T h a i l a n d AUS13333 A Quality Education for All Report No: AUS13333 T h a i l a n d A Quality Education for All GEDDR EAST ASIA AND PACIFIC Document of the World Bank May 1, 2015 Standard Disclaimer: This volume is a product of the staff of the International Bank for Reconstruction and Development/The World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Copyright Statement: The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development/The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750-4470, http://www.copyright.com/. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. Contents Acknowledgements vi Abbreviation vii EXECUTIVE SUMMARY 1 I Introduction 1 II Strategies to improve performance 3 1 Overview 5 2 An assessment of system performance 9 2.1 Expanded access to education 10 2.2 Greater emphasis on education quality 11 2.3 Improved learning outcomes (as measured by international assessments) 14 2.4 Functional illiteracy 15 2.5 Widening disparities in student performance 16 3 Why is performance falling short of desired levels? 21 3.1 Remaining system-level issues affecting education quality 22 3.2 Factors contributing to the performance gap between schools 24 3.2.1 The system of educational resource allocation 24 4 Strategies to improve the quality of education for all 31 A Improving the quality of education for all 32 1 Increasing school autonomy 32 2 Strengthening the use of information to hold teachers and schools 33 accountable for performance B Reducing inequities in the education system 35 1 Utilizing existing resources more effectively 35 2 Increasing/improving financing 40 3 Improving teaching resources for small and remote schools 42 4 Increasing awareness and understanding of the small school challenge 42 References 44 Annexes 49 List of Tables Table 1 Basic comparison of small schools and all OBEC schools (in 2010 2 unless otherwise indicated) Table 1.1 Proportion of functionally illiterate 15-year-olds on the 2012 8 PISA reading assessment, by type of location Table 3.1 Basic comparison of small schools and all OBEC schools (in 2010 25 unless otherwise indicated) List of Figures Figure 1 Distribution of Thailand’s 15-year-olds on the 2012 PISA reading 1 assessment Figure 1.1 Average PISA 2012 Scores vs. GDP Per Capita (constant 2005 USD) 6 Figure 1.2 Distribution of Thailand’s 15-year-olds on the 2012 PISA reading 7 assessment Figure 2.1 Adjusted primary and secondary net enrolment rates – by wealth 11 quartile Figure 2.2 Adjusted primary and secondary net enrolment rates – by gender 11 Figure 2.3 Average monthly wages of individuals with bachelor’s degree by level 14 of experience – basic education teachers and other occupations (2013 Thai baht) Figure 2.4 Contributions of changes in quality and in student characteristics to 15 changes in PISA reading scores in Thailand from 2003 to 2012 Figure 2.5 Percentage of 15-year-olds attaining each level of proficiency in the 16 PISA 2012 reading assessment in Thailand and Vietnam Figure 2.6 Proportion of functionally illiterate 15-year-olds in 2012 by type of 16 location Figure 2.7 Urban-rural differences in student performance in PISA 2012 17 mathematics Figure 2.8 Contributions of changes in quality to changes in PISA reading scores 17 for Thailand from 2003 to 2012 across the test score distribution Figure 2.9 Socio-economic inequality in learning outcomes in Thailand – student 17 performance index by average per capita consumption level Figure 2.10 Socio-economic-related student performance inequality – average PISA 18 scores by ESCS quantile Figure 2.11 Contributions of changes in quality and in student characteristics to 19 changes in PISA reading scores in Thailand from 2003 to 2012, by location Figure 2.12 Contributions of changes in quality to changes in PISA reading scores 19 in Thai villages from 2003 to 2012 across the test score distribution Figure 2.13 Contributions of changes in quality and in student characteristics to PISA 19 reading score changes for the lowest-performing 40 percent of village students, students in large cities, and the Thai student population Figure 3.1 Per-student public expenditure (Thai baht per year) – by province 24 Figure 3.2 Number of teachers and students and student-teacher ratio – 24 primary school Figure 3.3 Average school characteristics – Bangkok and Mae Hong Son 26 iv Wanted: A Quality Education for All Figure 3.4 Teacher shortage index by selected countries and for Thailand and the 27 OECD by type of location Figure 3.5 Quality of material resources index and quality of physical infrastructure 28 index for Thailand and OECD, by type of location Figure 3.6 Number of students and projected number of small schools from year 29 1993 to 2034 (assuming the current trend continues) Figure 4.1 Example of school mapping exercise – Ubon Ratchathani province 36 Figure 4.2 Nationwide school network optimization 37 List of Boxes Box 3.1 Guidelines for teacher allocation for OBEC schools 26 Box 4.1 Brazil’s experimentation with schools report cards 33 Box 4.2 Mexico’s reforms to increase parental participation 34 Box 4.3 China’s rural primary school merger program 39 Box 4.4 The Kangjan case study of school networking 40 Wanted: A Quality Education for All v Acknowledgements This report was prepared by Dilaka Lathapipat with much-appreciated contribution from Lars M. Sondergaard and Theepakorn Jithitikulchai (consultant), and further inputs from Minna Hahn Tong (consultant) in drafting and editing. The work also benefited from a background paper titled “Grouping Thailand’s Schools into Four Categories” written by the Thailand Development Research Institute (TDRI) under a contract with the World Bank Group. The report was conducted under the guidance of Lars M. Sondergaard, Harry Anthony Patrinos, Luis Benveniste, Ulrich Zachau, Constantine Chikosi, and Shabih Ali Mohib. The team is grateful for their ongoing support and guidance provided. The authors would like to thank the National Reform Council’s “Committee on Educational Reform and the Development of Thailand’s Human Resources,” the Ministry of Education (especially the Office of the Basic Education Commission (OBEC), the Office of the Permanent Secretary, and the Office of the Education Council (OEC)), the Bureau of the Budget, the National Economic and Social Development Board (NESDB), and the Quality Learning Foundation (QLF) for numerous discussions and valuable contributions. The World Bank team would also like to thank H.E. Professor Yongyuth Yuthavong (Deputy Prime Minister), H.E. Admiral Narong Pipatanasai (Minister of Education), H.E. Police General Adul Saengsingkaew (Minister of Social Development and Human Security), and H.E. Pichet Durongkaveroj (Minister of Science and Technology), for a follow up meeting and helpful advice after completion of the report. The team wishes to acknowledge Kitti Limskul (Faculty of Economics, Chulalongkorn University), Rangsan Maneelek (OBEC), Direk Patmasiriwat (National Institute of Development Administration), Khunying Sumonta Promboon, Chan Tantithamthavorn (OEC), Lilin Songpasook (OBEC), Somkiat Tangkitvanich (TDRI), Supanutt Sasiwuttiwat (TDRI), Tirnud Paichayontvijit (TDRI), Pokpong Junvith (TDRI), Pumsaran Tongliemnak (MoE), Supakorn Buasai (QLF), and Kraiyos Patrawart (QLF) for the extensive comments and advice provided at several stages of the preparation. Peer reviewers for the report were: Anna Olefir, Dina N. Abu-Ghaida, and Juan Diego Alonso. The report also benefited from comments and suggestions provided by Harry Anthony Patrinos and Prateek Tandon. We also thank Paul Daniel Risley, Trinn Suwannapha, Kanitha Kongrukgreatiyos, Leonora Aquino Gonzales, and Buntarika Sangarun for providing excellent assistance in external relations and production of multimedia materials, and Noppakwan Inthapan for providing administrative support. Photography by Seksan Pipattanatikanunt, report cover and layout design by QUO Bangkok. This report would not have been possible without school-level data from OBEC and the Ordinary National Education Test (O-NET) exam data from The National Institute of Educational Testing Service (NIETS). vi Wanted: A Quality Education for All Abbreviation AGE Apoyo a la Gestión Escolar (School Management Support) ASEAN Association of Southeast Asian Nations ESA Educational Service Area ESCS Economic, Social and Cultural Status FY Fiscal year GDP Gross domestic product IEA International Energy Agency Lao PDR Lao People’s Democratic Republic MOF Ministry of Finance NIETS National Institute of Educational Testing Service OBEC Office of the Basic Education Commission OEC Office of the Education Council OECD Organisation for Economic Co-operation and Development ONESQA Office for National Education Standards and Quality Assessment O-NET Ordinary National Education Test PACER Parent Advocacy Coalition for Educational Rights PEC Quality Schools Program PISA Programme for International Student Assessment SBM School-Based Management SNED Sistema Nacional de Evaluación del Desempeño de los Establecimientos Educacionales Subvencionados TDRI Thailand Development Research Institute THB Thai baht TIMSS Trends in International Mathematics and Science Study TOEFL Test of English as a Foreign Language USD United States dollar Wanted: A Quality Education for All vii Regional Vice President Axel van Trotsenburg Country Director Ulrich Zachau Senior Practice Director Claudia Costin Practice Manager Harry Patrinos Task Team Leader Dilaka Lathapipat EXECUTIVE SUMMARY I Introduction Over the past two and a half decades, For example, in the most recent PISA Thailand has made great progress in reading assessment (in 2012), one-third expanding basic education, closing the of Thai 15-year-olds knew the alphabet gap in attendance between socio- and could read, but they could not locate economic groups, and putting more information or identify the main messages focus on the quality of education. in a text –they were “functionally illiterate,” Participation in primary education is nearly lacking critical skills for skilled jobs universal, and secondary school enrolment (Figure 1). In contrast, the average score has risen impressively. Inequality between for students in lower-income Vietnam was socio-economic groups in their ability to 66 points higher than the average score for access schooling has continued to decline. Thai students, which implies that, on In terms of quality improvement, the 1999 average, a 15-year-old in Vietnam is National Education Act (amended in 2002) approximately 1.5 academic years ahead put long-overdue reforms on the national of the average Thai student. agenda, with objectives such as decentralizing education administration, Functional illiteracy can be seen across establishing educational standards and the various types of schools in Thailand, quality assurance mechanisms, enhancing indicating that there are still system- the quality of teachers and education wide issues affecting the quality of personnel, and laying the foundation for education. Schools in cities, towns, and more creative student-centered teaching villages all produce students who are not and learning processes. functionally literate by the age of 15. The greatest concentration is found in villages, Building on this progress, it appears that where 47 percent of their 15-year-old even more could be done to maximize students are functionally illiterate. However, the potential of Thailand’s students to three-quarters of the functionally illiterate become productive workforce students in Thailand are found in schools participants. Closer examination of PISA located in cities, towns, or small towns scores shows that many Thai students still where most of the student population is do not have the skills and competencies enrolled. While various factors are at work, needed in an increasing number of jobs. three major factors are likely affecting Figure 1 High performers Functionally illiterate Distribution of Thailand’s (level 5 and (below level 2) 15-year-olds on the 2012 above) 32% PISA reading assessment 1% Average performers (levels 2, 3, and 4) 67% Source: OECD PISA 2012 data. Wanted: A Quality Education for All 1 Executive summary Table 1 Small schools Basic comparison of small (less than 120 students) All OBEC schools schools and all OBEC schools (in 2010 unless Total number of schools in 2011 14,669 31,211 otherwise indicated) (primary and secondary) Average per student subsidy THB 41,551 THB 31,476 (primary and secondary) Average class size (primary and secondary) 8.8 21.8 Average student-teacher ratio* (primary and secondary) 13.4 21.4 Average number of teachers per class (primary and secondary) 0.73 1.15 Percentage of students who reported being poor** 73.3% 48.8% Student performance index in the 2010 41.3 42.1 O-NET exams*** * This is the weighted average student-teacher ratio, where the weights are school enrolment size. ** Poor students are those who reported household incomes below THB 40,000 per month. *** The student performance index (which ranges from 0 to 100) is a weighted index of the 2010 Ordinary National Education Test (O-NET) exams for Grades 6, 9, and 12. Source: World Bank staff calculations based on OBEC school data 2010. quality across the education system: Village schools face their own set of i) lack of school autonomy in personnel challenges that stem from being management which prevents schools from “remote”, small, and getting smaller. ensuring that they have the highest-quality As a result of falling birth rates, the total teaching staff possible; ii) underutilization number of primary school students in of information on student learning Thailand fell from 7.45 million in 1982 to outcomes (i.e. national standardized test just 5 million in 2012. At the same time, results) to hold teachers, schools, and the number of teachers barely changed other actors accountable for performance; over the same period. Yet over a timespan and iii) inefficiencies in government of just eight years, the number of small education spending, as reflected in chronic schools with less than 120 enrolled teacher shortages despite low students1 increased dramatically from student-teacher ratios. 10,899 (or 33 percent of OBEC schools) in 2003 to 14,669 (47 percent) in 2011. The situation appears to be particularly These small schools, which have very small acute for one group of students: classes and are much more expensive to students enrolled in village schools, operate, predominantly serve the especially the lowest-performing socioeconomically disadvantaged student 40 percent among them. The lowest- population (Table 1). Furthermore, under an performing 40 percent of village students OBEC program introduced over 20 years were the furthest behind in 2003, and they ago, many primary schools in rural villages continue to fall further behind. In 2003, were “upgraded” to secondary schools by these 15-year-old students were, on adding three more grades (to help fulfill the average, 125 “points” behind their peers in government’s commitment to providing large city schools, corresponding to more secondary education for all). However, little than three academic years of schooling. was done to ensure that these former That gap widened to 139 points by 2012. primary schools could provide a quality In 2012, almost half of those students were secondary school education. In general, functionally illiterate. village schools are hindered by a lack of 1 Using the OBEC definition, a small school is a school with less than 120 students enrolled. However, this standard definition could be inappropriate in some contexts. For example, should a school that teaches grades 1 to 4 with 25 students per grade (100 students in total) be considered “smaller” than a school that teaches grades 1 to 9 with 15 students per grade (135 students in total)? The key is to use a measure that is comparable across schools. In the following section of this Executive Summary, a small school is defined as a school with 20 students or less per grade on average. Using OBEC definition, the total number of small schools in 2010 is 14,159. The number of small schools in 2010 increases to 19,864 if the alternative definition is employed. 2 Wanted: A Quality Education for All Executive summary adequate teachers, material resources, and them break out of the cycle of poverty. physical infrastructure. Looking at the bigger picture, having a workforce with stronger analytical Improving educational outcomes among reasoning and problem solving skills – these poorer-performing students can skills that extend well beyond simply being have major impacts at the individual functionally literate – can help Thailand level and for Thailand’s economic move up the value-added ladder to a more growth prospects. For individuals, being knowledge-based economy. Therefore, equipped with the necessary skills and addressing the remaining gaps will enable competencies to obtain productive Thailand to improve its competitiveness, employment can help them secure a better economic growth, and prosperity. future and, for those who are poor, help II Strategies to improve performance Building on the major progress made in placing greater emphasis on improving access to education, Thailand improvements in student learning now has the opportunity to improve the outcomes; and iii) requiring publication quality of education for all and more fully of school budgets and resource tap the potential of its future workforce. allocations across schools to enable This section highlights some key areas parents and communities to monitor in which reforms could have a significant the efficiency of resource usage by impact on the quality of education, both for their schools. students throughout the education system and more specifically for students in rural B Reducing inequities in the village schools. education system The growing inequities in education call A Improving the quality of for a shift, from focusing on providing education for all schooling access to providing a quality To help improve the quality of education education in Thailand’s small village across Thailand’s education system, schools, which are at the heart of this actions are needed on multiple fronts. problem. The overall objective of this This list is by no means comprehensive, shift would be to bring learning standards but some key priorities include: everywhere to the same level as Bangkok. Furthermore, in focusing existing resources • Increasing school autonomy. more strategically so they can be put to Assessments of implementation of optimal use in improving learning school autonomy and accountability outcomes where help is needed most, policies in Thailand and elsewhere have such reforms have the potential to improve shown that increasing school autonomy the efficiency of Thailand’s education over personnel management can spending tremendously. Some options for improve student learning, in particular consideration include: at better-performing schools. Autonomy could perhaps first be increased for • Utilizing existing resources more better-performing schools and delayed effectively. A detailed mapping of for other schools until they have a schools shows that the vast majority sufficient level of capacity and proper of schools are small and located accountability for results. within 20 minutes from another school – 16,943 out of 19,864 small schools2 • Strengthening the use of information are non-isolated. With careful planning to hold teachers and schools and support, these schools could be accountable for performance. reorganized into fewer but larger and Several measures could be considered: better-resourced schools to provide a i) making school-level results on higher-quality education. Such standardized exams publicly available; reorganization could take the form of: ii) in school and teacher evaluations, 2 A small school is defined here as a school with 20 students or less per grade on average. Wanted: A Quality Education for All 3 Executive summary i) school mergers, which would involve public subsidies. Schools would also merging two or more schools within the likely become more cost-efficient as same area to form a bigger school; teachers and educational personnel are ii) school networking, which would no longer free resources from a involve reorganizing classes and the school’s perspective. structure of schools within the same area so they can share resources; and/ • Improving teaching resources for or iii) redefining poorly performing small and remote schools. OBEC schools to cover fewer and could introduce measures to improve lower grades. training in multi-grade teaching, so schools with severe teacher shortages • Increasing/improving financing. It could then consolidate classrooms and is estimated that for every province provide multi-grade education more to bring their learning standards up effectively. Building on experience from to Bangkok standards, the share of the health sector, OBEC could also government expenditure on education explore options for providing stronger would need to increase from 24.0 to incentives for quality teachers to be 28.8 percent. Furthermore, providing all deployed in small, remote schools. schools with the level of teacher quality and number of teachers per classroom • Increasing awareness and necessary to achieve Bangkok-level understanding of the small school learning standards would require challenge. Decisions on the approach recruiting, training, and deploying to reducing disparities in education 164,000 new teachers – an increase of could benefit from further research to over 40 percent of the teaching force understand the small school challenge, in Thailand in 2014. Another financing particularly at the primary school level. option that could be considered is And importantly, greater willingness financing schools based on the number of politicians and government officials of students they have enrolled rather to explain and discuss the problem of than on the inputs they employ. This small schools and the available options financing approach would incentivize to improve quality of learning would schools to improve quality and attract help generate a more informed dialogue more students in order to earn larger on this issue. Summary of recommended actions Improving the quality of education for all Reducing inequities in the education system Increase school autonomy Utilize existing resources more over personnel management effectively through school reorganization Strengthen the use of information to hold teachers and Increase/improve financing schools accountable for performance for schools Improve teaching resources for small and remote schools Increase awareness and understanding of the small school challenge 4 Wanted: A Quality Education for All Overview 1 Overview Over the past two and a half decades, decentralizing education administration, Thailand has made great progress in establishing educational standards and expanding basic education, closing quality assurance mechanisms, enhancing the gap in attendance between the quality of teachers and education socioeconomic groups, and putting personnel, and laying the foundation for more focus on the quality of education. more creative student-centered teaching Participation in primary education is nearly and learning processes. universal, and secondary school enrolment has risen impressively. Inequality between These achievements have likely socioeconomic groups in their ability to contributed to the steady improvement access schooling has continued to decline. in Thai students’ performance on In terms of quality improvement, the 1999 international assessments over the past National Education Act (amended in 2002) decade. On the OECD’s Program for put long-overdue reforms on the national International Student Assessment (PISA),3 agenda, with objectives such as Thailand’s average score has now reached 570 Figure 1.1 Average PISA 2012 Scores Singapore vs. GDP Per Capita Hong Kong SAR, China 550 (constant 2005 USD) Korea, Rep. Japan Finland Macao 530 Estonia SAR, China Canada Netherlands Poland New Zealand Germany Ireland Switzerland Vietnam Czech Australia 510 Belgium Republic Slovenia Austria Norway United Denmark Latvia France Luxembourg 490 Hungary Portugal Kingdom United States Lithuania Slovak Spain Italy PISA scores Croatia Republic Iceland Russian Sweden 470 Federation Israel Greece Turkey 450 Serbia Romania Cyprus Thailand Bulgaria United Arab 430 Chile Emirates Costa Rica Kazakhstan Malaysia Mexico 410 Montenegro Uruguay Tunisa Brazil Jordan Colombia 390 Albania Indodesia Peru Qatar 370 <3,000 3,000- 4,000- 5,500- 7,000- 9,500- 15,000- 22,000- 30,000- 37,000- 40,000- ≥48,000 3,999 5,499 6,999 9,499 14,999 21,999 29,999 36,999 39,999 47,999 GDP per capita (constant 2005 USD) Note: The average of PISA scores in mathematics, science, and reading is used in this graph. Source: OECD 2012 PISA and World Development Indicators. 3 The PISA is an international survey that aims to evaluate education systems worldwide by testing the skills and knowledge of 15-year-old students. The tests are designed to assess the extent to which students can apply their knowledge to real-life situations and be prepared for full participation in society. To date, students from more than 70 countries have participated in the assessment, which is conducted every three years (see www.oecd.org/pisa/ for more details). 6 Wanted: A Quality Education for All Overview Figure 1.2 High performers Functionally illiterate Distribution of Thailand’s (level 5 and (below level 2) 15-year-olds on the 2012 above) 32% PISA reading assessment 1% Average performers (levels 2, 3, and 4) 67% Source: OECD PISA 2012 data. a level slightly above what would be Functional illiteracy is not an isolated expected for a country at Thailand’s level challenge and can be seen across the of per capita income (Figure 1.1). With a various types of schools in Thailand. GDP per capita of USD 3,390 (in constant As Table 1.1, shows, schools in cities, 2005 USD) and an average PISA score of towns, and villages all produce students 437, Thailand’s performance is roughly who are not functionally literate by the age similar to those of Bulgaria, Romania, and of 15. Most of the functionally illiterate Chile and well above those of Malaysia, students are where most of the students Brazil, and Mexico, which have higher are enrolled, namely in schools located in levels of per capita income. The one outlier a city, town, or small town. Although “only” is Vietnam with a GDP per capita of USD one-third of their students are assessed as 986, whose 15-year-old students functionally illiterate, they comprise over performed at the level of students in much three-quarters of all functionally illiterate richer countries such as Australia, students in Thailand. Germany, and New Zealand. The situation appears to be particularly Building on this progress, it appears that acute for one group of students: even more could be done to maximize students enrolled in village schools, the potential of Thailand’s students to especially the lowest-performing 40 become productive workforce percent among them. As shown in Table participants. Closer examination of PISA 1.1, the lowest-performing 40 percent of scores shows that many Thai students still village students were the furthest behind in do not have the skills and competencies 2003, and they continue to fall further needed in an increasing number of jobs. behind. In 2003, these 15-year-old For example, in the most recent PISA students were, on average, 125 “points” reading assessment (in 2012), one-third of behind their peers in large city schools, Thai 15-year-olds knew the alphabet and corresponding to more than three could read, but they could not locate academic years of schooling. That gap information or identify the main messages widened to 139 points by 2012. In 2012, in a text – they were “functionally illiterate,” almost half of the village students were lacking critical skills for skilled jobs functionally illiterate. Moreover, the (Figure 1.2). In contrast, the average score problem is very likely worse than what for students in lower-income Vietnam was is captured by the numbers, as many 66 points higher than the average score for 15-year-olds in villages dropped out earlier Thai students, which implies that, on and are not even in school. If they had average, a 15-year-old in Vietnam is been kept in school and had been tested approximately 1.5 academic years ahead as part of PISA, the disparities likely of the average Thai student. would have been even wider since Wanted: A Quality Education for All 7 Overview Table 1.1 Functionally Point gap with Proportion of functionally illiterate in 2012 large cities illiterate 15-year-olds on the 2012 PISA reading % of % of Breakdown In In assessment, by type 15-year-old PISA PISA functionally of national 2003 2012 of location student 2003 2012 illiterate level, by population reading reading students in location in 2012 score score location Village 16% 394 410 47% 7% 68 73 Lowest-performing 337 344 125 139 40% in village Highest-performing 432 454 30 29 60% in village Small town, town, city 77% 422 444 31% 24% 39 39 Large city 7% 461 483 16% 1% – – Thailand 100% 420 441 32% Source: World Bank staff calculations based on OECD PISA data 2003 and 2012. dropouts generally tend to be This report aims to help policymakers lower-performing students. identify opportunities to strengthen the quality of education for all Thai students Improving educational outcomes among in order to build a more skilled and these poorer-performing students can competitive workforce. The purpose of have major impacts at the individual the study is to answer two overarching level and for Thailand’s economic questions: 1) How well is Thailand’s growth prospects. For individuals, being education system doing in preparing its equipped with the necessary skills and students to become productive workers? competencies to obtain productive and 2) What more can be done to improve employment can help them secure a better student performance and to close the future and, for those who are poor, help attainment gap between socioeconomic them break out of the cycle of poverty. groups? The rest of the report is structured This is especially important since recent as follows. Chapter 2 provides an overview research on Thailand suggests that private of system performance, reviewing the returns to higher education as well as the strengths and current challenges in the returns to educational quality have been education system. Chapter 3 takes a closer rising consistently over the last couple of look at the challenges, examining why decades (Lathapipat, forthcoming). student performance may be falling short Looking at the bigger picture, having a of desired levels and what factors may workforce with stronger analytical be contributing to the growing inequities. reasoning and problem solving skills – Based on this analysis, Chapter 4 provides skills that extend well beyond simply being various options for addressing these functionally literate – can help Thailand challenges and helping Thailand improve move up the value-added ladder to a more the quality of education for students across knowledge-based economy. Therefore, the country, drawing from experiences addressing the remaining gaps will enable both within and outside Thailand. Thailand to improve its competitiveness, economic growth, and prosperity. 