Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Social Protection Registries and Inclusion Exclusion Errors

Abstract. Social protection registries are administrative infrastructures through which states discover potentially eligible households, record socioeconomic circumstances, reconcile information and support referrals to programs. Their measurable effectiveness depends on much more than the predictive accuracy of a poverty score. A person may have a valid legal entitlement and remain absent from a registry; a registered household may be incorrectly classified; a correctly classified household may be denied enrollment or payment. These failures require separate observation units, denominators, dates and accountability procedures. This monograph examines social registries as access infrastructure, distinguishes registry coverage from legal entitlement, eligibility determination and benefit receipt, and develops an operational analysis of inclusion and exclusion across the full delivery process. Its evidence consists of International Labour Organization standards, World Bank delivery-system research and statistical methodology, peer-reviewed targeting studies, and dated institutional documentation from Brazil, Chile, Pakistan, the Philippines and Indonesia. It establishes how periodic census-style registration, income-data latency, identity and household matching, proxy means testing, administrative burdens and grievances interact to produce observed errors. It also identifies which conclusions can be inferred from official coverage counts and which require independent household surveys, randomized evaluations, record-level audits or longitudinal follow-up. The central finding is that registry accuracy and effective protection are analytically related but empirically nonidentical: improvements in one stage can coexist with persistent exclusion elsewhere. An evidence-led assessment must therefore report the entire progression from need to lawful entitlement, discoverability, enrollment, timely payment, correction and appeal.

Keywords: Social protection registries; social assistance; social insurance; inclusion error; exclusion error; targeting; proxy means testing; dynamic registration; administrative burden; data reconciliation; social rights; benefit take-up; grievance redress; social protection delivery systems; comparative public administration.

Introduction

The social registry has become one of the central administrative instruments through which public protection systems recognize household needs. Its significance lies in the fact that the record precedes many subsequent decisions. When a person is not reachable by outreach, cannot authenticate an identity, or is omitted from registration, the state may never reach the substantive eligibility assessment. Registration is thus a distributional event in its own right. The World Bank's social registry guidance describes social registries as inclusion systems and as information systems; both functions must be examined simultaneously (Leite et al. 2017). The delivery-systems Sourcebook situates registries in the wider chain of outreach, intake, assessment, enrollment, provision and management (Lindert et al. 2020).

The scale of the underlying policy challenge is considerable. The International Labour Organization's World Social Protection Report 2024–26 reports that 52.4 percent of the global population was effectively covered by at least one social protection benefit in 2023, while 3.8 billion people remained entirely unprotected. This is an aggregate measure of benefit coverage, not an estimate of errors in any particular social registry. The World Bank's State of Social Protection 2025 reports approximately two billion people in low- and middle-income countries lacking adequate protection. The two estimates describe different populations and definitions of protection and must not be subtracted from, or treated as confirmations of, one another.

A registry can be large yet still miss those most difficult to enumerate. It can also identify accurately the households that already appear in its database while the underlying enrollment process omits persons outside that database. Most published policy statistics do not contain enough information to distinguish these possibilities. The aim of this study is to make the distinction explicit, to assess the supported cross-country evidence and to specify auditable indicators that connect registry administration with actual benefit access.

This is a comparative documentary and analytical monograph. It draws on statutory materials and official administrative descriptions, global institutional syntheses and empirical journal research. Country cases are chosen for their distinct administrative approaches and the availability of identifiable primary documentation, not as a representative or random sample. Legal provisions are reported for their stated jurisdiction and effective context. Where documents describe intended improvements, the manuscript identifies them as design claims rather than observed causal effects. No original field survey, confidential administrative microdata analysis, independent national beneficiary audit or new theory is claimed.

Conceptual and Historical Foundations

Rights, protection and administrative recognition

Social protection encompasses several distinct institutional forms. Contributory social insurance is usually anchored in qualifying contribution or work histories and statutory contingency coverage. Noncontributory social assistance distributes resources under legislated or administratively defined eligibility rules, frequently using income, demographic categories or vulnerability indicators. Tax-financed universal benefits, public employment measures and access to services may operate alongside both. A household-centered social registry is commonly most visible in means-tested assistance, although it may facilitate referral to other social protection and social service schemes. A social insurance contribution ledger, by contrast, establishes work and payment history and has different evidence and correction requirements.

The Social Security (Minimum Standards) Convention, 1952 (No. 102) sets international standards across nine branches of social security. Its legal implications depend on ratification and accepted branches, and no social registry itself creates these entitlements. The Social Protection Floors Recommendation, 2012 (No. 202) calls for universality, nondiscrimination, accessible complaint and appeal procedures, and protection of private information in social security data systems (ILO 2012, paragraphs 3, 7 and 23). The Recommendation is normative guidance, while binding domestic benefit conditions and the legal force of any international convention require jurisdiction-specific examination.

The resulting sequence of concepts is precise. Legal coverage denotes the scope of protection contemplated by law; potential eligibility denotes satisfaction of program conditions at a defined time; registry inclusion denotes presence within a specified administrative database; administrative selection denotes the result of assessment; enrollment denotes a formal program relationship; receipt denotes actual benefit delivery; adequacy denotes the value and practical sufficiency of what is received. Each variable can vary independently. Evidence that a household appears in a registry cannot establish that it received a transfer; evidence of a transfer cannot establish that all legally eligible households received one.

