternative coverage, and ability to pay.
Let E_t denote the set of prior participants reaching a valid renewal opportunity in period t, and R_t the subset that renews during the declared window. The renewal ratio is |R_t|/|E_t| when the denominator is nonzero. Defining E_t requires an explicit treatment of deaths, contractual maturities, changes of residence, loss of eligibility, product withdrawal, and unresolved records. The researcher should also report the unconditional continuation of the original cohort, because exclusion of persons without a renewed offer otherwise conceals loss of market access.
Renewal of a subsidized contract at a subsidized price differs from renewal at a fully paid premium after support ends. An increase in initial enrollment may expand the composition of a cohort to households that were previously unable or unwilling to purchase. Subsequent renewal among this expanded cohort can be lower even when the subsidy has a positive effect on population-level coverage. The conditional renewal ratio is therefore an incomplete measure of the success of an acquisition intervention.
Contrasting health insurance experiences
In Nicaragua, the informal-sector insurance experiment found that fewer than 10% of original enrollees remained insured following expiration of the subsidy at the reported one-year horizon. The study also found a change in use toward covered facilities and a reduction in out-of-pocket spending that was smaller than premiums, without evidence of increased total utilization. This is evidence of modest persistence under the studied arrangement, rather than a general causal estimate of what all subsidies do to renewal.
Ethiopia's pilot produced a different longitudinal pattern. Mebratie and colleagues (2015), in Dropping out of Ethiopia’s scheme, reported an increase in membership from 41% to 48%, with 82% of prior members renewing and 25% of prior nonmembers joining. Knowledge and experience were associated with lower dropout. The figures describe the pilot and its sample. The observed relationships cannot be interpreted as randomized effects of knowledge or affordability.
Panda and colleagues (2016) examined three community schemes in rural India. Their longitudinal renewal study reported approximately 23% initial membership in 2011 and retention of 17% of that original cohort in 2013 in the discussion's specific cohort calculation. Receipt of benefits was positively associated with renewal; the authors identified possible roles for limited benefits and the claims process. The two-year endpoint differs from a single annual renewal ratio, and the association with payments is subject to selection.
The Ghana randomized subsidy experiment of Asuming, Kim, and Sim (2024) examined one-time subsidies of different levels at seven months and three years. All subsidy groups exhibited increased enrollment relative to the comparison at both horizons, while long-run health care utilization increased only in the partial-subsidy group. The article reports limited evidence of protection against out-of-pocket expenses. The three-year enrollment endpoint is a point-in-time treatment outcome rather than proof of uninterrupted three-year coverage.
Taken together, these studies show why an observed decline following a subsidy and a durable positive subsidy effect can coexist across settings. One result concerns retention among original enrollees; another concerns enrollment in randomized groups years later. Differences in the financial transition, benefit delivery, population, and endpoint can account for part of the variation. A causal claim about the relative importance of these cross-study differences would require harmonized data or a direct comparison under an appropriate design.
Experience, claims, and beliefs
Claims create opportunities for a participant to observe insurance performance. A successful indemnity payment can signal that the contract is effective, provide cash with which to renew, or alter the perceived probability of future payment. A failure to receive benefits can reflect a year without a covered event, an excluded loss, an index mismatch, incomplete documentation, a pending claim, or wrongful rejection. These states carry different implications for consumer experience and provider accountability.
In the Indian commercial rainfall-index market, Stein (2018) reports in Dynamics of Demand that customers receiving a payout were 9–22 percentage points more likely to buy again the following year. The author concludes that behavioral effects better account for the result than the tested alternatives, while repurchasing remains low overall. This discussion relies on the published abstract and treats the 2014 working-paper version as the same research family, not a second independent study.
A claim-conditioned comparison must establish how exposure to a loss determines both receipt of benefits and later purchasing. Loss severity can influence next-period wealth. A payout may mitigate that change, while location-specific weather influences future beliefs and outcomes. Even if recipients are more likely to renew, the descriptive contrast between recipients and nonrecipients cannot by itself isolate the causal effect of claim service. A randomized servicing intervention, a credible rule-based threshold design, or another defensible identification strategy may support that narrower question under explicit assumptions.
