Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Essays on education programs in Costa Rica
(USC Thesis Other)
Essays on education programs in Costa Rica
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ESSAYS ON EDUCATION PROGRAMS IN COSTA RICA by Jaime Andrés Meza-Cordero A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (Economics) August 2014 Copyright 2014 Jaime A. Meza-Cordero ii To my father and all those who guided me… iii Acknowledgements I would first like to thank my advisors and friends John Strauss and Jeffrey Nugent. Their continuous guidance significantly helped me become a better researcher, and for their countless hours of advice I will always be grateful. I would also like to thank Anant Nyshadham and Guilbert Hentschke for being part of my dissertation committee, and Geert Ridder, Eileen Crimmins and Richard Easterlin for being part of my qualifying committee, their feedback was very helpful. This dissertation would have not been possible without the contributions of Aldrich Chan, Horacio Alvarez-Marinelli, Daniel Castro-Johanning, Cesar Rodríguez-Barrantes, Leonardo Garnier and Jose Antonio Li; my most sincere gratitude to each of them. I also have to thank the Department of Economics of the University of Southern California; and particularly to Young Miller and Morgan Ponder for their valuable assistance during the last years. I especially would like to thank the Quirós-Tanzi Foundation for trusting me to evaluate their project, and I gratefully acknowledge funding from the Department of Economics of the University of Southern California and the Inter-American Development Bank for the creation of an original data set. I am also very indebted to the Costa Rican Institute of Census and Statistics (INEC) for allowing me access to their databases. I was very blessed with having classmates that were always willing to help during the most complicated stages of the PhD program. I will always be grateful to all of them, but especially to Fei Wang, Mehdi Shoja, Qi Sun and Malgorzata Switek who are greatly responsible of my accomplishments. I could not forget to mention how much I appreciate my friends Danilo Beteto, Vlad Radoias, Manuel Castro, Brijesh Pinto and Juan Sotes- Paladino for making this journey such a good experience. Finally, but most importantly, I would like to thank my family and friends, their constant support made this dissertation possible. iv Table of Contents Acknowledgements iii List of Tables vi List of Figures viii Abstract ix 1 Introduction 1.1 Program Evaluation in Education ………………….………………………. 1 1.2 Literature Review ……………………………….……………………….…. 2 1.3 Education in Costa Rica …………………………………………………… 5 2 The Formation of Technologically Skilled Human Capital in School: Evaluation of the One Laptop per Child Program in Costa Rica 2.1 Introduction ………………………………………………………………… 7 2.2 Literature Review ………………………………………………………….. 10 2.3 The Conectándonos Program ……………………………………………… 12 2.4 Data ……………………………………………………………………...… 14 2.5 Empirical Stategy ………………………………………………………….. 21 2.6 Results …………………………………………………………………...… 24 2.6.1 Computer Usage …………………………………………………...… 24 2.6.2 Time Allocation ………………………………………………...…… 24 2.6.3 Student and Parental Aspirations ……………………………………. 28 2.6.4 Test Scores …………………………………………………………... 28 2.6.5 Heterogeneous Effects ………………………………………………. 31 2.6.6 Attrition ………………………………………………….…………... 36 2.6.7 Pooled Cross Section Estimation …………...…...…………………... 36 2.7 Conclusion ………………………………………………………………… 42 v 3 The Effects of Subsidizing Secondary Schooling: Evidence From a Conditional Cash Transfer Program in Costa Rica 3.1 Introduction …………………………………………..…………………… 44 3.2 Literature Review …………………………………………………………. 45 3.3 The Avancemos Program …………………………………..………….….. 47 3.4 Data: The Costa Rican Household Survey …………………………….….. 48 3.5 Theoretical Framework …………………………………….……………... 54 3.6 Empirical Strategy ………………………………………………………... 58 3.7 Results …………………………………………………………………..… 63 3.7.1 Difference in Difference Propensity Score Matching ……………...… 63 3.7.2 Nearest Neighbor Propensity Score Matching …………………….… 64 3.7.3 Caliper Propensity Score Matching ……………………………….… 64 3.7.4 Kernel Propensity Score Matching ………….………………………. 65 3.7.5 Difference in Difference ………...……………….………………….. 65 3.7.6 Heterogeneous Effects ………………………………………………. 65 3.8 Conclusion …………………………………………………………...….… 68 4 The Effects of Childhood Health on Education and Economic Outcomes in Adulthood 4.1 Introduction ……………………………………………………………...… 69 4.2 Literature Review …………………………………………………………. 70 4.3 Data ……………………………………………….…………………….… 72 4.4 Empirical Strategy ………………………….…………………………..… 75 4.5 Results ………………..………………………………….……………..… 76 4.6 Conclusion ……………………………………………………………...… 77 5 Conclusion 78 Bibliography 80 Appendix 85 vi List of Tables 2.1 Baseline Characteristics: Treatment and Control Schools ……………...……… 15 2.2 Baseline Characteristics: Information from Students …………………….……. 17 2.3 Baseline Characteristics: Information from Parents ……………………..…….. 18 2.4 Effects of the OLPC Program on Weekly Computer Use ……………………... 25 2.5 Effects of the OLPC Program on Specific Computer Uses ……………………. 26 2.6 Effects of the OLPC Program on Weekly Time Allocation …………………… 27 2.7 Effects of the OLPC Program on Aspirations ………………………………….. 29 2.8 Effects of the OLPC Program on Test Scores …………………………………. 30 2.9 Heterogeneous Effects of the OLPC Program on Weekly Computer Use …...... 32 2.10 Heterogeneous Effects of the OLPC Program on Specific Computer Uses …… 33 2.11 Heterogeneous Effects of the OLPC Program on Weekly Time Allocation …... 34 2.12 Heterogeneous Effects of the OLPC Program on Aspirations ………………… 35 2.13 Differential Attrition …………………………………………………………… 37 2.14 Pooled Cross Section: Computer Use Reported by the Students ………………. 38 2.15 Pooled Cross Section: Effects of the OLPC Program on Specific Uses ………. 39 2.16 Pooled Cross Section: Effects of the OLPC Program on Time Allocation ……. 40 2.17 Pooled Cross Section: Effects of the OLPC Program on Aspirations …………. 41 3.1 Summary Statistics: Eligible Students …………………………………………. 51 3.2 School Outcomes: Students Ages 12-18 …..…………………………………… 53 3.3 Probability of Selection for the Program at Baseline …………………………... 62 3.4 Effects of the Conditional Cash Transfer Program on School Completion ……. 66 3.5 Heterogeneous Effects of the CCT Program on School Completion ………...... 66 3.6 Effects of the Conditional Cash Transfer Program on Hours Worked ………… 67 3.7 Heterogeneous Effects of the CCT Program on Hours Worked ……………...... 67 vii 4.1 Summary Statistics ………………………………………………………….….. 74 4.2 Estimates of Measures of Childhood Health on Economic Outcomes ……….... 76 viii List of Figures 2.1 Timeline of the Data Collection ………….………….……………………….… 16 2.2 Access to Technology ………………………………….………………….…… 20 2.3 Educational Objective for the Child …………………………….…….……….. 20 2.4 Desired Profession of the Parents for the Students …………………………….. 20 3.1 Average Years of Schooling 2000 to 2010…………………………….……...... 52 3.2 Average Hours Worked 2000 to 2010 …………………………………………. 52 3.3 Reasons Given by School Drop-Outs ……………………………………...…... 52 3.4 Distribution of Propensity Score: Treatment and Control 2007 ……………...... 61 3.5 Years of Schooling by Age at Baseline ….………….……………………..…... 63 3.6 Years of Schooling by Age Post-Intervention………….………………………. 63 4.1 Self-Reported Childhood Health ……………………………………………..... 74 A.1 The XO Laptop ………………………….……………………….……….……. 85 A.2 Map of the 4 Treatment Districts ………………………….…………………… 85 A.3 Parental Questionnaire ……………………..…………………….……….……. 86 A.4 Student Questionnaire ………….………………………….…………………… 90 ix Abstract This dissertation analyzes three different determinants for education in Costa Rica. The study consists of three essays; the first essay evaluates a program designed to expand the skills learned during primary school, the second essay evaluates a program aimed at improving education completion in secondary school, and finally, the third essay studies the existing association between health in early life and education outcomes in adulthood. The first essay (Chapter 2) assesses the impacts of a program that donates laptop computers to children in public elementary schools in Costa Rica. This initiative aims to improve the quality of the education received by teaching the students to be technologically literate. In collaboration with the NGO implementing the program, we collected baseline and post-intervention information from the 15 primary schools that were treated and from 19 primary schools serving as a control group. From these two sets of schools, we were able to construct a database of approximately 6500 observations and 71 variables. Using a difference in difference design, this essay estimates the effects of the program on 4 outcomes: computer usage, time allocation, aspirations and test scores. The main finding is that the program leads to an increase in computer use of about 5 hours per week for the treated students. Moreover, this chapter provides evidence that the treated students use the computer for browsing the internet, doing homework, reading and playing games. There is also evidence that the program leads to a decline in the time that students spend on homework and outdoor activities. No evidence is found of the program increasing computer usage by other family members. These results confirm the importance of computer access in low-income communities in order to promote a more technologically skilled labor force. The second essay (Chapter 3) evaluates the effects of a conditional cash transfer program that the Costa Rican government introduced in 2007. Following the success of Conditional Cash Transfers Programs in numerous Latin American countries, Costa Rica unraveled its very own program under the name Avancemos. This program was based on x giving students’ parents a monthly subsidy, conditional on mandatory school attendance for their children. The data used in this chapter came from a national household survey, which is conducted every year by the Costa Rican Government. This essay empirically tests the hypothesis that the program is improving school completion and consequently reducing dropout rates. The effects of the program on years of school completed are estimated using a difference in difference propensity score matching technique. The main result is that after 2 years of intervention, the conditional subsidy increases the years of school completed of treated students by 0.62 years. No evidence is found regarding impacts of the program on hours worked by eligible adolescents. The fourth essay (Chapter 4) studies the association between early life health and education and economic outcomes in adult life. This chapter relies on 2 measures of childhood health: a retrospective four-point scale assessment of health status while growing up, and adult height to the knee, which is known to be a good measure of nutrition and overall health during early ages. The data used in this essay came from the Costa Rican Longevity and Healthy Aging Study, which is a longitudinal study of a nationally representative sample of 2827 adults ages 60 and over. This chapter provides evidence of a strong correlation between early age health and adult education and economic outcomes, for both males and females. 1 Chapter 1 Introduction “Education is the most productive direct investment. Every step in this direction for any country in the Americas will contribute to broaden the political, economic and spiritual horizons of the hemisphere.” José Figueres Ferrer 1.1 Program Evaluation in Education Human capital plays a very important role in economic development. Increased quantity and quality of education, leads to increased productivity; ultimately improving wages and profits. Having a well-educated labor force also facilitates the adoption and efficient use of the most modern technologies, which has the potential to help less developed countries develop faster. In general, governments are well aware of the importance of education, for this reason expenditures in education are commonly one of the top components of government spending. From a public polity standpoint, the efficient use of resources toward education has been and will always be an essential topic to discuss. Many studies have shown that the relation between years of education and income has decreasing returns (see Psacharopoulos, 1985). Therefore, it is very important to guarantee that students remain in school during their childhood and adolescence. It is also a well-known fact that schooling and poverty are negatively correlated. For this reason governments have the responsibility to stimulate social mobility by providing opportunities to the most unprivileged children. Education in the developing world is currently facing many challenges, which in some cases are persistent and widespread across countries. Some of the most recurring problems are: low attendance rates, high dropouts rates, and bad overall teaching quality. Many 2 factors are known to influence school outcomes. Some of the best-known factors are the quality of the school infrastructure and teachers, distance to the nearest school, health of the students, and the opportunity cost associated with attending school. While institutional changes are needed to improve overall educational performance, targeted initiatives seem to have a better political acceptance in the short run. Several countries have been utilizing this approach, where targeted programs are developed to improve education. Some of the most popular programs provide monetary incentives to families, conditional on school attendance. Other programs focus on improving the school inputs, such as infrastructure and teacher quality. Glewwe and Kremer (2006) describe different programs aimed at improving school performance. They list interventions such as: raising the teachers’ incentives, providing vouchers to increase the variety of school choices, and decentralization of the school system. The authors argue that the results from these programs are mixed and inconclusive. For this reason, they recommend the use of randomized evaluations as the best tool for evaluating education programs. 1.2 Literature Review The literature on the effects of school inputs on performance is very broad; Hanushek’s (1997) review on this topic concludes that there is no consistent relationship between school resources and student performance. Nowadays, research on the evaluation of education programs is expanding as new programs are being implemented at small, medium or even national scales. One program that has become very widespread for increasing school completion is Conditional Cash Transfers (CCT). The best-known CCT program is the “Progresa/Oportunidades” in Mexico 1 . This type of program is targeted to combat high dropout rates, by providing a monetary stipend to the family of the student, conditional on the student attending school regularly, and in some countries, also on other 1 Well documented by Schultz (2004), Behrman et al (2005), Parker et al (2007) and Behrman et al (2011). 3 requirements such as periodic medical check-ups. The literature studying this program has found significant increases in school enrollment and completion ranging between 0.6 and 0.9 additional years depending on the study and duration of the intervention. A very noteworthy CCT program is Bolsa Escola in Brazil. This program gave a subsidy to unprivileged families with children aged 6-15 conditioned on school attendance. Costa Resende & De Oliveira (2008) found significant effects on food consumption quality, diversification and quantity. Another well-known CCT program in the region is Colombia’s Familias en Acción. As presented by Attanasio et al. (2005), this program had two components, for children ages 0-7 a monetary subsidy intended to improve nutrition, and for children 7-17 a monetary transfer conditioned on attending school regularly. On this study, the authors found a significant increase of 10% on school attendance in rural areas and of 5% in urban areas. There is a growing literature on the effects of technology adoption in school. As an attempt to improve school quality many countries have been engaging into programs that teach students how to use computers. Angrist and Lavy (2002), Banerjee et al (2007), and Barrera-Osorio and Linden (2009) studied the effects of incorporating a computer laboratory in schools of Israel, India and Colombia, respectively. All these studies find very small or no effects on test scores after having a computer laboratory. The overall conclusion from this literature is that unless the computers are put into action for guided use toward a specific class, test scores will remain unchanged. In a recent study, Muralidharan et al (2012) examined the household responses toward changes in school inputs. They show that when schools receive unanticipated grants, test scores increase. However, if the grants were anticipated, households would reallocate resources, and then the test scores remain unaltered. Therefore, if household and school inputs are substitutes, unanticipated inputs are likely to have a larger effect in improving performance. A research area that is related to changes in school inputs has to do with connectivity and the access to information. In his paper, Jensen (2007) shows how access to timely information leads to more efficient economic outcomes. Furthermore, in a more recent 4 paper, Jensen (2010) performed an experiment on high school students in the Dominican Republic. He randomly selected schools and provided information about jobs with higher measured returns to the students in the treatment group. The study finds that treated students completed significantly more years of schooling than their non-treated counterparts. Another very important determinant of education is student health. As presented by Behrman (1996), child health and nutrition are very important determinants of educational achievement. His study indicated that child health and nutrition are strongly associated with educational attainment; however, he warns that this does not necessarily imply causality. Another study that analyzes the associations between health and education is Ross and Mirowsky (1999). The authors found a significant correlation between health and years of schooling for adults ages 18 and older. In their conclusion it is argued that these results are attributed to economic conditions and a healthy lifestyle. Different studies have focused on the effects of nutrition on education achievements. Using data from the Philippines, Glewwe et al. (2001) found that malnourished children have lower performance in school, even after controlling for unobservable characteristics. The authors argue that good nutrition allows for starting education at a younger age and for greater learning productivity. In another study, Miguel and Kremer (2004) evaluated a program in Kenya that provided de-worming drugs to randomly selected schools. Their study found that this program significantly improved health and school participation of both treated and untreated students, due to positive externalities. Maluccio et al. (2009) analyzed the effects of childhood nutrition on educational outcomes in adulthood. Their paper estimates the impact of a community-level randomized nutritional intervention in rural Guatemala. This program gave a nutritional supplementation to randomly selected children during their first 3 years of life. The main findings of this study are an increase in school completion by treated women, and positive effects on reading comprehension and nonverbal cognitive tests scores for treated men and women. 5 1.3 Education in Costa Rica Costa Rica is a middle-income developing country in Central America with about five million habitants. According to the United Nations’ Human Development Report for 2011, the literacy rate in Costa Rica is one of the highest in Latin America at 96.1%. One of the main reasons for this high rate is that since 1870 public primary education is by constitutional law of universal provision and of no cost to students. At the same time, the government was set with the obligation to spend at least an 8% of the gross domestic product on public education, both primary and secondary. Despite the good quality and coverage of the public school system in this small country, in recent years, school dropout in secondary school has become a problem. The 2006 Costa Rican Household Survey shows that mean school completion is about 8.6 years. While about 99% of primary school aged children (ages 6-12) attend school, about 14% of the students that started high school did not finish. Many of these students dropped out of school to work 2 . In 2006, only 79% of secondary school aged adolescents (ages 12-17) were attending school. In recent years the Costa Rican Government has been very active in trying to facilitate and improve educational conditions. Some of the most important policies included eliminating national tests for 6 th graders as an attempt to reduce the number of dropouts that were occurring during the transition from primary school to secondary school. However, without a doubt, the most relevant initiative was giving a monthly subsidy to selected households conditioned on having their children attend secondary school regularly. This program was intended to decrease dropouts, improve attendance and increase school completion. This dissertation consists of evaluating two different education programs and assessing the importance of health on education in Costa Rica. The structure of this dissertation is as follows: Chapter 2 studies the effects of a program that donates laptop computers to students in elementary school, Chapter 3 analyses the impact of a program that gives monetary subsidies to poor families conditional on the school attendance of their kids, and 2 The minimum age for formal work is 15, however informal work is common for unprivileged children. 