8 Wanted: A Quality Education for All An assessment of system performance 2 An assessment of system performance As noted in Chapter 1, while Thailand focus on quality. It then describes the has made impressive progress in remaining challenges being faced in improving access to education, delivering a quality education to all challenges persist in helping all students students and addressing inequities in the achieve better learning outcomes and education system. reducing inequities in student performance. This chapter provides a The good news: more children are brief assessment of the performance of the enrolling and staying in school Thai education system, beginning with a (including poor children), and education review of recent achievements in quality is similar to other expanding access and strengthening the middle-income countries 2.1 Expanded access to education Thailand has made great progress in secondary enrolment rate among females expanding basic education and was 82 percent, around 8 percentage narrowing inequities in schooling access points higher than that among males between socio-economic groups. As (Figure 2.2). illustrated by Figure 2.1 which presents primary and secondary net enrolment rates Thailand’s success in achieving by wealth quartile, enrolment inequality near-universal primary education and between socio-economic groups has significantly raising secondary declined considerably. The left graph on enrolment comes from sustained efforts primary enrolment shows that participation to expand school coverage and in primary education has become nearly compulsory education. Under the 1977 universal for all wealth groups. Moreover, National Scheme of Education which substantial progress has been made in encompassed three 5-year educational increasing secondary net enrolment, which development plans (the Fourth, Fifth, and rose impressively from 31 percent in 1990 Sixth National Education Development to around 78 percent in 2011. Inequality in Plans), the policy agenda began to widen secondary enrolment between the rich and to address poverty and inequality issues the poor also narrowed significantly over (Office of the National Education the past couple of decades. Commission, 1999). In particular, the Fifth National Education Development Plan Throughout this period, gender gaps in aimed to expand compulsory education primary education enrolment have been (6 years of formal schooling) to all sub- virtually non-existent. In fact, the gender districts (tambons) of Thailand in the 1982 gap has reversed at the secondary level, academic year (Bhumirat et al., 1987). In with female enrolment surpassing that of 1987, the “Educational Opportunity males from 1992 onward. As of 2011, the Expansion School” program4 was 4 The “Educational Opportunity Expansion School” program was initiated to help fulfill the government’s commitment to providing secondary education for all. However, apart from adding additional grades, little was done to ensure that these former primary schools could provide a quality secondary school education. As discussed in Chapter 3, these rural schools are generally understaffed and are inadequately endowed with material resources (science laboratory equipment, library materials, instructional materials, etc.) and physical infrastructure. 10 Wanted: A Quality Education for All An Assessment of System Performance Figure 2.1 Adjusted primary net enrolment Adjusted primary and 100% secondary net enrolment rates – by wealth quartile 98% 96% Net enrolment rate 94% 92% 90% 88% Quartile 1 (Poorest) 86% Quartile 2 84% Quartile 3 Quartile 4 (Richest) 82% 1986 1989 1992 1995 1998 2001 2004 2007 2010 Year Secondary net enrolment 90% 80% 70% Net enrolment rate 60% 50% 40% 30% Quartile 1 (Poorest) 20% Quartile 2 10% Quartile 3 Quartile 4 (Richest) 0% 1986 1989 1992 1995 1998 2001 2004 2007 2010 Year Note: Children are divided into four wealth quartiles (the poorest in Quartile 1 and the richest in Quartile 4) according to their family per capita monthly expenditure,5 expressed in adult-equivalent units.6 Source: World Bank staff calculations based on Thailand Household Socioeconomic Survey (various years). Figure 2.2 Adjusted primary net enrolment Adjusted primary and 100% secondary net enrolment Net enrolment rate rates – by gender 98% 96% 94% 92% Male Female 90% 1986 1989 1992 1995 1998 2001 2004 2007 2010 Year Secondary net enrolment 90% Net enrolment rate 80% 70% 60% 50% 40% Male 30% Female 20% 1986 1989 1992 1995 1998 2001 2004 2007 2010 Year Source: World Bank staff calculations based on Thailand Household Socioeconomic Survey (various years). 5 Per capita monthly expenditure was used instead of household income to rank individuals into wealth groups, since expenditure data are generally recorded more accurately than current income in socioeconomic surveys of households. Current spending is also less volatile and is a better proxy for family wealth than income. 6 The Thai Household Socioeconomic Survey (SES) datasets contain measures of monthly household income and expenditure. In order to compare expenditures across households, it is important to correct for household composition and household size by dividing total consumption expenditure by the number of “adult equivalents” in the household to obtain the per capita monthly expenditure in adult-equivalent scale. Each child under age 15 is treated as equivalent to 0.5 adults. All individuals in each round of the SES dataset are then classified into four wealth quartiles based on their household’s per-capita expenditure. Wanted: A Quality Education for All 11 An Assessment of System Performance established to add lower secondary level and for students in the upper classes to existing rural primary schools, secondary level through undergraduate without charging tuition fees. Within school. The 1999 National Education eight years, the total number of these Act (amended in 2002) institutionalized schools increased to around 6,600, compulsory education of nine years and accommodating 21 percent of lower guaranteed the right of all children living secondary students (Varavan, 2006). in Thailand to receive basic education of A Student Loan Fund was also established quality free of charge for 12 years. As of in 1995 to provide funds for students from 2009, mandatory free education was low-income families who continue non- extended further to 15 years, including formal education at the lower secondary three years of pre-primary schooling. 2.2 Greater emphasis on education quality The ambitious 1999 National Education community groups, LAOs, alumni, and Act (amended in 2002) led to long- academicians. In addition, LAOs can overdue, comprehensive reform of the provide education services at any or all Thai education system. The Act aims to levels commensurate with their readiness, promote the decentralization of education suitability, and the requirements of the administration to Educational Service local area9 (Office of the Education Council, Areas (ESAs), educational institutions, and 2007). local administration organizations (LAOs). Teachers and institutions are allowed more At the same time, education planning freedom to set curricula and mobilize was consolidated at the central level. resources for the provision of education. In accordance with the 1999 National Furthermore, the Act lays the foundation Education Act, the Ministry of Education is for more creative student-centered responsible for promoting and overseeing teaching and learning methods. It also all levels and types of education under the stresses the importance of quality administration of the state. With regard to assurance in education and the continuous decentralizing educational administration process of teacher development. to LAOs, the Ministry prescribes criteria and procedures for assessing the Emphasis was placed on decentralizing readiness of the LAOs to provide education administrative responsibilities to the services and assists in enhancing their local level. The decentralization process capability in line with the policies and led to the establishment of 175 ESAs in required standards. It also advises on the 2003.7 Each ESA is comprised of a budgetary allocations provided by LAOs Subcommittee for Teachers and (Office of the Education Council, 2007). Educational Personnel and an Area Committee for Education, whose office is To help improve the quality of education, responsible for approximately 200 two major tasks were undertaken to educational institutions and a student develop education standards and the population of between 300,000 to 500,000. quality assurance system. The Office of Administration and management relating the Education Council (OEC) is responsible to academic matters, budgets, personnel, for formulating national education and general affairs are now the standards in cooperation with the offices responsibility of the educational institutions responsible for basic, vocational, and themselves.8 Oversight is through a board higher education. The national education of 7-15 members consisting of standards also serve as the basis for representatives of parents, teachers, setting assessment standards for internal 7 In 2010, the total number of ESAs increased to 225 (183 for primary and 42 for secondary level). 8 In reality, however, while schools are enjoying more autonomy with regard to curriculum and budget, they still have very little influence over personnel management. This issue is discussed further in Chapter 3. 9 It should be noted that the transfer of educational institutions to be under the jurisdiction of LAOs has progressed very slowly. As of academic year 2013, 1,730 educational institutions (out of 38,010 public and private schools) were under local supervision. This represents an increase of only 795 schools since academic year 2003 (Office of the Permanent Secretary, Ministry of Education, 2003, 2013). 12 Wanted: A Quality Education for All An Assessment of System Performance and external quality assurance and improving the teaching-learning mechanisms. A continuous process of process, efforts were made to modify the internal quality assurance was instituted admission system at both the basic and as part of education administration. higher levels to avoid placing too much The Office for National Education emphasis on examinations that depend Standards and Quality Assessment mainly on rote learning (Office of the (ONESQA), established in 2000, has Education Council, 2004). been tasked with overseeing the external assessments of educational institutions, In order to equip teachers with the skills which should receive external quality needed for the new student-centered evaluations at least once every five years approach, the Ministry of Education (Office of the Education Council, 2007; undertook measures in 2002-2003 to National Education Act 1999). reform the teacher education curriculum. Up until the late 1990s, basic In accordance with the new national education teachers in Thailand generally education standards, schools have been obtained either a two-year diploma or a required to shift from a teacher-centered bachelor’s degree from a variety of teacher approach to a learner-centered one. To colleges and universities (IEA 2009). As support this transition, the OEC identified part of the education reforms, numerous 586 individuals as “Master Teachers” who initiatives were undertaken to improve would then train 8,848 teachers in their teacher education. For example, the network. A “Research and Development Rajabhat Universities (formerly Rajabhat of Learner-Centered Learning Models” Institutes or teacher training colleges) project was conducted by the OEC in began developing several higher education collaboration with the Master Teachers and curricula, including the five-year their network teachers to develop and test bachelor’s degree curriculum for teaching various teaching-learning techniques that new teachers for basic education10 and would enhance the thinking processes of master’s degree programs in teaching and their students. Teaching models developed education administration. Two-year around these techniques were piloted in 90 bachelor’s degree courses were also basic education schools between developed to upgrade in-service teachers 2005-2006, with encouragingly positive and administrators possessing two-year results. The teaching models were diplomas. The latter were developed in compiled in multimedia form and response to the Teachers Council’s new disseminated to schools across the professional standards which required country, and seminars were organized in teachers to have at least a bachelor’s the four regions of the country so teachers degree in education or a bachelor’s could participate and learn from each other degree in other fields, plus completion of a (Office of the Education Council, one-year graduate certificate in education 2005-2006, 2007). In addition to (Office of the Education Council, 2004, developing the basic education curriculum 2005-2006). 10 The curriculum includes four years of coursework and a one year of teaching practice at approved schools (IEA, 2009; Office of the Education Council, 2004). Wanted: A Quality Education for All 13 An Assessment of System Performance Because teachers are critical to 1988 Figure 2.3 Average monthly wages of improving the quality of education, 60,000 Monthly wage (Thai baht) Teachers individuals with bachelor’s considerable efforts have been made 50,000 degree by level of Non-teachers to raise the status and standards of the 40,000 experience – basic education teachers and teaching profession. As shown in Figure 30,000 other occupations (2013 2.3, wage differentials between teachers Thai baht) 20,000 and non-teaching professionals grew from 1988 to 1998 in favor of non-teaching 10,000 professionals with more work experience.11 0 0-5 6-10 11-15 16-20 21-25 26-30 30+ Unsurprisingly, studies conducted around Years of experience that time found that students entering teacher education generally had low 1998 60,000 examination scores, and concerns were Monthly wage (Thai baht) Teachers raised about the quality of the teacher 50,000 Non-teachers workforce (Fry, 1999). As part of its reform 40,000 efforts, the government has initiated 30,000 scholarship and job guarantee programs 20,000 in order to attract more qualified students 10,000 into the teaching profession. For example, between 2004-2006, five-year scholarships 0 0-5 6-10 11-15 16-20 21-25 26-30 30+ were provided to 7,500 qualified students Years of experience who were required to teach in basic 2013 education institutions upon graduation 60,000 (Office of the Education Council, 2004). Monthly wage (Thai baht) Teachers The teacher salary structure has also been 50,000 Non-teachers adjusted upward over the past decade. 40,000 Given the finding from a 1998 survey of 30,000 1,000 graduates that low salaries were the 20,000 greatest deterrent to entering teaching 10,000 (Fry, 1999) and the fact that wage gaps 0 have narrowed significantly over the last 0-5 6-10 11-15 16-20 21-25 26-30 30+ decade (Figure 2.3), there are reasons to Years of experience be optimistic that the new breed of Source: Thai Labor Force Surveys and World Bank staff teachers will be of higher quality. calculations. 2.3 Improved learning outcomes (as measured by international assessments) The performance of 15-year-old Thai quantify how much of the increase was students in PISA assessments has attributable to changes in the background improved markedly since 2003, characteristics12 of the student population apparently thanks in large part to and how much was attributable to improvements in education quality or improvements in the “quality” or effectiveness. In the PISA reading effectiveness of the Thai education system assessment, for example, the average in transforming students’ characteristics score rose from 419.6 in 2003 to 441.4 into learning outcomes.13 As shown in in 2012. An analysis was conducted to Figure 2.4, improvements in education 11 For Figure 2.3, data from Thai Labor Force Surveys was used to estimate the average monthly wages (including overtime and bonuses) for basic education teachers as well as for non-teaching professionals who had attained bachelor’s degree qualification at the time of the surveys. Estimates of monthly wages were made at seven different levels of work experience across three different points in time (1988, 1998, and 2013). 12 Theobservable student characteristics used in the analysis include gender, grade level, and the PISA index of economic, social, and cultural status (ESCS). The ESCS index was derived from the following three indices: highest occupational status of parents, highest education level of parents, and home possessions. The index of home possessions comprises all items on the indices of family wealth, cultural possessions, home educational resources, as well as books in the home. 13 SeeAnnex Section A2.1 for details on the statistical technique used in this study to decompose the change in student test scores over time. 14 Wanted: A Quality Education for All An Assessment of System Performance Figure 2.4 450 Contributions of changes 440 in quality and in student PISA reading score 15.8 characteristics to changes 430 in PISA reading scores in 6.0 Thailand from 2003 to 2012 420 410 400 419.6 419.6 Quality 390 Student 2003 380 Thailand 2003 Thailand 2012 Source: World Bank staff calculations based on OECD PISA 2003 and 2012 data. quality accounted for 15.8 points of the at levels similar to their counterparts in 21.8-point increase in overall PISA reading much richer economies. For example, the scores for Thailand from 2003 to 2012. average reading score of 483 for students in large Thai cities is almost identical to the On average, learning outcomes are average reading scores achieved by broadly similar to those of other students in Sweden and Iceland. However, countries at Thailand’s level of income. this is the case for students in Thailand’s As mentioned in Chapter 1, the large cities, and the situation is very performance of Thai students as measured different for students in other locales as by average PISA 2012 scores in discussed below. mathematics, reading, and science is now slightly above what would be expected for The challenges: functional illiteracy is a country at Thailand’s level of per capita high, and the gaps in learning outcomes income. It also appears that 15-year-old are widening Thai students in large cities are performing 2.4 Functional illiteracy As mentioned in Chapter 1, test scores Vietnamese students failed to reach level 2 indicate that too many Thai students proficiency in the reading assessment. reaching the end of their compulsory education (grade 9) are not well- Functional illiteracy can be seen across prepared for further education and/or the various types of schools in Thailand. labor market entry. Figure 2.5 shows the As illustrated by Figure 2.6, schools in percentages of 15-year-old students in cities, towns, and villages all produce Thailand and Vietnam who attained each students who are not functionally literate by level of proficiency in the PISA 2012 the age of 15.15 The greatest concentration reading assessment. Nearly one-third of is found in Thai villages, where 47 percent Thai students scored below level 2,14 which of their 15 year-old students are is regarded as the minimum level for functionally illiterate. However, as Figure “functional literacy” needed to manage 2.6 shows, most of the functionally illiterate daily living and employment tasks that students in Thailand are found in schools require reading skills beyond a basic level. located in cities, towns, or small towns In contrast, in neighboring and lower- where most of the student population are income Vietnam which participated in enrolled. These students comprise around the PISA assessments for the first time three-quarters of all functionally illiterate in 2012, only 8 percent of 15-year-old students in Thailand. 14 “Some tasks at this level require the reader to locate one or more pieces of information, which may need to be inferred and may need to meet several conditions. Others require recognizing the main idea in a text, understanding relationships, or construing meaning within a limited part of the text when the information is not prominent and the reader must make low level inferences. Tasks at this level may involve comparisons or contrasts based on a single feature in the text. Typical reflective tasks at this level require readers to make a comparison or several connections between the text and outside knowledge, by drawing on personal experience and attitudes” (OECD, 2014). 15 OECD PISA defines a village as a community with less than 3,000 people; a small town has 3,000 to 15,000 people; a town has 15,001 to 100,000 people; a city has 100,001 to 1,000,000 people; and a large city has over 1,000,000 people. Wanted: A Quality Education for All 15 An Assessment of System Performance 45% Figure 2.5 Percentage of 15-year-olds 40% attaining each level of 35% proficiency in the PISA 2012 reading assessment 30% in Thailand and Vietnam 25% 20% 41% 38% 32% 15% 27% 23% 24% 10% 5% Thailand 8% 6% Vietnam 0% Functionally Level 2 Level 3 Level 4 illiterate or higher (below level 2) Source: OECD PISA 2012 data. Percentage of functionally Breakdown of national-level Figure 2.6 illiterate students in each location functionally illiteracy by location Proportion of functionally illiterate 15-year-olds in 50% Large city 3% 2012 by type of location 45% Village 40% 22% 35% 30% 25% 47.2% 20% 15% 30.6% 10% 15.5% 5% Small town, 0% town and city Village Small town, Large city 75% town and city Source: OECD PISA 2012 data. 2.5 Widening disparities in student performance An international comparison shows that score distribution confirms a significant compared to high-performing countries, increase in student performance the Thai education system is relatively inequality in Thailand, due largely to unequal. Even though positive changes in education quality. In Figure relationships between socio-economic 2.8, the dotted line shows total changes in background and student performance PISA reading scores for 15-year-old Thai are very common across many countries, students across percentiles of the test Figure 2.7 shows that three low-performing score distribution. For example, over the education systems – namely, Thailand, 2003-2012 period, the performance of Thai Malaysia, and Indonesia – are relatively students ranked at the 50th percentile (or more unequal than high-performing the median) improved by 24.1 points. In systems such as the OECD countries, comparison, the reading score of students Korea, and Japan. Significant urban-rural ranked at the 20th percentile improved by differences in student performance indicate only 19.4 points. The solid line on the that students outside of a few elite schools graph shows the portion of the change in large cities are not acquiring the same in test scores that can be attributed to level of mastery of key cognitive skills. improvements in education quality, and the gap between the “Overall” and “Quality” An examination of changes in the PISA lines represents the effect of changes in reading score over time across the test observed student characteristics.16 The 16 SeeAnnex Section A2.1 at the end of this report for details on the decomposition of the change in student test scores over time. The distribution graphs for the test scores are also presented in Figure A2.1 in the same section. 16 Wanted: A Quality Education for All An Assessment of System Performance Figure 2.7 600 Urban-rural differences in student performance in PISA 2012 mathematics 550 PISA mathematics score 500 Korea, Rep. 450 Japan OECD 400 Thailand Malaysia Indonesia 350 Town, small town City Large city and village Source: World Bank staff calculations based on OECD PISA 2012. Figure 2.8 30 Contributions of changes Change in PISA reading score 25 in quality to changes in PISA reading scores for 20 Thailand from 2003 to 2012 across the test score 15 distribution 10 5 0 -5 Overall Quality -10 0.0 0.2 0.4 0.6 0.8 1.0 Quantile Note: The median or 0.5 quantile is equivalent to the 50th percentile, the 0.2 quantile is equivalent to the 20th percentile, and so on. Source: World Bank staff calculations based on OECD PISA 2003 and 2012 data. Figure 2.9 50 Bangkok Socio-economic inequality 48 Student performance index in learning outcomes in Thailand – student 46 performance index by average per capita 44 consumption level 42 40 38 36 34 30,000 50,000 70,000 90,000 110,000 130,000 150,000 Household annual per capita consumption (THB) Source: World Bank staff calculations based on Thailand Household Socioeconomic Survey 2011 and NIETS 2010. figure suggests that inequality in test disparities in student performance can be scores was attributable more to changes seen across schools serving students from in education quality than to changes in diverse socio-economic backgrounds. observed student characteristics. Figure 2.9 plots provincial average household per capita annual consumption Closer examination of student in 2011 against the student performance performance by socio-economic index for the 2010 Ordinary National background indicates that the weaker Education Test (O-NET) exams.17 A clear performers tend to come from more positive and exponential relationship can disadvantaged backgrounds. Large be seen, with per capita consumption 17 The student performance index (which ranges from 0 to 100) is a weighted index of the 2010 O-NET exams in mathematics and science for students in Grades 6, 9, and 12. Details on the computation of the index are provided in Section A5.3 in the Technical Appendix to Annex 5 at the end of this report. Wanted: A Quality Education for All 17 An Assessment of System Performance 550 Figure 2.10 2012 Socio-economic-related 2003 student performance 500 inequality – average PISA Average PISA score scores by ESCS quantile 450 400 350 0 2 4 6 8 10 ESCS quantile Source: World Bank staff calculations based on OECD PISA 2003 and 2012 data. alone explaining around 16 percent of the students has increased only marginally. total provincial variation in student In fact, the test scores for the poorest performance. A similar analysis using 10 percent of Thai students have improved school average PISA 2012 test scores in the least, indicating that student mathematics and science and the PISA performance inequality between the richest index of economic, social, and cultural and poorest has been rising sharply over status (ESCS) of the student body reveals the last decade. that average ESCS accounts for as much as 49 percent of the total variation in Looking more closely by type of school average test scores across schools. location, it appears that the These results strongly suggest that the disadvantaged, poorer-performing socio-economic background of the student students are concentrated in lagging body is one – if not the most important – village schools. Breaking down Thailand’s determinant of school performance (secondary) school system by location type in Thailand. reveals that performance has been improving at very different rates Not only is socio-economic-related (Figure 2.11). A large disparity in learning student performance inequality high, outcomes is apparent, with students in but it also has been rising over the last villages having the lowest average decade. Figure 2.10 presents the performance in any given year. More estimated relationships between household importantly, the performance gap between socio-economic status (as measured by students in village schools and those in city the PISA ESCS index) of 15-year-old and large city schools widened significantly students in Thailand and their performance over the period. As represented by the on the PISA reading assessments in 2003 green portion of the bars in Figure 2.11, and 2012.18 Unsurprisingly, for both time the widening urban-rural disparity in periods, well-off students generally learning outcomes was driven largely by performed better on the reading differences in education quality. assessment than their more disadvantaged peers. Taking the difference between the Notably, the analysis also reveals that two lines for 2003 and 2012 at each ESCS students in small town schools quantile indicates the change in the benefited the most from improvements average reading score across the two in education quality over the last cohorts at that particular quantile. The decade. As shown in Figure 2.11, the results show that while test scores of average reading score of students in small Thai students have improved significantly towns actually surpassed that of students over the last decade, the richest students in town schools in 2012. Important lessons (ranked at the top end of the ESCS could be learned from investigating the distribution) have had the largest factors that contributed to these improvement in test scores, while the impressive gains, which is beyond the performance of the most disadvantaged scope of this report. Nevertheless, the gap 18 Specifically, for each year of data, the student population was ranked into quantiles according to their family socio-economic status, as measured by the PISA ESCS index. Average PISA reading scores for students were then estimated along the entire socio-economic status distribution. 