decentralized targeting of anti-poverty programs
TRANSCRIPT
Decentralized Targeting of Anti-Poverty Programs
• Central government delegates authority over the targeting and delivery of services to local communities,
• But retains control over the allocation of budgets to local government areas.
Trade-off? Information vs. accountability? • Yes, better information is available locally on who is poor
(lower costs monitoring, enforcement, local needs)• But are local organizations accountable to their poor? • Local capture when communities are heterogenous
(Bardhan and Mookherjee) • Anecdotal evidence of program capture by local elites
Community-based targeting
What determines outcomes for the poor?
• Targeting of communities by the center, or of families by the communities?
• What explains inter-village differences in performance at reaching the poor? – Does targeting performance vary with amount
received from the center?– What role is played by other factors, including local
institutions?
Two case studies of decentralized programs: Bangladesh’s Food-for-Education Program and Argentina’s Trabajar Program
Case Study 1: Bangladesh’s Food for Education Program
• Food to households with primary school children, • conditional on attending 85% of classes• 2 million participants in 1995-96
Evidence of significant gains in school attendance • A stipend with a value < ½ the mean child wage was enough
to assure nearly full school attendance for participants. • Only modest foregone income through displaced child labor,
so sizeable transfer benefits (Ravallion and Wodon, 2000).
But has the program reached the currently poor?
Two stages of targeting
1st stage: • “Economically backward”/”low literacy” Union Parishads
are chosen; one from each Thana, then top up.
• Anecdotal evidence of political manipulation/lobbying
2nd stage: • Households are identified within each Union by the
School Management Committee (teachers, parents, donors, local representatives)
• Eligibility criteria (widows, day-laborers, lowly occupations, landless, children)
1: Theory and measurement
Theory of Community-Based Targeting
• The center
– allocates a fixed aggregate budget across communities:
– does not know how the budget is allocated within villages
• The community
– allocates budget between ‘poor’ and ‘non-poor’ within the community
piG n
iG
niGi ,...,1
• Proportion Hi of the population is poor • Pareto efficiency: collective decision-making can be
represented by weighted sum of the aggregate (p.c.) welfare function for the each subgroup (poor/non-poor).
• Weight poor/non poor • Optimal allocation:
• “Targeting differential”:
),(),()1(),(),(max iiiiiiiii XGWXGHXGWXGH nnnnppppi
iiiii GGHGHts np )1(..
The local allocation problem
),,( iii XHGpGpiG ),,( iii XHGnGn
iG
),,(),,( iiiiii XHGnGXHGpGiT
npkXG iki
kki ,0),(
The center’s allocation problem
• Own weights on the poor/non poor
• Information set
(non overlapping with the one available locally)
• Choose Gi to maximize:
s.t. + local optimal allocations
Optimal outlays:
iiini
nni
n
ii
pi
ppi NZXGUHXGUHE ]),()1(),([ *
1
*
GNG i
n
ii
1
),( ii ZGG
npkk ,0* ),(),( iiii ZHX
Measuring targeting performanceNotation
• Each participating household receives same amount • The “targeting differential”:
• Targeting differential can be calculated at the national/local level (decomposable)
Poor? Yes No
Program?Yes s11 s12 G
No s21 s22 1-G H 1-H 1
1)1(
1 2211
H
s
H
sGGT np
2: Empirical model and resultsTargeting differential for FFE and its decomposition
• little sign of trade-off; the center is not more pro-poor in targeting villages than the villages in targeting households
• heterogeneous performance across communities (negative in 24% of the communities)
-0.0490.1460.134**0.3150.462Participating villages only
0.0030.0360.039**0.0790.118All villages
InterIntrapG nG T
G(i) 0 1
0
1
G(i) 0 1
0
1
piG n
iG
iG
G(i) 0 .94
-.16
1
G(i) 0 .94
-.24
.57
pi
pi
pi GGGEG )|( n
ini
ni GGGEG )|(
iH
Empirical model of program allocations
• Model allocations within villages between poor and non-poor as a function of the poverty rate, the budget received by the center, and village characteristics.
• Model the center allocation across villages as a function of village characteristics in the center’s information set.
Information structure and identificationTesting exogeneity of central allocation at village level and
exogeneity of information available to center Observed by: Examples Identifies exogenous
variation in: Center Community
Yes yes Commonly known
data such as from the census
Yes no Community’s relative position
on eligibility criteria
center’s allocation across communities
when explaining local targeting
No yes Inequality within
the village; transfers within
the village
potentially endogenous data on eligibility criteria
when explaining center’s allocation
Empirical models1. Intra-village allocations:
pi
pii
ppi XGG
ni
nii
nni XGG
for the sample of participating villages (Gi >0).