6 Chapter 4 assesses the association between health in early life and education and economic outcomes in adulthood. Finally, Chapter 5 has concluding remarks and policy implications. By studying these three different determinants of schooling outcomes, this dissertation aims to contribute to understanding the importance of education programs and of student health in a middle-income developing country. 7 Chapter 2 The Formation of Technologically Skilled Human Capital in School: Evaluation of a One Laptop per Child Program 2.1 Introduction Information and communication technologies (ICTs) have permanently transformed the economic markets as the global economy now heavily depends on the availability of immediate information. This transformation has led to major changes on the labor market, as the new higher paying jobs require a labor force with technological skills. The big challenge for developing countries is being able to raise their human capital to take advantage of these new opportunities. As stated by the World Bank Group (2012), “ICTs have great promise to reduce poverty, increase productivity, boost economic growth, and improve accountability and governance”. In an effort to instruct students in technological skills, computers are now entering classrooms. Assuming this trend continues, the real challenge becomes how to train these students in the use of computers, enhancing while not compromising their learning process. In recent years, different programs aiming to modernize school inputs have been introduced in many parts of the world, and have subsequently allowed technology to reach more children. However, such initiatives are often very expensive, and the lack of empirical evidence showing the overall effects of these programs has limited the ability to make well-informed decisions about implementation and improvement. For this reason, it is imperative that these initiatives be evaluated. One very widespread program is the One Laptop per Child (OLPC). The students that belong to an OLPC program are given a laptop computer, which they must carry into the classroom every school day, and in some countries they can take it home throughout the 8 school year. This revolutionary project was created at the MIT Media Lab, and was designed to provide students in elementary schools with low cost and durable laptop computers. The main objective of this project is to provide unprivileged kids with the opportunity to access a computer, helping to close the digital divide that exists across different SES groups and across countries. This initiative is strongly linked to what’s known as 21 st Century Skills, which suggests that society and human capital require new skills that are not well incorporated into the current education systems. The OLPC program also belongs to the group of initiatives known as Information and Communication Technologies for Development (ICT4D), which argues that access to more and better information furthers the socio-economic development of society. OLPC programs have expanded rapidly across the globe; as of today, 2.4 million computers have been distributed in 44 different countries, mostly in the developing world 3 . Latin American countries that have implemented the program include: Brazil, Peru, Colombia, Uruguay, Haiti, Paraguay, Nicaragua and recently Costa Rica (Nugroho and Londsale 2009). Furthermore, many governments are currently facing strong pressures to adopt a more modern education system, in many cases suggesting the universal distribution of laptops or tablets in their public school systems. Despite its popularity, critics of these OLPC initiatives argue that the program is: expensive, disrupts the educational process, puts an additional burden on the teachers, results primarily in using the computer for playing games, and that it is usually abandoned after the initial computers break down. Nevertheless, most do tend to agree that this type of program revolutionizes the conventional teaching techniques and tools, in which teachers and textbooks are the only source of information, by now allowing students to find knowledge by themselves. Most of the existing literature on technology adoption in education has based its analysis on the effects on school performance; finding little or no evidence of positive outcomes. The main unanswered question remaining is: why does the use of computers not seem to improve performance? I list two possible explanations. The first one is that the computers are not being used much, or at least not being used properly. The second explanation is 3 A complete and updated list can be accessed at http://one.laptop.org/map 9 that the computers are being used, but that the skills developed are orthogonal to the topics tested. This chapter’s premise is that, even with the computer being considerably used, the academic performance in elementary school is not going to be significantly affected, since laptops are not a tool destined to substitute for a blackboard, paper, or pencils; but are a tool designed to be complementary to the above mentioned. Under this premise, this chapter aims to contribute to the literature of technology adoption in schools by estimating the effects of the OLPC program from previously unexplored standpoints: computer usage intensity, changes in time allocation and changes in aspirations. The first focal point of this study is the most fundamental matter of interest, computer usage. The analysis will aim to determine the intensity of use of the laptop inside and outside of the household. Once usage is assessed, this study moves on to explore the specific uses that the students give to the computers; if they are using the tool for educational purposes (such as doing homework or finding information online) or for leisure (such as drawing or playing games). The second objective of this research is to determine how having a computer affects the intra-household relationships and time allocations within the household. To assess these effects, the time expenditures of the student doing homework, household duties, and outdoor activities will be determined. Further analyzed is the time that the parents devote to helping the student with homework, and if the parents or other family members share the computer with the student. The third objective of this chapter focuses on how the access to information affects student and parental aspirations. It is explored whether there is a change in the type of occupation desired for the student in adulthood, and also if there is a change in parental expenses on education. Finally, the effects on school performance are examined using 2 different tests. Given that the software loaded on the laptops contained applications designed to assist with math problems, a math test on the 5 th graders was applied. At the same time, since other installed applications were designed to improve the creativity and non-verbal abstract reasoning of the students, a cognition test was applied on the 6 th graders. 10 In this chapter I evaluate the one-year impacts of the OLPC Program in Costa Rica. With the collaboration of the Quirós Tanzi Foundation (the NGO responsible for funding and implementing the program), a treatment group of 15 primary schools was selected to receive laptops during the first week of classes in February 2012 4 . A control group of 19 primary schools located in the same districts as the treatment schools was also selected. The effects of the program are identified using a difference in difference estimator. My findings indicate that the program leads to a very large increase in computer usage, as treated students use a computer almost 5 hours more per week than their non-treated counterparts. The results also show that the treated students use the equipment for various applications such as browsing the Internet, reading and playing games. There is evidence of the students reducing their time doing homework and performing outdoor activities in about 1 hour per week. No evidence is found suggesting that the computer is being shared among other family members. Furthermore, no evidence is found suggesting a change in aspirations. The chapter consists of the following sections: Section 2 reviews the related literature on school inputs and OLPC programs. Section 3 describes the program, while Section 4 explains the stages of data recollection. Section 5 presents the empirical strategy used and Section 6 the results obtained. Finally, Section 7 concludes and describes future research plans. 2.2 Literature Review The existing literature about the incorporation of computers into standard education is still evolving and has mostly focused on the effects of these programs on test scores. Angrist and Lavy (2002) studied a program in Israel that distributed 35,000 computers to schools; they found low effects on performance. Further, using a randomized experiment Banerjee et al (2007) studied the impact of increased computer use upon students in India. In this program the treatment group employed a computer for two hours a week 4 The school year in Costa Rica goes from February to November. 11 during mathematics classes. They found positive and significant effects on math test scores, but no effects on test scores in other subjects. These results are suggestive that guided use of the equipment is necessary to improve test scores. Barrera-Osorio and Linden (2009) analyzed a program that donated computers to public schools in Colombia. They found no effects on test scores. Malamud & Pop-Eleches (2011) studied a program in Romania that provided vouchers to low-income families to facilitate the purchase of home computers. Contrary to what was expected, the authors found that the students from families that bought computers had worse performance in school. Their explanation for this outcome is that the students with home computers spent more time playing games and less time doing homework and studying. The literature on OLPC programs is new and limited, and so far has mostly focused on the effects of the laptops on performance. In the largest scale OLPC study to date, Cristia el al (2012) evaluated the program in Peru making use of a randomized experiment. The authors found no significant effects on Math and Language test scores. They did, however, find a positive effect in a progressive matrices cognition test and in a coding test. The reason given to explain these outcomes is that the lack of teacher training, education resources and connectivity did not allow the program to achieve its potential for school performance. In another study regarding the OLPC program in Peru, Beuermann et al (2012) focus on the effects generated on the students that were able to take their computer home after class. Through the use of a randomized experiment, they find that the students who where able to use the computers at home were more likely to perform home duties, but were less likely to spend time reading books. This study did not find any effects on student skills using a Windows based PC and an internet browser. Sharma (2012) found no effects on test scores when analyzing the OLPC program in Nepal. He concludes that, even though his study failed to find positive effects on school performance, utilizing a different evaluation design that includes variables concerning family members’ computer usage may show some positive effects of the program. 12 2.3 The Conectándonos Program In 2010, the Quirós Tanzi Foundation, decided to implement an OLPC program in a set of public elementary schools 5 in Costa Rica. The laptop computer provided is known as XO and costs approximately $209 per unit. This computer was designed in 2005 and it operates under free software. Since its early stages of construction the XO was oriented toward children, as shown by its waterproof and shockproof design. At the same time, the XO has all of the standard features of any other laptop computer, such as: wireless connectivity, USB ports, speakers, microphone, camera, and headphone jack. Figure A.1 in the Appendix depicts the XO laptop. In the first year of the program (2012), 1,550 laptops were distributed within 15 public primary schools in four districts of Costa Rica. In the second year of the program, another 1,500 laptops were provided; and in the upcoming years, 25,000 more laptops are going to be distributed. The program has a one-to-one scheme, each student is given a computer that is used about one hour daily during class and that can be taken home after school. A difference between this program and other OLPC programs is that the Conectándonos Program does not only provide the computers, it is a quality-oriented program in which teachers are guided and broken laptops are repaired within two weeks. For doing so, the NGO trains the teachers before the start of every school year, and a multidisciplinary team composed of educators, technicians and coordinators visits every school weekly to guarantee that there are no problems with the teaching techniques or with the equipment. The computers are collected for maintenance in December every year and given back to the returning students in the following February. In order to have a good idea of the costs of this program, five main things should be considered: the XO laptop, school infrastructure fixed costs, internet/electricity bills, teacher training and the cost of the weekly follow-ups by the multidisciplinary team. According to the NGO, the initial fixed costs, including: a laptop for each student and teacher, infrastructure and teacher trainings amounts to approximately $225 per student. In addition, the annual variable costs, which include: internet services, electricity, 5 Public schools in Costa Rica charge no tuition or any other fees. 13 technical support and pedagogic follow-up visits amount to approximately $40 per student. The majority of these variable costs come from fixing and replacing broken equipment. According to the NGO, 40% of the equipment needs at least one repair per year. The most common fixes are replacing the keyboard, screen and charger. The central question for determining the data collection and empirical strategy to follow was the school selection process for the first year of the program. In a first stage, the Costa Rican Ministry of Education gave the NGO a list with of all the public elementary schools in the country. It was decided that the schools that already had computer laboratories were to be ineligible. As a second step it was decided that very small schools (schools with only one teacher) and very big schools (schools with over 400 students) would not be considered. Finally, with the weekly follow-up visit plan in mind, it was decided that the schools that were more than a three-hour drive away from the metropolitan area would be excluded. After all of these filters were applied, the four districts that had more than three eligible schools were selected. The first district selected was Río Cuarto, which had ten eligible schools and two ineligible schools. Four schools were selected to be a part of the program in Year 1 and the other six were selected to incorporate in Year 2. The second district selected was San Isidro, which had five eligible schools and three ineligible schools. The five schools were selected to be a part of the program in Year 1. The third district was Santa Teresita, which had six eligible schools and two ineligible schools; three schools were selected for Year 1 of the program, and the remaining three were incorporated in Year 2. The fourth district was Curridabat, with four eligible schools and two ineligible schools. Three schools were selected for Year 1 of the program and the remaining school was incorporated in Year 2 6 . The first three districts mentioned are located in rural areas where the main economic activity is agriculture. The last district is located in an urban 6 Please note that due to the fact that the NGO prioritized the success of the follow-up visit scheme over the evaluation design, it was not possible to randomize the selection of the treatment schools during Year 1 of the program. The first group of treatment schools was chosen by the NGO in order to facilitate their weekly logistics. 14 area close to the capital city. Figure A.2 in the Appendix shows a map of the four districts in which the program began. 2.4 Data The data used in this chapter come from two household surveys that I conducted on the students and parents of a treatment group and a control group. The baseline survey was conducted in February 2012, and as mentioned in the previous section, the treatment group consisted of 15 schools located in four districts. In order to create an adequate control group, I selected the 10 schools (from the same four districts) that were chosen to be a part of the program in the second year. From here on I will call this group Control 10. However, since the treatment group was twice as large as Control 10, the remaining 9 untreated schools within the four districts were also selected, allowing an expanded group of 19 control schools (Control 19). In summary, this research utilizes 15 treatment schools and 19 control schools covering approximately 3,300 students, which represent the population of all the schools in the four treated districts. Table 2.1 details the location and characteristics of the treatment and control schools. The household survey consists of two independent components, a survey for the students and a survey for the parents. For the first round of surveys the student questionnaire was completed in class with teacher instruction and assistance in order to guarantee that the kids understood how to complete it. The parental questionnaire was distributed together with a contract accepting the laptop one week before the distribution. In order to get the laptop the first day of classes, it was mandatory to complete both the questionnaire and contract. This way the NGO took care of surveying the treatment group with near perfect response rates. The questionnaires for the control group were delivered and collected the last week of classes of the previous school period; this way contamination of information was avoided 7 . The student survey in the control group was applied identically as in the treatment group. However, since this group was not to receive computers, the parent 7 Since some control schools are located very close to treatment schools and the next year they would know about an OLPC program in their district, possibly affecting their responses. 15 survey was sent to the household as homework, and the schoolteacher would collect it the next day. For the control group there is a non-response rate of around 20%. To address the issue of response bias I will perform an attrition analysis in the last section of the results. Table 2.1 Baseline Characteristics: Treatment and Control Schools School Treatment Status Total Students Total Teachers Number of Classrooms Library Computer Laboratory Water & Electricity District: Rio Cuarto Río Cuarto Treatment 304 14 8 Yes No Yes IDA La Victoria Treatment 55 5 3 No No Yes Santa Isabel Treatment 119 6 5 No No Yes San Rafael Treatment 124 5 3 No No Yes La Flor Control10 38 2 1 No No Yes Carrizal Control10 40 5 2 No No Yes El Carmen Control10 50 4 3 No No Yes IDA Los Lagos Control10 64 5 3 No No Yes La Españolita Control10 48 3 2 No No Yes La Tabla Control19 98 3 2 No Yes Yes Santa Rita Control19 293 14 8 No Yes Yes José M. Herrera Control10 216 9 3 No No Yes District: San Isidro La Laguna Treatment 55 3 2 No No Yes Silvia Montero Treatment 159 10 6 No No Yes Mario Aguero Treatment 85 5 3 No No Yes Carbonal Treatment 98 7 3 No No Yes Enrique Riba Treatment 114 7 5 No No Yes Luis Sibaja Control19 237 9 6 No Yes Yes Itiquis Control19 240 9 6 Yes Yes Yes Timoleón Morera Control19 164 9 4 No Yes Yes District: Santa Teresita Carlos Luis Castro Treatment 54 8 3 No No Yes Palomo Treatment 25 3 1 No No Yes Jorge Rossi Treatment 23 2 1 No No Yes Guayabo Abajo Control10 62 5 2 No No Yes Cimarrón Control10 73 5 2 No No Yes Colonia Guayabo Control10 42 6 3 No No Yes Santa Teresita Control19 133 9 5 Yes Yes Yes El Cas Control19 110 4 2 No Yes Yes District: Curridabat Jose M. Zeledón Treatment 125 4 4 No No Yes La Lía Treatment 95 9 3 No No Yes Yerbabuena Treatment 113 9 3 No No Yes Cipreses Control10 43 2 2 No No Yes Carolina Bellelli Control19 243 14 6 Yes Yes Yes Quebrada Fierro Control19 179 8 5 No Yes Yes Treatment Group 1550 Control 10 Group 710 Control 19 Group 1700 Source: Quirós Tanzi Foundation. Note: Control 19 includes the original set of Control 10 schools plus 9 additional schools that already had computer laboratories. 