Schooling and Learning: Understanding Inequalities and What To Do About Them Deon Filmer Development Research Group The World Bank DEC Policy Research Talk The World Bank March 18, 2014
Structure of this presentation Understanding inequalities in schooling attainment Addressing school participation shortfalls by working on the demand side Learning achievements and how to improve them
Schooling attainment inequality across the developing world: A global take
Education Attainment and Enrollment Around the World Database covers 316 datasets 97 countries Data are 191 DHS 49 MICS 76 IHS Regional coverage 39 SSA; 19 LAC; 16 ECA; 12 EAP; 4 MENA; 6 SA Variables include School attainment (ages 15-19) School participation (ages 6-10; 11-14) School survival curves (ages 10-19) All by Quintile Rural/Urban Gender Source: Filmer. 2014. Education Attainment and Enrollment around the World: An International Database. http://econ.worldbank.org/projects/edattain.
Angola Mali Niger Burkina Faso Comoros Central African Liberia Madagascar Rwanda Burundi Guinea Ethiopia Mauritania Senegal Togo Cote d'ivoire Sierra Leone Congo, Dem. Rep. Mozambique Uganda Gambia, The Malawi Benin Sao Tome and Cameroon Nigeria Zambia Ghana Swaziland Lesotho Tanzania Congo, Rep. Kenya Namibia Gabon South Africa Zimbabwe Proportion Inequalities related to poverty: Grade 6 completion of 15-19 year olds in the richest and poorest quintiles. Within-country inequalities are as big if not bigger than cross-country inequalities Sub-Saharan Africa 1 0.8 0.6 0.4 0.2 0 Richest quintile Poorest quintile Average grade 6 completion
Inequalities related to poverty: Grade 6 completion of 15-19 year olds in the richest and poorest quintiles. Within-country inequalities are as big if not bigger than cross-country inequalities Sub Saharan Africa Latin America and Caribbean Middle East/ North Africa South Asia East Asia and Pacific Europe and Central Asia 1 0.8 0.6 0.4 0.2 0
0 Proportion.2.4.6.8 1 Shared Prosperity? Proportion of 15-19 year olds who complete grade 6 Overall, and richest and poorest quintiles 0.2.4.6.8 1 Overall proportion Average Quintle 5 Quintle 1
0 Proportion.2.4.6.8 1 Shared Prosperity? Proportion of 15-19 year olds who complete grade 6 Overall, and richest and poorest quintiles 0.2.4.6.8 1 Overall proportion Average Quintle 5 (smoothed) Quintle 1 (smoothed)
0 Proportion.2.4.6.8 1 Shared Prosperity? Proportion of 15-19 year olds who complete grade 6 Overall, and richest and poorest quintiles 0.2.4.6.8 1 Overall proportion Average Quintle 5 (smoothed) Quintle 1 (smoothed) Quintle 2 (smoothed)
Shared Prosperity? Each 1 percentage point increase in average grade 6 completion is associated with the following percentage point increase in grade 6 completion within each quintile Below the median completion rate Above the median completion rate 2.0 2.0 1.8 1.8 1.6 1.6 1.4 1.4 1.2 1.2 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 0.0 Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Note: Bars show coefficient estimates; Dots show coefficient estimates including country fixed effects
0 Proportion.2.4.6.8 1 Shared Prosperity? Proportion of 15-19 year olds who complete grade 6 Overall, and richest and poorest quintiles 5 6 7 8 9 GNI per capita (ln, constant PPP 2005) Average Quintle 5 (smoothed) Quintle 1 (smoothed)
0 Proportion.2.4.6.8 1 Shared Prosperity? Proportion of 15-19 year olds who complete grade 6 Overall, and richest and poorest quintiles 5 6 7 8 9 GNI per capita (ln, constant PPP 2005) Average Quintle 5 (smoothed) Quintle 1 (smoothed) Quintle 2 (smoothed)
Shared Prosperity? Each 10 percentage point increase in GNI per capita is associated with the following percentage point increase in grade 6 completion within each quintile Below GNI PC of $1000 Above GNI PC of $1000 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0.0 Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 0.0 Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Note: Bars show coefficient estimates. GNI PC in 2005 constant PPP $
A Methodological Point: Using Asset Indices
Rationale for Asset Indices DHS/MICS do not include per-capita household expenditures, the typically preferred indicator of household poverty Use a ranking of households based on an index of assets and dwelling characteristics When using DHS/MICS, quintiles of the population based on this ranking
Albania Brazil Ghana Nepal Nicaragua Panama PNG South Africa Uganda Vietnam Zambia How do different approaches to measuring welfare compare? 1 Proportion of 15- to 19-year-olds who completed grade 6, by quintile.8.6.4.2 0 Per capita HH expenditures IRT index Predicted per capita HH expenditures Share weighted average PC index, all indicators Count index PC index, assets only Per capita value of durable goods Note: Symbols indicate the poorest quintile. Each marking shows the predicted gap from the previous quintile after controlling for dummy variables for age and gender. Source: Filmer and Scott, 2011. Assessing Asset Indices Demography.
