dec lecture series, world bankpubdocs.worldbank.org/en/194861518548502794/dec-lecture... · duflo,...
TRANSCRIPT
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
DEC Lecture Series, World Bank
The Impact of Free Secondary Education:Experimental Evidence from Ghana
Gratefully acknowledging funding from NIH, NSF, PCD, J-PAL PPE Initiative, the IGC, 3ie, Nike Foundation+ the work of many wonderful surveyors and research assistants over the past 10 years
EstherDuflo
PascalineDupas
MichaelKremer
MIT Stanford Harvard
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Introduction• With growth of primary education, increased calls for
free secondary education, internationally (it is one of the SDG)
• Many anticipate large economic and social impacts, especially for girls
• Others less optimistic • Secondary education is expensive precisely• Will students learn? • Is the curriculum adapted for a terminal secondary degree or
really a preparation for tertiary? (Goldin, 1999) • Will they get jobs?
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Introduction: Current Evidence
• Plenty of evidence on primary, but little high-quality evidence on impact of secondary education in developing countries
• Evidence from developed countries not necessarily relevant.• Developing countries have vastly greater levels of education than
developed countries had at comparable income levels (Pritchett, 2001)
• Quality of education in developing countries may be very different (Hanushek and Woessman, 2012)
• Ozier (2016) RD study based on Kenya’s secondary school admission test
• Delayed fertility onset for women• Limited labor market outcomes, but increased formal employment for
men• “Marginal” candidates who just got in secondary
Duflo, Dupas, Kremer Returns to Secondary Education: Ghana
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
This Study• Ghana. Ongoing longitudinal study started in 2008
• Sampled 2,064 students admitted to particular secondary school and major, but did not enroll (mostly due to lack of fund)
• 682 (randomly selected) received 4-year scholarships to attend day secondary school
• Examine impact of opportunity of free secondary education holding admission criteria and other means to finance school constant.
• Relevant policy, in Ghana and elsewhere.
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
RoadmapI. IntroductionII. Background and DataIII. Impact on Secondary Education
• Impacts on SHS Participation & Completion• Learning Outcomes
IV. Impacts on Fertility, Marriage and Health BehaviorV. Earnings, Tertiary education VI. Conclusion
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Background and Data
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Ghana’s Education System• 95% primary school enrollment; 75% JHS enrollment• 70% of JHS students take JHS finishing exam
• 60% of test takers pass• 80% of those who pass enroll in SHS (Ajayi 2014)
• Tuition for day (non-boarding) SHS students in 2011: 500 Ghana cedis (Per capita GDP in 2011: 2400 Ghana cedis)
• Only 700 SHS nationwide (compared to 9000 JHS)• Girls 6 p.p. (20%) less likely to reach SHS
• Some who do not enroll in SHS enroll in (much smaller) TVI (technical vocational institute) system
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Experimental Design• 2,064 students sampled from 177 SHS across 5 regions
of rural southern Ghana. • admitted into a particular Senior High School (SHS) and major
based on JHS exam score• not enrolled in any SHS as of December 2008• some girls who graduated from JHS in 2007
• 682 randomly selected to receive 4-year SHS scholarship • Stratified by district, SHS, JHS, gender and year took JHS exam
• Scholarship covers the full tuition and fees for a day student for 4 years
• Average Cost = GHX 1920 (~US$480)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Timing and randomization • May 2008: Wrote JHS exit exam• July 2008: Graduated from JHS• August 2008: Admitted into SHS, placed into specific
major• September 2008: School year 2008-2009 starts• December 2008: Sampled• January 2009: Randomized into treatment or control
• July 2012: Eligible to graduate from SHS• September 2013: Earliest can enroll in tertiary if
graduated from SHS in July 2012
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Senior High School Curriculum• All students take English, Math, Science, Social Studies, but