What a social registry records

Leite et al. (2017) distinguish the applicant-oriented social registry from the beneficiary registry. The social registry records people and households considered for programs, potentially including those ultimately excluded from benefits; beneficiary registries record enrolled participants and program transactions. Their registry typology and architecture also distinguish periodic collection from continuous access to registration. An integrated information system may exchange data across several programs, but institutional integration neither guarantees a universal population register nor establishes equal legal eligibility across programs.

The unit of observation can be an individual, family, household, dwelling, claim, or benefit episode. These are not interchangeable. A household may contain more than one family; a family may split or move; a benefit may be legally assigned to an individual but calculated using household income. Cross-program systems must specify which household definition controls, how time-dependent membership is recorded, and whether children, informal caregivers and displaced persons can be represented through changing household relationships.

A registry has an important temporal structure. Let t be an assessment date, t₀ the date a household reported income, and t₁ the date the underlying payroll or tax record became available. A classification based on income measured at t₀ may be formally correct for that historical period and substantively unsuitable for a benefit claim at t. Data latency is therefore an analytical variable, not merely a technical defect. This matters particularly where income is seasonal, informal, interrupted or rapidly changing.

From enumerations to dynamic inclusion

Social program administration historically relied on local lists, paper certificates, caseworker assessments, periodically conducted household surveys and separate program files. Electronic systems made it possible to reuse household data, create cross-program gateways and compare records held by different public institutions. The World Bank's 2017 comparative study of 20 countries documents substantial diversity of institutional arrangements, information systems and enrollment practices; it does not establish a universal evolutionary sequence.

A census-like sweep can produce a large snapshot with well-defined fieldwork dates. Its weakness is that newly poor or newly resident households may have no access until a later sweep. A continuously accessible registry allows new applications and updates, but only if outreach, offices, staff, connectivity and accessible channels make the theoretical opportunity practicable. The World Bank's guidance on shock-responsive delivery describes dynamic registries as living platforms that support ongoing intake and welfare reassessment. The transition from static to dynamic administration must therefore be evaluated by observed access and data recency, not by the software's label.

Legal and statistical meaning of an error

A targeting error is always relative to a reference rule. A person considered poor under an independently measured poverty threshold may fail a program-specific means test for reasons that are legally authorized; the outcome is analytically important but cannot automatically be labeled an unlawful denial. Conversely, an eligible beneficiary denied a benefit because of a wrong household record may have suffered a legal-administrative error even if that person is not poor under a researcher's poverty definition.

The terms inclusion and exclusion are often used with different denominators. Some sources define inclusion error as the fraction of recipients who are not target-eligible; others report the fraction of noneligible persons who receive benefits. The first is a false-discovery or leakage share; the second is a false-positive rate. These can diverge widely where the target population is small. The World Bank's targeting review examines errors in assigning benefits to intended households, while the competing definitions of exclusion, inclusion and leakage require explicit denominators. A scientific report must give the numerator, denominator and poverty/eligibility rule whenever it states an error rate.

Administrative exclusion is also distinct from absence of effective coverage. A universal program can have no income-based targeting and still exclude people through deficient civil registration, residence restrictions, inaccessible applications or payment failures. The degree of registry centralization is similarly distinct from the degree of protection: one database can serve programs with narrow eligibility, while several coordinated databases can support broader entitlements. This distinction guides the empirical analysis that follows.

Registry Coverage

Population, frame and reach

Registry coverage is the proportion of a declared reference population represented by valid records at a stated date. A meaningful measure requires the boundaries of the population frame and the state of the record. Let N denote the number of persons in the defined population, and R the number with an identifiable, current, nonduplicated registry record. The descriptive registry coverage ratio is C_R = R/N. For household registers, the equation uses households rather than persons; converting the result to population coverage requires verified household composition and appropriate weighting. A count of records, a count of people associated with records, and a count of currently eligible persons are three different quantities.

A headline national count is best treated as a measure of administrative scale. It may include inactive records, persons living abroad, deceased persons whose records are awaiting correction, reconstituted households or duplicate identities. Conversely, some persons who receive benefits may use a benefit platform or legacy enrollment route outside the social registry. To estimate completeness, the analyst needs an external reference frame such as a census, probability-based survey or high-quality population register and must identify the groups that the frame itself may omit.

The ASPIRE methodology records the distinction between administrative program spending and household-survey-based performance indicators. Its household survey indicators cover specific survey years and program classifications and may include indirect beneficiaries. Administrative registry counts therefore cannot be equated with ASPIRE's participation or benefit incidence statistics without reconciling the target population, reference period, person-household conversion and program categories. The ASPIRE program definitions explicitly separate social assistance, social insurance and labor-market measures.

The denominator problem

Suppose a register holds information on 800,000 households while a survey identifies 1,000,000 households nationally. The nominal ratio is 80 percent. If the registry contains 60,000 obsolete or duplicate household records, the effective numerator may be 740,000 and the comparable rate 74 percent. These numbers are an illustrative calculation only, not estimates for any country. They show why identifying the population unit and the record validation procedure is prior to reporting coverage.

The more relevant denominator for a targeted program may be all potentially eligible households, which is not necessarily observable in registry records. If eligibility is determined only after registration, a registry-based estimate of the fraction of eligible people served is subject to selection on being registered. Conditioning on the administrative database turns the most consequential missing people into statistically invisible cases. An independent survey with the statutory conditions applied to survey records can reduce this problem, although incomplete survey variables may require explicit assumptions.

