Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Insurance Financial Inclusion Channel Design Experiments

Abstract. Insurance financial inclusion depends on the institutional arrangements that connect people facing risk to recognizable, affordable and enforceable protection. This monograph examines distribution-channel design as an experimental object, synthesizing field evidence from agricultural index insurance, voluntary social health insurance and subsidized private health plan enrollment in India, China, Ethiopia, Ghana, Kenya, Burkina Faso, Nicaragua, the Philippines, Indonesia and the United States. It distinguishes the assignment of an institutional channel from the use of that channel, completed enrollment, continuous coverage, payment of eligible claims and household welfare. A comparative analysis of randomized experiments and their follow-up studies shows that price relief, administrative assistance, social learning and provider availability operate through different mechanisms and may be complementary. Apparent improvements in sales can coexist with failed registration, rapid lapse, unserved claims or unchanged use of care. The study specifies cluster-randomized and factorial designs, estimands, administrative measurement, comparative cost accounting and conduct safeguards appropriate to agents, cooperative groups, banks, employers and assisted digital channels. Its original contribution is the integration of causal identification with the insurance lifecycle and the allocation of institutional responsibility. The evidence supports context-dependent channel decisions, rather than a universal ranking of channels. It does not report a new field experiment or a pooled cross-country treatment effect.

Keywords: Insurance distribution; financial inclusion; insurance intermediaries; field experiments; effective coverage; cluster randomization; administrative burden; digital insurance; claims; consumer protection; external validity.

Introduction

The economic function of insurance is the transfer or pooling of defined risks through an enforceable arrangement that specifies coverage, financing and obligations. The distribution function connects this arrangement to the person or organization exposed to risk. For an underserved household, the difference between an available insurance product and accessible protection may consist of a day of travel, a missing identity record, an unaffordable premium at the wrong point in the agricultural season, an inaccessible language, uncertainty about the intermediary's authority, or the absence of practical assistance after a loss. Each is an institutional property of the transaction as experienced by the household.

The question of this research is how alternative channel configurations change effective insured protection among eligible underserved populations. Its primary concern is the experimentally identifiable contribution of distribution institutions. Individual demand, risk aversion, framing, pricing and behavioral elicitation enter when they interact with the channel or threaten identification of its independent effect. The product class need not be designated microinsurance: a conventional health, life, property, agricultural or income protection contract may serve a low-income or otherwise excluded customer. A social insurance program may offer useful institutional evidence when its enrollment and service processes resemble those of an insurance scheme; its financing, entitlement rules and public obligations remain distinct from private underwriting.

Research questions follow the insurance lifecycle. Which configurations increase access to suitable offers? Which remove administrative barriers rather than merely increase attempted transactions? How can their effects be separated from premiums, subsidies and information? Which improve coverage over time, valid claims settlement and the capacity to withstand adverse events? The answer matters for both market participants and authorities pursuing access without sacrificing fair treatment or solvency.

This is a structured comparative narrative synthesis, not a systematic review with a registered search protocol. Source discovery used targeted searches for original experiments of institutional distribution, assisted registration, social channels, remote payments, insurance retention and welfare; primary publications were preferred, and official standards supplied institutional context. Published author abstracts, indexed methods and results, accessible institutional records and official text were checked at the claim level; source-specific access boundaries appear in the evidence ledger. Findings from the same experiment are not counted as independent replications. The selection includes unsuccessful interventions, missing endpoint evidence and differences between populations. No claim is made to exhaustive literature retrieval, raw-data reproduction, complete national legal verification, a new causal estimate, or study-level risk-of-bias adjudication based on inaccessible appendices. Quantitative examples identified as hypothetical are calculations for design, not findings.

Insurance Financial Inclusion and Channel Institutions

The unit of inclusion

The natural population denominator is people eligible for a defined insurance protection during a specified interval, including those whom the existing system does not reach. Eligibility should be determined independently of the intervention. For elective retail insurance, this may require a geographic sampling frame with a predeclared risk, age and product eligibility rule. For a public scheme, the relevant legal entitlement and contribution rules define the initial frame. In both cases an inaccessible offer, inability to supply required records and inability to pay are separate variables.

Insurance inclusion is multidimensional. A person may be eligible for several policies, possess a payment account and still lack the cover needed for a specific hazard. Conversely, a household may receive protection through a group master contract without possessing an individual policy document. The unit of observation for inclusion is therefore a person, household or enterprise exposed to specified risk, linked to a contract, coverage interval and legal rights. Counting policy certificates measures issuance. Counting distinct protected individuals with in-force coverage measures reach. Counting serviceable insured exposures addresses whether the protection can actually be invoked.

The World Bank's Global Findex 2025 describes an adult survey for 2024 covering approximately 148,000 respondents across 141 economies, with information about accounts and digital connectivity (World Bank 2025). Such measures inform sampling and access constraints. A mobile account provides a possible payment rail, not an insurance entitlement. An internet-enabled device provides a possible communication rail, not proof of understanding or control over credentials. Market penetration expressed as premiums divided by gross domestic product similarly reflects expenditure and product composition rather than the fraction of underserved persons adequately protected.

The distribution relationship

In insurance, a channel performs a set of tasks: identifying potential insured persons, eliciting their needs, communicating terms, enabling application and payment, effecting binding, delivering policy records, helping with changes and renewals, and connecting insured persons to claims and complaints services. Organizational arrangements may split these tasks across firms. A bank may collect payments while a licensed insurer accepts risk; a cooperative may convene members while a separate agent explains terms; an employer may administer enrollment but an insurer determines claims. A phone application can be merely an interface over the same institutional chain.

Agents combine personal communication with knowledge of local circumstances. An independent broker may offer comparison across contracts, whereas a tied agent's menu can be limited to a particular carrier. Cooperative associations and community groups may supply existing contact networks, collective trust and low-cost meeting arrangements. Their internal rules can also exclude people, allocate assistance unevenly, or create pressures to join. A bank or microfinance organization has a financial relationship with customers and a payment mechanism but may conflate borrowing with insurance, or leave noncustomers outside the population reached.

Employer enrollment aggregates people at a place of work, making recurrent contributions feasible; its limitations include job loss, informal employment and the possibility that dependents or contingent workers do not qualify. Mobile-network and digital platforms can reduce the marginal cost of interaction and support remote payment. Assisted digital systems add a human representative for language, identity verification, error correction or claims support. These are hypotheses about institutional capabilities. Their relative success is a research question that varies by product and context.