18 Wanted: A Quality Education for All An Assessment of System Performance to students in large city schools remains lower quantiles. In fact, the quality of substantial at 45 points, which schools serving the lowest-performing corresponds to more than one academic 40 percent of the village student year of schooling. population declined the most over the period under study. In contrast, school quality for the lowest-performing students in villages Figure 2.13 below shows that worsening has actually deteriorated over time. school quality accounted for a 7.8-point The problem of deteriorating education decline in the reading score for the quality is particularly concerning for the lowest-performing 40 percent of the poorest-performing children who attend village student population. At the other schools in Thai villages. As shown in end of the performance spectrum, students Figure 2.12, changes in education quality in Thailand’s large cities benefited from (represented by the solid line) actually had an increase in education quality, which a negative effect on overall test scores accounted for a 17-point increase in (represented by the dotted line) for the their performance. Figure 2.11 490 Contributions of changes 480 in quality and in student 17.0 characteristics to changes 470 in PISA reading scores in 4.4 Thailand from 2003 to 460 12.4 PISA reading score 2012, by location 450 8.6 440 15.8 430 6.6 24.2 2.6 6.0 461.5 420 410 8.1 440.7 10.1 424.4 400 419.6 Quality 6.1 405.3 Student 390 393.7 2003 380 Thailand Village Small town Town City Large city Source: World Bank staff calculations based on OECD PISA 2003 and 2012 data. Figure 2.12 50 Contributions of changes Change in PISA reading score 40 in quality to changes in PISA reading scores in 30 Thai villages from 2003 to 2012 across the test 20 score distribution 10 0 -10 -20 Overall Quality -30 0.0 0.2 0.4 0.6 0.8 1.0 Quantile Note: The median or 0.5 quantile is equivalent to the 50th percentile, the 0.2 quantile is equivalent to the 20th percentile, and so on. Source: World Bank staff calculations based on OECD PISA 2003 and 2012 data. Figure 2.13 25 Change in PISA reading score Contributions of changes 20 in quality and in student characteristics to PISA 15 reading score changes 17.0 for the lowest-performing 10 15.3 40 percent of village 5 students, students in 4.4 large cities, and the Thai 0 student population -7.8 Quality -5 Student -10 Lowest 40% village Large city Source: World Bank staff calculations based on OECD PISA 2003 and 2012 data. Wanted: A Quality Education for All 19 An Assessment of System Performance It should be noted that the disparities assume that the majority of these children are likely much greater than what is are socio-economically disadvantaged captured by the above analysis of PISA and/or academically ill-prepared, and they results, as many 15-year-olds in villages are therefore very likely to be functionally dropped out earlier and are not even in illiterate. Therefore, if the 15-year-olds in school. While the analyses of educational villages who dropped out earlier had been performance inequality in this chapter were kept in school and had been tested as carried out using PISA data on 15-year-old part of PISA, the disparities in results likely Thai secondary students, it is important to would have been even wider since note that some 20 percent of secondary dropouts generally tend to be school-age children are not even enrolled lower-performing students. in secondary schools. It is probably fair to 20 Wanted: A Quality Education for All Why is performance falling short of desired levels? 3 Why is performance falling short of desired levels? As a basis for considering the various challenges related to small village schools directions and options for education that are contributing to the growing reform, it is useful to understand the inequities. This chapter digs deeper into causes behind Thailand’s current each of these challenges and examines education challenges. It appears that their underlying causes, in order to help performance is being held back by two identify some appropriate policy options factors: i) remaining system-level issues for tackling these issues going forward. affecting education quality and ii) particular 3.1 Remaining system-level issues affecting education quality While much greater policy emphasis protect job tenure. The selection process is has been placed on education quality in based on compliance with teacher recent years, there are still unfinished standards requirements (such as system-level reforms that could completion of a university degree), without advance the agenda further. As provision for more local control. Because described in Chapter 2, considerable the hiring process is bureaucratic efforts have been made to develop better rather than selective on the basis of quality, teaching techniques, improve teacher schools have very little room for education, and raise the status of the maneuvering when assigned inadequate teaching profession (through increased teachers. Moreover, since teacher salaries salaries to teachers) to attract more are regulated, there is limited scope for qualified individuals. These reforms could implementing a system of rewards and be complemented by further progress on sanctions that could improve incentives system-level reforms that improve teacher (Arcia and Patrinos, 2013). management and increase accountability for the quality of education delivered. Another system-level challenge is that despite having a relatively sophisticated Teacher management is still highly system for assessing student learning centralized, so schools have little outcomes, the information is not being influence over personnel choices. While fully utilized. The National Institute of one of the main objectives of the 1999 Educational Testing Service (NIETS), National Education Act is to promote the established in 2005, has been assessing decentralization of education all 6th, 9th, and 12th graders for nearly a administration to Educational Service decade using the Ordinary National Areas (ESAs), educational institutions, and Educational Test (O-NET). However, local administration organizations, schools Thailand does not utilize this information have very little influence over personnel as much as other countries with a similarly management. Teacher management – rich assessment culture. Some examples: hiring and firing, disciplining, deployment, and payroll administration – has been • The information has little impact on evaluation of teachers (and their decentralized to the ESA level, but only salaries and career prospects). Teacher following the regulations set by the Teacher salaries in Thailand are determined by Civil Service Commission which tend to 22 Wanted: A Quality Education for All Why is performance falling short of desired levels? two factors: base salary and academic an ESA and an OBEC official), nor do position. The current base salary they seem to be part of any budget adjustment guideline from the Office of discussion with MOF (e.g., justifying the Teacher Civil Service Commission more/less spending, changes in does not place any weight on student spending, or new programs). In performance. Regarding the academic addition, while national-level results for position adjustment, student standardized exams are publicly performance in national standardized available, school-level results are not. exams accounts for only 3.3 percent of This limits the ability of key the total evaluation score. The rest is stakeholders to monitor and evaluate based on understanding of schools effectively. teaching materials, ability to assess student learning, use of Information Another challenge relates to education and Communications Technology, and spending: while a relatively large share understanding of the code of ethics of GDP (and government resources) (TDRI, 2013). goes to education, those resources could be used more effectively. • The Office for National Education Illustrating the government’s commitment Standards and Quality Assessment to education, the budget for the education (ONESQA) is tasked with conducting sector remains one of the largest items on “assessments” of all schools every five the budget. For FY 2015, the budget for years but has done so using criteria that education totals THB 531 million focus mainly on inputs and processes, (comprising 20.6 percent of the with very little weight on the student government’s overall budget and 4 assessment data produced by NIETS. percent of GDP), of which THB 388 million ONESQA follows a policy called is for pre-primary, primary, and secondary “amicable assessment” which education (Bureau of the Budget, 2015). emphasizes support for improvement However, there are various indications rather than high-stakes accountability. that this funding could be put to more While this is laudable approach, it is optimal use to improve education quality. unclear whether the results generated One sign of inefficiency is that Thailand are being used to make important has relatively low student-teacher ratios operational or policy decisions,19 and a – for example, the student-teacher ratio great amount of effort and resources in primary schools fell from nearly 30:1 in (on average, approximately USD 1,500 the late 1970s to just 16:1 in 2012—but per assessment) is being spent on almost one-third of Thai classrooms face these amicable assessments (World chronic teacher shortages (less than one Bank, 2011).20 teacher per classroom) due to ineffective • Assessment results (whether from teacher allocation (see Table A3.1 in Annex NIETS or ONESQA) are not used to hold 3). As discussed below, resources are schools or other actors accountable being spread thinly across schools that for their performance. For example, are standing half-empty due to dwindling the results do not appear to be used to student numbers. Many schools are in dire inform performance discussions among need of massive capital investments but any actors in the system (e.g., between do not have the funds for upgrading. Thus, the head of an Education Service Area there is considerable scope for improving and a school, or between the head of how educational resources are allocated. 19 According to the manual of ONESQA, the third round of evaluations (2011-2015) will place more weight on O-NET exam results. 20 World Bank (2011): Thailand Public Finance Management Report Discussion Paper 5: Analysis of Efficiency of Education. Wanted: A Quality Education for All 23 Why is performance falling short of desired levels? 3.2 Factors contributing to the performance gap between schools 3.2.1 The system of educational the total number of primary school resource allocation students in Thailand fell from 7.45 million in 1982 to just 5 million in 2012. However, While public spending appears to be the number of teachers barely changed pro-poor, the key question is whether over the same period (Figure 3.2.1). As a this translates into better educational result, the student-teacher ratio in primary resources for disadvantaged schools. schools fell from nearly 30:1 in the late At first glance, it is encouraging to see that 1970s to just 16:1 in 2012 (Figure 3.2.2). public subsidies for basic education are Therefore, at the macro level, Thailand pro-poor, although the correlation seems certainly has an adequate number of rather weak, as shown in Figure 3.1. These teachers relative to the number of subsidies encompass all recurrent public students. Nonetheless, as discussed expenditure on schools under the Office of below, many provinces are facing severe the Basic Education Commission (OBEC), teacher shortages due to ineffective which includes salaries of teachers, teacher allocation. administrators, and other education personnel. A key question is: does the The proliferation of small schools, in observed pro-poor public spending mean particular, is putting pressure on public that students in socio-economically educational resources.21 Despite disadvantaged schools are allocated declining student numbers, the number of relatively better educational resources? small schools with less than 120 students The rest of this chapter attempts to answer increased dramatically from 10,899 (or this question through rigorous analysis of 33 percent of OBEC schools) in 2003 to the available data. 14,669 (47 percent) in 2011. These small schools, which have very small classes As a starting point for investigating and are much more expensive to operate, inequities in public education resource predominantly serve the socioeconomically allocation, it is important to look at the disadvantaged student population adequacy of teacher allocations across (Table 3.1).22 schools. As a result of falling birth rates, 45,000 Figure 3.1 Per-student public Average per-student subsidy expenditure (Thai baht per 40,000 year) – by province 35,000 30,000 Bangkok 25,000 20,000 30,000 50,000 70,000 90,000 110,000 130,000 150,000 Household annual per capita consumption (THB) Source: World Bank staff calculations based on Thailand Household Socioeconomic Survey 2011 and OBEC. 21 Using the OBEC definition, a small school is a school with less than 120 students enrolled. However, this standard definition could be inappropriate in some contexts. For example, should a school that teaches Grades 1 to 4 with 25 students per grade (100 students in total) be considered “smaller” than a school that teaches Grades 1 to 9 with 15 students per grade (135 students in total)? The key is to use a measure that is comparable across schools. In the analysis presented in Chapter 4, a small school is defined as a school with 20 students or less per grade on average. 22 For a more detailed breakdown of key characteristics by school size, see Table A3.1 in Annex Section A3. 24 Wanted: A Quality Education for All Why is performance falling short of desired levels? Figure 3.2.1 8,000,000 400,000 Number of teachers and 7,500,000 students – primary school 350,000 7,000,000 Number of teachers Number of students 6,500,000 300,000 6,000,000 5,500,000 250,000 5,000,000 200,000 4,500,000 Teachers Students 4,000,000 150,000 1971 1975 1979 1983 1987 1991 1995 1999 2003 2007 2011 Year Figure 3.2.2 35 Student-teacher ratio – 33 primary school 31 Student-teacher ratio 29 27 25 23 21 19 17 15 1971 1975 1979 1983 1987 1991 1995 1999 2003 2007 2011 Year Source: EdStats. Table 3.1 Small schools Basic comparison of small schools and all OBEC (less than 120 students) All OBEC schools schools (in 2010 unless Total number of schools in 2011 14,669 31,211 otherwise indicated) (primary and secondary) Average per student subsidy THB 41,551 THB 31,476 (primary and secondary) Average class size (primary and secondary) 8.8 21.8 Average student-teacher ratio* (primary and secondary) 13.4 21.4 Average number of teachers per class (primary and secondary) 0.73 1.15 Percentage of students who reported being poor** 73.3% 48.8% Student performance index in the 2010 41.3 42.1 O-NET exams*** * This is the weighted average student-teacher ratio, where the weights are school enrolment size. ** Poor students are those who reported household incomes below THB 40,000 per month. *** The student performance index (which ranges from 0 to 100) is a weighted index of the 2010 Ordinary National Education Test (O-NET) exams for Grades 6, 9, and 12. Source: World Bank staff calculations based on OBEC school data 2010. While small schools may offer some sufficient resources to deliver a quality advantages in certain contexts, education. Despite small class sizes and Thailand’s small village schools are low student-teacher ratios, Thailand’s severely under-resourced. While one village schools do not have adequate could argue that small schools offer numbers of teachers—on average, these potential advantages such as a small small schools are staffed with less than learning environment and greater one teacher per classroom (Table 3.1). This teacher-student interaction, the benefits means that students across all grades are realized only if the school has cannot be taught at the same time, unless Wanted: A Quality Education for All 25 Why is performance falling short of desired levels? Share of teachers Average teacher Average number of Figure 3.3 with graduate degree experience (years) teachers per classroom Average school characteristics – Bangkok 20% 28 1.8 and Mae Hong Son 1.6 24 16% 1.4 Teachers experince (year) Teachers per classroom Teachers with degrees 20 1.2 12% 16 1.0 19.7% 12 0.8 1.61 23.5 8% 0.6 8 8.7% 0.4 4% 10.8 0.71 4 0.2 0% 0 0 Bangkok Mae Hong Son Bangkok Mae Hong Son Bangkok Mae Hong Son Source: World Bank staff calculations based on OBEC school data 2010. Box 3.1 Guidelines for teacher Schools with 120 enrolled students or less: allocation for OBEC • 1 to 20 students: 1 principal 1 teacher schools • 21 to 40 students: 1 principal 2 teachers • 101 to 120 students: 1 principal 6 teachers Schools with 121 enrolled students or more: • Pre-primary: teacher: student ratio = 1:25 student: teacher ratio = 30:1 • Primary: teacher: student ratio = 1:25 student: teacher ratio = 40:1 • 121 to 359 students: 1 principal • 360 to 719 students: 1 principal 1 assistant principal • 720 to 1,079 students: 1 principal 2 assistant principals • 1,080 to 1,679 students: 1 principal 3 assistant principals • 1,680 or more students: 1 principal 4 assistant principals the schools rely on multi-grade teaching average per capita consumption, and has and/or teachers cover a much broader the lowest population density – are range of subjects than in a larger school. allocated less qualified teachers with the least amount of teaching experience, and A comparison of Bangkok (where their classrooms are severely understaffed, schools, on average, are the largest) and with an average of only 0.71 teachers Mae Hong Son province (where schools, per classroom.24 on average, are the smallest) illustrates the wide disparities in adequacy and Considering the current teacher quality of teaching resources.23 As shown deployment guidelines (presented in in Figure 3.3, Bangkok enjoys much larger Box 3.1), it is not hard to envisage how shares of teachers with higher than a the allocation process would be bachelor’s degree, teachers with more problematic for the majority of primary years of experience, and more teachers schools with less than 120 enrolled per classroom. In contrast, schools in Mae students. For example, under the Hong Son province – which is located in guidelines, a small primary school with 20 the northern region bordering Myanmar students spread across a few grade levels and is the poorest province in terms of would be entitled to only one teacher. 23 For more detailed comparisons of educational resources across all Thai provinces, ranked by student performance, see Figure A3.1 in Annex Section A3. 24 In Bangkok, average enrolment is 753 students for primary school and 2,298 for secondary school. In Mae Hong Son, average enrolment is 83 students for primary school and 328 students for secondary school. 26 Wanted: A Quality Education for All Why is performance falling short of desired levels? Figure 3.4.1 Teacher shortage index, selected countries Teacher shortage index by 1.0 selected countries 0.8 0.6 0.4 0.2 0 -0.2 -0.4 D na a n . a e m nd ep or si si pa EC na hi la ne ay  R ap Ja  C et ai O a, al do ng R, Vi Th re M In SA Si Ko ng Ko g on H Figure 3.4.2 Rural-urban differences in teacher shortage index, Teacher shortage index for Thailand and the OECD Thailand and the OECD by 1.4 type of location 1.2 1.0 0.8 0.6 0.4 0.2 0 Thailand OECD -0.2 Village Small Town City Large town city Source: OECD PISA 2012. Given the severe staffing constraint with have been in service for over two years, the sole teacher being responsible for and it does not provide the right incentives teaching students in all grades and to educational personnel to work in remote subjects, it would be unreasonable to areas. This results in a disproportionately expect the school to deliver large share of teachers with relatively few high-quality education. years of experience in remote schools. The highly centralized teacher Compared to international peers, the management system, as discussed Thai school system is severely lacking earlier in this chapter, is further in qualified teachers, particularly in rural compounding the teacher shortage villages. On average, school principals in problem for small, remote schools. Thailand reported shortages of teaching High-caliber teachers generally do not staff in key subjects that seriously hindered want to go to remote schools, especially student learning. Compared to regional if the schools are tiny since it limits their peers and the OECD average, Thai schools interaction with and learning from peers. have the highest PISA teacher shortage Furthermore, the workload in remote index25 by far (Figure 3.4.1). Figure 3.4.2 schools may be much higher as the which presents the teacher shortage index teachers must cover a wider range of by location, confirms the earlier finding that subjects and/or grades. Currently, the schools in rural areas are more severely centralized teacher deployment process understaffed than their urban counterparts, allows teachers to be redeployed to any and this teacher allocation inequality is location of their own choosing once they much worse than that in OECD countries. 25 The PISA index on teacher shortage was derived from four items measuring the school principal’s perceptions of potential factors hindering instruction at school. The four items indicate shortages of qualified teachers in: i) science, ii) mathematics, iii) the test language within the economy, and iv) other subjects. More positive values on this index indicate higher rates of teacher shortage at a school. Wanted: A Quality Education for All 27 Why is performance falling short of desired levels? Rural-urban differences in quality of Rural-urban differences in quality of Figure 3.5 material resources index physical infrastructure index Quality of material 0.4 0.2 resources index (left) and quality of physical 0.2 Physical infrastructure index 0 infrastructure index (right) Material resources index 0 for Thailand and OECD, by -0.2 -0.2 type of location -0.4 -0.4 -0.6 -0.6 -0.8 -0.8 -1.0 -1.0 -1.2 Thailand Thailand -1.2 -1.4 OECD OECD -1.6 -1.4 Village Small Town City Large Village Small Town City Large town city town city Source: OECD PISA 2012. In addition to shortages in human provided low-quality education at a resources, Thai schools are hindered relatively high cost. Notably, schools in by inadequate material resources and Bangkok require relatively low per-student physical infrastructure. School principals subsidies, even though they are endowed in Thailand reported shortages or with more qualified and more experienced inadequacy of material resources26 and teachers as well as a higher number of physical infrastructure27 which limited the teachers per classroom on average (see capacity of schools to provide quality bar charts in Figure A3.1 in Annex A3). instruction. Compared to international Figure A3.1.6 in Annex A3 shows that the peers, Thai schools are more severely average personnel salary per student for hindered in these dimensions. Once again, schools in Bangkok is among the lowest in schools primarily serving disadvantaged the country. Given that these schools are children in rural areas are generally much allocated more and better educational more lacking in material resources and resources, the lower per-student physical infrastructure than those in urban expenditure was achievable only because areas, and this resource allocation these schools operate on much larger inequality is much worse than that in OECD scales in terms of enrolment and countries (Figure 3.5). This level of under- class sizes. resourcing means that for Thailand’s small village schools which are already expensive Given Thailand’s demographic trends, to operate (as reflected in high per-student the number of low-quality small schools subsidies), closing the performance gap will likely increase in the near future, would require a massive outlay that would which also means that performance make those schools even more expensive. gaps are likely to widen. As shown in Figure 3.6, the declining trend in birth rates Thus, while public education spending means that the number of students is in Thailand may appear to be pro-poor, expected to drop to 5.6 million by 2034, the small schools serving disadvantaged while the number of small schools with students face major resource less than 120 students is projected to constraints that affect their ability to continue increasing to 18,291 schools.28 provide a quality education. As indicated The growing number of small schools by the above analysis, socioeconomically resulting from dwindling numbers of disadvantaged children in Thailand are students means that more and more 26 The PISA index on the school’s material resources was computed on the basis of six items measuring the school principals’ perceptions of potential factors hindering instruction at school. These are shortage or inadequacy of: i) science laboratory equipment, ii) instructional materials, iii) computers for instruction, iv) internet connectivity, v) computer software, and vi) library materials. All items were reversed for scaling so that more positive values on this index indicate higher quality of material resources at a school. 27 The PISA index on quality of physical infrastructure was computed on the basis of three items measuring the principals’ perceptions of potential factors hindering instruction at school. The three items indicate shortages or inadequacy of: i) school buildings and grounds, ii) heating/cooling and lighting systems, and iii) instructional space. Once again, all items were reversed for scaling so that more positive values on this index indicate higher quality of physical infrastructure at a school. 28 The projection of the total number of small schools is based on the assumption that no school will be closed down as a result of the falling number of students. 