16 The second round of surveys took place in February 2013. Since the laptops were gathered by the NGO at the end of the 2012 school year, and returned to the students with a new contract in the beginning of the next school year, I repeated the method of attaching the survey to the contracts. A near perfect response rate was obtained from both the original treatment group and from the 10 schools that were controls in the first round and became treatment in the second round. For the remaining part of the control group, the surveys were again sent as homework with the assistance of the principals and schoolteachers. Two clarifications regarding the second round of surveys are in order. First, 6 th graders were interviewed at the end of the 2012 school year because they would enter high school the next year. Second, some students transferred schools, thus missing students were searched for in all the schools of the same district. If these students were found, the second observation was included as if they had remained in the original school. Figure 2.1 depicts the timeline of the data collection. Figure 2.1 Timeline of the Data Collection November 2011 February 2012 November 2012 February 2013 April 2013 | | | | | -Control Schools were Surveyed. -Treatment Schools were Surveyed. -Laptops were distributed. -All 6th Graders were Surveyed. -All Schools were Surveyed. -Laptops were distributed. -5th & 6th Graders were Tested. The student questionnaire gathered: the type of occupation desired in adulthood, the favorite class, school satisfaction, hours of computer usage, uses given to the computer, and with whom the computer is used. Table 2.2 provides the baseline summary statistics from the information provided by the students. Column 1 includes all the schools, column 2 includes the treatment schools, column 3 includes the Control 10 schools, and column 4 includes the Control 19 schools. One can observe that the means of the variables are quite similar between treatment and control schools. T-tests are reported in columns 7 and 8. 17 Table 2.2 Baseline Characteristics: Information from Students Full Sample Full Sample Treatment Control 10 Control 19 10th Percent. 90th Percent. (2)=(3) p-value (2)=(4) p-value (1) (2) (3) (4) (5) (6) (7) (8) Age 9.55 9.26 9.81 9.80 7.00 12.00 0.00 0.00 (2.03) (2.07) (2.16) (1.99) Male Respondent 0.52 0.54 0.56 0.51 0.39 0.21 (0.50) (0.50) (0.50) (0.50) Wants to Stay in County 0.71 0.75 0.75 0.67 0.94 0.00 (0.45) (0.43) (0.43) (0.47) School Enjoyment 9.12 9.26 9.16 9.00 7.00 10.00 0.23 0.00 (1.63) (1.33) (1.70) (1.85) Hours of Computer Use Home 1.94 1.85 1.04 2.22 0.00 6.00 0.00 0.04 (4.20) (3.88) (3.15) (4.77) Hours of Computer Use Outside 1.06 0.60 0.61 1.41 0.00 2.00 0.90 0.00 (2.85) (1.89) (2.18) (3.14) Observations 3174 1517 616 1657 Data: Data: OLPC Costa Rica Baseline (1 st Wave). Ho: No difference in means. The parental questionnaire contained a very broad set of topics. The first subsection gathered socio-demographic information from the household, intended to help identify the family in the future, and also to collect variables that will be used as pre-program controls. These variables include: individual’s name, ID and phone number, age, gender, size of the household, number of children in the household, family income, expenses on the children, gender of the head of the household, education completion and presence of a computer and internet at home. The second part of the questionnaire contained questions about aspirations, such as: the educational objective and desired profession for the student, and their preference for the child to continue living in the same town in adulthood. The next section of the questionnaire concerns behavioral characteristics such as: time that the parents spend helping the children with their homework; and time that the student spends doing homework, home duties and outdoor activities. Finally, questions about computer use at home were asked, including: weekly computer use by the child and other members, functions for which the computer is believed to be useful, and if the parents prefer the kids to use a computer more or less. Table 2.3 presents the baseline summary statistics 18 from the information provided by the parents. The student and parental questionnaires are presented in the Appendix. Table 2.3 Baseline Characteristics: Information from Parents Full Sample Full Sample Treatment Control 10 Control 19 10th Percent. 90th Percent. (2)=(3) p-value (2)=(4) p-value (1) (2) (3) (4) (5) (6) (7) (8) Age of the Parent 36.23 36.02 36.31 36.45 27.00 47.00 0.55 0.24 (8.27) (8.39) (8.44) (8.22) Male Respondent 0.26 0.27 0.23 0.29 0.08 0.45 (0.44) (0.45) (0.42) (0.45) Number of Household Members 4.90 4.97 4.74 4.93 3.00 7.00 0.02 0.66 (1.73) (1.75) (1.63) (1.75) Number of Kids in the Household 1.90 1.99 1.87 1.86 1.00 3.00 0.07 0.02 (1.20) (1.16) (1.23) (1.22) Male Head of the Household 0.72 0.73 0.70 0.72 0.28 0.55 (0.45) (0.45) (0.46) (0.45) Monthly Income of Household 450.66 434.56 457.10 435.45 160.00 800.00 0.19 0.47 (291.342) (260.35) (338.72) (288.10) Monetary Expenditure on Student 44.80 43.64 41.46 42.32 10.00 100.00 0.62 0.65 (33.55) (33.32) (31.59) (32.99) Monetary Expenses Other Kids 76.09 73.07 68.59 74.36 20.00 160.00 0.36 0.64 (72.32) (67.47) (72.01) (73.34) Ownership of Computer at Home 0.32 0.28 0.31 0.29 0.37 0.81 (0.47) (0.45) (0.46) (0.45) Having Internet at Home 0.21 0.18 0.20 0.19 0.35 0.40 (0.40) (0.38) (0.40) (0.39) Weekly Hours on Home Duties 3.06 3.05 3.26 3.40 0.00 7.00 0.31 0.06 (4.10) (3.89) (4.43) (4.62) Weekly Hours Help Homework 5.19 5.60 5.00 4.95 1.00 10.00 0.02 0.00 (4.46) (4.47) (4.24) (4.65) Weekly Computer Use Student 2.12 2.15 1.55 1.99 0.00 7.00 0.01 0.33 (3.71) (4.05) (2.66) (3.38) Student has Ever Used Computer 0.58 0.53 0.47 0.58 0.05 0.01 (0.49) (0.50) (0.50) (0.49) Anyone else has used a Computer 0.67 0.61 0.62 0.66 0.73 0.03 (0.47) (0.49) (0.49) (0.47) Weekly Computer Use Others 4.24 4.28 3.86 3.90 0.00 12.00 0.39 0.31 (8.56) (8.76) (8.34) (8.20) Want Child to Stay in County 0.81 0.83 0.81 0.76 0.43 0.00 (0.39) (0.38) (0.39) (0.42) Weekly Hours Outdoor Activities 5.90 6.78 5.61 5.22 0.00 14.00 0.00 0.00 (6.07) (6.77) (4.93) (5.43) Weekly Hours Homework 5.74 6.14 5.29 5.55 1.00 10.00 0.00 0.00 (4.48) (3.88) (4.07) (5.12) Encourage Student Use Computer 0.92 0.96 0.89 0.92 0.00 0.00 (0.27) (0.20) (0.31) (0.27) Observations 3174 1517 616 1657 Data: Data: OLPC Costa Rica Baseline (1 st Wave). Ho: No difference in means. 19 Households in the studied districts on average have 5 members, 2 of which are children under age 12. The monthly income of the household is approximately $500. Out of that total income, about 10% is destined to the schooling expenses of the student and about 18% to the expenses for the other children. The education completed of the parents is mostly primary. Only about 30% of the households have a computer at home and 20% have Internet, as shown in Figure 2.2. On average, parents help students with homework about 5 hours per week. Only half of the students claim to have ever used a computer, but 70% of the households report at least one other family member had used one. Figure 2.3 shows how parents mostly want their children to go to college. The most desired professions are: medical doctor, engineer and teacher. All the occupational responses are shown in Figure 2.4. The students on average spend a little over 6 hours per week performing outdoor activities, and a similar amount of time doing homework. The students also claim that they use a computer around 2 hours per week at home, and 1 hour outside; while the other household members report using a computer around 4 hours per week. Finally, 90% of the parents reported a desire for their children to use a computer more time. 20 Figure 2.2 Access to Technology Figure 2.3 Educational Objective for the Child Figure 2.4 Desired Profession of the Parents for the Students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he last section of the data collection consists on test scores obtained from a subsample of treatment and control schools. There was no baseline test scores information; therefore, I was forced to perform a single difference approach 8 . I chose control 10 as the control group that would minimize the selection bias. Then, for each control 10 school I chose the treatment school from the same district that was more comparable based on observable characteristics. This way a subsample of 10 treatment schools was selected. Two tests were applied in the months of March and April of 2013. The Math Test is an adaptation taken from the World Bank Math and Reading Student learning achievement documentation, and specifically the instrument developed for the Mongolian education system 5 th graders as a part of the READ Project. The Cognition Test selected was a progressive matrices test, which previous research considers being a good indicator of the effects of computer use on general intelligence (Malamud and Pop-Eleches, 2011). This test is an adaptation of the Wechhsler Scale (WISC-R III) designed for kids under age 15, and it consists of a set of progressive matrices, which are incomplete and intended to measure the cognitive ability of the sampled students. The specific topics evaluated were: attention, observation, detail perception, concentration and order. 2.5 Empirical Strategy To assess the impact of this program, one needs to calculate the difference in the outcomes of the treated students as compared to their potential outcome having not been treated. However, since a student cannot be selected to participate and not selected to participate at the same time, we have the usual missing information problem. For this reason, I selected a control group of participants who resemble the treated students and who live in the same district, but were not selected to be a part of the program during its first year. Given that the selection of the treatment schools was not randomized, a single difference estimator between treatment and control schools would lead to biased effects of the program. The danger of bias stems from the fact that students in selected schools 8 Since treatment was not randomized, a single difference estimator is going to have selection bias. 22 may systematically differ from the students in the non-selected schools. One initial difference between the treatment and control schools that would lead to selection bias is the fact that some control schools were not selected because of having a computer laboratory. A difference in difference estimator would eliminate the selection bias and lead to consistent estimates by assuming that the time trend is the same for both groups of schools, and that the unobservable characteristics of the students are time invariant. I define Y i to be the outcome variable for student i (computer use, change in aspirations, expenditures on education, time allocation, etc.). I also define time T=0 as the baseline period and T=1 as the post-intervention period. D=1 denotes the students that belong to the treatment schools and D=0 denotes the students in the control schools. Therefore, the difference in difference average treatment effect on the treated (ATT) will be: ATT i = [E(Y 11i | D i =1, T=1) - E(Y 10i | D i =1, T=0)] – [E(Y 01i | D i =0, T=1) - E(Y 00i | D i =0, T=0)] (1) A generic regression used to estimate the difference in difference average treatment effect on the treated is: Y = ! + "D + #T + $DT + %X 0 + & (2) Where X 0 is the set of baseline control variables. The only requirement for using this structure is having repeated cross sections of data; the parameter $ captures the ATT. However, given that I was able to follow the same individuals over time, I can make use of a panel data strategy that will provide better estimates. For this I determine the outcome variables by the following structure: Y it = ! + "D i + #T + $D i T + % t X i + µ i + & it (3) Where µ i represents any unobservable fixed effects for individual i, and the effects of the baseline control variables, % t , are allowed to change in time. By subtracting the baseline 23 period from the post-intervention period at the individual level, I am able to eliminate any time-invariant fixed effects, resulting in the following equation: 'Y i = # + $D i + (X i + ) i (4) Where ( is equal to '% and ) i is equal to '& i . Since D i is the dummy variable indicating treatment, $ will indicate the average treatment effect on the treated. My preferred specification uses Control 19 as the control group because of the balanced number of observations. However, since there is the possibility that the assumption of same time trends does not hold for students in schools with a computer lab, I also estimate the regression with Control 10 as the reference group. To improve the accuracy of the estimates, I include a broad set of pre-program control variables, which include: age of the child, age squared of child, age of the parent 9 , gender of the child, gender of the parent, gender of the head of the household, educational completion of the parent, monthly income, number of members in the household, number of children in the household, having a computer at home, having internet connectivity at home, student has ever used a computer, other family member knows how to use a computer, school year, number of classrooms in the school, having a library in the school, the teacher to student ratio, a dummy variable for rural area and a dummy variable for the district. To test for heterogeneous effects of the program, I also include in equation (4) an interaction term consisting of the treatment dummy variable and a pre-program observable variable. I test if the program has different effects for different school grades, for males or females, for households that already had a computer and for households where the parent has at least secondary schooling completed. As a robustness check I also estimate equation (2) using a pooled cross section, which allows me to increase the sample size as I can now include observations that exist in only one period. All the regressions are estimated using OLS and the standard errors are clustered at the school level. 9 Age, gender and education of the parent refer to the parent who filled the survey. 24 2.6 Results 2.6.1 Computer Usage I begin by estimating the effects of being treated on weekly computer usage in Table 2.4. Students in the program use a computer at home approximately 3 additional hours. This result is consistent using either Control 10 or Control 19 as the control group, and using information reported either from the students or from the parents. At the same time, students in the program use a computer outside their home approximately 2 additional hours. I find no significant effects on computer usage by other family members. Table 2.5 indicates the specific uses that the students give to the computers. The results suggest that the program significantly increases the likelihood of the students using a computer for browsing the internet, doing homework, self-learning online, drawing and playing videogames. These results are consistent using either control group. 2.6.2 Time Allocation In this subsection I explore how the program affected the time allocation of both the student and family members. Using information from the parental questionnaire, Table 2.6 shows that the program leads to the decline of the time doing homework by about one hour. The results also indicate that the program does not seem to lead to any change in the time spent by the parents helping the student with homework nor with the time that the student does home duties. The program does seem to reduce the time that the student spends doing outdoor activities by approximately one hour. 25 Table 2.4 Effects of the OLPC Program on Weekly Computer Use Control 10 Control 19 Baseline Mean (1) (2) (3) Student Survey Weekly hours of use at home 3.116 (0.710)*** 2.705 (0.337)*** 1.94 (4.20) Weekly hours of use outside 2.581 (0.718)*** 1.769 (0.333)*** 1.06 (2.85) Parent Survey Weekly hours of use by the student 2.714 (0.820)*** 2.561 (0.344)*** 2.12 (3.71) Weekly hours of use by others -0.509 (0.606) -0.258 (0.735) 4.24 (8.56) Control Variables Yes Yes Observations 2074 2590 Notes: Panel data estimation. The variables on the left are the outcome variables. The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Column (1) uses Control 10 as the control schools; column (2) uses Control 19 as the control schools. Column (3) indicates the pre-program means and standard deviations for the full sample of treatment and control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 26 Table 2.5 Effects of the OLPC Program on Specific Computer Uses Control 10 Control 19 (1) (2) Student Survey Uses for Internet 0.279 (0.089)*** 0.179 (0.051)*** Uses for Homework 0.288 (0.089)*** 0.044 (0.061) Uses for Self-Learning 0.210 (0.093)** 0.085 (0.053)* Uses for Reading 0.334 (0.091)*** 0.162 (0.058)*** Uses for Drawing 0.417 (0.091)*** 0.214 (0.063)*** Uses for Playing Games 0.500 (0.093)*** 0.299 (0.064)*** Uses for Phone calls 0.017 (0.022) 0.007 (0.017) Control Variables Yes Yes Observations 2074 2590 Notes: Panel data estimation. The variables on the left are the outcome variables (yes=1, no=0). The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Column (1) uses Control 10 as the control schools; column (2) uses Control 19 as the control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. *Significant at 10% ** significant at 5% *** significant at 1%. 27 Table 2.6 Effects of the OLPC Program on Weekly Time Allocation Control 10 Control 19 Baseline Mean (1) (2) (3) Parent Survey Weekly hours of student doing homework -0.896 (0.656) -1.059 (0.366)*** 5.74 (4.48) Weekly hours helping student with homework -1.061 (0.610)* -0.296 (0.305) 5.19 (4.46) Weekly hours of student doing home duties 0.464 (0.857) 0.386 (0.324) 3.06 (4.10) Weekly hours student performing outdoor activities 0.548 (0.898) -0.952 (0.558)* 5.90 (6.07) Control Variables Yes Yes Observations 2074 2590 Notes: Panel data estimation. The variables on the left are the outcome variables. The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Column (1) uses Control 10 as the control schools; column (2) uses Control 19 as the control schools. Column (3) indicates the pre-program means and standard deviations for the full sample of treatment and control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 28 2.6.3 Student and Parental Aspirations In Table 2.7 I turn to examine the potential effects of the program changing the aspirations of the students and his/her parents. The results do not show evidence of a transition from desiring blue-collar occupations toward white-collar occupations. If anything, the treated students seem to desire less a white-collar job in the future. Since the white/blue collar distinction does not necessarily denote if the use of a computer is needed, I also separated the occupations into jobs that require computer skills for everyday tasks, and jobs that do not. I determined that the following occupations require computer skills: engineering, administration and management, information and communication technology, general and keyboard clerks, customer services clerks and numerical recording clerks. I also did not find positive effects using this specification. These results are most likely due to the short time that the students and parents have had the computer, and to the predominance of white-collar occupations desired at baseline. I also find an insignificant increase in the expenditures of the parents toward the student and other children in the household. 2.6.4 Test Scores Table 2.8 shows that there are no statistically significant effects on mathematics and cognition tests. For the case of the mathematics test the results show a negative yet insignificant effect on treatment status, and this result is consistent with and without the addition of control variables. On the other hand, the cognition test shows a negative yet insignificant effect when not using control variables, and the effect becomes positive yet still insignificant when adding the controls. 