A participant who received no payment may undervalue risk transfer if evaluations depend excessively on recent realized returns. Yet the absence of renewal after no claim is not sufficient evidence of cognitive bias. The customer may have anticipated low value, learned about limited coverage, adopted a substitute, or faced an income shock. The observed outcome is consistent with multiple explanations, some behavioral and others fully compatible with realistic constrained choice.
Delavallade's (2017) Kolkata preventive-care experiment reported a 53% rise in willingness to pay for renewal following a randomized free checkup offer, a doubling of hypothetical renewal, and a ten-percentage-point increase in consultation of a qualified practitioner during subsequent illness. The reported renewal outcome is hypothetical. The study supports a distinction between experiencing service quality and actually maintaining insurance coverage, rather than an observed doubling of contract renewal.
Repurchase after expiry
The distinction between on-time renewal and subsequent repurchase is empirically important for voluntary life and accident coverage. Sane and Thomas (2020), examining repurchase among low-income households in three Indian states, report that the probability of repurchase is concentrated in the first two months after expiration and subsequently declines. A database that labels every expired policy permanently lost on its anniversary misses these later returns; a database counting eventual repurchases as uninterrupted renewals exaggerates continuity. An analysis should therefore report both the distribution of elapsed time to repurchase and the accumulated uncovered exposure. For benefits contingent on death or accident, events occurring in the interval cannot be assumed insured. Product terms and eligibility at reentry determine whether the restored coverage is equivalent to the original contract. A repeated-purchase sequence is thus a distinct empirical outcome from sustained protection.
Timing and payment continuity
Continuing payment depends on when obligations fall due relative to available funds. Short installment intervals can reduce the immediate amount required but increase the number of opportunities for collection failure. An annual payment may simplify administration while creating a large liquidity requirement. Flexible payment may accommodate variable income but complicate the treatment of unpaid premiums and benefit eligibility. Each design should be examined against the underlying contractual rules and administrative costs.
Consider two otherwise identical hypothetical policies. One requires twelve monthly contributions, with independent probability 0.98 of success on each due date; the other requires one annual contribution with probability 0.85 of success. If every monthly failure terminates coverage and reinstatement is impossible, the hypothetical probability of surviving all twelve payments is 0.98^12, approximately 78.5%. The annual counterpart is 85%. If failed monthly payments can be cured within a grace period, the calculation changes. The example isolates a frequency mechanism and does not estimate real failure probabilities or recommend one schedule.
Automatic continuation can remove an attention burden, provided the legal authority, advance notice, price disclosure, cancellation path, and payment method are valid. However, insufficient funds, expired payment credentials, a lost device, or a changed phone number can cause unauthorized gaps in protection. A collection system should identify failures, record notices, and support correction without silently imputing a customer preference.
An experiment that randomizes reminders should specify whether its intended endpoint is successful payment, timely renewal, or continued effective protection. A reminder may increase payment while leaving misunderstanding about terms unchanged. A valid consumer-oriented evaluation should include cancellation access, erroneous collections, complaints, and whether participants can distinguish renewal from acquiring a new product.
Payment channels also change the cost of acting on a renewal decision. Nsiah-Boateng et al. (2023) examined Ghana’s mobile contribution payment system using administrative data for 57,993 members and propensity-score matching. Their matched comparison estimated a 17.4 percentage point higher probability of renewal among mobile-channel users relative to office-channel users. Selection into the channel remains a possible explanation for some of the difference: matching adjusts for observed covariates and cannot guarantee exchangeability with respect to latent motivation, digital access, or income stability. The result supports investigating transaction friction rather than treating the point estimate as a universally transferable causal effect.
A subsequent qualitative investigation of mobile renewal by Ibrahim et al. (2025) involved 13 technical experts and 96 informal workers in 17 focus groups. Participants suggested reminders, savings support and automatic renewal while identifying trust, literacy, infrastructure and transparency concerns. These are design hypotheses rather than estimated effects. Together the two Ghana studies establish a rationale for testing payment-channel access, informed consent, reminder timing and reliable confirmation separately. A low-friction digital path should be paired with a practicable alternative for policyholders who cannot access it, and tests should record coverage continuity alongside completion of payment transactions.
Actuarial significance of selective continuation
For a portfolio containing exposure units i with expected covered cost q_i, selection is relevant whenever the mean of q_i among continuing insured units differs from the mean in the reference pool. Let r_i denote the probability of renewal under defined conditions. The expected mix of continuing exposures is weighted by r_i. Under suitable finite-population or large-sample assumptions, the ratio of expected aggregate cost to expected continuing exposures is sum(r_i q_i)/sum(r_i). This ratio is a planning approximation to the pool's mean cost, not the expectation of a random ratio in every finite realization.
