Key Concepts and Practice of Targeting POV Summer University Phillippe Leite Based on 2017 SSN core course presentation by Leite and Tesliuc, and Targeting debate presentation by Yemtsov et all 2017 1
Why target a SSN program? Support the implementation of the Social Protection Strategy
Monetary poverty Social Protection Strategy within a multidimensional poverty framework P L EPL EXTREME multidimensional poverty ESVL MODERATE multidim. poor SVL Decent and economically independent lives Social vulnerabilities Social policies aims to protect the poor, prevent poverty, build resilience and build economic empowerment Equity and Efficiency Supporting systems to implement social development interventions Social Registry Case Management System
A good targeting system provides Transparency and consistency Clear and consistent application of centralized criteria Low political interference and manipulation Maximum inclusion of the target population People who think they are eligible should be able to apply on an ongoing basis Budget and outreach Minimum leakage to the not desired As technically possible to the near poor, errors rather than fraud Cost-efficiency Under 10% of costs at scale 4
Challenge 1: Targeting when everybody needs? Focus on children: not losing the next generation, politically acceptable (even if they do not vote) AIDS and its stigma Giving transfers to children? When poverty (crisis) is very deep: Should you target the poor who have a chance? Should you give a chance to those who would sink? Households with «able-bodied» workers or not (who defines?) We know the PMT does not function very well Source: Kenya CT-OVC Who takes the decision? Make the criteria as extensive as possible to minimize the arbitrariness at the local level but politically difficult How to support communities, build appeals and grievance and genuine participation? 5
Challenge 2: Implementation Despite the method, implementation matters a LOT for optimizing targeting outcomes Moving from population to beneficiary is not simple. General population Budget implications, coordination, administration and transparency Target population Budget, develop a Monitoring and Information system, determine a targeting method; design an information and outreach campaing, ensure low cost for potential beneficiaries, set payment level 6
Challenge 3: Politics matters Development and poverty reduction are intrinsically political Reaching the desired population is a particular challenge Do they deserve it? Will richer & more powerful groups support investments for the poorest? Research shows that politics has been central to the success and failure of social protection Politics viewed here as an enabling as well as constraining force 7
Challenge 4: Poverty may be linked to your objective Malnutrition Poor education Poverty Unemployment, underemployment Vulnerability Targeting on your objective may undermine it The malnourished children of Bolsa Alimentação The orphans in Kenya Sometimes other categories may work Widows in rural Africa Families with no able-bodied workers 8
Challenge 5: Targeting a program or a system? A Social Registry may be used for different programs with different cut-offs Applicant Beneficiary Eligibility Enrollment Use different sets of the information (multi-dimensions of poverty) => a planning tool The idea is to focus programs on the needs of poor households and communities Cadastro Unico (Brazil) and popular housing, training and literacy, micro-credit Ethiopia: efforts to merge different databases Respect confidentiality/privacy among different systems. 9
But 10
Targeting is NEVER perfect Never 100% accurate What do these errors cost? Efficiency Social and political capital Inclusion: Media attention Exclusion: disenfranchisement What does it take to address them? A fine balance between the costs of accuracy and errors and the goals of targeting. 11
Inclusion and Exclusion Errors Income or Consumption, per capita or adult equivalent Overall Population Non-Poor population Eligibility Threshold Errors of Inclusion Of Non-Poor Beneficiaries of social Assistance Program Errors of Exclusion Poor Population 12
The treatment of Bolsa Familia in the media For your information Source: Lindert and Vincensini, 2010 The press paid more attention to inclusion errors in electoral periods 13
How to target? Methods 14
Targeting methods Categorical Geographical (Proxy) Means Test Combination Self-selection Communitybased
Geographical targeting When location is an important determinant of poverty Macro regions Micro-area poverty maps: based on census and household surveys Can be important when administrative capacity is low Often used as a first step: Panama s Red de Protección Social (CCT) Program 16
Self-targeting Open to everyone but only the poor will be interested Food subsidies of staples consumed by the poor: are they really consuming less? Midly progressive at best. Little exclusion and stigmatization but high inclusion errors. Example: Food subsidies in MENA Labor intensive public works with wages set very low: works for targeting. Stignatization can be high, exclusion errors can be high. Example: Trabajar in Argentina Some elements of self-targeting in a lot of programs: long waiting lines, compliance with conditionalities 17