29 Table 2.7 Effects of the OLPC Program on Aspirations Control 10 Control 19 Baseline Mean (1) (2) (3) Student Survey Seeks white collar occupation -0.012 (0.046) -0.062 (0.029)** Parent Survey Monthly expenses on student (In US Dollars) 4.606 (3.826) 4.081 (3.174) 44.80 (33.55) Monthly expenses on other kids (In US Dollars) 12.369 (9.328) 9.278 (5.736) 76.09 (72.32) Seeks white collar occupation (for student) -0.034 (0.057) -0.049 (0.034) Control Variables Yes Yes Observations 2074 2590 Notes: Panel data estimation. The variables on the left are the outcome variables. The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Column (1) uses Control 10 as the control schools; column (2) uses Control 19 as the control schools. Column (3) indicates the pre-program means and standard deviations for the full sample of treatment and control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 30 Table 2.8 Test Scores Math Test 5 th Graders Cognition Test 6 th Graders (1) (2) (3) (4) Treatment -5.131 (7.496) -5.303 (6.636) -8.008 (5.758) 2.865 (6.667) Age 8.965 (8.875) -21.795 (18.322) Age Squared -0.304 (0.431) 0.799 (0.909) Male 1.843 (2.544) 2.620 (3.832) Education Completion Parent 0.916 (0.931) 1.096 (1.411) Male Head of Household 5.593 (4.030) -2.879 (4.182) Monthly Income -0.005 (0.006) 0.004 (0.006) Size of Household -0.540 (0.902) -0.772 (1.541) Children in the Household 0.283 (1.172) 0.983 (1.568) Computer at Home 4.211 (3.310) 1.451 (5.806) Internet at Home 2.265 (4.770) -5.361 (6.95) Has Used Computer 4.211 (3.311) 2.451 (6.695) Others Have Used Computer 2.586 (3.562) 2.419 (4.425) District Control Variables No Yes No Yes Observations 201 167 182 145 Notes: Test Score is the outcome variable. The coefficient for Treatment denotes the ATT, the standard errors are robust and have been clustered at the school level. Cross sectional data. * Significant at 10%; ** significant at 5%; *** significant at 1%. 31 2.6.5 Heterogeneous Effects In this subsection, I consider whether or not the effects of technology access vary across subgroups. In Table 2.9 the odd columns show the ATT coefficient of equation 4 with the addition of an interaction term between the treatment dummy variable and 4 different pre-program variables. The even columns indicate the coefficient of this interaction term, which denotes the existence of heterogeneous effects for each specific subgroup. Column 2 shows how the students in grades 1 through 4 use the computer less time outside of the household, or put in other words, the older students use the computer more time outside of their household. Column 4 indicates that male students use the computer more than female students at home. In column 6 it stands out that the hours of use by others is significant if the household already had a computer. This means that the other family members who already knew how to use a computer do use the laptop. I find no different changes contingent on the parent having completed secondary education. Table 2.10 shows how students in the lower grades seem less likely to use the computer for reading, while males seem to be less likely to use the computer for doing homework. I find no effects on the changes based on having a computer at home at baseline and from the parent having completed secondary education. In Table 2.11 the heterogeneous effects on time allocation are indicated. The results suggest that the students in the lower grades use the computer more for doing homework than their upper class counterparts. Its also seems that the parents of these younger students spend more time helping them with their homework. No differential effects on time allocation were found based on computer possession or education of the parent. Heterogeneous effects on aspirations are studied in Table 2.12. The results suggest that there are no different effects conditional on grade level or on gender. I do find that the household that had a computer at home before the program started increased their expenses on the student’s education and on their other children. However, they also seem to decrease their intention for the student working in a white-collar job. Similar results are obtained when considering parents with secondary schooling. 32 Table 2.9 Heterogeneous Effects of the OLPC Program on Weekly Computer Use Treatment Treatment* Grades 1-4 Treatment Treatment* Male Treatment Treatment* Computer Treatment Treatment* Educated (1) (2) (3) (4) (5) (6) (7) (8) Student Survey Weekly hours of use at home 2.927 (0.518)*** -0.458 (0.612) 1.888 (0.521)*** 1.569 (0.718)** 2.712 (0.385)*** -0.019 (0.532) 2.711 (0.395)*** -0.024 (0.649) Weekly hours of use outside 2.368 (0.435)*** -1.242 (0.445)*** 1.423 (0.385)*** 0.677 (0.458) 1.852 (0.432)*** -0.231 (0.521) 1.808 (0.367)*** -0.149 (0.323) Parent Survey Weekly hours of use by the student 2.272 (0.497)*** 0.599 (0.593) 2.848 (0.565)*** -0.760 (0.847) 2.253 (0.463)*** 0.866 (0.616) 2.627 (0.412)*** -0.255 (0.571) Weekly hours of use by others -0.325 (0.793) 0.136 (0.694) 0.201 (0.843) -0.886 (0.823) -1.133 (0.739) 2.441 (0.943)** -0.342 (0.718) 0.321 (0.845) Observations 2590 2590 2590 2590 2590 2590 2590 2590 Notes: Panel data estimation. The variables on the left are the outcome variables. The coefficients of the odd columns denote the ATT after the inclusion of the interaction term, while the coefficients of the even columns denote the heterogeneous treatment effect by subgroup. The 4 subgroups are: student belonging to grades 1 through 4, gender, household having a computer pre-treatment and parent having at least secondary education completed. The standard errors are robust and have been clustered at the school level. For all regressions the control group used was Control 19 and control variables were included. * Significant at 10% ** significant at 5% *** significant at 1%. 33 Table 2.10 Heterogeneous Effects of the OLPC Program on Specific Computer Uses Treatment Treatment* Grades 1-4 Treatment Treatment* Male Treatment Treatment* Computer Treatment Treatment* Educated (1) (2) (3) (4) (5) (6) (7) (8) Student Survey Uses for Internet 0.172 (0.052)*** 0.016 (0.049) 0.156 (0.051)*** 0.046 (0.052) 0.183 (0.058)*** -0.009 (0.066) 0.179 (0.056)*** 0.002 (0.058) Uses for Homework 0.053 (0.055) -0.017 (0.073) 0.114 (0.066)* -0.131 (0.050)** 0.052 (0.732) -0.022 (0.051) 0.055 (0.067) -0.043 (0.065) Uses for Self-Learning 0.101 (0.061)* -0.034 (0.051) 0.068 (0.060) 0.023 (0.052) 0.078 (0.055) 0.017 (0.055) 0.087 (0.056) -0.011 (0.046) Uses for Reading 0.203 (0.061)*** -0.083 (0.042)* 0.167 (0.063)** -0.011 (0.048) 0.152 (0.066)** 0.028 (0.054) 0.157 (0.060)** 0.018 (0.051) Uses for Drawing 0.242 (0.065)*** -0.060 (0.061) 0.223 (0.062)*** -0.025 (0.040) 0.200 (0.068)*** 0.038 (0.048) 0.197 (0.064)*** 0.061 (0.051) Uses for Playing Games 0.319 (0.060)*** -0.039 (0.069) 0.287 (0.073)*** 0.017 (0.054) 0.323 (0.073)*** -0.064 (0.054) 0.287 (0.066)*** 0.046 (0.067) Uses for Phone calls -0.007 (0.021) 0.029 (0.264) 0.014 (0.018) -0.010 (0.019) 0.019 (0.020) -0.031 (0.032)* 0.001 (0.017) 0.025 (0.023) Observations 2590 2590 2590 2590 2590 2590 2590 2590 Notes: Panel data estimation. The variables on the left are the outcome variables (yes=1, no=0). The coefficients of the odd columns denote the ATT after the inclusion of the interaction term, while the coefficients of the even columns denote the heterogeneous treatment effect by subgroup. The 4 subgroups are: student belonging to grades 1 through 4, gender, household having a computer pre-treatment and parent having at least secondary education completed. The standard errors are robust and have been clustered at the school level. For all regressions the control group used was Control 19 and control variables were included. * Significant at 10% ** significant at 5% *** significant at 1%. 34 Table 2.11 Heterogeneous Effects of the OLPC Program on Weekly Time Allocation Treatment Treatment* Grades 1-4 Treatment Treatment* Male Treatment Treatment* Computer Treatment Treatment* Educated (1) (2) (3) (4) (5) (6) (7) (8) Parent Survey Weekly hours of student doing homework -1.566 (0.526)*** 1.035 (0.619)* -1.122 (0.543)** 0.134 (0.639) -0.842 (0.406)** -0.614 (0.392) -1.004 (0.417)** -0.220 (0.507) Weekly hours helping student with homework -0.795 (0.388)** 1.016 (0.396)** -0.166 (0.456) -0.226 (0.684) -0.255 (0.360) -0.114 (0.348) -0.208 (0.337) -0.337 (0.448) Weekly hours of student doing home duties 0.228 (0.411) 0.327 (0.321) 0.677 (0.437) -0.576 (0.497) 0.266 (0.342) 0.340 (0.383) 0.363 (0.322) 0.089 (0.395) Weekly hours student on outdoor activities -0.938 (0.719) -0.029 (0.665) -0.984 (0.589) 0.158 (0.674) -1.073 (0.591)* 0.347 (0.393) -0.985 (0.523)* 0.128 (0.567) Observations 2590 2590 2590 2590 2590 2590 2590 2590 Notes: Panel data estimation. The variables on the left are the outcome variables. The coefficients of the odd columns denote the ATT after the inclusion of the interaction term, while the coefficients of the even columns denote the heterogeneous treatment effect by subgroup. The 4 subgroups are: student belonging to grades 1 through 4, gender, household having a computer pre- treatment and parent having at least secondary education completed. The standard errors are robust and have been clustered at the school level. For all regressions the control group used was Control 19 and control variables were included. * Significant at 10% ** significant at 5% *** significant at 1%. 35 Table 2.12 Heterogeneous Effects of the OLPC Program on Aspirations Treatment Treatment* Grades 1-4 Treatment Treatment* Male Treatment Treatment* Computer Treatment Treatment* Educated (1) (2) (3) (4) (5) (6) (7) (8) Student Survey Seeks white collar occupation -0.075 (0.036)** 0.027 (0.040) -0.042 (0.029) -0.042 (0.050) -0.071 (0.033)** 0.024 (0.038) -0.072 (0.032)** 0.035 (0.026) Parent Survey Monthly expenses on student (In US Dollars) 5.571 (3.836) -2.971 (3.894) 3.953 (3.525) 0.853 (3.713) 1.609 (3.512) 7.041 (3.752)* 2.779 (3.205) 5.598 (2.807)** Monthly expenses on other kids (In US Dollars) 10.808 (6.246)* -3.009 (6.815) 10.209 (7.598) -1.431 (10.571) 0.752 (6.797) 23.968 (5.692)*** 6.359 (6.035) 13.104 (7.572)* Seeks white collar occupation (For student) -0.063 (0.037)* 0.027 (0.042) -0.032 (0.033) -0.020 (0.050) -0.027 (0.037) -0.066 (0.023)*** -0.043 (0.035) -0.030 (0.026) Observations 2590 2590 2590 2590 2590 2590 2590 2590 Notes: Panel data estimation. The variables on the left are the outcome variables. The coefficients of the odd columns denote the ATT after the inclusion of the interaction term, while the coefficients of the even columns denote the heterogeneous treatment effect by subgroup. The 4 subgroups are: student belonging to grades 1 through 4, gender, household having a computer pre-treatment and parent having at least secondary education completed. The standard errors are robust and have been clustered at the school level. For all regressions the control group used was Control 19 and control variables were included. * Significant at 10% ** significant at 5% *** significant at 1%. 36 2.6.6 Attrition One problem identified in the data collection stage is that the attrition is not balanced across the treatment and control groups. The response rates were around 99% for the treatment group, and around 80% for the control group. For an individual in the treatment group, the database has the baseline and post-intervention observation around 92% of the time. The missing 8% mostly came from students who transferred out or into the school during the study period. In the control group, the panel gathered both observations approximately two-thirds of the time; since in each round about 20% did not respond and this 20% is not necessarily comprised of the same students across both periods. In order to determine if those students who are missing are similar across groups, I identify the students that appeared in the baseline survey, but did not appeared in the post- intervention survey. Then, I estimate a logit model on the students’ baseline observable characteristics, with a dummy variable for attrition as the dependent variable. In Table 2.13 it can be seen how treatment significantly affects attrition in a negative way, in other words, it is more likely for the students in the control group to be attritors. With the sole exceptions of age and age squared, there are no other significant effects from observable characteristics, so it would be expected that there are no big differences in the unobservables. 2.6.7 Pooled Cross Section Estimation Finally, I perform the difference in difference estimation using the dataset as repeated cross sections. Since the fixed effects cannot be eliminated at the individual level, this specification is not as precise as the panel data strategy. However, one advantage of using the repeated cross section strategy is that the number of observations will substantially rise because the observations of those who attrit as well as from the students who entered the school after the baseline period will be included. Tables 2.14 to 2.17 show how the results are very consistent to those obtained with the panel data strategy. The main differences are found in the specific computer uses, where the results still show positive changes, but now become insignificant. 37 Table 2.13 Differential Attrition (1) (2) Treatment -0.259 (0.064)*** -0.258 (0.042)*** Age -0.351 (0.085)*** Age Squared 0.020 (0.004)*** Male 0.002 (0.112) Education Completion Parent 0.011 (0.006) Male Head of Household -0.036 (0.030) Monthly Income -0.000 (0.000) Size of Household -0.002 (0.006) Children in the Household 0.005 (0.011) Computer at Home 0.024 (0.029) Internet at Home -0.044 (0.022) Has Used Computer 0.010 (0.027) Others Have Used Computer -0.034 (0.028) Number of Classrooms 0.017 (0.019) District Control Variables No Yes Observations 2990 2442 Notes: The outcome variable is a dummy for attrition. The coefficients denote the probability to attrit; the standard errors are robust and have been clustered at the school level. * Significant at 10%; ** significant at 5%; *** significant at 1%. 38 Table 2.14 Pooled Cross Section: Computer Use Reported by the Students Control 10 Control 10 Control 19 Control 19 Baseline Mean (1) (2) (3) (4) (5) Student Survey Weekly hours of use at home 1.771 (0.278)*** 2.210 (0.396)*** 2.100 (0.297)*** 2.897 (0.398)*** 1.94 (4.20) Weekly hours of use outside 1.372 (0.254)*** 1.686 (0.314)*** 1.477 (0.220)*** 1.723 (0.298)*** 1.06 (2.85) Parent Survey Weekly hours of use by the student 2.328 (0.456)*** 2.429 (0.484)*** 2.001 (0.577)*** 2.769 (0.556)*** 2.12 (3.71) Weekly hours of use by others 0.648 (1.146) 0.568 (0.916) 0.347 (1.292) 0.983 (1.270) 4.24 (8.56) Control Variables No Yes No Yes Observations 4144 4144 6258 6258 Notes: Repeated cross section estimation. The variables on the left are the outcome variables. The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Columns (1) and (2) use Control 10 as the control schools, while columns (3) and (4) use Control 19 as the control schools. Column (5) indicates the pre-program means and standard deviations for the full sample of treatment and control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 39 Table 2.15 Pooled Cross Section: Effects of the OLPC Program on Specific Uses Control 10 Control 10 Control 19 Control 19 (1) (2) (3) (4) Student Survey Uses for Internet 0.135 (0.074)* 0.093 (0.081) 0.132 (0.066)** 0.114 (0.073) Uses for Homework 0.217 (0.071)*** 0.185 (0.077)** 0.097 (0.068) 0.098 (0.072) Uses for Self-Learning 0.120 (0.071)* 0.076 (0.067) 0.096 (0.065) 0.092 (0.067) Uses for Reading 0.140 (0.067)** 0.154 (0.068)** 0.054 (0.061) 0.086 (0.066) Uses for Drawing 0.142 (0.060)** 0.123 (0.059)** 0.173 (0.070)** -0.196 (0.101)* Uses for Playing Games 0.174 (0.085)** 0.173 (0.087)** 0.204 (0.073)*** 0.238 (0.080)*** Uses for Phone calls -0.013 (0.019) -0.002 (0.012) -0.021 (0.019) -0.004 (0.016) Control Variables No Yes No Yes Observations 4144 4144 6258 6258 Notes: Repeated cross section estimation. The variables on the left are the outcome variables (yes=1, no=0). The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Columns (1) and (2) use Control 10 as the control schools, while columns (3) and (4) use Control 19 as the control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 40 Table 2.16 Pooled Cross Section: Effects of the OLPC Program on Time Allocation Control 10 Control 10 Control 19 Control 19 Baseline Mean (1) (2) (3) (4) (5) Parent Survey Weekly hours of student doing homework -0.775 (0.438)* -0.179 (0.444) -0.439 (0.337) -0.231 (0.350) 5.74 (4.48) Weekly hours helping student with homework -0.246 (0.432) 0.169 (0.552) -0.213 (0.349) -0.017 (0.407) 5.19 (4.46) Weekly hours of student doing home duties -0.220 (0.542) 0.017 (0.510) 0.051 (0.257) 0.297 (0.314) 3.06 (4.10) Weekly hours student performing outdoor activities -0.847 (0.665) -0.289 (0.688) -0.595 (0.534) -0.255 (0.609) 5.90 (6.07) Control Variables No Yes No Yes Observations 4144 4144 6258 6258 Notes: Repeated cross section estimation. The variables on the left are the outcome variables. The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Columns (1) and (2) use Control 10 as the control schools, while columns (3) and (4) use Control 19 as the control schools. Column (5) indicates the pre-program means and standard deviations for the full sample of treatment and control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 41 Table 2.17 Pooled Cross Section: Effects of the OLPC Program on Aspirations Control 10 Control 10 Control 19 Control 19 Baseline Mean (1) (2) (3) (4) Student Survey Seeks white collar occupation -0.023 (0.031) 0.009 (0.027) -0.048 (0.027)* -0.026 (0.030) Parent Survey Monthly expenses on student (In US Dollars) 10.255 (6.675) 8.458 (5.345) 1.752 (5.515) 3.036 (5.576) 44.80 (33.55) Monthly expenses on other kids (In US Dollars) 9.475 (8.656) 12.536 (7.570) 3.145 (7.142) 9.175 (6.799) 76.09 (72.32) Seeks white collar occupation (For student) -0.038 (0.040) 0.043 (0.046) -0.029 (0.019) -0.036 (0.027) Control Variables No Yes No Yes Observations 4144 4144 6258 6258 Notes: Repeated cross section estimation. The variables on the left are the outcome variables. The coefficients denote the ATT and the standard errors are robust and have been clustered at the school level. Columns (1) and (2) use Control 10 as the control schools, while columns (3) and (4) use Control 19 as the control schools. Column (5) indicates the pre-program means and standard deviations for the full sample of treatment and control schools. Control variables include: age, age squared, school year, age of the parent, gender of child, gender of respondent, education completion of parent, gender of the head of the household, monthly income, size of the household, number of kids in the household, having computer and/or internet at home before program, student and/or someone in the household knowing how to use a computer before the program, ratio of teachers to students, number of classrooms, library dummy, rural dummy and district dummy. * Significant at 10% ** significant at 5% *** significant at 1%. 