DEVELOPMENT OUTREACH AREN’T CHILDREN LEARNING? WHY by Abhijit V

36 DEVELOPMENT OUTREACH AREN’T CHILDREN LEARNING? WHY BY Abhijit V. Banerjee and Esther Duflo

APRIL 2011 37 SPECIAL REPORTS

e are five What’s keeping children from learn- colleagues. Overall, teachers spend less ing? Or to reverse the question, what than half the time they are supposed to years away enables them to learn? In this article, we be teaching actually doing so (Chaud- offer a reading of the recent evidence, hury et al. 2006). from 2015, primarily but not exclusively from randomized trials, that, we hope, contrib- INCENTIVES HELP the year utes to an answer.

when the THE FIRST STEP IS SHOWING UP There is not enough pressure on teachers to teach. When such pressure WMillennium Development Goal of is brought to bear, they do teach more, Ensuring that children have access and students’ test scores improve, sug- universal education is supposed to schools and actually spend time there gesting that students can indeed be does matter. The data clearly show that taught (something teachers often ques- to be achieved, and the school children who spend more time in school tion), and that teachers know how to have better life outcomes (for example, teach (something education experts, attendance numbers do look Duflo 2001, Spohr 2003). The trouble who tend to insist on the need for train- is that while being enrolled is obviously ing, sometimes doubt). A randomized good. In many parts of both East a necessary condition for this, there are evaluation in nonformal schools in many reasons why enrollment by itself Rajasthan, India found that linking and West Africa, and almost all may not translate into much more ef- teacher compensation to attendance, fective schooling. The school year in by verifying attendance with objec- of South Asia, school enrollment India is only about 140 days, and each tive impersonal means (such as photos school day often lasts only 3 hours. By taken with tamper-proof date and time has grown rapidly, with primary contrast, children in most OECD coun- stamps), was effective. Teacher absenc- tries spend between 180 and 200 days es fell by half, from 42 percent to 21 school enrollment now exceeding in school, with longer school days of 6 percent. And, students learned more: to 8 hours. test scores rose by 0.16 standard de- 90 percent in many areas viations, and children were 50 percent ...AND TEACHING more likely to pass the exam allowing (UNESCO 2009). them to join formal schools (Duflo et al. 2010a). Another evaluation in India So why aren’t we celebrating? Moreover, being in the classroom is found that basing teacher pay on stu- The problem is that the children are less useful if the teacher is not there. dent performance was highly effective in school, but they are not learning. In In 2002 and 2003, the World Absentee- at improving student learning (Mu- India, for example, nearly 60 percent of ism Survey of six countries, led by the ralidharan and Sundararaman 2009). children in grade 4 cannot read a sim- World Bank, concluded that in Bangla- In Kenya, teachers hired on short con- ple story at grade 2 level, and 76 percent desh, Ecuador, India, Indonesia, Peru tracts, under supervision by the school cannot do simple division (Pratham and Uganda, teachers miss one day of committee, were much more likely to 2005). In neighboring Pakistan, 80 work out of five on average, and the ra- be present then regular teachers, and percent of children in grade 3 cannot tio is even higher (one in four) in India their students had higher test scores read a grade 1 paragraph (Andrabi et and Uganda. Their data from India also than those of regular teachers, even al. 2009). In Kenya, 27 percent of grade find that teachers who are in school do though the contract teachers had no 5 children cannot read even a simple not necessarily teach—they read the prior teaching experience (Duflo et al. paragraph (Uwezo 2010). newspaper, drink tea, or chat with their 2010b).