If a higher expected-cost group renews more frequently, the pool can become more costly even when total membership declines. If healthier or lower-risk members are more able to pay than vulnerable members, the composition can change in the opposite direction. In either case aggregate renewal is inadequate for estimating risk selection. Risk classification must also respect applicable legal restrictions, the purpose of the inclusive arrangement, and appropriate treatment of personal data.
Consider a hypothetical one-term pool of 800 exposures with expected benefit cost 20 currency units per exposure and 200 with cost 100. Its initial expected cost is 36 per exposure. If 50% of the first group and 80% of the second renew, the expected cohort consists of 400 and 160 exposures. Total expected cost is 24,000 and mean expected cost is approximately 42.86. A decrease in headcount from 1,000 to 560 accompanies an increase of about 19.0% in expected cost per exposure. This is a compositional illustration with fixed risk classes, not evidence of realized adverse selection in a specific insurer.
Renewal also changes expense allocation. New enrollment can require up-front distribution costs, while subsequent servicing may be cheaper or more expensive depending on collection, claims, and administration. A portfolio evaluation must account for earned premiums over actual exposure periods, incurred claims rather than merely paid benefits, allocated acquisition and servicing expenses, reinsurance, capital needs, and subsidies. Persistency can improve scale while an inappropriate product remains unprofitable or harmful. Conversely, a program with social funding may be financially sustainable under its public commitments even if individual contributions do not cover all costs.
Dropout Identification
Defining events through evidence
Dropout is a departure from a defined participation or coverage state before a scientific endpoint. Its operational definition depends on the product and inquiry. Failure to renew by the contractual deadline might indicate a lapse when the opportunity was valid. A participant dying while insured represents a terminating insured event or loss of eligibility, not a choice to discontinue. Product withdrawal and provider exit terminate opportunities independently of customer preferences.
The event record should distinguish voluntary nonrenewal, cancellation before expiry, payment lapse, loss of eligibility, group termination, provider exit, transfer to replacement coverage, death, scheduled maturity, and missing observation. Several events may occur together. A prespecified priority rule, or a richer event history allowing multiple labels, can maintain interpretability without falsely identifying a unique cause of departure. Legal status and self-reported reason should be retained as separate fields.
A gap in observation should not be coded as dropout solely because an interview is missing. The same person may be continuously insured, insured with another provider, or uninsured. Right censoring at the observation boundary records the time through which status is known. When censoring depends on unobserved participation and risk, naive complete-case analysis and ordinary independent-censoring estimators may be biased.
Temporarily inactive members who later reactivate need their own history. If a person is uninsured for two months and renews subsequently, the final status is active but uninterrupted coverage has failed. A strictly cohort-based continuous-survival outcome should count the interruption; a point-prevalence measure can count the person as active at the endpoint. Both are legitimate when their definitions are explicit.
Denominators, competing events, and missing status
Let an initial cohort contain 1,000 participants reaching the end of a study period. Suppose 600 renew, 250 decline or lapse, 100 lose eligibility, and 50 remain unresolved. These are hypothetical counts. The confirmed observed continuation of the original cohort is 60%. Excluding known loss of eligibility but retaining unresolved cases yields a confirmed renewal proportion of 600/900, or 66.7%, among those not known to have lost eligibility. Restricting the denominator to the 850 known eligible resolved cases gives 600/850, or 70.6%.
The eligible denominator is uncertain if unresolved cases may have lost eligibility. An explicit scenario treating all 50 unresolved as eligible gives bounds of 600/900 and 650/900, or 66.7% to 72.2%, depending on their renewal status. These are logical scenario bounds, not confidence intervals. If some unresolved people are ineligible, the population definition and denominator change. The analysis should report which scenario and assumptions are used, rather than assigning the 50 to lapse by default.
Risk sets for survival analysis include participants under observation and still eligible for the event immediately before each time point. Under independent censoring conditional on relevant variables, a Kaplan–Meier estimator can describe time to first interruption. Its product-limit form multiplies terms 1-d_j/n_j over event times j, where d_j counts interruptions and n_j is the risk set. For a particular exit cause, other exits are competing events. Treating them as ordinary censoring and estimating one minus the cause-specific Kaplan–Meier curve generally does not recover the observed cumulative probability of that cause. A competing-risks cumulative incidence function weights each cause-specific hazard by survival from all causes.
