Smarter Social Protection?
presented by Marta Favara, Catherine Porter, Tassew Woldehanna
CSAE Conference Presentation, University of Oxford
March 21, 2016
Model Town (Delhi) 9953330565 Escorts, Call Girls Services
PSNP CSAE presentation
1. Introduction Methods Data Results Conclusions and next steps
Smarter social protection?
Marta Favara, Catherine Porter, Tassew Woldehanna
CSAE Conference Presentation, University of Oxford
catherine.porter@hw.ac.uk
March 21, 2016
M Favara, C Porter, T Woldehanna
PSNP-CSAE
2. Introduction Methods Data Results Conclusions and next steps
Overview
1 Introduction
2 Methods
3 Data
4 Results
5 Conclusions and next steps
M Favara, C Porter, T Woldehanna
PSNP-CSAE
3. Introduction Methods Data Results Conclusions and next steps
Context and previous literature
Evaluations of (Conditional) cash transfers have shown
improvements in nutrition outcomes (e.g.Behrman and
Hoddinott, 2005)
Evidence on improvement in enrollment, grade, school
attendance (Schady et al 2005 review)
Less evidence on cognitive achievement
Mixed results on India’s NREGA the largest workfare
programme in the world:
Mani et al (2014) find strong positive effects on grade
progression and a number of cognitive skills tests for the
Young Lives cohorts
Shah and Steinberg (2015) find the opposite - NREGs
hampers progression and cognition due to time demands -
though this is mainly for children over 12
M Favara, C Porter, T Woldehanna
PSNP-CSAE
4. Introduction Methods Data Results Conclusions and next steps
PSNP Background
2005 Productive Safety Net Scheme introduced after long
history of responsive food aid and drought
Workfare programme for 80% of participants, UCT for others
(labour-poor hh)
Cash-first principle (tho in practice 60-40)
Co-ordination between donors, and with Government
innovation
In 2013 7.2 million beneficiaries (roughly 10% of the national
population) in 8 of 10 regions.
Phase 3 from 2010-2015 attempted to improve timeliness and
predictability of transfers, strengthen public works and
accountability,
M Favara, C Porter, T Woldehanna
PSNP-CSAE
5. Introduction Methods Data Results Conclusions and next steps
Cognitive development child model
Todd and Wolpin (2003, 2007) note that under the assumption
that effect of inputs (both observed and unobserved) as well as that
of initial ability decline geometrically over time, then a “lag value
added” model can be specified, using only the immediate lag of
achievement serving as a proxy for all previous inputs, and ability.
Aika = αXika + γAik,a−1 + eika (1)
Where Aika is achievement of child i in household k at age a.
Vector of X controls includes PSNP participation, plus other
exogenous household controls (household size, shocks, prices).
In our model child is aged 12 years in 2013. Lagged ability is
measured at age 8.
M Favara, C Porter, T Woldehanna
PSNP-CSAE
6. Introduction Methods Data Results Conclusions and next steps
Empirical Strategy (1)
PSNP enters directly into equation (1)
Main challenge is non-random targeting
Difference-in-differences strategy (test parallel trends using
2002-2006 HAZ)
Include lag child achievement as per conceptual model, to
control for ability and child-hh unobservables
Cluster fixed effects to control for village unobservables
(school quality etc)
M Favara, C Porter, T Woldehanna
PSNP-CSAE
7. Introduction Methods Data Results Conclusions and next steps
Empirical Strategy (2): Selection of Comparison Group
Selection of comparison group: never treated may be
systematically different from beneficiaries
We use those who received food aid in 2002 or 2006 as the
control group, plus hh that reported being on the shortlist for
PSNP (129 hh)
We split PSNP beneficiaries by timing in table 1:
Table: PSNP Beneficiaries
PSNP Groups Obs Percentage
Control 129 25.34
2009 only 129 25.34
2009 & 2013 226 44.40
2013 only 25 4.91
Total 509 100
M Favara, C Porter, T Woldehanna
PSNP-CSAE
8. Introduction Methods Data Results Conclusions and next steps
Young Lives dataset
Cohort children born in 2000
20 sentinel sites
Multi-purpose survey including cognitive skills, nutrition,
education, non-cognitive skills
4 rounds (2002, 2006, 2009, 2013)
Round 3 and 4, sibling information collected (for one younger
sibling)
Attrition rate around 2.2 percent (very low)
M Favara, C Porter, T Woldehanna
PSNP-CSAE
9. Introduction Methods Data Results Conclusions and next steps
Figure 1. PSNP and survey timing
Figure: PSNP and survey timing
M Favara, C Porter, T Woldehanna
PSNP-CSAE
10. Introduction Methods Data Results Conclusions and next steps
Table: Control and Treatment groups: baseline
characteristics
Control PSNP beneficiaries t-test
Mean Std.Dev Mean Std.Dev p-value
In 2006 (age 5)
Height-for-age (z-score), age 5 -1.56 0.087 -1.57 0.052 0.860
Food expenditure 79.53 4.303 80.28 3.231 0.902
Non-food expenditure 30.19 2.101 27.40 1.165 0.233
% Education/Total expenditure 0.01 0.002 0.01 0.001 0.750
Caregiver aspirations 0.57 0.044 0.58 0.025 0.760
In education 0.06 0.021 0.04 0.010 0.288
At least one shock 0.88 0.029 0.80 0.021 0.046
Changes between 2002-06
Parallel trends assumption:
Height-for-Age (z-score) -0.17 0.157 -0.09 0.098 0.685
Observations 129 380
Note: Average values measured in round 1 and 2 at the age of 1 and 5; standard deviation
reported in parentheses. Food, non-food and education expenditure are per month/adult and
in current Birr. The p-value for a t-test for differences in means between Control group and
PSNP beneficiaries is reported in the last column.