Coverage should be disaggregated by district, remote location, disability-related accessibility, age group, household composition, migration or displacement status, and other classifications that are lawful and statistically supportable. Disaggregation makes clear whether low national omission is masking very high omission for smaller populations. Administrative counts must be accompanied by uncertainty where matching to a household survey introduces sampling error, undercoverage, nonresponse or imperfect linkage.

Outreach, registration and take-up

Enrollment begins before a form is completed. Knowledge of a program, willingness to approach an agency, travel distance, documentation, literacy, digital access and perceived stigma all influence registration. Herd and Moynihan's study of administrative burden distinguishes learning, compliance and psychological costs of citizens' interactions with government (Herd and Moynihan 2018). A registry's promise of self-service can reduce one set of costs and introduce another through authentication requirements or internet dependence.

The probability that an eligible person receives a timely benefit may be expressed descriptively as the product of successive conditional transition probabilities: P(receipt | eligible) = P(reached | eligible) × P(registered | reached, eligible) × P(correctly assessed | registered, reached, eligible) × P(enrolled | correctly assessed, registered, reached, eligible) × P(paid on time | enrolled, correctly assessed, registered, reached, eligible). This is the probability chain rule under a specified sequence, not a causal model and not an assumption of independence. If every stage achieves a 95 percent conditional success rate, the illustrative total is about 77.4 percent. High performance within each administrative office can therefore coexist with large end-to-end losses.

A successful registry requires more than accepting applications. Applicants must know where and how to update circumstances, obtain decisions, identify the reason for an adverse classification and access a remedy. ILO Recommendation No. 202 requires accessible and inexpensive complaints and appeals in its paragraph 7 guarantees. A registry that records complaints but does not correct downstream eligibility or unpaid periods has measured contact without achieving effective remediation.

Registration without effective entitlement

The ILO's reported effective coverage series places global coverage at 52.4 percent in 2023 for at least one benefit, but a person can be registered without appearing in that indicator. The reverse is also possible where the person receives a contributory benefit through a separate social insurance administration. This prevents the use of registry penetration as a proxy for universal social protection.

A public register may extend to many households because several programs use it. In such cases, a larger registry is a potentially valuable platform investment; its distributive value depends on program eligibility breadth, fiscal allocations, service quality and take-up. The World Bank's 2025 registry review reports social registries in 62 countries covering approximately 1.1 billion people as of 2024. These World Bank administrative figures describe the scale of registry systems, whereas the ILO's population coverage estimate describes benefit receipt. The figures refer to distinct statistical objects and are presented side by side only to explain their different meanings.

Dynamic inclusion and the continuity of access

The practical test of dynamic inclusion is whether a newly vulnerable person can make an effective claim during a change in circumstances. Performance indicators include the median time from attempted contact to successful registration, the fraction of submitted updates processed within the official standard, the prevalence of rejected or incomplete intake attempts, and the number of households registered through accessible outreach channels. An automated timestamp is reliable only if failed attempted visits, abandoned applications and inaccessible contacts are also visible to evaluators.

The challenge of shock responsiveness illustrates the relationship between preparedness and access. Prior enrollment can speed up payments to known households; sudden unemployment, disaster displacement or livelihood loss can create eligible populations absent from previous files. Cross-sector data sharing may help identify these households, but cannot by itself establish their consent, legal eligibility or reachability. The World Bank's adaptive social protection guidance points to continuous intake, updated data and interoperable delivery components. Its institutional guidance supports design analysis, not an unconditional claim of superior outcomes for every implementation.

Inclusion and Exclusion

The confusion matrix and its competing denominators

For a specified program, date and eligibility standard, classify each person by actual eligibility E and actual receipt B. Let TP denote eligible recipients, FN eligible nonrecipients, FP ineligible recipients, and TN ineligible nonrecipients. The matrix is:

Benefit received Benefit not received
Eligible under the stated rule TP: reached eligible population FN: eligible population excluded
Ineligible under the stated rule FP: ineligible recipients TN: correctly unserved ineligible population

The exclusion error rate among eligible persons is FN/(TP + FN). The inclusion error share among beneficiaries is FP/(TP + FP); another convention uses the false-positive rate FP/(FP + TN). These definitions answer different questions. Overall accuracy (TP + TN)/N can be misleading where the eligible population is relatively small, because a system rejecting everyone may still classify most of the population correctly while failing its distributive purpose.

A complete assessment reports at least the recall of eligible persons TP/(TP + FN), the eligible-recipient share TP/(TP + FP), and the target population prevalence (TP + FN)/N. It states whether receipt is measured as any payment, an adequate amount, or an on-time payment. Where benefits vary in value, a program can have a low recipient-level inclusion error and still transfer disproportionate monetary resources to noneligible recipients. The ASPIRE documentation on beneficiary and benefit incidence defines separate distributional indicators; neither is identical to the classification matrix above.

For illustration, consider 1,000 households of which 300 qualify and 700 do not. If 180 eligible households and 70 ineligible households receive a benefit, the exclusion rate is 120/300 = 40 percent, the beneficiary-based inclusion error is 70/250 = 28 percent, and the false-positive rate is 70/700 = 10 percent. These figures demonstrate denominator dependence. They are neither observed program results nor parameter estimates.

