Report No. 28779 Books, Buildings, and Learning Outcomes: An Impact Evaluation of World Bank Support To Basic Education in Ghana April 27, 2004 Operations Evaluation Department Document of the World Bank i Foreword Bank support to basic education has increased greatly over the last two decades. The Ghanaian experience provides a useful test case o f the effectiveness o f this support. Since 1986 there have been ten Bank education sector projects in Ghana, o f which five have directed support to basic education: the Health and Education Rehabilitation Project, which supplied school learning materials; two education sector adjustment credits in support o f the reform program; the Primary School Development Project; and the Basic Education Sector Investment Credit. The main questions addressed in the OED study are: (1) what has happened to educational outputs (school attendance and learning); (2) what are the main determinants o f those outputs; (3) which educational interventions have the largest and most cost effective impact on educational outputs; (4) to what extent have Bank-supported activities promoted interventions which support improved educational outputs; and (5) how do improved educational outputs support better welfare outcomes? These questions were addressed through a variety o f means, including a nationwide survey carried out by OED in collaboration with Ghana Statistical Service and the Ministry o f Education, Youth and Sports. The survey followed up on a living standards survey conducted in 1988 that included data on test score outcomes and school quality. The study is thus in a unique position to analyze school-level changes over the 15 year period, 1988-2003. A major finding o f the study i s that both the quantity and quality o f schooling have improved over the last fifteen years. Enrolments in basic education have increased by over 10 percent compared to 15 years ago. Moreover, 15 years ago nearly two-thirds o f primary school graduates were illiterate, as shown by the fact that they scored two or less o n a simple eight question multiple choice English test - the same as guessing. Less than one in five do so badly today. Statistical analysis shows that these improvements in leaming outcomes are clearly and strongly linked to better welfare as measured by higher income, better nutrition, and reduced mortality. Analysis o f the economic rate o f return to education shows that there i s no retum to simply attending school, but there i s a retum to learning achievements. The majority o f children now benefit from attending school, both educationally and economically, which was not the case 15 years ago. The data show that gains in educational outputs are directly linked to better school quality, manifested in improved infrastructure and greater availability o f school supplies. Today it is the norm to have one textbook per child for math and English, rather than one per class as was common before the advent o f reforms. Text book provision i s amongst the most cost effective means o f improving test scores. School building has contributed to higher enrolments. In one area surveyed in which a new school was constructed enrolments more than tripled. These gains are impressive, but there remains substantial room for improvement. Enrolments lag in some parts o f the country, and while test scores are improved they are s t i l l weak. Statistical analysis shows that increased school quality can in turn be linked to the Bank’s support which has financed the construction o f 8,000 classroom blocks and provided 35 million textbooks over the last 15 years. Nationally, the Bank supported school building .. 11 and rehabilitation program has increased enrolments by around four percent. Moreover Bank support helped sustain initially unpopular reforms, demonstrating the efficacy o f working in partnership with a government committed to a well-defined sectoral strategy. The downside o f this positive story is that increased reliance o n community and district financing means that schools in poorer areas get l e f t behind, especially those in off- road rural communities. There are still some schools with very poor facilities in which little learning takes places. The lessons drawn from this study are: 0 increasing the availability and quality o f classrooms and instructional materials directly contributes to both educational attainment and achievement; 0 supervision o f teachers by the head teacher and circuit supervisor matter, as do the teaching methods adopted by the teacher, including the language used as the medium o f instruction, so efforts should also be made to retain trained teachers, to improve teacher morale, and to expand in-service training; 0 a class o f schools in poorer communities are very poorly resourced, so resources should be directed to the most needy schools to overcome the bias that results from community-based financing; and 0 the private sector has been neglected, although it i s o f growing importance; attention needs to be paid to i t in both government strategy and Bank support. Gregory K. Ingram Director-General Operations Evaluation ... 111 Contents Preface.............................................................................................................................. .. vi1 Executive Summary ......................................................................................................... ix 1. Introduction.................................................................................................................. 1 Scope and Purpose ofthe Study ............................................................................... 1 Education and the International Development Agenda ............................... 1 Education in Ghana...................................................................................... 1 Evaluation Questions: What Explains Educational Performance? ......................... 2 Approach ...................................................................................................... 5 Outline.......................................................................................................... 5 2 . Changes in Basic Education Since the 1980s............................................................. 7 The Education System Before 1986 ......................................................................... 7 The 1986 Reform Program ...................................................................................... 8 FCUBE and Decentralization .................................................................................. 9 Budget .................................................................................................................... 11 School-Level Inputs................................................................................................ 12 Physical and Material Inputs...................................................................... 12 Material Inputs ............................................................................... 13 Physical (building) Inputs.............................................................. 14 Allocation o f Material and Physical Inputs ................................... 15 Teachers ......................................................................................... 16 School Management................................................................................... 18 3 . The Bank’s Education Portfolio in Ghana .............................................................. 20 Overview ................................................................................................................ 20 The Sector Adjustment Credits: EdSAC I and 1 1 .................................................... 22 Objectives .................................................................................................. 22 U s e o f Funds .............................................................................................. 22 Conditionality and Reforms ....................................................................... 23 Investing in Basic Education: PSD and BESIP ..................................................... 25 Objectives .................................................................................................. 25 Project Components and U s e o f Funds ...................................................... 26 The Role of Other Donors...................................................................................... 27 iv 4 . The Bank’s Impact on Education Policies and Outputs......................................... 28 The Bank and Educational Reform ........................................................................ 28 Sustainability o f Reform ............................................................................ 31 Donor Coordination ................................................................................... 31 Education Sector Outputs ...................................................................................... 32 Budget ........................................................................................................ 32 Activities .................................................................................................... 33 5 . Educational Performance Has Improved ................................................................ 36 School Attainment: Higher Enrolments and Better Completion Rates ..................36 Improved Test Scores ............................................................................................. 38 6 . Causes and Consequences o f Improved Educational Performance ........... 42 Better Inputs ........................................................................................................... 42 Improving School Efficiency..................................................................... 42 Improvements in School Facilities............................................................. 42 Improving Teacher Conditions .................................................................. 42 From Inputs to Outputs (Attainment and Achievement) ........................................ 43 Increasing Enrolments ............................................................................... 43 Determinants o f Test Scores ......................................................................45 From Outputs to Outcomes .................................................................................... 46 Education and Social Outcomes ................................................................ 46 Education and Economic Outcomes .......................................................... 48 7 . Lessons Learned and Progress Toward t h e MDGs ................................................ 50 Progress Toward the Millennium Development Goals......................................... -50 Lessons Learned..................................................................................................... 51 Annex A: Test Examples ................................................................................................. 53 Annex B: Budget Analysis ............................................................................................... 57 Annex C: School Costs..................................................................................................... 68 Annex D: School-Level Changes in Inputs, Management, and Methods ................... 76 Annex E: Tables Of School Quality Variables ............................................................ 116 Annex F: Variable Definition........................................................................................ 130 Annex G: Analysis o f Test Scores ................................................................................. 133 Annex H: Data on Educational Performance .............................................................. 151 Annex I:School Attainment .......................................................................................... 171 V Annex J: Conditions Attached to Bank Credits ............................................................. 185 Annex K: Education and Welfare Outcomes ................................................................. 196 Annex L: Evaluation Approach Paper ........................................................................... 202 References .......................................................................................................................... 215 Table 1.1: Coverage o f data collection instruments ................................................................ 5 Table 3.1: World Bank support to Ghana’s education sector. 1986-2003 ............................ 21 Table 3.2: Allocation o f resources under EdSAC I and I1.................................................... 23 Table 3.3 : Allocation o f resources under PSD project and BESP ....................................... 26 Table 4.1: Activities in World Bank basic education projects.............................................. 34 Table 5.1: Drop-out rates are low (2003) .............................................................................. 38 Table 5.2: Student performance i s returning to the levels attained 40 years ago (average test scores o f primary school leavers by decade) ............................................. 40 Table 6.1 : Results from studies o f education and social outcomes in Ghana ....................... 47 Table 6.2: Percentage reductions in welfare outcomes caused by higher school attainment ............................................................................................................ 48 Table 7.1 : Progress toward the education related MDGs ..................................................... 50 Figures Figure 1.1: Ghana’s education system went into decline in the mid-1970s ... starting a slow but steady recovery since the mid-1980s .................................................. 2 Figure 1.2 How educational inputs affect welfare outcomes .................................................. 4 Figure 2.1 : Restructuring o f education system ....................................................................... 9 Figure 2.2: Government spending on education has risen: central government education expenditure ..................................................................................................... 11 Figure 2.3: Schools in nearly all areas have more material inputs than before: cluster-level material inputs to school quality ................................................................ 13 Figure 2.4: The quality o f school infrastructure has improved in most areas: cluster-level physical inputs to school quality ................................................................ 14 Figure 2.5: The proportion o f teachers who are trained rose ............................................... 16 Figure 2.6(a): Efficiency gains have been realized by increasing the pupil-teacher ratio ................................................................................................................................. 17 Figure 3.1: Bank disbursements o n an annual and project basis, FY86-02 .......................... 21 Figure 5.1 : More children are attending school (attendance rates by age group) ................. 37 v1 Figure 5.2: And gender and regional gaps are closing (enrolment differentials. 6-1 1 year olds) ................................................................................................................ 37 Figure 5.3: M o r e children finish school (primary completion rates ) ................................... 38 Figure 5.4: School outputs have improved: test scores plotted against years of schooling ..................................................................................................................... 39 Figure 5.5: Criterion Reference Test scores in public schools have risen each year. 1992-2000 ....................................................................................................................... 41 vii Preface This Operations Evaluation Department report has been prepared in response to a request from the W o r l d Bank’s Board o f Executive Directors that the department resume work o n impact evaluation. OED has a long history o f conducting such studies, in which the meaning o f impact has been interpreted in various ways. I nthis report, impact is taken to imply a concern with final welfare outcomes, and the attempt to establish the counterfactual to isolate the effects o f different determinants o n those outcomes. The study thus traces the causal chain from inputs through to welfare outcomes. The data constraints and methodological challenges facing such analysis are well known. OED hopes to demonstrate that these challenges can be overcome to illustrate how the Bank’s activities contribute toward the alleviation o f global poverty. This report has been prepared by Howard White with the assistance o f Edoardo Masset. Preparation o f the study was assisted by Alain Barbu and Martha Ainsworth and contributions from the peer reviewers Kwame Akyeampong and Paul Glewwe and from Dean Nielsen. Thanks are due to the co-operation o f Ghana Statistical Service (GSS) and the Ministry o f Education (MoE) in preparing and implementing the survey -KB Danso-Manu and Thomas Coleman deserve particular mention - and to the Bank’s Ghana education team (BenBit M i l l o t and Eunice Dapaah) for their cooperation. M a r y Esther Dakubu, o f the Institute o f Afkican Studies, and K w e k u Osam, Linguistics Department, University o f Ghana, were responsible for the preparation o f the local language questionnaire used in this study. The following World Bank staff provided comments: Helen Abadzi, Victoria Elliot, Deon Filmer, N i l s Fostvedt, Patrick Grasso, BenBit Millot, Dean Nielsen, Halsey Rogers, and Yvonne Tsikata. William Hurlbut edited the report. Administrative support was provided by Pilar Barquero. Collaboration with GSS and M o E extended beyond the survey to data analysis as part o f OED’s support for evaluation capacity building. Staff from both agencies, with support from OED, conducted analysis o f the data collected for this study, which has been drawn o n in the preparation o f this report. This study was carried out under the partnership agreement between OED and the UK Department for International Development (DFID). ix Executive Summary 1. The Millennium Development Goals aim for universal primary education by 2015 and gender equality in enrolments at all levels o f education. The Education for All (EFA) initiative lays out a strategy for achieving these goals. The Bank’s o w n strategy stresses the school quality aspects o f EFA, emphasizing the need to focus o n preserving learning outcomes while access to education is expanded. This report assesses the impact to date o f the efforts over the past 15 years toward increasing the quantity and quality o f basic education in one African country, Ghana. 2. Ghana typifies many o f the challenges faced by African countries as they strive to meet the MDGs. Having established one o f the best education systems in Africa, the number o f children attending primary school began to fall in the mid-70s. School quality was falling with non- salary recurrent expenditures being squeezed out. Many schools had no more than one textbook to a class and the majority o f primary school graduates were illiterate. In 1986, the government embarked on an ambitious reform program to increase efficiency by restructuring pre- university education and increasing cost recovery among senior secondary and tertiary students, enabling resources to be re-allocated to basic education. In the mid-1990s a policy o f free, compulsory universal basic education (FCUBE) was launched. Since 1997, the education sector has been decentralized with increased community management and the introduction o f School Management Committees and School Performance Assessment Meetings. 3. The government’s efforts to improve education have been supported by the World Bank and other donors. The Bank’s assistance began with the Health and Education Rehabilitation Project (HEW), which supplied school learning materials. The reform program was supported by two education sector adjustment credits (EdSAC I and 19.These adjustment credits were followed by two investment projects: the Primary School Development Project and the Basic Education Sector Investment Credit (BESP). The resources provided by the Bank have been predominately used for school building and rehabilitation, and textbook supply. Through these five projects close to 35 million textbooks have been provided and 8,000 school pavilions constructed. Despite the emphasis o n the importance o f “software” in contemporary education strategies, the Bank’s lending has focused almost exclusively o n “hardware” and instructional materials (textbooks and teachers’ guides). 4. The Bank’s support helped the government carry out a reform program that was resisted by the teaching profession and some segments o f the population. The strong commitment shown by government, and firm actions it took to implement the reforms, demonstrate the high degree o f ownership. In that favorable context, the Bank’s financing reinforced the government’s position, allowed textbooks to be printed for the new syllabus in Junior Secondary Schools (JSS) and training for 40,000 JSS teachers to prepare them for the new system. W h i l e there was opposition to the cost recovery measures in second-cycle and tertiary institutions, the majority o f parents were more concerned with the quality o f basic infrastructure, the improvement o f which -with substantial Bank support -helped maintain the momentum behind the reforms. Alongside formal conditionality for restructuring the education system and introducing cost recovery measures, informal policy dialogue was greatly facilitated by the Bank’s senior education specialist being resident in Accra and developing a close working relationship with senior ministry officials. The Bank operated behind the scenes in facilitating donor co-ordination for the education sector, although donor competition meant that the anticipated sector-wide approach to education in the latter part o f the 1990s failed to materialize. 