38 DEVELOPMENT OUTREACH score 0.31 standard deviations higher village, it is possible to account for PRIVATE SCHOOLS DO BETTER, BUT NOT BY A HUGE MARGIN in reading and 0.22 standard deviations roughly half to a third of the estimated higher in arithmetic. This likely re- overall gain in test scores from private mains an overestimate of the impact of schooling just by virtue of the fact that Another way to look at incentives is private school in India, if parents send private school teachers are more likely to compare children in private schools the most able children to private school to be at work. The rest may be the result with children in government schools. In or if they provide them with other ad- of teacher effort while in school, or bet- Colombia, students who won a lottery ditional inputs. ter pedagogy. for a private school voucher were 15 The net effect of private school is percent more likely than losers to attend thus not that much higher than the BUT INCENTIVES ARE ONLY PART private school, and scored significantly effect of improving incentives in the OF THE STORY higher on a standardized test (Angrist NGO (nongovernmental organization) et al. 2002). In Pakistan, students in pri- schools in Rajasthan. Indeed, part of vate schools increase average achieve- the effect of private school may be due In the 2000s, Pratham, a large NGO ment by 0.25 standard deviations each to the fact that private school teachers in India, trained balsakhis (children’s year, compared to students in public attend school more often: using the friends) to provide remedial education school. Self-selection is obviously an is- effect of teacher attendance estimated to the lowest performing 3rd and 4th sue here, but Sonalde Desai and others from the Rajasthan study combined graders in Vadodara and Mumbai mu- try to deal with it by comparing siblings with the estimate from the World nicipal schools. Balsakhis were mostly in India who belong to the same family. Bank’s study on absenteeism that pri- local high school girls with a week’s They found that, compared to their sib- vate school teachers in India are 8 per- training who were paid a relatively lings in public schools, primary school centage points less likely to be absent low salary of 1,000 Rupees per month, age children attending private school than public school teachers in the same ($62.50, at purchasing power parity).

APRIL 2011 39 40 DEVELOPMENT OUTREACH The primary focus was to teach basic participated in the program, so the ef- read) attended the evening remedial literacy and numeracy skills to students fect on those who did would have to be sessions. With the summer schools, the who were lagging behind. After one five times larger or about one standard corresponding number was 18 percent. year, these students’ test scores were a deviation(JPAL, 2009). very large 0.6 standard deviations high- A fourth study, also with Pratham, EDUCATION AS A LOTTERY er than those of similarly low achieving shows that even children who have children in comparison schools (Baner- mastered the basics can benefit from jee et al. 2007), and students initially at these types of programs though the ef- We propose a very simple theory the bottom of the class scored a whole fect may be smaller. In Bihar, India, an- to account for all of this, which we call standard deviation higher in the pro- other supplemental education program the education-as-lottery hypothesis. gram schools. was targeted at all children, including Surveys of parental aspirations suggest Another evaluation of a Pratham those who could already read. Pratham that the average semieducated or un- program measured the results of a vol- provided educational materials and educated parent sees education mainly unteer teacher program in Jaunpur, trained volunteers to use them. The as a way to secure a government or India, where school attendance is only evaluation suggests that children who other salaried job. For this reason, they 50 percent. More than 60 percent of the were taught by these volunteers saw think that education is only worthwhile children aged 7 to14 could not read and large gains as well (0.15 standard devia- if their child can get through the gate- understand a simple, first-grade level tions in math and 0.16 in language for keeping public exams that restrict access story. Pratham recruited and trained children in grades 3 to 5) [JPAL 2009]. to these kinds of jobs. All the evidence local volunteers in 65 randomly select- However, when Pratham trained gov- suggests that they are probably wrong. ed villages to conduct evening “camps” ernment school teachers in these tech- That is, while the evidence suggest that for two months. The volunteers typi- niques, rather than volunteers, and the the return to an extra year of educa- cally had a high school education and teachers were asked by the government tion in developing countries is more or received only a week of training, but the to use these techniques during the reg- less constant, parents believe that the children benefited from these camps. A ular school year, we see no evidence of returns are concentrated at the higher year later, children who initially could similar gains. levels of education: in Morocco for ex- not read anything were 60 percentage ample, parents believe that each year points more likely to decipher letters IT’S PUZZLING of primary education increases a boy’s than children in comparison schools. earning by 5 percent, but each year of Those who initially could already deci- secondary education by 15 percent. pher letters were 26 percentage points First, many of these gains seem The pattern was even more extreme more likely to be able to read and un- large relative to the gains from private for girls: parents believed each year of derstand a story (Banerjee et al. 2010). school. Why don’t the private schools primary education was worth almost In another program, in Bihar, In- adopt Pratham-style pedagogical tech- nothing, 0.4 percent. But each year of dia, government schoolteachers re- niques to improve their performance, secondary education was perceived to ceived special training from Pratham since it takes only a week’s training? increase earnings 17 percent. As a re- to conduct summer school, focusing Second, why do government school sult they believe that education is much on basic skills. Participating children teachers use the Pratham techniques more of a lottery than it really is. showed large learning gains. On aver- during the summer, but not during the Several implications follow from age, they tested 0.2 standard deviations school year? Third, why did parents and this hypothesis: higher than children in the compari- children not respond more enthusiasti- Given the winner-take-all nature of son group—comparable to the private cally to the offer of Pratham’s remark- education, it is very important to identify school effect—even though the sum- ably effective remedial programs? In the child who has the best chance of be- mer school program lasted only four Jaunpur only 8 percent of the children ing a winner as early as possible and put- weeks and less than one in five children (13 percent of those who could not ting all the resources behind him or her.