also admitted to a major• 40% of sample admitted to academic majors
• General Arts (includes French, more social science)• General Science (advanced math, chemistry, physics, bio)
• 60% in vocationally-oriented majors: Home Economics, Visual Arts, Agriculture, Technology, and Business
• Supplementary fees for vocational majors, general science• Differing preparation for tertiary, labor market• Over 20% of academic admits switch to vocational, 30% of
vocational admits switch to academic major• Implies we may understate impact of differences by major
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Timeline of Surveys• Fall 2008: Baseline Survey, distributed cell phones to improve follow up rate
• Spring 2013: In-person Follow-up Survey (5-year impacts)• 96.3% survey rate • Include cognitive testing questions based on PISA
• Spring 2015: follow-up survey by phone (home tracking for 15% stragglers) (7-year impacts)
• Spring 2016: follow-up survey by phone (home tracking for 10% stragglers) (8-year impacts)
• 96.4% survey rate • Improves labor market survey questions compared to 2015
• Summer 2017: Follow up survey by phone (home tracking for 8% stragglers) (9-year impacts)
• Average age at time of survey: 26• 95.4% survey rate • Further improves labor market surveys (but also have comparable data to 2016)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Baseline: Beliefs about Education• Sample believed there were high returns to SHS
• 276% increase in earnings if complete SHS
• Government employment is seen as the pathway to these high returns
• Over 70% believed SHS would lead to government employment or employment in a profession dominated by government employees (teacher or nurse) by age 25
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Impact on Secondary Education and Learning
1. Impact on secondary school participation2. Impact of knowledge/skills (2013)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
1. Impact of Scholarship on SHS Enrollment
0%20
%40
%60
%80
%10
0%Sh
are
enro
lled
in S
HS
Term 1----------
Term 2---2008/09----
Term 3----------
Term 1----------
Term 2---2009/10----
Term 3----------
Term 1------2010
Term 2/11-------
Term 32011/12
Treatment: BECE'08 Boys BECE'08 Girls BECE'07 Girls
Control: BECE'08 Boys BECE'08 Girls BECE'07 Girls
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Impacts on SHS Participation & Completion
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Impacts on SHS Participation & Completion
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Impacts on SHS Participation & Completion
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer18
What does it mean for the debate on free secondary education?
• Ghana: Protracted political debates around this question in the past three elections (December 2008, December 2012, December 2016)
• Proponents focus on benefits• Opponents focus on costs
• Many “inframarginals” – people who would have paid on their own.
• Dec 2016: Pro-Free SHS camp won election• Sep. 2017: free SHS launched
• GoG estimates that the policy led 90,000 additional children to enroll this fall
• But debate goes on, with opposition asking where the money is going to come from
• Our study: quantifies these costs and benefits• Many domains
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Fiscal Cost of Free SHS policy
• Scholarship winners = 3.09 years in SHS; non-scholarship winners = 1.83
• Scholarship paid for 3.09 years of education per 1.26 additional years in our sample.
• Cost of free education: Upper bound: no effect of scholarship on JHS pass rate
• Assume 80% of qualified students complete SHS regardless, other 20% behave like our sample
• Free SHS requires paying for 15 years of schooling for each additional year of attainment
• If promise of free secondary education leads 25% of students not passing exam to pass
• Free SHS requires paying for ~6 years of schooling for each additional year of SHS attainment
• Important margin: only 40% of those who start JHS pass final exam
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. Learning Outcomes
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. Learning Outcomes
All Female Male(1) (2) (3)