5. This study conducted a survey in collaboration with Ghana Statistical Service and the Ministry o f Education, Youth and Sports covering 1,740 households, 706 basic schools, and 3,129 teachers. This nationally representative survey was carried out in the same 85 areas o f the country as the education module o f the second round o f the Ghana Living Standards Survey in 1988/89, enabling a unique and detailed picture o f changes in schools over the 15- year period. These data show large improvements in school quality, especially with respect to material inputs. For example: 0 In 1988, less than h a l f o f schools could use all their classrooms when i t was raining, but in 2003 over two-thirds can do so. 0 Fifteen years ago over two-thirds o f primary schools reported occasional shortages o f chalk, only one in 20 do so today, with 86 percent saying there i s always enough. 0 The percentage o f primary schools having at least one English textbook per pupil has risen fi-om 21 percent in 1988 to 72 percent today and for math books in Junior Secondary School (JSS) these figures are 13 and 71 percent, respectively. 6. School quality has improved across the country: in poor and non-poor communities alike. But there is a growing disparity within the public school sector. Increased reliance on community and district financing means that schools in relatively prosperous areas continue to enjoy better facilities than do those in less well o f f communities. Future investments in school quality cannot be solely demand driven, which will tend to favor the better off. Demand-drivenprograms should be complemented by interventions in disadvantaged schools, which can be identified through the annual school census conducted as part o f the Education Management Information System (EMIS). 7. The importance o f the private sector has increased greatly in the last 15 years. Close to 20 percent o f the schools in the survey areas are private, compared to fewer than 5 percent five. years ago. Private schools are not all elite schools. Many are in relatively poor areas and many do not perform well on quality measures. 8. Improving school quality h as been accompanied by increased enrolments, which have grown by over 10 percent over the 15 years. By 2000, over 90 percent o f Ghanaians aged 15 and above had attended school compared to 75 percent 20 years earlier. In addition, drop-out rates have fallen, so completion rates have risen: by 2003,92 percent o f those entering grade 1 complete Junior Secondary School (grade 9). Gender disparities have been virtually eliminated in basic enrolments. Primary enrolments have risen inboth disadvantaged areas and amongst the lowest income groups. The differential between both the poorest areas and other parts o f the country, and between enrolments o f the poor and non-poor, have been narrowed but are still present. 9. Rising attainment has been accompanied by higher achievement. I t is no longer the case that most primary graduates are illiterate. The survey undertaken for this study conducted math and English tests among 9-55 year olds identical to the tests carried out 15 years ago, enabling xi a direct comparison o f learning outcomes. Today, less than a fifth o f those who have completed grades 3-6 scored two or less out o f eight o n the short Englishmultiple choice test -the same as guessing -compared to nearly two-thirds in 1988. Test scores are significantly higher today for both math and English. Children completing the nine years o f basic education in 2003 scored higher that those with ten years o f basic education under the o l d system 15 years ago. But the shortening o f post-basic education from seven to three years has had a small adverse impact o n learning outcomes amongst secondary graduates. 10. Using the English test results to measure literacy shows that the literacy rate among those aged 15-24 (one o f the MDG indicators) has risen from 49 percent to 68 percent between 1988 and 2003. The increase in school quality (higher scores achieved by those enrolled in school) accounts for over h a l f (57 percent) the increase in literacy, with the remainder coming f r o m increased quantity (higher enrolments). 11. Statistical analysis o f the survey results shows the importance o f school infrastructure o n enrolments. Building a school, and so reducing children’s travel time, has a major impact o n enrolments. While the majority o f children live within 20 minutes o f school, some 20 percent do not and school buildinghas increased enrolments among these groups. In one area surveyed, average travel time to the nearest school was cut by 45 minutes with enrolments increasing from 10 to 80 percent. In two other areas average travel time was reduced by nearly 30 minutes and enrolments increased by over 20 percent. Rehabilitating classrooms so that they can be used when it i s raining also positively affects enrolments. Complete rehabilitation can increase enrolments by as much as one third. Across the country as a whole, the changes in infrastructure quantity and quality have accounted for a 4 percent increase in enrolments between 1988 and 2003, about one third o f the increase over that period. A large part o f this improvement can be attributed to the W o r l d Bank, which has been overwhelmingly the main funder o f better infrastructure in this period. 12. Learning outcomes depend significantly o n school quality, including textbook supply. Bank-financed textbook provision accounts for around one quarter o f the observed improvement in test scores. But other major school-level determinants o f achievement such as teaching methods and supervision o f teachers by the head teacher and circuit supervisor have not been affected by the Bank’s interventions. The Bank has not been heavily involved in teacher training and plans to extend in-service training have not been realized. Support to “hardware” has been shown to have made a substantial positive contribution to both attainment and achievement. But when satisfactory levels o f inputs are reached -which i s s t i l l far from the case for the many relatively deprived schools - future improvements could come f i o m focusing o n what happens in the classroom. However, the Bank’s one m a i n effort to change incentives -providing head teacher housing under the Primary School Development Project in return for the head teacher signing a contract o n school management practices -was not a great success. Others, notably DFID and USAID, have made better progress in this direction but with limited coverage. 13. School building and rehabilitation has been a cost effective means o f increasing enrolments. Other activities are most cost effective in improving test scores, with textbook supply being one o f the most effective. The question for the Bank i s the balance to be xii maintained between these traditional, proven and s t i l l necessary activities and other activities such as promoting community management and enhancing the effectiveness o f teaching. 14. Better education leads to better welfare outcomes. Existing studies o n Ghana show h o w education reduces fertility and mortality. Analysis o f the survey data shows that education improves nutritional outcomes, with this effect being particularly strong for children o f women living in poorer households. Regression analysis shows there is n o economic retum to primary and JSS education, but there i s a retum to cognitive achievement. Children who attain higher test scores as a result o f attending school can expect to enjoy higher income; but children who leam little in school will not reap any economic benefit. 15. The lessons o f the Ghana education experience are: 0 Increasing the availability and quality o f classrooms and instructional materials directly contributes to both educational attainment and achievement. However, such a “hardware” approach will become less relevant as all schools attain the desired level o f quality. Ghana is not yet in that position: substantial inputs are still required for the most disadvantaged schools. Even where good school quality i s achieved, educational outcomes, while improved, are still far from satisfactory. 0 The evidence i s clear that supervision o f teachers by the head teacher and circuit supervisor matter, as do the teaching methods adopted by the teacher, including the language used as the medium o f instruction. Since attempts to remove untrained teachers have been unsuccessfbl, and since not all trained teachers appear familiar with improved methods anyway, there i s a strong case for pushing forward with in- service training. Efforts should also be made to retain trained teachers and t o improve teacher morale. Achieving both o f these means better teaching conditions, including paying teachers o n time. 0 The downside o f community and district financing o f schools is that it leads to disparities in resource availability. There remains a class o f schools inpoorer communities - particularly but not only inrural areas -which are very poorly resourced. Resources should be directed to the most needy schools to overcome the bias that results from community-based financing. School mapping continues to play an important role, which means that support to E M I S i s important. 0 While not a major part o f this study, it i s clear that the private sector has been neglected. But it i s o f growing importance so attention needs t o be paid to it in both govemment strategy and Bank support. 1 1. Introduction SCOPE AND PURPOSE OF THE STUDY Education and the International Development Agenda ‘Yll agree that the single most important key to development and to poverty alleviation is education. This must start with universal primary education for girls and boys equally... ” James D. Wolfensohn, January 1999l 1.1 Education i s central to international poverty reduction goals, as reflected in i t inclusion in two o f the Millennium Development Goals (MDGs): universal primary education and gender equality in school enrolments. Support for education has also manifested itself in the Education for All (EFA) initiative. Launched at Jomtien (Thailand) in 1990, the movement gained international support through a partnership o f UNESCO, UNICEF, UNDP, and the World Bank, and was given a further boost by the Dakar World Education Forum in April 2000.2 1.2 The World Bank’s Education Sector Strategy (World Bank 1999) is complementary to the framework o f action adopted at Dakar, with a stress on quality. The goal i s to “ensure that, by 2015, every b o y and girl in the developing world has access to and completes a free and compulsory primary education o f good q ~ a l i t y ” .The ~ emphasis on quality has led to a focus on issues such as parental and community participation and improved teaching methods, which are increasingly incorporated into project design. This study examines the impact o f external support provided by the World Bank on the achievement o f education goals in the case o f one African country, Ghana. Education in Ghana 1.3 Ghana’s education sector, once one o f the most respected in Africa, has faced difficult challenges in the past two decades. Basic education was expanded following independence, as was the case in neighboring countries. But by the mid-seventies the number o f children attending primary school in Ghana started to fall (Figure 1.1). In 1975 there were over 2.3 million children in primary school: this figure had fallen by over one million by the early eighties. Quality as well as quantity suffered. Non-salary recurrent expenditures were squeezed out; both falling real wages and frequent late payments demoralized the teaching force. The majority o f primary school graduates were illiterate. Meanwhile, government spending was excessively oriented toward the tertiary sector. 1. Quoted in World Bank Education Sector Strategy, July 1999, p. iii. 2. UNESCO The Dakar Framework for Action, Paris, 2000. 3. World Bank (2002) OpeningDoors: Education and the World Bank. 2 Figure 1.1: Ghana's education system went into decline in the mid-1970s starting a ... slow but steady recovery since the mid-1980s (index of total primary enrolments) 8 350 -? II E 0 250 ?i - '0 E 150 1970 1975 1980 1985 1990 1995 2000 -Ghana -Togo - - - - C6te d'lvoire Source: World Bank World Development Indicators, 2003 1.4 The government embarked on an ambitious reform program in 1986 to restructure pre-university education and introduce greater cost recovery at secondary and tertiary levels. These changes, together with the higher economic growth resulting from the economic reform program, led to a steady recovery in the number o f child attending school (Figure 1.1). While in principle there has always been free universal primary education in Ghana, fees charged at the local level have been one factor in restraining enrolments. Free compulsory universal basic education (FCUBE), introduced in 1996, aimed at eliminating these fees.4 Since 1997 education services have been decentralized, including the introduction o f School Management Committees and School Performance Assessment Meetings for increased community management and accountability. 1.5 The World Bank has supported these developments through 10 projects, o f which 5 have assisted basic education: the Health and Education Rehabilitation Project, the Education Sectoral Adjustment Credits I and 11, the Primary School Development Project, and Basic Education Sector Improvement Credit. Since 1986 the Bank has lent close to $260 million in support o f education in Ghana, accounting for close to h a l f o f all external assistance to the sector. QUESTIONS: WHATEXPLAINS EVALUATION PERFORMANCE? EDUCATIONAL 1.6 Many factors contribute to educational outcomes. Access to, and quality of, school facilities are important. But so i s the home environment, including the importance parents put o n their child's education and the time the child has to spend working in household or other enterprises. To what extent can improved educational outputs and the resulting welfare outcomes be attributed to the changes in school inputs and management and the support the 4. Basic education in Ghana i s primary (grades 1-6) and Junior Secondary School (JSS, grades 7-9). 3 Bank has provided to these? The challenge for this report is to answer the following five questions: a What changes have occurred to school attainment and achievement (education o ~ t p u t s ) ,including ~ the MDG indicators o f completion and gender equality in enrolments, in Ghana since the start o f reforms in 1986? a What are the determinants o f changes in basic educational outputs for children o f basic school age in Ghana? a Which education interventions have the greatest impact o n the determinants o f educational outputs? a What has been the role o f the Bank and other extemal donors in promoting education interventions that result in improved school attainment and achievement? a D o improved school attainment and achievement support better welfare outcomes as captured in the MDGs, such as lower child mortality, better nutrition, and reductions in income-poverty? 1.7 This report i s thus primarily concemed with determining changes in education outputs and outcomes and attributing, or not as the case may be, any improvements to activities supported by the Bank and other agencies. The study does not therefore cover the same ground as a country sector study, and i s less concemed with topics o f relevance, efficiency, and efficacy, which are usually central to OED’s approach. 1.8 The framework for this analysis i s provided by Figure 1.2. The ultimate concern is improved welfare, which i s the outcome of, among other things, the higher level o f education outputs. Cognitive development i s an output o f the education system. Producing this output requires that students attend and stay at school, with the quality o f the output depending o n the quality o f the various inputs, both hard (physical infrastructure) and soft (learning environment and methods). The World Bank has supported the inputs into the educational process both directly (e.g., financing school building) and indirectly (support t o policy reform). 5. “Educational attainment” refers to the highest level o f education and “educational achievement” to test scores. 4 Figure 1.2 H o w educational inputs affect welfare outcomes Education budget Policy reform allocation plus (influenced by donor programs policy dialogue) L School Learning Teacher Teacher School rehabilitation, materials training Pay supervision 1 1 - 1 1 1 - 1 - 1 1 111 111 1 9 1 1 1 \ 7 * Teacher morale Teacher quality I4 INTER- V V V Teaching and learning environment 1 OUTPUTS (school quaIity) 1 1 Education Outputs * Attainment (completion) Achievement OUTPUTS (learning gains, test scores) 1 * - Welfare outcomes OUTCOMES 5 OVERVIEW OF THE STUDY Approach 1.9 The evaluation framework for this study was developed through a literature review o f the determinants o f school attainment and achievement, a review o f the Bank’s portfolio o f education investments and an inception visit to Ghana.6Data collection focused on a household and school survey replicating the data collected in the second round o f the Ghana Living Standards Survey (GLSS2) in 1988/89. Interviews were carried out in 84 o f the 85 clusters covered by the 1988 survey, including 1,740 households, 704 schools and 3,129 teachers (Table 1.1 and Box 1.1). Achievement tests were taken by Table 1.1: Coverage of data collection instruments over 3,500 people. 1988 2003 Clusters 170 whole survey 84a 1.10 Quantitative data collection 85 education module was supplemented by fieldwork in Household survey Ghana interviewing key informants, Householdsb 3,190 1,740 visits to district offices and to Individualsb 14,924 7,191 schools in urban and rural areas. Tests‘ 3,718 3,582 Existing reports on education in School survey Ghana and other donor projects were Primary 286 417 collected and a review carried out o f Middle/JSS 233 289 the relevant World Bank project Teachers 0 3,129 files. a. One cluster was no longer inhabited in 2003 b. In 1988 approximately half of these numbers were in clusters Outline covered by the education module c. Number of people taking the Raven’s test. 1.11 Chapters 2 and 3 describe the inputs into Ghana’s education system. The former reviews the changes that have taken place in basic education since reforms were initiated in 1986 and chapter 3 reviews the Bank’s education portfolio together with that o f other donors. Chapter 4 brings these two strands together, identifjmg the impact o f the Bank and other external agencies on education policies and basic education outputs. The analysis o f the determinants o f educational attainment and achievement in Ghana i s presented in Chapter 5, linking these determinants to the interventions supported by the Bank and others. Chapter 6 goes on to examine the relationship between education outputs and welfare outcomes. Chapter 7 concludes with lessons learned and implications for future support to education. The technical annexes present more detailed analysis to substantiate the arguments made in the report. 6. The evaluation methodology i s given in more detail in the approach paper (Annex L) and in the design paper for the evaluation (available o n the study website). 6 Box 1.1 Evaluation design: costs and benefits The main data collection instrument for the impact evaluation was the re-surveying o f households and schools in the 85 communities covered in the education module o f the 1988/89 Ghana Living Standards Survey (GLSS2). The total cost o f this survey, f i o m the household and school listing through to data entry and cleaning, was US$263,000. Household surveys typically cost US$lOO per household, suggesting that the survey o f 1,740 households accounted for just less than one half o f the total survey budget. The school and teacher questionnaires (the latter including application o f the English, math and local language tests) cost just under US$50 each. The unique featui-e o f the study design was the application o f the same English and math tests used 15 years earlier. The nationally representative random sample o f people talung the same test over this period gives a firm basis for mapping progress in learning outcomes. The study i s unusual in linlung data on both school and household characteristics with student test scores, allowing analysis o f the factors behind changes in school attainment and achievement. The data also allow analysis o f changes in school-level inputs over the period o f the study. The quantitative data were supplemented by qualitative information f i o m fieldwork and a review o f the literature. Two trips were undertaken during which key informant interviews were carried out with government officials at central and district level, representatives o f the teachers’ union and NGOs. Schools were visited outside o f Accra, meeting with teachers, parents and pupils in different parts o f the country. The data were collected by Ghana Statistical Service, working in collaboration with the Ministry o f Education, who advised on the design o f the school and teacher questionnaires and provided enumerators for the school survey. Data analysis undertaken by both these organizations has been incorporated into the report. 