Self targeting through consumption subsidies PROS Administratively simple Few errors of exclusion Universal benefit may be politically very popular CONS Hard to find really inferior goods May be hard to transfer large amounts Hard to reform Technical Requirements An inferior good with a suitable marketing chain A service supplied by public and private sector where amenities can differ Appropriate Circumstances Low administrative capacity For your information 18
For your information Self-targeting through workfare PROS Administratively simple Keeps work incentives Eliminates concerns about shirkers Automatic exit criteria CONS Organizing public works is not administratively simple Not applicable for many programs or target groups Foregone earnings reduce net benefit Technical Requirements Wage set below going wage for hard, physical labor A works program that does high value-added projects Appropriate Circumstances Unemployment; Crisis and chronic poverty settings 19
Categorical (demographic) targeting Characteristics that are linked to poverty or vulnerability Age: pre-school children and old-age Marital status: single parent Ethnicity: scheduled castes in India, native American PROS Administratively simple Low cost CONS Weak correlation with poverty Technical Requirements Good civil registry Appropriate Circumstances When targeting specific vulnerabilities (malnutrition) 20
Community-based targeting Uses a group of community members or leaders (whose functions are not related to the program) They must identify those most in need according to program criteria (often OVC, elderly, hh w/o able-bodied adult) Good results Community meeting SCT Zambia 21
22
Community-based targeting PROS Good information Low(on the books) administrative cost Local monitoring may reduce disincentives CONS Unknown effects on roles of local actors Costly for the community May reinforce existing power structures or patterns of exclusion May generate conflict and divisiveness Local definitions may vary Technical Requirements Intensive outreach to decision-makers Cohesive, well-defined communities Appropriate Circumstances Low administrative capacity Strong community structures, political economy Low benefit that must be finely targeted Cost to communities Scalability 23
Proxy-means testing Target group Indicator well-being Households with low consumption levels Consumption, per capita or per adult equivalent Eligibility criteria? Based on a representative household survey with information on consumption and household characteristics Consumption is estimated based on observable household characteristics correlated with poverty, based on representative household survey Most frequent technique: regression model on log consumption Variables used to estimate consumption include: location, housing quality, assets/durables, education, occupation and income, and a variety of others (disability, health, etc.) The regression model is used to estimate a household score The score is compared to a threshold, determined to separate the predicted poorest x% of the population from the rest Appropriate conditions high degree of informality, seasonality, or in-kind earnings; chronic poor are the target group; benefits will be granted for long periods of time 24
Selected Not selected Total Exclusion Inclusion Niveau de vie Pauvre 560 256 816 Non pauvre 113 150 263 31.4% 16.8% Quintiles de bien-être économique Q1 556 236 792 Q2 28 61 89 29.8% 17.1% Q3 87 31 118 Déciles de bien-être économique D1 521 205 726 D2 35 31 66 D3 5 30 35 28.2% 19.0% D4 23 31 54 D5 59 22 81 Total 673 406 1079 25
70 60 50 40 30 20 10 0 20.6 16.7 14.6 17.6 37.5 Colombia 51.3 53.1 54 Inclusion errors increasing over time: how to deal with 12.9 14.4 13.0 14.7 49.3 EE EI EE EI 61.9 63.7 66.4 Type of household- Sisbén Vs ECV 78.869.8 Sisbén III 2011 Sisbén III 2013 ECV 2011 ECV 2013 95 95.6 Casa o apartamento 20.729.6 Cuarto 4.4 4.3 Pob_ingreso IPM 2008 2010 2011 2012 Type of sewage Sisbén Vs ECV 73.2 64.9 Sisbén III 2011 Sisbén III 2013 ECV 2011 ECV 2013 94.6 91.4 17.4 25.5 5.4 8.6 Source: DNP De uso exclusivo del hogar Compartido con otros hogares
Sisbén New SISBEN Information system designed to identify potential households beneficiaries for social programs, and be used by local authorities and implementers of social policy on the national agenda. Optimizing its operability Update PMT Increase internal validation and checks Improve IT platform Offer additional services to improve targeting Characterizing the population Use spatial information Use local variables Work with local authorities Strengthening interinstitutional relations Set the norms and rules Define Interoperability Have a better information flow
http://documents.worldbank.org/curated/en/2014/06/22671265/effective-targeting-poor-vulnerable
Means Testing (MT) Target group Indicator of wellbeing Eligibility criteria? Appropriate conditions Households with low levels of income and assets (means) Income, per capita or per adult equivalent Asset ownership or value (wealth) Based on a representative household survey with information on income and assets Eligibility threshold = income cut-off that separate the target group (poor) from the rest Administrative vs economic income: not all types of income sources may be included, but the large majority are Issues: income under-reporting in the survey, informal income Assets: used to filter-out asset-rich households Pilot before going to scale Incomes, expenditures, wealth are formal, monetized and well-documented; Benefits are high OECD 29