42 2.7 Conclusion This chapter analyzed the effects of an OLPC program that in addition to donating laptop computers to each student in eligible primary schools also provided periodic follow-up visits to guarantee the year-around functioning and maintenance of the program. The results one year after the implementation of the program provide evidence of the treated students using a computer 5 hours more per week than their non-treated counterparts. The results also suggest that the male students use the computer more time than females, and that the older students use the computer more time outside the household than the younger students. An unexpected finding is that the computer usage of other family members did not change. When checking for heterogeneous effects I find that the only case in which other family members share the computer is when the household already had a computer before the program. These results suggest that the reason why other family members did not use the computer is because they do not know how to use it. My findings also indicate that students in the program are significantly more likely to use a computer for browsing the internet, doing their homework, learning by themselves, drawing and playing videogames. These results provide evidence that in a program with connectivity the students are able to learn how to use diverse applications, which would lead to a more technologically skilled labor force. I also find that the increase in computer use seems to lead to a decrease of the time doing homework and performing outdoor activities, reinforcing the view of parental guidance being fundamental for the laptop to be used in a productive way and for the child to stay healthy (Malamud and Pop-Eleches, 2011). A result from this chapter that is consistent with the existing OLPC literature (see Cristia et al, 2012 & Sharma, 2012) is the lack of significant effects on test scores. The reason for this result is that the program is not of guided use toward specific courses. The Conectándonos Program differs from other OLPC initiatives because it focuses on the quality of the program rather than of the quantity of computers to be distributed. This chapter shows how a program that ensures year around functioning equipment and connectivity leads to a very high and diverse usage. NGOs and governments should revise the implementation methods of OLPC programs, as my findings suggest that 43 spending on infrastructure, repairs and teaching training lead to a significantly more productive use of the laptops. In this chapter I also show that parents and siblings who did not know how to use a computer did not share the computer with the student. A policy implication to easily increase the population that benefits from this type of program would be to provide computer training sessions to the parents in the beginning of the school year, and also distributing the XO computers that come with a standard sized keyboard. Due to the high implementation costs and difficulty tracking the effects of these types of initiatives, policymakers in developing countries should plan the evaluation designs of large-scale interventions beforehand, keeping in mind that the quality of the evaluation depends on the response rates of the data collection. A timely assessment of this type of intervention would help correct the program’s flaws before compromising large monetary investments. This research was fortunate because the conditionality of receiving treatment created an incentive for the students and parents to provide reliable information that allowed the construction of a very rich panel of data. Thus, I will advocate that treatment, or even eligibility for treatment, be conditioned on the provision of the information needed for the evaluation of this type of programs. As a final point, I recognize that this chapter has the limitation that it studies the program only one year after it began. Longer-term evaluations are required for a comprehensive understanding of the many outcomes that do not exist or are impossible to capture in the short run. This chapter was not able to determine if the access to information is going to change the professional aspirations of the students and their parents; further research is needed. To end on a good note, the data collection was made in a way in which individuals can be tracked over time, so newer waves will help shed more light on the effects of the OLPC program, with labor market outcomes as the final decisive results. 44 Chapter 3 The Effects of Subsidizing Secondary Schooling: Evidence From a Conditional Cash Transfer Program 3.1 Introduction The United Nations’ Millennium Development Goals state the need for developing countries to guarantee free and universal primary education to their children. Costa Rica has successfully satisfied this provision to its citizens for many years now. However, an issue that has not yet been solved is how to prevent high school students from dropping out of the system. Since the unprivileged population has lower access to risk reduction mechanisms, they are the most vulnerable during economic downturns. When an economic crisis arises, the poorest families usually react by taking their children out of school in order to help raise income for the family, which leads to lower schooling and intergenerational poverty. It is well known that failure to complete secondary school leads to poverty, and at the same time poverty leads to underage employment in the informal sector, which in turn leads to even lower schooling. As stated in the Education for all in the Americas regional meeting from the UNESCO World Education Forum (2000): “The countries commit to give priority to policies and strategies aimed at decreasing repetition and dropout, assuring permanence, progress and success of boys and girls and of adolescents in basic education systems.” A proposed solution to combat dropouts is Conditional Cash Transfer programs (CCT). CCT programs intend to provide short-term financial relief to participating households and lead to long-term wealth redistribution through human capital. In 2005 the attendance 45 rate for high school in Costa Rica had fallen to an all-time low at 78.4%, with only 20% of the students finishing the five years of high school on time. As an attempt to alleviate this problem, in 2006 the Costa Rican Government created the Avancemos (Let’s Advance) Program. The program consists on a monthly monetary subsidy conditional on school attendance and a yearly health check for eligible adolescents. According to the Minister of Education, Leonardo Garnier, high school attendance increased from 78.4% in 2005 to 82.9% in 2009, confirming the expected short-term positive effect on attendance rates. This chapter tries to answer: What are the short-term impacts of the Avancemos program on years of school completed and hours worked? To answer this question I will empirically estimate the effects of the program 2 years after it started with the treated group and a control group created to match observable characteristics of the treated group. The data used comes from yearly household surveys, which contain information of numerous socioeconomic characteristics as well as the main variables of interest: receiving the conditional cash transfer, education completion and hours worked. The structure of this chapter is as follows: A review of part of the existing literature on CCT programs and a description of the Avancemos Program, followed by a description of the data used, the theoretical framework and the empirical strategy. Finally, the results and concluding remarks are presented. 3.2 Literature Review Dropouts from school are a very common problem in underdeveloped countries. For this reason in the last years, CCT programs have come to play a very important role in education. Many authors have acknowledged the importance of evaluating this type of intervention and have tried to measure its effects. Parker et al. (2008) point out: “The main motivation of conditional cash transfer programs is the linking of benefits to human capital investment, particularly of children. The aim is to alleviate current poverty 46 through monetary transfers as well as future poverty, by increasing human capital of children”. The most studied CCT program to this day is the Progresa/Oportunidades program in Mexico, which started under the name Progresa as a randomized experiment in 1998. It included elementary and high school students who lived in rural areas. In one of the first studies made, Skoufias, Davis and Behrman (1999) found an increase of 10% in school attendance. In another study Behrman, Sengupta & Todd (2001) found that the program increased grades completed by 0.6 years and school attendance by 19%. Schultz (2004) calculated the increase in the probability of finishing another year of schooling, conditioned on having finished the previous year. His results are that the program increased schooling by 0.81 years using a single difference approach and 0.66 years using a difference in difference approach. In a set of newer studies, Behrman et al. (2006) and Parker et al. (2008) explained how Progresa was renamed Oportunidades as it was extended to urban Mexico. With this expansion, the program became based on eligibility requirements and self-selection, as individuals were now required to apply for the program. Applicants were attended in a module installed in each town, where socioeconomic status was identified and later verified by an in situ visit. Behrman et al. (2006, 2011) provided estimates of the program’s effect on enrollment and grade completion based on a difference in difference matching approach. Parker et al. (2008) also used matching to create comparison groups. Both studies found positive significant effects on school enrollment and grade completion, which ranged from an additional 0.5 to 0.9 years of schooling depending on the duration of the intervention. Another well-known CCT program in Latin America is Brazil’s Bolsa Escola, which started in 1995 and expanded in 2001. In this program, the subsidy was given to families with children aged 6-15 who had a per capita income lower than R$90. Costa Resende & De Oliveira (2008) found that this program led to an increase in food consumption quality, diversification and quantity. Colombia also developed a CCT program named Familias en Acción. The program started in 1999 and it had two components: from ages 0-7 there was a monetary transfer to the mothers, intended to improve the nutrition of the 47 child, conditional on medical check ups, and from ages 7-17 there was a monetary transfer to parents, conditional on keeping the student(s) in school. Attanasio et al. (2005) find that this intervention led to an increase in attendance of 5.2% in urban areas and of 10.1% in rural areas. In another study, Duryea and Morrison (2004) evaluated a small scale CCT program in Costa Rica named Superémonos, which took place in the early 2000’s. Since this program was not randomized, the authors used propensity score matching to estimate its effects. They found significant increases in school attendance but no effects on performance and child labor. Not all the existing literature about CCT programs has found positive results. Ravallion & Wodon (1999) studied a program in Bangladesh in which rice was given conditioned on school attendance. They argue that this conditional transfer program led to lower leisure for the kids due to the fact that the parents forced them to stay working. In another study, Skoufias & Parker (2001) claim that CCT are not necessarily the best way to spend money in the educational system. They argue that school inputs, such as investing in school infrastructure and having better teachers, should be also considered. 3.3 The Avancemos Program In mid 2006 the Costa Rican Government created a pilot CCT program that covered 8,000 unprivileged students in urban districts. In 2007 this program expanded nationally. Each district in the country was assigned a regional quota that was determined by the district’s Index of Social Development. This way, the regions with a lower index would get more resources. The program’s coverage kept growing every year: 8,000 students in 2006, 94,621 in 2007, 136,000 in 2008, 165,749 in 2009 and 185,000 for 2010 and 2011; which is about 51.4% of the currently enrolled students in public secondary schools. The Avancemos program’s main objective is to keep students in the school system by offsetting the potential income obtained from dropping out of school and working. The monthly transfer was designed to rise with school years to offset the increasing opportunity cost of not working. For 7 th grade the monthly payment was of around $30, 48 $40 for 8 th , $50 for 9 th , $70 for 10 th and $80 for 11 th . The jump in 10 th grade is due to the fact that the dropout rate after 9 th grade was especially high. To get an idea of the magnitude of these transfers, the minimum wage at that time was of about $400 per month. Although the program doesn’t forbid having a job, at the very least it limits the time available for working, and it is expected that a student not working would have better school attendance and higher grades. It is assumed that those who didn’t get the CCT are not affected by this policy, i.e. they don’t dropout from school only because they weren’t selected, and that larger class sizes would not lead to more dropouts or lower school quality. For this program only high school students were eligible and all households were uniformly notified of the programs implementation through the national media. There is also no restriction on the number of eligible students per family as long as they haven’t repeated a grade more than once. Selection into the program consists of many steps. First, the family has to be identified by the IMAS (Mixed Institute for Social Aid), and must have filled a technical survey named FIS that contains 56 variables which are corroborated in situ regarding the socioeconomic status of the family. Using this information a vulnerability score is computed for each family. The poorest families (about a third of the country) are covered through these surveys and they are categorized into 4 groups depending on their socioeconomic status. The second step consists on a parent applying for the program with a signed contract. In the last step, the IMAS establishes a regional cutoff score under which all the students who applied are selected for the subsidy. After the students are selected, the money is deposited on a monthly basis to the mother’s bank account. 3.4 Data: The Costa Rican Household Survey The data used in this chapter comes from the Costa Rican Household Survey for Multiple Purposes (Encuesta Nacional de Hogares con Propositos Multiples). This survey consists of cross sections collected on a yearly basis, where one fourth of the households 49 interviewed are randomly replaced every year. The primary unit is the household and the sample is chosen from 826,541 households, which account for the whole country. The sample is stratified in order to have a good representation of all the regions in the country. In a first stage clusters of households are selected, and in a second stage individual households are randomly selected. The final sample consists of 10,890 households. As was mentioned earlier, the program started with a small pilot plan in 2006, but it was in 2007 when it expanded nationwide. The household survey was conducted in the months of June and July, while the school year in Costa Rica starts in February. Since the main outcome variable is school completion, students who started the program in February 2007 are going to report receiving the program when asked in June/July, however, their school completion is going to be unaltered from late 2006 (the time when they where still not in the program). For this reason the survey from 2007 is going to be used as the baseline year, as it contains current treatment status and the level of school completion before treatment began. Table 3.1 shows the summary statistics of the eligible students, in this case, students between ages 12 and 18. At baseline, 36% of the students live in urban areas, the size of the families if of about five members, out of which approximately one is a kid under age 12. Most households possess a refrigerator and a television, while about half of them have a cellular phone. Only about a forth of the households have a computer, a car and cable/satellite television. About a fifth of the households have at some point been awarded a household subsidy. In 0.71 of the cases a male is the head of the household, and the education completion of the head of the household is on average of only elementary school completed. Finally, about 15% of the eligible adolescents report working. Figure 3.1 shows how average years of schooling for adolescents between ages 12 to 20 were increasing very slowly in the first part of the decade and how after the program was installed in 2007 there was a substantial increase. At the same time, Figure 3.2 shows how after 2007 the hours worked by adolescents has decreased. Figure 3.3 shows the evolution of the reasons given by school dropouts. Not being interested in formal 50 education, can’t affording studies, and having to work are reported as the main reasons for quitting school. Table 3.2 contains information on the school outcomes for adolescents between ages 12 to 18 for the years 2000, 2003, 2005, 2007 and 2010. It can be seen how even before the program started in 2007 there was a tendency towards attending more secondary school. When analyzing the evolution of school completion by age, it can be seen how there are slow increases in years completed for all ages during this decade. However, there seems to be a stop in 2007, the year when the program started (yet still had no effects on school completion). For 2010, there is an increase in school completion for all ages. In 2010, after 3 years of the program, it can be seen how reasons for dropping out of school change. There was a decrease in the proportion of dropouts due to having to work and for not affording school. However, there was an increase in the proportion of dropouts due to lack of interest in formal education and because of having problems studying. 51 Table 3.1 Summary Statistics: Eligible Students Baseline 2007 Post-Intervention 2009 (1) (2) Years of School Completed 6.71 7.05 (2.06) (2.14) Age of the Student 14.97 15.05 (2.02) (1.94) Male Student 0.53 0.51 (0.50) (0.49) Lives in an Urban Area 0.36 0.38 (0.48) (0.48) Family Size 5.15 4.91 (1.82) (1.69) Kids Under Age 12 0.92 0.81 (1.04) (0.98) Ownership of a Cell Phone 0.56 0.75 (0.50) (0.43) Ownership of a Refrigerator 0.91 0.94 (0.29) (0.31) Ownership of a Computer at Home 0.28 0.35 (0.45) (0.43) Ownership of a Car 0.27 0.36 (0.44) (0.48) Ownership of a TV 0.94 0.95 (0.23) (0.24) Ownership of a Cable or Satellite 0.20 0.25 (0.40) (0.43) Ownership of a Internet 0.07 0.18 (0.26) (0.38) Awarded a Housing Subsidy 0.22 0.26 (0.41) (0.43) Male Head of Household 0.71 0.68 (0.46) (0.43) Schooling of Head of the Household 1.42 1.37 (0.75) (0.74) Student Has to Work 0.14 0.07 (0.35) (0.26) Student Hours Worked 5.13 2.45 (13.99) (9.84) Working Student’s Wage 4177.05 8383.29 (17734.34) (39379.2) Family Income 439349.81 455311.80 (508530.41) (560865.35) Observations 6627 5828 Data: Costa Rican Household Survey 2007 52 Figure 3.1 Average Years of Schooling 2000 to 2010 Figure 3.2 Average Hours Worked 2000 to 2010 Figure 3.3 Reasons Given by School Dropouts 53 Table 3.2 School Outcomes: Students Ages 12-18 2000 2003 2005 2007 2010 (1) (2) (3) (4) (5) Characteristics Receiving Avancemos -------- -------- -------- 0.163 0.324 Currently in Secondary School 0.69 0.71 0.75 0.79 0.80 12 Year Olds School Completion 5.34 5.50 5.84 5.71 5.94 13 Year Olds School Completion 5.62 5.83 6.09 6.05 6.14 14 Year Olds School Completion 6.32 6.25 6.68 6.67 6.77 15 Year Olds School Completion 6.60 6.74 7.17 7.11 7.27 16 Year Olds School Completion 6.98 7.26 7.58 7.53 7.80 17 Year Olds School Completion 7.27 7.49 7.63 7.81 8.09 18 Year Olds School Completion 7.31 7.49 7.64 7.75 8.32 Reason for Dropping School Has to Work 0.12 0.05 0.12 0.08 0.09 Rather Work ------ 0.08 0.05 0.07 0.06 Has to Work at Home 0.05 0.03 0.03 0.03 0.05 Can’t Afford School 0.21 0.21 0.22 0.18 0.15 Access Problems to Schools 0.07 0.05 0.05 0.07 0.06 Problems Studying 0.11 0.08 0.13 0.08 0.12 Not Interested 0.32 0.23 0.31 0.24 0.39 Pregnancy or Marriage 0.04 0.03 0.03 0.04 0.03 Disability 0.04 0.04 0.02 0.03 0.02 Observations 4156 4223 4418 4411 3904 Data: Costa Rican Household Survey 2000-2010 54 3.5 Theoretical Framework The Avancemos program was not a randomized experiment, the families that wanted to receive the treatment had to be eligible and apply for the program. Therefore, the control group consists of the students that were eligible but that did not receive treatment. Possible reasons for not being treated would be that the parents did not fill the application, a problem verifying eligibility status, or that the program was not present in their community in 2007. Since post-program differences between treated and untreated groups could come from pre-program differences, comparing means between these groups would not reveal the unbiased effects of the program. Therefore, an appropriate control group will be obtained through matching techniques. To account for the selection bias problem, a difference in difference propensity score matching (DIDPSM) strategy will be applied. Y i is defined to be the outcome variable for student i (school completion and hours worked per week). A student receiving the transfer would be Y 1i and Y 0i if not. D=1 denotes the students that belong to the treatment schools and D=0 denotes the students in the control schools. The average treatment effect on the treated (ATT) will be: ATT i = E(Y 1i - Y 0i | D i =1) = E(Y 1i | D i =1) - E(Y 0i | D i =1) (1) Since an eligible student i can not be simultaneously in the treatment and in the control group, then E(Y 0i | D i =1), which is the counterfactual, does not exist. At this point I assume that there exists a similar group of untreated, such that: E(Y 0i | D i =1) = E(Y 0i | D i =0). According to the identification hypothesis: in selection, people with the same observable characteristics have the same chance of being selected as treatment or control, therefore, (Y 0i , Y 1i ! D i | X) and E(Y 0i | X i , D i =1) = E(Y 0i | X i , D i =0). Using the observable characteristics in selection X we have: ATT i = E(Y 1i – Y 0i | D i =1, X) = E(Y 1i | D i =1, X) - E(Y 0i | D i =1, X) (2) 55 Performing the matching would be difficult due to the large set of observable variables. However, Rosenbaum and Rubin (1983) demonstrated how reducing the dimension to a univariate propensity score would be suitable. In order to yield the conditional probability of being selected for treatment, a propensity score is generated based on observable variables, such that: P(X) = P(D=1 | X), and 0<P(X)<1. Rosenbaum & Rubin’s (1983) Propensity Score Matching (PSM) technique allows constructing a control group that is matched on observable characteristics. Nowadays, this method is very popular due to its easy application and diverse advantages. According to Ravallion (2008): “In evaluating anti-poverty programs in developing countries, single-difference comparisons using PSM have the advantage that they do not require either randomization or baseline data.” To be able to identify the ATT, propensity score matching assumes that the observable characteristics X contain all the information about the potential outcome in the absence of treatment. Therefore, treated and non-treated students are comparable to the non-treated outcome, conditional on a set of pre-treatment observable variables and within a region of common support of their propensity scores. It is important to mention that the estimation of the propensity score requires the use of observable variables that are not influenced by the presence of the program. Therefore, the variables selected are pre-program baseline characteristics that identify the regional and socioeconomic status of the household. After generating a propensity score for each individual, equation (2) can be expressed as: ATT i = E(Y 1i – Y 0i | D i =1, P(X)) = E(Y 1i | D i =1, P(X)) - E(Y 0i | D i =1, P(X)) (3) There are different methods developed to perform the PSM. The ones used in this chapter are: nearest neighbor, caliper and kernel matching. 