Evidence on institutional channels

In Nicaragua, Thornton et al. (2010) varied premium costs and enrollment sites in a voluntary public health insurance scheme for informal workers. Their randomized enrollment evaluation reported roughly 20% overall participation and a slightly adverse enrollment result associated with microfinance institution involvement. An intermediary already familiar with credit customers need not be a successful insurance outlet. Its reputation, contractual relationship with the insurer, proximity and role in completing paperwork are distinct characteristics. The study establishes a boundary to the assertion that any existing financial network automatically expands insurance coverage.

In Ethiopia, Dercon et al.'s (2012) informal risk-sharing group experiment marketed rainfall protection to existing groups and varied the training emphasis on sharing residual basis risk. The reported uptake result suggests that a formal contract can complement locally organized risk-sharing. Group enrollment is especially relevant when an index payment does not match an individual household's actual loss. The experiment concerns information within a group-based arrangement, rather than a randomized comparison of all group and independent-agent architectures.

Cai, de Janvry and Sadoulet's (2015) weather insurance network experiment in rural China assigned intensive product information to selected farmers. The authors estimate that the added influence of one treated friend on an untreated farmer's take-up corresponded to about a 13% premium reduction; they identify diffusion of knowledge rather than peer purchase decisions as the central explanation. Social networks can thus be part of the channel's mechanism, with exposure extending beyond a direct sales contact. This is different from establishing a cooperative's legal authority to distribute insurance.

The analytical distinction is between a channel's institutional attributes and its surface identity. A branded mobile channel may rely on an agent to register consumers; a group-based channel may rely on a digital payment platform; a bank branch may be a point of contact while claims are administered elsewhere. The treatment has to specify the division of functions and the incremental resources introduced. Otherwise a comparison of named channels is an uninterpretable comparison of unrecorded bundles.

The credit-linked distribution contrast is sharpened by Bauchet, Damon, and Larsen's randomized microfinance bundling study. In Colombia, a voluntary crop microinsurance offer accompanying a loan application was randomized against an offer presented later. According to the publisher's study abstract, the delay did not significantly change overall insurance take-up or coverage amount, although crop-specific responses differed and comprehension remained low. This directly identifies the timing of an offer within a microfinance relationship. It does not estimate the average causal effect of mainstream bank branches compared with independent insurance intermediaries, nor the effect of the intervention on later insured claims.

Employer-linked insurance requires distinguishing plan presentation from enrollment under an actual administrative system. Atanasov and Baker's employer health-plan choice study combines a survey after a new high-deductible employer plan was introduced with a framed choice experiment. The publisher and PubMed abstracts report that only 3% chose the new plan in the surveyed setting, that additional financial information by itself failed to increase selection in the framed experiment, and that perceived access to physicians and the composition of the comparison menu affected choices. The framed experimental outcome is not an estimate of the effect of employer distribution versus brokerage on actual insured population coverage or claims. Its contribution here is to identify menu clarity and perceived service access as measurable components of an employer-linked channel.

The ILO study of banking and retail correspondents, by Leach, Menon, and Ncube, provides a complementary descriptive institutional examination of four microinsurance cases. This case-study record can guide measurement of the reach and service economics of a bank or retail partnership. Its purposive descriptive design does not identify causal channel treatment effects and cannot substitute for independently ascertained claims outcomes.

Channel reach and the publicly relevant margin

A distribution experiment is inclusion-oriented when it begins with the exclusion problem, not a list of existing clients who are convenient to randomize. In a rural district the sampling frame may cover households within travel-distance bands, including people who lack identity documents or mobile money accounts. In a labor-market context it can include seasonal workers and those working outside participating employers. For a health marketplace, the relevant stages may be nonapplicants, applicants without plan selection, and enrollees facing renewal. Eligibility and contactability should be recorded separately.

Within an eligible cohort, increasing enrollments among already-served buyers can replace existing cover rather than extend protection. An agent moving a customer from one insurer to another changes market share. It expands inclusion only to the degree that distinct insured exposure, continuity or appropriateness actually improves. A delivery channel can improve one without changing the others. An evaluation should capture prior cover and transitions, including joint coverage, displacement, reinstatement and uninsured intervals.

Access Barriers, Product Eligibility and Service Costs

Identifying the barrier

The household's economic burden includes premium expenditure, required contributions, travel and waiting time, communication costs, application effort and anticipated claim friction. A premium subsidy changes the payment price. An agent visit can change information, travel costs and trust simultaneously. Application assistance changes the resources needed to complete a transaction; a relaxed documentary rule changes eligibility or verification. The causal question is which component has been deliberately varied.

Cole et al. (2013) found several consequential nonprice frictions in Indian rainfall insurance experiments, including constrained liquidity, trust and limited salience. Price sensitivity in such a setting cannot by itself reveal a household's valuation of the risk protection. A person unable to advance cash before a crop season has a different problem from a person who expects a contract not to pay. Design implications may include payment timing, credible explanation or assistance, each with a distinct experimental contrast.

The Philippines provides a sharp distinction between information-price relief and administrative support. Capuno et al.'s (2016) voluntary health insurance experiment assigned municipalities to an information kit and a temporary half-premium subsidy; the reported initial enrollment effect was three percentage points, statistically indistinguishable from zero at the conventional 5% level (p = 0.11). A later assistance treatment for still-eligible nonenrollees produced a much larger increase of 29 percentage points (p < 0.01). The later intervention bundled completion, submission and delivery of administrative documents. Its result applies to that selected, eligible stage and is not a direct contrast with the first-stage population. Removing trips and handling paperwork was potentially valuable in the particular institutional setting.

Information can also fail to alter uptake. The randomized home-visit intervention described by Bocoum et al. (2019) in rural Burkina Faso modestly improved insurance knowledge without detecting increased membership. The scientific lesson is that knowledge and purchase behavior are distinct endpoints. Even an effective explanation may be followed by an informed decision not to purchase, financial constraints, or a lack of confidence in available care. A null result for enrollment does not establish that information is socially valueless.

Administrative capacity and verification

Insurance application often includes multiple records: names, dates, eligibility evidence, dependent relationships, address or location, contact method, payment authorization and acceptance of contract terms. In an assisted channel a representative can help supply or reconcile these records, but may have no authority to alter a registry. Under fragmented administration, each failed procedure can become a point of exclusion.

Banerjee et al. (2021) examined Indonesia's national health insurance through a randomized study involving approximately 6,000 households. Assistance increased attempted enrollment substantially more than it increased completion; more than half of those attempting enrollment did not succeed. This result identifies the practical distinction between motivating demand and processing applications. The same study reports that temporary subsidies attracted lower-cost enrollees in part by reducing strategic timing. Price relief can therefore change risk-pool composition as well as coverage.