Example: Means-Tested Programs in US For your information 1. Eligibility determined based on a number of tests: Asset tests: Financial assets compared to threshold; Vehicles (fair market value) Gross income tests: Gross earned and un-earned income Compared to threshold Takes into account household size Net income tests: Gross income minus set of standard deductions Compared to threshold Takes into account household size 3 0 2. Extensive verification of information: Verification mainly covers two aspects: Identity and household composition Income, assets Two main instruments for verification: Documentation Automated computer matches, verification 3. Benefit calculations: Benefits calculated as the maximum benefit minus the income Taking into account household size Results in graduated benefits
Hybrid-Means Testing (HMT) Target group Indicator of wellbeing Households with low levels of means: income & assets Income, per capita or per adult equivalent Asset ownership or value (wealth) Eligibility criteria? Based on a representative household survey with information on income and assets Administrative income = formal income sources + estimated informal income Eligibility threshold is the administrative income cut-off that separates the target group from the rest Assets: used to filter-out asset-rich households Pilot before going to scale Appropriate conditions More than half of income in formal sector, good asset or business registries; Benefits are moderate to high Central and Eastern Europe, Central Asia, Brazil 31
Implications for the design of HMT programs Effective HMT programs use: 1. Verification of formal income 2. Imputations to address the underreporting of informal income 32
Hard & Easy to Verify Income Bulgaria Kyrgyz Republic 33
PTM or HMT? Pros &Cons PMT captures long-term, chronic poor; harder to capture welfare changes Census registration approaches with limited or absent program frontline units results in static caseloads for many years, followed by political economy issues at recertification Calibration of a new PMT model depends on the availability of a recent HH survey with a good consumption module HMT can capture welfare changes, both chronic and recent poor Depends on the availability/ quality of administrative data, frequency of updates. Eligibility criteria can be easily improved Open eligibility depends on availability of local offices close to beneficiaries HMT less reliable for: (i) groups relying on informal earnings 34
Any method Any Means tested Proxy-means tested Community assessment Any Geographical Age:Elderly Age:Young Other categorical Any Public works Consumption Community bidding % of benefits / % of population No single method is best Targeting performance by targeting method 2.5 2.0 1.5 1.0 Huge variation within method according to implementation 0.5 - Individual assessment Categorical Any selection method 75th perc. 25th perc. Median Handa et al., CBT 2010 Coady, Grosh and Hoddinott, 2004 35
Eligibility Criteria for Programs Eligibility determined based on a number of tests For example: Net income tests: Net income Less a income disregards Normalized per adult equivalent or per capita Compared to threshold 3 6 Asset tests: Asset value compared to threshold (e.g. financial savings) Yes/No filters (e.g. second house, vehicles) Proxy CBT Extensive verification of information Documentation provided by applicant payroll statements, benefit letters, banking statements, house ownership and vehicle documentation, etc. Third party documentation, usually automated tax records, social security registry, unemployment registry, banking information Benefit calculations Benefits level = Fixed + variable Or maximum benefit minus administrative income Or Taking into account household size Or access to services
Complex Administrative Infrastructure Supports programs Frontline units close to beneficiaries Country Number of administrativeterritorial tiers, and total population Subnational tiers involved in program administration Regional level Local level Albania 2 tiers, 3.6 million 12 Regional Service Administrations 385 offices; Armenia 2 tiers, 3.2 million 11 Departments 55 Centers Bulgaria 2 tiers, 7.2 million 28 Regional Directorates 272 Directorates Kyrgyz Republic 3 tiers, 5.2 million 7 oblast Departments 40 rayon Departments; 477 rural local governments Lithuania 2 tiers, 3.5 million No role 60 Departments; 550 wards Romania 2 tiers, 21.5 million 42 Directorates of Social Assistance 3,176 local governments Uzbekistan 37 382 rayon Departments; 3 tiers, 25 million 12 Oblast Departments 12,000 mahalla committees
Complex Administration cont d Census-Sweep, Beneficiary list or on demand registration (self-selection) The composition of assistance units, formal incomes, and some assets are verified including through home visits & via cross-checks Frequent recertification and mandatory updates of documents (quarterly or annually) Sometimes additional conditions (community works) 38