28 Wanted: A Quality Education for All Why is performance falling short of desired levels? schools in Thailand will become chronically more and more children in rural areas to understaffed (assuming that the current be enrolled in larger and better-resourced teacher allocation rule still applies in the schools located in urban areas. Those left future) and poorly equipped. Furthermore, behind will likely be the most socio- improvements in transportation will enable economically disadvantaged children. Figure 3.6 20,000 10 Number of students and projected number of small schools from year 1993 to 18,000 9 Number of students (million) 2034 (assuming the current Number of schools trend continues) 16,000 8 Number of students Number of small schools 14,000 7 12,000 6 10,000 5 93 95 97 99 01 03 05 07 09 11 13 15 17 19 21 23 25 27 29 31 33 19 19 19 19 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 Year Note: The estimation is based on the enrolment rate provided by International Futures at the Pardee Center (2014) and the number of Thai children aged 3-17 reported by the United Nations, Department of Economic and Social Affairs. This was adjusted with the number of students in OBEC using Statistic of Thailand Education (from the Office of the Education Council, 2013). Source: Authors’ calculations Wanted: A Quality Education for All 29 Strategies to improve the quality of education for all 4 Strategies to improve the quality of education for all Building on the major progress made in This chapter presents some potential improving access to education, Thailand policy options that could help ensure now has the opportunity to improve the a high-quality education for all Thai quality of education for all and more fully students. It highlights some key areas tap the potential of its future workforce. in which reforms could have a significant A range of deep and sustained actions impact on the quality of education, both for aimed at improving the quality of education students throughout the education system could help Thailand improve student and more specifically for students in rural learning and develop a more competitive village schools. This list is by no means workforce. Efforts to bring learning comprehensive but is meant to illustrate standards throughout the country – the range of reforms that could be including in small and remote villages – to undertaken in light of the challenges the level of learning standards in Bangkok discussed in the previous chapters. could also help disadvantaged students It draws upon experiences from within and improve their future earning opportunities outside Thailand as well as more in-depth and break out of poverty. Tackling these analysis of particular options and their challenges successfully will enable potential impacts. Thailand to maximize its human capital, reduce inequality, and foster economic growth. A Improving the quality of education for all 1 Increasing school autonomy if the schools are already performing well. Analysis of PISA data for Thailand An assessment of implementation of (Lathapipat, 2015) shows that enhanced school autonomy and accountability autonomy, especially regarding personnel policies in Thailand showed that management (which encompasses increasing school autonomy over selecting teachers for hire, firing teachers, personnel management can improve establishing teachers’ starting salaries, and student learning. The study finds that determining teachers’ salary increases), Thai schools with an average PISA score significantly improves learning for better- above 500 points showed higher indicators performing schools (those in the top half of personnel management autonomy than of the performance distribution), while the schools with PISA scores below 400 points opposite seems to be true for schools with and that PISA scores are significantly weaker performance (bottom half of the affected by personnel autonomy (Arcia and performance distribution). The analysis also Patrinos, 2013). These findings suggest shows that learning improves the most in that greater autonomy in personnel a system with strong accountability (using management could improve performance. levels of parental monitoring as proxy for accountability). However, evidence from within Thailand suggests an important caveat: school These findings are consistent with autonomy does not make sense evidence from other countries. Analysis everywhere. Increased autonomy can of PISA data for 42 countries found that improve learning outcomes in schools, 32 Wanted: A Quality Education for All Strategies to improve the quality of education for all Box 4.1 Brazil’s experimentation with schools report cards An example of an effort to generate and disseminate information to improve education service delivery is the experimentation with school report cards in the Brazilian state of Parana between 1999 and 2002. The report card contained school data on test-based performance for 4th and 8th graders and parent opinions about their children’s school. It also included student flow data (promotion, retention, and dropout rates), school characteristics (e.g., average class size, share of teachers with college degrees) from the annual school census, student information (e.g., family socioeconomic status) from questionnaires attached to statewide achievement tests, and principals’ statement about their management style. Whenever possible, comparative municipal and state averages of key indicators were provided so parents and teachers could compare the performance of their school with that of neighboring schools. Schools were also reported to be performing at, below, or above their expected performance, controlling for the socio-economic background of the students. The three-page summary of indicators was disseminated to parents and teachers through various local-level workshops, and results were published in the state education secretariat’s monthly newsletter and widely disseminated through press releases and press conferences. While no rigorous evaluations of this experiment have been undertaken, anecdotal evidence suggests positive results. Parents engaged in discussions with teachers about how they might improve school performance and, through school councils, increased their voice in policy debates about education. Source: Burns et al., 2011 and Winkler, 2004. autonomy affects student achievement teachers and schools can be held negatively in developing and low- accountable. As discussed earlier, while performing countries but positively in national-level results for standardized developed and high-performing countries exams are publicly available, school- (Hanushek et al., 2011). Furthermore, level results are not. Making school-level recent empirical evidence from Latin results publicly available would enable America shows very few cases in which key stakeholders to monitor and evaluate school autonomy made a significant performance effectively. Making difference in learning outcomes – in most assessment results more readily available of the 30 cases examined, Patrinos (2011) has helped improve state-level finds that school autonomy was increased accountability and student achievement in through simple mechanisms for increasing the United States (Hanushek and parent participation and financial Raymond, 2005; Carnoy and Loeb, 2002), accountability, but with little consequence Brazil (Burns et al., 2011), and Mexico to educational personnel. The study (Alvarez et al., 2007).29 A particularly strongly suggests that giving schools inspiring example of an effort to generate operational autonomy can have a and disseminate information to improve significant impact on learning if it affects education service delivery is the teacher motivation. All of these findings experimentation with school report cards in indicate that increased autonomy across Brazil conducted between 1999 and 2002 the board may not be desirable – perhaps (see Box 4.1). autonomy could first be increased for better-performing schools and delayed for Strengthening the linkage between other schools until they have a sufficient teacher/school evaluation and student level of capacity and proper accountability performance could help incentivize for results. teachers and schools to improve the quality of teaching. In evaluating teachers and schools, greater emphasis could be 2 Strengthening the use of placed on improvements in learning information to hold teachers outcomes, while taking into consideration and schools accountable for differences in student background performance characteristics. Appropriate incentives for It would be useful to make data on education personnel that are directly linked standardized exams available so to improvements in student learning could 29 Using PISA 2012 data, Lathapipat (2015) finds evidence that teacher evaluation and reward mechanisms that are directly linked to student learning outcomes have positive impacts on student performance in Thailand. Achievement data disclosure is also found to have significant positive impacts for schools throughout the entire performance distribution. Wanted: A Quality Education for All 33 Strategies to improve the quality of education for all Box 4.2 Mexico’s reforms to Mexico’s school-based management (SBM) programs grew out of a concern for equity and for increase parental poor, rural, and heavily indigenous schools, which led to a large-scale compensatory education participation program. That program included a small-scale parental participation program, the Support to School Management (or AGE), introduced in 1996. AGE consists of monetary support and training to parent associations. The parent associations can spend the money for the purpose of their choosing, although spending is limited to small civil works and infrastructure improvements. They are not allowed to spend money on wages and salaries for teachers. Despite being a limited version of SBM, the AGEs represent a significant advance in the Mexican education system, where parent associations have tended to play a minor role in school decision-making. The AGE financial support consists of quarterly transfers to APF school accounts, varying from USD 500 to USD 700 per year according to the size of the school. AGE helps generate significantly higher levels of school participation and communication – both among parents, and with teachers and school principals – because of the projects that parent associations undertake, but more so because of the training they receive and the meetings they undertake. The AGE helps articulate expectations and promotes social participation. Many parents believe that the AGEs put pressure on school principals and teachers to help their children. AGE also motivates parents to follow their children’s progress. Rigorous impact evaluations have shown that AGE improves parental participation and improves the school climate (Gertler et al., 2008). It has also been shown that AGE leads to improvements in schooling outcomes such as reduced grade repetition and failure, and better test scores (Shapiro and Moreno, 2004; Lopez-Calva and Espinosa, 2006). Mexico’s successful experience with SBM in rural areas led to the creation of an urban, now nationwide, more advanced program known as the Quality Schools Program (PEC) in 2001 with the goal of expanding autonomy and improving learning in Mexican schools. Participation in PEC entails the following: preparation of a plan by staff and parents of a school that outlines steps for improvement; five-year grants to schools to implement the activities; parental participation in designing and implementing plans; and training of school principals. Several qualitative evaluations find positive effects on test scores, with the largest gains in schools that had the poorest students, and a positive impact on school climate and processes (Loera 2005). PEC also leads to higher accountability and transparency levels (Patrinos and Kagia, 2007). After only a few years of implementation, participation in PEC significantly decreases dropout rates, failure rates, and repetition rates (Skoufias and Shapiro, 2006). Source: Barrera et al. (2009), quoted in Patrinos et al. (2010). be explored. Real consequences for poor schools facing relatively good chances of performance could also be adopted to winning the award. help ensure that students are being taught by high-caliber, effective teachers. As In addition, increasing the transparency documented in Vegas (2005), international of school budgets and allocations would evidence shows that teachers respond to help stakeholders hold schools incentives. International experience also accountable for resource usage. To offers lessons on the pitfalls to avoid (e.g., improve transparency, schools could be teachers could stop cooperating or could required to publish their budgets, while start teaching to the test). Chile’s SNED local authorities could be required to program30 offers one example: since 1996, publish the allocation of resources across monetary bonuses have been offered to schools (through a transparent and schools that show excellent performance in equitable per-student funding formula, terms of student achievement. Teachers in as discussed below). Such transparent winning schools receive what has typically per-student allocation formulae are used amounted to half of one month’s salary, or in a number of countries around the world, between 5-7 percent of a teacher’s annual including Romania (since 2009), Bulgaria salary. Although impact evaluations of the (since 2006), and many others SNED are difficult due to the absence of a (Sondergaard et al., 2011). This would natural control group, Vegas (2005) enable parents and communities to provides some preliminary evidence that monitor the efficiency of resource usage the incentive has had a cumulative positive by their schools. As illustrated by Mexico’s effect on student performance for those reforms to increase parental participation 30 SNEDstands for the Sistema Nacional de Evaluación del Desempeño de los Establecimientos Educacionales Subvencionados program. 34 Wanted: A Quality Education for All Strategies to improve the quality of education for all (Box 4.2), an increased role for parents in for performance can have a significant holding schools and teachers accountable impact on schooling outcomes. B Reducing inequities in the education system The growing inequities in education call communities at the forefront with OBEC for a shift from focusing on providing providing overall leadership and direction, schooling access to focusing on providing incentives to ESAs and local providing a quality education in communities to identify solutions, Thailand’s small village schools, which anticipate and address bottlenecks, and are at the heart of this problem. The share best practices on how to use overall objective of this shift would be to resources more effectively. At the same bring learning standards everywhere to time, OBEC could also explore some the same level as Bangkok. Furthermore, more overarching “structural” options for in focusing existing resources more addressing teacher shortages such as strategically so they can be put to optimal improving the incentives for teachers to use in improving learning outcomes where work in remote areas and providing more help is needed most, such reforms have support on how to provide multi-grade the potential to improve the efficiency teaching effectively. These options are of Thailand’s education discussed in greater detail below. spending tremendously. 1 Utilizing existing resources This shift could build on existing more effectively programs such as OBEC’s “Educational Opportunity Expansion School” Thailand is in the fortunate position of program. As mentioned in Chapter 2, this having the opportunity to improve its OBEC program, which was introduced over resource use without having to go two decades ago, extends primary schools through the difficult transition process with three more grades to allow them to faced by other countries in similar offer secondary education. Although this situations. As discussed earlier, Thailand’s helps expand access to secondary oversized school network no longer fits the education, these expanded primary current and projected school population. schools are most likely the schools that are Unlike countries in Eastern Europe which struggling to provide a quality education. face a similar problem, Thailand fortunately With sufficient staffing and resources, this does not face a financial necessity to OBEC program could provide a platform downsize (in part, because Thailand does for new initiatives that shift the focus from not have large and growing social improving access to education to assistance outlays, as is the case in aging improving the quality of education Eastern Europe); rather, the impetus for provided, not only at these secondary addressing this problem in Thailand is to schools but also more broadly for all small create optimal learning environments and and rural primary schools. make the best use of the teacher workforce. Thailand thus has some Given the large scale and complexity of breathing room to carefully plan for a the challenge and the different transition to a smaller network, learn what circumstances facing each of Thailand’s works, and bring stakeholders on board. 19,864 small schools,31 a “one-size-fits- Because the Eastern European countries all” and/or a top-down approach is waited until they faced a fiscal crunch, unlikely to work. More promising by the time they had to act, they were approaches would place the need of faced with the difficult challenge of having 31 Recallthat the previous chapters utilized OBEC’s definition for a small school, which refers to a school with less than 120 students enrolled. As mentioned earlier, this standard definition can be problematic in certain contexts as it does not allow direct comparisons to be made between schools that offer different numbers of grade levels. This chapter therefore uses a measure that is comparable across schools: a small school is defined here as a school with 20 students or less per grade on average. Wanted: A Quality Education for All 35 Strategies to improve the quality of education for all Figure 4.1 N Map Legend Example of school Big cchool mapping exercise – Outside network school Ubon Ratchathani province Teacher per class > 1 (Education service area 1) Teacher per class < 1 Extended school Road Kilometers 0 2.5 5 10 15 20 Source: School mapping exercise carried out for this report. to lay off 10-20 percent of the teacher access to education.33 The vast majority workforce.32 In contrast, Thailand could of small schools are located within embark on a 10-year journey to 20 minutes from another school (16,943 consolidate its school network without out of 19,864 small schools34). As such, having to lay off a single worker in with careful planning and support, these the process. schools could be reorganized into fewer but larger and better resourced-schools Broadly speaking, Thailand’s 10-year (which an analysis of Thailand’s journey to improve the use of existing assessment data in Annex A4 shows would resources and address teacher provide high-quality education35). The shortages would likely involve a remaining 15 percent (or 2,921) of small reorganization of schools through a schools are isolated, with no other schools combination of approaches. These nearby.36 These schools are unable to approaches could include school mergers, share resources with other schools due to school networking, and redefining schools. their remote locations and should not be Each of these options is described in merged, since access would be adversely greater detail below. affected. Figure 4.1, which presents a visualization of the school mapping A careful mapping of schools shows that exercise conducted for schools in 85 percent of small schools are located Education Service Area 1 of Ubon in relatively close proximity to other Ratchathani province, illustrates that the schools and so could be reorganized majority of small schools are located in relatively easily without impairing close proximity to each other or to other 32 Thechallenges faced due to dwindling student numbers in Eastern European countries is well-documented in World Bank public expenditure reviews, see for example World Bank (1998): “Romania Public Expenditure Review,” World Bank (2012): “Bulgaria: Public Expenditure for Growth and Competitiveness,” and World Bank (2013): “Serbia: Municipal Finance and Expenditure Review.” 33 An accompanying study to this report (“Grouping Thailand’s Schools into Four Categories”) was conducted by the Thailand Development Research Institute Foundation (TDRI) under the guidance and contract with the World Bank Group. The school grouping exercise classified OBEC schools into four distinct categories: 1) isolated small schools, 2) isolated large schools, 3) non-isolated small schools, and 4) non-isolated large schools. “Small schools” were defined as those with 20 students or less per grade. 34 These numbers exclude 137 schools with no students in 2010. 35 AnnexA4 presents empirical evidence that staffing classrooms in poor schools with adequate numbers of good-quality teachers could have a large impact on student learning. 36 Aschool is defined as isolated if there is no school of a similar type (meaning some/all grade levels taught at the schools overlap) located within 20 minutes from it or if the sub-district where the school is situated is more than 500 meters above sea level. 36 Wanted: A Quality Education for All Strategies to improve the quality of education for all Figure 4.2 35,000 Nationwide school 30,000 network optimization Number of schools 9,421 25,000 20,000 15,000 16,943 Non-isolated large schools 14,252 10,000 Non-isolated small schools 5,000 Isolated large schools 1,908 1,908 2,921 2,921 Isolated small schools 0 Status quo School network optimization Source: Authors’ calculations based on OBEC school data 2010. larger schools (as indicated by the pink shows that they need to be planned with lines connecting the schools). The extreme care and sensitivity to local paragraphs below discuss how resource concerns. Local stakeholders – parents, sharing could help such schools improve students, local administration, and local the utilization of resources (particularly politicians – may understandably be teachers) and thus improve resistant to the idea of merging their local student learning. schools with others. In anticipation of this, numerous measures could be adopted to School mergers help alleviate such concerns, such as: School mergers would involve merging • Establish the notion of a “central two or more schools within the same school” (or “receiving school”), area to form a bigger school, with the providing additional resources to build objective of creating larger, better- their capacity to absorb more students resourced schools rather than spreading from nearby areas (e.g., through resources thinly across numerous small additional support teachers, transport, schools. The school mapping exercise canteen meals, and semi-boarding suggests that at the national level, the facilities). Bulgaria introduced such 16,943 non-isolated small schools could schools in 2008 as part of an effort to be merged with each other and/or to the consolidate its school network, and 9,421 non-isolated large schools to create Moldova introduced such schools with 14,252 large schools (a total reduction of the support of a World Bank project 12,112 schools) without impairing access starting in 2013.37 to education. This represents around 39 percent of total schools under OBEC • Establish the notion of a “protected supervision. The hypothetical reform would school” as a school that cannot and reduce teacher shortages substantially, should not be consolidated (otherwise reducing the total number of classrooms access to education would be hurt), staffed with less than one teacher from using clearly defined and objective 110,725 to just 12,600 and reducing the data-driven criteria. Once such schools total number of understaffed schools from are clearly defined (and distinguished 14,159 to only 1,739. The reform would from other small schools), it will be also result in a massive increase in the easier to mobilize additional resources average number of teachers per in terms of teachers and materials. classroom, from 1.15 to 1.39 (an increase Again, the Bulgaria experience provides of more than 21 percent) – importantly, an example of this. without needing to hire an extra teacher. • Provide bussing and/or pay for transportation. For example, ahead of While school mergers can bring major their big attempts at merging schools, efficiency gains and are relatively quick Bulgaria and Moldova purchased to implement, international experience 37 For a detailed description of the policy measures adopted by Bulgaria, please see Ministry of Education and Science (2008): National Report on the Development of Education in Bulgaria (available at http://www.ibe.unesco.org/National_Reports/ ICE_2008/bulgaria_NR08.pdf). For details on the Moldova project, please see World Bank (2012): Project appraisal document for Moldova Education Reform Project. Wanted: A Quality Education for All 37 Strategies to improve the quality of education for all minibuses and distributed them to local One successful example of a school authorities. This was done as part of a merger in Thailand is the Jai-Prasan-Jai dialogue with local governments about model in Lopburi province. In this case, the constraints they faced in OBEC officials selected one school to consolidating their networks. teach all levels, so the students, teachers, and principals of nearby schools all moved • Provide stipends (conditional on to this school. The school facilities with attendance) to students involved in no students are still maintained for local school mergers. Providing such events and community use. conditional stipends might help reduce resistance from parents and also help A rural primary school merger program reduce the risk that children might drop in China is one of the few for which out (as a result of having to move to rigorous evaluation of program impacts a different school with a better but on student academic performance has also more challenging been undertaken. Evidence from the learning environment).38 impact evaluation indicates that students’ • Give parents a school choice if their academic performance improved closest school is to be merged with significantly from the transfer to better- another – they might wish to send their resourced schools. However, there are also child to the new merged school, or they indications of negative boarding school may prefer a different school. effects (see Box 4.3). • Provide additional academic support to School networking students in their new school. Evidence on school mergers in the United States School networking would involve (which is one of only a few places reorganizing classes and the structure where rigorous analysis has been of schools within the same area so they carried out to document the impact of can share resources without mergers on students’ learning consolidating schools. All stakeholders – outcomes) suggests that students including the Ministry of Education, ESA, moved tend to struggle academically local authorities, principals, teachers, and during their first year in the new school. parents – would work together to form a Given the large differences in network and design the shared education performance between Thai students, programs. This approach offers the such difficulties can be expected in advantage of being less likely to meet Thailand, as well. As such, it makes resistance compared with school mergers, sense to have well-designed programs so it is more likely to be implemented. of support ready for students during However, the networking process could their first difficult year(s).39 take a longer time to implement and could fail as more people are involved in • Carefully monitor the attendance and the process. academic performance of children involved in a merger to ensure that no Several useful examples of school harm is done and to gather information networking and shared education on the impact of the policy efforts. programs can be found within Thailand: • Expand boarding facilities at high- • In the Kangjan model in Loei province, performing secondary education four schools (with a total travel distance schools to ensure that children in of 10 kilometers between schools) remote areas can access these agreed on the education years (grades) schools – again, doing this as part of a for which they would like to be program to designate “central schools,” responsible. Students are moved to the giving certain schools a special status school that teaches their grades. Prior and more resources. to the establishment of the network, 38 Bulgaria provides painful lessons of what can happen if care is not taken to ensure that children affected by a school reorganization stay in school. Schady et al. (2009) conducted an impact evaluation that showed that children affected by a school merger were far more likely to drop out than similar children not affected. See Schady, N., L. Sondergaard, C. Bodewig, T.P. Sohnesen. 2009. “School Closures Impact on Dropout Rates: Main Results and Lessons for the Future.” World Bank and the Task Force on Impact Evaluation for more details. See also World Bank (2010): A review of the Bulgaria school autonomy reforms. 39 The evidence on U.S. school mergers is summarized in PACER (2013): “School closings policy,” Issue Brief. 38 Wanted: A Quality Education for All Strategies to improve the quality of education for all Box 4.3 China’s rural primary During the late 1990s and early 2000s, China embarked on an ambitious primary school merger school merger program program in an effort to improve the overall quality of education and address rural-urban disparities. Students in small rural village schools were transferred to new and larger centralized schools in towns and county seats, and the number of primary schools in rural China fell by 24 percent from 2001 to 2005. Taking advantage of scale economies, these new schools were equipped with better facilities and higher-quality teachers. Unlike in the small rural schools, teachers in these new larger schools were able to focus on a single grade (and in many cases, on a single course). Furthermore, while the curriculum in small rural schools was often restricted to math and Chinese language due to teacher shortages, the new central schools were able to offer a much richer curriculum. The merger program also entailed building boarding facilities since these new schools were often located far from students’ homes. These youngsters were encouraged or sometimes required to leave the comfort of their homes and care of their parents to board at schools during the week. Mo et al. (2012) conducted an impact evaluation in one of the poorest counties in Shanxi province, which is also one of the poorest in China. The study evaluated all 7th graders in all ten junior high schools in the county at the beginning of the 2009 academic year. It found that transferring from small village primary schools to more centralized town and county schools had large positive and statistically significant impacts on standardized math test results.40 However, the study also found significant negative effects when students stayed in boarding schools. Nevertheless, in comparing the transfer effect with the boarding effect, the study found that even if students boarded after transfer, they still benefited academically from transferring to new and better-resourced schools. The study suggests that extra attention be given to attenuating the negative boarding school effect so students might be able to take more advantage of the additional resources made available by the program.41 Source: Mo et al. (2012) each of the four schools taught eight in which they are enrolled.42 One major classes with three teachers per school, limitation of this model is that one of which meant that each teacher needed the schools must have the capacity to to teach at least two classes. With the accommodate all students. school network, each school now has two classes and three teachers, which Redefining schools means that each class now has at least Another option would be to explore one teacher. After the intervention, whether some schools can be student performance in the redefined to cover fewer and lower standardized O-NET exams improved grades. Some of the schools under the significantly (Box 4.4). OBEC program described above that are • In the Tripakeechan 1 model in not providing a quality secondary Janthaburi, each school teaches education could perhaps be converted pre-primary students, but the primary back into a primary school, offering instead and secondary students and teachers to pay for transportation of students in that have been merged together in area to a nearby secondary school. one school. Similarly, another option might be to convert a poorly performing primary school • In the Ban Yang Noi model in Ubon currently offering all six grade levels into a Ratchathani province, one school in new type of school only offering pre-school the area teaches core subjects to all and Grades 1 and 2 (and offer to transport students from all small schools in the children from Grade 3 onward to better area three days a week. For the rest of schools). The advantage of this approach the week, students study at the schools is that every community would keep some 40 The test was a 30-minute standardized math test administered by the researchers themselves to ensure that there was no coaching for the test before the survey. 41 As mentioned in Mo et al. (2012), a number of studies found that when boarding schools in the program were poorly managed, students performed worse in school. Some studies also found that poor nutrition and health in some boarding facilities negatively affected educational performance (Luo et al., 2009 and Shi, 2004). 