The layered sources of error

Errors may arise at six analytically distinct points. Outreach error occurs when eligible people are unaware of, or cannot access, the application opportunity. Registration error arises from omission, refusal, duplication, identity problems or transcription. Data error comes from stale income, incorrect household composition, incomplete assets, inconsistent disability assessment or inaccurate administrative matches. Rule-application error occurs when correctly collected facts are processed under an incorrect legal or computational rule. Enrollment error occurs where a favorable eligibility decision is not operationalized. Payment error occurs when an enrolled person does not receive the lawful benefit accurately and on time.

An observed exclusion event may combine several of these mechanisms. Describing every denial as a defect in the proxy means test would conflate the predictive scoring stage with barriers that existed before scoring. Conversely, a technically perfect registry could support a program whose lawful eligibility threshold leaves out many poor people. That is a design choice and poverty-policy issue distinct from the accuracy of the administrative record.

The 2022 World Bank synthesis Revisiting Targeting in Social Assistance emphasizes that the choice among methods depends on policy objectives and context. It recognizes both the budget advantages of prioritizing poorer households and the importance of robust delivery, with no single technique appropriate everywhere. Its assessment is compatible with the layered error distinction: econometric precision is one component of access and not the sole objective of social policy.

Proxy means tests and measurement limits

A proxy means test estimates welfare or consumption from observed indicators such as housing quality, durable assets, demographics and location. It is used when direct earnings or consumption data are expensive, unreliable or unavailable. The predicted ranking may be informative for stable structural poverty, but households with similar observables may have substantially different current incomes, debts, needs or exposure to shocks. The prediction's residual error remains relevant even when an algorithm has been fitted properly.

Brown, Ravallion and van de Walle's nine-country African targeting analysis shows that conventional proxy-means tests can exclude many poor households, and that more poverty-focused specifications can improve some outcomes without solving the underlying distributional problem (Brown et al. 2018). These results concern studied household survey settings and policy simulations. They do not establish a universal numerical error rate for all registries, nor do they identify the effects of particular registration offices.

In an Indonesian field experiment across 640 villages, Alatas et al. (2012) compare proxy-means tests, community rankings and a hybrid procedure. Community approaches performed somewhat worse under the study's consumption-based poverty measure, while communities expressed greater satisfaction; the authors find that differing concepts of poverty explain part of the discrepancy. The experiment demonstrates that a benchmark chosen by an evaluator is part of the meaning of accuracy. It does not prove that consultation should replace formal eligibility law or that any local ranking is corruption-free.

Targeting, welfare and horizontal equity

With a fixed budget, concentrating payments on those with lower welfare can increase the amount available to each recipient. Yet narrowing eligibility can increase administrative costs, erroneous exclusions, discontinuities at the cutoff and perceptions of arbitrary treatment among otherwise similar households. Hanna and Olken's comparison of universal and targeted transfers discusses the welfare advantages that targeting can retain despite measurement imperfections, together with horizontal equity violations and implicit marginal disincentives (Hanna and Olken 2018). An efficient budget allocation, an accurate poverty classifier and a rights-consistent delivery system are related but different evaluative objectives.

The correct criterion for judging false positives also varies with the program. In a means-tested transfer, payment to a household whose legally countable income exceeds the threshold may be classified as an administrative inclusion error. In a universal child benefit, the same family's receipt may be fully lawful. If a national study imposes a poverty-line benchmark on the universal program, the resulting 'leakage' measures a distributional share rather than a violation of the benefit's eligibility rule.

The consequences of errors are asymmetric. Wrongly rejecting a household near subsistence can be more damaging than modestly overpaying a household just above a means-test threshold. The relevant costs depend on the social objective, the degree of poverty, urgency of need, lost entitlement duration, appeal expenses and public fiscal impact. Monetary loss and probability of misclassification should therefore be reported separately rather than compressed into an unexplained composite score.

Distinguishing error, misconduct and uncertainty

An inclusion error is not synonymous with fraud. A household may provide truthful information while authorities act on stale records; an agency may legitimately pay benefits during a statutory transition; administrative databases may disagree about a name or family relationship. A fraud allegation requires proof of the relevant prohibited conduct and mental element under applicable law. Similarly, statistical model uncertainty cannot be equated with a legally erroneous decision without specifying which rule governed the household at the decision date.

Evaluation should preserve contested cases as a distinct category until the factual and legal dispute is resolved. If all pending or appealed claims are automatically counted as ineligible, exclusion will be underestimated. If all later-successful appeals are treated as original administrative errors, the analysis may overlook intervening changes in lawful eligibility. The date of the underlying event, application, original decision and reconsideration must remain linked.

Data Reconciliation

Record matching as an administrative decision

Reconciliation joins observations about the same person, household or event across sources. Its purpose can be to resolve duplicates, update income, identify changes in family composition, connect disability or school records, or verify payments. The data architecture must specify the authoritative field, evidence date, source agency, matching confidence, correction route and decision history for each material attribute. A person's name and date of birth are not a complete national identity. Even a unique identity number may be missing, incorrectly assigned, shared in error or stored in an outdated household relationship.