Standardized score, Reading test (2013) Treatment effect 0.147 0.151 0.136 Standard error (0.046)*** (0.064)** (0.064)** Comparison mean -0.000 -0.096 0.100 p-value on equality of effects (2)= (3): .875Standardized score, Math test (2013) Treatment effect 0.129 0.170 0.078 Standard error (0.047)*** (0.065)*** (0.065) Comparison mean -0.000 -0.191 0.199 p-value on equality of effects (2)= (3): .316
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. Total Standardized Score: By gender/cohort
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. By gender/major
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. By Quartile of JHS exit exam score
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. By School Category
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. Learning Outcomes: Other MeasuresAll Female Male(1) (2) (3)
National political knowledge standardized score (2013) Treatment effect 0.098 0.120 0.062 Standard error (0.047)** (0.064)* (0.064) Comparison mean 0.000 -0.239 0.250
(2)= (3): .521International political knowledge standardized score (2013) Treatment effect 0.067 0.005 0.104 Standard error (0.047) (0.059) (0.059)* Comparison mean 0.000 -0.402 0.419
(2)= (3): .240Knows how to use the internet (2015) Treatment effect 0.066 0.088 0.034 Standard error (0.023)*** (0.030)*** (0.030) Comparison mean 0.593 0.417 0.775
(2)= (3): .201Knows how to use the internet (2016) Treatment effect 0.039 0.047 0.018 Standard error (0.022)* (0.030) (0.029) Comparison mean 0.639 0.474 0.811
(2)= (3): .481
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Other outcomes
All Female Male(1) (2) (3)
Has WhatsApp account (2017) Treatment effect 0.030 0.080 -0.027 Standard error (0.024) (0.033)** (0.033) Comparison mean 0.572 0.489 0.659 p-value on equality of effects (2)= (3): .021**Has Facebook account (2017) Treatment effect 0.053 0.077 0.016 Standard error (0.023)** (0.032)** (0.032) Comparison mean 0.553 0.408 0.706 p-value on equality of effects (2)= (3): .171
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Returns to a year of secondary education: Comparing OLS and IV
• OLS estimate: estimate in the control group. • IV estimate: use scholarship as instrument for years of
education. • Caveats:
• Direct impact of getting scholarship on outcome:• Financial [positive and negative]• Self confidence • Psychological incentive effects.
• A few children in the control group go to technical institute: • To the extent quality is lower, returns to education are under-
estimated.
• Interpretation:• IV is the estimate for compliers: potentially those with
highest return (if they are credit constrained).
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. Learning Outcomes: OLS vs. IV
All
OLS IV
(1) (2)
Total standardized score (2013)
Effect of year of education 0.213 0.124
Standard error (0.014)*** (0.034)***
p-value on equality of effects OLS=IV: .014***
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
2. Learning Outcomes: OLS vs. IV
Total Standardized Score (2013)
CombinedAll Female Male
OLS IV OLS IV OLS IV(1) (2) (3) (4) (5) (6)
Secondary Education (Lower Bound)Effect of year of education 0.213 0.124 0.253 0.155 0.160 0.088
Standard error (0.014)*** (0.034)*** (0.019)*** (0.048)*** (0.016)*** (0.047)*p-value on equality of effects (1)=(2): .014*** (3)=(4): .058* (5)=(6): .158
Secondary Education + TVI (Upper Bound) Effect of year of education 0.211 0.135 0.246 0.162 0.157 0.102
Standard error (0.014)*** (0.037)*** (0.015)*** (0.050)*** (0.016)*** (0.054)*p-value on equality of effects (1)=(2): .055* (3)=(4): .111 (5)=(6): .339
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Putting Machine Learning to the test• OLS appears to over-estimate. Can causal effect be
recovered by controlling for the very rich set of control variables that we do have?
• Since there are many many potential control variables, we apply the Double Machine Learning method proposed in Chernozhukov et al. (2017)
• Briefly, the idea is similar to partialing out the effect of the slew of unobservables from both schooling and test scores.
• We also do the same thing, weighing the observations by the heterogeneity in treatment effect in the first stage [to get the effect on people who observationally look like the compliers]
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
For test scores, ML does well.