42 OBEC and ONESQA (2012). Wanted: A Quality Education for All 39 Strategies to improve the quality of education for all Box 4.4 The Kangjan case study Compared to other merger/networking models in Thailand, the Kangjan networking model is of school networking relatively well-documented. Tables B4.1 and B4.2 compare school conditions before and after the networking in terms of number of students, classes, and teachers in each school; number of students who had to travel to other schools after the intervention; and average O-NET scores by subject. The networking started in 2011 with four schools in ESA 1 in Loei province. The four schools, which are located within 10 kilometers of each other, are (1) Bann Namo, (2) Bann Pak Mung Hauy Tab Chang (Pak Mung), (3) Bann Hat Kampee, and (4) Bann Cokwao. Prior to 2011, all four schools taught pre-primary 1 to primary 6. With the same total number of teachers and students as before, the school networking organized classes much more efficiently. Prior to the establishment of the network, each school had three teachers (excluding the principal) and taught eight classes, so they were critically short of teachers. After the networking intervention, the number of classrooms in each school was reduced to two so each class/grade had 1.5 teachers on average. Table B4.1 also shows that 180 out of 247 students needed to move to other schools after the networking. Table B4.1 Pre- and post-networking school conditions Number of students Number of students in each grade transferred Number Number to other of of Schools K1 K2 P1 P2 P3 P4 P5 P6 Total schools classes teachers Before Cokwao 3 4 4 1 4 4 7 6 33 25 8 3 network Hat Kampee 11 10 10 8 9 8 16 6 78 56 8 3 Pak Mung 10 8 6 11 6 9 5 6 61 42 8 3 Namo 8 8 10 10 10 9 12 8 75 57 8 3 Total 32 30 30 30 29 30 40 26 247 180 32 12 After Cokwao 29 30 59 2 3 network Hat Kampee 40 26 66 2 3 Pak Mung 30 30 60 2 3 Namo 32 30 62 2 3 Table B4.2 compares the school average O-NET test results in eight core subjects before and after the networking. The table also shows the corresponding national average scores in 2010 and 2011. As shown in the last column, test scores for students in the network improved substantially more than the national average in all subjects except physical education over the 2010-2011 period. Table B4.2 Impact on student performance in O-NET exams Kangjan National Changes in Kangjan model Changes Changes minus changes Year Year from 2010 Year Year from 2010 in the national Subject 2010 2011 to 2011 2010 2011 to 2011 average Thai 29.8 49.2 19.4 31.2 50.0 18.8 0.6 Mathematics 26.5 53.9 27.4 34.9 52.4 17.6 9.9 Science 38.8 43.6 4.8 41.6 40.8 -0.7 5.5 Social science 42.1 47.9 5.8 47.1 52.2 5.2 0.7 Physical education 51.5 51.9 0.5 54.3 58.9 4.6 -4.1 Arts 33.1 44.8 11.7 41.1 46.8 5.7 6.0 Vocational skills 46.6 53.9 7.3 52.5 55.4 2.9 4.4 English 13.9 41.5 27.6 21.0 38.4 17.4 10.2 Source: TDRI (2015). type of school while initiating a dialogue 2 Increasing/improving financing about how to provide a quality education in the grades where better teachers are While increasing funding for small needed. Ideally, this option would be schools could help reduce disparities, pursued as part of a “networking strategy” the key question is whether it is possible to ensure that each grade level is filled. to provide the level of resources necessary to bring learning standards to Bangkok levels. Small schools in 40 Wanted: A Quality Education for All Strategies to improve the quality of education for all Thailand are so underfunded that per that so many quality teachers could be student spending for small schools would hired, trained, and would be willing to go to need to be increased substantially. For small, remote village schools. example, schools in Mae Hong Son would require at least43 a 64 percent increase in Another financing option that could the average per-student amount in order incentivize schools to become larger to provide students in the province with and more efficient would be to finance adequate resources to attain the same schools based on the number of level of educational outcomes as students students they have enrolled rather than in Bangkok.44 on the inputs they employ. By tying money to the number of students enrolled Calculating the estimated change in the in an ESA (as opposed to how many average personnel salary per student45 schools there are and/or how many necessary for each province to bring teachers are employed), OBEC could their school standards up to Bangkok incentivize local communities to seek ways standards (in terms of teacher quality to create larger schools. This type of and number of teachers per classroom46) “demand side financing” has been the reveals that total recurrent spending on most widely adopted policy change in OBEC schools would need to increase Eastern Europe, where student numbers by a massive 31 percent. The average have been plummeting for more than two per-student public subsidy would rise from decades.47 The funding formula48 could be THB 31,475 to THB 41,250 per year. based on an “adequacy concept,” meaning Overall, the share of government that the per-student amount allocated expenditure on education would increase should in theory be adequate for all by nearly 5 percentage points, from 24.0 to schools to reach a pre-specified “student 28.8 percent. performance standard” (but not so generous that it fails to provide the proper Even if financing could be increased incentive49). This mechanism is more significantly, the impact would depend equitable since it creates a direct link on whether enough qualified teachers between funding and the purpose of the could be mobilized, particularly for activity funded. However, shifting to such village schools. Providing all schools with financing would have to be done gradually, the level of teacher quality and number since all new formulas would end up with of teachers per classroom necessary to “losers” (i.e. areas and schools that would achieve Bangkok-level learning standards get less under the proposed formula than would require the monumental task of what they are getting with the current recruiting, training, and deploying 164,000 system) and “winners,” and the only way new teachers. This represents an increase for “losers” to be able to adjust would of over 40 percent of the teaching force be to downsize. Thus, schools and ESAs in Thailand in 2014. Therefore, this is not would need time to prepare for a just a fiscal challenge – it is highly unlikely new formula. 43 “Atleast” because the estimate has not taken into account the required incentives (such as salary increases) needed to attract teachers and other education personnel to rural schools. 44 This is estimated using an educational cost function-based per-student funding formula (see Box A5.1 in Annex 5) which takes into account differences in student performance, school and class sizes, and the school student-body characteristics (poor students, students with disability, etc.). For example, students in small remote schools are poorer and less-prepared academically, so it will take more resources to bring those schools to Bangkok-level learning standards. 45 Only the salary portion of the per-student subsidy was calculated since the non-salary portion does not vary with teacher quality or the intensity of staffing (see Section A5.2 in Technical Appendix to Annex 5 for details of the non-salary items). 46 Aninput-based approach was employed to approximate the increase in the average salary per student amount. Specifically, we used the teacher salary regression equation (Table A5.2 in Technical Appendix to Annex 5) to estimate the change in the provincial average salary per student if schools in the other provinces were to have the same proportion of teachers with higher than bachelor’s degree qualification, teachers with the same average experience, and classrooms staffed with the same number of teachers as the average school in Bangkok. 47 For example, see Juan Diego Alonso and Alonso Sanchez (2011): Reforming education finance in transition countries” or Sondergaard et al (2011). 48 An example application of the funding formula is given in Annex 5 (Table 5.1). Interested readers are referred to Technical Appendix to Annex 5 for details of the computation of the “cost function-based” per-student funding formula for Thailand. 49 See discussion in Box A5.1 in Annex 5. Wanted: A Quality Education for All 41 Strategies to improve the quality of education for all 3 Improving teaching resources 4 Increasing awareness and for small and remote schools understanding of the small school challenge In order to provide better education to students in severely under-resourced Decisions on the approach to reducing schools, OBEC could introduce disparities in education could benefit measures to improve training in multi- from further research to understand the grade teaching. Schools with severe small school challenge, particularly at teacher shortages could then consolidate the primary school level. Much of the classrooms and provide multi-grade analysis conducted for this report utilized education more effectively. Colombia’s the available information for a small sample “Escuela Nueva” (New School), which is from PISA survey data, namely widely recognized as a successful example 15-year-olds in secondary schools. of rural primary educational innovation, Because many students do not even make accepts multi-grade teaching as an it to the secondary level, it is important to unavoidable condition in small schools of know what is happening at the primary rural areas. It encourages the development level. One possible starting point would be of special materials and teaching methods to use data from the Trends in International for multi-grade teaching. Notably, the Mathematics and Science Study (TIMSS) academic achievement of students in results for 4th graders to examine whether Escuela Nueva is even higher than in urban the performance gap between village schools (UNESCO, 2004). After officials schools and large city schools is already from the Ministry of Education in Vietnam there at grade 4 or whether it only appears visited Colombia, the Ministry launched later in the school cycle. As noted in a version of Escuela Nueva aimed at Chapter 2, another area for further work improving the quality of rural schools is to understand the factors that have in Vietnam. contributed to the impressive gains made among small town schools, which could OBEC could also explore options for help inform policies aimed at the providing stronger incentives for quality village level. teachers to be deployed in small, remote schools. As shown in the health sector The proposed reform options to improve (and documented in Putthasri et al., 2013), education system performance and taking an evidence-based approach is address educational attainment more useful than trying to find the perfect inequality could be piloted first in some solution from international experience that areas. As discussed above, a few will work in the Thai context. In particular, examples of school mergers or school such an evidence-based approach networking can be found in Thailand. would entail: However, it appears that none of them have been subjected to rigorous 1) Acknowledging that there is a evaluation. The same can be said of problem; financing and school-based management 2) Identifying ways to quantify the reform experimentation. International problem in order to establish a experience suggests that it would be a baseline and track progress; good idea to conduct well-planned pilot studies on the reforms whenever possible 3) Experimenting with different so policymakers and local communities programs to tackle the problem – for can better understand their impacts on example, in health, there have been students, teachers, schools, communities, experiments with the size of the and other stakeholders. Well-informed “financial penalty” for shortening decisions could then be made to scale up mandatory service and with the the options that have demonstrated way/location where doctors did their favorable impacts. clinical training; 4) Evaluating which experiments have Thailand’s leaders have the opportunity worked; and to champion education reform – including effective ways to address the 5) Building on the experiments that challenges of small schools and improve worked. the quality of learning. A lively public debate can make much difference. 42 Wanted: A Quality Education for All Strategies to improve the quality of education for all The little public debate that has occurred Such public debate on education could on small schools usually revolves around benefit from the type of institutions drama created around one of the few which have been guiding some of the school closures that has taken place in major reforms undertaken in the health past years. Furthermore, in policy cycles, sector in the past 25 years. Specifically, the debate tends to focus on what seems the Health System Research Institute impossible to fix—the small, isolated (HSRI) and the Thailand Health Promotion schools and their unique, large challenges Foundation (Thaihealth) were instrumental – as opposed to the small schools that are agencies to create and sustain a public not isolated. Greater public awareness and debate on the need to reform, and discussion is needed on the poor quality of underpin reform proposals with relevant Thailand’s small schools and on the evidence. On the education front, the opportunities for offering a better Quality Learning Foundation (QLF) has the education to the hundreds of thousands of potential to become an agency playing the children currently enrolled in such schools. same role as the Thailand Health It is hoped that this report can help Promotion Foundation but it has yet to contribute to such a dialogue. secure a reliable source of funding.50 50 QLF was established in 2010, through the Prime Minister’s Office Regulation on Learning Society BE2553. QLF inherits the system approach from the HSRI and the social mobilisation approach to mobilise multi-sectorial and multidisciplinary partners from both domestic and international level from the Thaiheatlh which has been successful in reforming the health sector in the past 25 years. In the past 5 years, QLF has produced a number of promising policy system researches and evidence-based education reform proposals which have potential to scale-up into a national education reform policy in the near future. For instance, the National Education Accounts (NEA), the Area-based education reform programs and the School-based information system for education decentralisation. As of June 2015, a drafted QLF law is waiting for the cabinet consideration to be passed by the National Assembly. Wanted: A Quality Education for All 43 References Aigner, D. J., C. A. K. Lovell, and P. Schmidt (1977): “Formulation and Estimation of Stochastic Frontier Production Function Models,” Journal of Econometrics, 6, 21-37. Alonso, J. D. and A. Sanchez (2011): “Reforming education finance in transition countries: Six Case Studies in Per Capita Financing Systems,” The World Bank, June 2011. Arcia, G. and H. A. Patrinos (2013): “Assessment of Policy Intent and Policy Implementation of School Autonomy and Accountability in Thailand,” mimeo, The World Bank, March 2013. Alvarez, J. V. Garcia Moreno and H. A. Patrinos (2007): “Institutional Effects as Determinants of Learning Outcomes: Exploring State Variations in Mexico,” Well-Being and Social Policy, 3(1), 47-68. Barrera, F., T. Fasih, and H. A. Patrinos, with L. Santibáñez (2009): “Decentralized Decision-Making in Schools. The theory and evidence on School-based management,” Washington DC, The World Bank. Bhumirat, C., S. Kidchanapanish, P. Arunrungrueng, and R. Shinatrakool (1987): “Research and Evaluation on the Quality of Primary Education in Thailand,” Paper funded through the BRIDGES Project, A Cooperative Agreement between Harvard University and S&T/Ed. United States Agency for International Development (DPE 5824-A00-5076-00). Blinder, A. S. (1973). “Wage Discrimination: Reduced Form and Structural Estimates,” The Journal of Human Resources, 8(4), 436-455. Burns, B., D. Filmer, and H. A. Patrinos (2011): “Making Schools Work: New Evidence on Accountability Reforms,” Washington DC, The World Bank. Carnoy, M. and S. Loeb (2002): “Does external accountability affect student outcomes? A cross-state analysis,” Educational Evaluation and Policy Analysis, 24(4), 305-331. Chakraborty, K and J. Poggio (2008): “Efficiency and Equity in School Funding: A Case Study for Kansas,” International Advances in Economic Research, International Atlantic Economic Society, 14(2), 228-241, May. Costrell, R., E. A. Hanushek, and S. Loeb (2008): “What Do Cost Functions Tell Us About the Cost of an Adequate Education?” Peabody Journal of Education, 83, 198-223. DiNardo, J., N. M. Fortin, and T. Lemieux (1996): “Labor Market Institutions and the Distribution of Wages, 1973-1992: A Semiparametric Approach,” Econometrica, 64(5), 1001-1044. Duncombe, W., J. Ruggiero, and J. Yinger (1996): “Alternative approaches to measuring the cost of education,” In H. L. Ladd (Ed.), Holding schools accountable: Performance-based reform in education, 327-356. Washington, DC: Brookings. Duncombe, W. and J. Yinger (2005): “Estimating the Costs of Meeting Student Performance Outcomes Adopted by the Kansas State Board of Education,” Report Prepared for The Kansas Legislative Division of Post Audit. Duncombe, W. (2007): “Estimating the cost of meeting student performance standards in the St. Louis public schools,” Syracuse, NY: Syracuse University, January. Firpo, S., N. M. Fortin, and T. Lemieux (2007): “Decomposing Wage Distributions using Recentered Influence Function Regressions,” mimeo, University of British Columbia. Firpo, S., N. M. Fortin, and T. Lemieux (2009): “Unconditional Quantile Regressions,” Econometrica, 77(3), 953-973. Fry, G. (1999): “Teaching personnel strategy in Thailand: A review and recommendations,” Report prepared for UNESCO-Bangkok, Bangkok: UNESCO. 44 Wanted: A Quality Education for All References Gertler P., H. A. Patrinos, and M, Rubio-Codina (2008): “Empowering Parents To Improve Education: Evidence From Rural Mexico,” Impact Evaluation Series No. 4. World Bank Policy Research Working Paper 3935, Washington DC. Greene, W.H., (2003): “Simulated Likelihood Estimation of the Normal-Gamma Stochastic Frontier Function,” Journal of Productivity Analysis, 19, 179-190. Gronberg, T. J., D. W. Jansen, L. L. Taylor, and T. K. Booker (2004): “School Outcomes and School Costs: The Cost Function Approach,” Joint Select Committee on Public School Finance, Austin, TX, http://bush.tamu.edu/research/faculty/TXSchoolFinance/. Gronberg, T. J., D. W. Jansen, L. L. Taylor, and T. K. Booker (2005): “School Outcomes and School Costs: A Technical Supplement,” Joint Select Committee on Public School Finance, Austin, TX, http://bush.tamu.edu/research/faculty/TXSchoolFinance/. Guan, Z., S. C. Kumbhakar, R. J. Myers, and A. O. Lansink (2009): “Measuring Excess Capital Capacity in Agricultural Production,” American Journal of Agricultural Economics, 91, 765-776. Hampel, F. R. (1968): “Contribution to the Theory of Robust Estimation,” Ph.D. Thesis, University of California at Berkeley. Hampel, F. R. (1974): “The Influence Curve and Its Role in Robust Estimation,” Journal of the American Statistical Association, 69(346): 383-393. Hanushek, E. A., J. F. Kain, S. G. Rivkin (1998): “Teachers, Schools, and Academic Achievement.” NBER Working Paper 6691. Hanushek, E. A. and M. E. Raymond (2005): “Does school accountability lead to improved student performance?” Journal of Policy Analysis and Management, John Wiley and Sons, Ltd., vol. 24(2), pages 297-327. Hanushek, E. A., L. Susanne, and L. Woessmann (2013): “Does school autonomy make sense everywhere? Panel estimates from PISA,” Journal of Development Economics, Elsevier, 104(C), 212-232. Imazeki J. (2008): “Assessing the Costs of Adequacy in California Public Schools: A Cost Function Approach,” Education Finance and Policy, MIT Press, 3(1), 90-108, January. International Association for the Evaluation of Educational Achievement (IEA) (2009): “Do Countries Paying Teachers Higher Relative Salaries Have Higher Student Mathematics Achievement?” Teacher Education and Development Study in Mathematics (TEDS-M). Karakaplan M. U. and L. Kutlu (2013): “Handling Endogeneity in Stochastic Frontier Analysis: A Solution to Endogenous Education Cost Frontier Models,” mimeo, Georgia Institute of Technology. Kumbhakar, S. C., and C. K. Lovell (2000): “Stochastic Frontier Analysis,” Cambridge University Press. Lathapipat, D. (forthcoming): “Inequality in Education and Wages,” in Phongpaichit, P. and Baker, C. ed, Unequal Thailand. Causes, Consequences, Reforms. NUS Press, Singapore. Lathapipat, D. (2015): “School-Level Governance: Decentralised Decision-Making for Improved Learning Outcomes,” Unpublished mimeo. Lindert K., E. Skoufias, J. Shapiro (2006): “Redistributing Income to the Poor and the Rich: Public Transfers in Latin America and the Caribbean,” World Bank, Washington, DC. Loera, A. (2005): “Cambios en las Escuelas que Participan en el PEC 2001-2004. Evaluacion Cualitativa del Programa Escuelas de Calidad,” Reportes Nacionales, Fase 3. Mexico City, Mexico Lopez-Calva, L. F. and L. D. Espinosa (2006): “Efectos diferenciales de los programas compensatorious del CONAFE en el aprovechamiento escolar,” In A.B. Lerín, ed., Efectos del Impulso a la Participación de los Padres de Familia en la Escuela. Mexico: Consejo Nacional de Fomento Educativo. Wanted: A Quality Education for All 45 References Luo, R., Y. Shi, L. Zhang, C. Liu, S. Rozelle, and B. Sharbono (2009): “Malnutrition in China’s rural boarding schools: the case of primary schools in Shanxi Province,” Asia Pacific Journal of Education 29(4), 481-501. Meeusen, W., and J. van den Broeck (1977): “Efficiency Estimation from Cobb-Douglas Production Function with Composed Error,” International Economic Review 8, 435-444. Ministry of Education and Science (2008): National Report on the Development of Education in Bulgaria (available at http://www.ibe.unesco.org/National_Reports/ICE_2008/bulgaria_NR08.pdf). Mises, R. V. (1947): “On the Asymptotic Distribution of Differentiable Statistical Functions,” The Annals of Mathematical Statistics, 18(3), 309-348. Mo, D., H. Yi, L. Zhang; Y. Shi, S. Rozelle, and A. Medina (2012): “Transfer Paths and Academic Performance: The Primary School Merger Program in China,” International Journal of Educational Development, 32(3), 423-431. National Education Act B.E. 2542 (1999) and Amendments (Second National Education Act B.E. 2545 (2002)), Office of the National Education Commission, Office of the Prime Minister, Kingdom of Thailand. Nye, B., S. Konstantopoulos, and L.V. Hedges (2004): “How Large are Teacher Effects?” Educational Evaluation and Policy Analysis, 26(3), 237-257. Oaxaca, R. (1973): “Male-Female Wage Differentials in Urban Labor Markets,” International Economic Review, 14: 693-709. Office of the Education Council. Education in Thailand 2004. Bangkok, 2004. Office of the Education Council. Education in Thailand 2005-2006. Bangkok, 2006. Office of the Education Council. Education in Thailand 2007. Bangkok, 2008. Office of the National Education Commission. Education in Thailand 1999. Bangkok, 1999. Office of the Permanent Secretary. Educational Statistics in Brief 2003. Ministry of Education, Thailand. Office of the Permanent Secretary. Educational Statistics in Brief 2013. Ministry of Education, Thailand. Organization for Economic Cooperation and Development (OECD) (2014). PISA 2012 Technical Report. Paris: OECD. PACER (2013): “School closings policy,” Issue Brief. Patrinos, H. A. and R. Kagia (2007): “Maximizing the Performance of education Systems – The Case of Teacher Absenteeism,” Chapter 2 within “The Many Faces of Corruption – Tracking Vulnerabilities at the Sector Level,” Edited by Campos, J. Edgardo and Pradhan, Sanjay, The World Bank, Washington D.C. 2007. Patrinos, H. A. and T. Fasih (2010): “A Review of the Bulgaria School Autonomy Reforms,” World Bank Report No. 54890-BG, Washington, DC. Patrinos, H. A. (2011): “School-Based Management,” In Bruns, B., D. Filmer, H. A. Patrinos (Ed.) “Making Schools Work: New Evidence on Accountability Reforms,” Washington DC: The World Bank. Putthasri W., R. Suphanchaimat, T. Topothai, T. Wisaijohn, N. Thammatacharee, and V. Tangcharoensathien (2013): “Thailand special recruitment track of medical students: a series of annual cross-sectional surveys on the new graduates between 2010 and 2012,” Human resources for health 11(1): 47. doi:10.1186/1478-4491-11-47. Rivkin, S. G., E. A. Hanushek, J. F. Kain (2005): “Teachers, Schools, and Academic Achievement,” Econometrica, 73(2), 417-458. Schady, N., L. Sondergaard, C. Bodewig, T. P. Sohnesen (2009): “School Closures Impact on Dropout Rates: Main Results and Lessons for the Future,” World Bank and the Task Force on Impact Evaluation. 46 Wanted: A Quality Education for All References Shapiro, J. and J. M. Trevino (2004): “Compensatory Education for Disadvantaged Mexican Students: An Impact Evaluation Using Propensity Score Matching,” World Bank Policy Research Working Paper 3334, June 2004. Shi, R. (2004): “Experience drawn from research on school consolidation in foreign countries,” Comparative Education Research 12. Skoufias E. and J. Shapiro (2006): “Evaluating the Impact of Mexico’s Quality Schools Program: The Pitfalls of Using Nonexperimental Data,” World Bank Policy Research Working Paper No. 4036. World Bank, Washington, DC. Sondergaard L., M. Murthi (2011): “Skills, Not Just Diplomas, Directions in Development,” World Bank. Stevenson, R. (1980): “Likelihood Functions for Generalized Stochastic Frontier Functions,” Journal of Econometrics, 13, 57-66. Thailand Development Research Institute (2013): “Strategies for Reforming Basic Education,” Thailand, Research Paper for the Office of The Basic Education Commission. Thailand Development Research Institute (2015): “Grouping Thailand’s Schools into Four Categories,” May 2015. UNESCO (2004): “Improving the quality of primary school through distance education,” Report prepared by the International Research Foundation for Open Learning for the EFA Monitoring Team, UNESCO, Paris. April 2004. Varavarn, K. (2006): “Thailand’s Experience in Addressing the Challenges of Secondary Education,” Ministry of Education, Thailand. Vegas, E. (2005): “Incentives to Improve Teaching: Lessons from Latin America,” Washington, DC: World Bank. Winkler, D. R. (RTI) (2004): “Increasing Accountability in Decentralized Education in Paraná State, Brazil,” Unpublished paper, October 2004. World Bank (1998): “Romania Public Expenditure Review,” Report No. 17743-RO. World Bank (2010): “A Review of the Bulgaria School Autonomy Reforms.” World Bank (2011): “Thailand Public Finance Management Report Discussion Paper 5: Analysis of Efficiency of Education.” World Bank (2012): “Bulgaria: Public Expenditure for Growth and Competitiveness,” Report No. 62774-BG. World Bank (2012): “Project Appraisal Document for Moldova Education Reform Project.” World Bank (2013): “Serbia: Municipal Finance and Expenditure Review,” Report No. 76855-YF. Wanted: A Quality Education for All 47 Annexes Annex A2 A2.1. Decomposing the Change in the PISA Test Score Over Time Theoretical Framework The focus of this part of the study is on analyzing changes in the PISA test score distribution between two points in time, dated = 0,1. The method used here is essentially the first stage of the two-stage decomposition procedure proposed by Firpo, Fortin, and Lemieux (2007) – FFL 2007 from hereon. Assume that the test score function1 depends on some observed and unobserved attributes ( , ) of individual student indexed by ∈ {1, … }, where = 0 + 1 is the total number of the combined observations for both time periods, ∈ ⊂ ℝ , and ∈ ℝ. The test score function can thus be expressed as: = ( , ) for = 0,1 and = 1, … , (1) Define the observed test score for student as = 1 + 0 (1 − ), where = 1 if the outcome variable is observed in time period 1, and = 0 otherwise. If we view as a treatment indicator, we can only observe 0 | = 0 for untreated units, and 1 | = 1 for treated units, but 0 (∙) 1 (∙) not both. Their corresponding distribution functions are denoted by , and respectively. In effect, we are faced with a missing data problem. However, it is possible to conceive of the (∙) counterfactual quantity 0 | = 1~ , whose identification requires further assumptions stated (∙) below. In words, the counterfactual test score distribution function is the distribution that would have prevailed under the test score function of year 0, but with the observed and unobserved student attributes ( , ) jointly distributed as in year 1. Denote by the test score at a particular quantile , whose change over time we seek to decompose. The “overall change” in , ∆ , from date 0 to 1 is divided into the “school quality effect”, ∆ , and the “student effect”, ∆ : 1) )) ) 0 )) ∆ = ∆ + ∆ = ( ( − ( + ( ( − ( (2) Assume that (i) the distribution of the unobserved characteristics, , is independent of after conditioning on observed covariates , and (ii) < Pr( = 1| = ) < 1 − for all ∈ and for some > 0. Assumption (i) is called the “ignorability of treatment” assumption and is written as ⊥ | = for all ∈ . Assumption (ii) is called the “overlapping support” assumption. Together, these two assumptions are called the “strongly ignorable treatment assignment” (∙) assumptions, and are sufficient for the identification of , and ensure that the student effect ∆ only reflects changes in the distribution of observable student attributes (Theorem 2 in FFL (2007)). 