APRIL 2011 41 This is the child who gets sent to private This tendency to pick winners early those techniques focus on helping the av- schools, and we often see parents refer- and focus on them would explain why erage child master the basic concept bet- ring to her as the only smart child in the parents are not very excited by remedial ter and distract from “finishing” the sylla- family. In Pakistan, children perceived by education. If their child needs remedial bus. On the other hand, during summer their parents as more intelligent are four education, they feel, he is probably be- school, they were there explicitly to help times more likely to be enrolled in private yond help. the children to catch up and therefore schools (Andrabi et al. 2009, p. 100). In Because parents are focused on the willing to do what Pratham suggested. Burkina Faso, a study found that adoles- lottery, it is no surprise that the edu- What is true for government cents were more likely to be enrolled in cation system gets designed to reflect schools is probably even more so for school when they scored high on a test those preferences. Since the bet is on the private schools, which depend for their of intelligence, but they were less likely highest performing children, the focus existence on pleasing the parents. Why to be enrolled in school when their sib- in class is always to cover the whole syl- would we expect them to use techniques lings had scored high. The result is that labus even if the average child is totally that are meant for the average child? many children (perhaps a majority) get a lost. Think of those fourth graders who signal from their parents relatively early cannot read but get geography and his- THE EVIDENCE FOR OUR in their lives (the private school/public tory and science thrown at them. The HYPOTHESIS? school choice, for example, often happens whole system conspires against them at the primary school level) that they are on this—India’s Right to Education likely to be unsuited to education. It is Bill makes finishing the syllabus a legal A study by Trang Nguyen is highly no wonder that after this, many of them requirement. In Kenya, providing ad- consistent with this view. She finds that are mostly going through the motions in ditional textbooks benefited only those in Madagascar some parents consider- school, waiting for when they can drop students who were already at the top of ably overestimate the return to educa- out. This would explain why, for example, their class since the textbooks were far tion and some substantially underesti- child attendance rates in India are 70 per- too advanced to be useful to the rest of mate it, though on average they get it cent, worse even than teacher attendance the children (Glewwe et al. 2007). about right. However, they dramati- (ASER 2005). This explains why teachers do not use cally overestimate (by a factor of two) the Pratham techniques in class, since the chance that those who graduate