7 2. Changes in Basic Education Since the 1980s I n 1986 the Government o f Ghana embarked on an ambitious program o f education reform. The main element o f this reform - the restructuring o f the education system - was successfully carried out. A second stage o f reforms to decentralize the school system is still underway. There have been substantial improvements in school-level inputs to the education system over the past 15 years. The availability o f material inputs - chalk, textbooks, and desks - has risen markedly. The development o f school infrastructure has kept pace with growing enrolments and has improved in quality. Some negative aspects can be noted. First, the percentage o f trained teachers has fallen and in-service training remains scant. Second, teacher absenteeism has risen and the quality o f teaching and supervision o f teachers by head teachers and circuit supervisors is uneven. Hence, while physical and material inputs have improved, there is less strong evidence o f improved teaching within schools. Finally, the reliance on communityjnancing widens the gap between well-resourced schools in afluent communities and badly resourced ones in the poorest areas. Ifeducation targets are to be met, attention necessarily needs to be paid to the latter group where enrolments, attainment, and achievement are lowest. THEEDUCATION SYSTEM BEFORE1986 2.1 From a position o f having been one o f the best in Africa, Ghana’s education system was by the early eighties in the throes o f a crisis with several underlying sources. Prolonged economic decline prior to the introduction o f reforms had led to a compression o f educational expenditure fiom 6.4 percent o f GDP in 1976 to just 1.5 percent by 1983. This spending was skewed in two ways: (i) large subsidies to secondary and tertiary levels, meaning that only one-third o f education expenditure went to the primary sector, and ( ii)recurrent expenditure was almost entirely absorbed by wages o f teaching and non-teaching staff, a problem exacerbated by the large number o f “ghost workers.’” The physical quality o f basic education facilities was very poor; schools structures were dilapidated and many lacked chairs, desks, and even chalk. The structure o f the system was inefficient, the school year was short, as was the school day at just four hours. However, pre-university education could extend to a staggering 17 years. 2.2 At independence in 1957 Ghana’s education system consisted o f six years o f primary education, followed by five years in secondary leading to 0-levels, and a further two years (“sixth form”) to the A-levels required for university admission. Entrance to secondary was by means o f a common entrance exam. However, the majority o f students went from primary to middle school for up to four years.’ Many children from better-off homes attended private primary schools and were able to skip the middle school stage: in 1985,30 percent o f secondary entrants were from private primary schools, most o f the rest coming f i o m the fourth year o f middle school. Thus the majority went through a 6,4,7 system, totaling 17 years o f pre- University education. 7. That is, people o n the pay role who no longer work in that position or may not even ever have existed. 8. Middle schools were created by the Accelerated Development Plan for Education in 1951, replacing the senior primary schools that had been introduced a few years earlier (Graham 1971: Chapter 11). 8 2.3 Between Independence in 1957 and the mid-1980s there were nine attempts at educational reform, starting with the Botsio Commission in 1960.’ Most important was the 1972 Dzobo Commission whose report, “The N e w Structure and Content o f Education,” formed the basis for the 1986 reforms. The Dzobo Commission recommended that middle schools be replaced with Junior Secondary Schools (JSS), with a stronger vocational orientation, following which 118 JSSs were created on an experimental basis. However, opposition from the middle classes and the teaching profession, including the Ghana Education Service (GES) created in 1974, forestalled extension o f the reforms. But 14 years after the Dzobo Commission the PNDC government finally implemented the proposed changes. THE1986 REFORM PROGRAM 2.4 The education reform program adopted in 1986 sought to: 0 Change the structure o f the school system by replacing the 6,4,7 system with 6,3,3, shortening pre-university education from 17 to 12 years. Middle schools were t o be replaced by JSSs, which would be an integral part o f the system for a l l children, and 0 and A-levels replaced with the secondary certificate. 0 Improve the teachindleaming process by increasing school hours and the quality o f teachers, including the phasing out o f untrained teachers (i.e., those with n o formal teaching qualification, often called “pupil teachers”). 0 Increase cost recovery at the secondary and tertiary levels. 0 Make educational planning and management more effective. All four elements o f the reform program were implemented and most sustained. 2.5 The restructuring was phased as shown in Figure 2.1. The last cohort o f middle school students was admitted in 1986/87; when they graduated in 1989/90 middle schools ceased to exist. Meanwhile, the first JSS cohort was admitted in 1987/88, so that schools simultaneously contained both JSS and middle school students for three years. The first JSS students took the new ninth grade Basic Education exam at the end o f the 1989/90 academic year, the successful candidates forming the first cohort to enter the new SSS system in January 1991,completing in December 1993. 2.6 From 1987 to the mid-90s there was a substantial drop in the percentage o f untrained teachers f i o m 50 to 20 percent in primary schools, and 35 to 14 percent in JSSs (see paragraph 2.26 below). This decline has been reversed in recent years, particularly in primary schools. The reversal is partly because o f the growth o f the private school sector, in which most teachers are untrained. 9. The first Education Committee had been in 1908 (McWilliam and Kwamena-Poh 1975: Chapter 7). More important was that o f 1942, whose proposals laid the basis o f the Accelerated Development Plan for Education the next decade, which provided the framework for a substantial rise in enrolments in the 1950s (Graham 1971: Chapter 11). 9 Figure 2.1 : Restructuringo f education system I I 86/87 I 87/88 1 88/89 I 89/90 I 90/91 I 91192 I 92/93 I 93/94 I 94/95 I Middle schoo’ :tr Ladmitted Last cohort graduate Middle schools closed First cohort First JSSnot take BE JSS cohort yet begun admitted exam (end of year) sss Last Form Lastcohort Old (old 1 cohort complete system system) admitted Form 5 finished sss First cohort First cohort (new New SSS system not yet begun admitted complete System) (Jan 1991) (Dec 1993) 2.7 The reform also included three forms o f cost recovery: (1) increased charges for textbooks, (2) removing boarding and feeding subsidies for secondary and tertiary institutions, and (3) removal o f student subsidies for tertiary education. Charges for textbooks were raised to cost-recovery levels, with the intention o f setting up a revolving fund. However, the fund was not well managed (e.g., BESIP S A R : 12) and did not become a basis for sustainable textbook supply, which has continued to be supported by extemal donors. Moreover, textbook charges were abolished for primary students in 1995. Boarding and feeding subsidies were removed first through an increase in the parental contribution followed by the removal o f the government’s contribution. Removal o f subsidies for university students was delayed for some time on account o f i t s political unpopularity manifested in frequent protests but eventually proceeded with some modifications. The University Rationalisation Study was completed in March 1988 and in September o f that year the government announced i t s intention o f removing subsidies from the tertiary sector. However, two months later, the government proposed a loan scheme for tertiary students that contained an element o f subsidy. While cost recovery has not been as extensive as at first envisaged, parental contribution to costs for senior secondary and tertiary education have become an established part o f the education system in Ghana. 2.8 Improvements to educational planning centered around strengthening the Ministry o f Education. The Policy, Budgeting, Monitoring, and Evaluation division was created (with UNDP technical support partly financed by the Bank). The divisions o f Curriculum Design and Development, and Supplies, were both relocated in the Ministry from GES. A school mapping was carried out in 1987 (under the project preparation facility from the Bank) and educational statistics began to be collated on a systematic basis since 1988 resulting in the later establishment o f the Education Management Information System (EMIS) with World Bank and U S A I D support. FCUBE AND DECENTRALIZATION 2.9 Once the new structure was in place, sector policy was outlined in 1996 in the strategy document “Free Compulsory Universal Basic Education (FCUBE),” which stated the government’s commitment “to making schooling from Basic Stage 1 through 9 free and 10 compulsory for a l l school-age children by the year 2005.. . [and] to improving the quality o f the education services offered” (GoG [MoE], FCUBE, April 1996: 1). In principle, this statement did n o t signal any change in policy, but was one o f the periodic attempts by government to abolish unsanctioned fees that proliferate at the local level.” 2.10 The significance o f FCUBE was twofold: (1) it provided a basis for a coordinated sector program providing a framework for donor support to education; and (2) i t l a i d out the institutional and other measures to support the nascent decentralization program, including increased community participation in school management. 2.1 1 F C U B E had three costed components: a Improving quality o f teaching and learning, consisting o f (1) the review and revision of teaching materials in line with a revised, more focused, syllabus, (2) new measures o n teacher incentives, including teacher prizes and teacher housing in rural areas, and (3) a shift to in-service teacher training using distance learning materials. a Strengthening management at both central and district level; and a Improving access and participation, though, inter alia, facility construction and rehabilitation and pilot scholarship schemes to encourage girls’ participation at primary level. n addition t o the above, measures were to be undertaken to ensure the financial I sustainability o f the education sector. 2.12 There h as been progress regarding the first two elements o f the first component, but the shift to in-service teacher training has not really taken off. The GSS/OED survey data show that less than 5 percent o f basic school teachers receive such training o n a regular basis. The largest changes have taken place with respect to decentralization. The Local Government Acts of 1988 and 1993 shifted responsibility for the administration o f education to the districts, and the 1995 Ghana Education Service Act created District Education Oversight Committees (DEOCs) as well as community-level School Management Committees (SMCs). Whereas PTAs had been expected to play a largely revenue raising f i c t i o n , the SMCs were to act like school boards, which already existed at secondary level. Annual School Performance Assessment Meetings (SPAMs) were to be key events at which the SMC, teachers, and the rest o f the community could meet together. Armed with data from the most recent Performance Monitoring Test (PMT), which ranks each school in the district based o n test results, they are to prepare a plan to improve school performance. 10. Primary school fees were f n s t abolished o n January 1, 1952 (Graham 1971: Chapter 11). The Education Act of 1961 c o n f i i e d this position (though materials could be charged for) and made primary schooling compulsory, though the government at the time acknowledged this was not practical in all locations (McWilliam and Kwamena-Poh 1975: Chapter 13). Free compulsory basic education i s enshrined in the 1992 constitution. 11 BUDGET 2.13 In the early 1980s government expenditure f e l l below 10 percent o f GDP. At around one fifth o f total spending, education spending was just 1.5 percent o f GDP. From 1984-87 education expenditure grew rapidly for three reasons: education claimed a growing share o f a budget that was a growing share o f a growing GDP (Figure 2.2)." Real expenditure grew at an average rate o f 35 percent a year over this period, and the share o f education spending in GDP more than doubled (see Annex B). The growth in real spending exceeded the growth in student numbers so real spending per student also increased. Figure 2.2: Government spending on education has risen: central government education expenditure -- 5.0 -- 4.0 ~ 201 P (3 0 -- 3.0 c C Q 2 ... 2 CI -- 2.0 ? I 1 Share of government expenditure (left scale) lo 0.0 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 Source: MoE and GSS data 2.14 These increases were sustained into the early 1990s. Real spending and education's share o f GDP continued to rise, passing 5 percent in 2001, Total spending on education rose faster s t i l l as a result o f (1) increased parental contributions, (2) the growth o f the private sector in the 199Os, (3) substantial donor support to the sector since 1990, and (4) the introduction o f GETFund in 2001 (see footnote 11). However, the share o f education in central govemment spending has fallen, though partly mitigated by the one-third o f Common Fund resources that are spent by District Assemblies on schools.'2 11. The figure excludes the GETFund created in August 2001 and financed mainly from VAT. In 2002, the GETFund disbursed 140 billion cedis, o f which 125 billion were to tertiary education (90 billion o f that being student subsidies). The introduction o f GETFund thus increases the share o f education in government expenditure but reduces the share o f basic education in that expenditure. 12. Since 1993,5 percent o f central government revenues are paid to the District Assembly Common Fund (DACF) for investment expenditure by districts. 12 2.15 The share o f basic education in total education spending has fluctuated around an average of 67 percent over the period 1989-2001,being above this average in the early 1990s and again in the most recent years. N o substantial reorientation o f the education budget appears to have taken place in the period since 1989.13 However, at an average for the period o f 42 percent, the share going to primary education is above the one-third reported for the early 1980s, showing that the shift took place during the major expansion in funding in the mid-1980s. SCHOOL-LEVEL INPUTS 2.16 School quality can be measured by four different types o f inputs: 0 Material inputs, such as chalk and textbooks 0 Physical inputs, such as classrooms and blackboards 0 Teachers 0 School management. Data were collected on each o f these aspects in both 1988 and 2003 school surveys and are used here to show how the situation in schools has changed over time (Annex D provides a more detailed analysis). Physical and Material Inputs 2.17 The main message from the GSS/OED school survey i s the overwhelming improvement in physical and material inputs. For example: 0 In 1988 less than h a l f o f schools could use all their classrooms when it was raining, but in 2003 over two-thirds can do so. 0 94 percent o f schools have a blackboard in every classroom today compared to 78 percent 15 years ago. 0 Fifteen years ago over two-thirds o f primary schools reported occasional shortages o f chalk, but today 86 percent say there is always enough. 0 The percentage o f primary schools having at least one English textbook per pupil has risen from 2 1 percent in 1988 to 72 percent today; and the percentage o f JSS having at least one math book per pupil has risen from 13 to 7 1 percent.14 13. Although there was a substantial reduction in the length o f senior secondary education it was accompanied by increased enrolments at that level, limiting the savings realized by the efficiency gain for reallocation at the basic level. 14. It i s whether textbooks are being used or not that matters. Responses from the teacher questionnaire show that, where books are available, they were used by over 90 percent o f teachers in their most recent math or English class. A study in the mid-90s found that textbooks were indeed used in the classroom provided there were sufficient to go round (Okyere et al. 1997). 13 2.18 Despite this overwhelmingly positive message there remain some schools, most typically in poor rural areas, in which conditions, while improved, remain poor (see para. 2.25). Material Inputs 2.19 The four materials inputs for which data can be compared between 1988 and 2003 - availability o f chalk, math and English book availability, and desks'' -were combined into an For each o f these four variables there has been a highly significant index o f material inputs. l6 improvement in the level o f inputs at both primary and JSS level, and the index shows an improvement in nearly every area surveyed (Figure 2.3). Figure 2.3: Schools in nearly all areas have more material inputs than before: cluster-level material inputs to school quality (a) Material Primary (6) Material Midde/JSS 1.o 1.o 0.8 0.8 X ! J a c '0 .- '0 C 'Z 0.6 5 0.6 Kl .- C I 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 Cluster Cluster _____ 1988 -2003 -2003 (public only) _ - _- * 1988 -2003 -2003 (public only) Source: GLSS2 and GSS/OED school survey 2.20 Figure 2.3 shows the cluster level average o f the material input index for 1988 and 2003, calculated separately for primary and middle/JSS. In each graph the clusters have been ranked according to the value o f the index in 1988, so that the clusters with the schools with the fewest material inputs in that year appear to the lefl o f the scale. Where the line for 2003 lies above that for 1988 there has been an increase in the material input index for that cluster. T w o points jump out from these graphs: 15. School furniture has been included i nthe material index although it should arguably be included among the measures o f physical quality. However, the latter are restricted to infrastructure. 16. The index i s the simple average o f the four variables scaled over the range 0- 1. 14 0 There has been a substantial increase in the level o f material inputs across the country, especially in primary schools. In only two clusters (which had the maximum value o f 1 in 1988) has the level o f material inputs declined at primary level. For middle/JSS there have been an improvement in all but 9 o f the 76 clusters 0 The improvement has been greatest the lower the initial level o f the index, meaning that the clusters in which schools that were the most deprived in 1988 have seen the largest improvements in material inputs. 2.21 The share of private schools in the sample increased f i o m 5 to 20 percent between 1988 and 2003. But the increase in school quality does not result f i o m the better quality o f private schools. Figure 2.3 also shows the material input index for 2003 calculated for public schools alone. Ingeneral this line i s not far removed fiom the overall cluster average. Indeedit i s above it, indicating that public schools have a higher level o f material inputs than do private ones, in 22 o f the 41 clusters that have private schools. When the changes i nthe index and its components are calculated for public schools only these changes all remain significant at the 1percent level (Annex E). Physical (building) Inputs 2.22 Physical inputs have also increased, though to a lesser extent. The indicators used are the adequacy o f the number o f classrooms, the proportion that can be used when raining, the proportion with a blackboard and the quality o f those boards, the presence o f a library and own water supply. T w o o f these have not improved (number o f classrooms and library) for either type o f school, one (library) has not for primary schools, and another (classrooms that can be used when raining) for middle/JSS. The lack o f change o f there being sufficient classrooms shows that classroom building has kept pace with growing student numbers. The number o f classrooms has increased, but been matched by more students. Overall, there has been a significant increase in the index o f physical inputs (Figure 2.4). 15 Figure 2.4: The quality o f school infrastructure has improved in most areas: cluster-level physical inputs to school quality (a) Physical Primary (b) Physical Midde/JSS 1.0 -f 1.0 1 0.8 x 0 .- U S 2 0.6 Q ._ - S 8 'R 0.4 E 0.2 0.0 Ij; I ' , 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 Cluster Cluster _____ 1988 -2003 -2003 (public only) - _ - _ _1988 -2003 -2003 (public only) Source: GLSS2 and GSS/OED school survey 2.23 Figure 2.4 shows the change in physical inputs in the same way as Figure 2.3 showed material inputs. Well over half o f the clusters have experienced an overall improvement in physical inputs. 2.24 Once again, although private schools perform better in some respects, their increase does not account for the improvement in school quality that has taken place. In 2003, private schools had superior inputs with respect to the percentage o f classrooms that could be used when raining and having their own water supply. They also had slightly better average quality chalkboards, although the difference i s not quite statistically significant. There i s no difference with respect to having sufficient classrooms, chalkboards, or a library. Allocation of Material and Physical Inputs 2.25 There were biases in the allocation o f material inputs in 1988. By 2003 these had been eliminated, with the exception o f desks. But there has been a continued bias against poorer areas in the distribution o f physical inputs. The source o f these differences is the basic school financing and distribution system. Chalk and textbooks are supplied centrally through GES to their district offices, which distribute them to schools. This system was not functioning in 1988 owing to lack o f materials and transport. But today i t works so as to ensure sufficient supplies in the majority o f school^.'^ However, infrastructure i s the responsibility of districts, which may also supply desks, with additional support from the 17. I t i s plausible that the efficiency of the distribution system has been enhanced by decentralization, which has placed more GES officers at district level. This question i s beyond the scope o f this study. 16 PTAs. Schools in wealthier districts will benefit from both higher levels o f district support and higher parental contributions, resulting in discrepancies in resource availability. The worst-resourced schools are “bush schools” that i s schools in off-road rural communities. Such schools have difficulty in attracting teachers’8 and parents who can illafford any cash contributions. There is growing dichotomy within the public sector between these schools and those o f relatively more affluent parents in urban areas.’’ Teachers 2.26 The number o f primary teachers rose from 47,900 in 1980 to 84,400 in 2001. For JSS these numbers are 22,500 and 43,000 respectively. In line with the reform program, the proportion o f teachers who are trained rose, particularly in primary school reaching nearly 80 percent from a l o w ofjust 50 percent (Figure 2.5). But this trend was reversed in the mid- ~ ~that today only 60 percent o f primary teachers are trained. This i s partly because o f 1 9 9 0 so the growth o f private schools, which typically do not require their teachers to be trained. I nthe 2003 GSS/OED school survey 87 percent o f public basic school teachers were trained, whereas just 12 percent o f teachers in private schools had teacher training. A second explanation i s that trained teachers are taking study leave and not retuming to basic education -either joining the administration, teaching in secondary school, or leaving education altogether.20 Figure 2.5: The proportion of teachers who are trained rose.. . and then fell again 100 , E Q)40- n 30- 20 - 10 - 0 , I / 1 / / / / , , 2.27 Ghana has a l o w pupil-teacher ratio (PTR) compared to other countries. Official policy i s to raise the PTR in the interests o f efficiency. The increase in the average ratio for primary 18. The two bush schools visited by the study team both only had one teacher, the others having refused to take up their posts (see Hedges 2002, for further discussion o f the failure o f some teachers to take rural postings). In neither case was the teacher present o n the day o f the visit. 19. This phenomenon was documented in the reported entitled A Tale of Two Ghanas (Kraft et al. 1995). 20. After a certain number of years service teachers qualify for paid study leave, during which they continue to draw their salary while pursuing full-time further education. Being a primary school teacher i s thus a well-established stepping-stone to other careers (Hedges 2002; and Akyeampong and Stephens 2002). 17 schools fi-om 30.6 to 36.0 between 1988 and 2003 therefore represents greater efficiency. Only 13 percent o f primary schools now have a low PTR (defined as less than 20) compared to 18 percent 15 years ago (Figure 2.6(a)). But more schools suffer from having too few teachers, defined as a PTR o f 50 or more, especially in northern regions where 54 percent o f primary schools had a high PTR (Figure 2.6(b)). 2.28 The quality o f teachers i s measured by teacher training and the methods they employ, including student supervision and time on task. The proportion o f trained teachers has fallen and the provision o f in-service training is unsatisfactory: 35 percent o f the 3,129 teachers interviewed in the GSS/OED survey stated that they received no teacher in-service training at all in the past year. O f those who have received such training, 70 percent have received it three times a year or less. Less than 3 percent o f teachers benefit from in-service training once a month or more. Figure 2.6 (a): Efficiency gains have been Figure 2.6 (b): But the PTR i s too high in the realized by increasing the pupil-teacher ratio Savannah region (PTR by zone, 2003) 80 I 80 I I 70 70 $ 60 60 5 5 0 0 - 50 50 u) u) % 40 % 40 c C o 30 E 30 2 2 c 20 c 20 10 10 0 0 Low Medium High Coastal Forest Savannah Source: GLSS2 and GSSlOED school survey 2.29 Teaching methods can be broken down into use o f improved methods, the frequency with which teachers set homework and time on task. In 2003, teachers were asked three questions to test their familiarity with improved teaching methods. About a third o f teachers use a student-centered learning approach and use simulations (role play) on a regular basis, though about a fifth o f the latter could not explain them properly. About one-fifth use cues to help explain difficult words. I nsummary, improved teaching methods are far from unknown, but not widespread, being utilized by a minority o f teachers. Trained teachers are significantly more likely to use improved methods than untrained ones, although there is not a significant difference between teachers who have received university-level teacher training and those trained by TTCs (Annex D, section D.5). In-service training also helps. Head teacher supervision o f teachers has a significantly positive impact on the use o f improved methods, as does the teacher having direct contact with the circuit supervisor. 2.30 In 2003, data were also collected on the frequency with which teachers set homework, look at and assess students’ work for both math and English. Homework i s set at least once a week by over 95 percent o f teachers for math and English, and work assessed with the same frequency by h a l f the teachers surveyed. Less attention i s paid to local 18 languages with homework set frequently by 80 percent o f teachers and far fewer assessing work o n a regular basis. On average one-third o f the time in the classroom i s spent o n task based o n a narrow definition, but 72 percent using a broader definition. There is considerable variation around these averages. 2.3 1 Teacher absenteeism has increased over the past 15 years. In 2003 nearly 13 percent o f teachers had been absent in the past month for reasons other than sickness,21compared to just over 4 percent in 1988. Correspondingly, more schools are affected by absenteeism today than in 1988. Fifteen years ago 85 percent o f schools did not suffer at all, whereas this figure has n o w fallen to 61 percent. There is a substantial difference between public and private schools: 80 percent o f private schools have no problem with absenteeism, compared to not much more than h a l f o f public schools. Absenteeism i s greater in rural areas, probably for the following reasons: (1) teachers m a y live in town some distance from the school and suffer transport problems, (2) they have to travel to town once a month to collect their pay, which they m a y find i s not yet there, and (3) rural teachers attend to their farming activities.22 M o r e generally, absenteeism i s linked t o l o w teacher morale and poor working conditions, in particular not receiving pay o n time (see Annex D). School Management 2.32 The focus on software rather than hardware means an increased focus o n issues o f school management. At the school level the majority o f head teachers are actively involved in the different types of supervision. Notably, less than 5 percent o f teachers say that the head teacher does not look at their lesson plans o n a regular basis. However, fewer than h a l f say that the head actually discusses the lesson plan with them. And, while the large majority o f schools have visits from the circuit supervisor, nearly half (44 percent) o f teachers have n o direct contact with him or her. This latter finding confirms that from the evaluation o f the Primary School Development project in the late 1990s, which found that many circuit supervisors merely checked staffing numbers and enrolments rather than observing teachers in the classroom or other activities that might positively affect learning (Fobih et al. 1999: p.33). 2.33 In 1988, circuit supervisors visited schools just over once every two months o n average. By 2003 the mean number o f visits rose from 6 to 9 a year for primary schools and a bit less for JSS. There is little variation between areas o f the country as to the frequency o f visits, but 45 percent o f private schools receive infrequent supervision visits, compared to only 7 percent o f public schools. 2.34 Virtually all schools have a PTA. Over 99 percent o f public basic schools had them in 2003, as did 95 percent o f private schools. However, it i s not the mere presence o f a PTA that will make the difference, but the extent to which it provides support to the school. There is considerable variation in the extent to which PTAs have provided support to schools and in 21. A more detailed study by EARC (2003), taking into account late arrival and not being present in the classroom, finds an even higher degree o f absenteeism. 22. A main source o f income for urban teachers i s extra classes, which necessarily do not take place during school hours. Rural communities, which are more cash constrained, offer fewer opportunities for extra classes. 19 the value o f parents’ monthly contributions. Econometric analysis shows that the level o f community support to the school through the PTA is closely related to the community’s economic well-being. On average, schools in the better-off areas among the survey areas can expect to receive 10 times as much in PTA contributions as can schools in the least well o f f areas.23The actual range i s far higher, with several schools not requesting a PTA contribution compared to the maximum o f 150,000 cedis per child ($20) (Annex C). 2.35 School Management Committees are also widespread, being present in over 80 percent o f the schools ~urveyed.’~However, in only h a l f o f schools had SMCs met in the preceding month or provided support in the past year, and in even fewer helped the school in dealings with outside agencies. The lower prevalence o f SMCs than PTAs is largely explained by the fact that they are not required at private schools: over 90 percent o f primary schools have SMCs. 2.36 Virtually all public primary schools (92 percent) have had a School Performance Assessment Meeting, at 98 percent o f which an action plan was agreed. However, knowledge o f SMCs and the SPAM among households i s far less common than the school-level data suggests it should be and participation rates correspondingly low. Only 6 percent o f households say that someone attended a S P A M at their child’s school. 23. This result follows from statistical analysis o f PTA contributions with respect to average community income. This elasticity i s found to be close to unity (see Annex D, section D.4). 24. See Condy (1998) for detail o n the setting up and intended role o f SMCs. 20 3. The Bank’s Education Portfolio in Ghana Five Bank projects have provided support to basic education: the education component o f HERP, EdSAC I and II,PSD, and BESIP. The money from all o f these projects has been largely devoted to hardware and instructional materials, mainly school building and rehabilitation, and textbooks and schoolfurniture. The Bank’s contribution to more recent changes in school management, such as school management committees and increased emphasis on in-service training, has been rather more limited. OVERVIEW 3.1 Following the engagement o f the World Bank with Ghana’s Economic Recovery Program in 1983 the Bank undertook initial education sector analytical work in 1984. Further discussions the following year resulted in a preparation mission in September 1985, which proposed a sector approach.25 Prior to the sectoral adjustment credit a $0.3 million project preparation facility enabled some planning activities, such as a school mapping exercise as well as purchase o f essential school materials such as pens and pencils, and further emergency support was provided under the Health and Education Rehabilitation Project (1986-91, see Table 3.1). The first two education projects were sector adjustment credits EdSAC I (1986-91) and I1(1 990-94), the first o f which was the first SECAL for education by the Bank and foreshadowed the later adoption o f the sector approach more generally.26These two projects, which were directly linked to the reforms described in the previous chapter, were complemented by two investment projects: (1) Community Secondary School Construction (CSSC, 1991-95) to create the extra capacity at the secondary level, especially in under-served areas, expected to be generated by the reforms; and (2) Tertiary Education (1992-98), which was left out o f EdSAC I1because o f i t s political sensitivity. 3.2 T w o further projects also supported formal basic education: Primary School Development (PSD, 1993-98) and the Basic Education Sector Improvement Project (BESIP, 1996-02). Basic education can include adult education programs, and these have been supported by two additional projects: Literacy and Functional Skills (1991-95) and National Functional Literacy Program (1992-98) and National Functional Literacy. A final project focused on vocational skills training as part o f a broader informal sector project. 3.3 Over the period 1986-2002 the Bank disbursed $260 million to projects supporting Ghana’s education sector, an average of $17 million a year, peaking at nearly $40 million in 1995 when five projects were disbursing simultaneously (Figure 3.1). 25. Project design also drew o n the 1985 Public Expenditure Review. 26. The evolution o f the portfolio -both the sequencing o f reforms under EdSAC I and I1(and associated Community Secondary Schools Project), and the timing o f the related investment projects for primary and tertiary education -can also be compared to the more recent development o f adaptable program lending. 21 Table 3.1: World Bank support to Ghana's education sector, 1986-2003 Budget Outcome rating Project Approved Closed IDA Total Bank OED ratina Health 8, education rehabilitation (HERP) 18.0 18.1 Not S 1I86 12/91 olw education component 6.1 rated Education sector adjustment 38.3 45.5 S S 12/86 12/91 Education sector adjustment II 53.2 S MU 5/90 12/94 Community secondary school construction 14.7 19.6 S MS 619 1 6/95 Literacy and functional skills 27.8 S S 3/92 12/97 Tertiary education 44.8 51.0 MS' MS 10192 9/98 Primary school development (PSD) 53.2 56.6 U MU 6/93 12/98 Basic education (BESIP) 47.9 241.6 S S 6/96 12/02 Vocational skills and informal sector 5.8 U U 3/95 6/01 National functional literacy program 23.7 S n.a. To close 12/04 Key: S = Satisfactory, MS = Moderately/marginallysatisfactory: MU = Moderatelylmarginally unsatisfactory; U = Unsatisfactory. I / The rating system used by the Bank's operational staff allow for only satisfactory or unsatisfactory ratings, but in this case it was stated that the project outcome was "barely satisfactory". Source: World Bank project documents. Figure 3.1 : Bank disbursements on an annual and project basis, N86-02 35 - Nationalfunctional literacyd 30 - UU BESIP* OVocational skills* 25 - PSD 20 - €i Tertiary education project Literacy and functional skills 15 - cssc 0 EdSAC II 10 EdSAC I 5 H W 0 1987 1989 1991 1993 1995 1997 1999 2001 *OED estimate of annual disbursement from project supervision data. Source: World Bank project documents 22 3.4 This study focuses on the five credits that have supported formal basic education: HEW, EdSAC I and 1 1, Primary School Development, and the Basic Education Sector Investment Project. THESECTOR ADJUSTMENT CREDITS: EDSAC 1AND 11 Objectives 3.5 The objectives o f the two EdSACs were linked to the reforms begun in 1986: 0 EdSAC I(l986-91): (a) Change the structure o f the education system; (b) improve pedagogic efficiency and increase access; (c) improve budgetingprocedures and effect cost savings and cost recovery measures. 0 EdSAC II (1990-94): (a) Complete the restructuring o f the school system to a 12-year cycle; (b) extend the reform to senior secondary education; (c) consolidate the basic education reforms so that primary and JSS leavers acquire the cognitive skills needed to take advantage o f education offered at higher levels; and (d) ensure the financial sustainability o f the new system. Use o f Funds 3.6 Since both EdSAC I and 11were budget support, it may seem that the attribution o f the funds to specific expenditure items is not worthwhile. However, although the funds were budget support, they were disbursed against expenditures on a schedule prepared against a positive list and agreed on a tranche-by-tranche basis with the Ministry o f Ed~cation,2~ with agreement on the overall education budget as one o f the conditions for tranche release. Procurement was carried out by a Project Management Unit (€‘MU)located in the ministry. The Bank’s task manager, who was based in Accra fiom late 1987, was involved in monitoring procurement decisions and procedures. Hence it makes more sense to say that the Bank financed these items than would usually be the case with budget support (with the partial exception o f an early disbursement, which was retroactive finance for textbooks the govemment had already printed).” 3.7 Under the two EdSACs just under 30 percent o f IDA funds were used for school building and rehabilitation, and a similar amount for school fumiture and equipment (Table 3.2). Other expenses includes items such as vehicles, so about two-thirds o f total funds went on “hardware.” The next largest item was teaching materials, which includes both the development and printing o f teacher materials and textbooks. A relatively small amount (only 2 percent under EdSAC 1 1) was spent on teacher training. However, under each o f 27. Only EdSAC I and I1and one credit to Nigeria operated in t h i s way as the Bank’s legal department ruled that program aid funds could not be used to finance local expenditures. This decision became irrelevant after February 1996 since when it has not been necessary to account for the use o f funds from adjustment loans. 28. T h e region objected to OED’s assessment o f EdSAC I1o n the grounds that it was not what happened to the money that mattered but rather the reforms that were supported. I t i s argued here that both the use o f funds and reform matter. 23 EdSAC I and I1over 20,000 JSS teachers received teacher training to orient them to the new system. Under EdSAC I124 percent o f the funds were allocated to primary education and another 15 percent to JSS.29Nearly 95 percent o f the funds spent in the primary sector were used for civil works (mainly the construction o f school pavilions), as were just under one- third o f the funds benefiting JSSs with most o f the remainder (61 percent) for textbooks. Table 3.2: Allocation o f resources under EdSAC Iand I1 EdSAC I I EdSAC II IDA Total 1 IDA Total I IDA US$ Percent US$ Percent millions millions School building and 11.3 17.5 29.4 38.3 15.2 28.5 rehabilitation Teacher training 3.4 3.4 8.8 7.5 1.I 2.1 Teaching materials 8.1 8.1 21.o 17.8 12.0 22.5 School furniture and 9.4 10.3 24.4 22.5 18.2 34.1 equipment Other expenses 6.3 6.3 16.4 13.9 6.8 12.8 Total 38.6 45.7 100.0 100.0 53.3 100.0 Source: calculated from project data in Bank implementationcompletion reports. 3.8 The system o f administering budget support has systemic effects, that is, the way in which the donor’s system for managing the aid inflow (ie., procurement, monitoring, and reporting requirements) affects the govemment’s resource allocation procedures. Such effects are frequently negative as donor systems can impose large transaction costs o n the borrower (see White and Dijkstra 2003: Chapter 12). However, in the case o f EdSAC they appear to have been positive, with the Ministry o f Education requesting more frequent supervision. Procurement procedures are prone to corruption and bureaucratic delay. The presence in the field o f the Bank’s task manager facilitated timely and detailed comments o n bidding procedures and familiarized ministry staff with competitive tendering procedures, which were adopted for all ministry procurements in the early 1990s (an EdSAC I1condition). These interventions would have been unnecessary if procurement had been problem free, but it had not. Or they could have been costly if there were many donors imposing different procedures. But the Bank was the only donor o f substance prior to 1990 (other donors supported education by co-financing EdSAC I). The first significant bilateral support was USAID’s $35 million Primary Education Project (PREP, 1990-05) o f which $32 million was budget support channeled through the P M U responsible for EdSAC procurement. Conditionality and Reforms 3.9 Both credits consisted o f three tranches. EdSAC I1conditionality was simplified to the same set o f six conditions for each tranche release, but complicated by the introduction o f 29. EdSAC I1supported the second phase o f the reforms, which was focused o n the second and third cycles. While data are not available, EdSAC I ,which supported the f i r s t phase (reform o f basic education), will have allocated a higher proportion o f funding to the basic level. 24 performance indicators that were used to judge progress but which did not have legal status. This ambiguity might explain some o f the tensions that emerged between GoG and the Bank in the later period (see below). 3.10 Although the government implemented an impressive range o f reforms, this does not mean that the conditionalities attached to the two EdSACs were problem free. Far from it. The policy conditions under the two credits can be divided into four areas (the conditions are listed in full in Annex J): 0 Restructuring: these conditions matched the government's own timetable for the introduction o f the new system and were met accordingly. Bank reviews o f the adjustment credits noted an initial lack o f trained teachers for all subjects and teaching materials for JSS. However by June 1987,7,000 JSS teachers had been trained and the army mobilized to distribute textbooks, indicating that whatever shortcomings there were did not arise from government complacency. 0 Budget: EdSAC I required that there should be agreement on the education budget, which was met each time. This condition was kept for EdSAC I1with the added requirement that actual expenditure should be in line with the budget and that the share o f basic education should stay at least i t s 1989 level (62 percent). The second o f these targets was met, but the first was not with larger amounts going to tertiary and vocational training, resulting in tensions between the Bank and government, the latter accusing the Bank o f bringing up arbitrary c~nditionalities.~' 0 Cost recoven: (1) Boarding and feeding subsidies: Government subsidy to feeding and boarding costs for secondary school students were reduced - although not to the level required by the Bank. This partial slippage was allowed to pass, and the subsidies were later removed altogether. The condition to eliminate the feeding and boarding subsidy at the tertiary level was postponed and the Bank accepted the introduction o f a subsidized student loan scheme. Delays in completion o f the University Rationalization Study (URS) and its implementation were major factors behind the delayed release o f the second and third tranches o f EdSAC I (2) Book charges were introduced and . increased at a rate to ensure full cost recovery, with the proceeds paid into a revolving fimd. In January 1995, the charge was abolished at primary level. The revolving fimd did not become a basis for sustainable textbook purchases with textbook supply continuing to depend o n external finance. 