In discrete annual terms, suppose 80% of a fixed cohort renews at the first opportunity and 75% of those renewing also renew at the second, with no competing exits and no reinstatement. Continuous two-term retention is 0.80 × 0.75 = 60%. Averaging the two percentages gives 77.5% and answers a different question. If later renewal denominators include newcomers or returning participants, the multiplication ceases to describe the original continuously covered cohort.
Cohort accounting and coverage exposure
A portfolio can grow even while a majority of prior participants leave. In a simplified closed-status accounting period, closing active membership equals opening active membership plus first activations plus returns minus exits, after reconciling transfers and duplicate records. If 500 opening members leave and 700 new members activate, a base of 1,000 grows to 1,200 despite 50% retention of the opening cohort. The increased closing stock cannot establish improved persistence.
Exposure-based measurements register the accumulated time insured. In a hypothetical year of 365 days, two eligible persons would contribute 730 person-days. If one is covered for the whole year and the other for 180 days, the covered proportion of their combined eligible time is 545/730, approximately 74.7%. A count of people ever covered during the year would report both as reached, while the end-of-year active proportion could differ depending on the distribution of their covered days. This calculation clarifies the time dimension of protection.
Periods of partial protection require peril-specific accounting. An exposure may be insured against accidental death for all covered days but have restricted sickness coverage under another contractual provision. A researcher should define a benefit-weighted measure only when the weights and desired welfare interpretation are stated. Simply combining unequal benefits into a single binary flag can erase substantial differences in coverage adequacy.
The Ethiopian figures in Mebratie and colleagues’ study and the Indian figures in Panda and colleagues’ analysis demonstrate the need to distinguish original cohorts, latest membership, and intervals. Their different retention estimates do not order overall welfare or actuarial sustainability. Meaningful comparison requires like periods, benefits, eligibility rules, and consistent treatment of exit.
Data collection and causal identification
A longitudinal study should record a stable pseudonymous exposure identifier, eligibility start and end dates, offer and consent timestamps, product and contract version, premium due dates, payment outcomes, activation dates, termination and reinstatement events, claim event dates, and relevant characteristics measured before exposure to interventions. Privacy rules and lawful access should govern collection and linkage. Personal identifiers unnecessary to the scientific task should remain outside analytical datasets.
Sampling must include nonparticipants as well as current and former members when the objective is target-market penetration. Household sampling can use survey weights reflecting unequal selection probabilities and nonresponse adjustments. Clustering by village, employer, distribution channel, or household affects sampling uncertainty and should be reflected in variance estimation when relevant. Administrative records of all current policies may eliminate one component of sampling error but retain measurement errors and incomplete eligibility denominators.
Attrition from a survey must be distinguished from attrition from insurance. Report interview completion by treatment and baseline status. Compare observed characteristics of respondents and nonrespondents. If missing outcomes plausibly depend on observed variables, weighting or multiple imputation can be investigated under a clearly stated missingness assumption. Sensitivity analysis for outcomes missing not at random remains necessary when the assumption is uncertain.
Randomized interventions should be assessed by intention to treat: the difference in the probability of the prespecified outcome among the groups originally assigned to the intervention and comparison. An effect of realized participation can be estimated using assignment as an instrument only under appropriate relevance, independence, exclusion, and monotonicity conditions. The resulting local effect applies to the relevant compliers, rather than necessarily to every eligible household.
A regression estimated only on initial purchasers can introduce post-treatment selection when the experimental intervention itself affected enrollment. The subsequent comparison may contrast people of different underlying risk, financial means, or willingness to participate. A credible analysis can report the unconditional probability of initial enrollment followed by renewal within each randomized arm and separately describe conditional renewal as a selected-cohort outcome. This distinction is necessary even in a well-executed experiment.
For example, suppose in a hypothetical trial 200 of 1,000 control households enroll and 100 of those renew; 400 of 1,000 treated households enroll and 160 renew. Conditional renewal falls from 50% to 40%, while population-level enrollment followed by renewal rises from 10% to 16%. With valid random assignment and complete follow-up, the latter difference of six percentage points estimates the effect of the offer on the joint outcome. The conditional ten-point decline alone does not establish that the offer reduced the propensity of otherwise comparable individuals to renew.