42 DEVELOPMENT OUTREACH from school will get a government job, in classes that were split without tracking, where a program that allowed pairs of making education more of a lottery and these gains persisted even one year af- children to play math learning games for (Nguyen, 2008). ter the program ended and all the students two hours a week generated gains of 0.39 Nguyen also finds that when - par were put back in the same class (Duflo et al. standard deviations, and those gains were ents who underestimate the returns are forthcoming). The children in the less ad- obtained at all levels of the distribution of given information about actual returns vanced tracked classes benefitted presum- test scores (Banerjee et al. 2007). to education, their children perform ably from the fact that, although the teacher The ultimate solution, however, has much better: their test scores improved was probably still teaching to the top of the to involve a wholesale attitude shift by by 0.37 standard deviations. An earlier (new) class, they were now nearer the top. everyone in the system from parents to study by Jensen also finds that in the educators. The good news is that if this Dominican Republic giving students shift takes place, very large gains can WHAT DOES THIS MEAN FOR information about the returns to edu- EDUCATION POLICY? follow. cation reduced the chance of dropping out (Jensen, 2010b). More recently, a Abhijit V. Banerjee is the Ford Foundation randomized evaluation in three North- A proper answer to this question International Professor of Economics ern States in India (also by Jensen) goes beyond the scope of this article. A at the Massachusetts Institute of found that once parents became aware few remarks however seem warranted. Technology (MIT). Esther Duflo is the of the high-paying jobs available to First, there is now huge pressure all over Abdul Latif Jameel Professor of Poverty educated young women through a re- the world to hire more teachers, but if Alleviation and Development Economics cruitment drive for call centers, they we are right, just cutting class size with- at MIT. Both authors are founders and were more likely to keep their daugh- out changing pedagogy will not work. directors of the Abdul Latif Jameel ters in school. In other words, this This is indeed what was found in India Poverty Action Lab (J-PAL) at MIT. convinced them that investing in their in the 1990s (Banerjee et al. 2005), and daughters was a better lottery ticket also in Kenya more recently (Duflo et References than they thought (Jensen 2010a). On al. 2010b). the other hand, interestingly, this study Second, because the long-term incen- Andrabi, Tahir, Jishnu Das, Asim Khwaja, also found that in response to the drive tives are distorted by the assumption of Tara Vishwanath, and Tristan Zajonc. parents reduced educational invest- a lottery, creating short-term rewards for 2009. “Pakistan Learning and Educational ment for boys they wanted to keep with educational success are all the more im- Achievement in Punjab Schools (LEAPS): them on the farm, and they increased portant. A program in Kenya that offered Insights to Inform the Education Policy the education of boys they wanted to girls who scored in the top 15 percent of Debate.” Washington, D.C.: The World Bank. send to the city. an exam a scholarship for the next year Angrist, Joshua, Eric Bettinger, Erik Bloom, A more indirect but compelling piece of worth about twenty dollars, not only got Elizabeth King, and Michael Kremer. evidence comes from a randomized evalu- the girls to do much better but also put 2002. “Vouchers for Private Schooling in ation of a tracking program in Kenyan gov- pressure on the teachers to work harder Colombia: Evidence from a Randomized ernment schools. Extra teachers were hired, (to help the girls), which meant that boys Natural Experiment.” The American and classes were split to allow for smaller did better too, even though there was no Economic Review, December: 1535–1558. class sizes. Some randomly selected classes scholarship for them (Kremer et al. forth- were divided into a more advanced and a coming). A computer-based teaching Banerjee, Abhijit, Rukmini Banerji, Esther less advanced class based on the children’s program that rewards successful learning Duflo, Rachel Glennerster, and Stuti Khemani. performance, while other classes were split by allowing kids to play games, should 2010. “Pitfalls of Participatory Programs: at random—what is sometimes called also work well in this environment, be- Evidence from a randomized evaluation in tracking. Children in the tracked classrooms cause, apart from everything else, it is a education in India.” American Economic (both those in the advanced and the less ad- way to create short-term incentives. This Journal: Economic Policy 2 (1): 1–30. vanced class) learned more than children is in fact what was found in Vadodara,