0 Staffing: (1) A payroll audit was undertaken to eliminate ghost workers, with 5,722 ghosts removed from second-cycle institutions by January 1987. A freeze o n new posts was breached in 1988 with GES employment increasing by close to 7,000 (to a total o f 158,102) as a result o f the hiring o f untrained middle school leavers, in contravention o f both the condition and government's o w n policy that no new untrained teachers should be hired. Extensive staff cuts brought GES employment 30. The two main points o f dispute were: (1) the Bank objecting to tertiary's share exceeding 20 percent, which GoG said was a convention not a condition, and (2) that the government reduced fimding for education materials, apparently exploiting fungibility as more donor funds became available. 25 down t o 146,000 by mid 1990. During EdSAC I1negotiations a ceiling o f 153,000 was agreedY3’ which was as good as kept until the third tranche32 was approved, but breached shortly thereafter rising to nearly 155,000 in 1994. The Bank wrote to the government asking i t to keep to “its ceiling” o f 153,000 but had no leverage since the h n d s were disbursed. (2) Freezing the size o f GES at a time o f growing enrolments had the desired effect o f increasing the pupil-teacher ratio. At senior secondary level the condition that class sizes for optional subjects be at least 20 was not met (the Bank had originally proposed 25), the Bank responding merely by requesting government to send a further instruction to schools to reduce the number o f options taught in schools missing the target. The letter was sent but not complied with by a l l schools. 3.1 1 The EdSAC I1targets not contained in the legal covenant included designing and implementinga plan for in-service training, the introduction o f the new circuit supervisor system and the introduction o f performance testing (the Criterion Reference Test). Each o f these things was done, though the funding and technical support to do so was provided by U S A I D rather than the Bank. IN BASICEDUCATION: INVESTING PSD AND BESIP Objectives 3.12 B o t h the Primary School Development Project (PSD) and Basic Education Sector Improvement Program (BESIP) emphasized increasing access and improving the quality o f education: 0 PSD: The overall goal o f this project was to increase learning achievements and enrollments in primary schools throughout the country. In order to accomplish this, the project had the specific objective o f increasing the amount and improving the quality o f instructional and learning time in primary schools, particularly as far as 1,983 o f the least well-endowed primary schools are concerned. 0 B E S P was intended to help the Government o f Ghana to implement FCUBE, specifically aiming to (a) improve the teaching process and learning outcomes; (b) strengthen management o f the basic education system through better planning, monitoring and evaluation by MOE/GES at central, regional and district levels, and by promoting active involvement o f communities in the management o f schools; (c) improve access to basic education, especially o f girls, the poor and other disadvantaged segments o f the population; and (d) ensure financial sustainability o f the Government program for basic education over the longer term. 31. The Bank had wanted 152,000 and the government 155,000. 32. There was a negligible excess, With a figure o f 153,513 in August 1982. 26 Project Components and Use o f Funds 3.13 For PSD two main areas o f activity were identified: 0 Policy and management changes: (1) increased instructional time, (2) reducing student fees and levies, (3) improve skills and motivation o f head teachers, (4) community involvement in selection o f head teachers, (5) orientation o f district officials and community leaders, (6) support to school supervision, and (7) school mapping. 0 Investment in physical infrastructure: (1) construction o f classrooms, (2) construction of head teachers’ housing, (3) provision o f roofing sheets. Communities were to be responsible for building the external walls (“cladding”) for pavilions constructed by the project. 3.14 These activities were to be carried out in the 1,983 most deprived schools. This number of schools covered by the project was later increased to 2,178 in response to pressure from MPs. Inthe mid to late 1990s there were approximately 11,200 public primary schools, meaning that about 20 percent o f all schools received support from the Primary School Development project. Eighty-five percent o f PSD finds were spent on civil works (Table 3.3), constructing a school pavilion (a cement floor and roof with girder supports) and house for the head teacher in each beneficiary school. I nreturn for the accommodation the head teacher was to sign an agreement with PTA and DEOC o n holding meetings out o f school time, providing teacher training, community relations, and attending training. The communities were to s i g n contracts to clad the pavilion (i.e., construct external walls) within six months o f completion. Table 3.3: Allocation o f resources under PSD project and BESIP Primary School Development BESIP US$ Percent US$ Percent School building and rehabilitation 38.0 67.1 16.3 34.2 Head leachers’ Housing 10.5 18.6 0.0 0.0 Training materials 2.1 3.7 1.3 2.7 Training 1.3 2.7 Teaching materials 0.0 0.0 2.0 4.3 School furniture 0.0 0.0 4.2 8.9 Textbook supply 0.0 0.0 16.4 34.3 Other expenses 6.0 10.6 6.2 13.0 Total 56.6 100.0 47.8 100.0 3.15 The project was restructured at the mid-term review to better support the FCUBE, incorporating the Education Management Information System (EMIS), provision o f teaching materials, a Schooling Improvement Fund (SIF), and an information, education, communication (IEC) program. These changes made little difference to the allocation o f funds. Less than 4 percent was spent o n training materials and training. 27 3.16 While components may be important even if they do not have much money spent on them, the Bank’s implemenation completion report rated the project as unsatisfactory noting that many required reforms had been only partially implemented. For example, schools did not provide the required length o f instructional time, community involvement was negligible other than in some SIF schools, and there was little impact from orientation and training o f officials, community leaders, and teachers. PSD’s main achievement was the provision o f physical infi-astructure . 3.17 The BESIP S A R stated that “despite increased resource inputs and enrollments, the reform movement has had very limited success so far in improving the quality o f teaching and learning outcomes” (p.5) so that “more attention has to be paid to software” ( S A R : 14). However, most Bank resources for the project were devoted to hardware and instructional materials, especially following the Mid-term Review when the project was restructured to focus o n three components: (1) c i v i l works, (2) textbook supply, and (3) EMIS. A s a result, the allocation to c i v i l works and goods increased by about $19 million, giving rise to the large share devoted to hardware and materials in project expenses: $15.4 m i l l i o n o f the total budget o f $47.9 were spent o n c i v i l works and a further $25.8 m i l l i o n o n goods (presumably mostly textbooks and furniture, though also including vehicles and other equipment). Approximately one-third was spent o n school building and rehabilitation, just over another third o n textbook supply and just under 10 percent o n school furniture: in total 77 percent o f the project budget was spent o n hardware and materials inputs. OF OTHER DONORS THEROLE 3.18 The Bank’s role should be put in perspective against the contribution o f other donors. The main agencies active in basic education are USAID and DFID, and some support to school buildingthrough the EU’s Micro-projects Program (Annex B). The largest contributions - USAID through QUIPS and DFID through Whole School Development (WSD) -have complemented rather than competed with the Bank’s inputs since there has been a focus o n software (district management, community participation, teacher training, etc.). QUIPS contains small grants to beneficiary schools that have been used for construction in many cases, but the program will only cover three schools in each district (totaling 330 schools) by the end o f 2004. O f more significance i s the EU MPP, which has financed some 1,500 classroom blocks (a block usually contains three classrooms) around the country. Regarding textbooks, the main input was $10 million from U S A I D under PEW in 1991. T o some extent these books would have replaced those supplied by HERP, which had become worn out, though U S A I D also supplied books for social science and sciences, which had been largely neglected in HERP procurements. In summary, the other donors active in basic education have by and large not overlapped in supplying the items provided by the Bank. Where they have overlapped the contribution o f other donors are not insubstantial, but are o n a smaller scale than those o f the Bank. 28 4. The Bank’s Impact on Education Policies and Outputs The Bank has provided bothfinance andpolicy support to the education sector over the past 15 years. Despite the clear government ownership o f the education reform program, the Bank can be argued to have played an important role in its implementation. FKhile critics argue that the reforms were carried out too quickly, it is at least as plausible that delays would have resulted in failure. The Bank’s policy conditions underpinned the reforms, its finance helped them be realized, assisted by technical support. Over the past I 5 years the Bank has provided close to 35 million textbooks andfinanced the construction o f 8,000 school pavilions, being the main provider o f both these types of support. THEBANKAND EDUCATIONAL REFORM 4.1 Critics o f the World Bank argue that it forces reform o n unwilling countries.33The evidence in this case suggests a contrary position. Here was a set o f reforms the government wished to undertake, which it used Bank assistance to carry out. The strong domestic ownership was shown by: 0 The strong domestic dynamic to education policy issues.34The reforms were not designed by World Bank staff, but based o n the recommendations o f the 1972 Dzobo Commission, restated by the Education Commission o f 1985. In the view o f the Bank task manager o f the time the reforms were accepted by the government in 1973, and the Bank merely helped bring them back to l i f e and simplify the curriculum, ensure that books would be available, and that schools would not be closed because there was n o food. 0 Aspects o f the reforms were not favored by the Bank, notably the increased vocationalization o f the curriculum. The Bank accepted this policy in order to retain its position supporting the education sector, staff saying that they saw no alternative at the time in view o f the strong position taken by the senior M O E official. The government wished to expand senior secondary education more rapidly than the Bank thought wise. In the end the Bank supported the Community Secondary Schools Project, for which it had tried unsuccessfully to find another donor. As a final example, the Bank quickly accepted the government’s view that it was politic to bring in a subsidized student loan scheme at tertiary level once feeding subsidies were eliminated. 0 The government, including the President, publicly reaffirmed their commitment to the reforms and made the case for them to the public. The reform program was first announced in national radio and TV broadcasts in October 1986. The more difficult “second phase” dealing with second-cycle and tertiary reforms beginning in 1990 33. For example, Heyneman states t h s position as “local policy makers have become passive recipients o f the Bank’s agendas” (2003: 3 15). 34. The strong domestic dynamic to the politics o f the education sector continues to t h i s day, as recently illustrated by the government’s decision in late 2002 to make English the mediumo f instruction from Grade 1 and the strong reaction from both domestic constituencies and donors. 29 received renewed support from the President. Furthermore, the Minister o f Finance frequently spoke o f the need for cost recovery in health and education, this case being repeated in the 1987 National Program for Economic Development. a The government took several decisive steps in support o f the reform prior to i t being launched and to ensure it was followed through, including substantial increases in education spending. 0 Finally, the reforms made sense given the political position o f the ruling Provisional National Defense Council (PNDC) at the time. 4.2 Why did PNDC embrace reforms that had proved politically difficult for well over a decade, and how was i t able to successfully implement them? The opposition to the reforms came from the middle class elite, which were not PNDC’s political base. During Rawling’s first year in power he directly attacked wealth and implemented stringent anti-corruption measures. H i s subsequent adoption o f the liberalization agenda can be attributed to the fact that it would undermine rent seekers to the benefit o f the wider p~pulation.~’ PNDC was not overly concerned about middle-class opposition. Students were a special case, since Rawlings did have support in the student-based June the Fourth Movement (EM).But J F M was o n the left wing o f the party, which was alienated by the adoption o f an IMF program in 1983. The loss o f this support base, and Rawlings’ populist inclination, implied a need to broaden PNDC’s appeal. Reform o f the education sector was an obvious candidate. The children o f the rural poor were either not attending school at all, or attending second or third-rate facilities, whereas the children o f the better o f f were enjoying the bulk o f govemment spending. Moreover, the benefits o f economic reform would take some time to reach rural residents outside o f the cocoa producing region, so expanding educational provision and improving quality would build support for reform more generally.36 4.3 The political commitment o f P N D C is clear from the decisive manner in which reform was handled. As the reforms got underway, key c i v i l servants were replaced and a new PNDC Secretary for Education appointed. She was joined by another prominent P N D C member as Deputy Minister who was to remain in the post for nine years and is widely recognized to have been the central figure in steering through the reforms. A second Deputy Minister, responsible for higher education, was in place for seven years. This team moved to end corruption, weeding out ghost workers (by the end o f 1986,5,722 ghost workers had been removed from second-cycle institutions alone) and regaining control o f educational policy from GES by relocating three divisions (Supplies, Curriculum Research and Development, and PBME) within the ministry. T o circumvent possible delays from GES opposition the army was mobilized to distribute textbooks to the new JSS schools. Student unrest was also tackled with a firm hand, with arrests and closure o f the universities -these 35. For an elaboration o f this argument see Sowa and White (2003). 36. The government also invested in rural infrastructure (roads and electricity). Bringing electricity to every district was a strongly held desire o f Rawlings, which was also supported by the Bank. See Tsikata (2001) for the argument that the PNDC used the aid-financed expansion o f services to buildpolitical support. Van Donge (2002) makes t h i s argument specifically for the case o f education. 30 strong moves did not threaten the government’s popularity since the universities were widely regarded as elitist (Tsikata, 2001: 73 and Nugent, 1995: 118). 4.4 Three roles can be identified for the Bank in supporting the reforms: (1) money, (2) technical assistance, and (3) donor mobilization. 0 The role o f money. Unlike some macroeconomic adjustment programs in which there may be nothing obvious to finance, the educational reforms in Ghana required financial support. The main requirements were teacher training in the new curriculum, textbooks and other teaching materials for that curriculum, and school building and rehabilitation for the expansion o f enrolments. The Bank supported each o f these activities. Even with the growth in spending o n education, the government was covering not much more than salaries, so the Bank finding paid for many o f these requirements. Bank assessments o f the impact o f the EdSACs argued that the local cost financing provided by these credits was central to the implementation and sustainability o f the reforms, allowing them to be completed before opposition could mount. Paying for activities that facilitated growing enrolments helped build support for the government’s educational policy. Money also contributed to the restructuring o f the ministry, which helped government to increase its control over GES. It used the reform program implementation and the EdSAC credit to cany out this agenda, such as the relocation o f key activities within the Ministry supported by Bank technical assistance. 0 Technical assistance. The Bank financed technical assistance for studies that played a role in planning, policy, and implementation. The project preparation facility financed both the school mapping exercise and the University Rationalization Study (URS). Technical inputs o n textbook design were provided, as well as more day-to-day support o n managing procurement. There was also informal influence o n these various aspects, in particular from the Bank’s education specialist resident in For example, he commented o n drafts o f the U R S before it was officially submitted to the Bank. H i s role in budget monitoring was mentioned in Chapter 3. 0 Mobilizing donor support. The Bank helped present the government’s case to outsiders. This was the first sectoral adjustment credit in education and the Bank was undoubtedly instrumental in coordinating donors in a w a y so as to support the reforms. 4.5 W h i l e the reforms were government-driven, the Bank did have some influence o n the shape o f the program. For example, the government was persuaded to restrict vocational training at JSS level to an introduction to tools. But there were other areas where the Bank was the one to give ground. For example, the Bank went ahead and supported senior secondary schools in a more full-fledged program that proved a costly failure, with $18 m i l l i o n wasted on workshop equipment that not 37. T h e task manager’s own view i s to be found in Bennett (no date). 38. A Bank review o f EdSAC I1estimated that $18 million were wasted o n workshop equipment for vocational training lying under-used and unmaintained. 31 Sustainability of Reform 4.6 The restructuring o f education i s w e l l entrenched. The 1996 manifesto o f the main opposition party, the National Patriotic Party (NPP),39 criticized the restructuring on the grounds that it was rushed and done with inadequate c o n s ~ l t a t i o nN .~o~intention was announced to reverse to the reforms. T o the contrary the origin o f the JSS system was traced to Busia’s Progress Party41with a commitment to ensure access up to JSS 3 for all Ghanaians. While cost recovery measures were criticized, the proposed policies put the state’s role as paying teachers’ salaries, with communities responsible for much else. Since NPP came t o power in 2000 there have been n o signs o f a policy reversal. The recent education sector strategy makes no reference t o changing the structure o f the system and reaffirms the decentralization measures introduced in the second half o f the 1990s. Donor Coordination 4.7 Donor support for EdSAC I had the features o f a sector-wide approach (SWAP) a decade before the term came into usage and the Bank can claim some credit for achieving this degree o f donor coordination. A pre-condition for a SWAP i s a clearly defined, government-owned sector strategy, which the first wave o f educational reforms clearly were.42The Bank was active in promoting donor coordination by facilitating donor discussion o n the sector and mobilizing co-financing for EdSAC I . An important stage in this process was a donor meeting held in Vienna in September 1987.43 Having a clear lead donor helps take a sector program forward and the Bank occupied this position, though it probably helped that the meeting was jointly sponsored by UNICEF, since donors m a y have resisted being directed solely by the Bank. Donor coordination continued in the early 1 9 9 0 helped ~~ by the fact that the PMU for EdSAC, which received technical assistance from the Bank, assumed responsibility for the management o f a l l external projects,44though not their policy n 1990 the Bank shared with U S A I D consultants functions, which rested in the ministry. I who designed the latter’s new project for basic education. The Bank went o n to play a role in setting up a donor forum for the education sector in August 1994. 4.8 Similar efforts were made to ensure a coordinated approach to the Basic Education Sector Investment Project (BESIP), but these were far less successful. Having been at the forefront o f donor coordination in the early 1990s, Ghana has had no education sector program 39. In the 1996 elections the NPP won 63 out o f 200 seats in Parliament, other parties took 6 seats with the remaining 131 going to the ruling NDC. 40. NPP (1996) Development in Freedom. Agendafor Change, Accra. 4 1. NPP traces i t s political heritage back to Busia and Danquah ( T s h t a 2000: 70). 42. Other pre-conditions relate to the overall policy and budgetary environment, which were satisfied in Ghana at least up until 1992. 43. A Bank staff member drafied the government document “The reform and rehabilitation o f the education system, 1987-89” showing the fmancing gap requiring donor support, which was discussed at the Vienna meeting. 44. The importance of t h s organizational change has been highlighted in a speech by the Minister for Education in the mid-90s (see Sawyer, 1997). 