A field experiment should log the reasons for failed applications before the insurer rejects them permanently. Examples include missing national records, incompatible identifiers, inability to verify dependent relationships, failure of payment confirmation, insufficient staff authority or an unexplained system error. Distinguish preventable operational failures from legal ineligibility. A design that recruits previously excluded individuals but cannot complete their enrollment requires additional infrastructure, not a more aggressive sales script.

A two-sided capacity experiment can randomize front-end help and separately vary processing support across administrative offices. Consumers assigned to usual care must retain existing rights and legally required support; the treatment adds a permissible improvement. Administrative office congestion can create interference: successful applicants in the intervention arm can delay control-group applications processed by the same office. Randomization should follow the operational service boundary or explicitly model such exposure.

Economic cost of channel inclusion

A distribution institution faces costs for recruitment, sales training, travel, verification, communications, payment collection, data systems, customer support, audit and complaint management. Fixed costs should be allocated over a defined horizon and incremental volumes. A channel may lower insurer acquisition expense by requiring unpaid time from households or community leaders. A social evaluation must recognize those resource costs even when they do not appear in the insurer's expense line.

Define the incremental resource cost per additional continuously insured person-year as the difference in resources used between intervention and comparison divided by the additional covered person-years attributable to assignment. For illustration, if incremental resources cost 12,000 currency units and intervention and control yield 420 and 300 covered person-years in equally sized populations, the incremental cost is 100 units per additional covered person-year. If the difference in person-years is zero or statistically uncertain, the ratio becomes unstable; report costs and effects jointly with uncertainty rather than presenting a spurious ranking.

In insurer accounts, premium is revenue and a claim payment is an obligation arising from the insured event. In household accounts the same items are expenditures and receipts. In an economy-wide resource analysis the transfer must not be treated as a pure consumption of resources on both sides; administrative costs, forgone labor and financing distortions are relevant resource effects. Explicitly identify payer, price year, currency and whether the subsidy is a budgetary cost, a transfer or both in different perspectives.

A channel can increase cost per person reached but decrease cost per suitable in-force policy-year. Two channels can have identical sales costs and different outcomes if one regularly assists policyholders after loss. Longitudinal cost accounting therefore follows the entire contract life, including servicing, claims assistance, dispute resolution and exit. A short experiment may establish acquisition costs only; medium-term recurring costs require a continuing observation window.

Cluster-Level Channel Design Experiments

From treatment label to experimental intervention

A research protocol specifies the factual content of each arm: eligible products, premiums, exclusions, insured sum, distribution functions, permissions, hours of availability, assistance duration, incentives, languages, geographic coverage, fallback routes, and duration. Agents in each arm should receive defined workloads, training and scripts. The comparison should specify whether it represents the existing market, an enhanced usual-care channel, or an alternative new channel.

A plausible three-arm trial compares usual retail access, a trained visiting representative and a trained representative supported by a digital workflow. A fourth arm offering self-service access can identify the effect of assistance within the digital design. Product terms, pricing, eligibility and lawful disclosures are common, unless those factors are independently randomized. Because the representative's presence itself can change product awareness and confidence, the difference between visiting and remote channels is an effect of the configured service package.

A factorial trial can assign channel assistance and premium subsidy independently. Let Z denote assistance assignment and S subsidy assignment, each binary. For population-average enrollment or covered person-months, the difference E[Y|Z=1,S=s] minus E[Y|Z=0,S=s] is the channel assignment effect at subsidy level s. The interaction is the difference between those effects across subsidy levels. If assistance is particularly effective when a subsidy is available, a pooled main effect may conceal complementarity. Random assignment, adequate sample size in each cell and consistent implementation are needed. The same method can independently allocate claims assistance, provided no arm is deprived of required services.

Clustering, assignment and interference

Insurance agents, community groups, shops and local enrollment offices serve overlapping populations. Randomizing a village, service catchment, workplace or branch can reduce treatment contamination. Assignment should be stratified by population size, baseline coverage, travel burden, region and relevant risk exposure where those are known before treatment. When matched pairs are used, inference must respect the matching.

The CONSORT cluster trial extension (Campbell et al. 2012) emphasizes reporting details of cluster trial design alongside conventional trial reporting. For distribution experiments, this includes the number of stations, agents or villages randomized; numbers recruited and analyzed per arm; recruitment timing relative to assignment; loss of clusters; and analysis that recognizes correlation within the assignment unit. Put the cluster allocation mechanism, randomization seed where appropriate, and original untouched sampling list into controlled research records.

A rough planning calculation under equal cluster size m and intracluster correlation rho inflates individual-randomization variance by the design effect 1 + (m − 1)rho. For m = 40 and rho = 0.05, the factor is 2.95. Four thousand participants would then yield an approximate independent-observation equivalent of 1,356 for a simple mean contrast. This is a planning approximation, not an actual trial power analysis, and depends on equal sizes, outcome structure and the assumed correlation. Unequal clusters, rare claims, multiple arms and covariate adjustment require design-specific power simulation. Increasing participants within few clusters cannot fully substitute for increasing independent clusters.

Network spillovers have direct substantive meaning. A treated group can communicate terms to people assigned to control; a competitor can alter its offerings after an agent enters a village. The potential outcomes of an individual can then depend on others' assignment. The causal approach developed by Hudgens and Halloran (2008), causal inference with interference, provides a methodological basis for separating own-assignment and exposure effects. A saturation design can randomize communities to different proportions encouraged to use a channel and randomize individuals within communities. Analysis must define exposure beforehand and acknowledge uncertain cross-community interactions.

Intention to treat and treatment received

The primary effect of assignment to channel availability is the intention-to-treat (ITT) difference between outcomes under randomized offer and comparison. ITT includes inability to contact participants and failed adoption. It therefore represents the expected impact of implementing the specified offer on the intended population. Per-protocol comparisons between people who use channels and those who do not generally lose the original randomized comparability.

In an encouragement experiment, assignment may shift use without requiring it. The Wald ratio divides the ITT on the outcome by the ITT on channel use. Under random or as-if random assignment, relevance, exclusion, monotonicity and appropriate no-interference or exposure assumptions, this can identify a local average effect among compliers; see Angrist, Imbens and Rubin's (1996) instrumental variables analysis. For example, if 70% of assigned and 30% of comparison consumers use a channel and coverage differs by eight percentage points, the illustrative ratio is 0.08/0.40 = 0.20, or 20 percentage points for compliers under the assumptions. This is neither the overall population effect nor the effect on every observed buyer.