Identifying reasons without collapsing distinct mechanisms
Qualitative interviews can identify practical barriers absent from transaction data. An interviewer should distinguish the immediate event leading to termination from the respondent's longer-term explanation. A missed payment might result from a banking failure, low liquidity, a conscious decision to stop, or lack of awareness of the deadline. Respondents can report more than one explanation and may reconstruct motives after the event.
The distribution of reported reasons among reachable former members does not automatically describe all former members. Some who move or lose contact may be difficult to reach. An operational investigation should pair a stratified follow-up sample with objective records, document response rates and the time elapsed after exit, and identify cases where the administrative and reported causes differ. Disagreement is valuable evidence of measurement or service problems, not always grounds to discard one record.
A causal conclusion that changing an operational process prevents dropout requires variation in that process independent of competing causes, or transparent identification assumptions. Randomizing the timing or channel of valid reminders may estimate their effect on renewal. A policy change affecting a particular eligibility group might permit a quasi-experimental design if the comparison groups and timing support credible counterfactual inference. Reporting any association as causal without these conditions risks inefficient intervention and erroneous attribution of responsibility.
Integrated Scientific Synthesis and Research Agenda
Comparative interpretation of evidence
The source corpus contains several distinct estimands. The rainfall-insurance acquisition experiments estimate responses to offers or interventions. The Nicaragua and Ghana studies offer randomized variation in participation incentives and follow-up outcomes. The Ethiopia and rural India studies describe longitudinal membership and factors associated with it. The BASIX study examines subsequent purchase in relation to payout experience. The Kolkata experiment includes stated or hypothetical renewal demand. These findings are related but cannot be pooled as if they were identical renewal outcomes.
| Evidence setting | Observed endpoint | Design and interpretive boundary |
|---|---|---|
| Rural India rainfall insurance | Initial purchase | Randomized acquisition experiments; linked working-paper summary covers price and nonprice frictions |
| Managua informal-sector health cover | Purchase and one-year membership | Randomized offer and enrollment procedure; retention conditional on enrollment |
| Ethiopia community-based health insurance | Membership across two survey waves | Longitudinal pilot; correlates of renewal need not be causal |
| Bihar and Uttar Pradesh community-based insurance | Original-cohort retention through 2013 | Longitudinal cohort analysis; benefit receipt association potentially selected |
| Ghana social health insurance | Enrollment at seven months and three years | Randomized subsidy assignment; point-in-time enrollment does not establish continuity |
| BASIX rainfall-index insurance | Next-period purchase after payment experience | Published abstract supports reported range; working paper is same research family |
| Kolkata health microinsurance | Hypothetical renewal and care seeking | Randomized service offer; willingness is not completed renewal |
The table is an analytical classification rather than a pooled quantitative estimate. Its most important implication is that variation in the endpoint can create apparent contradictions. Another is that service quality may enter the value of insurance through different channels in health and agricultural products. Neither implication requires a novel named theory.
A more complete international comparison would incorporate further research on life and accident microinsurance, informal risk-sharing groups, funeral products, bundled credit cover, digital premium collection, and compulsory or automatically renewed arrangements. Those subjects remain empirical extensions of the scope of the actual corpus. The present review's geographical and product composition limits its external validity; it does not provide a global mean renewal rate.
Interpreting sustainable inclusion
The commercial objective of acquiring customers, the actuarial objective of maintaining a sufficiently diversified pool, the regulatory objective of fair treatment, and the household objective of managing financial risk overlap in important cases. They can also diverge. A contract with a low premium may facilitate access but transfer only a limited portion of a catastrophic loss. An insurer may improve retention through default settings while weakening the customer's control of repeated payments. A generous subsidy can increase enrollment while leaving service capacity inadequate.
Sustainable participation should be interpreted relative to the durability of the financing arrangement. For an unsubsidized product, a long-term evaluation concerns premium adequacy, expected claims, operational cost, and continuity of suitable cover. For a public scheme, financing includes legitimate government contributions and risk equalization where applicable. A program's ongoing dependence on subsidies is not by itself proof of failure; the scientific question concerns the stability, targeting, and fiscal consequences of the financing commitment alongside the achieved protection.