Albania Brazil Ghana Nepal Nicaragua Panama PNG South Africa Uganda Vietnam Zambia How do different approaches to measuring welfare compare? 1 Proportion of 15- to 19-year-olds who completed grade 6, by quintile.8.6.4.2 0 Per capita HH expenditures IRT index Predicted per capita HH expenditures Share weighted average PC index, all indicators Count index PC index, assets only Per capita value of durable goods Note: Symbols indicate the poorest quintile. Each marking shows the predicted gap from the previous quintile after controlling for dummy variables for age and gender. Source: Filmer and Scott, 2011. Assessing Asset Indices Demography.
Albania Brazil Ghana Nepal Nicaragua Panama PNG South Africa Uganda Vietnam Zambia How do different approaches to measuring welfare compare? 1 Proportion of 15- to 19-year-olds who completed grade 6, by quintile.8.6.4.2 0 Per capita HH expenditures IRT index Predicted per capita HH expenditures Share weighted average PC index, all indicators Count index PC index, assets only Per capita value of durable goods Note: Symbols indicate the poorest quintile. Each marking shows the predicted gap from the previous quintile after controlling for dummy variables for age and gender. Source: Filmer and Scott, 2011. Assessing Asset Indices Demography.
Grade completion curves by quintile Patterns of grade completion are very different across countries Niger 2012 1 Cambodia 2010 1 Indonesia 2012 1 0.8 0.8 0.8 0.6 0.6 0.6 0.4 0.4 0.4 0.2 0.2 0.2 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 Poorest quintile Quintile 2 Quintile 3 Quintile 4 Richest quintile Poorest quintile Quintile 2 Quintile 3 Quintile 4 Richest quintile Poorest quintile Quintile 2 Quintile 3 Quintile 4 Richest quintile Source: Filmer,. 2010. Education Attainment and Enrollment around the World: An International Database. http://econ.worldbank.org/projects/edattain.
Changes over time in grade completion curves by quintile Niger 1992 Niger 1997 Niger 2006 Niger 2012 1 1 1 1 0.8 0.8 0.8 0.8 0.6 0.6 0.6 0.6 0.4 0.4 0.4 0.4 0.2 0.2 0.2 0.2 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 Poorest quintile Quintile 2 Quintile 3 Poorest quintile Quintile 2 Quintile 3 Poorest quintile Quintile 2 Quintile 3 Poorest quintile Quintile 2 Quintile 3 Source: Filmer,. 2010. Education Attainment and Enrollment around the World: An International Database. http://econ.worldbank.org/projects/edattain.
Other Correlates of Schooling Attainment Inequality
Other dimensions of inequalities Can extend analysis to analyze Gender Orphan status (in particular as a results of HIV/AIDS) Disability
0 1 2 3 4 5 6 7 8 Interaction between poverty and gender Grade 6 completion gaps by gender are larger in the poorest than the richest quintile Male / female ratio in grade 6 completion Poorest versus richest quintile Richest versus poorest quintiles 45 o line 0 1 2 3 4 5 6 7 8 Male-Female ratio in poorest quintile Male/Female ratio 45 degree line Source: Updated from Filmer, 2006. Gender and wealth disparities in schooling: Evidence from 44 countries International Journal of Educational Research.