1 In this study, the test score function is viewed as the “Quality” or effectiveness if the Thai education system in transforming its student characteristics (both observed and unobserved) into learning outcome. To divide the overall change in test score over time into the school quality and student characteristic effects, we employ a reweighting procedure proposed by DiNardo, Fortin, and Lemieux (1996) – DFL (1996) from hereon. Specifically, the test score distribution functions for each time period is non-parametrically identified from observed data, and their empirical distribution counterparts based on a random sample {1 , … , } of size are given by: 1 ̂ () = ∑∈ 1( ≤ ), for = 0,1 (3) where = {: = } is the index set for observations at time and 1(∙) is an indicator function () equal to unity if the expression in the parentheses is true. Note also that can be written as: () () = ∫ | (|) (4) () where ⊂ ℝ is the support of and is the conditional distribution of | = . Assuming the “strongly ignorable treatment assignment” assumptions discussed above and using equation (4), the counterfactual distribution function is well-defined and can be constructed as follows: () 0 1 () = ∫ | (|) 0 1 () 0 () = ∫ | (|) 0 () 0 0 () = ∫ | (|) () (5) where () is the reweighting function. Applying Bayes’ rule to the function in the same fashion as DFL (1996), the “inverse probability weighting function” or IPW is expressed as (Theorem 1 in FFL (2007)): (=0) Pr(=1|=)Pr⁡ () 1− () = Pr(=0|=)Pr⁡(=1) = (1−()) ( ) (6) where () = Pr( = 1| = ) is the propensity score and can be estimated using a flexible logit ( = 1). The empirical IPW is normalized to sum to unity for convenience: regression, and = Pr⁡ ̂ ( ) ̂ ( ) = ∑ ̂ , for ∈ 0 (7) ∈ ( ) 0 The empirical counterfactual test score distribution is thus given by: ̂ () ̂ ( )1( ≤ ) = ∑∈0 (8) ̂ Equations (3) and (8) allow us to estimate the sample quantiles of interest; ( 0 ̂ ), ( 1 ), ̂ and ( ), which can be used to compute the estimated school quality, ∆ , and student effects, ∆ , as per equation (2). Notice that the methodology discussed in this section is effectively a generalization of the classical Oaxaca-Blinder decomposition to any marginal quantile of interest. Therefore, if we apply the empirical IPW technique to estimate the means of the various distributions, then the resulting decomposition would be equivalent to that obtained using the classical method. Decomposing the Change in PISA Reading Score from 2003 to 2012 The estimation results from the (flexible) logit regression model are presented in Table A2.1 at the end of this section. The dependent variable is the indicator function which is equal to unity if a student is observed in year 2012 and zero otherwise, while the observable student characteristics (the right hand side variables) used in the analysis include gender, grade level, and the PISA index of economic, social, and cultural status (ESCS). Using the regression coefficients, we are able to estimate the propensity scores and compute the normalized empirical IPW for each individual observation as per equation (7). First, we demonstrate the IPW technique by applying it to the mean of the test score distribution for 15 year-old Thai students. Specifically, the counterfactual mean test score is computed as follows: ̅ = ∑∈ ̂ ( ) 2003 This equation is analogous to equation (8) and represents the average test score that would have prevailed under the test score function of year 2003 but with the observed and unobserved student attributes ( , ) jointly distributed as in 2012 (sampling weights are also used in the analysis, but are omitted here to simplify notation). The mean test score for each time period is simply ̅ = 1/ ⁡∑∈ , for = 2003,2012. Figure 2.11. Contributions of Changes in Quality and in Student Characteristics to Changes in PISA Reading Scores in Thailand from 2003 to 2012, by Location 490 480 Quality 17.0 470 Student 4.4 460 2003 12.4 450 8.6 440 15.8 430 6.6 24.2 2.6 420 6.0 461.5 410 8.1 440.7 10.1 424.4 400 419.6 6.1 390 405.3 393.7 380 Thailand Village Small town Town City Large city Source: World Bank Staff Calculations based on OECD PISA 2003 and 2012 Analogous to equation (2), the overall change in the average PISA reading scores for Thai students from 2003 to 2012 can be decomposed into the portion attributable to changes in the background characteristics of the student population and the remaining “unexplained” portion, which we interpret as representing improvements over time in the “quality” or effectiveness of the Thai education system in transforming student characteristics into learning outcomes. Specifically, ̅ ̅ ̅ ̅ 1 ̅ ̅ ̅ 0 ∆ = ∆ + ∆ = ( − ) + ( − ) (9) The decomposition results for Thai students on the whole are shown on the left-most bar in Figure 2.11 in Chapter 2, which is reproduced here for ease of exposition. It is estimated that improvements in the observable characteristics of the student population accounted for 6 points of the 21.8-point increase in the overall PISA reading scores for Thailand from 2003 to 2012. This means that the remaining 15.8-point increase can be attributed to an apparent improvement in educational “quality.” The decomposition analysis can also be carried out for subgroups of the Thai student population. Figure 2.11 presents the decomposition results for students attending schools in five distinct locations of the country: villages, small towns, towns, cities, and large cities (see Table A2.2 for the total number of 15-year-old students attending schools by location as well as their average reading test scores in each year). The disaggregated analysis is simply done by applying equation (9) to the student population subgroup of interest. Much richer analysis can be done if we look beyond the effects at the means. As mentioned earlier, the FFL (2007) decomposition framework allows us to generalize the classical Oaxaca-Blinder decomposition to any marginal quantile of interest. We apply the reweighting procedure to analyze changes at different quantiles of the test score distribution. The resulting distributions of PISA Reading Scores in 2003, 2012, and the counterfactual distributions (produced using equations (4) and (5)) for Thailand and for subgroups of the student population in five different areas are shown in Figure A2.1. Also drawn on each chart in the Figure is a vertical line at the PISA reading score of 407, which is defined as the threshold level of functional literacy. The decomposition of the overall changes in the test score from 2003 to 2012 into the quality and the student effects across quantiles of the test score distribution are shown graphically in Figure 2.9 in Chapter 2. Notice that for ease of exposition, we have included only the “Overall effect” and the “Quality effect” line graphs in these charts. From equation (2), we can easily see that the gaps between these two line graphs necessarily represent the effects on test scores of changes in the observed student characteristics over time. Figure A2.1. Distributions of PISA Reading Scores in 2003, 2012, and the Counterfactual Distributions (the vertical lines represent the threshold level of functional literacy) .015 .006 .01 .004 .005 .002 0 0 200 300 400 500 600 700 200 300 400 500 600 700 PISA reading score PISA reading score Thailand 2012 Thailand 2003 village 2012 village 2003 counterfactual counterfactual .006 .006 .004 .004 .002 .002 0 0 200 300 400 500 600 700 200 300 400 500 600 700 PISA reading score PISA reading score small town 2012 small town 2003 town 2012 town 2003 counterfactual counterfactual .006 .005 .004 .004 .003 .002 .002 .001 0 0 200 300 400 500 600 700 200 300 400 500 600 700 PISA reading score PISA reading score city 2012 city 2003 large city 2012 large city 2003 counterfactual large counterfactual Source: World Bank Staff Calculations based on OECD PISA 2003 and 2012 Table A2.1. Logit Regression Results Variables Dependent variable: Year 2012 Female -0.050 (0.045) Grade level: (reference "G7") Grade 8 1.895 (1.346) Grade 9 2.878** (1.286) Grade 10 4.002*** (1.287) Grade 11 4.633*** (1.296) ESCS 0.120 (0.073) ESCS2 -0.214*** (0.045) ESCS3 -0.094** (0.041) ESCS4 0.020 (0.014) Location: (reference "Town") Village -0.296*** (0.070) Small town 0.452*** (0.062) City 0.804*** (0.063) Large city -0.552*** (0.085) Intercept -3.706*** (1.286) Observations 11,776 Pseudo R-squared 0.121 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table A2.2. Number of 15-Year-Old Students Attending Schools by Location and Their Average PISA Reading Test Scores 2003 2012 PISA reading # Students Student share PISA reading # Students Student share Village 394 170,356 27% 410 110,148 16% Small town 405 112,541 18% 438 184,211 26% Town 424 185,551 29% 433 182,739 26% City 441 78,432 12% 462 173,700 25% Large city 461 83,602 13% 483 49,959 7% Thailand 420 630,481 100% 441 700,757 100% Source: OECD PISA 2003 and 2012 Annex A3 Table A3.1. Key Characteristics of OBEC Schools – by School Size Category Primary Secondary All Per- Per- Per- School Size Average Total # student Average Total # student Average Total # student Category class size Schools subsidy class size Schools subsidy class size Schools subsidy Less than 50 4.6 3,785 53,504 4.4 10 97,653 4.6 3,795 53,635 50 to 69 7.6 3,296 43,144 5.8 30 63,433 7.6 3,326 43,330 70 to 89 10.1 3,260 39,673 8.1 72 58,635 10.0 3,332 40,089 90 to 119 13.0 3,489 36,768 10.4 217 51,818 12.8 3,706 37,672 120 to 149 16.6 2,878 34,208 12.7 584 45,828 15.8 3,462 36,199 150 to 199 21.2 2,175 30,400 16.6 1,284 38,486 19.2 3,459 33,454 200 to 279 26.0 1,337 27,329 21.6 2,183 33,230 23.1 3,520 31,026 280 to 499 26.3 874 26,368 26.6 2,720 29,383 26.5 3,594 28,656 500 to 749 30.5 324 25,454 29.7 1,032 28,109 29.9 1,356 27,474 750 to 1149 33.1 166 25,634 34.1 493 28,803 33.9 659 28,008 1150 to 1999 36.6 114 25,444 37.8 386 28,960 37.5 500 28,162 2000 or above 41.9 50 25,275 43.1 434 31,052 43.0 484 30,544 Overall 15.6 21,748 32,489 29.1 9,445 30,843 21.8 31,193 31,476 Primary Secondary All Student- # Student- # Student- # School Size teacher Teacher Total # teacher Teacher Total # teacher Teacher Total # Category ratio per class Classes ratio per class Classes ratio per class Classes Less than 50 10.3 0.52 28,010 6.3 0.95 85 10.3 0.52 28,095 50 to 69 12.6 0.65 25,926 7.8 0.82 313 12.6 0.65 26,239 70 to 89 13.8 0.78 25,740 9.1 0.99 728 13.7 0.78 26,468 90 to 119 15.0 0.92 27,717 9.8 1.14 2,206 14.7 0.93 29,923 120 to 149 15.9 1.09 22,953 10.8 1.23 6,192 15.0 1.12 29,145 150 to 199 18.7 1.18 17,535 13.6 1.26 13,683 16.8 1.22 31,218 200 to 279 22.3 1.23 11,915 17.1 1.32 24,009 19.0 1.29 35,924 280 to 499 23.6 1.17 11,968 21.0 1.33 37,304 21.7 1.29 49,272 500 to 749 26.0 1.25 6,327 23.4 1.34 20,613 24.0 1.32 26,940 750 to 1149 25.9 1.32 4,579 24.8 1.46 13,313 25.1 1.42 17,892 1150 to 1999 28.1 1.38 4,645 26.6 1.50 15,325 26.9 1.47 19,970 2000 or above 27.8 1.55 3,160 26.3 1.69 28,952 26.4 1.67 32,112 Overall 19.4 0.93 190,475 22.7 1.40 162,723 21.4 1.15 353,198 Source: World Bank Staff Calculations based on OBEC 10 15 20 25 30 10 15 20 25 30 35 40 45 50 0 5 0 5 10% 12% 14% 16% 18% 20% 22% 0% 2% 4% 6% 8% 10 - Bangkok Metropolis 10 - Bangkok Metropolis 12 - Nonthaburi 10 - Bangkok Metropolis 12 - Nonthaburi 83 - Phuket 12 - Nonthaburi 83 - Phuket 75 - Samut songkhram 83 - Phuket 75 - Samut songkhram 22 - Chanthaburi 75 - Samut songkhram 22 - Chanthaburi 51 - Lamphun 22 - Chanthaburi 51 - Lamphun 55 - Nan 51 - Lamphun 55 - Nan 54 - Phrae 55 - Nan 54 - Phrae 73 - Nakhon pathom 54 - Phrae 73 - Nakhon pathom 56 - Phayao 73 - Nakhon pathom 56 - Phayao 45 - Roi et 56 - Phayao 45 - Roi et 18 - Chai nat 45 - Roi et 18 - Chai nat 76 - Phetchaburi 18 - Chai nat 76 - Phetchaburi 24 - Chachoengsao 76 - Phetchaburi 24 - Chachoengsao 52 - Lampang 24 - Chachoengsao 52 - Lampang 92 - Trang 52 - Lampang 92 - Trang 20 - Chon buri 92 - Trang 20 - Chon buri 36 - Chaiyaphum 20 - Chon buri 36 - Chaiyaphum 21 - Rayong 36 - Chaiyaphum 21 - Rayong 40 - Khon kaen 21 - Rayong 40 - Khon kaen 39 - Nong bua lam phu 40 - Khon kaen 39 - Nong bua lam phu 35 - Yasothon 39 - Nong bua lam phu 35 - Yasothon 70 - Ratchaburi 35 - Yasothon 70 - Ratchaburi 90 - Songkhla 70 - Ratchaburi 90 - Songkhla 93 - Phatthalung 90 - Songkhla 93 - Phatthalung 23 - Trat 93 - Phatthalung 23 - Trat 44 - Maha sarakham 23 - Trat 44 - Maha sarakham 57 - Chiang rai 44 - Maha sarakham 57 - Chiang rai 72 - Suphanburi 57 - Chiang rai 72 - Suphanburi 26 - Nakhon nayok 72 - Suphanburi 26 - Nakhon nayok 65 - Phitsanulok 26 - Nakhon nayok 65 - Phitsanulok 48 - Nakhon phanom 65 - Phitsanulok 48 - Nakhon phanom 41 - Udon thani 48 - Nakhon phanom 41 - Udon thani 74 - Samut sakhon 41 - Udon thani 74 - Samut sakhon 61 - Uthai thani 74 - Samut sakhon 61 - Uthai thani 61 - Uthai thani 86 - Chumphon 86 - Chumphon 86 - Chumphon 17 - Sing buri 17 - Sing buri 17 - Sing buri 66 - Phichit 66 - Phichit 66 - Phichit 47 - Sakon nakhon 47 - Sakon nakhon 47 - Sakon nakhon 11 - Samut Prakan 11 - Samut Prakan 11 - Samut Prakan 80 - Nakhon si thammarat 80 - Nakhon si thammarat 80 - Nakhon si thammarat 53 - Uttaradit 53 - Uttaradit 53 - Uttaradit 13 - Pathum thani 13 - Pathum thani 13 - Pathum thani 27 - Sakaeo 27 - Sakaeo 27 - Sakaeo 67 - Phetchabun 67 - Phetchabun 67 - Phetchabun 46 - kalasin 46 - kalasin 46 - kalasin 15 - Ang thong 15 - Ang thong 15 - Ang thong 19 - Saraburi 19 - Saraburi 19 - Saraburi 77 - Prachuap khiri khan 77 - Prachuap khiri khan 77 - Prachuap khiri khan 50 - Chiang mai 50 - Chiang mai 50 - Chiang mai Figure A3.1.1. Student Performance Index 49 - Mukdahan 49 - Mukdahan 49 - Mukdahan 42 - Loei 42 - Loei 42 - Loei 33 - Si sa ket 33 - Si sa ket 33 - Si sa ket Figure A3.1.3. Average Teacher Experience (Years) 91 - Satun 91 - Satun 91 - Satun 84 - Surat thani 84 - Surat thani 84 - Surat thani 62 - Kamphang phet 62 - Kamphang phet 62 - Kamphang phet 60 - Nakhon sawan 60 - Nakhon sawan 60 - Nakhon sawan 34 - Ubon ratchathani 34 - Ubon ratchathani 34 - Ubon ratchathani Figure A3.1.2. Share of Teachers with Higher than Bachelor's Degree 14 - Phra nakhon si ayutthaya 14 - Phra nakhon si ayutthaya 14 - Phra nakhon si ayutthaya 63 - Tak 63 - Tak 63 - Tak 82 - Phangnga 82 - Phangnga 82 - Phangnga 64 - Sukhothai 64 - Sukhothai 64 - Sukhothai 85 - Ranong 85 - Ranong 85 - Ranong 71 - Kanchanaburi 71 - Kanchanaburi 71 - Kanchanaburi 16 - Lop buri 16 - Lop buri 16 - Lop buri 43 - Nong khai 43 - Nong khai 43 - Nong khai 30 - Nakhon ratchasima 30 - Nakhon ratchasima 30 - Nakhon ratchasima 81 - Krabi 81 - Krabi 81 - Krabi 25 - Prachin buri 25 - Prachin buri 25 - Prachin buri 37 - Amnat charoen 37 - Amnat charoen 37 - Amnat charoen 32 - Surin 32 - Surin 32 - Surin 31 - Buri ram 31 - Buri ram 31 - Buri ram 58 - Mae hong son 58 - Mae hong son 58 - Mae hong son 95 - Yala 95 - Yala 95 - Yala Figure A3.1. Key Characteristics of OBEC Schools – Ranked by Student Performance 94 - Pattani 94 - Pattani 94 - Pattani 96 - Naratiwat 96 - Naratiwat 96 - Naratiwat 10 15 20 25 30 35 40 45 0 5 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 10 - Bangkok Metropolis 10,000 15,000 20,000 25,000 30,000 5,000 0 10 - Bangkok Metropolis 12 - Nonthaburi 12 - Nonthaburi 10 - Bangkok Metropolis 83 - Phuket 83 - Phuket 12 - Nonthaburi 75 - Samut songkhram 75 - Samut songkhram 83 - Phuket 22 - Chanthaburi 22 - Chanthaburi 75 - Samut songkhram 51 - Lamphun 51 - Lamphun 22 - Chanthaburi 55 - Nan 55 - Nan 51 - Lamphun 54 - Phrae 54 - Phrae 55 - Nan 73 - Nakhon pathom 73 - Nakhon pathom 54 - Phrae 56 - Phayao 56 - Phayao 73 - Nakhon pathom 45 - Roi et 45 - Roi et 56 - Phayao 18 - Chai nat 18 - Chai nat 45 - Roi et 76 - Phetchaburi 76 - Phetchaburi 18 - Chai nat 24 - Chachoengsao 24 - Chachoengsao 76 - Phetchaburi 52 - Lampang 52 - Lampang 24 - Chachoengsao 92 - Trang 92 - Trang 52 - Lampang 20 - Chon buri 20 - Chon buri 92 - Trang 36 - Chaiyaphum 20 - Chon buri 36 - Chaiyaphum 21 - Rayong 21 - Rayong 36 - Chaiyaphum 40 - Khon kaen 21 - Rayong 40 - Khon kaen 39 - Nong bua lam phu 39 - Nong bua lam phu 40 - Khon kaen 35 - Yasothon 39 - Nong bua lam phu 35 - Yasothon 70 - Ratchaburi 70 - Ratchaburi 35 - Yasothon 90 - Songkhla 90 - Songkhla 70 - Ratchaburi 93 - Phatthalung 93 - Phatthalung 90 - Songkhla 23 - Trat 23 - Trat 93 - Phatthalung 23 - Trat 44 - Maha sarakham 44 - Maha sarakham 44 - Maha sarakham 57 - Chiang rai 57 - Chiang rai 57 - Chiang rai 72 - Suphanburi 72 - Suphanburi 72 - Suphanburi 26 - Nakhon nayok 26 - Nakhon nayok 26 - Nakhon nayok 65 - Phitsanulok 65 - Phitsanulok 65 - Phitsanulok 48 - Nakhon phanom 48 - Nakhon phanom 48 - Nakhon phanom 41 - Udon thani 41 - Udon thani 41 - Udon thani 74 - Samut sakhon 74 - Samut sakhon 74 - Samut sakhon 61 - Uthai thani 61 - Uthai thani 61 - Uthai thani 86 - Chumphon 86 - Chumphon 86 - Chumphon 17 - Sing buri 17 - Sing buri 17 - Sing buri 66 - Phichit 66 - Phichit 66 - Phichit 47 - Sakon nakhon 47 - Sakon nakhon 47 - Sakon nakhon 11 - Samut Prakan 11 - Samut Prakan 11 - Samut Prakan 80 - Nakhon si thammarat 80 - Nakhon si thammarat Source: World Bank Staff Calculations based on OBEC and NIETS 80 - Nakhon si thammarat 53 - Uttaradit 53 - Uttaradit 53 - Uttaradit 13 - Pathum thani 13 - Pathum thani 13 - Pathum thani 27 - Sakaeo 27 - Sakaeo 27 - Sakaeo 67 - Phetchabun 67 - Phetchabun 67 - Phetchabun 46 - kalasin 46 - kalasin 46 - kalasin 15 - Ang thong 15 - Ang thong Figure A3.1.5. Average Class Size 15 - Ang thong 19 - Saraburi 19 - Saraburi 19 - Saraburi 77 - Prachuap khiri khan 77 - Prachuap khiri khan 77 - Prachuap khiri khan 50 - Chiang mai 50 - Chiang mai 50 - Chiang mai 49 - Mukdahan 49 - Mukdahan 49 - Mukdahan 42 - Loei 42 - Loei 42 - Loei 33 - Si sa ket 33 - Si sa ket 33 - Si sa ket Figure A3.1.6. Average Personnel Salary Per Student 91 - Satun 91 - Satun 91 - Satun 84 - Surat thani 84 - Surat thani 84 - Surat thani Figure A3.1.4. Average Number of Teachers per Classroom 62 - Kamphang phet 62 - Kamphang phet 62 - Kamphang phet 60 - Nakhon sawan 60 - Nakhon sawan 60 - Nakhon sawan 34 - Ubon ratchathani 34 - Ubon ratchathani 34 - Ubon ratchathani 14 - Phra nakhon si ayutthaya 14 - Phra nakhon si ayutthaya 14 - Phra nakhon si ayutthaya 63 - Tak 63 - Tak 63 - Tak 82 - Phangnga 82 - Phangnga 82 - Phangnga 64 - Sukhothai 64 - Sukhothai 64 - Sukhothai 85 - Ranong 85 - Ranong 85 - Ranong 71 - Kanchanaburi 71 - Kanchanaburi 71 - Kanchanaburi 16 - Lop buri 16 - Lop buri 16 - Lop buri 43 - Nong khai 43 - Nong khai 43 - Nong khai 30 - Nakhon ratchasima 30 - Nakhon ratchasima 30 - Nakhon ratchasima 81 - Krabi 81 - Krabi 81 - Krabi 25 - Prachin buri 25 - Prachin buri 25 - Prachin buri 37 - Amnat charoen 37 - Amnat charoen 37 - Amnat charoen 32 - Surin 32 - Surin 32 - Surin 31 - Buri ram 31 - Buri ram 31 - Buri ram 58 - Mae hong son 58 - Mae hong son 58 - Mae hong son 95 - Yala 95 - Yala 95 - Yala 94 - Pattani 94 - Pattani 94 - Pattani 96 - Naratiwat 96 - Naratiwat 96 - Naratiwat Annex A4: What is the evidence that organizing learning with better teachers will improve learning? This annex presents the empirical evidence that staffing classrooms in poor schools with adequate number of good quality teachers will have a large impact on student learning. Our empirical analysis in this annex employs the concept of an Educational Production Function where it is conceptualized that schools employ some production technology to combine various inputs into producing student learning.2 Using this framework, we estimate the causal relationships between school-average student achievement3 and key measures of school quality, namely, the proportion of teachers with higher than bachelor’s degree qualification, the teacher workforce average years of teaching experience, the “unobserved teacher quality index”,4 and the average number of teachers per classroom. Due to data limitations, however, we will not be able to evaluate the impacts of material educational resources and physical infrastructure on student learning.5 Since the focus of this annex is to evaluate potential interventions that are needed to close the school performance gap, we will be estimating the effects of these school quality variables on the entire distribution of school-level student performance outcome. An estimation method that “goes beyond the mean” will have to be employed for this particular purpose. In this study, we use the “Unconditional Quantile Regression” (UQR) method introduced by Firpo, Fortin, and Lemieux (2009) to estimate the causal impacts of the key measures of school quality on student achievement at different performance quantiles. Issues relating to the estimation of the educational production function model using the UQR technique is described in Technical Appendix to Annex A4, where detailed discussions of the results are also provided. The rest of this annex highlights the key findings. All four measures of school quality show positive and statistically significant effects on student performance. Our analysis begins with analyzing the effects of measured school quality variables on average student performance outcome at the school level. In particular, it is found that 1) an increase in the unobserved teacher quality index of one standard deviation is associated with an increase of 0.9 percent in student performance, 2) a 10 percentage point increase in the share of teachers with higher than bachelor’s degree qualification is expected to raise student performance by 0.27 percent, 3) an increase of 10 years in the average experience of the teacher workforce is estimated to improve student performance outcome by 1 percent, and 4) allocating one more teacher for each classroom is expected to raise performance by as much as 2.6 percent, holding other factors constant. Gains in student learning from having better quality teachers are found to be much larger for low-performing schools. The estimated causal effects of the three measured teacher quality variables show greater impacts on student achievement for schools ranked at the lower end of the performance 2 The empirical study in this chapter uses a 2010 cross-sectional school data collected by the Office of the Basic Education Commission (OBEC), the Ministry of Education in Thailand (see Annex section A4.3). 3 Student achievement is measured by the Student Performance Index, which is a weighted index of mathematics and science scores in the 2010 Ordinary National Education Test (O-NET) exams for Grades 6, 9, and 12. The index is constructed as explained in detail in Section A5.3 in Technical Appendix to Annex A5. 4 As described in the Annex section A4.3, the “unobserved teacher quality index” approximately captures variations arising from the discretionary wage component (such as performance pay), the average academic ranking of the teacher workforce, and other school average teacher characteristics unobserved by the researcher. 5 Nevertheless, all is not lost as a rich body of research shows that effective teachers are the most important factor contributing to student learning (see for example Rivkin, Hanushek, and Kain (2005), Hanushek, Kain, and Rivkin (1998), and Nye, Konstantopoulos, and Hedges (2004)). distribution. The three graphs in Figure A4.2 (reproduced from Technical Appendix to Annex A4) show the estimated marginal effects of the three teacher quality variables on the entire distribution of schools, ranked in accordance with their student performance outcome. The dotted lines in the graphs represent the 95 percent confidence band for the estimated effects. Consider low-performing schools that are ranked at the 10th percentile of the performance distribution. The study finds that 1) an increase in the unobserved teacher quality index of one standard deviation is expected to raise student performance by 1.14 percent (Figure A4.2a), 2) a 10 percentage point increase in the share of teachers with higher than bachelor’s degree qualification is expected to increase student performance by 0.48 percent (Figure A4.2b), and 3) an increase of 10 years in the average experience of the teacher workforce is estimated to improve student performance outcome by 2.6 percent (Figure A4.2c). The estimated marginal effects of the teacher quality variables on schools ranked at the 10 th percentile are therefore much larger than the estimated effects for the average school discussed in the previous paragraph. The marginal effects on other percentiles can analyzed analogously. Figure A4.2. The Estimated Marginal Effects of Measured Teacher Quality Variables on Student Performance a) Unobserved teacher quality b) Share of teachers with c) Teaching Experience index higher than bachelor's degree 0.004 0.025 0.12 0.003 0.020 0.08 0.002 0.015 0.001 0.04 0 0.010 0.00 -0.001 0.005 -0.002 0.000 -0.04 -0.003 -0.005 -0.08 -0.004 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.02 0.18 0.34 0.5 0.66 0.82 0.98 Quantile Quantile Quantile The study also finds the greatest positive impacts on student learning for the lowest- performing schools from easing severe teacher shortages in their classrooms. Turning now to our measure of the adequacy of teacher quantity, we can see from Figure A4.3 that allocating one more teacher to each classroom is expected to raise performance for schools at the 2 nd percentile of the performance distribution by as much as 15.8 percent. The effects are estimated to fall to within a range of 4.8-6.5 percent for schools ranked between the 6th and the 20th percentiles before falling off gradually thereafter. The large and positive effects for schools at the bottom end of the performance distribution should not come as a surprise, considering the fact that teacher shortages are very critical for these schools (see Technical Appendix to Annex A4 for more detailed discussion). The results shown in Figure A4.3 alone are strong evidence that staffing shortages in Thai classrooms are adversely affecting student learning, especially in small schools that primarily serve socioeconomically disadvantaged students. Figure A4.3. The Estimated Marginal Effects of the Number of Teachers per Classroom on Student Performance 0.20 0.15 0.10 0.05 0.00 -0.05 -0.10 0.02 0.18 0.34 0.5 0.66 0.82 0.98 Quantile The results from our empirical study in this annex show that eliminating teacher shortages, both in terms of quality and quantity, would result in significant improvements in student achievement and the impacts would be greatest for lower-performing schools. The analyses of the effects of measured teacher quality and the number of teachers per classroom unambiguously suggest that allocating more and better teachers to small and low-performing schools would result in significant improvements in student learning. In light of these findings, we conclude that improving the quality of teachers and addressing the severe teacher shortages, especially for the vast number of small rural schools, should be at the center of Thailand’s reform initiatives if the country is serious about tackling the widespread low quality education and the high disparity in educational achievements across socioeconomic groups. Technical Appendix to Annex A4 A4.1. Using School Level Educational Production Function to Determine the Effects of Measured Teacher and School Characteristics on Student Performance Our empirical analysis in Annex A4 uses school level data (see Annex A4.3) to estimate the educational production function for schools under the jurisdiction of the Office of the Basic Education Commission (OBEC). Specifically, we conceptualize schools as employing some production technology to combine various inputs into producing student learning. This analytical framework is particularly useful in helping us determine the relationships between measured teacher and school characteristics with student achievement, as indicated by the Student Performance Index in the 2010 O-NET exams. Complications arise in the model estimation stage, however. Due to data limitations, we do not observe all potentially important family background characteristics of the student body that capture the quality of early education the students received or the home environments that are conducive to learning. These background characteristics are crucial in determining a child’s cognitive ability. Furthermore, it is conceivable that these relevant family factors are related to both student achievement and the selection of schools and teachers by the parents.6 In other words, the omitted family background characteristics of the student body are confounding our estimates of the school and teacher effects on student achievement. To mitigate the problem, this study statistically controls for some of this confounding by including as covariates the available variables that capture the socioeconomic status of the student body.7 Furthermore, a particularly important covariate to include in the model is student achievement in the previous school year. We argue that prior student achievement can be seen as summarizing the effects of the remaining confounding socioeconomic factors. Including the variable as a covariate therefore permits a comparison of student performance across schools. In our empirical investigation of the educational production function, we are especially interested in the effects of three measures of teacher quality on student achievement. These are; the proportion of teachers with higher than bachelor’s degree qualification, the teachers’ average years of teaching experience, and the “unobserved teacher quality index”.8 Furthermore, to assess the school effect on student performance, we have decided to include as a covariate the average number of teachers per classroom instead of a more conventional average class size variable. From Table A3.1 in Annex A3, we can see that most small rural schools in Thailand have very small average class sizes. However, instead of reflecting provision of high quality education, the small average classes actually reflects the severity of teacher shortages in small schools. Hence, the number of teachers per classroom variable is a more appropriate variable to include in our model. Finally, recall that the primary focus of this Annex is to evaluate interventions that are needed to close the performance gap between schools. Since our underlying question of policy interest concerns the entire distribution of student performance across schools, estimation methods that “g o beyond the mean” have to be employed. In this study, we use the “Unconditional Quantile Regression” method introduced by Firpo, Fortin, and Lemieux (2009) to estimate the impact of changes in the covariates on the unconditional quantiles of the outcome variable (school performance). The method is briefly described in section A4.2 in this Annex. What do the empirical evidences tell us? Before we go on to analyze the estimation results from the educational production function model, let us first consider simple relationship between enrolment size and student performance. The left-hand graph in Figure A4.1 presents a scatter plot of school enrolment size against the percentile ranking of the school student performance index in 2010. While we can see that many small schools managed to score highly in terms of student achievement, it cannot be denied that the majority of low-performing schools are small schools. Furthermore, the right-hand graph in the same figure shows a scatter plot of the average number of teachers per classroom against enrolment size. Three features of the graph are particularly intriguing. First, teacher shortage is clearly relatively more severe for 6 For example, highly educated parents are more likely to have prepared their children better since birth to be school ready and to have chosen neighbourhoods with schools that are well-resourced and have high quality teachers. 