1. Nearest neighbor matching: This method assigns a weight of one to the non-treated observation with the closest score towards the treated individual, and zero to all others, i.e., only the non-participant with the value P j that is closest to P i is matched. 56 2. Caliper matching: This is a variation of nearest neighbor matching which attempts to avoid bad matches. This method defines a region of tolerance (e.g. 0.0001 to 0.00001), and then selects the non-participant that most closely matches the propensity score of the participant within the region of tolerance. 3. Kernel matching: This nonparametric matching estimator constructs the counterfactual for each program participant using a kernel-weighted average over multiple persons in the comparison group, not only the closest. This reduces the variability of the nearest neighbor estimator. The outcome of the matched control individual is calculated by: ! ˆ y i = K p(x) i " p(x) j h # $ % & ' ( j) D { = 0} * y j K p(x) i " p(x) j h # $ % & ' ( j) D { = 0} * , with weights: ! w ij = K p(x) i " p(x) j h # $ % & ' ( K p(x) i " p(x) j h # $ % & ' ( j) D { =0} * Where P(X) i and P(X) j are the propensity scores of the treatment and control groups respectively, h is the bandwidth chosen, and K is the kernel function. PSM allows for an estimation of the effects of Avancemos using only post-intervention data. The problem that arises using this method alone has to do the existence of unobservable characteristics in selection, which may lead to systematic differences between treated and untreated outcomes. Even though PSM considers differences based on the observable characteristics of the students, the existence of unobservable characteristics such as: distance to the school and public administration offices, student’s motivation, and expectations of the parents can also lead to a selection bias that PMS does not account for. In this case the conditional exogeneity assumption by Rosenbaum and Rubin (1983) is not satisfied, making the PSM method unreliable. To account for these unobservable characteristics that led to self-selection we need to incorporate a difference in difference (DID) strategy, which considers a baseline and post-intervention period, and assumes that the unobservable characteristics are time- 57 invariant. Time is defined as T=0 for the baseline period and T=1 for the post- intervention period. The DIDPSM average treatment effect on the treated is then: ATT i = [E(Y 11i | D i =1, T=1, P(X)) - E(Y 10i | D i =1, T=0, P(X))] – [E(Y 01i | D i =0, T=1, P(X)) - E(Y 00i | D i =0, T=0, P(X))] (4) Heckman, Ichimura and Todd (1997) developed a DIDPSM strategy that disentangles the selection problem. According to Todd (2008): “Difference in difference matching estimators identify treatment effects by comparing the change in outcomes for treated persons to the change in outcomes for matched, untreated persons. Difference in difference matching estimators allow for selection into the program to be based on unobserved time-invariant characteristics of individuals … difference in difference matching would solve the problem of the unobservables.” This approach is analogous to the standard DID regression estimator, but it reweights the observations according to the weighting functions implied by the matching estimators. After having panel data from the baseline and post-intervention period, the next step is to use a logistic model and a set of pre-program covariates to perform the propensity score matching. Finally, matching methods are used to construct matched non-treatment outcomes for each treated individual. Depending on the matching method used, each treated individual could be matched to one non-treated individual with the closest score, or to a group of non-treated to which weights are assigned. Nearest neighbor matching has the particularity that it constructs a control group of the same size as the treatment group, and since it only selects the closest match within a region of tolerance, then it by design satisfies the common support requirement. Using this method yields a treatment group and an adequate control group for the 2 periods. With this information a difference in difference estimator can be obtained in regression form, where the average treatment effect on the treated will be coefficient !: Y = " + #D + $T + !DT + % (5) 58 However, in a more general setting, by using more observations per treated individual, kernel matching reduces the variability of the estimator when compared to nearest neighbor matching. These matched untreated individuals are going to serve as the optimal control group. For the estimator to be consistent, the means of the covariates across groups should not be significantly different. Balancing tests are used to obtain the best propensity score specification. These tests examine whether the distribution of the covariates is independent of treatment, conditional on the propensity score. Finally, the standard errors are calculated with a bootstrap technique. 3.6 Empirical Strategy In order to estimate the effects of the program, one needs to compare the results of a treated student with the results of the same student having not been treated. However, since a student cannot be treated and non-treated at the same time, there does not exist a counterfactual. In order to make the comparisons, a matching method is necessary. The underlying motivation for the matching method is to reproduce the treatment group among the untreated, this way re-establishing the experimental conditions in a non- experimental setting. The effects of the program will be obtained by estimating the post- intervention differences across groups and time. The treatment group naturally consists of those students who report receiving the subsidy. The control group is composed of the eligible, yet non-treated students. Since selection was not randomized, the control group used in the estimations will be constructed with PSM. Participation in Avancemos is the variable to be explained. Using a logistic model, a propensity score of being selected is calculated for each treated and non-treated student. To obtain this propensity score, 24 observable characteristics reported in the Costa Rican household survey at baseline were used as covariates. These variables relate to regional, socio-economic and family characteristics and are chosen to be orthogonal to treatment, i.e. they help determine participation in the program but the program does not affect them. 59 The covariates used as the explanatory variables are: age, gender, total kids under age 12 in the household, total members of the household, ownership of the house, type of walls, type of floor, number of bedrooms, number of bathrooms, water availability, having a refrigerator, having a car, having a housing subsidy, regional characteristics (region of the country and a dummy for rural or urban zone), type of dwelling (house, apartment, etc), having electricity, a list of luxury goods (cell phone, computer, television, cable/satellite services, internet), head of the household’s education level (none, elementary, secondary, college) and family income. To ensure that the propensity score specification was appropriate, different balancing tests summarized in Todd and Smith (2005) were implemented. The first test used is the one presented in Dehejia and Wahba (2002). This procedure stratifies the treatment and control groups into strata based on the estimated propensity scores, and then tests for significant difference between the covariates within each stratum. The groups were initially split into quintiles. After conducting t-tests for the covariates of both groups within each stratum, 3 of the 24 covariates presented significant differences. Therefore, a new specification was developed. In order to satisfy the balancing property, higher order terms were added to the propensity score estimation. As noted in Dehejia and Wahba (2002), these interaction and higher order terms were added as to control better for the unbalanced covariates. The terms added were: age squared, age cubed, interactions between age and gender, age and size of the family, age and number of siblings, gender and number of siblings, and family size squared. The number of strata that ensures that the means of the covariates are not different for treated and controls in each stratum is of 8. The second balancing test conducted is the one proposed in Rosenbaum and Rubin (1985), which calculates standardized differences for each covariate between the treatment and control groups. Only 2 of the covariates presented standardized differences above the value of 20%. Table 3.3 shows the coefficients and z-statistics of the logistic regression, which denotes the effects of the covariates on the probability of selection at baseline. Most of the signs of the coefficients are as expected. In particular: Older students have a significantly higher chance of being selected and most durable goods have negative and significant 60 effects on selection. Figure 3.4 shows the distribution of the propensity scores for the treatment and control group. The region of common support includes matches between the values of 0.0098 and 0.7354; about 10% of the treatment observations will not be used, as their value is greater than the upper bound. The first specification implemented is difference in difference propensity score matching (DIDPSM), which requires panel data. This is the preferred specification as it accounts for selection bias by assuming that the unobservable characteristics are time invariant. As it was mentioned before, the household survey consists on cross sections of data that change 25% of its composition every year. Thus, within 2 waves 75% of the households surveyed are the same, within 3 waves 50% of the household surveyed are the same and so on. The households are uniquely identified within different waves, but the individuals within the household are not. To address this problem, each individual was uniquely identified, and confirmed to match across years based on observable characteristics. To construct the panel 3 steps were followed. First it was decided to use 2009 as the post- intervention period, this way a panel would be built with 50% of the observations from the cross sections. The students considered were of ages 14 to 19 in 2009, so would have 2 years of exposure to the intervention. The baseline year is still 2007, in which the considered students are of ages 12 to 17. The second step consisted on merging these 2 datasets with the uniquely identified household as the primary unit, the 50% of households that only appear in one cross section were not considered. The third step uniquely identified the individuals within the household, and these individuals had to match across both waves. The variables used for confirming that the individual was the same across both years were: same sex, age+2 in the second wave, and same or higher level of education completion in the second wave. Having panel data and the propensity score for each individual, the effects of the program were obtained through a kernel DIDPSM estimator. Using baseline information and an Epanechnikov kernel function with a bandwith of 0.06, the comparison group was created. The final step was to use a difference in difference estimator to obtain the effects of the program. Standard errors were calculated with a bootstrapping technique. 61 PSM estimators were also calculated using cross sectional data 10 . The first method used was nearest neighbor matching. Replacement was allowed, such that the propensity scores were the closest even if the untreated student had already been selected before. The next matching method implemented was caliper matching. The upper bound for matches to be kept was 0.00000001, matches with greater propensity scores differentials than this cutoff were deleted. To account for the high variability of using only one control observation for each treated student, a kernel matching method was also implemented. Finally, a difference in difference estimator was also calculated. For this specification, the control group consisted of all the untreated students who were eligible, based on the government set eligibility score. This score was computed using the same observable characteristics and weights determined by the government. The baseline comparison of years of schooling across groups is plotted in Figure 3.5, while Figure 3.6 shows the comparison in 2009. After estimating the effects for the whole sample of students, heterogeneous effects were analyzed by splitting the sample by age, gender, and region. Figure 3.4 Distribution of Propensity Score: Treatment and Control 2007 10 Post-intervention data from 2009. Density 0 .2 .4 .6 .8 Propensity Score Untreated Treated Distribution of Propensity Score 62 Table 3.3 Probability of Selection for the Program at Baseline Baseline 2007 Baseline 2007 Age 14.922 Tenure of House -0.008 (4.432)*** (0.025) Age Squared -0.084 Walls 0.041 (0.295)*** (0.024)* Age Cubed 0.015 Floor 0.028 (0.006)** (0.051) Gender 1.707 Bedrooms 0.001 (0.605)*** (0.043) Family Size 0.113 Bathrooms -0.565 (0.245) (0.110)*** Total Kids -0.669 Water -0.164 (0.413) (127) Urban -0.125 Type of House -0.103 (0.081) (0.069) Male Head of the Household 0.047 Electricity 0.035 (0.074) (0.028) Head of the Household Schooling -0.327 Cellular phone 0.143 (0.057) (0.079)* Family Income -0.000 Refrigerator -0.357 (0.000)*** (0.124)* Age*Gender -0.102 Computer -0.212 (0.039)*** (0.089)** Age*Family Size 0.025 Car 0.560 (0.015)* (0.093)*** Age*Number Kids 0.033 Television -0.496 (0.026) (0.162)*** Gender*Number Kids 0.016 Cable/Satellite Television 0.425 (0.070) (0.098)*** Family Size Squared -0.031 Internet -0.190 (0.245) (0.129)* Hours Worked Parent -0.044 Housing Subsidy 0.474 (0.010)** (0.074)*** Observations 6627 Notes: Absolute value of z statistics in parentheses * Significant at 5%; ** Significant at 1% 63 Figure 3.5 Years of Schooling by Age at Baseline Figure 3.6 Years of Schooling by Age Post-Intervention 3.7 Results 3.7.1 Difference in Difference Propensity Score Matching The preferred estimator uses a Difference in Difference Propensity Score Matching (DIDPSM) technique. This method has the advantage in that it accounts for selection bias 64 that comes from the parents having to apply for the program for their children. In order to use DIDPSM panel data is required. After having a panel with baseline and post- intervention information for treated and non-treated students, a DIDPSM kernel estimator was obtained. The results show that the program leads to a significant increase of 0.62 years of schooling and an insignificant decrease of 0.7 hours worked for the treated students. The full results for school completion are listed in Table 3.4, while the results for hours worked are listed in Table 3.6 3.7.2 Nearest Neighbor Propensity Score Matching (Cross Sectional) In this subsection I explore the results using the method of nearest neighbor propensity score matching. This specification selects for every treated student the non-treated student that is most alike based on observable characteristics. This method was done with replacement, such that every non-treated student can be selected as the nearest neighbor for each treated student, even if this non-treated student had already been previously paired with another treated student, guaranteeing that the best possible matches are found. Using this method I found a significant increase of 0.97 years of additional schooling for the treated students and an insignificant decrease of 0.9 hours worked after a two-year exposure to the program. 3.7.3 Caliper Propensity Score Matching (Cross Sectional) Caliper propensity score matching improves nearest neighbor matching as it excludes matches in which the propensity scores differentials are further from an established cut-off point. After setting a cut-off of 0.00000001, 4% of the nearest neighbor matches were excluded. Using this specification I find that the program lead to a significant increase of 0.96 years of schooling for the treated students and an insignificant decrease of 1.1 hours worked. 65 3.7.4 Kernel Propensity Score Matching (Cross Sectional) The next method used is kernel propensity score matching. This method does not rely only on one matched non-treated student for each treated, but a kernel-weighted average over multiple students with a comparable propensity score. Using this method I find a significant increase of 0.96 additional year of schooling and an insignificant decrease of 0.6 hours worked for the treated group. 3.7.5 Difference in Difference For this estimation, all the non-treated students that were eligible for the program, but who do not report receiving the transfer, are used as the control group. The results show a significant increase of 0.72 additional years of schooling and an insignificant decrease of 0.74 hours worked for the treatment group, yet these results do not account for any selection bias. 3.7.6 Heterogeneous Effects The results for heterogeneous effects use DIDPSM and are presented in Tables 3.5 and 3.7. The first subgroup analyzed are the students in the first years of secondary school, ages 12 to 15, and the second subgroup are the students in the last years of secondary school, ages 16 to 18. The estimates obtained for both groups are almost the same as for the whole sample, a significant increase of 0.6 years of school completion. The second subgroups analyzed are only male and only female students. The results suggest that male students benefit more from the program than females, as they almost complete one additional year of education, while females complete only 0.4 additional years of education. This result is expected as many males had higher drop out rates in order to start to work earlier in life. Finally, the sample is split between urban and rural areas. The effect in the urban areas is of only 0.36 additional years of school; while in the rural areas there is a much larger effect of 0.84 additional years of schooling. Regarding hours worked per 66 week, the results find only insignificant negative effects that range from 0.15 hours to 2 hours; the full results are listed in Table 3.7. Table 3.4 Effects of the Conditional Cash Transfer Program on School Completion DID PSM Nearest Neighbor PSM Caliper PSM Kernel PSM DID (1) (2) (3) (4) (5) ATT 0.623*** (0.178) 0.971*** (0.067) 0.962*** (0.051) 0.963*** (0.073) 0.725** (0.352) Observations 1996 2798 2616 2828 2995 Notes: Column (1) uses Kernel DIDPSM. Notes: The estimates are obtained using kernel DIDPSM. Table 3.5 Heterogeneous Effects of the Conditional Cash Transfer Program on School Completion Ages 12-15 Ages 16-19 Male Female Urban Rural (1) (2) (3) (4) (5) (6) ATT 0.568*** (0.198) 0.559*** (0.212) 0.984*** (0.266) 0.473*** (0.238) 0.364 (0.266) 0.842*** (0.184) Observations 1102 866 938 1044 887 1031 67 Table 3.6 Effects of the Conditional Cash Transfer Program on Hours Worked DID PSM Nearest Neighbor PSM Caliper PSM Kernel PSM DID (1) (2) (3) (4) (5) ATT -0.649 (3.173) -0.943 (2.864) -1.078 (2.882) -0.572 (1.902) -0.743 (5.138) Observations 1253 1784 1747 1782 1313 Notes: Column (1) uses Kernel DIDPSM. Notes: The estimates are obtained using kernel DIDPSM. Table 3.7 Heterogeneous Effects of the Conditional Cash Transfer Program on Hours Worked Ages 12-15 Ages 16-19 Male Female Urban Rural (1) (2) (3) (4) (5) (6) ATT 0.149 (2.963) -2.023 (3.812) -0.692 (3.113) -0.533 (3.238) -0.632 (3.56) -1.302 (3.184) Observations 662 595 601 652 463 768 68 3.8 Conclusion CCT programs were created to increase schooling quantity, specifically education completion. They are an important tool for increasing school attendance and reducing underage working, but they could also be very expensive programs. This large monetary investment by the government is the main reason why I considered it very important to quantify the short-term effects of the Avancemos program. According to my preferred specification, DIDPSM, after the first two years of the program, selected students were able to increase their years of schooling by 0.62. It seems that males are the ones taking more advantage of the program. At the same time, the effects seem to be larger in rural areas. No significant effects were found for hours worked. Further research is needed to address very important issues that I left untouched in this chapter, such as why males and students in rural areas have higher benefits from the program. An important topic left untouched in this paper is regarding returns from education to quantify the economic value of one more year of education. This would allow making a cost-benefit analysis of the long run results. It is also important to make reviews as newer data becomes available, in order to quantify the longer-term effects of the program. This CCT program has allowed selected students to significantly increase their schooling. It is also true that the program is very expensive and that these resources could have been spent differently (teacher incentives, school infrastructure, etc.). The most important conclusion is to recognize that many underprivileged adolescents have had the chance to stay in school a little longer, and that they now carry more education completed, which only time could prove as a good or bad investment. 