The analytic distinction is between accurate data linkage and sound entitlement determination. Two linked records may accurately belong to the same individual and still describe different periods or meanings of income. An annual tax return, a monthly payroll entry and a contemporaneous self-reported loss of earnings are not interchangeable. A system that systematically privileges the oldest but most formally verified source may create delay-driven exclusions. A system that accepts every unverified declaration may increase the risk of inaccurate assessments. These tradeoffs require rules for evidence currency, conflict adjudication and provisional entitlements.

The World Bank's social registry assessment framework identifies data management, interoperability, governance and updating as different building blocks of registry capability (Leite et al. 2017). A technical interface alone does not specify which administration bears responsibility for an incorrect record, how the household is notified, or how correction changes decisions across several programs. Those are institutional and legal questions.

Brazil: a multidimensional federal gateway

Brazil's Cadastro Único is a significant example of a shared registry serving different public programs. The Ministry of Development and Social Assistance identifies Decree No. 11,016 of March 29, 2022 as the regulatory instrument for the federal registry and Law No. 8,742 of 1993 as part of its statutory framework. These authorities apply to the Brazilian national system; they are not model statutes automatically applicable elsewhere.

The ministry's Cadastro Único registration guidance states that households should update information when material circumstances change and at least every 24 months. Federal Law No. 15,077 of 2024, Article 2, formalized a maximum 24-month updating requirement for relevant federal transfer programs and specified advance notification provisions. Its Article 1 addresses biometric documents with exceptions where access conditions cannot be supplied. The statutory details matter because stringent verification rules combined with inadequate outreach may turn record maintenance into an entitlement barrier; the exceptions and notice rules are part of the legal design, not peripheral conveniences.

The ministry's Bolsa Família rule explanation describes cross-checks with other government and social security records and procedures for identifying stale or inconsistent information. It also distinguishes transitional payment protection after certain earnings increases. An income match that triggers review therefore need not imply immediate lawful termination. Analysts should compare the reason for the alert, the applicable protection rule and the eventual payment outcome.

The World Bank's 2025 registry program account states that Cadastro Único connected 94.5 million people as of September 2025 to more than 60 federal programs. This is an institutional coverage and program-use statistic, not the share of Brazilian residents receiving Bolsa Família, and does not estimate the number of erroneous exclusions. Brazil's example demonstrates the administrative advantage and shared risk of a cross-program data gateway: a correct update may improve several eligibility decisions, while a mistaken attribute can propagate across several assessments unless program-specific rules and correction processes remain visible.

Chile: record correction and reference periods

Chile's Registro Social de Hogares combines household-declared information with government administrative databases. Its official household registry questions explain that recorded employment income may reflect a 12-month income reference period and that administrative data can be delayed. A household may have reported recent unemployment yet still have earlier earnings represented in the calculation. The official explanation makes it possible to distinguish genuine record error from a formally defined lookback period, even when the current hardship of the household warrants separate policy attention.

Chile also publishes procedures for updating administrative income, including evidence requirements for dependent and independent work. Its update and rectification procedures give a documented route for reconciling declared information against institutional data. The possibility of filing a rectification is not evidence that every claim is resolved promptly; actual accessibility requires information on elapsed processing time, successful corrections and outcomes for households without online credentials or supporting documents.

The Chilean case is analytically useful because it exposes a frequent source of disagreement about 'accuracy.' A data item may faithfully represent historical earnings and yet poorly represent current ability to meet urgent needs. The legal rule governing the reference period, the means of updating and the availability of exceptional relief jointly determine whether delayed records translate into wrongful or lawful exclusion.

Pakistan: a documented move toward continuous intake

Pakistan's National Socio-Economic Registry illustrates the distinction between periodic survey rebuilding and an application-accessible system. An official Senate parliamentary response describes the February 2023 transition to the Dynamic National Socio-Economic Registry, with 647 Dynamic Registration Centers and 25 mobile registration vehicles. The institutional figures are an official description of implementation rather than a causal evaluation of benefit access.

A government social protection yearbook for FY 2024–25 reports a registry database of approximately 35 million households and 18.6 million household registrations through dynamic centers in that fiscal year, alongside 25 mobile outreach vans. These numbers measure different things and should not be added. They do not, on their own, demonstrate that all unregistered eligible households had access or that predicted poverty status corresponded to actual benefit receipt.

The Pakistani transition offers testable evaluation questions: Did waiting time between a household shock and a valid registration fall? Which geographic and gender groups used the mobile channels? How many applications remained incomplete? What share of households experienced corrections after reinterview? Did the proportion of newly eligible households receiving benefits change independently of budget size? A before-and-after comparison also needs to account for the period's economic shocks, program rule changes and changes in the population frame.

The Philippines: enumeration, selection and program enrollment

The Philippines' Listahanan 3 provides a clearly dated case of a national household assessment used to identify potential social assistance recipients. The Department of Social Welfare and Development's Listahanan 3 public profile reports 15,487,655 assessed households and 5,599,091 classified as poor. A Philippine government announcement states that the database was finalized in the final quarter of 2021 and that proxy means testing generated the poverty classification. Those counts are assessment outputs at a specific period and are not an estimate of national social protection enrollment.

Listahanan demonstrates why the terms 'poor household,' 'assessed household' and 'program beneficiary' require different columns. The 5.6 million assessed poor households need not equal the number eligible for any particular conditional cash transfer, which may impose age, school attendance, residence and other program conditions. Conversely, households not contacted in the enumeration cannot be categorized as nonpoor by the absence of a Listahanan poverty flag.