7 In fact, from Table A4.3 in this section we can see that we have available to us only two variables that could potentially capture the socioeconomic status of the school student body. These are: 1) the share of students that are poor and 2) the distance from the school to the nearest city. 8 As described in the Annex A4.3, the “unobserved teacher quality index” approximately captures variations arising from the discretionary wage component (such as performance pay), the average academic ranking of the teacher workforce, and other school average teacher characteristics unobserved by the researcher. smaller schools. Second, many small schools are seen to be staffed with relatively high number of teachers per classroom. Third, a lot of the schools that are not classified as “small” by OBEC definition (having 120 enrolled students or less) also face serious teacher shortage problem. Notice that it is the variations in key variables such as these that enable us to estimate the effects of measured teacher and school characteristics on student performance. Figure A4.1. Enrolment vs. Student Achievement (Left) and Teacher per Classroom vs. Enrolment (Right) Enrolment Size vs. Test Score Percentile Ranking Teachers per Classroom vs. Enrolment Size 6000 2 1.5 Teachers per classroom 4000 Enrolment size 1 2000 .5 0 0 0 .2 .4 .6 .8 1 0 2000 4000 6000 Test score percentile ranking Enrolment size Source: World Bank Staff Calculations based on OBEC and NIETS The functional form for the educational production function used in this study can be seen from Table A4.2 in this section. Also reported in the Table are the coefficient estimates from the unconditional quantile regression model (UQR) for some selected quantiles of the school performance distribution, as well as the coefficient estimates obtained using conventional ordinary least squares (OLS) regression. Let us first analyze the estimation results from the OLS regression model which is reported in the final column of Table A4.2. All of the coefficients can be seen to have the expected signs and are mostly significant at conventional statistical levels. However, since two of the four variables of interest enter the model in quadratic form, they are not straight forward to interpret. Therefore, to ease exposition, we also report in Table A4.1 the estimated average marginal effects of the four measured teacher and school characteristics on (log) student performance. The final column of Table A4.1 shows that an increase in the unobserved teacher quality index of one standard deviation is associated with an increase of 0.9 percent in the student performance index for an average school, ceteris paribus. Similarly, a 10 percentage point increase in the share of teachers with higher than bachelor’s degree qualification is expected to raise student performance by 0.27 percent. An increase of 10 years in the average experience of the teacher workforce is estimated to improve student performance outcome by 1 percent. Lastly, we find that allocating one more teacher for each classroom is expected to raise performance by as much as 2.6 percent, holding other factors constant. Notice that the estimated marginal effects are all statistically significant at the 1 percent significance level. Table A4.1. Average Marginal Effects of Selected Variables from the Educational Production Function Model – OLS and Unconditional Quantile Regression Q10 Q30 Q50 Q70 Q90 OLS Unobserved teacher quality index 0.011*** 0.010*** 0.009*** 0.007*** 0.007*** 0.009*** (0.002) (0.001) (0.001) (0.002) (0.003) (0.001) Share of teachers with more than bachelor’s degree 0.048*** 0.050*** 0.045*** 0.018 -0.014 0.027*** (0.017) (0.012) (0.011) (0.014) (0.021) (0.010) Average years of teaching experience 0.003*** 0.002*** 0.002*** 0.001*** -0.002*** 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Average number of teachers per class 0.058*** 0.036*** 0.021*** 0.009 -0.013* 0.026*** (0.007) (0.005) (0.004) (0.005) (0.008) (0.004) Delta method standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 While OLS can be used to estimate the partial effects of the covariates on the performance outcome for an average school, the narrow focus on the mean outcome obscures the effects of the teacher and school characteristics on other important features of the school performance distribution that are of policy relevance. Much richer analysis can be carried out using UQR as is illustrated in the first five columns of Table A4.1, where the estimated marginal effects of the key variables on school performance are presented for schools ranked at the 10th, 30th, 50th, 70th, and 90th performance percentiles. Alternatively, the UQR marginal effects can be presented graphically as shown in Figure A4.2 for the three variables on measured teacher quality, and in Figure A4.3 for the number of teachers per classroom variable (see Figure A4.4 in this Annex for all the graphs of the UQR coefficient estimates). Figure A4.2a shows the estimated marginal effects of the unobserved teacher quality index on the entire distribution of schools, ranked in accordance with their student performance outcome index in 2010. The dotted lines in the graph represent the 95 percent confidence band for the estimated effects. Immediately apparent are the larger effects of a one standard deviation increase in the unobserved teacher quality index on student performance for schools ranked towards the lower end of the performance distribution. Figure A4.2. Unconditional Quantile Marginal Effects – Teacher Quality Variables a) Unobserved teacher quality b) Share of teachers with c) Teaching Experience index higher than bachelor's degree 0.004 0.025 0.12 0.003 0.020 0.08 0.002 0.015 0.001 0.04 0 0.010 0.00 -0.001 0.005 -0.002 0.000 -0.04 -0.003 -0.005 -0.08 -0.004 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.02 0.18 0.34 0.5 0.66 0.82 0.98 Quantile Quantile Quantile Similarly, we can see from Figure A4.2b that the effects on student performance from increasing the share of teachers with higher than bachelor’s degree qualification are largest for schools that performed below the median in the performance distribution. For schools that are ranked higher than the 82nd percentile, the effects of increasing the share of highly qualified teachers turn out to be negative. However, it can be inferred from the width of the 95 percent confidence band that the estimated effects are not significantly different from zero. The final measured teacher characteristic of interest is the average years of experience of the teacher workforce. Once again, Figure A4.2c shows the estimated effects of allocating more experienced teachers to lower-performing schools to be very high. For instance, an increase of 10 years in the average teacher experience is estimated to improve student performance outcome for schools at the 12th percentile by as much as 2.7 percent. This is much greater than the 1 percent impact estimated for the average school using OLS. Finally, it is not obvious why the marginal effects of average teacher experience should turn negative and statistically significant for schools which are ranked above the 82nd percentile. Figure A4.3. Unconditional Quantile Marginal Effects – Teachers per Classroom 0.20 0.15 0.10 0.05 0.00 -0.05 -0.10 0.02 0.18 0.34 0.5 0.66 0.82 0.98 Quantile Turning now to our measure of the adequacy of teacher quantity, we can see from Figure A4.3 that allocating one more teacher for each classroom is expected to raise performance for schools at the 2nd percentile of the performance distribution by as much as 15.8 percent (or 0.147 log points), other factors held constant. The effects drop down to within a range of 4.8-6.5 percent for schools between the 6th and the 20th percentiles before falling off gradually thereafter. The large and positive effects for schools at the bottom end of the performance distribution is not surprising considering the fact that teacher shortages are very severe for these schools.9 9 For schools ranked at or below the 2nd percentile of the performance distribution, the average number of teachers per classroom is 0.79. For schools ranked between the 2nd and 4th percentiles, the figure improves slightly to 0.93. The figure improves further to 1.0 for schools ranked between the 4th and 6th percentiles, and to 1.06 for schools ranked between the 6th and 20th percentiles. For those schools that are ranked above the 20 th percentile, the average number of teachers per classroom rises to 1.18. These figures once again confirm that teacher shortage is a very serious problem constraining Thai schools. Table A4.2. OLS and Unconditional Quantile Regression Results for the Educational Production Function Model Dependent variable: Log student performance index in 2010 Q10 Q30 Q50 Q70 Q90 OLS Log student performance index in 2009 0.286*** 0.291*** 0.354*** 0.446*** 0.517*** 0.377*** (0.009) (0.006) (0.005) (0.007) (0.013) (0.005) Unobserved teacher quality index 0.011*** 0.012*** 0.009*** 0.007*** 0.007*** 0.009*** (0.002) (0.002) (0.001) (0.002) (0.003) (0.001) Share of primary students -0.018 -0.028 -0.038** -0.056** -0.057* -0.040** (0.029) (0.022) (0.019) (0.023) (0.034) (0.017) Share of lower secondary students 0.050* -0.052** -0.136*** -0.172*** -0.163*** -0.103*** (0.026) (0.021) (0.018) (0.021) (0.031) (0.015) Share of upper secondary students 0.022 0.052** 0.000 -0.025 -0.065* -0.001 (0.031) (0.025) (0.021) (0.025) (0.035) (0.017) Average number of teachers per class 0.222*** 0.164*** 0.039*** -0.016 -0.082*** 0.070*** (0.022) (0.017) (0.013) (0.016) (0.022) (0.012) Average number of teachers per class2 -0.079*** -0.056*** -0.009 0.012* 0.033*** -0.021*** (0.009) (0.007) (0.005) (0.007) (0.009) (0.005) Share of students poor -0.036*** -0.031*** -0.007* 0.012** 0.033*** -0.005 (0.006) (0.005) (0.004) (0.005) (0.007) (0.003) Share of teachers with more than bachelor's degree 0.048*** 0.046*** 0.045*** 0.018 -0.014 0.027*** (0.017) (0.013) (0.011) (0.014) (0.021) (0.010) Average years of teaching experience 0.014*** 0.010*** 0.004*** 0.003*** 0.005*** 0.007*** (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) Average years of teaching experience2 -0.000*** -0.000*** -0.000** -0.000** -0.000*** -0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Distance in km to the nearest city -0.001*** -0.001*** -0.001*** -0.000** 0.000 -0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Intercept 2.113*** 2.274*** 2.336*** 2.159*** 2.120*** 2.222*** (0.044) (0.032) (0.025) (0.032) (0.053) (0.025) Observations 30,120 30,120 30,120 30,120 30,120 30,120 R-squared 0.078 0.104 0.147 0.144 0.090 0.212 Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table A4.3. Summary Statistics (at the school level) Mean S.D. Min Max Log student performance index in 2010 3.690 0.207 2.303 4.554 Log student performance index in 2009 3.664 0.233 2.708 4.474 Unobserved teacher quality index 0.003 0.993 -17.232 15.985 Share of primary students 0.651 0.225 0 1 Share of lower secondary students 0.118 0.196 0 1 Share of upper secondary students 0.031 0.109 0 1 Average number of teachers per class 1.038 0.384 0.100 4.000 Share of students poor 0.662 0.319 0 1 Share of teachers with more than bachelor's degree 0.120 0.117 0 1 Average years of teaching experience 22.747 7.273 0.000 39.417 Distance in km to the nearest city 12.201 11.000 0 136 Number of schools 30,120 Figure A4.4. Unconditional Quantile Regression Coefficient Estimates Log student performance Unobserved teacher quality Share of primary students index in 2009 index 0 0.6 0.014 0.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.01 0.5 0.012 -0.02 0.01 -0.03 0.4 0.008 -0.04 0.3 0.006 -0.05 0.2 0.004 -0.06 0.1 0.002 -0.07 0 0 -0.08 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.09 Quantile Quantile Quantile Share of lower secondary Share of upper secondary Teacher per class student students 0.7 0.20 0.08 0.6 0.15 0.06 0.5 0.10 0.04 0.4 0.05 0.02 0.3 0.00 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0 0.2 -0.05 0.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.02 0.1 -0.10 -0.15 -0.04 0 0.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.20 -0.06 -0.1 -0.25 -0.08 -0.2 Quantile Quantile Quantile Teacher per class squared Share of poor students Share of teachers with higher than bachelor's degree 0.10 0.05 0.04 0.07 0.05 0.06 0.03 0.05 0.00 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.02 0.04 -0.05 0.01 0.03 0.02 -0.10 0 0.01 0.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.01 0 -0.15 -0.02 -0.010.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.20 -0.02 -0.03 -0.03 -0.25 -0.04 -0.04 Quantile Quantile Quantile Figure A4.4. Unconditional Quantile Regression Coefficient Estimates (continued) Average teaching experience Average teaching experience Distance to the nearest city (years) squared (km) 0.030 0 0.0002 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.025 -0.0001 0.0000 0.02 0.18 0.34 0.5 0.66 0.82 0.98 0.020 -0.0002 -0.0002 0.015 -0.0003 -0.0004 0.010 -0.0004 -0.0006 0.005 -0.0005 -0.0006 -0.0008 0.000 0.02 0.18 0.34 0.5 0.66 0.82 0.98 -0.0007 -0.0010 Quantile Quantile Quantile Intercept 2.8 2.6 2.4 2.2 2 1.8 1.6 1.4 0.02 0.18 0.34 0.5 0.66 0.82 0.98 Quantile A4.2. Unconditional Quantile Regression This section describes the unconditional quantile regression method introduced by Firpo, Fortin, and Lemieux (2009) – FFL from hereon. The method enables an evaluation of the impact of changing the distribution of explanatory variables on the quantiles of the “unconditional” (or marginal) distribution of the outcome variable using a familiar regression framework. Consider a real-valued statistical functional ( ), where is an underlying distribution function for (any) random variable . The influence function (; , ) introduced by Hampel (1968, 1974) is a widely used tool in studies on local robustness properties of functionals, and is defined as, (,. ) ( +( − ))−( ) (; , ) = | = lim ,⁡⁡⁡⁡⁡0 < < 1 (1) ↓0 =0 if this limit is defined for every point ∈ ℝ, and denotes the probability measure that puts the mass 1 at the value . If a statistical functional (von Mises functional due to Mises (1947)) is Gâteaux differentiable at , FFL show that the following approximation holds for some distribution function close to , ( ) = ( ) + ∫ℝ (; , )( − )() + (2) where is a remainder term. Noting that ∫ℝ (; , ) () = 0 by definition, we have ( ) = ( ) + ∫ℝ (; , ) () + . For a particular case that = , FFL call this first order approximation term the “Recentered Influence Function,” (; , ) = ( ) + ∫ℝ (; , ) ()⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡ = ( ) + (; , ) (3) FFL recognise several interesting properties of the (; , ), the most important of which is that it integrates up to the functional of interest ( ); that is, ∫ℝ (; , ) () = [(; , )] = ⁡( ). Applying the law of iterated expectation to this expression yields, [[(; , )|)]] = [ ()] = ( ) (4) where [∙] makes explicit that the expectation is taken over the support of , and () denotes the RIF regression model with regard to the statistical functional . As mentioned at the outset, this study applies the RIF Regression method to a whole range of quantiles of the dependent variable , which are generically denoted by ( ) for any quantile ∈ (0,1) of interest. The influence function for a quantile is given by: −(≤ ) (; , ) = (5) ( ) where (∙) denotes an indicator function, equal to 1 if the expression in the parentheses is true and 0 otherwise. Using the definition of RIF given in (3) for the ℎ quantile of the marginal distribution of , the feasible version of the (; , ) can be expressed as: ̂ ) (1−) ̂ (; ̂ ) = (> ̂ , + ̂ − (6) ̂ ̂ ( ) ̂ ̂ ( ) Note that the estimator of the ℎ population quantile of the marginal distribution of , ̂ , is nonparametrically identified from observed sample. Furthermore, the empirical density of is estimated using the kernel density procedure as follows: 1 − ̂ () = ∑ =1 ( ) (7) ℎ ℎ where (∙) is the kernel density function, ℎ is the kernel bandwidth,10 is the sampling weight,11 and ̂ . is the total sample size. The empirical density function is evaluated at A4.3. Data This study employs a 2010 cross-sectional school data collected by the Office of the Basic Education Commission (OBEC), the Ministry of Education in Thailand. A rich set of information from 31,330 schools were collected on teacher and personnel salaries, number of students in each education level, number of poor students (students whose household incomes are below 40,000 Baht per month), number of classes, number of teachers by their educational qualification, their average years of teaching experience, etc. Hence, the data are aggregated at the school level by nature. Using the raw data, we were able to construct school-level variables needed in estimating the educational production function. The constructed variables that serve as inputs into the production function capture the average student body, teacher, and school attributes. Particularly, these are: the shares of students by broad schooling level (pre-primary, primary, lower secondary, and upper secondary), the share of students that are poor, the proportion of teachers with higher than bachelor’s degree qualification, the teachers’ average years of teaching experience, the “unobserved teacher quality index”, the average number of teachers per classroom, and the distance in kilometers from the school to the nearest city (Amphur). The “unobserved teacher quality index” is constructed from the “composition-adjusted teacher salary” variable which is computed as explained in Section A5.4. As discussed at length in section A5.4, the variation in the composition-adjusted teacher salary encompasses variations arising from teacher “price” differences across geographical areas, the discretionary wage component (such as performance pay), the average academic ranking of the teacher workforce, and other school average teacher characteristics unobserved by the researcher. The unobserved teacher quality index is then computed by normalizing the composition-adjusted teacher salary to have zero mean and unit 1 2 10 The Gaussian kernel is used in this study, where () = √2 − /2 11 Notice that the sampling weight is normalized to sum to . That is, ∑ =1 = variance. Notice that since the index also encompasses teacher’s wage variation arising from teacher prices, we therefore regard it as an approximate measure of unobserved teacher quality. The final key input variable in the educational production function is the school level “student performance index in 2009”, which is a weighted index of mathematics and science scores in the 2009 Ordinary National Education Test (O-NET) exams for Grades 6, 9, and 12. The index is constructed as explained in great detail in A5.3. The output variable for the educational production function is the (log) student performance index in 2010. The summary statistics for the variables used in estimating the educational production function model are presented in Table A4.3. Notice that a small percentage of the observations were dropped due to the presence of missing data on one or more variables. Furthermore, the 137 schools that had no students were discarded. The final sample size is 30,120 schools. Annex A5: Per-Student Funding Formula based on an Educational Cost Function Approach Table 5.1. Funding Formula Example Application School Characteristics Shares by level Baseline Simulation Pre-school 13.4% 30.0% Primary 45.5% 50.0% Lower secondary 27.6% 20.0% Upper secondary 13.5% 0.0% School level composition index (1) 1 0.875 Enrolment size Baseline Simulation less than 50 0 0 50 to 69 0 0 70 to 89 0 0 90 to 119 0 0 120 to 149 0 1 150 to 199 0 0 200 to 279 0 0 280 to 499 0 0 500 to 749 1 0 750 to 1149 0 0 1150 to 1999 0 0 2000 or above 0 0 Enrolment size index (2) 1 1.268 Class size Baseline Simulation less than 10 0 0 10 to 19 0 1 20 to 29 1 0 30 to 34 0 0 35 to 39 0 0 40 to 44 0 0 45 or above 0 0 Average class size index (3) 1 1.103 Student Body Characteristics Base case Simulation Share of students who are poor 48.9% 60.0% Student poverty index (4) 1 1.001 Base case Simulation Share of disabled students 0.2% 0.0% Student disability index (5) 1 0.999 Table 5.1 Funding Formula Example Application (continued) Provincial Labor Market Characteristics Base case Simulation Teacher wage index 26,816 26,816 Input price index (6) 1 1.000 Base case Simulation Overall cost index (7)* 1 1.225 Per student spending 33,506 41,035 Student performance index 50 **Note (7)=(1)x(2)x(3)x(4)x(5)x(6) Box A5.1. Predicting the Per-Student Cost of Meeting Student Performance Standard This section applies the cost function model to predicting the per-student cost for a hypothetical “baseline school” that is adequate for supporting students in attaining some educational performance outcome standard. The estimated regression coefficients from the cost function model (see column “Frontier IV” in Table A5.4) are then used to construct cost indices for each school in the estimation sample. These cost indices reflect differences relative to the baseline school of key student body and school characteristics that are outside the control of the school or local community in question. An overall cost index is then computed by multiplying together the various component indices. This overall index is effectively a measure of relative variation in per-student cost for the school compared to the baseline school. For example, a cost index of 1.5 indicates that the school in question will require 50% more spending per student than the baseline school to achieve the same average student performance outcome. The manner in which the cost indices are estimated is completely transparent and the indices can be easily applied to predicting the required per-student spending for any school in the country. We begin with using the estimated per-student cost function to generate the predicted subsidy required for each school to reach some average student performance outcome. This is made possible because the cost function specification directly ties spending to performance, while accounting for differences in school and student-body characteristics. In this study, we set the required Student Performance Index12 at 50 (referred to as the performance standard from hereon), or at around the 84th percentile in the student performance outcome distribution. Then, we estimate the per-student cost of reaching this performance standard for a hypothetical “baseline school” which has as its characteristics the mean shares of students by the three broad schooling levels (pre-primary, primary, lower secondary, and upper-secondary), the mean shares of poor and disabled students, and the national average teacher salary index (see Annex Section A5.4). Furthermore, we set the enrolment size of the baseline school at between 500 to 749 students and an average class size at between 20 to 29 students. Finally, we allow the baseline school to have an average level of cost inefficiency (see equation (12) in Section 12 The computation of the Student Performance Index is explained in Section... A5.1). Using equations (9), (10), and (11) from Section A5.1; the predicted log per-student cost of the baseline school can thus be written as: ̂ ln⁡ ̂1 ln() + ̂0 + ( ) = ̂ + ̂2 ln(50) + ∑ ′ =3 where is the provincial average composition-adjusted teacher salary, is a vector of the baseline school characteristics set at values as stated in the previous paragraph, is the average level of cost ̂ ’s are the estimated marginal effects. inefficiency, and the We use equation (14) from Section A5.1 to convert the predicted log per-student cost of the baseline school into the predicted cost in level term. The per-student cost of this hypothetical baseline school to meet the required performance standard is predicted to be 33,506 Baht per annum (see Table 5.1). The adequate per-student cost for each school to meet the performance standard based on the estimated marginal effects is the predicted cost of the baseline school multiplied by an “overall cost index” for the school. The overall cost index is in turn a product of different component indices pertaining to schooling level composition of the student body, enrolment size, average class size, student poverty, and the share of students who are disabled. For example, consider estimating the schooling level composition index for school which is composed of ℎ_ share of primary-level students, ℎ_ share of lower-secondary students, and ℎ_ share of upper-secondary students. Notice that the respective shares for the baseline school are 0.455, 0.276, and 0.135 respectively. The schooling level composition index for school can be computed using the formula: exp{(ℎ_ − 0.455) ̂ℎ_ + ̂ℎ_ + (ℎ_ − 0.276) (ℎ_ − 0.135)̂ℎ_ } ̂ℎ_ and ̂ℎ_ , where ̂ℎ_ are the estimated Frontier IV marginal effects for the “Share of primary students”, “Share of lower secondary students”, and , “Share of upper secondary students” variables respectively. Conducting the preceding exercise for all schools in the data set, it is predicted that the average per student public expenditure adequate for achieving the stated student performance standard is 36,331 Baht per annum – a 15.4 percent increase from the actual average per-student subsidy of 31,475 Baht per annum. In order to see the required change in per-student public expenditure at a more disaggregated level, we average the required per-student subsidies for all provinces in Thailand. The required change in the average per-student amount by province is plotted against the provincial average annual per capita consumption and the provincial average student performance index in Figures BA5.1 and BA5.2 respectively. The broad patterns of the two figures indicate that poorer provinces with lower educational outcomes would require larger increases in per-student public expenditures in order to raise their students’ average educational performance to the required performance standard. In other words, public expenditure would need to be even more progressive than it currently is (recall Figure 3.1 in Chapter 3). Notice also that there will be losers from implementing the funding formula as four provinces would see their average per student public allocation decline. On the other hand, the network of small schools in Mae Hong Son (the poorest province in Thailand in terms of per capita consumption expenditure) would need a massive increase of 64 percent in per student financing. Figure BA5.1 Required Change in Per Student Subsidies vs. Per Capita Consumption – by Province 70% Required change in average per-student Mae Hong Son 60% 50% 40% public subsidy 30% 20% R² = 0.291 10% BKK 0% -10% -20% 30,000 50,000 70,000 90,000 110,000 130,000 150,000 Average annual consumption per capita (THB) Figure BA5.2 Required Change in Per Student Subsidies vs. Average Student Performance Index – by Province 70% Required change in average per-student Mae Hong Son 60% 50% 40% public subsidy 30% 20% R² = 0.260 10% 0% BKK -10% -20% 34 36 38 40 42 44 46 48 50 52 Average student performance index in 2010 Source: World Bank staff calculations based on Thailand Household Socioeconomic Survey 2011, OBEC 2010, and NIETS 2010. Distinguishing between factors that are and are not under the control of the school and/or local community Recall that the funding formula gives schools funding premiums for factors that are outside the control of the school or the local community, such as the share of students by level of schooling in the area, the shares of students who are poor or disabled, and the local price level of educational inputs. However, as we have seen from the analysis given in Chapter 4, most small schools need not be small since they are located in close proximity to other schools. Therefore, these non-isolated schools may not be regarded as small and may not receive the same funding premium that a similarly-sized isolated school would receive. The importance of the scale effects of enrolment and class sizes is clearly depicted in the bar charts shown in Figure BA5.3. Consider the left bar chart which shows the reduction in required per-student subsidy as the schools gets larger. The required per-student subsidy for a school with an enrolment size between 500 to 749 students is a massive 43 percent lower than that for a school with less than 50 enrolled students, other things held constant. The scale effect of average class size can be analyzed analogously.13 Figure BA5.3. Estimated Scale Effects of Enrolment and Average Class Size Percent Reduction in Cost Compared to Schools Percent Reduction in Cost Compared with Less than 50 Students to Schools with Average Class Sizes Less than 10 50% 45% 32% 40% 28% 35% 24% 30% 25% 20% 20% 16% 15% 12% 10% 8% 5% 4% 0% 0% Enrolment size Class size Source: World Bank staff calculations based on OBEC 2010. For the reasons given in the previous paragraph, it is crucial that a school mapping exercise be conducted prior to determining the required per-student subsidies. We calculate the amount of subsidies for all 19,081 schools that would remain after the hypothetical school consolidation reform. The national average amount of required funding for all schools to achieve a score of 50 on the Student Performance Index is estimated at 33,863 Baht per student. This amount is 7.6 percent higher than the actual average per-student subsidy of 31,475 Baht, which is considerably lower than the 36,331 Baht needed prior to the hypothetical school network rationalization reform. However, in order for Thailand to realize these efficiency gains, more than 12,000 schools will need to be consolidated. The strategies that Thailand could follow to realize this vision is discussed in Chapter 4. 