The exclusion restriction is especially demanding here. Assignment to an adviser can improve understanding or help a household navigate an existing contract even when the person never buys through the adviser. In that circumstance using assignment as an instrument for purchase does not isolate the effect of insurance alone. The offer ITT is still meaningful, but the instrumental-variable interpretation requires a different intervention or explicit acknowledgment of the violation. Report first-stage strength, event timing and the set of treatments potentially affected by assignment.

Myerson et al. (2022) provide a relevant retail-market example in California marketplace telephone outreach. Personalized assistance directed to people who had applied without selecting coverage was associated with an enrollment increase of 2.7 percentage points among consumers who received a call because of randomized assignment. The reported estimate is a call-receipt effect for induced recipients; it must not be described as the population-wide effect of assignment or of an agent's physical availability. The sampling stage and receipt-versus-assignment distinction are crucial.

Rollouts, power and data integrity

A staggered rollout may be appropriate when operational capacity permits only phased implementation. Randomly ordering clusters improves credibility compared with allowing offices to enter when ready. Define the time-specific comparator, account for seasonal cycles, and distinguish a short-run effect on new contracts from cumulative coverage and learning. If rollout is not randomized, the analysis requires defensible assumptions regarding confounding, anticipation and counterfactual time trends. A simple before-after comparison across treated sites cannot identify a channel effect.

Participants may leave a job, relocate, lose contact or experience a shock before interview. Administrative coverage can be linked to individuals through pseudonymous records; absence from a database cannot automatically be classified as uninsurance if other insurers may cover them. Survey attrition needs reporting by arm, with prespecified sensitivity analysis. Data collectors should use common instruments and, when practical, avoid knowing the recipient's assignment. Missingness due to treatment may itself reveal a service problem.

Prespecify primary endpoints, meaningful subgroup contrasts, uncertainty intervals, adjustment for multiple testing and treatment fidelity measures. A distribution manager may have incentives to redefine a failed application or renewal; independent readback from policy administration and claims ledgers reduces this risk. Raw application records, including abandoned and rejected applications, should remain traceable. Research access to sensitive identifiers and health information requires lawful authorization, minimization, and role-based controls.

Informed Enrollment and Coverage Continuity

Enrollment as a longitudinal process

Enrollment occurs after an eligible person becomes aware of an offer, evaluates terms, makes a decision, submits a valid application, meets acceptance conditions, and completes payment or qualifying public enrollment. Insurance coverage commences on the effective date established by the contract or governing scheme. Its termination, lapse, suspension and reinstatement follow relevant terms and law. A survey's reported purchase is therefore a weaker indicator of effective inclusion than verified coverage dates and benefits.

The correct numerator for a stage depends on the research question. The proportion attempting registration among eligible assigned persons measures population-level engagement. The completion proportion among all eligible persons captures the joint outcome of motivation and administrative capacity. Completion conditional on attempting isolates operational description, but conditioning on a post-assignment event can bias a cross-arm causal comparison. Both unconditional ITT estimates and descriptive conditional ratios should be reported.

For coverage continuity, let C_it equal one when person i has suitable coverage in force on day t and zero otherwise. The total covered person-days divided by the product of eligible population size and observation days is average protected time, with a precisely defined eligible-cohort rule. If 100 people receive certificates and half lose coverage after three months, a six-month endline count hides substantial variation in exposure. Event-date records allow the analysis to estimate time to lapse and interrupted duration. Where eligibility itself changes endogenously, report a fixed baseline cohort and an explicit alternative dynamic eligibility denominator.

Immediate and persistent effects

Baillon et al.'s (2022) Philippine incentive follow-up finds a persistent effect from a temporary subsidy and less durable enrollment among the additional participants persuaded by assisted application, although assistance generated a larger initial response. Because this is a follow-up of the same experimental setting, it extends the temporal evidence rather than supplying an independent geographic replication. A channel can induce consumers with very different willingness to renew, even when everyone faces the same future payment rules.

Chemin's (2018) Kenyan informal group studies reports that mobile payment options, individual registration help and smaller subsidies produced no significant enrollment response in an initial study. Full subsidy produced 45% take-up with negligible persistence once it ended. In another area, meetings with existing groups produced 12% take-up and 7% at one year without a premium subsidy. These are reported levels in distinct experimental contexts; comparing them as treatment effects of a common randomization would be incorrect. Their joint value is to demonstrate that first purchase and subsequent retention depend on the channel's social and economic setting.

The contrast with the earlier Philippine assistance result is scientifically productive: administrative help can remove a binding friction in one setting and fail when another friction dominates. An observed lack of incremental enrollment is compatible with insufficient trust, unsatisfactory services, limited resources, weaker assistance, or poorly adapted product conditions. Distinguishing those mechanisms requires designed variation or additional validated measurements, not retrospective attribution from an average effect.

Comprehension and meaningful choice

An informed enrollment measure tests whether the consumer understands the material terms necessary for a decision: covered event, major exclusions, premium schedule, benefit amount or limit, waiting period, documentation required at claim, and cancellation or renewal process. Questions should be adapted to the contract rather than using a generic financial-literacy score. Measure unaided and aided comprehension separately; repeated testing may itself affect behavior.

Enrollment consent must be distinguishable from research consent. A person may volunteer for a survey while declining the product. A member of a group can decline without losing unrelated group benefits. A borrower needs a clear account of whether any associated insurance is compulsory, whether substitutes are permitted and who receives the benefit. Automated renewal should require valid authority and suitable advance information, with an accessible correction and cancellation mechanism under applicable law.

Research design should recognize that a legitimate increase in informed rejection can be compatible with better distribution quality. A channel disclosing a material exclusion may lower sales while improving understanding and reducing later disputes. Prespecify safeguards against evaluating agents solely on quantity purchased. Assess informed choice using common independent outcome assessments rather than allowing the representative responsible for sales to certify understanding.

Digital enrollment and continuity

Remote channels can reduce travel and permit action outside branch opening hours. They can also introduce dependence on devices, network connectivity, account control, digital skills and electronic identity matching. A shared handset may expose confidential information or leave a household member unable to complete renewal independently. Digital support must include recovery after changed phone numbers, lost credentials, payment failures and interrupted connectivity.

Evaluate alternatives that use low-bandwidth interfaces, assisted enrollment, paper confirmation or authorized telephone support. Maintain an accessible service path for people without a smartphone. An experiment restricted to connected existing users estimates an effect within that population only. A population experiment offering a digital-only option must record nonreach, failures and exclusion; it should not silently omit the individuals unable to use the interface.