0
0.05
0.1
0.15
0.2
0.25
OLS (no cont.)* OLS* IV* DoubleML* DoubleML(Weighted)*
Total Standardized Score
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Impacts on Fertility, Marriage and Health Behavior
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Delayed fertility (and marriage) for Women
0.120.16
0.23
0.29
0.48
0.59
0.64
0.7
0.110.15
0.220.26
0.4
0.50.54
0.63
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
2009 2010 2011 2012 2013*** 2015*** 2016*** 2017**
Ever pregnant
Control Treatment
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
(Reported) fertility for Men is also impacted but effects are much more modest
0.010.02
0.03
0.06
0.11
0.21
0.25
0.33
00.02
0.03
0.06
0.09
0.16
0.23
0.33
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
2009 2010 2011 2012 2013 2015 2016 2017
Ever had a pregnant partner
Control Treatment
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Ever pregnant
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Living Arrangements
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Marriage and Fertility (2017)Combined
All Female Male(1) (2) (3)
Ever lived with partner(2016)Treatment effect -0.063 -0.091 -0.027Standard error (0.020)*** (0.028)*** (0.028)Comparison mean 0.241 0.344 0.134p-value on equality of effects AF=AM=VF=VM: .419 F=M: .106
Ever pregnant/had a pregnant partner (2016)Treatment effect -0.071 -0.107 -0.023Standard error (0.024)*** (0.032)*** (0.031)Comparison mean 0.403 0.582 0.213p-value on equality of effects AF=AM=VF=VM: .164 F=M: .061*
Number of children ever had (2016)Treatment effect -0.130 -0.217 -0.030Standard error (0.040)*** (0.054)*** (0.054)Comparison mean 0.519 0.814 0.212p-value on equality of effects AF=AM=VF=VM: .051* F=M: .014**
Had unwanted first pregnancy (full sample) (2016)Treatment effect -0.071 -0.115 -0.019Standard error (0.024)*** (0.032)*** (0.031)Comparison mean 0.375 0.566 0.181p-value on equality of effects AF=AM=VF=VM: .108 F=M: .032**
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Marriage and Fertility• Timing of results suggests it’s unlikely to be an
“incarceration effect”• Potential mechanisms:
• (1) Increase in the opportunity cost of bearing and raising children (Becker, 1991)
• Evidence of increased earnings, tertiary education• (2) Education may shape/change preferences for children
• for females academic admits, scholarship reduced desired fertility by age 50 by 0.23 children (15.7% decrease)
• No evidence for female vocational admits• (3) Ability to make better choices thanks to better decoding
of information (Rosenzweig and Schultz, 1989)• Health behavior
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Health BehaviorAll Female Male(1) (2) (3)
Desired fertil ity: # of children by age 50 (2013) Treatment effect -0.029 -0.043 -0.015 Standard error (0.049) (0.070) (0.070) Comparison mean 3.629 3.639 3.619 p-value on equality of effects (2)= (3): .779Index of risky sexual behavior(safe--> risky)(2013) Treatment effect -0.040 0.001 -0.075 Standard error (0.029) (0.040) (0.040)* Comparison mean 0.000 0.096 -0.099 p-value on equality of effects (2)= (3): .178Index of STI risk exposure (2013) Treatment effect -0.058 -0.040 -0.071 Standard error (0.028)** (0.039) (0.039)* Comparison mean -0.000 0.092 -0.096 p-value on equality of effects (2)= (3): .573Preventative health behavior (3 questions) (2013) Treatment effect 0.109 0.118 0.104 standard error (0.037)*** (0.052)** (0.052)** Comparison mean 1.624 1.691 1.555 p-value on equality of effects (2)= (3): .841
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Marriage and Fertility : ML vs OLS (2017)
-0.18
-0.16
-0.14
-0.12
-0.1
-0.08
-0.06
-0.04
-0.02
0OLS (no cont.)* OLS* IV* DoubleML*
DoubleML(Weighted)*
Number of Children Ever Had
2017 2016
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor Market Impacts
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Macro Context During Study Period
• Rapid growth ends in 2012, induces fiscal retrenchment:
• Nursing and Teacher Training programs allowances and quotas removed in 2014
• Common for participants to wait two years before getting admission to tertiary education due to quotas
• Government hiring freeze in 2015
• SHS length shortened from 4 to 3 years in 2009/2010• Study participants graduated in a double cohort with
students who enrolled a year later
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor Market Effects• Bleak labor market:
• 2016: only 46% of women and 69% of men in control group had positive earnings in last month.