13 The scale effects are calculated based on the estimated coefficients of the cost function model shown in Table A5.4. The required change in the average per-student amount by province post-school consolidation reform is again plotted against the provincial average annual per capita consumption and the provincial average student performance index in Figures BA5.4 and BA5.5. We can see that the broad patterns where poorer provinces with lower educational outcomes would require larger increases in per-student public expenditures remain unchanged. However, there will be more losers as 23 provinces would now face reductions in the allocation of per-student public subsidies. Figure BA5.4 Required Change in Per Student Subsidies vs. Per Capita Consumption – by Province (Post School Network Rationalization) 70% Required change in average per-student 60% 50% Mae Hong Son 40% public subsidy 30% 20% R² = 0.239 10% BKK 0% -10% -20% 30,000 50,000 70,000 90,000 110,000 130,000 150,000 Average annual consumption per capita (THB) Figure BA5.5 Required Change in Per Student Subsidies vs. Average Student Performance Index – by Province (Post School Network Rationalization) 70% Required change in average per-student 60% Mae Hong Son 50% 40% public subsidy 30% 20% R² = 0.208 10% 0% BKK -10% -20% 34 36 38 40 42 44 46 48 50 52 Average student performance index in 2010 Source: World Bank staff calculations based on Thailand Household Socioeconomic Survey 2011, OBEC 2010, and NIETS 2010. Technical Appendix to Annex A5 A5.1. Educational Cost Function This study uses a Cobb-Douglas per-student cost function specification. Specifically, (∑ (0 ) 1 2 exp⁡ ( , , ; ) = exp⁡ ′ =3 ) (1) where denotes teacher salary which is a proxy for the prices of school educational resources, is the level of output (student performance), is a vector of variables that controls for possible variation in resource needs due to differences in school and student body background characteristics that affect their performance, and the ’s are parameters to be estimated. Taking logarithm of equation (1) yields a log-linear model of the form: ( ) = 0 + 1 ln( ) + 2 ln( ) + ∑ ln⁡ ′ =3 (2) However, as pointed out in Duncombe and Yinger (2005) and Duncombe (2007), the observed data are per-student spending, not cost, which is defined as the minimum expenditure required to achieve a certain performance outcome given school and student body characteristics. To estimate the cost of education requires taking into account differences in school efficiency in resource utilization. A school may have higher per-student spending compared to another school with similar student performance level, school and student background characteristics, and input prices they face because of inefficient use of resources. It is becoming increasingly common in recent literature to attempt to directly control for variables that are related to these inefficiencies in the right hand side of equation (2). For example; Duncombe and Yinger (2005) and Duncombe (2007) use “efficiency related measures”, which are broadly categorized as fiscal capacity, competition, and factors affecting voter involvement in monitoring the government. Imazeki (2008) focuses instead on local competition by using the Herfindahl Index (HHI) as a proxy for local competition in school districts. As argued in Costrell, Hanushek, and Loeb (2008), finding a set of observed and measurable factors that determined efficiency in schools is clearly difficult. As a result, existing “efficiency controls” are rarely convincing measures of the full range of efficiency and do little to explain variations in education spending. This study employs a Stochastic Frontier Analysis (SFA) approach introduced by Aigner, Lovell, and Schmidt (1977) and Meeusen and van den Broeck (1977) to identify the points at or near the bottom of the observed spending patterns. This statistical method has been used by a number of studies to convert an expenditure equation to a cost function (see for example Duncombe, Ruggiero, and Yinger, 1996; Gronberg, Jansen, Taylor, and Booker, 2004, 2005; Chakraborty and Poggio, 2008). In particular, consider per-student expenditure, , incurred by school . By definition, it must be the case that ≥ ( , , ; ) (3) because ( , , ; ) is an education cost function. To incorporate the fact that per-student expenditure may be subject to random shocks outside the control of the school, rewrite equation (3) as { } ≥ ( , , ; )exp⁡ (4) where exp⁡ { } is { } captures the effects of random shocks on each school and ( , , ; )exp⁡ the stochastic cost frontier. The remaining excess of expenditure over the minimum attainable cost is attributed to a degree of inefficiency, exp{ }. Specifically, { } = ( , , ; )exp{ }exp⁡ (5) An appropriate measure of cost efficiency (Kumbhakar and Lovell, 2000) in a stochastic frontier setting is given by: ( , , ;) exp{ } = = exp{− } (6) which is the ratio of minimum cost attainable to observed expenditure. From equations (4) and (5), it follows that ≤ 1, with = 1 if and only if attains its minimum feasible value. Assuming that the cost frontier takes the log-linear Cobb-Douglas functional form given in equation (2), the stochastic frontier model in equation (5) can now be written as: ( ) = 0 + 1 ln( ) + 2 ln( ) + ∑ ln⁡ ′ =3 + + (7) where ≥ 0 is a one-sided error term representing cost inefficiency and is a two-sided error term. The two error components of the asymmetric “composed error” term = + are assumed to be independently distributed from each other. This study follows Meeusen and van den Broeck (1977) and makes the following distributional assumptions: 2) ~(0, ~. . . That is, the two-sided errors ’s are assumed to be i.i.d. normally distributed and the one- sided errors ’s are assumed to have i.i.d. exponential distribution.14 Furthermore, when the two error components are also independently distributed from the regressors in equation (7), then maximum likelihood will consistently estimate the parameters along with the distribution parameters and . The log-likelihood function for a sample of schools can be shown to take the following form (see Meeusen and van den Broeck, 1977): 2 2 − ln (, , 2 ) = ∑ =1 {− ln + 22 + ln Φ ( )− } (8) where Φ(∙) is the standard normal cumulative distribution function, and is the composed error. Dealing with Endogenous Explanatory Variables The assumption that the error components are independently distributed from the explanatory variables in equation (7) is problematic. Karakaplan and Katlu (2013) argue that increasing cost 14Other distributional assumptions have been proposed for the non-negative inefficiency term, such as the half normal (Aigner, Lovell, and Schmidt, 1977), the truncated normal (Stevenson, 1980), and the Gamma distribution (Greene, 2003). efficiency may release additional resources that can be reallocated to enhance student performance. As a result, school output may be correlated with the inefficiency component. Costrell, Hanushek, and Loeb (2008) assert that some school districts are composed of citizens that are more education- oriented than others and this characteristic is not captured by observable variables. These districts may therefore tend to spend more and have more highly performing students. This statistical association would clearly cause an upward bias in the relationship between spending and performance. Furthermore, since the cost function is derived under the assumption of perfectly competitive market, it is implicitly assumed that schools are “price-takers.” This assumption is unlikely to be tenable as variation in teacher salaries likely includes some discretionary variation which may be correlated with student performance and per-pupil spending. The presence of endogenous teacher salaries and student performance variables would result in inconsistent parameter estimates if conventional maximum likelihood estimation of stochastic frontier model is performed. A proper instrumental variable (IV) framework is needed to deal with the endogeneity issue in the SFA setting. One of only a few studies trying to address the endogeneity problem in SFA is Guan, Kumbhakar, Myers, and Lansink (2009). The authors propose a two-step procedure to handle endogenous explanatory variables. In the first step, consistent estimates of the slope parameters are obtained using instrumental variable regression. Notice that the estimated intercept term from the first step is, however, inconsistent for the cost frontier intercept. In the second step, they use the residuals from the first stage as the dependent variable to get the maximum likelihood estimates of the stochastic frontier distributional parameters and a “shift parameter” for the first stage-intercept. Specifically, in the first step estimate equation (7) using instrumental variable regression to obtain ln⁡ ̂1, ln( ) + ̂0, + ( ) = ̂2, ln( ) + ∑ ′ ̂ =3 , + ̂ (9) In the second step, take the residuals ̂ ’s from the first stage and use them as the dependent variable in the following stochastic cost frontier model with only an intercept term ̂ = 0 + + (10) The model in equation (10) is estimated using maximum likelihood as per equation (8) to 2 obtain consistent estimates of 2 = 2 + , = / , and the shift parameter 0 . Finally, a consistent estimate of the stochastic cost frontier intercept from equation (7) is constructed by ̂0, + ̂0 = ̂0 (11) Cost efficiency estimates for each school can be obtained using equation (6). However, this study uses an alternative point estimator for proposed by Battese and Coelli (1988): 1−Φ(∗ −∗ /∗ ) 1 2 = [exp{− }| ] = [ ] exp {−∗ + 2 ∗ } (12) 1−Φ(−∗ /∗ ) 2 where ∗ = − / and ∗ = for the normal-exponential model. Predicting the Per-Student Cost in Level Using equations (9), (10), and (11); the predicted log per-student expenditure of a particular school can be written as: ̂ ln⁡ ̂1 ln() + ̂0 + () = ̂2 ln() + ∑ ̂ =3 ′ + (13) where is the provincial average composition-adjusted teacher salary, is the level of output (student performance index), is a vector of school and student body characteristics, is the national average ̂ ’s are the estimated regression coefficients. Notice that we have level of cost inefficiency, and the omitted the subscript to simplify notation. To convert the predicted log per-student expenditure of the school into the predicted cost in level term (in Thai Baht unit), we apply the following formula: ̂ ̂ = exp{ln⁡ ()}exp{ 2 ̂ /2} (14) 2 ̂ where is the variance of the predicted two-sided residuals. A5.2. Data The education cost function described in the previous section of this report is estimated based on 2010 cross-sectional school data collected by the Office of the Basic Education Commission (OBEC), the Ministry of Education in Thailand. A rich set of information for 31,330 schools were collected on teacher and personnel salaries, number of students in each education level, number of disabled students, number of classes, number of teachers by their educational qualification, personnel teaching experience, etc. However, a small percentage of observations were dropped due to the presence of missing data on one or more of the variables summarized in Table A5.1. The per student cost summarized in the same Table is calculated from the sum of subsidies schools received under the government’s “15-Year Free Basic Education Program” and the total expenditure on education personnel, divided by the number of students enrolled in that school. The subsidies under the 15-year free basic education program can be classified into five categories as follows; direct per student tuition subsidy, uniforms, learning materials, textbooks, and other activities which promote quality improvements among students. The expenditure for education personnel covers salaries of school administrators, teachers, and other educational staffs. All calculations are based on an annual basis. Also included in Table A5.1 are the summary statistics of the variables used in estimating the cost function model. All the variables are categorized into four major groups as follows; 1) teacher characteristics, 2) cost variables, 3) Student body characteristics, and 4) School characteristics. A5.3. Student Performance Outcome The student performance outcome measure used in this report is a weighted index of mathematics and science scores in the 2010 Ordinary National Education Test (O-NET) exams for Grades 6, 9, and 12. These tests are centrally administered by the National Institute of Educational Testing Service (NIETS) and all students in these respective grades are required to sit the exams. The rest of this subsection describes the procedure used in constructing the overall school outcome measure needed for our cost function estimation. For each school, we have information on the total number of students who took the exams by grade and subject. Also reported are the number of students that falls into each of the twelve test score intervals ranging from; (0), (0.01-10), (10.01-20), (20.01-30),…, (80.01-90), (90.01-99.99), and (100). Histograms representing densities of the 2010 O-NET scores in mathematics and science for the three grade levels are presented in Figure A5.1. To calculate the average test score for each exam, we first assign students within each test score interval the upper bound score for that particular interval (see table below). The school average test score for each exam is then simply calculated as a weighted average of the assigned scores, where the weights are the number of students in the test score intervals. The school average O-NET score for each grade level is in turn a weighted average of the mathematics and science test scores, where the weights are the total number of students who took the exams. The resulting empirical cumulative distribution and density functions for the student performance measures in the three grade levels are depicted in Figure A5.2. Assigning 2010 O-NET Scores to Students in each Test Score Interval 0.01- 10.01- 20.01- 30.01- 40.01- 50.01- 60.01- 70.01- 80.01- 90.01- Test score interval 10 20 30 40 50 60 70 80 90 100 Assigned score 10 20 30 40 50 60 70 80 90 100 Notice that the distributions of the average test scores in the three grade levels are quite different from each other. The challenge here is to find a way to appropriately combine the average scores for the three different grades into a single measure of student performance outcome for each school. This study uses the distribution of the Grade 6 O-NET scores (G6 scores from hereon) as the reference distribution, where the scores from the other two grade levels are projected onto. Specifically, we first compute the empirical cumulative distribution function (ECDF) for the G6 scores using the formula: ̂ ( 6 ≤ ) = ∑ ̂6 () ≡ 6 (6 ≤ ) (15) ̂6 (∙) is the ECDF of the G6 scores, 6 is the random variable representing the school mean where ̂ (∙) denotes probability, and (∙) is an indicator function equal to unity if the expression G6 score, 6 6 ∑ 6 6 in the parentheses is true. Each school is assigned a weight = / , where is the total number of students in school who took the G6 O-NET exams. Equation (15) is evaluated at some school average test score . In the next step, we similarly compute the ECDF’s for the test scores in the other two grade levels and then apply the inverse of the reference ECDF (also known as the quantile function), ̂6 −1 (∙) , to the values of the ECDF’s of the other test scores to obtain the rescaled scores. For example, consider the school average G9 score at the ℎ quantile (9 ), where ∈ (0,1). The rescaled G9 ℎ score at the quantile (9_ ) is computed as follows: ̂6 −1 ̂ ( )) (9 9 = 9_ (16) In the final step the student performance index is computed as a weighted average of the rescaled test scores from the three standardized exams. Specifically, for any school we compute the student performance outcome index, , using the formula: 6 +9 12 6 9_ + 12_ = 6 +9 +12 (17) The empirical cumulative distribution and density function for the constructed student performance index are presented in Figure A5.3. Figure A5.3: Cumulative Distribution and Density Functions for the Student Performance Index CDF - Student Performance Index Density Function - Student Performance Index .06 1.1 1.0 0.9 0.8 0.7 .04 Density 0.6 0.5 0.4 .02 0.3 0.2 0.1 0.0 0 20 40 60 80 100 0 Student performance index 0 20 40 60 80 100 Student performance index Source: World Bank staff calculations based on OBEC and NIETS. It is decided that the student performance index of 50 represents the required standard of proficiency used in this report. This is equivalent to the 84th percentile of the national student performance outcome distribution. In other words, 84 percent of students in Thailand performed at or below this level in the 2010 O-NET exams in mathematics and science. Figure A5.1. Distribution of O-NET Scores in Mathematics and Science by Grade Level (2010) ONET Mathematics - Grade 6 ONET Science - Grade 6 30% 25% 25% 20% 20% 15% 15% 10% 10% 5% 5% 0% 0% ONET Mathematics - Grade 9 ONET Science - Grade 9 45% 35% 40% 30% 35% 25% 30% 25% 20% 20% 15% 15% 10% 10% 5% 5% 0% 0% ONET Mathematics - Grade 12 ONET Science - Grade 12 50% 60% 45% 40% 50% 35% 40% 30% 25% 30% 20% 15% 20% 10% 10% 5% 0% 0% Source: World Bank staff calculations based on OBEC and NIETS. Figure A5.2: Cumulative Distribution and Density Functions for the 2010 O-NET Scores Cumulative Distribution Function - G6 ONET Density Function - G6 ONET 2010 1.1 .05 1.0 0.9 .04 0.8 0.7 .03 Density 0.6 0.5 0.4 .02 0.3 0.2 .01 0.1 0.0 0 20 40 60 80 100 0 School average G6 ONET score 0 20 40 60 80 100 School average G6 ONET score Cumulative Distribution Function - G9 ONET Density Function - G9 ONET 2010 .15 1.1 1.0 0.9 0.8 .1 0.7 Density 0.6 0.5 0.4 .05 0.3 0.2 0.1 0.0 0 20 40 60 80 0 School average G9 ONET score 0 20 40 60 80 School average G9 ONET score Cumulative Distribution Function - G12 ONET Density Function - G12 ONET 2010 .2 1.1 1.0 0.9 .15 0.8 0.7 Density 0.6 .1 0.5 0.4 0.3 .05 0.2 0.1 0.0 0 20 40 60 80 0 School average G12 ONET score 20 40 60 80 School average G12 ONET score Source: World Bank staff calculations based on OBEC and NIETS. A5.4. Composition-Adjusted Teacher Salaries From the summary statistics shown in Table A5.1, it can be inferred that total salary expenditure makes up 57 percent of total per-student public subsidy in Thai schools. Furthermore, teacher salary accounts for most (around 94 percent) of total salary expenditure per student. It is therefore very clear that teacher salary is the most important resource price to include in the cost function model. Moreover, it also serves as a proxy for the market prices for other school educational resources such as non-teaching staff. However, while data on school average salary of teachers are readily available, they do not only contain variation in the “prices” of teachers across schools. For instance, average teacher salaries can also vary across schools due to differences in teacher educational qualification, years of teaching experience, past performance assessment, as well as academic standing allowance and position allowance of teacher civil service. Figure A5.4: Unadjusted Average Teacher Salary (THB per month) 35,000 30,000 25,000 20,000 15,000 10,000 5,000 0 Nakhon ratchasima Sakon nakhon Chaiyaphum Sakaeo Maha sarakham Lampang Nan Bangkok Metropolis Saraburi Phichit Phetchabun Suphanburi Samut Prakan Pathum thani Rayong Si sa ket Ubon ratchathani Uttaradit Samut songkhram Ranong Yala Trat Trang Buri ram Khon kaen Udon thani Loei Nakhon sawan Prachuap khiri khan Phangnga Songkhla Yasothon Chiang mai Lamphun Phayao Uthai thani Tak Phitsanulok Nakhon pathom Naratiwat Ayutthaya kalasin Satun Lop buri Chon buri Chachoengsao Prachin buri Nong bua lam phu Nong khai Roi et Nakhon phanom Phuket Nonthaburi Mukdahan Samut sakhon Chiang rai Sukhothai Ratchaburi Surat thani Chumphon Pattani Ang thong Sing buri Chai nat Mae hong son Chanthaburi Nakhon nayok Surin Phrae Kamphang phet Kanchanaburi Nakhon si thammarat Krabi Phatthalung Phetchaburi Amnat charoen Source: World Bank staff calculations based on OBEC 2010. Figure A5.4 presents the raw or “unadjusted average teacher salaries” in OBEC schools across 76 provinces in Thailand. We can immediately see the large provincial differences in average teacher salaries. In order to measure teacher prices more accurately, we need to measure average salaries for teachers of comparable average characteristics across schools. This can be done by adjusting average teacher salaries using regression. It is important to note that for each school we observe only the shares of teachers by educational qualification and the average years of teaching experience in our dataset. We do not, however, observe other teacher quality measures such as past performance assessment and teacher academic standing which are clearly associated with salaries. Therefore, in addition to pure differences in the price of teachers, the resulting “composition-adjusted average teacher salaries” will also inevitably contain variations arising from differences in the unobserved teacher characteristics. To construct the composition-adjusted average teacher salaries, we first regress the log of average teacher salary on the shares of teachers in the school with doctorate, master’s, bachelor’s, and lower than bachelor’s degree, and a quartic in the average years of teaching experience. The regression results presented below in Table A5.2 indicate that teacher educational qualification and experience explain as much as 96 percent of the total variation in average salaries across schools. In the second step, we then predict what the logarithm of average teacher salary would be for each school if teacher education and experience are at the country average levels. It is important to note that we need to add the regression residuals from the first step to the predicted log average salary in this second step. The variation in the adjusted average teacher salaries is therefore due to the residuals of the log salaries from the first step alone. The residual wage variation therefore encompasses variations arising from teacher price differences, the discretionary wage component, and other teacher characteristics unobserved by the researcher. Figure A5.5: Composition-Adjusted Average Teacher Salary (THB per month) 30,000 25,000 20,000 15,000 10,000 5,000 0 Nakhon sawan Khon kaen Phichit Chaiyaphum Saraburi Sakaeo Rayong Maha sarakham Lampang Nan Suphanburi Bangkok Metropolis Phetchabun Samut songkhram Ranong Samut Prakan Pathum thani Si sa ket Ubon ratchathani Uttaradit Phayao Trang Trat Prachuap khiri khan Buri ram Songkhla Ayutthaya Yasothon Udon thani Loei Chiang mai Lamphun Uthai thani Tak Phitsanulok Nakhon pathom Phangnga Lop buri kalasin Naratiwat Nakhon phanom Satun Yala Ang thong Chon buri Chachoengsao Prachin buri Nong bua lam phu Nong khai Roi et Mukdahan Nonthaburi Phrae Chiang rai Mae hong son Samut sakhon Krabi Phuket Ratchaburi Surat thani Chumphon Phatthalung Pattani Sing buri Chai nat Nakhon nayok Chanthaburi Nakhon ratchasima Surin Sakon nakhon Kamphang phet Sukhothai Kanchanaburi Nakhon si thammarat Phetchaburi Amnat charoen Source: World Bank staff calculations based on OBEC 2010. In the final step, we simply exponentiate the predicted log average salaries to obtain the predicted average salaries in level terms across schools. The results for provincial average adjusted salaries presented in Figure A5.5 clearly show relatively little variation across provinces compared to the unadjusted salaries in Figure A5.4. A5.5. Instrumental Variables As discussed at length in the previous section, it is highly likely that the student performance variable is endogenous in the cost equation and instruments are needed in order to consistently estimate the model parameters. This study follows a similar approach taken by Duncombe and Yinger (2005) and Gronberg, Jansen, Taylor, and Booker (2005) and uses as instrument the median student performance of schools in the same province, but not in the same district as the school in question. Since there is only a single instrument for each endogenous variable, it is not possible to test the validity of the instrument using an overidentifying restriction test. However, the instrument does pass the necessary rank condition test as can be seen from the results from the first stage regressions shown in Table A5.3 where the instruments enter the equations highly significantly with a p-values of less than 0.01. Table A5.1. Descriptive Statistics Variables Observations Mean S.D. Min Max Student performance index 30,366 42.13 7.97 10 95 Cost variables Per-student expenditure 30,366 31,470 8,388 12,197 833,013 Salary expenditure per student 30,366 17,840 8,659 902 820,494 Teacher salary per student 30,366 16,792 8,349 902 816,312 Non-salary expenditure per student 30,366 13,631 2,359 10,497 18,389 Teacher characteristics Teacher salary 30,852 26,673 5,677 2,647 47,113 Share of teachers with doctorate degree 30,852 0.001 0.006 0.000 0.250 Share of teachers with master’s degree 30,852 0.124 0.099 0.000 1.000 Share of teachers with bachelor’s degree 30,852 0.835 0.123 0.000 1.000 Share of teachers with lower than bachelor’s degree 30,852 0.041 0.088 0.000 1.000 Years of potential experience 30,852 22.610 6.654 0.000 39.917 Student body characteristics Share of pre-primary students 30,366 0.134 0.109 0.000 0.642 Share of primary students 30,366 0.455 0.334 0.000 1.000 Share of lower secondary students 30,366 0.276 0.255 0.000 1.000 Share of upper secondary students 30,366 0.135 0.201 0.000 1.000 Share of poor students 30,366 0.489 0.362 0.000 1.000 Share of disabled students 30,366 0.002 0.011 0.000 0.297 School characteristics Enrolment size: Less than 50 students 30,366 0.107 0.309 0.000 1.000 50 to 74 students 30,366 0.107 0.309 0.000 1.000 75 to 119 students 30,366 0.108 0.311 0.000 1.000 120 to 159 students 30,366 0.121 0.326 0.000 1.000 160 to 199 students 30,366 0.113 0.317 0.000 1.000 200 to 299 students 30,366 0.113 0.317 0.000 1.000 300 to 499 students 30,366 0.115 0.319 0.000 1.000 500 to 699 students 30,366 0.118 0.322 0.000 1.000 700 to 899 students 30,366 0.044 0.206 0.000 1.000 900 to 1499 students 30,366 0.022 0.145 0.000 1.000 1500 to 2499 students 30,366 0.016 0.126 0.000 1.000 2500 students or above 30,366 0.016 0.124 0.000 1.000 Class size: Less than 10 students 30,366 0.265 0.441 0.000 1.000 10 to 19 students 30,366 0.376 0.484 0.000 1.000 20 to 29 students 30,366 0.252 0.434 0.000 1.000 30 to 34 students 30,366 0.060 0.238 0.000 1.000 35 to 39 students 30,366 0.028 0.166 0.000 1.000 40 to 44 students 30,366 0.013 0.114 0.000 1.000 45 students or above 30,366 0.006 0.076 0.000 1.000 Table A5.2. Teacher Salary Regression Results Log teacher Variables salary Teacher qualification: (Share of teachers with lower than bachelor’s degree) Share of teachers with doctorate degree 0.295*** (0.087) Share of teachers with master’s degree 0.090*** (0.007) Share of teachers with bachelor’s degree 0.029*** (0.006) Average years of teaching experience 0.094*** (0.003) Average years of teaching experience2 -0.003*** (0.000) Average years of teaching experience3 /100 0.008*** (0.001) Average years of teaching experience4 /1000 -0.001*** (0.000) Intercept 8.950*** (0.013) Observations 30,852 R-squared 0.958 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table A5.3. First Stage Regression Results Log student Variables Log teacher salary performance Log teacher salary instrument 0.719*** 0.055 (0.025) (0.105) Log student performance instrument 0.017*** 0.611*** (0.006) (0.028) Share of primary students 0.007 0.005 (0.006) (0.024) Share of lower secondary students 0.012** -0.176*** (0.005) (0.024) Share of upper secondary students 0.038*** 0.073* (0.007) (0.038) School enrolment size: (Less than 50 students) 50 to 69 students 0.000 0.027*** (0.002) (0.006) 70 to 89 students 0.004* 0.040*** (0.002) (0.007) 90 to 119 students 0.005** 0.054*** (0.003) (0.009) 120 to 149 students 0.008*** 0.060*** (0.003) (0.009) 150 to 199 students 0.007** 0.063*** (0.003) (0.010) 200 to 279 students 0.008*** 0.068*** (0.003) (0.011) 280 to 499 students 0.009*** 0.065*** (0.003) (0.012) 500 to 749 students 0.011*** 0.059*** (0.003) (0.012) 750 to 1149 students 0.014*** 0.083*** (0.003) (0.013) 1150 to 1999 students 0.012*** 0.123*** (0.004) (0.015) 2000 students or above 0.019*** 0.206*** (0.004) (0.019) Class size: (Less than 10 students) 10 to 19 students -0.003 0.006 (0.002) (0.007) 20 to 29 students -0.002 -0.006 (0.002) (0.009) 30 to 34 students -0.003 -0.025** (0.003) (0.010) 35 to 39 students 0 -0.037*** (0.003) (0.014) 40 to 44 students 0.001 -0.014 (0.004) (0.019) 45 students or above 0.001 0.056** (0.004) (0.022) Table A5.3. First Stage Regression Results (continued) Log student Variables Log teacher salary performance Share of students poor -0.004*** -0.034*** (0.001) (0.005) Share of students disabled 0.003 -0.300* (0.034) (0.161) Intercept 2.794*** 0.867 (0.254) (1.056) Observations 30,366 30,366 R-squared 0.165 0.207 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Table A5.4. Results from the Cost Function Model Variables Frontier IV OLS Log teacher salary 0.788*** 0.611*** (0.159) (0.034) Log student performance 0.699*** 0.188*** (0.044) (0.008) Share of primary students 0.708*** 0.713*** (0.030) (0.028) Share of lower secondary students 0.966*** 0.892*** (0.029) (0.025) Share of upper secondary students 0.681*** 0.724*** (0.036) (0.029) School enrolment size: (Less than 50 students) 50 to 69 students -0.191*** -0.179*** (0.009) (0.008) 70 to 89 students -0.224*** -0.204*** (0.010) (0.009) 90 to 119 students -0.262*** -0.237*** (0.013) (0.011) 120 to 149 students -0.316*** -0.287*** (0.013) (0.011) 150 to 199 students -0.369*** -0.340*** (0.013) (0.012) 200 to 279 students -0.437*** -0.406*** (0.014) (0.013) 280 to 499 students -0.507*** -0.478*** (0.015) (0.013) 500 to 749 students -0.553*** -0.527*** (0.016) (0.014) 750 to 1149 students -0.566*** -0.524*** (0.018) (0.016) 1150 to 1999 students -0.583*** -0.521*** (0.020) (0.018) 2000 students or above -0.596*** -0.484*** (0.022) (0.019) Class size: (Less than 10 students) 10 to 19 students -0.094*** -0.092*** (0.010) (0.009) 20 to 29 students -0.192*** -0.199*** (0.012) (0.011) 30 to 34 students -0.250*** -0.268*** (0.014) (0.013) 35 to 39 students -0.279*** -0.303*** (0.017) (0.014) 40 to 44 students -0.324*** -0.339*** (0.020) (0.017) 45 students or above -0.375*** -0.348*** (0.022) (0.018) Table A5.4. Results from the Cost Function Model (continued) Variables Frontier IV OLS Share of students poor 0.010* -0.013** (0.006) (0.005) Share of students disabled 0.287** 0.196 (0.142) (0.185) Intercept -0.344 3.377*** (1.565) (0.344) ln(σv2) -3.279*** (0.037) ln(σu2) -6.544*** (0.830) Observations 30,366 30,366 R-squared 0.458 Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Thailand A Quality Education for All THE WORLD BANK 30th Floor, Siam Tower, 989 Rama 1 Road, Pathumwan, Bangkok 10330 Tel: (66) 0-2686-8300 Fax: (66) 0-2686-8301 E-mail: thailand@worldbank.org www.worldbank.org/th