Targeting has costs (remembering SR) Intake Registry Lots of set-up costs, as programs scale-up Difficult to measure b/c of shared staff and functions Documents (IDs, proof of status) Need to go to an office, spend time, work requirement in workfare Stigma (public list) Work effort: benefit levels, sliding withdrawals, periodicity Crowding out private transfers or complementing them Fertility effects: quantity and quality of children Is a program for the poor a poor program? 39
Targeting costs vs other program costs Conceptualizing Administrative Costs Illustration based on a Categorical vs Targeted Child Allowance Program 40
Administrative costs: Despite the programs complexity, admin costs are low to moderate 41
Combining methods may improve accuracy Often a first step is geographical targeting Then collect some information at the household-level Triangulate from several sources: Respondent Community Administrative records at local and central level Grievance and redress mechanisms No matter which combination, implementation is key. 42
Targeting debate 43
Recent debate on targeting Emerged in the context SPF, SDG and USP Populated by critiques of the targeting approaches from more serious (broadly focused on defining poverty ) Promoting Millennium Development Ideals: The Risks of Defining Development Down - Lant Pritchett and Charles Kenny, CGD Working Paper 338. Washington, DC: Center for Global Development. 2013 to pure advocacy writing Exclusion by design: An assessment of the effectiveness of the proxy means test poverty targeting mechanism - Kidd, Stephen; Gelders, Bjorn; Bailey-Athias, Diloá - International Labour Office, Social Protection Department (SOCPRO). - Geneva: ILO, 2017 ESS Working Paper No. 56); Proxy Means Testing: it s Official! - Nicolas Freeland Development Pathways, March22nd 2017; Has seen an escalation after the publication of Brown, Ravallion and de Walle, 2016 A Poor Means Test? (henceforth BRW) published by The World Bank (Policy Research working paper series) 44
What critiques are getting? RIGHT Econometric OLS-based PMT has large errorsespecially to predict extreme poverty (poorest Critics are based on simulations as a test on actual performance of targeting methods. 5-10%) OLS-based PMT Targeting is challenging in low income countries. Those are often also the places with low capacity and low budgets relative to need. We face a large development problem, such as how to make more efficient use of limited budget PMT (with or without enhancement) in itself is not the solution to it. WRONG Targeting performance is measured only via exclusion and exclusion errors or impact on poverty, while measuring performance considers its efficacy to identify the correct group, the coverage, generosity and distributional impacts. Impact evaluation, targeting assessments of actual SSN projects not part of the discussion PMT are presented as one time formula setting but policymakers should know that it have to be recalibrated with actual data on beneficiaries. There is no documented best practice advice on Most likely the [implementation] will lead to methods targeting by the WBG (most is in authored performing less well than our calculations suggest. No, papers) there is evidence that it can be improved. At a country level, there are times when WBG is PMT are promoted by the WBG in 9 countries covered associated with advocacy for PMT in the paper- Completely wrong. 45
What else critiques are missing? (1/2) Critiques claim : Accuracy of targeting is delusional, because poverty is multifaceted and hard to measure. But good targeting is not synonym of poverty reduction. Significant work on improving the targeting is conducted at the Bank. The use of quantile regressions in AFR was promoted and presented in Del Ninno & Mills (2015) and have been promoted, informally, since then in many countries (BRW findings is not new) BRW show that PMT is performing only slightly better than random or universal assignment. This is not true everywhere, and only for FGT0 in few cases, and no other accuracy measure as FGT1 was considered Debate is not new. History of interaction on targeting: articles, blogs and face to face discussions AUSAID/DFAT; and practice of using PMT where appropriate these so far have not influenced the critiques. 46
What else critiques are missing? (2/2) Critiques offer no real alternative. BRW end up saying that enhanced PMT is still the least worst option and other critiques are saying that in limited budget there is nothing inherently wrong with PMT to ration the access, what is wrong is a claim for accuracy. But that is precisely our practice approach. PMT is a method not a policy. Program implementation matters. Rarely, poverty targeted programs are designed in isolation, most typical - as part of a deeper reform that would prioritize support for a population that is not benefiting from overall growth policies or universal access policies. Targeting must not be seen as a social policy as many foreseen it. Targeting is a method that supports the principal of universality, equity and human rights and if properly applied, enhances the effectiveness of universal social programs. 47