• 2017: 47% and 70%, so almost no change! • When we include back wages, shadow wages (from working
with parents) and/or in-kind income we get 63% and 85% respectively
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
What are they up to in 2017?All Female Male(1) (2) (3)
Enrolled in formal study/ training (2017) Treatment effect 0.022 0.046 -0.003 Standard error (0.014) (0.019)** (0.019) Comparison mean 0.081 0.068 0.096 p-value on equality of effects (2)= (3): .069*Wage worker (2017) Treatment effect 0.049 0.042 0.048 Standard error (0.022)** (0.031) (0.030) Comparison mean 0.282 0.212 0.356 p-value on equality of effects (2)= (3): .888Actively searching for a job (2017) Treatment effect 0.092 0.111 0.069 Standard error (0.021)*** (0.030)*** (0.030)** Comparison mean 0.249 0.203 0.297 p-value on equality of effects (2)= (3): .328Min. training for your job: SHS (2017) Treatment effect 0.068 0.058 0.076 Standard error (0.017)*** (0.024)** (0.023)*** Comparison mean 0.121 0.102 0.142 p-value on equality of effects (2)= (3): .582Public sector wage employee (2017) Treatment effect 0.032 0.041 0.023 Standard error (0.009)*** (0.012)*** (0.012)* Comparison mean 0.024 0.019 0.028 p-value on equality of effects (2)= (3): .286
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Big differences by Gender/MajorFemale Male Female Male
(5) (6) (8) (9)Enrolled in formal study/ training (2017) Treatment effect 0.080 0.015 0.022 -0.019 Standard error (0.031)** (0.031) (0.026) (0.025) Comparison mean 0.081 0.093 0.061 0.100 p-value on equality of effects (5)= (6): .137 (8)= (9): .251Wage worker (2017) Treatment effect 0.090 -0.007 0.016 0.096 Standard error (0.049)* (0.049) (0.041) (0.040)** Comparison mean 0.223 0.351 0.213 0.358 p-value on equality of effects (5)= (6): .164 (8)= (9): .161Actively searching for a job (2017) Treatment effect 0.104 0.061 0.110 0.077 Standard error (0.049)** (0.048) (0.040)*** (0.039)** Comparison mean 0.194 0.331 0.199 0.275 p-value on equality of effects (5)= (6): .529 (8)= (9): .563Min. training for your job: SHS (2017) Treatment effect 0.026 0.099 0.085 0.063 Standard error (0.038) (0.038)*** (0.032)*** (0.031)** Comparison mean 0.129 0.109 0.088 0.164 p-value on equality of effects (5)= (6): .171 (8)= (9): .619Public sector wage employee (2017) Treatment effect 0.033 0.010 0.051 0.034 Standard error (0.020) (0.020) (0.017)*** (0.016)** Comparison mean 0.018 0.036 0.021 0.022 p-value on equality of effects (5)= (6): .422 (8)= (9): .470
Academic Major Admits Vocational Major Admits
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor market outcome: earnings (2016)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor market outcome: earnings (2017)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor Market Outcomes: Combined Earnings (2017)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor Market Effects: Discussion• Vocational Admit Males have very high return
• Credit Constraints?
• Academic Females• Consistent results across years: more likely to be wage employed• Big gains in tertiary• In 2016: wide range of possible outcomes for those not in school• In 2017: start seeing positive effect (including delayed wages)
• Vocational Female: large effects in 2016 which have gone back to zero.
• Academic Males• Current data suggests poor return, even for those not in school-
Why?
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor Market Outcomes: Discussion• Despite the fact that these youth are 26, they are still
not settled in the labor market1. Many are still intending to go back to school (tertiary); they
take effective investment in this direction (extra classes, applications)
2. They are still looking for a job, even when they don’t have one.
3. Those effects (taking steps to go to tertiary, looking for jobs) are even stronger for secondary school graduates.
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Schooling preparation
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Schooling preparation
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Labor Market Outcomes OLS vs. IV• OLS estimates consistently lower than the IV estimates
• Machine learning does nothing to “fix” it
• Consistent with (at least) two hypotheses• Financially constrained students have higher returns to
education.• Partly captured by observables, but not much.
