1 9/22/03PP253/ARE253 Alain de Janvry & Elisabeth SadouletFall 2003Handout #4Social programs and the art of targetingI. Targeting: How to reach the deserving population?- Targeting is an instrument to make programs more effective for the chosen purpose.- Important for cash transfers programs, safety nets programs, education/health/nutrition programs.- If objective of the program is poverty reduction, targeting requires identification of the currently poor.If objective of program is other (e.g., education, health, nutrition), targeting requires identification of peopleat risk of not going to school, ill-health, malnutrition.- Targeting has:Costs: e.g., identify who is poor and who is non-poor.Benefits: decrease errors of exclusion of poor and inclusion of non-poor.- Targeting is very difficult due to hidden/asymmetrical/strategic information (masquerading).II. Errors in targeting: errors of exclusion (Type I) and inclusion (Type II)Error of exclusion (Type I): categorize a poor as non-poor = poor are excludedError of inclusion (Type II): categorize a non-poor as poor = non-poor are includedInterpretation: classify households into two groups poor and non-poor according to the approximate measureˆy of income (y measured with error or proxy of poverty), and a corresponding poverty threshold ˆz.Let fypoor(y)ˆ() be the distribution of true income y (unknown) of the sub-population classified aspoor, and fynon - poor(y)ˆ() be the true distribution of those classified as non-poor.f y poor(ˆ y )()fynon - poor(ˆ y )()Income ypoverty line: zType IIType IClassification on the basis of approximate measure ˆy“Poor” “Non-Poor”Poor OK Type I errorClassification on the basisof true measure yNon-Poor Type II error OKNote: Can decrease Type I by increasing z (which increases the number of “poor”), but this increases TypeII error. Use a loss function to choose the optimum ˆz to minimize the aggregate cost of Type I and Type IIerrors. The loss function could weight Type I error more than Type II.Note: Different criteria for measuring poverty (P0, P1, or P2) lead to different targeting of welfare budget tominimize poverty.III. Methods of targetingBroad targeting (D. van de Walle): Target types of spendings used more by the poor (baladi bread in Egypt,primary education, health centers) and give equal access to all.Narrow targeting: Target categories of people. Alternative options to do this:1. Means tests (compare income or expenditures to poverty line): Observing income or expenditures isexpensive or imprecise.2. Indicator targeting (tagging, categorical targeting, statistical profiling): Use correlates of povertyi) Use individual assessment by welfare agents: subjective assessment, poverty rankingii) Use correlates of povertyPoor area, ownership of durables, quality of housing, type of employment, gender, ageiii) Use two-step procedure to predict incomeStep 1: Use income and expenditure survey to estimate income equation: y = f (X),where X are easily measurable variables such as durables owned, quality indicators of housing,occupation in farm work.Step 2: Run census of population to observe X for all households (Progresa census)Use ˆfX() to predict household incomes.Progresa in Mexico (7 determinants): secret formula to avoid masquerading.If predicted income is below poverty line, declare as poor. But imprecise at the individual level (sameproblem as in poverty maps) due to large confidence interval over predicted income.Useful for small area targeting (say groups of > 3000 people) like localities or municipalities (like in theconstruction of poverty maps): average income prediction has a variance that falls with the square of thenumber of people in the group.iv) Use asset position: Land ownership. Grameen Bank (land holdings £ 0.5ha)But large heterogeneity among small holders due to off-farm activities and transfers (pluriactivity).v) Use place of residence: Geographical targeting. Easy, but large leakages (Type II errors) if intra-regional heterogeneity is high. Precision declines with size of area.Location of fair price shops in poor neighborhoods (Egypt, India).Welfare programs targeted at marginal localities:- CONAPO marginality index in Mexico used by Progresa to target communities (followed byplacing a poverty line in the community to separate poor from non-poor).- PRAF in Honduras (with universal targeting in poor communities).India: Geographical targeting by states ineffective due to high income inequality within states.China: Geographical targeting by communes effective as low intra-community disparities, but largeinter-community disparitiesLatin America: Targeting by municipality not effective as intra-municipal Gini about as high as thenational Gini.3. Self-targeting: Queuing, location of services (transactions cost), poor people’s food, relief work program:use mechanism design principle to achieve self-selection.Note: Self-targeting is achieved by imposing a cost on participation that is higher for the non-poor than forthe poor.Effective cost: location of stores.Opportunity cost of time: workfare, queuing.Utility gain: inferior foodsHowever, the cost on poor reduces the net gains to participants.2 9/22/033.1. Queing: Imposes cost of time for access. Assumes that opportunity cost of time is less for thepoor. But cost of time may be irrelevant for non-poor household members (use maids, non-workingwives).3.2. Inferior goods: Yellow corn in Mozambique (David Sahn). Consumption declines as incomerises (MUpoor > 0, MUnon-poor < 0). This, however, restricts food subsidies to inferior foods whichhas a cost on the poor (they would prefer non-inferior foods).3.3. Guaranteed employment schemes: workfare versus welfare. (Besley and Coate, AER 1992)Conditional targeting: benefit from program if agree to work.Self-targeting: poor choose to participate; non-poor self-exclude.Example: India’s Maharastra Employment Guarantee Scheme, food-for-work programs (World FoodProgram)Assumptions:Two types of workers:Low ability (L) (also called poor) with frequency g and wage wLHigh ability (H) (also called non-poor) with frequency 1 -()g and wage wH.Poverty line: zIncome: y, with yL < z < yHTime worked in the private sector: lMechanical design: Workfare contract = (b, c):Cash transfer: bRequired time worked in the public sector (non-productive labor): ci) Targeted welfare program (full information)The policy maker can identify
View Full Document
Unlocking...