69 Chapter 4 The Effects of Childhood Health on Education and Economic Outcomes in Adulthood 4.1 Introduction Education completion levels and economic outcomes are hard to predict. One of the main challenges of understanding adult outcomes is that there are numerous and also cumulative determinants acquired throughout the life cycle. Some of these determinants come from the environment the person is born into: socioeconomic status, education accessibility and quality, labor opportunities, disease risk, pollution, and public policies, among others. Other determinants come from individual choices such as: motivation, diet, lifestyle behaviors and preventive health. The question that this chapter tries to answer is: What is the association between health in early life and educational/economic success in adulthood? To answer this question, a dataset containing information from elder adults in Costa Rica is used. This dataset gathers self-reported answers to questions about early life health, educational completion and income, as well as anthropometric measures that would help analyze important correlations. Costa Rica is worldwide famous for its high life expectancy. According to the WHO, life expectancy in Costa Rica was of 79 years in the year 2012. At the same time, the infant mortality rate (for 1000 live births) was of only 10. One of the main reasons for these outcomes is the existence of a universal health care system, which in 2012 accounted for 28% of total government expenditure. 70 This essay examines the association between early life health and early life nutrition on education and economic outcomes. The main hypothesis is that health during childhood and adolescence plays a fundamental role in the accumulation of human capital. People that are healthier when young have the opportunity to get more education, which leads to better jobs and higher incomes when older. Particularly, this chapter studies the associations between a self-reported measure of good health as a child with education completion and income level. As noted by Smith (2009), since the variable “good health as a child” could enclose recall bias depending on how long ago childhood was for the elder interviewed; the accuracy of the response is unreliable. However, in his study he finds that the elders’ responses are indeed quite accurate. Subsequently, the same associations will be analyzed using an objective measure of early age nutrition. As suggested by Huang, Lei, Ridder, Strauss and Zhao (2012), for the elder, height to the knee is a more appropriate measure of nutrition than total height because of height shrinkage. For this reason, this chapter uses height to the knee as the objective measure of nutrition and overall health in early life. The first part of the chapter consists of a literature review of existing work on determinants of adult outcomes. The next subsection describes the data used from Costa Rica. The subsequent subsection describes the empirical strategy, followed by the results obtained. Finally, the concluding remarks are presented. 4.2 Literature Review Early life health has recently gained attention in the economic development literature. Using the Mexican Health and Aging Study (MHAS), Kohler and Soldo (2003) studied the interplay between early life events, socioeconomic conditions throughout the life course, and health outcomes at old ages. They show how accumulation of diseases differs across socioeconomic status (SES) groups, and how parental background determines the pattern of disease accumulation over the life span of 71 individuals. They also explain two ways that may link parent’s education to children’s health: father’s education (assuming that SES affect health) and mother’s education (assuming that health is determined through knowledge about health care and healthy behaviors). Their study finds that childhood poor health and nutritional deprivation are good predictors for adult functional limitations. It has been presented in the literature that low SES leads to late detection of health conditions, affecting the capacity to work and leading to lower savings (Adams, Hurd, McFadden, Merril, Ribeiro, 2004). However, just as higher SES prevents health deterioration, better health also facilitates individuals to improve their SES (Strauss and Thomas, 1998). This is known as dual causality. Strauss and Thomas (2008) study the inter-relationships of health, other human capitals and prosperity. They find that good health in early life leads to better human capital, which allows better economic outcomes. In another study, Smith (2008) found that poor childhood health has a large effect on adult SES outcomes, particularly education, family income, household wealth, individual earnings and labor supply. Different studies have analyzed the relation between health conditions of children and their schooling outcomes. Miguel and Kremer (2004) performed a randomized experiment in Kenyan schools, where students in selected schools would receive de- worming drugs. In another experiment, Maluccio et al. (2009) studied the impacts of a nutritional supplementation treatment on adult outcomes in Guatemala. Both of these randomized experiments find that an improvement in the health of the students led to better schooling outcomes, especially for females. Pitt, Rosenzweig and Hassan (2012) analyze why better health during childhood, comparatively, tends to lead to more schooling for females and higher earnings for males. In their study, the authors test if returns to schooling and brawn differ by occupation. Their main finding is that men have a comparative advantage in brawn, for this reason they tend to pick occupations with lower returns to skill and more returns to physical capacity. This paper concludes that nutritional investments lead to more brawn for men, which increases their potential wages and opportunity cost of school attendance, leading to higher earnings but lower school completion. 72 This growing attention to the health of the elderly has generated a new wave of health research in Costa Rica. According to Rosero-Bixby, Dow and Laclé (2004), one particularity of Costa Rica is that life expectancy is greater than in the USA, while per capita expenditure in health is only one tenth. In this study the authors find insignificant effects of SES on mortality. In a more recent study, Rosero-Bixby and Dow (2009) try to determine how SES affects health in Costa Rica. The authors find mixed results as risky behaviors such as smoking and lack of exercise are common among low SES, but high calorie diets are prevalent among high SES. The study that this chapter follows more closely is the one by Smith, Shen, Strauss, Zhe and Zhao’s (2012). The authors explore the effects of childhood health on adult health and socioeconomic status using data from the China Health and Retirement Longitudinal Study (CHARLS) pilot, which was conducted in 2008 in the provinces of Zhejiang and Gansu in China. The authors measure childhood health through a retrospective self-evaluation question using a standard five-point scale of general health before age 16; and they use adult height to approximate early life nutrition. This study finds that good health as a child is associated with a 15 percent increase in per- capita expenses, and also that adult height is strongly related to years of schooling completed. In a very recent study Huang, Lei, Ridder, Strauss and Zhao (2012) argue that height shrinkage among the elder contaminates the associations between height and early age nutrition. Their study shows that height shrinkage is negatively correlated with early health conditions and socioeconomic status, so using adult height as a measure of early age nutrition could be problematic. 4.3 Data Recently, a growing number of countries are engaging in data collection of the elder populations, providing standardized international datasets on aging, embedded around the Health and Retirement Study (HRS) in the United States. The HRS has been a very important contribution to understanding old adults health. This study is a 73 multipurpose, longitudinal household survey, representing the United States’ population aged 50 and older. It includes information on physical, emotional and cognitive health characteristics; self-reported data on a range of chronic health conditions, socioeconomic status, retirement, demographic characteristics, family transfers and networks (Lee, 2010). Some of the surveys modeled after the HRS are: the Mexican Health and Aging Study (MHAS), the English Longitudinal Study of Ageing (ELSA), the Survey of Health, Ageing, and Retirement in Europe (SHARE), the Korean Longitudinal Study of Ageing (KLOSA), the Japanese Study on Aging and Retirement (J-STAR), the Chinese Health and Retirement Longitudinal Study (CHARLS), the Irish Longitudinal Study on Ageing (TILDA), the Longitudinal Ageing Study in India (LASI) and finally, the Costa Rican Longevity and Healthy Aging Study (CRELES). The data for this chapter comes from the first wave of CRELES, which is a longitudinal study of a nationally representative sample of 2827 adults born in 1945 or before (ages 60 and over during the first interview) with oversampling of the older. The first wave of surveys was conducted from November 2004 through August 2006. The initial sample size was obtained from a two-step procedure. From the 2000 census, 9600 individuals were randomly selected with stratification by 5-year age groups. Sampling fractions ranged from 1.1% among those born in 1941-1945 to 100% for those born before 1905. Next, a sub-sample of 60 health regions (out of 102 for the whole country) was selected. This included 5300 individuals. The non-response rates were the following: 19% deceased by the contact date, 18% not found on the field, 2% moved out, 4% rejected the interview. The data and samples were collected from the selected households. The participants responded a 90-minute questionnaire, and afterwards the interviewer took anthropometric measures. 25 percent of the interviews required help from a family member to answer the questions. The survey collected information on early age and current health, cognitive functioning tests and socioeconomic outcomes. The summary statistics for selected variables are presented in Table 4.1. Figure 4.1 shows the average self-reported childhood health at baseline. 74 Table 4.1 Summary Statistics All Male Female (1) (2) (3) Variable Childhood Health 2.204 (1.059) 2.203 (1.008) 2.203 (1.102) Height to Knee 48.769 (3.369) 51.014 (2.655) 46.820 (2.629) Education Completion 3.205 (2.723) 3.135 (2.785) 3.266 (2.668) Income 156.608 (298.639) 197.210 (339.231) 121.419 (253.268) Foreign Born 0.055 (0.228) 0.069 (0.254) 0.043 (0.203) Observations 2548 1183 1365 Data: CRELES Round 1 (2005) Figure 4.1 Self-Reported Childhood Health Data : CRELES Round 1 (2005) 75 4.4 Empirical Strategy The explanatory variables are self-reported childhood health and height to the knee, as they represent early life health and nutrition proxies respectively. The retrospective early life health question comes from the section Childhood Conditions. The specific question asked is: “How was your health for the majority of your childhood and adolescence?” The possible answers are: Excellent, Very Good, Good and Poor. Childhood health is defined as good for those who answered excellent or very good. Adult height to the knee is obtained from the Anthropometry and Mobility Questionnaire and it is measured in centimeters. The dependent variables are years of education completed and log income. Education completion comes from the section Personal Information, and is determined depending on the last year completed; being 0 no school completion at all, and 11 for secondary school completed. Log income comes from Section H, Employment and Income. Income is revealed through the question “During the last year, what has been your total monthly income?” The estimates are obtained using ordinary least squares. Every regression includes a year of birth cohort dummy variable aiming to capture the social environment, disease exposure, public health and education policies. A variable is included indicating if the interviewee is foreign born to control for different policies and infrastructure across countries. These results are presented in the odd columns of Table 4.2. A dummy variable is also included for each county to control for public health infrastructure, environment, pollution, and weather, among others. These results are presented in the even columns of Table 4.2. All the results are elaborated for the whole sample, and then split for male and female. In all models, I include both child health measures since they are typically only weakly correlated, so that the estimated effects are not sensitive to their simultaneous inclusion. 76 4.5 Results The results are presented in table 4.2. Education completion is positively and significantly correlated with early life health, for both male and female, but the estimates are much higher for female. Height to the knee has positive effects on education completion, but this effect is significant only for female. When testing for the significance of this difference across genders, the results show that the difference in height to the knee is significant at 10%. Income is also positively correlated with early life health, and this result is significant for both male and female. When testing for the significance of this difference across genders, the results show that this difference in self-reported childhood health is significant at 5%. Height to the knee also has positive and significant correlations with income, for both male and female. Table 4.2 Estimates of Measures of Childhood Health on Economic Outcomes All Male Female (1) (2) (1) (2) (1) (2) Education Completion Good Health Child 0.492*** 0.458*** 0.350** 0.275* 0.607*** 0.559*** (0.110) (0.108) (0.164) (0.166) (0.147) (0.147) Height to Knee 0.024 0.019 0.013 0.009 0.099*** 0.096*** (0.016) (0.016) (0.031) (0.320) (0.028) (0.028) Income Good Health Child 41.260*** 36.593*** 64.137*** 53.658*** 23.843* 18.017 (11.979) (12.014) (19.842) (20.218) (14.154) (14.345) Height to Knee 11.730*** 12.253*** 11.591*** 14.031*** 5.333** 5.748** (1.781) (1.802) (3.750) (3.903) (2.702) (2.801) Observations 2548 2548 1183 1183 1365 1365 Notes: Data source is CRELES Round 1. Column (1) includes age cohort and foreign born dummies. Column (2) includes age cohort, foreign born and district of birth dummies. Standard errors in parenthesis. * Significant at 10% ** significant at 5% *** significant at 1%. 77 4.6 Conclusions The motivation behind this chapter was to empirically test if early life health matters in adulthood. This study provides empirical evidence that show the importance of childhood health towards education completion and economic outcomes. The results suggest that childhood health is highly associated with education/economic outcomes, for both men and women in Costa Rica. As has been presented in the literature, healthier kids are most likely to achieve higher education completion, especially females. Also consistent with the existing literature, better nutrition seems to lead males to occupations that pay higher wages. A policy implication derived from this chapter is that governments should pay primordial attention to the nutrition and health care of children, as this will have significant repercussions in their lifecycle. Further research would help explain why height to the knee is associated with the education completion of females but not of males and how early life health can affect other determinants such as adult health and lifestyle behaviors. 78 Chapter 5 Conclusion Education is fundamental for development. The capacities of a country to guarantee a highly educated and productive labor force will not only improve the individual wellbeing of its citizens, but the collective wellbeing as well. For this reason developing countries have been implementing programs aimed to facilitate school completion and improve school quality. This is what motivated this dissertation to evaluate two targeted programs that have become very relevant in recent years: conditional cash transfers and technological literacy initiatives. After having studied interventions aimed to improve school quantity and quality, there is a third fundamental element that was also considered, health. Healthier students are able to stay in school longer and perform better. This dissertation tries to contribute to the literature on education in the developing world by exploring three very diverse standpoints that affect human capital accumulation: subsidies to the households, laptops to the students, and the health of the students. To accomplish this objective, three different data sets were used: a new dataset at the household level for the OLPC program, a national household survey for the CCT, and a representative survey for the elder with retrospective questions about education and health. In a summary of the results, from the case of Costa Rica we learned that the One Laptop per Child and Conditional Cash Transfer programs have been successful in accomplishing their objectives of having a positive impact in education. In Chapter 2 it was shown that OLPC leads to a significant usage of computers and the learning of very important skills through the use of different applications. This confirms the importance of computer access in low-income communities in order to promote a more technologically skilled labor force. In Chapter 3 it was found that the CCT program keeps high school students in school about 0.62 years longer, which will 79 allow these students to be more productive and therefore have higher wages in adulthood. In Chapter 4 a positive association between health in childhood and schooling outcomes was found, which underlines the importance of investing in the health of children. Education in developing countries can be greatly improved following three steps. The first one is the identification of the problems that lead to bad school completion rates and performance. For example, it is very different to have low attendance rates because of hard accessibility into the school system, as opposed to having low attendance because of students starting to work at young ages. A solution to the first problem would consist on improving the infrastructure and transportation, whereas the solution to the second problem could be a conditional cash transfer program to offset the wages from working. These solutions are precisely the second step, the implementation of a targeted program designed to solve the well-identified problem. The third step consists of a timely and well-developed evaluation of the program, which would allow for implementation corrections and a cost-benefit analysis. The 3 chapters in this dissertation provide empirical evidence of how two education programs and health have positive and significant effects on education. I conclude that governments should pay special attention to programs aimed at improving the technological proficiency of students, while facilitating a system in which students have the right incentives to complete their education. By improving the quality and quantity of the education received, and at the same time guaranteeing that the children are in good health, human capital accumulation will grow rapidly. 