• The underestimate is mainly driven by non participation—possibly in turn by the inability to match the tertiary result. Those who are more likely to go to school are more likely to go to tertiary and they are thus less likely to work.
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Comparing OLS, ML and IV (2016)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Comparing OLS, ML and IV (2017)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Satisfaction• Skeptics warn that gap between expectations and
reality could create dissatisfaction
• We find: • No indication of greater depression/unhappiness in general
(7 questions mental health index)• Women are more satisfied with finances
• Concentrated among vocational admits• But more dissatisfied with life in general (academic admits)• Employed men less satisfied with their job
• Satisfaction questions hard to interpret (ambition may have been affected by education)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Conclusion• Scholarships increased secondary school completion rates by 30 percentage
points• Secondary education leads to significant gains on cognitive scores• Secondary education delays fertility and marriage; enables healthier behaviors • Treatment effects for women are greater on a number of dimensions
• Learning, fertility and marriage, tertiary enrollment, wage employment
• Labor market outcomes are mixed• Many of the youth are still not settled in a job. Large fraction are not in school but not
working, or still doing on the job search• Male Vocational majors experience improved labor market outcomes• Female Academic majors increase tertiary enrollment and we start seeing positive effect
in 2017• Male Academic majors do not see significant benefits by age 26 –fell way short of sky-
high expectation at baseline• Consistent with a limited number of tertiary spots/government jobs and little cognitive gains for them.
• The youth seem to be looking for a ”real” job and are not willing to settle for less.
• Education accentuates this search: more likely to search, more likely to apply to tertiary, more likely to be dissatisfied with their current job
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Additional Slides & Tables
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
ReferencesCesur, R., & Mocan, N. H. (2013). Does secular education impact religiosity, electoral participation and the propensity to vote for Islamic parties? Evidence from an education reform in a Muslim country. Working Paper
Becker, Gary S. (1991). “An Economic Analysis of Fertility,” Democratic and Economic Change in Developed Countries, Gary S. Becker, ed., Princeton: Princeton University Press.
Duflo, E. (2004). The medium run effects of educational expansion: Evidence from a large school construction program in Indonesia. Journal of Development Economics, 74(1), 163-197.
Goldin, Claudia (1999). “America’s graduation from high school: The evolution and spread of secondary schooling in the twentieth century.” The Journal of Economic History, 58(02), 345-374
Gulesci, Selim and Erik Meyersson (2015). “ ‘For the Love of the Republic’: Education, Secularism, and Empowerment.” Working Paper.
Hanushek, E. A., & Woessmann, L. (2012). Do better schools lead to more growth? Cognitive skills, economic outcomes, and causation. Journal of Economic Growth, 17(4), 267-321.
Heckman, J. J. (1991). Randomization and social policy evaluation.
Keats, Anthony (2014). “Women’s Schooling, Fertility, and Child Health Outcomes: Evidence from Uganda’s Free Primary Education Program.” Mimeo, Wesleyan University.
Krueger, A. B., & Malečková, J. (2003). Education, poverty and terrorism: Is there a causal connection?. The Journal of Economic Perspectives, 17(4), 119-144.