A DSWD first-quarter 2025 program report gives a further example of the separation: the report discusses Pantawid Pamilyang Pilipino Program households not found in Listahanan 3 and records how payment eligibility and status checks were handled while assessments continued. That administrative evidence shows a reconciliation problem directly; it cannot be converted into a population-level exclusion percentage without information on the denominator of all legally eligible households and their final status. DSWD also provides an integrated grievance portal, but online submission requirements and complaint resolution performance require independent review.

Indonesia: an integrated registration initiative

Indonesia's Regsosek illustrates a comprehensive socioeconomic data initiative that can serve several policy functions. A 2024 BAPPENAS–UNICEF report describes the 2022 nationwide data collection and its intended role in coordinating planning and social protection delivery beyond the earlier DTKS-based arrangements. The report describes system objectives and institutional design; an intended reduction in exclusion or fraud should not be quoted as an observed realized percentage unless a relevant evaluation provides the evidence.

This case also shows why a nationwide collection exercise is not necessarily synonymous with dynamic registration. A broad original household frame can be valuable, but as incomes, families and addresses change, update channels and versioned records determine continuing suitability. Reuse across agencies should preserve program-specific eligibility law, data-use authority and an intelligible correction path for the household.

Comparative administrative interpretation

The five country cases show different observable features rather than a ranked league table. Brazil supplies a legal and operational shared-registration gateway with explicit periodic update rules; Chile documents how administrative-source latency and defined income periods are reconciled; Pakistan reports a transition from periodic surveys to demand-based registration; the Philippines shows how poverty classification, beneficiary enrollment and grievances occupy different stages; Indonesia provides an explicitly integrative national data-collection program. Because the sources differ in period, unit and purpose, no synthetic cross-country 'exclusion rate' is calculated.

A cross-country table can still be scientifically useful when it identifies documented mechanisms instead of suggesting comparable performance metrics:

Country/system Verified feature What the cited evidence does not establish
Brazil / Cadastro Único Federal legal rules for household updates and cross-program use The nationwide share of wrongfully excluded eligible claimants
Chile / Registro Social de Hogares Administrative-source income reconciliation and 12-month reference Average correction success across all households
Pakistan / Dynamic NSER Official transition and registration infrastructure counts Causal change in national benefit receipt
Philippines / Listahanan 3 Dated household poverty-assessment frame and program linkage National legal-entitlement coverage of all excluded applicants
Indonesia / Regsosek National integration initiative and 2022 data collection Verified effect of the registry on poverty or targeting error

This table is an evidentiary classification, not a new model or comparative score. Its purpose is to prevent the common mistake of inferring performance from architectural announcements.

Evidence Verification and Sensitivity

Evidence classes and levels of inference

This monograph uses four practical categories of evidence: primary legal instruments establish operative rules within their jurisdiction; official administrative documents describe implementation and reported counts; independent statistical surveys estimate outcomes under declared measurement assumptions; and empirical studies with credible designs identify selected comparisons or causal effects. A legal text cannot directly measure actual access. An administrative count cannot by itself measure omitted households. A randomized field experiment informs the mechanism examined in its experimental sample and context, not every unobserved national implementation.

The original experiment by Alatas et al. (2012) supports statements about how evaluated targeting methods perform under a particular welfare benchmark and sampling design. The Brown et al. (2018) analysis supports statements about proxy-means-test performance in nine African datasets. The World Bank's delivery delivery systems Sourcebook and 2022 targeting synthesis support cross-country process interpretations and policy comparisons. These sources differ in causal identification and should not be collapsed into a single quantitative estimate.

Survey ascertainment and eligibility coding

To estimate true program exclusion, a survey must first reconstruct statutory eligibility at the correct date. A survey that records monthly consumption but lacks a disabled person's certified status, a child's birth document, household residence history or verified contribution record cannot conclusively reproduce legal qualification. The analyst should distinguish 'survey-estimated eligible,' 'administratively eligible' and 'legally established eligible.' Imputation must be declared, with sensitivity estimates under alternative coding assumptions.

A survey may underrepresent unsheltered people, remote groups, informal settlements, institutionalized persons, refugees and frequently mobile workers. Some nonrespondents may have systematically different registry access. The resulting exclusion estimate can be biased even where survey weights reproduce the known population totals. High-quality validation requires comparison with geographic frames, civil registration where available, community-led outreach and independent field checks. It also requires protection of confidential respondent and beneficiary information.

Administrative extracts create their own selection problems. A dataset containing completed applications will miss people who attempted unsuccessfully to apply. A dataset of benefit appeals will overrepresent people who were informed of their decision and had sufficient knowledge and resources to contest it. If the assessment depends entirely on cases visible to the administration, unknown omissions remain outside the denominator. The solution is triangulation across an external population frame, direct outreach evidence and appeal and program data, with differences preserved rather than forced into an unsupported single count.

Reference dates and poverty measurement

A targeting evaluation should report at least four dates: household situation measurement, registration or last update, eligibility decision, and actual payment period. It should record whether welfare is measured before or after public transfers, and whether the analysis uses income, expenditure, assets or a statutory score. According to ASPIRE's indicator definitions, benefit incidence and beneficiary incidence are both sensitive to how populations are grouped into welfare quintiles, including pre- or posttransfer welfare. This methodological distinction prevents mechanically attributing a ranking change to registry improvement.