32 in recent years. Rather it has had three large donors (World Bank, USAID, and DFID) with remarkably similar projects under different management systems with an undoubted increased transaction costs for g o ~ e r n m e n tH .~ w did this situation arise? The structure appeared to be in o~ place for a sector program. There was already a donor coordination group and a government strategy (FCUBE). BESIP, the Bank’s project, was meant to be synonymous with FCUBE - the FCUBE document i s headed “the Basic Education Sector Investment Program” o n the cover page. Bank documentation during the preparation o f BESP fi-equently refers to the fact that a sector approach is to be adopted; the project budget - o f $250 million - covered the whole FCUBE program o f which $50 million was to come from the Bank. 4.9 The sector approach appears to have foundered o n donor competition, despite the efforts o f the Bank to encourage a government-led process. In July 1994, the Bank’s education specialist in Accra wrote to the Minister o f Education referring to discussions they had had o n developing a new approach to donor financing o f basic education in Ghana, proposing to invite donors to a preliminary assessment o f the sector later that month. The letter stressed the importance o f Government being seen to be firmly in charge o f developing the comprehensive basic education program. I t was suggested that the Minister formally write to other donors informing them o f the Ministry’s plans, and to undertake a Ghanaian- led analysis to develop strategy with a meeting in November to get donors o n board. But, whereas in 1987 other donors had n o experience in education and were willing to follow the Bank’s lead, this was not the case nine years later. The crucial episode appears to have been a workshop in London (supported by the Overseas Development Administration, n o w DFID) to develop a sector strategy: despite the fact that a strategy already existed and that n o other donors were invited to the meeting.46From this time onward first DFID and then U S A I D went their own way with programs to finance basic education. Only during 2003, with the new government strategy providing a basis, has a stronger degree o f donor coordination emerged. SECTOR OUTPUTS EDUCATION Budget 4.10 The Bank review o f B E S P estimated that the IDA credit represented about 8 percent of the annual MOE expenditures (recurrent + investment) o n basic education. This estimate under-states the importance o f the Bank’s resources since over 95 percent o f the government’s basic education spending i s for wages and salaries (see Annex B). Over the period 1989-2001, the value o f W o r l d Bank disbursements was one-third o f total government non-wage spending in education, but much higher for the basic sub-sector. This picture i s 45. The mid-term review for QUIPS states that “donor coordination under the FCUBE has been characterized as fragmented and lacking collective consultation on strategic plans and financing.. . between donors there has been l i t t l e regular sharing or coordination o f the key elements o f their programs” (Bonner et al. : 11) and “the lack o f coordination o f donor activities at the district level i s negatively affecting DE0 operations and attitudes” (ibid: 48). 46. Source: interview with DFID education advisor based i nAccra at the time. The incident i s also reported in DFID’s Development Efeectiveness Report, w h c h notes that both heavy DFID involvement i ndrawing up the strategy and the bilateral nature o f the meeting undermined the sector-wide approach (DFID, 2002, p.26 Box 5). 33 l i t t l e changed by taking into account the support schools receive from the districts, whose total spending is less than 5 percent o f govemment spending with about one-third going to education. W h i l e schools have benefited from these resources, they do not match the scale o f World Bank financing. 4.1 1 The scale o f the Bank’s operations has also matched that o f other donors. Bilateral aid to education totaled $350 million over the period 1989-2001, compared to the Bank’s $260 million. Within basic education the main players have been USAID ($88 million in the Primary Education Project and QUIPS), DFID (250 million in Whole School Development) and school building by the EU under i t s Micro-projects prog~am.~’ The value o f these bilateral programs approximately equals that o f the Bank. Activities 4.12 Table 4.1 summarizes the physical activities financed by World Bank resources.48As shown in the previous chapter, the bulk o f financing has been directed to c i v i l works and textbooks. 4.13 HEW began with the distribution o f 6.1 m i l l i o n textbooks to basic schools, which were mostly for math, English, and science.49In 1990, there were 2.8 m i l l i o n children in public basic schools, so they would have received, o n average, two textbooks each. Data collected toward the end o f HERP showed 100 percent coverage for 7 o f the 20 textbook titles printed, with an average o f 82 percent and a minimum o f 73 percent. For the 15 teacher guides printed average coverage was 78 percent, with complete coverage for 5 titles. HERP therefore turned the situation around f i o m one o f practically no textbooks in most classrooms to having one book per student in most schools for the three core subjects. This book supply supported the reform process by putting in place textbooks adapted to the new syllabus.so 47. These are not the only projects but the main ones. Other active donors include KfW (developing and printing local language textbooks), Japan, and UNICEF, and a large number of NGOs. 48. That is, funds used for technical assistance are not included. 49. The books were printedbased o n textbooks developed in the preceding years by a textbook committee. Books for other subjects, such as social sciences, were initially excluded as being less essential. Toward the end o f the project a small number o f social science texts were printed. 50. There were some delays in getting books into the classrooms in the first year. 34 35 4.14 However, intended textbook lifespan is only three years, the revolving fund for textbook procurement did not become well established, and fees for primary texts were dropped in 1995. Hence the bank has continued to supply textbooks, most recently 11 m i l l i o n books under BESIP. Under all projects combined the Bank has financed the provision o f close to 35 m i l l i o n textbooks. 4.15 School pavilions have been the main type o f c i v i l works, with over 8,000 o f these constructed under the various projects, for which the community was expected to provide the external walls. A s noted in various World Bank reports, this was frequently not done. Although the majority o f PSD-constructed schools were clad (Table 4.1) - econometric analysis shows that PSD made a significant contribution to schools having a greater proportion o f classrooms that can be used when it is raining (Annex D) -pavilions were constructed under all four o f the main Bank projects. M a n y pavilions remain unclad, frequently with l o w internal walls. PSD alone reached 25 percent o f primary schools, overall close to one-fifth o f public basic schools have benefited from World Bank c i v i l works in the past 15 years. 4.16 Other c i v i l works include head teachers’ housing under PSD, JSS workshops, and improved toilet facilities. There has been some provision o f school ftmiture, notably under BESIP, which rehabilitated 2,300 primary classrooms and provided furniture for them. 4.17 The Bank has been less active in other areas. A notable exception i s the teacher training provided at the time o f the reforms to both junior and senior secondary school teachers to ready them for the new syllabus. These were short one-off courses. As such they gave teachers some familiarity with the new JSS system and the new syllabus, hence supporting the reform program, but cannot be expected to have had a significant effect on teaching methods. But the further development o f in-service training has been much more the provenance o f other donor projects, notably QUIPS and WSD, as have support to SMCs, provision o f other teaching materials and encouraging improved teaching methods. 36 5. Educational Performance Has Improved Both educational attainment and achievement have risen in Ghana over the past 15 years. The enrolment rate has risen and dropouts reduced, so that completion has risen from 60 to 73 percent. The gender gap in primaly enrolments has been virtually eliminated and the gap in enrolments between childrenfrom poor and non-poor households narrowed. At the same time, test scores have improved. Children completing JSS today with nine years o f basic education per$orm better in the math and English tests than did children leaving middle school after ten years of schooling in the 1980s. SCHOOL ATTAINMENT: AND BETTER ENROLMENTS HIGHER COMPLETION RATES 5.1 The school system expanded throughout the reform period. The number o f basic schools increased by 50 percent fiom 12,997 in 1980 to 18,374 in 2000. This expansion has enabled rising enrolments. By 2000, over 90 percent o f Ghanaians aged 15 and above had attended school compared to 75 percent 20 years earlier (Annex H , Figure 3). The downtum in enrolments that had begun in the mid-70s was reversed. The basic school enrolment rate has risen steadily since the start o f the reforms, accumulating an increase o f over 10 percentage points between 1988 and 2001.51 5.2 GLSS data show continuously improving school attendance rates5*among children of primary and junior secondary school age (Figure 5. l).53 On the other hand, attendance rates at the secondary level showed a large initial increase but have since leveled off. 5.3 Growing enrolments have narrowed enrolment differentials. The gap between male and female enrolments has been virtually eliminated (Figure 5.2).54Closing o f the gender gap i s in part a function o f growing enrolments: when enrolments are 100 percent then there can be no gaps. The gender gap remains greatest where enrolments are lowest: notably in the Northemregion.55Enrolments have expanded most rapidly in the savannah (Northern and the two Upper regions), where the attendance rate for 7-12 year olds was just 52 percent in 1988. On the other hand, although rural enrolments have risen, they have not done so more quickly than those in urban areas so that the differential has remained. Finally, primary enrolments have risen more rapidly among the poor than the non-poor, although a substantial gap 51. Official data show no increase in enrolments from 1990 to 2000. Annex H shows the denominator (population) used in that calculation to be progressively under-estimated. Once this error i s corrected the Ministry o f Education data show the same rise in enrolments as that reported here f i o m GLSS data. 52. Attendance rate i s used here, as it i s by GSS, to mean the percentage o f an age cohort that i s currently enrolled in school. I t does not mean the percent present at school o n a particular day as a percent o f those enrolled. See Annex H for a discussion o f the different terms and the relationship between them. 53. The minimum age for enrolment i s 6, though most children begin school between the age o f 7 and 9. The age range 7-12 i s thus taken to correspond to primary children and 13-15 to JSS. Enrolment rates, reported in Annex H, tell the same story as these attendance rates. 54. Official data from the school census show a narrowing o f the gap, though it s t i l l remains. The most recent MoE publication reported 47.2 percent o f primary students to be female, compared to their population share in census data o f 49.6 percent (MoE 2002, Education Indicators at a Glance). 55. Beyond the scope of this study are the barriers girls face in accessing and completing school, including sexual harassment by teachers (on which see Leach et al. 2002). 37 remains. In 1988 only 60 percent o f the poorest quintile attended school compared to 80 percent o f the top quintile. By 2003 these figures were 77 and 94 percent, respectively. The narrowing o f the gap in enrolments between the poor and non-poor means that support o f the expansion o f primary education has been pro-poor. But for junior and senior secondary schools enrolments have grown more rapidly among the less poor (Annex H paragraph H.1.20). Spending for these sub-sectors also benefits the poor, but by less than the non-poor have benefited. Figure 5.1: M o r e children are attending school Figure 5.2: And gender and regional (attendance rates by age group) gaps are closing (enrolment differentials, 6-11 year olds) 0 . m ) c .P 0.70 . I 1 * ’ CL 0.60 7 to 12 13 to 15 16 to 18 1988 1992193 1998197 2003 1988 1992193 Age group 13 1998197 0 2003 -Female I I - Savannah - Rural Source: GLSS 2,3, and 4 and GSSlOED household survey 5.4 Repetition is not a large problem in Ghana since there i s automatic progression f r o m one grade to the next provided minimal attendance requirements are satisfied.56W o r l d Bank data show a repetition rate o f less than 5 percent in Ghana, compared to close to 30 percent in neighboring Togo.57Students may drop out before completing their education, though the data show this to be relatively rare, and declining. In 2003,95 percent o f those aged 15 or less who began primary school reached Grade 6, and 92 percent went o n to complete JSS.58 Fifteen years ago these figures were 86 and 73 percent, respectively, showing that retention, and so completion, has much improved. But variations remain. Completion rates are lower in rural areas, particularly in the savannah zone, where 9 percent o f students do not complete Grade 6 (Table 5.1). M a l e and female completion rates are comparable until grade 6, and a slightly higher proportion o f girls complete JSS than do boys. The poor remain more likely to drop out than the non-poor (Annex H , Table H.13). 56. Teachers in Ghana interviewed during fieldwork blamed poor student quality o n the policy o f automatic progression. In general, international evidence does not support the view that large-scale repetition improves student learning. N’tchougan-Sonou’s (2001) comparison o f Togo (which has repetition) and Ghana suggests that in the West African context there may be some effect, but she fails to allow for school quality. 57. World Bank World Development Indicators 2003. 58. These results are based o n a survival function, which takes account o f the censoring caused by children s t i l l in school (see Annex H for an exposition). 38 Table 5.1: Drop-out rates are low (2003) 5.5 Figure 5.3, which presents data from GLSS2 and the Percentage of those enrolling in grade I who complete GSS/OED survey, o n the percent o f Grade 4 Grade 6 JSS different age groups (all children) which have completed primary. Region These data c o n f i m the rising Coastal 97.4 97.4 92.5 completion rate in two ways. First, Forest 97.0 95.0 91.4 the line drawn from each survey is Savannah 93.2 91.5 n.a. downward sloping -within each Ruralhrban survey the data show that older age Urban 97.8 96.6 92.7 cohorts are less likely to have Rural 94.9 93.8 90.4 completed t ha n younger ones. Sex Second, the line for 2003 lies above Male 96.1 95.0 90.2 that for 1988. People aged 18-24 Female 96.9 95.5 93.1 , today are more likely to have Source: GSS/OED household survey. completed primary than the same age group 15 years earlier, both as enrolments have risen and drop-outs fallen. 5.6 The figure also shows completion rates for females from the 2003 data. The female completion rate has converged o n that for male over time, although a gap remains. Completion rates have improved for all income groups. In 1988 only 65 percent o f children entering P1 from households in the bottom quintile completed basic education; by 2003, 74 percent do so (some way below the figure o f 91 percent for the top quintile). TEST IMPROVED SCORES 5.7 In 1988, Ghana Statistical Service (GSS) visited 1,524 = - \ households in 85 different areas o f the country.59Each person aged 80 between 9 and 55 years and with 70 - at least three years o f schooling was asked to take a short English f .- 60- h 50 - reading test o f eight multiple 0 J 40 - choice questions and a math test Q) E 30 - E 4 o f eight sums (two addition, two 8 20 - e subtraction, two multiplication, s 10 - and two division). Those scoring 0 , five or more o n either test took a longer, more advanced test6' The results revealed the poor quality o f education being received by 59. This survey was the second round o f the Ghana Livings Standards Survey (GLSS2). 60. The short tests are in Annex A o f t h s report together with a sample o f the advanced tests. The full version of the advanced tests i s available on request. 39 Ghanaian children. Children who had completed three years o f primary education scored an average o f 0.8 o n the short English test -worse than if they had simply guessed all the answers.61Children who had completed all six years o f primary did not do much better, with an average mark o f only 3.1. In the simple math test the average score for primary graduates was 4.9. 5.8 Fifteen years later the GSS/OED survey re-visited the same 85 communities and carried out exactly the same tests in 1,740 households. The results clearly show that children are better educated today than they were 15 years ago. Primary graduates scored an average o f 5.6 o n the short English test and 5.7 o n the math test. These higher scores have been achieved in the context o f growing enrolments, so that a greater proportion o f those aged 9- 55 took the tests in 2003 than in 1988. 5.9 The improvement in the output o f the basic education system in Ghana is shown by Figure 5.4, which plots the regression-based mean test score against years o f education for 1988 and 2003.62The test score shown is the combined test score, which i s the s u m o f simple and advanced test scores, with a resulting maximum o f 37 for Englishand 44 for math.63 Several points emerge from these graphs. First, children at all levels o f basic education (grades 1-10 in 1988 and 1-9 today) score higher marks today than did their counterparts 15 years ago. Figure 5.4: School outputs have improved: test scores plotted against years o f schooling (a) M a t h (b) English English test 35 ! 30 , i 25 a I 20 I15 i I0 5- I t " 7 ' I I , I , I I ~ 04 , 0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18 2 Years of schooling Years of schoolina Source: GLSS2 and GSS/OED household survey n school, or who left it in the last year, who have completed Grade 3 but 61. The sample i s children currently i not Grade 4. 62. The line i s the fitted line f i o m the regression o f the combined test score on years o f education, using lowess (locally weighted regression) estimation. The sample comprises children in school or who have l e f t in the last three years. 63. See Annex G for a detailed discussion o f the analysis o f the test score data. 40 Second, Junior Secondary School graduates score higher than did Middle School graduates, despite the latter receiving 10 rather than 9 years o f education.64Third, the gain is larger at lower grades and for English i s reversed for secondary school graduates who score worse today than did their counterparts 15 years ago. Although not the subject o f this report, which focuses on basic education, the shortening o f pre-university education from 17 to 12 years m a y have been at the expense o f the quality o f senior secondary school graduates.65 More specifically, the data show that it is the compression o f the previous seven years o f secondary into three o f senior secondary that has caused this deterioration.66 5.10 W h i l e children o f better-off households on Table 5.2: Student performance average score higher, scores have improved for children i s returning to the levels attained o f households across the income distribution. There is 40 years ago (average test scores greater uniformity in performance across income o f primary school leavers by groups for primary school children today than 15 years decade) ago. Nevertheless for math the improvement has been English Math greatest for the children o f the relatively better o f f (Annex G, Table G.4). 1950s 5.4 4.2 1960s 2.5 4.1 5.1 1 A longer-term perspective is provided by looking 1970s 3.8 4.1 at the scores o f those leaving school after completing 1980s 0.9 3.2 Grades 5 or 6 across all age groups. Table 5.2 shows the 1990s 3.1 4.6 decade average scores o f primary school leavers. Both the English and math scores show a U shape, declining Notes: calculated for those leaving school after completing Grades 5 and 6. into the 1980s but then picking up in the past decade. 1990s is 1990-2002. Source: GSSlOED survey 5.12 The finding that educational outcomes are improving appears to run contrary to general concerns about the poor state o f Ghana’s basic education6’ There i s no contradiction here. While things have got better there i s s t i l l ample room for improvement. Nearly one-half (46 percent) o f children who have completed Grades 3-6 scored 5 or less on the simple English test, meaning they are barely literate and one-fifth (19 percent) scored 2 or less, i.e., the same as guessing, and so are illiterate. But 15 years ago these figures were 78 and 62 percent, respectively. Negative perceptions o f the state o f education arise im comparing the system today with that pre-crisis, some 30 years ago rather than 15. Such f o views also arise from continued middle-class discontent regardingthe reforms and their impact on senior secondary education. ~ 64. School reforms, discussed in detail in Chapter 2, reduced the length o f pre-university education from 17 to 12 years. 65. But it should also be bome inmind that recent graduates o f secondary education passed through basic education when quality was lower than today, which would have adversely affected their h t u r e learning. 66. The decline in quality o f secondary graduates i s commonly recognized. The restructuring of pre-university education was accompanied by an increase in the length o f B.A. degrees from three years to four in 1994. 67. For example, a recent report states that “not much has been achieved relative to improving the quality o f education in Ghana” (Educational Assessment and Research Center 2003). This point o f view was bome out in numerous interviews during fieldwork. 