Test exogeneity of Gi using relative position as an IV
2. Center’s allocations:iiii uZZG 2211
iii XZ 1
Test exogeneity of Zi (program eligibility criteria) using LIML for limited dependent variables (Smith Blundell)
Explanatory variables
• Eligibility criteria: landless, female headed, lowly occupations, children
• Structural characteristics: agricultural development, non-farm income diversification, illiteracy, schools, banks, shocks
• Openness: electricity, phone, accessibility • Inequality of land holdings• Local institutions: net transfers to the poor,
recreational clubs, cooperatives
Targeting
performance Participation rate for the
poor
Participation rate for the
nonpoor 0.324** 1.177** 0.853** Allocation
to village (2.30) (16.30) (9.63)
0.081 -0.145 -0.226** Poverty rate in the village (0.43) (0.99) (2.51) R2 0.07 0.73 0.66 N 62 62 62
Note: Robust t-statistics in parentheses. * denotes significant at 10% level; ** at 5% level.
Intra-village targeting performance of Bangladesh’s Food-for-Education Program
A village is better at reaching the poor if it:
• receives larger allocation from the center; targeting improves both absolutely and relatively (Gi )
• has a lower fraction of widow female heads (eligibility)
• has fewer schools, has high cropping intensity, has lower illiteracy rate, has lower incidence of shocks (structural)
• is less isolated: telephone, closer to Thana HQ (openness)
• has more equal land distribution (inequality)
• is not already helping the poor (informal net transfers)
The center targets villages that:
• are in regions with higher incidence of landlessness• have a higher fraction of households in low professions
(relative to region)• are hit by a shock (natural disaster/ epidemic /pests)• have a Grameen Bank branch• has a village member in the UP council
Political economy constraints:• Low predictive power of eligibility criteria• A wider range of communities are targeted than would
appear to be justified by the program’s stated objectives
Summary for Bangladesh case study• On measuring targeting performance
– Program is mildly pro-poor– The center is not good at targeting poor villages (political
constraints)– Local communities are more accountable to the poor; no
trade-off evident
• On explaining performance– Fraction poor/non poor receiving the program and targeting
increase with the size of the budget– early capture by the non-poor– Indications that differences in relative power matter– Role of local level inequality in determining outcomes– Local political economy helps perpetuate inequality in the
presence of central efforts at redistribution
Case Study 2: Argentina’s Trabajar Program
• Trabajar program aims to reduce poverty by:
– providing short-term work at low wages; self-select unemployed workers from poor families;
– locating the projects in poor areas (physical and social infrastructure; compete on a points system)
• Local groups (municipalities and NGOs) are the sponsoring agencies and must provide co-financing of non-wage costs.
• Concerns about reaching poor areas
Measuring targeting performance
• Targeting differential is estimated by the regression coefficient of program spending on poverty rate estimated across departments (n=510)
• This identifies difference in amounts going to the poor vs non-poor
• Overall targeting differentials are significantly positive
• Overall TD improved with program expansion and worsened with contraction, as in Bangladesh’s FFE
Panel-data test using inter-provincial allocation of spending
• Test equation with province fixed effects:
• Aggregate spending allocation is allowed to be endogenous in that it is correlated with the province effect
jtjjtjt μηGβαT
(j=1,22; t=1,6)
Budget effects on targeting of Argentina’sTrabajar Programs
Full
sample Trabajar II
Trabajar III
Program spending 3.13 (4.81)
3.55 (5.32)
10.39 (4.44)
R2 0.778 0.813 0.903
N 132 66 66 Robust t-statistics.
The dependent variable is the targeting differential given by the regression coefficient of Trabajar spending per capita at department level for each province and time period on the incidence of unmet basic needs per capita.
Each regression included province fixed effects.
Summary of Argentina Case Study
• As for Bangladesh’s FFE, targeting worsens (improves) as outlays contract (expand)
• Extra public action is warranted to protect the poor during fiscal adjustment
• Evaluations that ignore political economy can greatly underestimate the gains from successful add-on social programs
• Efforts of combine cuts with better targeting may violate political economy constraints
• A pro-poor shift in spending during adjustment will not be politically easy