APRIL 2011 43 Banerjee, Abhijit, Shawn Cole, Esther Ratio, Teacher Management and Education Muralidharan, Karthik and Venkatesh Duflo, and Leigh Linden. 2007. “Remedying Quality” (mimeo). Cambridge, MA: Sundararaman. 2009. “Teacher Education: Evidence from Two Randomized Massachusetts Institute of Technology. Performance Pay: Experimental Evidence Experiments in India.” Quarterly Journal of from India.” NBER Working Paper 15323. Economics 122(3):1235–1264. Duflo, Esther, Rema Hanna, and Stephen Cambridge, MA: National Bureau of P. Ryan. 2010. “Incentives Work: Getting Economic Research. Banerjee, Abhijit, Suraj Jacob and Michael Teachers to Come to School” (mimeo). Kremer, with Jenny Lanjouw and Peter Cambridge, MA: Harvard University and Nguyen, Trang. 2008. “Information, Lanjouw. 2005. “Moving to Universal Massachusetts Institute of Technology. Role Models and Perceived Returns Education: Costs and Trade offs” (mimeo). to Education: Experimental Evidence Cambridge, MA: Massachusetts Institute of Duflo, Esther. 2001. “Schooling and from Madagascar.” MIT Working Paper. Technology. Labor Market Consequences of School Cambridge, MA: Massachusetts Institute of Construction in Indonesia: Evidence from Technology. Chaudhury, Nazmul, Jeffrey Hammer, an Unusual Policy Experiment.” American Michael Kremer, Karthik Muralidharan, and Economic Review 91(4): 795–813. OECD (Organisation for Economic Co- F. Halsey Rogers. 2006.”Missing in Action: operation and Development). Key Indicators Teacher and Health Worker Absence in Glewwe, Paul, Michael Kremer, and Sylvie on Development. www.oecd.org/document/5 Developing Countries.” Journal of Economic Moulin. 2007. “Many Children Left Behind? 5/0,3746,en_2649_37455_46349815_1_1 Perspectives 20(1): 91–116. Textbooks and Test Scores in Kenya.” NBER _1_37455,00.html. Working Paper. No. 13300. Cambridge, MA: Chou, Shin-Yi, Liu, Jin-Tan, Grossman, National Bureau of Economic Research. Pratham Annual Status of Education Michael and Ted Joyce. 2010. “Parental Report. 2005. Final edition. Education and Child Health: Evidence from J-PAL South Asia. 2009. Evaluation of scripts.mit.edu/~varun_ag/readinggroup/ a Natural Experiment in Taiwan.” American Pratham’s Read India: July 2009 Report to images/1/14/ASER.pdf. Economics Journal: Applied Economics Bihar Government. 2(1):33–61. www.povertyactionlab.org/south-asia. Spohr, Chris. 2003. “Formal Schooling and Workforce Participation in a Desai, Sonalde, Amaresh Dubey, Reeve Jensen, Robert. 2010a. “Economic Rapidly Developing Economy: Evidence Vanneman and Rukmini Banerji. 2009. Opportunities and Gender Differences in from ‘Compulsory’ Junior High School “Private Schooling in India: A New Human Capital: Experimental Evidence in Taiwan.” Journal of Development Landscape.” India Policy Forum Vol. 5: 1–58. for India.” NBER Working Paper. W16021. Economics 70(2): 291–327. Suman Bery, Barry Bosworth and Arvind Cambridge, MA: National Bureau of Panagariya, eds. New Delhi: Sage. Economic Research. UNESCO (United Nations Educational, Scientific, and Cultural Organization). Duflo, Esther, Pascaline Dupas, and . 2010b. “The (Perceived) 2009. EFA Global Monitoring Report: Michael Kremer (forthcoming). “Peer Returns to Education and the Demand Overcoming Inequality: Why Governance Effects, Teacher Incentives, and the Impact for Schooling.” The Quarterly Journal of Matters. www.unesco.org/new/en/ of Tracking: Evidence from a Randomized Economics 125(2): 515–548. education/themes/leading-the- Evaluation in Kenya.” American Economic international-agenda/efareport/ Review. See also NBER Working Paper. Kremer, Michael, Edward Miguel, and reports/2009-governance. W14475. Cambridge, MA: National Bureau Rebecca Thornton. “Incentives to Learn” of Economic Research. (forthcoming). Review of Economics and Uwezo. 2010. “Are Our Children Learning? Statistics. Cambridge, MA: Massachusetts Annual Learning Assessment, Kenya 2010.” Duflo, Esther, Pascaline Dupas and Institute of Technology Press. http://www.uwezo.net. Michael Kremer. 2010. “Pupil-Teacher

44 DEVELOPMENT OUTREACH Technology and Labor Productivity

Worldwide, labor has become nearly twice as productive over the last 20 years— and even more so in the developing countries, with Asia in the lead.

Labor productivity is critical to economic success; and productivity growth has three main sources: • capital deepening: the increase in capital per worker, with ICT particularly important. Capital deepening requires improving the business environment to enhance investors’ confidence and make investment opportunities more attractive. • growth in labor quality: the increase in the proportion of workers with high levels of education and experience, and • total factor productivity (TFP) growth: reorganizing production HOW ICT POLICY CONTRIBUTES TO LABOR PRODUCTIVITY processes using more and better Contributes to… technology and management. Policies CAPITAL LABOR TFP DEEPENING QUALITY GROWTH Policies to boost productivity growth must be strategic and must foster ICT infrastructure, especially + + + + + simultaneous improvements in all three broadband such as Internet access areas. This means: Competition and regulation to + + + + ++ • investing in human capital and improve quality and reduce costs of improving technology for better ICT products and services access to information, e-government for transparency, + + + + + • making education more accessible efficiency, and effectiveness. For and affordable, and example, e-procurement. • investing in ICT as a strategy of Promote ICT-enabled services and + + ++ ++ choice for boosting economic content development to foster ICT growth and competitiveness. use across sectors, organizations, and households for better decision making* Promote ICT-enabled services ++ + + + + Note: Most of the policy options are adapted and content development to foster from OECD (2008). The authors use the “+” technology diffusion to business* sign to express their own judgment of the ICT for education such as online + + + + + expected effect of each policy option: courses and other distance learning Education for IT development and + + + + + + + = strong effect maintenance + + = significant effect + = some effect Source: Dale Jorgenson and Khuong Vu (2010). “Potential Growth of the World Economy.” Journal of Policy Modeling, 32: 615–631.

* Government ICT policy should follow market principles and encourage the participation of the private sector as much as possible. Enhancing the benefits that users can reap from ICT- enabled services and products is more effective than providing them with subsidies.

APRIL 2011 45