The relationship between claims and continued participation also requires care. A commercial portfolio with many denied claims may record low payouts and positive short-run accounting performance. A participant who never submits a claim may remain unaware of exclusions. A high renewal rate under such conditions cannot substitute for claims and consumer-outcome analysis. Conversely, high payouts resulting from legitimate severe losses may coincide with a difficult underwriting result while fulfilling the central purpose of insurance. The insurer's performance should be examined through both actuarial and beneficiary outcomes.
When participation is selective, interventions affecting enrollment and renewal can change expected costs. The price that supports a balanced portfolio after an acquisition campaign may differ from the price estimated beforehand. Evaluating this effect requires data on the evolving risk mix and credible estimation of expected incurred claims. It does not justify individual discrimination, concealment of exclusions, or changes in terms inconsistent with applicable legal obligations.
Evaluation design across the policy lifecycle
A practical research program can follow all eligible exposure units from the offer onward, including those never activated. Prespecify the time window for each transition, whether reinstatement counts as renewal, and how changes of product and eligibility will be handled. The study should align follow-up with the actual premium period and loss calendar. Health schemes may warrant observation of service use and financial burden; index insurance requires matching the relevant peril and indemnity trigger to actual loss experience where feasible.
Targeted interventions can be assigned to different stages: offer delivery, explanation of terms, premium financing, payment reminders, service assistance, or claim documentation. The study should avoid combining all interventions into one arm when the research question concerns their separate mechanisms. A factorial experiment, when operationally and ethically feasible, can identify interactions under the relevant assumptions but requires attention to multiplicity and sample size.
Investigators should publish a protocol stating the primary endpoint and analysis plan before examining outcomes when feasible, report intervention assignment and actual implementation, identify deviations, and distinguish confirmatory results from exploratory subgroup findings. Wherever local rules permit, deidentified data dictionaries, reproducible analysis code, and aggregate results can support scientific reassessment without disclosing vulnerable participants' identities.
The evaluation must also monitor consumer harms. Consequences may include mistaken deductions, double coverage, gaps after a failed payment, delayed claims, inadequate notice of changed terms, or exclusion of those unable to meet technological requirements. An intervention that raises the number of renewals but also increases unauthorized payments has a different welfare interpretation from one that expands voluntarily maintained effective protection. A narrow single-metric incentive creates risk of confusing sales performance with customer value.
Research questions that remain open
Further evidence is needed to quantify whether the same affordability intervention has different long-term consequences across seasonal agricultural products, fixed-term health insurance, funeral cover, and continuous accident insurance. The comparative corpus identifies plausible mechanisms and heterogeneous observations, rather than identifying causal effects of product type itself.
Longitudinal work should also separate prior experience with valid claims from the consequences of covered service quality, social information, payout timing, and changes in household resources. These mechanisms may mediate the observed relationship between prior benefits and renewal. Their separation requires credible temporal ordering and research designs capable of distinguishing direct and indirect effects.
Greater attention is warranted to participants who remain continuously covered despite financial constraints and to those leaving for an adequate replacement product. Their outcomes reveal successful forms of risk management that may otherwise be classified incorrectly. Equity assessments should include target-market nonparticipants, discontinuation reasons, and benefit adequacy, not simply compare the characteristics of current policyholders.
Finally, cross-country insurance research requires transparent comparability metadata. A common reporting template for the observation window, unit of analysis, offer population, contractual state, contribution regime, benefit type, competing events, and missing-data treatment would facilitate later synthesis. Such metadata are ordinary scientific and operational documentation, not a proposal for a new regulatory standard.
Conclusion
Initial enrollment, contract renewal, and uninterrupted coverage are separate measurable aspects of participation in inclusive insurance. Their interpretation requires the actual eligibility population, offer process, contractual term, financial obligations, and the status of each exposure. A count of people reaching one stage of the process cannot substitute for an estimate of progression through later stages.
The evidence reviewed establishes that premium incentives and administrative arrangements can influence first enrollment; it also shows that renewal and later membership vary across studied arrangements. The observed effects of service and claim experience deserve investigation with methods able to distinguish learning, liquidity, selection, and contractual performance. Existing experimental and observational evidence supports differentiated explanations while leaving the global magnitudes and causal interactions unresolved.
The actuarial consequences of selective participation arise from the changing risk composition and the length of actual exposure. The consumer consequences depend on the validity and suitability of continuing cover, the availability of covered services, and the fulfillment of benefit obligations. Financial sustainability and household financial protection therefore require investigation through different but connected measurements.