Digital continuity depends on synchronization between payment rails and the insurer's record. A successful transfer without accurate allocation to the policy can create a lapse that the consumer did not intend. Record payment time, settlement time, application to premium obligations, notification and any correction. Where the consumer law or scheme protects coverage during processing errors, reflect that rule in the in-force determination; do not use payment-system logs alone to infer legal status.

Claims Payment and Household Welfare

The insurance endpoint

The practical value of an insurance promise becomes observable when a covered event occurs. Claims evaluation should trace exposure, occurrence, notification, documentation, adjudication, settlement, appeal and payment. A low number of recorded claims may indicate little loss, an unsuitable product, incomplete notice, poor accessibility or legitimate contractual exclusions. The raw claim-to-policy ratio cannot determine which explanation applies.

In a representative study, report claim frequency per insured exposure-time, fraction of documented eligible loss events reported, paid fraction of assessed payable claims, timeliness of payment, reversal rates and reasons for denial or abandonment. The eligible-loss denominator requires independent event ascertainment and contract analysis; insurer claim records alone omit households who were unable to file. If adjudicated eligibility is influenced by the intervention or quality of documentation, the observed conditional acceptance rate is a descriptive service measure, not an uncomplicated causal effect.

For index insurance, payment is determined by a contractually defined index rather than the household's measured damage. Assess both positive triggers and loss without trigger, using independent loss data and verified formula inputs. Basis risk is an economic exposure created by imperfect alignment between the index and actual loss. A group may help share idiosyncratic losses, but a local association can itself face correlated shocks. A channel's trustworthy communication cannot repair the physical mismatch between index and loss.

Health insurance often pays providers rather than policyholders. Use paid covered amounts and net household out-of-pocket spending measured after reimbursement, separating copayment, uncovered costs, premiums, and indirect costs. A reduced direct payment at one provider may coexist with increased travel or prescription costs elsewhere. Determine whether reductions reflect actual financial protection, switching providers, deferred treatment or changed care volumes.

Household financial consequences

Thornton et al. (2010) report in their Nicaragua study that insured persons substituted toward covered facilities and spent less out of pocket, but the reduction was smaller than the premiums; they found no increased healthcare use and very low retention. This would be incomplete as a monetary welfare judgment: a reduction in variance and protection against rare severe expenses can be valuable even where average realized payments are small. Conversely, expected risk reduction cannot be assumed from enrollment if facilities fail to deliver covered services.

Karlan et al. (2014) provide evidence on a different mechanism through Ghanaian agricultural risk experiments. Rainfall insurance enabled increased agricultural investment and risk-taking in production choices. Their randomization compares insurance and capital-related interventions, not alternative distribution institutions. The finding establishes why channel evaluation should consider ex ante behavior as well as ex post claims: more credible coverage can change investment even when a loss does not occur during observation.

In Kakamega, Abajobir et al. (2024) examined subsidized digital health insurance through a pair-matched cluster trial covering 24 villages. The comparison varied access to fully subsidized enrollment with administrative support while facility improvements and community health volunteer training were shared by both arms. No differential statistically significant effect was detected on the measured reproductive, maternal, newborn and child health outcomes at the stated follow-ups. The authors report improvement over time in both arms. This is evidence about the incremental bundle in the study context; it does not isolate the effects of digital technology, premium subsidy, or facility improvement. The null estimates do not prove that all effects are zero.

Welfare measurement and rare events

Insurance affects expected consumption, the distribution of consumption across states, investment, labor decisions and willingness to undertake risk. Uninsured households may rely on precautionary savings, borrowing, asset sales and informal transfers. An experiment measuring these outcomes should distinguish changes before a shock from consequences after it. Predefine exposure intervals, covariate adjustment and event ascertainment. In a health scheme, record illness independent of care-seeking where feasible; otherwise measured incidence may rise because access improves.

A shock-by-assignment interaction can investigate whether coverage changes household adjustment to an adverse event. Exogeneity of the shock is essential for a causal conditional comparison, and the event itself may be affected by coverage. For instance, insurance can change preventive behavior or crop choices. Conditioning on realized losses may then select systematically different groups. Where the treatment potentially changes the event, report effects on joint outcomes such as the incidence of catastrophic expenditure or forced asset sale among all randomized households.

A small distribution experiment is rarely powered to measure rare catastrophic claims precisely. Its design can still quantify valid coverage, access to filing, simulated service capacity and actual process failures. Strong claims about household welfare and catastrophe protection require adequate event exposure, observation time and suitable outcome measures. Administrative linkages and repeated surveys can enable longer follow-up; post-trial permission and data governance need planning from the outset.

Cost and value without double counting

An insurance intervention has at least three evaluation perspectives. The consumer perspective includes the premium, time cost, uncovered loss, claim receipts and consumption stability. The insurer perspective considers premium, commissions, operating costs, expected claims, reserves and capital. The public perspective includes fiscal subsidy, supervision, real resource costs and distributional consequences. A transparent accounting keeps these perspectives separate and reconciles transfers.

For illustration, a household pays 100 units annually for coverage and receives a 1,000-unit benefit in a state with probability 0.08. Ignoring other features, expected gross benefit is 80 and expected cash loss relative to the premium is 20. This does not establish that the product is undesirable: the marginal value of funds in the loss state may be higher. Equally, a positive expected benefit is insufficient if claims are difficult to enforce or the buyer cannot finance the premium. Risk reduction, trustworthiness and liquidity all matter.

A channel that improves filing of eligible claims can increase the insurer's loss ratio. That result may indicate correction of an access failure rather than worsening underwriting. Conversely, selling small-premium policies with very low payouts can produce an apparent distribution success while offering little protection. Portfolio monitoring must distinguish adverse selection, misclassification, lawful payment of valid claims, benefit inadequacy, and service failure.

International Implementation and Regulatory Ethics

Institutional comparison

The reviewed experiments concern different legal and institutional populations: voluntary public or social health insurance in Nicaragua, the Philippines, Kenya and Indonesia; community health insurance in Burkina Faso; rainfall index protection in India, China, Ethiopia and Ghana; and subsidized private health plan enrollment on a United States state marketplace. Their findings illuminate overlapping institutional mechanisms rather than homogeneous regulatory markets.