Mocan, N. H., & Cannonier, C. (2012). Empowering women through education: Evidence from Sierra Leone. NBER Working Paper No. w18016/
Osili, Una Okonkwo & Bridget Long (2008). “Does Female Schooling Reduce Fertility? Evidence from Nigeria.” Journal of Development Economics 87(1):57-75
Ozier, Owen (2016). “The Impact of Secondary Schooling: A Regression Discontinuity Analysis.” Journal of Human Resources, forthcoming
Prichett, Lant (2001). Where has all the education gone? The World Bank Review, 15(3), 367-391
Rosenzweig, Mark and T. Paul Schultz (1989). “Schooling, information, and nonmarket productivity: Contraceptive Use and its effectiveness.” International Economic Review 30: 457-477
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
Rate of Return of SHS for vocational majors
• Costs:• Includes add’l SHS fees paid, add’l school costs and foregone
earnings per scholarship winner from respondents’ last term, adjusted for inflation and number of terms in school
• 1.19 yrs of add’l SHS fees cost approximately GHX 140/yr in 2016 GHX• Winners paid GHX 160/yr more in add’l school costs• Foregone earnings:
• Benefits:• Scholarship winners earn add’l GHX 31.6/mo (GHX 378.76/yr)• Assume benefits persist through 30-yr working career
• Overall 16% return on SHS• 12% if assumed that winners didn’t begin working until 2016• SHS education worth GHX 4,936 (3% discount rate), up to GHX
3,200 (5% discount rate)
2008-09 2009-10 2010-11 2011-12
GHX 264 GHX 192 GHX 110 GHX 12
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
3. Tertiary education: ApplicationsAcademic Major Admits Vocational Major Admits
All Female Male All Female Male(4) (5) (6) (7) (8) (9)
Plans to continue to tertiary (2013)Treatment effect 0.267 0.289 0.240 0.241 0.282 0.197Standard error (0.038)*** (0.054)*** (0.053)*** (0.032)*** (0.045)*** (0.044)***Comparison mean 0.450 0.388 0.516 0.426 0.350 0.503p-value on equality of effects (5)=(6): .520 (4)=(7): .595 (8)=(9): .180
Sat for WASSCE exam (2015)Treatment effect 0.295 0.286 0.301 0.263 0.288 0.237Standard error (0.038)*** (0.053)*** (0.053)*** (0.031)*** (0.045)*** (0.043)***Comparison mean 0.449 0.418 0.482 0.423 0.352 0.494p-value on equality of effects (5)=(6): .839 (4)=(7): .516 (8)=(9): .416
Applied for tertiary education (2015)Treatment effect 0.094 0.131 0.059 0.073 0.100 0.047standard error (0.030)*** (0.043)*** (0.043) (0.025)*** (0.036)*** (0.035)Comparison mean 0.177 0.167 0.188 0.153 0.117 0.189p-value on equality of effects (5)=(6): .237 (4)=(7): .595 (8)=(9): .300
if applied: number of programs applied to (2015)Treatment effect -0.037 0.119 -0.155 -0.097 -0.108 -0.074standard error (0.169) (0.235) (0.241) (0.150) (0.232) (0.197)Comparison mean 1.681 1.457 1.896 1.653 1.556 1.712p-value on equality of effects (5)=(6): .416 (4)=(7): .792 (8)=(9): .911
Admitted to a tertiary program (2015)Treatment effect 0.024 0.055 -0.008 0.035 0.047 0.023standard error (0.023) (0.033)* (0.032) (0.019)* (0.027)* (0.027)Comparison mean 0.096 0.073 0.122 0.073 0.055 0.090p-value on equality of effects (5)=(6): .173 (4)=(7): .713 (8)=(9): .532
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
3. Tertiary education: Years of Enrollment
Academic Major Admits Vocational Major AdmitsAll Female Male All Female Male(4) (5) (6) (7) (8) (9)
Years spent attending tertiary education (2016)Treatment effect 0.084 0.134 0.031 0.058 0.036 0.076Standard error (0.045)* (0.064)** (0.064) (0.037) (0.054) (0.052)Comparison mean 0.192 0.147 0.241 0.112 0.100 0.124p-value on equality of effects (5)=(6): .252 (4)=(7): .652 (8)=(9): .593