Enrollment rate: Orphans Enrollment rate: Orphans Enrollment, conditional on individual and household characteristics, among orphans and non-orphans ages 7 to 14. There is substantial heterogeneity in the association between orphan status and enrollment. Two-parent orphans, and maternal orphans often but not always have lower enrollment. 100 Paternal orphans 45 degree line 100 Maternal orphans 45 degree line 100 Two-parent orphans 45 degree line 80 80 80 60 60 60 40 40 40 20 20 20 0 0 0 0 20 40 60 80 100 Enrollment rate: Both parents alive 0 20 40 60 80 100 Enrollment rate: Both parents alive 0 20 40 60 80 100 Enrollment rate: Both parents alive Note: Graphs show predicted enrollment after controlling for sex, age, urban/rural residence, household economic status, and geographic region. Solid points indicate that the difference between orphans and non-orphans is significantly different from zero at the 5 percent level. Source: Ainsworth and Filmer, 2006. Inequalities in Children's Schooling: AIDS, Orphanhood, Poverty, and Gender World Development.
Bolivia Burundi Cambodia CambodiaSES Chad Colombia India Indonesia Jamaica Mongolia Mozambique Romania South Africa Zambia Comparison of deficits in school participation among children ages 6 to 17: partial marginal effects from multivariate model The schooling deficit associated with disability is typically larger than that associated with other characteristics 0.0-0.1-0.2-0.3-0.4-0.5-0.6-0.7 With disability Female Rural Poorest-Richest quintile Source: Filmer, 2008. "Disability, poverty and schooling in developing countries: Results from 14 household surveys. World Bank Economic Review
What to do to address schooling deficits?
Distance to school, poverty, gender and school participation Data collected in early- to mid-1990s that included access to facilities module 24 DHS surveys from 21 countries 15 Sub-Saharan 1 East Asian (Philippines) 2 South Asian (Bangladesh, India) 3 Latin American/Caribbean (Bolivia, Dom. Rep., Haiti) Analysis limited to rural areas
Bangladesh 1993-94 Bangladesh 1996-97 India 1992-93 India 1998-99 Philippines 1993 Dominican Rep. 1991 Bolivia 1993-94 Madagascar 1992 Benin 1996 Nigeria 1999 Tanzania 1991-92 Cote d'ivoire 1994 Uganda 1995 Haiti 1994-95 Morocco 1992 Burkina Faso 1992-93 Niger 1998 Cameroon 1991 Zimbabwe 1994 Niger 1992 C.A.R. 1994-95 Senegal 1992-93 Mali 1995-96 Chad 1998 Estimated relationship between distance and school participation (rural population 6-14) 8 7 6 5 4 3 2 1 0 Average distance to nearest primary school (km) Effect of 1km increase in distance to primary school on probability of enrollment 0.01 0.005 0-0.005-0.01-0.015-0.02-0.025-0.03-0.035 Statistically insignificant Statistically significant Source: Filmer, 2007. "If You Build It, Will They Come? School Availability and School Enrolment in 21 Poor Countries. Journal of Development Studies
Simulated enrollment But the relationship between distance and school participation is small compared to the shortfalls in participation Simulating the effect of reducing distances to primary and secondary schools on school participation 1.0 Simulation: Distance to primary school reduced to zero 45 degree line 1.0 Simulation: Distance to primary and secondary schools reduced to zero 45 degree line 0.8 0.8 Cameroon 0.6 0.6 C.A.R 0.4 Chad 0.4 Benin Chad 0.2 Mali 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Actual enrollment 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Actual enrollment Source: Filmer, 2007. "If You Build It, Will They Come? School Availability and School Enrolment in 21 Poor Countries. Journal of Development Studies
Why limited association between distance and school participation? Selective school placement Not likely driver Demand side constraints Evidence of impact of scholarships/conditional cash transfers Quality of schooling Has potential impact on demand
Promoting schooling through demand-side transfers
Studying scholarships in Cambodia 1) Track-record of demand-side incentives in middle-income countries/lac. can they work in a low-income setting? 2) What s the right amount of a transfer what is the price elasticity? 3) What are the spillover-effects beyond the recipient?