41 5.13 There i s corroborating evidence o f improved educational outcomes from the Criterion Reference Test (CRT) carried out since 1992, the mean English score rising from 29.9 to 36.9 between 1992 and 2000 and math by a similar amount (Figure 5.5). While covering a shorter time period than the two GSS surveys, the tests show the same clear improvement in test scores. W h i l s t the CRT confirms the improvement which has taken place i t also confirms that standards are s t i l l very low: the most recent CRT shows that in 2000 less than 10 percent o f children reached mastery level in math, and less than 5 percent did so in English. 5.14 The CRT scores also show Figure 5.5: Criterion Reference Test scores in public the better performance o f children schools have risen each year, 1992-2000 inprivate schools. In 2000 the mean CRT English score was 70 in private schools compared to 39 in public schools. It might be thought that the improvement found in the GSS/OED survey data can be attributed to increasing enrolments i inprivate schools. This i s not the case. The CRT data reported in 24 26 Figure 5.5 are for public schools 1992 1993 1994 1995 1996 1997 1999 2000 only and they show an Note: No data available for 1998 improvement in the 1990s. Source: MoE Similarly in the GSS/OED survey data, whilst students fiom private schools do better, it does not explain away the improvement in scores. For example, the average score in the simple Englishtest for primary graduates h as risen from 3.2 in 1988 to 4.9 in 2003. Considering public schools alone these figures are 2.8 and 4.6. The scores for public schools alone are lower, but the improvement is in fact a bit larger. 5.15 Since a larger proportion o f the population are n o w attending school, the improved test results mean that the average, quality-adjusted, level o f education (what economists call the stock o f human capital) i s rising. The most common education stock measure i s adult literacy, which i s usually measured indirectly as the percentage o f those aged 15 and above who have completed grade 5 (Annex H). However, the test score data allow calculation o f actual literacy measures, defining a person as literate in English if he or she scored 5 or more on the simple English test. Using this definition, the literacy rate among those aged 15-24 (which is the age group for the MDG indicator) has risen from 49 percent to 68 percent between 1988 and 2003, and in the population o f the whole from 37 to 45 percent. A decomposition analysis shows that the increase in school quality (higher scores achieved by those enrolled in school) accounts for over h a l f (57 percent) the increase in literacy, with the remainder coming from increased quantity (higher enrolments) (Annex H).68 68. Literacy rate = No. 1iterateJagecohort = No. literatelno. in school x no. inschoollage cohort, where these ratios measure the quality and quantity o f schooling respectively. The percentage change in the literacy rate may therefore be decomposed into the sum o f the percentage change o f these two components. 42 6. Causes and Consequences o f Improved Educational Performance Both educational attainment and achievement have risen in Ghana over the past 15 years. The World Bank, through its support o f the policy reform process and financing school-level improvements in quality (principally classroom construction and rehabilitation) and the availability of teaching materials, has contributed to these improvements. Improved educational performance has been onefactor behind better social indicators such as lower fertility and mortality and improved nutrition, as well as sustaining economic growth. INPUTS BETTER I m p r o v i n g School Efficiency 6.1 School efficiency has increased over the past 15 years in the following three ways: a The reduction in pre-university education from 17 to 12 years: the reduction o f basic education from 10 to 9 years was achieved with n o cost in terms o f children’s cognitive performance, indeed basic school graduates get better test scores today than they did 15 years ago. a Increasing instruction time: the school day was increased from four t o five hours. Although this reform took some time to take hold it has n o w become well-established (Annex D). 0 The increase in the pupil-teacher ratio: the PTR in primary schools increased from 30 to 36 over the period. 6.2 The Bank contributed to these efficiency improvements through i t s role in underpinning the reform process, documented in Chapter 4, and i t s pressure to restrain growth in GES employment. Improvements in School Facilities 6.3 The funds provided by the Bank have largely been used to improve school facilities, notably the supply o f nearly 35 m i l l i o n textbooks and construction o f close t o 8,000 school pavilions. The contribution of these to the level o f outputs was discussed in the preceding chapter. Prior to the Bank’s involvement, communities were required to construct their o w n schools, which were usually mud-walled structures o f limited suitability and lifespan. Neither other donors nor the government have been as active in the fields focused o n by the Bank so that it has been the major player in improving these aspects o f school quality. I m p r o v i n g Teacher Conditions 6.4 One channel through which school inputs affect student learning outcomes is through their impact o n teacher performance. A number o f teacher level variables, such as time o n task, the use o f improved teaching methods and their monitoring o f student perfonnance, have an 43 effect on test scores. These teacher variables depend, in tum on school facilities and management. The GSS/OED teacher survey asked respondents o f their perception o f their living and working conditions. Unsurprisingly, multiple regression analysis shows the former to be most strongly related to the availability o f water and electricity in the home and i f lodging is provided. Perceptions o f working conditions are related to classroom quality (if disrupted by noise, the presence o f internal wall, and quality o f the chalkboard) and the availability o f teaching materials. School management (an active PTA and contact with the circuit supervisor) also has a positive impact on perceived working conditions. 6.5 Teachers’ attitudes to both their working and living conditions are strongly influenced by whether o r not they receive their pay on time. The problem o f late pay, which i s more common for new teachers, has been considerably alleviated over the years but i s s t i l l an issue, with 28 percent o f teachers often not receiving their pay on time. 6.6 H o w teachers feel about their living and working conditions matters as it affects teacher morale, which affect both absenteeism and the likelihood o f remaining a teacher. Teachers were asked if they enjoy being a teacher and if they intend to remain one or not. These two variables were combined to construct a measure o f teacher morale. B o t h living and working conditions are significant determinants o f teacher morale. And teacher morale i s in turn a significant determinant o f the likelihood o f a teacher being guilty o f absenteeism (Table D.43). B o t h teacher absenteeism and the outflow o f trained teachers constitute considerable sources o f inefficiency in the system. INPUTS TO OUTPUTS (ATTAINMENT AND ACHIEVEMENT) FROM Increasing Enrolments 6.7 Multivariate analysis o f the child-level data from 1988 and 2003 shows that the following factors are significant determinant o f whether or not a child attends and stays in 0 Child characteristics: children with more siblings are less likely to attend school, especially those o f lower birth order. This finding fits with the common observation that older children work to pay for the education o f their younger sibling, being themselves deprived o f education. The female dummy was significantly negative in 1988 but not in 2003, indicating the closing o f the gender gap in enrolments. 0 Household characteristics: children o f higher income households are more likely to go to school, as are children o f more educated parents. The latter results means there i s an “inter-generational multiplier effect” as children sent to school 15 years ago as a result o f improvements in the education system are more likely t o send their o w n children to school today. nh e x I 69. The regression results are reported i .The results are from a Cox regression o f school attainment, which was used in preference to the censored ordered probit which has been more commonly i n the literature. However, the censored ordered probit model and a simple probit o f enrolment give similar results. A regression o f n enrolments also gave s i m i l a r results. the cluster-level change i 44 0 Proximity o f school facilities: the greater the distance to the nearest primary school the less likely is a child to be enrolled. This effect was stronger in 1988 than 2003, as school building means that the vast majority o f children now live close to a primary school so there i s little variation in the independent variable. In 2003 children were significantly more likely to attend school if there was a private school in the vicinity. 0 Quality o f school facilities: the school having an adequate number o f classrooms to cater for a l l grades significantly affects the likelihood o f a child going to school, as does the availability o f other materials such as chalk and desks. 70 0 Staffing: In 2003 parents were more likely to send their children to schools that had a l o w pupil teacher ratio and less likely to send them to schools with a high ratio. These results are picking up two phenomena. One i s that overcrowding deters parents from sending children to school. But a high pupil teacher ratio also results from having insufficient teachers - one or two to cover four or even all six grades, which i s not unknown in rural areas -which makes parents doubt that their child will receive an education. 6.8 Building a school, and so reducing children’s travel time, has a major impact o n enrolments. While the majority o f children live within 20 minutes o f school, some 20 percent do not and school building has increased enrolments among these groups. In one area surveyed, average travel time to the nearest school was cut by 45 minutes with enrolments increasing from 10 to 80 percent. In two other areas average travel time was reduced by nearly 30 minutes and enrolments increased by over 20 percent. Calculation using the regression estimates suggest that o n average building a school in a community in which the children previously had to walk an hour to school will increase enrolments in that community by around 5 percent (Annex I ) . Building a new classroom block to replace an unusable one will increase enrolments in the school’s catchment area by 7.5 percent. 6.9 The regression estimates can be used to examine which have been the most important factors behind enrolment growth: 0 The largest single effect comes from the elimination o f gender bias, accounting for a 4 percent increase in enrolments. This autonomous effect partly reflects the success o f efforts to get girls into school, though these are not something the World Bank has directly supported. 70. N o data were available in 1988 o n seating places. Typically a class w i l l have both desks and chairs or neither. Where chairs are not available students bring their own from home, being a substantial “in-kind” parental contribution. The desks variable may also be picking up h o w the cost o f providing a chair deters parents. 71. Thls i s done by decomposing the observed change in enrolments to the sum o f the product o f the regression coefficients and the difference between 1988 and 2003 o f the mean value o f each o f the explanatory variables. Ths analysis was carried out using the simple probit estimates o f enrolment (Annex I). The enrolment increases sum to more than the actual increase as they are offset by a negative s h f t in the survey dummy. An important caveat i s that the system relies upon government financing teachers’ salaries, but the importance o f that i s not captured in the analysis since government fulfilled h s function, leaving l i t t l e variation in the relevant explanatory variables. 45 0 Higher household incomes have accounted for enrolment growth o f about 2.5 percent over the period. Increased parental education accounts for close to another 2 percent. 0 Improved school facilities, including reduced distance to school, have accounted for about a 4 percent increase in enrolments between 1988 and 2003, about one third o f the increase over that period. A large part o f this improvement can be attributed to the World Bank, which has been overwhelmingly the main funder o f better infrastructure in this period. However, this attribution must be seen in the context o f a fimctioning education system in which government ensures a supply o f trained teachers. Determinants o f Test Scores 6.10 Linking children to the school they have attended allows a regression analysis o f the determinants o f test score outcomes incorporating both school and household characteristics. These regressions are reported in Annex G. Schooling improves test scores, each additional year o f schooling increasing the combined English score by 3.6 points and math by 4.9 points.72The 10 percent o f the age group attending school who would not have done so 15 years ago can be expected to increase their combined English score by 20 points if they complete primary (as 95 percent do) and 27 if they go o n to complete JSS (as do 86 percent). For math these figures are 16 and 2 1,respectively. 6.1 1 The direct impacts o f the recorded increase in material and physical items between 1988 and 2003 increased math scores by 1.6 and Englishby 2.0 points. These figures understate the gains in the most deprived areas. Ensuringthat a school has one math and English book per child compared to the situation in the mid-1980s o f one text per classroom will increase average English scores o f children in that school by 6 points and math scores by close to 10 points. 6.12 The 2003 school and teacher surveys collected data not collected in 1988, allowing a more detailed analysis o f test score determinants. Measures o f the quality o f school infrastructure, in particular if classes are disrupted by noise, the presence o f internal walls and chalkboard quality, all have a significant impact o n for test scores. The combined effect of these three infrastructure variables can improve English scores by 11.3 points and math by 10.1. 6.13 The regressions using the 2003 data also show that process matters. The most important single variable in determiningtest score outcomes is teaching methods. If all teachers in the school used improved methods then, compared to a situation in which none do so, children’s English scores would be 6.2 points higher and their math score 8.8 higher. Important determinants o f use o f improved methods are teacher training (notably for teachers in the coastal region), including in-service training. Supervision by the head-teacher and contact with the circuit supervisor also increase the likelihood that improved methods will be adopted. None 72. The years o f schooling slope dummy included in the regressions was not significant. That means that the school level factors accounting for better test scores are included in the model. 46 o f these are areas to which the Bank can be said to have contributed. Efforts to improve head- teacher performance through the provision o f housing were judged to be unsuccessfu1.73 6.14 Questions to teachers regarding the use o f classroom time allowed the construction o f a time o n task variable, and this too was found to significant affect test scores. Time o n task naddition, the itself i s a function of teacher training and contact with the circuit supervisor. I ability o f teachers in the school to speak the local language improves student math scores, presumably since they do not have to rely o n English, o f which students may have a poor grasp, to explain difficult concepts. 6.15 The results reported in the preceding paragraphs pose something o f a puzzle. Better teaching methods and time o n task improve test scores, and teacher training enhances both o f these things. But private schools largely recruit untrained teachers and it i s w e l l established that, o n average, private schools get better test results (chapter 5). There are two answers to this puzzle. The first is the finding that the private school dummy i s significant for English scores but not for math. So, once factors relating to both the student’s background and school facilities are controlled for, there i s no pure “private school effect” for math. There i s however one for English, perhaps reflecting the enforcement o f English as the language o f tuition in these schools. The second answer to the puzzle i s that there are indeed aspects o f private schools, such as teacher discipline, which are conducive to good learning outcomes. This fact does not contradict the finding that children will learn better still, even in private schools, if improved teaching methods are employed. 6.16 Home factors also matters to student performance. The two measures o f parental involvement in a child’s education (membership o f PTA and meeting with a teacher) give a combined impact o f 3.5 and 3.9 points o n math and English scores respectively. Income also matters; economic growth (the between sample rise in incomes) has increased average English scores by 2.2 points and math scores by 1.2 points. As in the case o f enrolments, to the extent that education affects these household characteristics there i s a multiplier effect whereby the educational performance o f children o f educated parents improves. 6.17 Textbook provision i s a very cost-effective means o f improving learning outcomes, with teacher training being the next most cost effective (Annex G). School infrastructure also has a beneficial effect o n learning outcomes, but its largest benefit is from enabling higher enrolments. FROM OUTPUTS TO OUTCOMES Education and Social Outcomes 6.18 There is a well-established literature linking educational outputs to welfare outcomes, both economic and social. Studies of education and social outcomes tend to focus o n the effect o f female education (Table 6.1). Where both male and female education i s included, 73. An internal Bank review o f the Primary School Development project judged that the lack o f improvement in teacher perfonnance and supervision showed that the provision of head-teacher housing was an ineffective strategy to improve head-teacher performance as school-level supervisors. 47 then the latter i s shown to be more important.74The most commonly studied outcomes have been fertility and child nutrition. All studies from Ghana find that the higher levels o f education reduce fertility, normally measured as the number o f births. Education measure Child survival Fertility Contracepfive prevalence Nu trition 1 Alderman (1990) Benefo and Female schooling Male schooling Mother’s education Positive l Positive l Insignificant Insignificant Schultz (1996) Glewwe and Desai Test scores (1999) insignificant (mother’s math positive) Gyimah (2002) Secondary or higher Positive Gupta and Mahy Maternal education: Positive (2003) None, 1-7 years, 8 or more Oliver (1999) Mother’s years of Positive schooling, test scores Rue1 et ai. (1999) None, primary, Positive secondary Maxwell et al. Mother’s education Positive (2000) level Sackey (2003) Mother’s years of Positive Positive Positive Positive schooling Father’s years of Positive Positive Insignificant Positive schooling OED analysis Maternal education Positive Paternal education Positive (indirect through income) 6.19 The findings from studies o f child nutrition are more ambiguous. An early study found no significant impact, but more recent studies find that education does improve nutritional status. Both Rue1 et al. (1999) and Maxwell et al. (2000) presenting different analyses o f the same data from Accra find that mother’s education is significantly associated with better child nutrition. In addition, there i s a considerable indirect effect from education o n improved childcare practices, which also improve nutrition. Good care practices, supported by education, can compensate for lower income. Hence the nutritional status o f children o f educated mothers 74. Equations that include income as an explanatory variable do not capture the indirect effect o f education through income. This w i l l be one o f the channels through which male education matters. Ths argument i s supported by Maxwell et al. (2000) who instrument for income with father’s education, finding it to be significant. The same i s true o f Alderman (1990), which may partly explain why the education term itself i s not significant. 48 at lower income levels can that Of Table 6.2: Percentage reductions in welfare children in higher-income families. outcomes caused by higher school attainment Children o f mothers with little education living in low-income households have the Fertility Mortality Nutrition worst nutritional status. Benefo and Schultz -2.4 -3.6 - 6.20 Analysis carried out using the Rue1 et al. - - 10.3-20.6 GSS/OED data supports the view that Sackey -4.8 -0.7 - education can substitute for income in achieving better nutritional outcomes. OED analysis 4.8-27.2 These data show that maternal education has a significant impact o n child nutrition Source: Annex K (Annex This impact is higher for women in poorer households. I nthe poorest households a woman completing JSS increases the expected nutritional status (measured by height for age) by 4.5 points, sufficient to move the child out o f the category o f being malnourished. 6.21 A smaller number o f studies establish the link between education and lower child mortality. For example, Benefo and Schultz (1996) find a weak impact o f mother’s education o n child mortality (but a strong one o n fertility). They show these effects to b e stronger when women live in a community with good access to water and weaker when there is poor access to health facilities, thus supporting the view that education facilitates the better use o f other amenities so as to improve welfare outcomes. The channels through which education operates are also shown by studies showing education to affect both income and child care practices. 