Total years of education to date (2016)Treatment effect 1.367 1.421 1.290 1.167 1.200 1.113Standard error (0.152)*** (0.214)*** (0.212)*** (0.126)*** (0.179)*** (0.173)***Comparison mean 11.251 11.029 11.490 11.117 10.766 11.471p-value on equality of effects (5)=(6): .664 (4)=(7): .311 (8)=(9): .728
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
4. Marriage and FertilityAcademic Major Admits Vocational Major Admits
All Female Male All Female Male(4) (5) (6) (7) (8) (9)
Ever lived with partner(2016)Treatment effect -0.059 -0.093 -0.014 -0.063 -0.088 -0.031Standard error (0.033)* (0.045)** (0.044) (0.027)** (0.037)** (0.036)Comparison mean 0.231 0.314 0.141 0.247 0.369 0.126p-value on equality of effects (5)=(6): .209 (4)=(7): .932 (8)=(9): .267
Ever pregnant/had a pregnant partner (2016)Treatment effect -0.085 -0.134 -0.020 -0.072 -0.112 -0.020Standard error (0.038)** (0.050)*** (0.050) (0.031)** (0.042)*** (0.041)Comparison mean 0.393 0.529 0.245 0.404 0.621 0.185p-value on equality of effects (5)=(6): .107 (4)=(7): .799 (8)=(9): .117
Number of children ever had (2016)Treatment effect -0.142 -0.209 -0.051 -0.131 -0.249 -0.001Standard error (0.065)** (0.086)** (0.087) (0.054)** (0.072)*** (0.070)Comparison mean 0.500 0.731 0.252 0.530 0.883 0.178
p-value on equality of effects (5)=(6): .196 (4)=(7): .898 (8)=(9): .013**Had unwanted first pregnancy (full sample) (2016)
Treatment effect -0.067 -0.118 -0.004 -0.081 -0.130 -0.022Standard error (0.038)* (0.050)** (0.050) (0.032)** (0.043)*** (0.040)Comparison mean 0.371 0.523 0.211 0.369 0.595 0.153p-value on equality of effects (5)=(6): .106 (4)=(7): .768 (8)=(9): .065*
Desired fertility: # of children by age 50 (2013)Treatment effect -0.156 -0.233 -0.082 0.017 -0.015 0.048Standard error (0.083)* (0.118)** (0.117) (0.069) (0.099) (0.097)Comparison mean 3.656 3.659 3.652 3.620 3.652 3.588p-value on equality of effects (5)=(6): .362 (4)=(7): .109 (8)=(9): .648
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
4. Health BehaviorAcademic Major Admits Vocational Major Admits
All Female Male All Female Male
(4) (5) (6) (7) (8) (9)
Index of risky sexual behavior(safe-->risky)(2013)
Treatment effect -0.064 -0.059 -0.058 -0.066 -0.008 -0.114
Standard error (0.047) (0.066) (0.065) (0.039)* (0.056) (0.054)**
Comparison mean 0.023 0.120 -0.080 -0.013 0.082 -0.109
p-value on equality of effects (5)=(6): .991 (4)=(7): .973 (8)=(9): .175
Index of STI risk exposure (2013)
Treatment effect -0.123 -0.142 -0.094 -0.054 -0.022 -0.078
Standard error (0.047)*** (0.065)** (0.065) (0.039) (0.055) (0.054)
Comparison mean 0.044 0.142 -0.060 -0.019 0.072 -0.111
p-value on equality of effects (5)=(6): .605 (4)=(7): .254 (8)=(9): .467
Preventative health behavior (3 questions) (2013)
Treatment effect 0.170 0.179 0.174 0.073 0.081 0.070
Standard error (0.061)*** (0.086)** (0.085)** (0.051) (0.073) (0.071)
Comparison mean 1.621 1.703 1.534 1.613 1.674 1.551
p-value on equality of effects (5)=(6): .969 (4)=(7): .222 (8)=(9): .911
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
3. Tertiary education: Marginals vs. Inframarginals
• Gap more pronounced for vocational admits: • 16.2% continuation to tertiary among inframarginal
vocational admits; 5.4% among marginals• 22.4% vs. 16.4% respectively for academic admits
• No marginal/inframarginal gap for females• 17.9% continuation among infra-marginal females; 18.7%
among marginals; • 20.4% vs. 4.2% for males• Gender discrimination within the household? • Violation of assumption that no effect on propensity that
female infra-marginal SHS graduates go on to tertiary? (maybe scholarship increases effort, reduces interruption of school?)
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
School bill
Feb 2018Returns to Secondary Education: GhanaDuflo, Dupas, Kremer
3. Tertiary education: Plans vs success