The national scheme rules in force at each historical experiment determine enrollment rights, contract terms and eligibility. This study relies on the primary authors' institutional descriptions for historical scheme identities; it does not certify current licensing, contribution, privacy or entitlement law in each jurisdiction. A prospective experiment must produce a jurisdiction-specific legal and regulatory schedule for its actual study period before assignment begins. The schedule should state the responsible supervisor, required permissions, applicable product authorization, agency and outsourcing rules, customer data authority, contract formation, payment and claims rights, and complaint or appeal processes.

International standards provide an analytical baseline. The IAIS Insurance Core Principles (ICP 18 and ICP 19, online text checked November 19, 2026) address professional and transparent intermediation and fair treatment throughout the insurance business lifecycle. Practical obligations vary by national law and the roles assigned to individual parties. A subject's participation in an academic experiment cannot excuse unfair sales, unsuitable advice, missing records or a failure to honor an issued contract.

In the European Union, Insurance Distribution Directive (Directive (EU) 2016/97, consolidated text January 9, 2024) supplies a legal reference for customer-oriented distribution, disclosure and demands-and-needs assessment, subject to its scope and national implementing provisions. A study manipulating the availability of advice cannot remove disclosures or other requirements to construct a control arm. Article-level applicability, exemptions and current transposition would need separate verification for any real EU protocol. EIOPA's (2026) Third IDD Application Report is based on supervisory and stakeholder evidence for 2024–2025; this helps situate market monitoring, rather than measuring experimental causal effects or authorizing a general ranking of intermediaries.

Channel-specific duties

In a visiting-agent design, define responsibility for record retention, verified authority, language accessibility, compensation, customer identification, payment acceptance and the transfer of documents. Test implementation through independent mystery shopping where legally and ethically appropriate and consent-based interviews. A complaint mechanism should remain available to participants and nonparticipants throughout the study.

For a cooperative or community group, insurers and investigators need safeguards against coercion, disclosure of private circumstances to neighbors, informal collection without receipts, diversion of contributions and preferential service to group leaders. Membership in the collective is a distinct relationship from membership in the insurer's contract. The group must be clear about what it can promise and what remains the carrier's obligation.

For bank-linked or employer-mediated coverage, disclose the relationship between the insurance and the underlying financial or employment service. A lender or employer can supply contact infrastructure while also exerting economic influence. Evaluate customers outside the institution's existing base where the inclusion objective concerns the whole population. A design with voluntary opt-in must not make access to unrelated services contingent on research participation or on buying a policy. Group contracts need clear information about individual beneficiary rights, coverage on exit and the handling of complaints.

For digital platforms, separate data collection needed for insurance from platform use for other purposes. Consent, data minimization, secure credentials, correction of inaccurate records, accessibility and escalation to human support are important design requirements. A user interface that changes contract terms or presents a bundled purchase as inevitable undermines informed consent. Researchers should preserve legal minimum protections in every arm and obtain oversight for any differential intervention involving payment collection or automated renewal.

Safety and ethics in experimental allocations

Coverage and claims are welfare-relevant. Withholding a legally mandated service, ordinary eligibility, required advice, or claim assistance to manufacture experimental variation is inappropriate. Randomized designs should allocate genuinely incremental capacity when available, with suitable comparison services and transparent participant rights. Resource-constrained rollout can be meaningful when allocation is equitable and no participant is denied established entitlements. Where a treatment is known to remedy serious harmful failures, equipoise and the continuing justification for comparison require review.

Informed research consent must be understandable, and refusal must not compromise insurance rights. If vulnerable families receive premium subsidies, the continuation or end of financial support needs transparent communication in advance. The trial should record adverse events, coercive enrollment, privacy incidents, unexplained lapse and unfair claim treatment. An independent escalation channel should be able to pause recruitment or modify the protocol when serious harm emerges, while documenting the implications for analysis.

Report intervention fidelity and risks as outcomes, rather than concealing them in implementation notes. Compensation schemes tied solely to gross enrollments create incentives to enroll inappropriate households. A trial can compare carefully designed service incentives with ordinary permissible compensation but should monitor suitability, complaints, cancellations and claim access. Findings should be interpreted alongside the cost of compliance and the resources necessary for supervision.

External Validity and Causal Mechanisms

A comparative evidence matrix

The literature is best interpreted as evidence about specific mechanisms and outcomes. The following comparison records the assigned contrast rather than treating the channel's name as a uniform treatment.

Setting and research Assigned or varied element Supported result Transfer boundary
Nicaragua, Thornton et al. (2010) Premium and enrollment location Low overall enrollment; location/administration matter; MFI involvement slightly negative Voluntary public program, not generic bank distribution
Philippines, Capuno et al. (2016) Information and subsidy; subsequent administrative assistance Larger response to assistance among eligible residual nonenrollees Two different experimental stages
Philippines, Baillon et al. (2022) Follow-up after temporary incentives Immediate and persistent effects differ Same research lineage, not an independent replication
Indonesia, Banerjee et al. (2021) Subsidy and registration assistance Attempts exceeded successful registrations Specific administrative scheme
Kenya, Chemin (2018) Individual interventions; existing group meetings in separate study Unequal uptake and persistence across configurations No common cross-area head-to-head comparison
Colombia, Bauchet, Damon, and Larsen (2017) Randomized loan-time versus delayed voluntary crop insurance offer No overall detectable take-up or coverage difference; crop heterogeneity and low understanding Microfinance clients; timing within one channel, not mainstream bancassurance or claims
United States employer plans, Atanasov and Baker (2014) Employer survey and framed high-deductible choice experiment Plan presentation and physician-access perceptions influence choice; financial information alone did not increase framed selection Survey/framed choices, not randomized actual employer-versus-broker enrollment
China, Cai et al. (2015) Random information exposure within social networks Spillovers in adoption Information channel, not legal intermediation mandate
Ethiopia, Dercon et al. (2012) Content of training for risk-sharing groups Evidence consistent with benefits from discussing informal sharing of basis risk Working-paper abstract; group setting
Burkina Faso, Bocoum et al. (2019) Home-based information package Knowledge improved modestly, no detected uptake increase Product and location specific
India, Cole et al. (2013) Price and nonprice interventions Liquidity, salience and trust matter Rainfall index insurance
United States, Myerson et al. (2022) Randomized phone outreach Enrollment increased for induced call recipients Existing marketplace applicants without plan selection
Kenya, Abajobir et al. (2024) Subsidized digital coverage with assistance, shared service upgrades No detectable differential effect on specified maternal/child health outcomes Bundle, limited outcomes and trial horizon
Ghana, Karlan et al. (2014) Insurance and capital interventions Changed agricultural investment and risk decisions Welfare mechanism, not channel comparison

No common effect estimate is calculated because target populations, randomization units, products, comparison conditions, endpoints and periods differ. An absolute percentage-point effect should not be combined with a relative effect or a conditional probability without conversion to a valid common population and outcome. Studies with repeated follow-up provide evidence of persistence and should retain their statistical dependency. The matrix is a structured comparison of selected evidence, not a ranking of publication credibility.