80 Bibliography Adams, Peter, Hurd, Michael D., McFadden, Daniel L., Merrill, Angela and Ribeiro, Tiago, (2003), Healthy, wealthy, and wise? Tests for direct causal paths between health and socioeconomic status, Journal of Econometrics, 112, issue 1, p. 3-56. Angrist, J. and Lavy, V. (2002). “New Evidence on Classroom Computers and Pupil Learning.” The Economic Journal, 112: 735-765. Attanasio, O.; Battistin, E.; Fitzsimons, E.; Vera-Hernandez, M.; (2005) How effective are conditional cash transfers? Evidence from Colombia. (IFS Briefing Notes BN54 ). Institute for Fiscal Studies: London, UK. Banerjee, A., Cole S., Duflo, E. and Linden, L. (2007). “Remedying Education: Evidence from Two Randomized Experiments in India.” Quarterly Journal of Economics, 122.3, 1235-1264. Barrera-Osorio, F. and Linden L. (2009). “The use and Misuse of Computers in Education: Evidence from a Randomized Experiment in Colombia”. Policy Research Working Paper 4836, The World Bank. Barrera-Osorio, F., Bertrand, M., Linden L. and Perez-Calle, F. (2011). “Improving the Design of Conditional Transfer Programs: Evidence from a Randomized Education Experiment in Colombia. American Economic Journal: Applied Economics 3, no. 2: 167-195. Behrman, J. (1996). "The Impact of Health and Nutrition on Education." World Bank Research Observer 11, no. 1, 23-37. Behrman, Gallardo-Garcia, Parker, Todd and Velez-Granjalez (2006). “How conditional cash transfers impact schooling and working behaviors of children and youth in urban Mexico.” Mimeo University of Pennsilvania. Behrman, J., Parker, S., and Todd, P. (2005) “Long-Term Impacts of the Oportunidades Conditional Cash Transfer Program on Rural Youth in Mexico.” N 122 Ibero-America Institute for Economic Research, 2005. Behrman, Parker and Todd (2011). “Do conditional cash transfers for schooling generate lasting benefits”. The Journal of Human Resources 46.1 Behrman, Sengupta and Todd (2001). “Progressing through Progresa: An impact assessment of a school subsidy experiment.” Economic Development and Cultural Change, 2005, vol. 54, issue 1, pages 237-75. Beuermann, D., Cristia, J., Cruz-Aguayo Y., Cueto, S., Malamud, O. (2012). “Home Computers and Child Outcomes: Short-Term Impacts from a Randomized Experiment in Peru”. Inter-American Development Bank Working Paper Series No. IDB-WP-382. Chapman, D. W. and Mahlck, L.O. (2004). “Adopting Technology for School Improvement: A Global Perspective”. International Institute for Educational Planning. Cristia, J., Ibarraran, P., Cueto, S., Santiago, A., Severin, E. (2012). “Technology and Child Development: Evidence from the One Laptop per Child Program”. Inter-American Development Bank Working Paper Paper Series No. IDB-WP-304. Contoyannis P. & Jones A. (2001). "Socioeconomic Status, Health and Lifestyle," Discussion Papers 01/19, Department of Economics, University of York. 81 Costa and De Oliveira (2008). “Avaliando Resultados de um Programa de Transferencia de Renda: O impacto de Bolsa-Escola sobre os Gastos das Familias Brasileiras”. Currie, Janet and Madrian, Brigitte C., (1998), Health, Health Insurance and the Labor Market, JCPR Working Papers, Northwestern University/University of Chicago Joint Center for Poverty Research. Cutler, David M. and Lleras-Muney, Adriana, Education and Health: Evaluating Theories and Evidence (2006). NBER Working Paper Series, Vol. w12352. Deaton (1997). “The Analysis of Household Surveys: A Microeconomic Approach to Development Policy.” The Johns Hopkins University Press. Dehejia, R., and Wahba, S. (2002). "Propensity score-matching methods for nonexperimental causal studies." Review of Economics and statistics 84, no. 1: 151-161. Duflo (2001). “Schooling and Labor Market Consequences of School Construction in Indonesia: Evidence from an Unusual Policy Experiment.” American Economic Review 91 (4), 795-813. Duflo, E., Glennerster, R, Kremer, M. (2007). “Using Randomization in Development Economics Research: A Toolkit.” In T. Paul Schultz and John A. Strauss, eds., Vol. 4 of Handbook of Development Economics, Elsevier, 2007, pp. 3895-3962. Duryea and Morrison (2004). “The Effect of Conditional Transfers on School Performance and Child Labor: Evidence from and ex-post Impact Evaluation in Costa Rica.” Inter-American Development Bank, Research Department Working Papers; 505. Fernández Rojas, X.; Méndez Chacón, E. (2007). Estilos de vida y factores de riesgo para la salud de las personas adultas mayores del proyecto CRELES- Costa Rica 2004-2006. Número especial: Proyecto CRELES - Costa Rica: Estudio de Longevidad y Envejecimiento Saludable. Revista Población y Salud en Mesoamérica. Volumen 5, número 1, Julio - Diciembre, 2007. Garnier (2008). “Estadisticas Generales MEP 2008.” Gersten, O.; Dow, W.; Rosero-Bixby, L. (2010). Stressors over the life course and neuroendocrine system dysregulation in Costa Rica. Journal of Aging and Health, 28 de mayo. Glewwe, P. and Kremer, M. (2006). “Schools, Teachers and Educational Outcomes in Developing Countries,” In Handbook of the Economics of Education, edited by Eric A. Hanushek and Finis Welch. Amsterdam: North Holland, 943-1017. Glewwe, P., Jacoby, H., and King, E. (2001). "Early childhood nutrition and academic achievement: a longitudinal analysis." Journal of Public Economics 81, no. 3: 345-368. Glewwe, P. and Miguel, E. (2008). “The Impact of Child Health and Nutrition on Education in Less Developed Countries.” Handbook of Development Economics, Volume 4, Chapter 56. Hanushek, E. (1997). “Assessing the Effects of School Resources on Student Performance: An Update”. Educational evaluation and policy analysis 19.2 (1997): 141-164. Heckman, Ichimura and Todd (1998). “Matching as an Econometric Evaluation Estimator.” The Review of Economic Studies. Vol. 65, No. 2 (Apr., 1998), pp. 261-294 Hsieh, C. and Urquiola, M. (2006). "The effects of generalized school choice on achievement and stratification: Evidence from Chile's voucher program." Journal of public Economics 90.8: 1477-1503. Huang, W.; Lei, X.; Ridder, G.; Strauss, J.; and Zhao, Y. (2012). “Height, Height Shrinkage, Health at Older Ages and SES: Evidence from China”. American Economic Journal: Applied Economics, Volume 5, Number 2, April 2013 , pp. 86-121(36) 82 Instituto Nacional de Estadistica y Censos (INEC). “Encuesta Nacional de Hogares con Propositos Multiples 1999-2010.” Jensen, R. (2007). “The Digital Provide: Information (Technology), Market Performance and Welfare in the South Indian Fisheries Sector,” Quarterly Journal of Economics, 122(3), p. 879 ! 924. Jensen, R. (2010). “The (Perceived) Returns to Education and the Demand for Schooling," Quarterly Journal of Economics, 125(2), p. 515-548. Kohler and Soldo (2003). Early Life Events and Health Outcomes in Late Life in Developing Countries—Evidence from the Mexican Health and Aging Study (MHAS). University of Pennsilvanya, Population Studies Center 2003; pp1-8. Lee, Jinkook (2010). Data Sets on Pensions and Health: Data Collection and Sharing for Policy Design. Santa Monica, CA: RAND Corporation, 2010. Lonsdale, M. and Nugroho, D. (2010) “Evaluation of OLPC Programs Globally: A Literature Review” Australian Council for Educational Research. Malamud, O. and Pop-Eleches, C. (2011). “Home Computer Use and the Development of Human Capital Outcomes.” Quarterly Journal of Economics 126: 987-1027 Maluccio, J., Hoddinott, J., Behrman, J., Martorell, R., Quisumbing, A., and Stein, A. (2009). "The impact of improving nutrition during early childhood on education among guatemalan adults*." The Economic Journal 119, no. 537: 734-763. Miguel, E., and Kremer, M. (2004). "Worms: identifying impacts on education and health in the presence of treatment externalities." Econometrica 72, no. 1: 159-217. Ministry of Education Culture and Science, Education Evaluation Center of Mongolia and World Bank. (2008). “Mongolian National Assessment of Primary Education Mathematics and Reading.” Rural Education and Development Project. Ulaanbaatar City. Mincer, J. (1974). “Schooling, Experience and Earnings”. Columbia University Press. 41-63. Molina & Fallas (2009). “Transferencias monetarias condicionadas en Costa Rica: el caso del programa Avancemos.” Muralidharan, K., Das, J., Dercon, S., Habyarimana, J., Krishnan, P. and Sundararaman, V. (2012). “School Inputs, Household Substitution, and Test Scores.” American Economics Journal: Applied Economics. Nugroho, D. Londsale, M. 2009. “Evaluation of OLPC Programs Globally: A Literature Review.” Mimeograph. Australian Council of Educational Research. Parker, S., Rubalcava, L. and Teruel, G. (2007) "Evaluating conditional schooling and health programs." Handbook of Development Economics 4: 3963-4035. Psacharopoulos, George. "Returns to education: a further international update and implications." Journal of human resources 20.4 (1985). Pitt, Mark M., Mark R. Rosenzweig, and Mohammad Nazmul Hassan (2012). "Human capital investment and the gender division of labor in a brawn-based economy." The American Economic Review 102, no. 7: 3531-3560. 83 Ravallion and Wodon (1999). “Does Child Labor Dispace Schooling? Evidence on Behavioral Responses to an Enrollment Subsidy.” World Bank Policy Research Working Paper 2116. Rao, D.B. (2003). “World Education Forum”. Discover Publishing House. Page 203. Rosenbaum, P. R. and Rubin, D. B. (1983). “The central role of the propensity score in observational studies for causal effects.” Biometrika 70 41--55. Rosero-Bixby, Dow and Laclé (2004). “Insurance and Other Socioeconomic Determinants of Elderly Longevity in a Costa Rican Panel.” Journal of Biosocial Science, Volume 37. Rosero-Bixby, L.; Dow, W. (2009). Surprising SES gradients in mortality, health, and biomarkers in a Latin American population of adults. The Journals of Gerontology. Social Sciences, 10.1093/geronb/gbn004. Ross, C., and Mirowsky, J. (1999). "Refining the association between education and health: The effects of quantity, credential, and selectivity." Demography 36, no. 4: 445-460. Quaglia, R., and Cobb, C. (1996). "Toward a theory of student aspirations." Journal of Research in Rural Education 12.3: 127-132. Schultz, P. (2004). "School subsidies for the poor: evaluating the Mexican Progresa poverty program." Journal of development Economics 74.1: 199-250. Sharma, U. (2012). “Can Computers Increase Human Capital in Developing Countries? An Evaluation of Nepal’s One Laptop per Child Program.” Working Paper. Skoufias, Davis and Behrman (1999). “An evaluation of the selectiion of beneficiary households in the education, health, and nutrition program (PROGRESA) of Mexico.” IFPRI. Skoufias and Parker (2001). :Conditional Cash Transfers and Their Impacts on Child Labor and Schooling: Evidence from the Progresa Program in Mexico.” IFPRI Smith, J., and Todd, P. (2005)"Does matching overcome LaLonde's critique of nonexperimental estimators?." Journal of econometrics 125, no. 1: 305-353. Smith, James P. (1999). "Healthy Bodies and Thick Wallets: The Dual Relation between Health and Economic Status," Journal of Economic Perspectives, American Economic Association, vol. 13(2), pages 145-166, Spring. Smith, James P. (2009). “Reconstructing Childhood Health Stories.” Demography 46(2):387-403. Smith, James P., Shen, Yan, Strauss, John, Yang, Zhe and Zhao, Yaohui. (2012). “The Effects of Childhood Health on Adult Health and SES in China”. Economic Development and Cultural Change. Oct 1, 2012; 61(1): 127–156 Strauss, John, Lei, Xiaoyan, Park, Albert Francis, Shen, Yan, Smith, James P., Yang, Zhe and Zhao, Yaohui. (2010). “Health Outcomes and Socio-Economic Status Among the Elderly in China: Evidence from the Charls Pilot”. Journal of population ageing 3.3-4): 111-142. Strauss, John & Thomas, Duncan (1998). “Health, Nutrition, and Economic Development”, Journal of Economic Literature, Vol. 36, No. 2 (Jun., 1998), pp. 766-817 Strauss, John & Thomas, Duncan (2007). "Health over the Life Course," Handbook of Development Economics 4, Elsevier. 3375-3474. Teo, T. and Noyes, J. (2008). “Development and Validation of a Computer Measure for Young Students (CAMYS).” Computers in Human Behavior 24 (2008) 2659-2667. 84 Thomas, D., Witoelar, F., Frankenberg, E., Sikoki, B., Strauss, J., Sumantri, C. and Suriastini, W. (2012) “Cutting the costs of attrition: Results from the Indonesia Family Life Survey.” Journal of Development Economics, 98, issue 1, p. 108-123 Todd, Petra E. (2007) "Evaluating social programs with endogenous program placement and selection of the treated." Handbook of development economics 4 (2007): 3847-3894. UNDP (United Nations Development Programme) (2011). "Human Development Report 2011. Sustainability and equity: A better future for all." Villatoro (2005). “Programas de transferencias monetarias condicionadas: Experiencias en America Latina.” Revista de la Cepal 86. The World Bank Group (2012). “ICT for Greater Development Impact: Sector Strategy”. Available from: http://siteresources.worldbank.org/EXTINFORMATIONANDCOMMUNICATIONANDTECHNOLOG IES/Resources/WBG_ICT_Strategy-2012.pdf World Health Organization. “World Health Statistics 2012”. Country Statistics, Costa Rica. Available from: http://www.who.int/gho/publications/world_health_statistics/2012/en/ 85 Appendix Figure A.1 The XO Laptop Source: http://laptop.org/en/laptop/index.shtml Figure A.2 Map of the 4 Treatment Districts 86 Figure A.3 Parental Questionnaire ! ! ! "#$%&'($)#$#*!+,#-,./!0!12,3%4!5#,!+.,%$'*! ! ! 1&6##78!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!9.'%8! :;!<)%$'=5=&.'=#$!#5!'6%!+.,%$'!>!?%*@#$*=A7%! ! "#! $%&'!()!*+'!,*-.'/*!!0000000000000000000000000000000000000000000000000000000000000000000000! ! "1! 2(-3!$%&'!!!0000000000000000000000000000000000000000000000000000000000000000000000000000! ! "4! 2(-3!56!$-&7'3!!!!8008008008008008008008008008! ! "9! ":'!!8008008! ! ";! ,'<!!!!!!!!!=!!!!!!#!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!">! ?+(/'!$-&7'3!!8008008008008008008008008! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!@!!!!!!!1! ! ! ! ! ! ! 87 !"#$%&'%()*%+,-./'%,*-3'%1# ! "#! $%&'()*+!,)(-!(-%!.(/0%+(!!!!!!!!!!!!!!!!!!!!!!!!!! ! ! 1'(-%2!!!!!!!!!!!!!!#!!!! ! ! ! ! ! ! ! ! ! 3*(-%2!!!!!!!!!!!4! ! ! ! ! ! ! ! ! ! 5(-%2! !!!!!!!!!!6! ! "4! 7*/2!80/9'()*+!:*;<&%()*+!! ! ! ! =2);'2>!!!!!!!!!!!#! ! ! ! ! ! ! ! ! ! ?)@-!.9-**&!!!!4! ! ! ! ! ! ! ! ! ! A+)B%2C)(>!!!!!!6! ! ! ! ! ! ! ! ! ! D%9-+)9'&!!!!!!!!E! ! ! ! ! ! ! ! ! ! F*+%!!!!!!!!!!!!!!!!!G! ! "6! F/;H%2!*I!?*/C%-*&0!3%;H%2C!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!JKKJKKJ! ! "E! F/;H%2!*I!L)0C!MA+0%2!N@%!#4O!)+!(-%!?*/C%-*&0!!!!!!!!JKKJKKJ! ! "G! D-%!?%'0!*I!(-%!?*/C%-*&0!)C!!!!!!!!!!!!!!!!!!!!!!!3'&%! #! ! ! ! ! ! !!!!!!!!!!!!!!! !!!!!!!!!!!!!!1%;'&%! 4! ! "P!!!!!!!!Q-'(!)C!(-%!3*+(-&>!R+9*;%!)+!(-%!?*/C%-*&0S!!!!!!JKKJKKJKKJKKJKKJKKJKKJ! ! "T!!!!!!!!?*,!3/9-!3*+%>!0*!>*/!3*+(-&>!R+B%C(!)+!(-%!80/9'()*+!*I!>*/2!:-)&0S!!!!!!JKKJKKJKKJKKJKKJKKJKKJ! ! "U!!!!!!!!?*,!3/9-!3*+%>!0*!>*/!3*+(-&>!.<%+0!*+!>*/2!*(-%2!:-)&02%+S!!!!!!JKKJKKJKKJKKJKKJKKJKKJ! ! "V! W*%C!(-%!?*/C%-*&0!-'C!'!:*;</(%2S!!!!!!!!!!!!!!!!!7%C! !!#! ! ! ! ! ! !!!!!!!! !!!!!!!! !!!!!!!F*!!!!!4! ! "#X! W*%C!(-%!?*/C%-*&0!-'C!R+(%2+%(S!! !!7%C!!!!!#! ! ! ! ! ! ! !!!!!!!!!!!!!!!!!!!F*!!!!!!4! 88 !"#$%&'()*+,'%#&'(#+-.#/('0.1+# ! "#! $%%&'(!)*+,-!./0.!./%!"/1'2!34%52-!*5!)*+-%!6+.1%-!!!!!!!!!!!!!!!!!!!!!!!7887887! ! "9!!!!!!!!$%%&'(!)*+,-!./0.!./%!:0,%5.-!/%'4!./%!"/1'2!;1./!)*<%;*,&!!!!!!!!!!!!!7887887! ! "=!!!!!!!)0-!./%!"/1'2!%>%,!+-%2!0!?*<4+.%,!15!./%!40-.@!!!!!!!!!!!!!!!!!!!!!!!!!A%-!!!!!!!!!#!! ! ! ! ! ! ! ! ! !!!!!!!!!!B*!!!!!!!!!9! !! !!!"C!$%%&'(!"*<4+.%,!+-0D%!E(!./%!"/1'2@!!!!!!!!!!!!!!!!7887887! ! "F! 6*%-!G5(*5%!%'-%!15!./%!)*+-%/*'2!&5*;-!/*;!.*!+-%!0!"*<4+.%,@!!!!!!!!!!!!!!A%-!! #! ! ! ! ! ! !! ! ! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!B*! 9! ! "H!!!!!!!!$%%&'(!"*<4+.%,!+-0D%!*I!*./%,!)*+-%/*'2!J%<E%,-!!!7887887! ! "K! $/0.!:,*I%--1*5!;*+'2!(*+!'1&%!I*,!(*+,!"/1'2@!!!!888888888888888888888888888888888888888888888! ! "L!!!!!!!!$/%,%!;*+'2!(*+!'1&%!(*+,!"/1'2!.*!M1>%!0-!05!G2+'.@!!!!!!!!!!!!!!!!!!!!!!!!!!!!30<%!"*+5.(!!!!!!!#! ! ! ! ! ! ! ! ! !!!!!!!!!!!!!!!!!!!!!!!!N./%,!"*+5.(!!!!!!9! ! "O! P2+?0.1*50'!NEQ%?.1>%!I*,!(*+,!"/1'2! ! ! :,1<0,(!!!!!!!!!!!!#! ! ! ! ! ! ! ! ! ! )1D/!3?/**'!!!!!9! ! ! ! ! ! ! ! ! ! R51>%,-1.(!!!!!!!!=! ! ! ! ! ! ! ! ! ! S%?/51?0'!!!!!!!!!!C! ! ! ! ! ! ! ! ! ! B*5%!!!!!!!!!!!!!!!!!!F! ! ! !!"#T! $%%&'(!)*+,-!./0.!(*+,!"/1'2!:%,I*,<-!34*,.-UN+.2**,!G?.1>1.1%-@!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!7887887! !!! 89 !!! !!"##! $%%&'(!)*+,-!./0.!(*+,!"/1'2!34%52!2*156!)*7%8*,&9!!!!!!!!!!:;;:;;:! ! !!"#<! $/0.!2*!(*+!./15&!0!"*74+.%,!1-!=**2!>*,9!!! ! !!!!!!!!!!!!!!!!!!!!!!!!!?5.%,5%.!!!!!!!!#! ! @A0,&!0''!./0.!044'(B! ! ! ! ! ! ! !!!!!!!!C'0(156!!!!!!!!!!<! ! ! ! ! ! ! ! ! ! ! !!!!!!!!)*7%8*,&!!!D! ! ! ! ! ! ! ! ! ! ! !!!!!!!!E*./156!!!!!!!!!F! !!! !!"#D!!!!!G*!(*+!./15&!(*+,!"/1'2!-/*+'2!+-%!A*,%!*,!H%--!0!"*74+.%,9!!! ! !!!!!!!!!A*,%!!!!!!!!!!!!!#! ! ! ! ! ! ! ! ! ! ! !!!!!!!!!H%--!!!!!!!!!!!!!!!<! ! ! ! 90 Figure A.4 Student Questionnaire ! ! "#$%&'($)#$#*!+,#-,./!0!12,3%4!5#,!1'2)%$'*! ! ! 1&6##78!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!9.'%8! D. Information of the Student D1 Full Name _____________________________ D2 School Year (1) (2) (3) (4) (5) (6) D3 Age |__|__| D4 Sex M 1 F 2 D5 What do you want to be when you grow up? __________________ D6 Where do you want to live when you grow up? Same County 1 Other County 2 91 D7 From 1 to 10, how much do you enjoy school? |__|__| D8 What is your favorite class? __________________________ D9 How many hours do you use a computer at home per week? |__|__| D10 How many hours do you use a computer outside per week? |__|__| D11 What do you use a computer for? Internet 1 !"#$%&#''&()#(&#**'+, Homework 2 Learning 3 Reading 4 Drawing 5 Playing 6 Phonecalls 7 Don’t use 8 D12 Who have you used a computer with? Teachers 1 !"#$%&#''&()#(&#**'+, Classmates 2 Family 3 Don’t use 4
Abstract (if available)
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on education: from Indonesia to Los Angeles, CA
PDF
Essays on development and health economics
PDF
Essays on education and institutions in developing countries
PDF
Essays in political economy and mechanism design
PDF
Essays on the empirics of risk and time preferences in Indonesia
PDF
The impact of agglomeration policy on CO₂ emissions: an empirical study using China’s manufacturing data
PDF
Essays on human capital accumulation -- health and education
PDF
Essays on family planning policies
PDF
Essays on economics of education and private tutoring
PDF
Essays on the economics of subjective well-being in transition countries
PDF
Selection and impacts of early life events on later life outcomes
PDF
Essays on health and aging with focus on the spillover of human capital
PDF
Three essays on health economics
PDF
Essays on development economics and adolescent behavior
PDF
Essays on the econometrics of program evaluation
PDF
Essays on development and health economics: social media and education policy
PDF
Essays in uncertainty and aggregate economic activity
PDF
The impact of globalization and multinational corporations on education in Costa Rica
PDF
Against the wind: labor force participation of women in Iran
PDF
Adaptation, assets, and aspiration. Three essays on the economics of subjective well-being
Asset Metadata
Creator
Meza-Cordero, Jaime Andrés
(author)
Core Title
Essays on education programs in Costa Rica
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
05/22/2014
Defense Date
04/21/2014
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Avancemos,Costa Rica,Education,OAI-PMH Harvest,One Laptop per Child,program evaluation
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Nugent, Jeffrey B. (
committee chair
), Strauss, John A. (
committee chair
), Hentschke, Guilbert (
committee member
), Nyshadham, Anant (
committee member
)
Creator Email
jaimemeza733@yahoo.com,mezacord@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-414880
Unique identifier
UC11296647
Identifier
etd-MezaCorder-2527.pdf (filename),usctheses-c3-414880 (legacy record id)
Legacy Identifier
etd-MezaCorder-2527.pdf
Dmrecord
414880
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Meza-Cordero, Jaime Andrés; Meza-Cordero, Jaime Andres
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
Avancemos
One Laptop per Child
program evaluation