WBG key publications on targeting Paper Topics Targeting of transfers in developing countries. Coady, Grosh, Hoddinott. 2004. The cost of poverty alleviation transfer programs: A comparative analysis of three programs in Latin America. Caldes, Coady, Maluccio. 2004. - Targeting methods and costs - Targeting performance (253 cases) - Performance and cost-efficiency for three Latin American countries Designing and implementing Household targeting systems: Lessons from Latin America and United States. Castaneda, Lindert, de la Briere. 2005. - Provides operational guidance on targeting processes - Targeting performance for 6 programs in Latin America From Protection to Promotion. Grosh, del Ninno, Tesliuc, Ouerghi. 2008. - Guidance on targeting options - Guidance on selection of methods Should Cash Transfers be Confined to the Poor? Pablo Acosta, Phillipe Leite and Jamele Rigolini 2011 Targeting Performance and Poverty Effects of Proxy Means-Tested Transfers: Trade-offs and Challenges. Stephan Klasen, Simon Lange 2015 - Use of simulations to compare performance of targeting (PMT) and universal approaches. - Trade offs between accuracy and poverty effect - PMT does better Community Based Targeting: A review. McCord. 2012. - Review of CBT approaches and performance Effective targeting mechanisms for the poor and vulnerable in Africa. Del Ninno and Mils. 2015. Last-resort income support to the poorest: A review of experience in Eastern Europe and Central Asia. Tesliuc, Pop, Grosh, Yemtsov. 2015. - Description of targeting methods - Detailed description of PMT models and their accuracy - Innovative work on identifying the vulnerable for shocks - Hybrid means test and use of filters - Eligibility analysis - Detailed analysis of targeting costs (so far unique) Operationalizing targeting. Eric Zapatero and Ana Verónica López 2015 - Detailed description of methods and processes of implementation, country studies How to target households in Adaptive Social Protection Systems? Relative Efficiency of Proxy Means Test and Household Economy Analysis in Niger Pascale Schnitzler 2016 - Latest paper on modifications of PMT 48
Conclusion Targeting is complex A single method does not dominate another Combination can work but attention is needed on the implementation arrangements Implementation arrangements have much in common: Verification strategies home visit versus computerized cross-checks of databases Outreach, re-certification, quality control, system design, staffing, etc. Performance of targeting can not be measured only as simple function of FGT0 or type I or II errors. Targeting performance can be measured in terms of its efficacy to identify the correct group, but performance is also a function of coverage, generosity and distributional impacts. 49
Conclusion Targeting is a method, not a policy Countries need both targeted and universal policies within a fiscal constraints. There is a full continuum of methods not the polarity Targeting is hard and needs constant development of methods. Well-designed poverty targeted programs acknowledge the challenges of identifying the desired population. Rarely, targeted programs are designed in isolation, but as part of a deeper reform that would prioritize support for a population that is not benefiting from overall growth policies or universal access policies. As such targeting is taken as part of overall social protection system reform. 50
Conclusion The World Bank does NOT advocating for specific formula or approach Adapting to context (political, administrative, technical) is necessary for any targeting to work In fact PMT is not the main approach in our project (see next slide) We are aware of PMT issues, this is why we already have developed methods to improve it or combine with other approaches Historically the focus was on minimizing inclusion errors, but it is shifting increasingly to reducing exclusion errors 51
Source: Operationalizing targeting, 2015, survey of TTLs/PADs, 155 programs 52
Conclusion Combining methods may improve accuracy Often a first step is geographical targeting Then collect some information at the household-level Triangulate from several sources: Respondent Community Administrative records at local and central level Grievance and redress mechanisms No matter which combination, implementation is key. 53
Conclusion Implementation matters Lowering barriers to participation Effective dissemination of information about the program and Minimize visits and waiting for application Minimize documentation required, free-of-charge provision of documents attesting eligibility Introduction of one-stop or one-window system; Single application for multiple benefits Lowering errors Use multiple targeting methods combined and Cross-check the information provided by applicants against other public databases; Perform home-visits to assess the means of the households and Frequent recertification Improving program administration: MIS, Staff training, Coordination,... 54
More information www.worldbank.org/safetynets Enrollment in the Safety Net, How-to Note Grosh, del Ninno, Tesliuc & Ouerghi, From Protection to Promotion: The Design and Implementation of Effective Safety Nets, Chapter 4 Tesliuc, Pop, Grosh & Yemtsov, Income Support for the Poorest: A review of experience in Eastern Europe and Central Asia Governance and service delivery, in SSN working papers series 55