Mechanism identification and transfer

Four mechanisms recur in the evidence: the monetary feasibility of paying for coverage, the burden of converting intention into valid enrollment, the circulation of usable information through institutions, and confidence in subsequent performance. They can operate simultaneously. A familiar representative can change information and confidence; an application assistant can reduce administrative costs and correct misunderstandings; a mobile payment facility can change travel burden and payment timing. Labeling an effect as one mechanism requires independent experimental variation or further credible identification.

The binding constraint is specific to the population and product. A person who knows the contract and trusts the insurer may need help supplying a record. Another person may have the correct documents but insufficient funds. In some locations health coverage may not be valuable when service quality or transport is inadequate. The organization that solves one constraint might be ineffective against another. Consequently, local diagnostic evidence should precede selection of a distribution experiment.

Transfer across countries requires comparison of baseline risk, existing protection, contract enforceability, affordability, documentation infrastructure, literacy, payment rails, availability of claims services, and the legally permitted functions of intermediaries. Transfer across product classes also requires consideration of duration and payout: a short agricultural index contract, a long-term life product and health insurance with frequent service use have very different pathways from coverage to welfare. Enrollment assistance that is useful across several classes may still have different renewal, claims and cost effects.

Designs for less-studied institutional channels

The directly checked corpus provides stronger causal evidence on assisted enrollment, administrative location, information and group channels than on employer and bank distribution of mainstream commercial insurance. Bank and employer channel comparisons are therefore a prospective research priority, not an established empirical ranking. A suitable experiment could randomize consenting bank branches to trained referral-only assistance or authorized full-service insurance intermediation while holding eligibility, carrier, coverage, premium, and lawful communications constant. Enroll an independent sample including noncustomers in catchment areas when the policy objective concerns population inclusion. Measure actual bank-client reach, displaced cover, complaint burden, claim assistance and renewal.

Employer-channel evaluation could compare an additional opt-in service at randomly assigned worksites with existing legally compliant access elsewhere, retaining the same eligibility, products and premium. Track workers who leave or change employers, dependent coverage, contribution continuity, independent consent, and transactions not available to informal workers. If the employer channel changes who is eligible, the intervention is a broader change in access rules and the denominator must reflect that distinction. A randomized encouragement to contact an existing broker would have a different estimand from introduction of a group policy.

A cooperative-to-agent comparison might randomize additional agent office hours across matched communities, with a separate arm offering group information and the same agent-supported application. If group participation is voluntary and spillovers are substantial, compare population ITT and network exposure under a defined design. Observe excluded nonmembers, not just those reached by the group. Claims and renewal assistance should be part of the evaluated service bundle, or explicitly held constant and assessed later.

These proposed designs are examples of applying established experimental methods. They do not constitute an original named theory or evidence that any one institution is superior. A prospective study must resolve regulatory authorization, sampling feasibility, confounding interventions, power and data governance before deployment.

Scale, competition and long-run outcomes

Experimental delivery may draw resources from nearby offices or require unusually intensive implementation. The marginal cost of an additional representative can change with scale, rural density, claim frequency and administrative capacity. An insurer may alter agent commission, underwriting or product prices when a channel grows; competitors may enter or exit. Such equilibrium changes do not follow from small trials conducted with fixed contract and market conditions.

Portability, service continuity after provider exit and recurring compliance costs matter. A digital pilot may depend on a single platform, while a cooperative intervention may rely on a motivated organizer who is difficult to replicate. Measure fidelity under routine supervision and distinguish one-time research activities from ongoing service costs. Independent replication across settings and longer follow-up are needed before generalizing to nationwide implementation.

A decision to expand a channel should consider population ITT on suitable coverage, cost per durable protected exposure, valid claim servicing, equity between groups, risk-pool effects, and operational feasibility. A null result should prompt focused diagnostic work, not automatic escalation of sales intensity. A positive acquisition result should lead to renewal and claims measurement, rather than premature inference of welfare gains. This sequence connects scientific evidence to institutional design while preserving uncertainty about downstream effects.

Conclusion

Insurance financial inclusion is established through a functioning distribution and protection relationship. The selected experimental literature shows that enrollment depends on the joint operation of price, administrative capacity, information and social context. Institutional arrangements determine which barriers are encountered and which can be removed. The evidence from the Philippines, Indonesia and Nicaragua distinguishes assistance and location from price or awareness. Results from Kenya, China, Ethiopia and Burkina Faso demonstrate the context-dependent contribution of social communication and the limitations of information-only or payment-channel interventions.

The appropriate causal estimand depends on the actual intervention: assignment to an available channel, encouraged use, contact or receipt, and completed enrollment are distinct. A comparison of users of two channels is not automatically randomized. Cluster designs and factorial assignments permit relevant contrasts when interference, treatment fidelity and service capacity are considered. Population coverage, person-time, continuity and valid claims provide more informative endpoints than conversion alone.

The Colombia randomized timing study, employer health-plan choice experiment, and ILO correspondent case studies sharpen the institutional boundaries of the comparative evidence: credit-linked offer timing, employer menu framing and descriptive banking-correspondent operations are distinct treatment objects. Their findings do not provide a common cross-channel causal estimate or demonstrate long-run claims effects.\n\nThe analysis supports three substantive conclusions. First, meaningful access requires measuring the whole sequence from eligibility and reach to successful in-force protection, including administrative failures. Second, the channel has an economic and ethical service function across the insurance lifecycle; acquisition and claims performance should be evaluated together with pricing, financing and consumer safeguards. Third, international transfer requires identifying the operative mechanism and matching the relevant institutional context, with direct experimentation where evidence is insufficient. The corpus is particularly informative about health and agricultural index insurance and does not establish a causal ranking for bank or employer delivery of all mainstream product classes.

The author's contribution is a comparative and methodological synthesis of existing empirical research, insurance operations and supervised distribution. It specifies how to distinguish channel effects from subsidies, product changes and downstream welfare, and integrates valid denominators, persistence, cost and conduct into a common experimental assessment. Original empirical estimates and universal policy conclusions are outside the verified evidence. Further research should prioritize long-run claims and household welfare, less-examined mainstream insurance distribution arrangements, and independently replicated designs with suitable common endpoints.