Program and Evaluation design At the level of each program lower secondary school Rank incoming students according to poverty score Establish 2 cutoff points: Applicants with the highest dropout risk offered $60 per year scholarship somewhat lower dropout risk offered $45 and others offered no scholarship Use Regression Discontinuity Design for evaluation Compare just above to just below cutoffs
Probability Probability Large program impact First $45 is much more cost effective (almost no discernable additional impact of $60 over $45) No scholarship versus $45 $60 versus $45 scholarship 1 1 0.8 0.8 Estimate of impact 0.6 Estimate of impact 0.6 0.4 0.4 0.2 0.2 0-25 -15-5 5 15 25 Relative ranking 0-25 -15-5 5 15 25 Relative ranking Recipients Non-recipients Recipients Non-recipients Source: Filmer, and Schady. 2011. Does More Cash in Conditional Cash Transfer Programs Always Lead to Larger Impacts on School Attendance?, Journal of Development Economics
Enrollment Enrollment Results from medium-term follow-up: Sizeable impacts beyond the scholarship period itself, especially for girls Girls Boys 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 2005 2006 2007 (Grade 7) (Grade 8) (Grade 9) 2008 (Grade 10) 2009 (Grade 11) 0.0 2005 2006 2007 (Grade 7) (Grade 8) (Grade 9) 2008 (Grade 10) 2009 (Grade 11) Impact estimate Non-recipient mean Impact estimate Non-recipient mean Source: Filmer and Schady. Forthcoming. The Medium-Term Effects of Scholarships in a Low-Income Country Journal of Human Resources
Other dimensions of impact Scholarship recipients were about 10 percentage points less likely to work for pay. No negative spillovers onto the enrollment or labor supply of siblings (i.e. non-recipient siblings are not withdrawn from school) No negative spillovers onto the enrollment of other non-recipient students in the same schools Source: Filmer, Ferreira and Schady. 2009. Own and sibling effects of conditional cash transfer programs : theory and evidence from Cambodia WBPRP No 5001
Summary of CCT/Scholarship programs evaluated to-date Country Age/Grade/Gender Baseline enrollment Impact Chile Ages 6-15 60.7 7.5*** Colombia Ages 8-13 91.7 2.1** Ages 14-17 63.2 5.6*** Ecuador Ages 6-17 75.2 10.3** Honduras Ages 6-13 66.4 3.3*** Jamaica Ages 7-17 18 days out of 20 0.5** Grades 0-5 94 1.9 Mexico Grade 6 45 8.7*** Grades 7-9 42.5 0.6 Nicaragua (1) Ages 7-15 90.5 6.6*** Nicaragua (2) Ages 7-13 72 12.8*** Bangladesh Ages 11-18 (Girls) 44.1 12.0** Cambodia (1) Grades 7-9 (Girls) 65 31.3*** Cambodia (2) Grades 7-9 65 21.4*** Malawi Ages 13-22 (Girls) 76.9 6.1*** Pakistan Ages 10-14 (Girls) 29 11.1*** Ages 6-11 93.3 4.5*** Philippinnes Ages 12-14 84.5 3.9 Ages 15-17 62.3-2.7 Tanzania Turkey Ages 0-18 (midline) Primary school 59 87.9 6-3.0* Ages 0-18 (endline) Secondary school 59 39.2 4 5.2 Impacts tend to Be larger when baseline enrollments are lower Be larger at transition points Be larger for poorer families Source: Updated from Fiszbein and Schady (2009)
-2-1 -2-1 0 1 2 Vocabulary score (normalized) 0 1 2 But No impact of program on measures of learning Mathematics Vocabulary Estimate of impact Estimate of impact -25-15 -5 5 15 25 Relative ranking Mean Quartic -25-15 -5 5 15 25 Relative ranking Mean Quartic Source: Filmer and Schady. Forthcoming. The Medium-Term Effects of Scholarships in a Low-Income Country Journal of Human Resources
Consistent with findings elsewhere Ecuador and Mexico: No increase in test scores among recipient students Mexico: No increase in test scores for those who were offered a transfer (despite increased enrollment)
Why limited impacts on learning? Marginal students? Selection into schooling based on expected gains Poor quality/inappropriate teaching? Although in Cambodian case impacts aren t larger in higher quality schools Other explanations?