For a threshold-based program, the measurement error near the cutoff is particularly important. The analyst should calculate error rates under plausible shifts in the threshold, measurement period and equivalence scale; report subgroup outcomes; and test whether classification reverses for households with volatile incomes. A small change in a proxy score can trigger a large benefit change when enrollment is discontinuous. Reporting an overall mean absolute prediction error alone will hide that distributive consequence.

Linkage uncertainty, false matches and missing matches

Where multiple registries are matched, record linkage creates false-positive and false-negative matches. A false positive merges records belonging to different people and may incorrectly combine incomes or assets. A false negative fails to connect records belonging to the same person and can create duplicate participation or an inability to establish eligibility. Quality metrics need an independently reviewed match sample and should be reported by subgroup and confidence threshold. A global match-rate figure can conceal concentrated errors among people with different naming conventions, changing addresses or incomplete civil documentation.

Reconciliation should retain a source and timestamp for each material field and a durable change log. When corrected information overturns an adverse decision, the downstream program should receive an effective correction and apply the relevant remedy for past benefit periods under applicable law. A correction recorded only in the master database may leave the person excluded from payments or services; the technical event and the legal remedy must be evaluated together.

Privacy, due process and automated decisions

Social registry modernization expands the number of institutions that can access sensitive data about earnings, disability, family status, location and benefits. The legal basis for collection and interagency sharing must be specified for each jurisdiction, along with purpose limitation, access controls, retention and accountability. The European Union's General Data Protection Regulation states principles of lawfulness, fairness, transparency, data minimization and accuracy in Article 5, and addresses automated individual decision-making with legal or similarly significant effects in Article 22. Its application to specific public benefit processing depends on the operative legal framework, available exceptions and safeguards; the GDPR is a European instrument and is not a global social registry statute.

The United Nations Special Rapporteur Philip Alston's 2019 report on digital welfare identifies risks of opaque automation, surveillance and weak accountability in public assistance systems (Alston 2019). This is an official human-rights analysis rather than an empirical estimate of overall error rates. It reinforces an administrative point: the ability to compute or cross-check an eligibility score is insufficient if households cannot understand, contest and correct a consequential decision.

Human review must be substantively capable of changing the outcome. A reviewer needs access to the source facts, applicable rule, data lineage and relevant alternative evidence. If a caseworker can only accept or reject an opaque output without correcting the underlying record, nominal manual review may leave the exclusion mechanism intact. Audit quality therefore depends on both traceability and remedial authority.

The ILO's privacy and appeals provisions support a joint evaluation of data protection and benefit access. Excessive collection may discourage application or expose households to misuse. Insufficient access to accurate information may produce incorrect decisions. These requirements should be addressed through lawful and proportionate data governance, not by assuming that all data linkage is beneficial or harmful.

A minimum empirical audit specification

An adequate registry audit should publish the reference period, program entitlement rule, age and household eligibility scope, administrative data sources, external population frame, record linkage method, independent validation strategy, outcome definitions and uncertainty. It should then report a cohort-based progression: potentially eligible population; people reached by outreach; attempted applications; completed registrations; initial classifications; program applications; final eligibility decisions; enrollments; timely payments; appeals; successful corrections; and back payments where lawful.

For each transition, the report should count people and households separately, calculate elapsed times, disaggregate by substantively relevant characteristics and document missing observations. Error rates must be defined relative to both the statutory test and any separate poverty or welfare benchmark. The audit should preserve reason codes for rejection, technical failure, missing identity records, uncertain matching and pending appeal; otherwise one aggregate 'not selected' category obscures actionable problems.

The minimum sensitivity program includes alternative eligibility coding near thresholds, reference-year changes, household-size definitions, probable duplicate handling, unresolved match cases, survey nonresponse bounds and missed-population scenarios. Confidence intervals are appropriate for sample-based statistics; administrative census counts instead require data-quality uncertainty and completeness assessments. None of these checks can be replaced by a general statement that a registry is 'accurate.'

Scientific Synthesis and Research Agenda

What the comparative record establishes

Across the documentary and empirical sources, three findings are mutually reinforcing. First, being recorded is an intermediate administrative status, while the public protection objective concerns enforceable access and effective receipt. Second, a targeting score is only one source of possible classification error; upstream registration and downstream delivery can dominate observed failures. Third, temporal mismatch is endemic in socioeconomic administration: accurate historical attributes can yield decisions that do not correspond to present circumstances, particularly where an eligibility test uses a fixed lookback window or outdated census data.

The country's institutional design alters how these problems manifest. Brazil's legal update requirements and shared gateway allow examination of notice and cross-program propagation. Chile's published income reconciliation rules make it possible to locate a mismatch between current and historical earnings. Pakistan's documented dynamic registration shift enables an analysis of access intervals and newly registered households. The Philippine assessment and program interface highlight differences between identified poverty and actual beneficiary enrollment. Indonesia's integration initiative provides a basis for studying cross-program reuse. These are source-grounded contrasts, not causal rankings or claims of a universally optimal national design.