References

  • Abajobir, Amanuel, Caroline Wainaina, Menno Pradhan, Wendy Janssens, and Estelle M. Sidze. 2024. “The Impact of Digital Interventions on Health Insurance Coverage for Reproductive, Maternal, Newborn and Child Health Services Utilization in Kakamega, Kenya: A Cluster Randomized Controlled Trial.” Health Policy and Planning 39 (10): 1007–1021. DOI: 10.1093/heapol/czae079. Source record.

  • Angrist, Joshua D., Guido W. Imbens, and Donald B. Rubin. 1996. “Identification of Causal Effects Using Instrumental Variables.” Journal of the American Statistical Association 91 (434): 444–455. DOI: 10.1080/01621459.1996.10476902. Source record.

  • Atanasov, Pavel, and Tom Baker. 2014. “Putting Health Back Into Health Insurance Choice.” Medical Care Research and Review 71 (4): 337–355. DOI: 10.1177/1077558714533821. Publisher record.

  • Baillon, Aurélien, Joseph Capuno, Owen O’Donnell, Carlos Antonio Tan Jr., and Kim van Wilgenburg. 2022. “Persistent Effects of Temporary Incentives: Evidence from a Nationwide Health Insurance Experiment.” Journal of Health Economics 81: 102580. Source record.

  • Bauchet, Jonathan, Amy Damon, and Vance Larsen. 2017. “Microfinance Bundling and Consumer Protection: Experimental Evidence from Colombia.” Journal of Development Effectiveness 9 (4): 443–461. DOI: 10.1080/19439342.2017.1363802. Publisher record.

  • Banerjee, Abhijit, Amy Finkelstein, Rema Hanna, Benjamin A. Olken, Arianna Ornaghi, and Sudarno Sumarto. 2021. “The Challenges of Universal Health Insurance in Developing Countries: Experimental Evidence from Indonesia’s National Health Insurance.” American Economic Review 111 (9): 3035–3063. DOI: 10.1257/aer.20200523. Source record.

  • Bocoum, Fadima, Michael Grimm, Renate Hartwig, and Nathalie Zongo. 2019. “Can Information Increase the Understanding and Uptake of Insurance? Lessons from a Randomized Experiment in Rural Burkina Faso.” Social Science & Medicine 220: 102–111. DOI: 10.1016/j.socscimed.2018.10.029. Source record.

  • Cai, Jing, Alain de Janvry, and Elisabeth Sadoulet. 2015. “Social Networks and the Decision to Insure.” American Economic Journal: Applied Economics 7 (2): 81–108. DOI: 10.1257/app.20130442. Source record.

  • Campbell, Marion K., Gilda Piaggio, Diana R. Elbourne, and Douglas G. Altman, for the CONSORT Group. 2012. “CONSORT 2010 Statement: Extension to Cluster Randomised Trials.” BMJ 345: e5661. DOI: 10.1136/bmj.e5661. Source record.

  • Capuno, Joseph J., Aleli D. Kraft, Stella Quimbo, Carlos R. Tan Jr., and Adam Wagstaff. 2016. “Effects of Price, Information, and Transactions Cost Interventions to Raise Voluntary Enrollment in a Social Health Insurance Scheme: A Randomized Experiment in the Philippines.” Health Economics 25 (6): 650–662. DOI: 10.1002/hec.3291. Source record.

  • Chemin, Matthieu. 2018. “Informal Groups and Health Insurance Take-up Evidence from a Field Experiment.” World Development 101: 54–72. DOI: 10.1016/j.worlddev.2017.08.001. Source record.

  • Cole, Shawn, Xavier Giné, Jeremy Tobacman, Petia Topalova, Robert Townsend, and James Vickery. 2013. “Barriers to Household Risk Management: Evidence from India.” American Economic Journal: Applied Economics 5 (1): 104–135. DOI: 10.1257/app.5.1.104. Source record.

  • Dercon, Stefan, Ruth Vargas Hill, Daniel Clarke, Ingo Outes-Leon, and Alemayehu Seyoum Taffesse. 2012. “Offering Rainfall Insurance to Informal Insurance Groups: Evidence from a Field Experiment in Ethiopia.” BREAD Policy Paper 032, 30 pp. Source record.

  • European Insurance and Occupational Pensions Authority (EIOPA). 2026. Third IDD Application Report. EIOPA-26-077. March 30, 2026. Official publication record. Source record.

  • European Parliament and Council of the European Union. 2016. Directive (EU) 2016/97 of 20 January 2016 on Insurance Distribution (Recast). Official Journal L 26, February 2, 2016, 19–59. Consolidated text as of January 9, 2024, Articles 1, 17 and 20. Source record.

  • Hudgens, Michael G., and M. Elizabeth Halloran. 2008. “Toward Causal Inference with Interference.” Journal of the American Statistical Association 103 (482): 832–842. Source record.

  • International Association of Insurance Supervisors (IAIS). n.d. Insurance Core Principles and Common Framework Online Tool. ICP 18 (Intermediaries) and ICP 19 (Conduct of Business). Accessed November 19, 2026. Source record.

  • Leach, Jeremy, Anand Menon, and Sandisiwe Ncube. 2014. “Achieving Scale and Efficiency in Microinsurance through Retail and Banking Correspondents.” Microinsurance Paper 37. Geneva: International Labour Organization. Institutional record.

  • Karlan, Dean, Robert Osei, Isaac Osei-Akoto, and Christopher Udry. 2014. “Agricultural Decisions after Relaxing Credit and Risk Constraints.” Quarterly Journal of Economics 129 (2): 597–652. DOI: 10.1093/qje/qju002. Source record.

  • Myerson, Rebecca, Nicholas Tilipman, Andrew Feher, Honglin Li, Wesley Yin, and Isaac Menashe. 2022. “Personalized Telephone Outreach Increased Health Insurance Take-Up for Hard-to-Reach Populations, But Challenges Remain.” Health Affairs 41 (1): 129–137. DOI: 10.1377/hlthaff.2021.01000. Source record.

  • Thornton, Rebecca L., Laurel E. Hatt, Erica M. Field, Mursaleena Islam, Freddy Solís Diaz, and Martha Azucena González. 2010. “Social Security Health Insurance for the Informal Sector in Nicaragua: A Randomized Evaluation.” Health Economics 19 (Suppl. 1): 181–206. DOI: 10.1002/hec.1635. Source record.

  • World Bank. 2025. The Global Findex Database 2025: Connectivity and Financial Inclusion in the Digital Economy. Washington, DC: World Bank. Official report landing page. Source record.