What is poor quality? Teaching time lost Percent of time officially allocated to schooling; when a teacher is present; and spent in teaching and learning activities 100 80 60 40 20 0 Official time Presence time Time on task
Severe Shortfalls in the Delivery of Education Services Service Delivery Indicators Kenya Senegal Tanzania Uganda (Public schools only) Share of teachers with minimum knowledge: English/French 10% 29% 9% 4% Mathematics 75% 75% 73% 36% Classroom teacher absence rate 47% 29% 53% 57% Classroom teaching time 2h 19m 3h 15m 2h 04m 2h 58m (scheduled teaching time) (5h 40m) (4h 36m) (5h 12m) (7h 20m)
Percent Service Delivery shortfalls have consequences 100 80 60 40 20 0-20 -40-60 -80-100 Percent of SACMEQ 6 th grade Math test-takers who score at each performance level 60 50 40 30 20 10 0 Percent of PASEC testtakers who perform at a level less than random guessing Competent and above (5,6,7,8) Beginning numeracy (Level 4) Pre, Emergent and Basic numeracy (Levels 1,2,3) Math French
Making Schools Work: New Evidence on Accountability Reforms Reviews evidence on impact evaluations of Information for accountability School-based management Teacher incentives
Return to Cambodia
Primary School Scholarships Program Schools randomly assigned to Treatment cohort of students that began receiving scholarships in 2008/2009 school year Control same cohort did not receive scholarships Half of the schools each randomly assigned to Poverty targeted scholarships Merit-based scholarships
-15-10 -5 0 5 10 15 Is there a targeting tradeoff? A) 21% B) 27% C) 27% D) 25% -15-10 -5 0 5 10 15 Relative poverty ranking
Primary school scholarships increased schooling participation attainment 0.35 0.30 *** 0.25 0.20 0.15 0.10 *** *** *** 0.05 0.00 Poverty targeted scholarships Probability of reaching grade 6 Merit targeted scholarships Highest grade completed Source: Barrera-Osorio. 2013. Incentivizing Schooling for Learning WBPRWP No. 6541
Standard Deviation increase Merit recipients performed better on tests, poverty recipients did not 0.20 0.15 0.10 0.05 0.00-0.05 * * -0.10 Poverty targeted scholarships Merit targeted scholarships Math test Digitspan test Source: Barrera-Osorio. 2013. Incentivizing Schooling for Learning WBPRWP No. 6541
Standard deviation increase Even poor merit-based scholarship recipients improved their test scores 0.25 0.20 0.15 * * 0.10 0.05 0.00-0.05-0.10 Poverty below median achievement Poverty above median achievement Merit below median poverty Merit above median poverty Math test Digitspan test Potential role of framing of cash transfer Source: Barrera-Osorio. 2013. Incentivizing Schooling for Learning WBPRWP No. 6541
Science of Delivery? 2005 Prior programs Evaluation of prior secondary scholarship programs Implement secondary scholarships with transparent targeting Evaluation of impact of secondary school scholarships 2010 Cambodia Education Sector Support Program (P070668) Education Sector Support Scale Up Action Program (P109925) Scale-up at lower transfer amount Implement primary school scholarships with evaluation about of how to increase learning Build better school readiness through early child development Evaluation of primary school scholarships Evaluation of ECD programs 2015 Cambodia Global Partnership for Education Second Education Support Project (P144715) Implement primary and secondary school scholarships targeting poor with academic potential Scale-up access to ECD with careful attention to implementation quality Proposed evaluation of redesigned ECD scale-up
What to make of all this? Inequalities in schooling are large Especially with respect to poverty Prosperity not always shared Demand-side interventions work But implications for learning limited Service delivery is poor in many counties With implications for learning outcomes Attention should be paid to framing of interventions Can affect motivation and effort with impact on outcomes
Thank you!