The empirical literature adds necessary discipline. Alatas et al. (2012) show that alternative targeting methods rank differently under a consumption benchmark and according to beneficiary perceptions. Brown et al. (2018) show that methods with improved predictive specifications still face substantial poverty targeting limitations. Hanna and Olken (2018) examine tradeoffs between targeted and broadly distributed benefits under fiscal constraints. Together these studies support context-sensitive assessment of administrative design and distributional consequences. They do not permit a global numerical prediction of social registry errors.

The relationship between inclusion and institutional accountability

Two registries can publish the same measured exclusion rate yet generate different social consequences. In one, households may be able to correct the error within days with a payment adjustment. In another, there may be no accessible appeal, and a household may lose several months of support. The classification rate alone fails to capture duration, procedural burden and lost value. A rigorous evaluation should therefore report both point-in-time error prevalence and the length and financial consequence of unresolved erroneous denials.

The analytical contribution of this monograph is to make the full sequence of decisions explicit and to identify where an observed failure can arise. It does so by combining established delivery-system distinctions, standard statistical classification measures and jurisdiction-specific documentary evidence. This is an original comparative synthesis, not an assertion of a new model, doctrine or discovery. Its core claim is testable: a registry's improvements in matching or predictive classification will not necessarily increase receipt among all eligible persons when outreach, registration, program financing, processing capacity or payment pathways remain restrictive.

The practical implication is that a registry modernization should have an evaluation plan established before deployment. The plan should compare eligible population frames before and after the change, measure entrance and exit from the registry, record correction and appeal outcomes, and examine changes in actual payments. If only the database size grows, the administration has measured expansion of recorded information. If beneficiaries rise, it has measured increased program reach. Neither event alone proves a reduction in wrongful exclusion because the underlying eligible population may also be changing.

Research priorities and identifiable evidence needs

Future comparative research should use longitudinal household panels to test how quickly vulnerability shocks are recognized by the registry and translated into payments. Independently sampled applicant and nonapplicant households should establish whether a program's omissions arise before registration or during scoring. Documented linkage-error studies should compare administrative matches with ground-truth verification, including subgroup differences by geography and identity documentation. Appeal outcome studies should distinguish procedural access, time to correction, original legal entitlement and recovery of unpaid benefits.

Quasi-experimental studies of staggered office openings, outreach interventions or registration reform may help attribute changes in take-up where credible comparison groups are available. Researchers should verify parallel-trend assumptions where applicable and account for simultaneous changes in benefit budgets and eligibility law. Randomized evaluations of communication methods or assistance with applications can identify effects on specified populations, provided consent, ethics and service access are protected. No national causal effect is inferred here without such an identification design.

Cross-country reporting would benefit from a consistent minimum disclosure standard: eligibility law and date, population frame, unit of observation, registration status, direct and indirect beneficiary treatment, income reference period, inclusion-error convention, exclusion-error denominator, payment period, appeal disposition and data-quality warnings. This is a recommendation concerning disclosure of established quantities, not the creation of an independent theoretical framework.

Research on rights and administrative burdens should also investigate households that never appear in any administrative record. Enumeration by its nature gives more precise information about those already observed than those absent. A registry-only evaluation faces a structural blind spot. Representative household surveys, targeted independent enumeration, qualitative research with unregistered communities and documented population benchmarks are therefore essential complements to database analytics.

Institutional implications for public protection

For policymakers, the strongest operational sequence is to establish the lawful benefit objective and eligible population; make outreach and registration accessible; collect information proportionate to the decision; specify which source and reference period control each field; disclose comprehensible decisions; preserve correction and appeal rights; and measure eventual payment. Doing these things requires administrative capacity, data safeguards and benefit financing. Greater computational complexity cannot substitute for any one of them.

For auditors, routine monitoring should separate three categories of findings: statutory and procedural noncompliance; distributional outcomes relative to a poverty or vulnerability benchmark; and data-system quality failures. These categories may overlap but require different remedies. A lawful but narrowly designed benefit may require legislative reform. An incorrect income match requires record rectification. A missed payment requires operations and payment-system correction. A person never contacted requires outreach and registration redesign.

For researchers, effect claims should be proportional to evidence. Official ministry pages establish current administrative statements; laws establish the operative text, with amendments and implementation requiring separate checks; development institutions offer comparative descriptive syntheses; peer-reviewed experiments provide evidence under their designs. The resulting scientific account can be both substantive and precise without presenting untested policy preferences as empirical results.

Conclusion

Social protection registries mediate access to public assistance and other services by deciding which people become administratively visible and which facts are used to assess them. Their quality has distributive consequences that extend beyond database integrity. An accurate record, an accurate targeting score, a legally correct determination and a delivered benefit are successive but distinct achievements.

The evidence supports continuous and accessible registration, context-sensitive use of socioeconomic information, auditable reconciliation, and effective remedies as important design conditions. It also demonstrates why no single enrollment count or targeting coefficient measures the success of a protection system. International standards establish rights-based objectives and procedural safeguards. National law determines benefit qualification and correction rights. Program administration determines whether people can reach those rights in practice. Independent statistical evidence makes omitted and misclassified populations observable.

The appropriate scientific result is an integrated assessment of the path from need and entitlement to actual receipt, with dates, denominators and error definitions recorded at every consequential stage. Comparative studies should value both the reduction of invalid payments and the prevention and timely repair of wrongful exclusion. The proportion of people that a registry can describe is meaningful; the proportion of eligible people who actually obtain lawful, timely and adequate protection remains the decisive public-policy outcome.

References