The original analytical synthesis is a reconciliation of the temporal, contractual, statistical, and welfare dimensions of participation. Cohort accounting, competing-event analysis, explicit denominator sensitivity, and careful interpretation of randomized assignments make reported patterns more informative. The hypothetical calculations establish the logical implications of measurement choices and risk selection; they do not represent observations of any actual insurer.
Inclusive insurance realizes its protective purpose when eligible households can obtain appropriate insurance under fair terms, maintain it when it continues to meet their needs, and receive promised protection when an insured event occurs. Research on participation should make visible the conditions under which every stage succeeds or fails. This conclusion defines the empirical work required to assess the durability of inclusion while preserving the distinction between observed insurance behavior and demonstrated welfare improvement.
References
International Association of Insurance Supervisors (IAIS). (2012). Application paper on regulation and supervision supporting inclusive insurance markets. Approved October 9, 2012. Official document
International Association of Insurance Supervisors (IAIS). (n.d.). Financial inclusion. Retrieved November 18, 2026. Institutional page
Thornton, R. L., Hatt, L. E., Field, E. M., Islam, M., Solís Diaz, F., & González, M. A. (2010). Social security health insurance for the informal sector in Nicaragua: A randomized evaluation. Health Economics, 19(Suppl.), 181–206. DOI: 10.1002/hec.1635. Article text
Panda, P., Chakraborty, A., Raza, W., & Bedi, A. S. (2016). Renewing membership in three community-based health insurance schemes in rural India. Health Policy and Planning, 31(10), 1433–1444. DOI: 10.1093/heapol/czw090. Article text
Mebratie, A. D., Sparrow, R., Yilma, Z., Alemu, G., & Bedi, A. S. (2015). Dropping out of Ethiopia’s community-based health insurance scheme. Health Policy and Planning, 30(10), 1296–1306. DOI: 10.1093/heapol/czu142. Article text
Asuming, P. O., Kim, H. B., & Sim, A. (2024). Selection and behavioral responses of health insurance subsidies in the long run: Evidence from a field experiment in Ghana. Health Economics, 33(5), 992–1032. DOI: 10.1002/hec.4797. Published article PDF
Stein, D. (2018). Dynamics of demand for rainfall index insurance: Evidence from a commercial product in India. The World Bank Economic Review, 32(3), 692–708. DOI: 10.1093/wber/lhw045. Publisher abstract
Delavallade, C. (2017). Quality health care and willingness to pay for health insurance retention: A randomized experiment in Kolkata slums. Health Economics, 26(5), 619–638. DOI: 10.1002/hec.3337. PubMed record
Cole, S., Giné, X., Tobacman, J., Topalova, P., Townsend, R., & Vickery, J. (2012). Barriers to household risk management: Evidence from India. IMF Working Paper 12/195. Related peer-reviewed version: American Economic Journal: Applied Economics, 5(1), 104–135 (2013), DOI: 10.1257/app.5.1.104. IMF working paper
Stein, D. (2014). Dynamics of demand for rainfall index insurance: Evidence from a commercial product in India. World Bank Policy Research Working Paper 7035. World Bank working paper
Sane, R., & Thomas, S. (2020). From participation to repurchase: Low income households and micro-insurance. Journal of Risk and Insurance, 87(3), 783–814. DOI: 10.1111/jori.12288. Article record
Nsiah-Boateng, E., Musah, M., Akuamoah, C. D., Asenso-Boadi, F., Andoh-Adjei, F.-X., & Boye, B. O. (2023). Effect of mobile phone-based health insurance contribution payment system on retention of coverage in the National Health Insurance Scheme in Ghana: An evaluation study. BMC Health Services Research, 23, 239. DOI: 10.1186/s12913-023-09236-7. Article record
Ibrahim, F., Neumann, A., Mensah, K. A., Opoku, D., Kwawukume, M. A., Nübler, L., Siegel, M., Owusu-Dabo, E., Quentin, W., Struckmann, V., & Waitzberg, R. (2025). Is mobile renewal enough? A qualitative study exploring stakeholders’ perspectives on mobile phone-based add-ons for national health insurance uptake in Ghana. BMC Public Health, 25, 1700. DOI: 10.1186/s12889-025-22857-3. Article record