M Favara, C Porter, T Woldehanna
PSNP-CSAE
11. Introduction Methods Data Results Conclusions and next steps
Table: Descriptive statistics
Variables Mean Std.Dev
Household Characteristics
Household size 6.09 0.082
Wealth Index (total) 0.30 0.005
Number of shocks 1.06 0.053
At least one shock 0.58 0.022
Shock: drought 0.28 0.020
Shock: flood 0.12 0.015
Shock: crop failure 0.36 0.021
Shock: illness of household member 0.19 0.017
Food expenditure 284.77 6.518
Non-food expenditure 125.85 6.167
% Education/Total expenditure 0.01 0.001
Individual Characteristics
Male 0.53 0.022
Age (in months) 145.42 0.181
Height-for-age (z-score) -1.59 0.041
Time use (hours/day)
Caring for household member 0.62 0.039
On household chores 1.73 0.058
In domestic tasks 2.51 0.099
In paid activities 0.07 0.023
At school 5.33 0.065
Studying (outside school) 1.25 0.037
Observations 509
Note: Average values measured at the age of 12; standard devi-
ation reported in parentheses. Expenditure are per month/adult
and in current Birr
M Favara, C Porter, T Woldehanna
PSNP-CSAE
12. Introduction Methods Data Results Conclusions and next steps
Preliminary results: Math score age 12 VA model
Value-added specification (Cluster Fixed effect)
(1) (2) (3) (4) (5) (6)
PSNP Graduates 0.284** 0.294** 0.285** 0.304** 0.304** 0.321***
(0.127) (0.125) (0.125) (0.131) (0.131) (0.117)
PSNP Continuing 0.132 0.134* 0.135 0.137 0.127 0.137
(0.081) (0.074) (0.081) (0.086) (0.090) (0.111)
PSNP new 2013 0.293** 0.294** 0.282** 0.240 0.240* 0.268
(0.136) (0.139) (0.135) (0.140) (0.140) (0.199)
Height-for-age (z-score) -0.077** -0.077*
(0.035) (0.041)
Maths at age 8 (z-score) 0.492*** 0.471*** 0.473*** 0.439*** 0.454*** 0.442***
(0.039) (0.036) (0.035) (0.040) (0.038) (0.041)
Controls x x x x x x
Household Inputs x x x x x
Education Inputs x x x x
Time use x x x
Non-cognitive skills x x x
Health x x
Expenditure x
Constant 2.464** 1.472 0.820 0.481 0.826 0.878
(1.149) (1.180) (1.179) (1.233) (1.210) (1.383)
Observations 509 509 509 509 509 509
R-squared 0.337 0.357 0.377 0.401 0.405 0.410
Note: The table reports the OLS estimates with standard errors (reported in parentheses) clustered at village
level, including village fixed effects. * p<0.1 ** p<0.05 *** p<0.01. The dependent variable (raw mathemat-
ics score) is measured at the age of 12 and standardized within the sample by age (round). All controls are
either time invariant or measured at the age of 12. All controls are included as specified.M Favara, C Porter, T Woldehanna
PSNP-CSAE
13. Introduction Methods Data Results Conclusions and next steps
Preliminary results: Summary
Results strongest for those who received PSNP who had
graduated by 2013
Impact of approx .25SD on maths score
Robust to inclusion of controls, nutrition, time use,
expenditure
M Favara, C Porter, T Woldehanna
PSNP-CSAE
14. Introduction Methods Data Results Conclusions and next steps
Table: Year of graduation for “2009 only” PSNP beneficiaries
Year of graduation Frequency Percent
2007 1 1.92
2008 4 7.69
2009 1 1.92
2010 6 11.54
2011 9 17.31
2012 26 50.00
2013 5 9.62
Total 52 100.00
Note: Much missing data, total 129 participants who reported
receiving PSNP in 2009, but not in 2013.
M Favara, C Porter, T Woldehanna
PSNP-CSAE
15. Introduction Methods Data Results Conclusions and next steps
Summary
We find positive impact of PSNP participation on cognitive
outcomes of children aged 12
Order of magnitude around 0.3 SD on maths test score
(standardised)
Relatively similar to CCT findings in the literature on younger
children
Fairly surprising given no conditionality, work requirement
(tho for now these are fairly young children)
M Favara, C Porter, T Woldehanna
PSNP-CSAE
16. Introduction Methods Data Results Conclusions and next steps
More work to do...
Mechanism of impact (time use, nutrition, why the
graduates?)
Sibling-differences model
Older cohort
Robustness: other cogntiive outcomes
Comments welcome
M Favara, C Porter, T Woldehanna
PSNP-CSAE