6.22 The results from the studies mentioned above can be used to investigate the scale o f education’s impact o n welfare outcomes. Infrastructural improvements will result in an increase in enrolments o f around 10 percent and reduce the dropout rate (Annex I). Hence, the average schooling o f mothers will rise. This rise in schooling leads to relatively small, though not negligible, changes in fertility and child mortality (Table 6.2).76The study o f Accra shows the impact o n nutrition (stunting, Le., the height for age z-score) a more substantial impact o f an improvement o f between 10 and 20 percent. The channels for this increase are both improved childcare practices and higher income. OED’s o w n analysis supports these substantial effects. The lower increases from education are realized among wealthier families, with the largest absolute gain to poorer households. Education and Economic Outcomes 6.23 The most comprehensive analysis o f the economic returns to education in Ghana i s that by Teal (2001), which brings together data from four rounds o f the GLSS (1987/88- 75. There i s an impact from father’s education in some specifications. But in general the impact offather’s education i s indirect though i t s effect o n household income. 76. T h e results shown here ignore the feedback loop that operates between lower fertility and reductions in mortality. 49 1998). H e finds that there is a positive return to all levels o f education, but that it i s higher for higher levels.” The rise in the average level o f education accounts for about one-third o f growth in per capita income that has taken place over the decade 6.24 Analysis o f the GSS/OED presents a clear message: education matters only to the extent that it results in higher cognitive achievement (Annex K). Education can affect earnings both directly -more educated people earn more, which may result simply from a screening function - and indirectly through raising their cognitive skills, which are rewarded with higher earnings. OED’s analysis shows that there was a direct return to education in 1988 but this i s no longer so for primary and JSS in 2003, for which the return in fact appears to be negative. But schooling raises test scores and those with higher test scores earn more. Those who get higher test scores as a result o f schooling do enjoy higher earnings. To the extent that poorer children in less well resourced schools are not reaping educational benefits from school attendance nor will they enjoy economic gains, generating a vicious circle of poverty. 77. In most areas o f the world the returns are highest to primary education. However, this i s not generally the case in Africa (Bennell, 1996). 50 7. Lessons Learned and Progress Toward the MDGs THE MILLENNIUM PROGRESS TOWARD GOALS DEVELOPMENT 7.1 The education MDG i s to “ensure that, by 2015, children everywhere, boys and girls alike, will be able to complete a full course o f primary schooling.” In addition, the third MDG o f gender equality has two education-related targets (progress toward gender equality in enrolments and literacy). Table 7.1 shows the progress made in the period 1988-2003 and extrapolates these trends to 2015. Ghana has made considerable progress toward both these goals, in particular the gender equality in education has been achieved with respect to primary enrolments and i s likely to be achieved, or close to achieved, for the other gender indicators. On the other hand, at current rates o f progress enrollments will fall short o f the target o f W E and up to nearly a quarter o f those aged 15-24 will be illiterate.78 Closing these gaps will require, among other things, focusing attention o n difficult-to-reach areas and ensuring that all schools receive the required level o f inputs.” This i s not simply a matter o f regional disparities: there are deprived schools in even the better o f f districts. Table 7.1: Progress toward the education related MDGs 1988 2003 Predicted for 2015a Low Hiah Goal 2: Universal Primary Education I Complete Grade Six Total 65.4 76.9 82.1 87.5 Girls 55.6 72.1 80.1 88.8 Net enrollment ratio in primary education 72.1 84.1 89.5 95.1 Proportion of pupils starting grade 1 who reach grade 5 88.5 95.8 98.9 100.0 Literacy rate of 15 to 24-year-oldsb 49.0 68.0 77.6 88.4 Goal 3: Promote gender equality and empower women I Ratio of girls to boys in primary and secondary education Primary 95.0 100.0 100.0 100.0 Secondary 78.0 86.0 88.8 91.7 Ratio of literate females to malesb 83.0 92.0 95.0 98.0 78. The literacy rates shown here are lower than the official figures since the numbers here are based o n actual reading slulls rather than school attainment (see Annex H). 79. T h i s report has demonstrated the importance o f material and physical inputs in supporting enrolments and student learning. I t may be the case that the “last 5 to 10 percent” comprise difficult to reach groups including street children, orphans and disabled and that separate, more costly, measures are required to get these children in school. 51 LEARNED LESSONS 7.1 The main conclusions from this report, and the lessons to be learned from them, are as follows: 0 The Bank focused its spending o n hardware and instructional materials, even when the rhetoric o f strategy and project documents turned toward software. This focus turns out to have been beneficial. The inputs the Bank has provided (books and buildings) have been shown to have made a significant contribution to both educational attainment and achievement. Two caveats are perhaps in order: (1) the focus o n hardware and materials took place within the context o f an agreed program o f educational reform with a government that has been committed, especially in the early years, to improving the quantity and quality o f education; and (2) the projects o f the other major donors (DFID and USAID) have focused o n software, though the extent o f their impact is not yet widespread. 0 The lesson to be drawn i s that getting enough classrooms and those classrooms being in decent shape i s a necessary ingredient o f educational strategy. But it cannot be the sole ingredient. Indeed, i t will become less important as all schools attain the desired level o f physical and material inputs. But Ghana i s not yet in that position: substantial inputs are s t i l l required for the most disadvantaged schools. Even where good school quality is achieved, educational outcomes, while improved, are still far from satisfactory. Improving them will indeed require attention to software. 0 The evidence in this report o f beneficial effects from community management is not strong. However, it can be argued that these changes have yet to really take h o l d so that their effect can be felt. I t i s shown that parental involvement does matter, but this could well be proxying for parental interest in child’s education. But the evidence i s clear that supervision o f teachers by the head teacher and circuit supervisor matter, as do the teaching methods adopted by the teacher. Since attempts to remove untrained teachers have been unsuccessful, and since not all trained teachers appear familiar with student-centered approaches anyway, there i s a strong case for pushing forward with efforts to emphasize the role o f in-service training. Efforts should also be made to retain trained teachers, which m a y suggest some reconsideration o f the current policy regarding study leave. Finally, teachers being able to speak the local language helps student math learning. 0 The downside o f community and district financing o f schools i s that it leads to disparities in resource availability. There remain a class o f schools in poorer communities -particularly but not only in rural areas -which are very poorly resourced and the community can do nothing about i t and often lacks the political connections to attract district finance. 7.2 Some immediate implications o f the analysis are: 52 e It i s necessary to focus resources o n the most needy schools. The bias that results from community-based financing needs to be overcome. School mapping continues to play an important role, so support to EMIS is important. e The private sector has been neglected, but i t i s important, so attention needs to be paid t o it in both government strategy and Bank support. Possible areas o f attention are enforcing registration to avoid very poor schools, but taking care to not be too restrictive. Teacher certification could be required, and while there does not seem a need to require formal teacher training, the provision o f in-service training would help promote better learning outcomes. e Teaching methods matter a lot for test outcomes. Teacher training seems to affect this, but differentially. This supports the idea that teacher training should look at method as well as content. In-service training matters, but there i s not much o f it at present. Teacher morale i s reasonable, but is affected by living and working conditions and especially if teachers get their pay o n time. Resolving payment problems will raise morale and reduce absenteeism. Inputs matter, textbooks in particular. While sustainable textbook financing i s a desirable goal, donors should not be averse to large-scale textbook provision, such as the W o r l d Bank has done. Thought might also be given to providing exercise books and pencils to the most needy basic schools (perhaps by geographical targeting to bottom 20 districts, but probably not given the substantial intra-district variation in school quality). 53 Annex A Annex A: Test Examples Short maths test 1. 1+2= 5. 24+ 17= 2. 5-2= 6. 33 - 1 9 = 3. 2x3= 7. 17x3= 4. 10+5= 8. 41+7= 54 Annex A Short English Reading Test John is.a small boy. H e lives in a village with his brothers and sisters. H e goes to school every week. In his school there are five teachers. John i s leaming to read at school. H e likes to read very much. H i s father i s a teacher, and his parents want him to become a school teacher too. 1. Who i s John? 5. What i s John doing at school? (A) Anoldman (A) Helping the teacher (B) A small boy (B) Tallung with his friends (C) A school teacher (C) Learning to read (D) Aschool (D) Teaching the class 2. Where does John live? 6. Who i s a school teacher? (A) Inavillage (A) John (B) Inacity (B) John’s father (C) Inaschool (C) John’s brother (D) In a forest (D) John’s mother 3. What does John do every week? 7. What do John’s parents want him to do? (A) Works with his father (B) Plays with h i s friends (A) Go to school (C) Helps his brothers and (B) Learn to read sisters (C) Obey his teachers (D) Goes t o school (D) Become a teacher 4. How many teachers are there at 8. The best title for this story i s John’s school? (A) John Learns to Read (A) om (B) Why Reading i s Important (B) Three (C) John’s Village (C) Five > (D) Schools in Ghana (D) Six 55 Annex A A sample o f questions from the Advanced Mathematics test 3. There are 4 rows o f chairs and 12 A I_ chairs in each row. H o w do you find out the total number o f chairs? 12 cm. (A) 12+4 (B) 12-4 B 12 cm. C (C) 12x4 (D) 1214 Note: figure not drawn to scale 14. If the perimeter o f the triangle 1 1 ABC i s 30 centimetres, what i s 8. - + - - - the length, in centimetres o f side 2 3 AB? 1 36. Which CANNOT b e the intersection o f 3 planes? (A) 1 point 13. 1% o f 400 i s (B) 1 line (C) 3 concurrent lines (D) 3 parallel lines 56 Annex A A sample page from the Advanced English test (...) The cat brushed against the o l d man. H e did not move. H e only stood, staring in the window o f the house. The party inside looked warm and friendly, but no one noticed him. The o l d man walked sadly on, followed by the cat 8. What was inside the house? (A) A Party (B) Some dogs (C) An old lady (D) A meeting 9. The man is described as being (A) Old (B) Young (C) Thin (D) Small Directions: For questions 10-15, read the passage below. Each l i n e o f the passage has a number. In each line, there i s a box with four possible choices. Pick the choice that best completes the sentence in each numbered line. Mark the letter (A,B,C, or D) o f the choice o n your answer sheet. (A) hears. 10. Sound i s something we I t comes to your (C) heard. (A) Eyes 11. (B) in different ways. I t might be pleasant, (C) ears I (D) mouth I (A) when 12. like the voice o f a fiend, unpleasant, like the velr> (...) 57 Annex B Annex B: Budget Analysis 1. Education expenditure data were obtained from the Ministry o f Education (MoEYS), which compiles annual data as provisional (budgeted) and actual, broken down to budget lines and functional classifications. These data were provided to the study team by the ministry for the period FY 1989-2001, although the data for 1993 could not be located. 2. This annex i s primarily a technical note that explains how the analysis was performed. Some observations are made as to the results insofar as are requiredto support the argument in the main report. Trends in Education Spending As a percent o f government expenditure and GDP 3. Data on expenditure as a percent o f GDP were collected from both the Ministry o f Finance and Ghana Statistical Services. However, the education expenditure data from these sources did not correspond to those providedby the Ministry o f Education, although the discrepancy was relatively minor (in most years i t was around 5 percent, which was the median, in only one year did it exceed 7 percent, reaching 17 percent). Since the MoF/GSS series were only available until 1997, the presentation here i s based on OED’s own calculations, using the MoEYS expenditure data and GDP data from World Development Indicators. The latter series i s identical to those from MoF/GSS other than 1990-92. For the years 1982-88, the MoF/GSS data were used. The discrepancy in 1988 (the year in which the series are joined) is just 0.2 percent o f GDP, so no adjustments were made in linking the data. However, five new budget lines were added under the Ministry o f Education budget heading in 1999.’ Since these budget lines are for existing institutions they have been excluded from the totals shown here.’ The resulting data show a substantial rise in education’s share o f GDP over the period 1984-87 and a continued, but erratic, rise since then. 4. Data on education’s share o f government spending are available from the Quarterly Digest o f Statistics published by Ghana Statistical Service. The tables provide the total o f recurrent and capital (development) expenditure o f education. The resulting percentage i s shown in Figure B.1. The share o f education rose from 1982-87 and then leveled o f f before falling in the 1990s. This decline has been slightly mitigated by spending through the districts, which i s discussed below. 5. In August 2001 the Ghana Education Trust Fund (GETFund) was established by act o f parliament to provide additional resources to all levels o f education financed out o f an equivalent to two and one half percent out o f the prevailing rate o f the Value Added Tax and any other contributions. I npractice the bulk o f the funds have been used for Tertiary education: in 2002 o f the 140 billion cedis disbursed, 125 billion (89 percent) were allocated to the tertiary level, the bulk o f it (90 billion) being a contribution to the student loan scheme. 1. The five are UNESCO Commission, West Africa Exams Council, Ghana Library Board, National Service Scheme, and the Ghana Book Development Council. These five budget lines account for j u s t over 2 percent o f total expenditure. 58 Annex B This use of general taxation to finance tertiary education is a regressive fiscal policy. The value o f GETFund disbursements is around 10 percent o f government spending o n education, thus boosting overall education spending but reducing the share o f basic education in that spending. Figure B.1: Trends in education expenditure 70 6.0 60 ~ -- 5.0 - u) C - .- 50 - -- 4.0 e n 4 40- (3 y. - Q 0 0 -- 3.0 c 2 C m Q 30- &- $ -- 2.0 - - * C 20- h 0 - m Q n \ 10 - I -- 1.0 0 1 , , , , , , , , , , , ' , , , , , , , 0.0 C 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 Share of government expenditure -Real expenditure -Share of GDP (right axis) Source: MoEYS and GSS data Real education expenditure 6. Analysis o f real expenditure trends often deflates expenditure by the consumer price index (CPI) or sometimes the GDP deflator. However, this procedure i s inappropriate if there has been a change in real wages, especially if wages are a substantial component o f total expenditure. The correct approach is to separate out wage and non-wage elements o f expenditure and deflate the former by a wage index and the latter by a price index. 7. The latter approach was followed here, using the non-food component o f the C P I as the price index, re-based for 1 9 8 9 4 00. There i s a break in Ghana's C P I with a new series reported since 1998 and n o GSS data provide an overlap between the two series. However, the IMF Statistical Annex reported an annual inflation rate for the non-food C P I for 1998. This figure was used to link the two series. 8. The wage index was constructed as follows. Data are available o n the personal emoluments for all budget headings. However, w e do not have employment data corresponding to each of these budget heads. But data are available o n the number o f teachers employed in primary and JSS. The implicit salary per person was calculated by dividing personal emoluments for primary by the number o f teachers in public primary schools. The same calculation was performed for junior secondary. A weighted average was 59 Annex B taken o f the two series using the respective weights in employment over the whole period (66:34 for primary:JSS). This wage series was re-based at 1989=100. 9. The wage and price series were used to deflate personal emoluments and the non- salary component o f expenditures respectively. The resulting series were summed to give total real expenditure. 10. The result i s shown in Figure B.1. There has been a continual, if slightly erratic, increase in real expenditure throughout the period 1989-99, which has been reversed in the last two years for which data are available, though GETFund began disbursements in 2001. Basic Education Share o f Education Spending Data and method 11. The education budget i s divided into the following headings: 140, covering M o E Y S headquarters, 141 for GES, 142 for regional services, which includes both the cost o f district offices and the funds flowing to school facilities themselves, 143 for special education for the handicapped, 144 for national culture (including archives), and 145 for tertiary education. Each o f these codes is divided into several lines corresponding to the departments and units within the various organizations. 12. The Ministry o f Education spreadsheets show provisional (budgeted) and annual expenditure by each budget line, separated into personal emoluments and non-salary items. Recent spreadsheets provide a different breakdown o f non-salary items, including investment costs, whereas data for earlier years refer solely to recurrent costs. The share to basic education i s calculated by applying a coefficient to each line item. For items solely dedicated to basic education (that is, the basic education staff within GES, and the primary school and middle/JSS budget lines) a coefficient o f one is assigned. The coefficient for other budget lines varies, the most common value being 0.6. All central M o E Y S budget lines have a weight o f zero. These coefficients are set by the ministry based o n their experience and are adopted here. 13. The basic education share shown in Figure B.2 was calculated from the M o E Y S spreadsheets using the MoEYS’s own methodology. The results were checked against the MoEYS’s own reported figures for the basic share and discrepancies reconciled.’ M o E Y S estimates o f the basic share were not available for 1999-2001, but replication o f MoEYS’s figures for 1989-98 i s a safeguard that the figures reported are accurate. 14. The MoEYS spreadsheets break down the basic share into primary and JSS by using a set ratio o f 0.7:0.3. Following this procedure results in the same trends being observed as can be seen in the basic education data. However, the procedure loses the information contained in the fact that the two largest single line items -primary and middle/JSS, which together 2, T w o revisions o f more than h a l f a percentage point were made to the MoEYS basic share figures. Inthe 1992 spreadsheet, the basic education line i t e m had been omitted f r o m the calculation, and in the 1995 spreadsheet, the central adrmnistration line for regional services had been given a weight o f 1rather than 0.6. 60 Annex B account for over h a l f o f total spending and over 80 percent o f basic education expenditure - can be allocated t o these two categories. The primary school share reported here was calculated by (1) deducting the sum o f the primary and middle/JSS budget lines from the basic school total; (2) adding the primary school line to the pro-rated (0.7) residual; and (3) calculating the resulting share. The equivalent procedure was followed for middle/JSS, using a share o f 0.3. Figure B.2: Basic education share o f central government spending s .- 5 50- P 40- e i30 : 20 1 -Provisional -Actual +Primary (actual) -JSS (actual) 15. This methodology implicitly distributes administrative costs, other than those o f MoEYS HQ,across the sectors proportional to their share o f spending. Hence “basic share” i s not a measure o f funds flowing to basic schools. An altemative approach i s to attempt t o separate out administrative costs into another line item. However, according to an academic researcher who has analyzed these data,3 the school budget lines (primary, etc.) also contain some administrative costs o f an unknown, and varying, amount, making an accurate figure for the administrative overhead difficult to obtain. Given these uncertainties, the appropriate procedure was to adopt the same approach as MoEYS. 16. The basic share has fluctuated, but with n o trend. District-Level Expenditure 17. The PNDC govemment expressed i t s commitment to decentralization in 1983, but the first concrete steps were taken five years later in 1988 when the current structure o f 110 districts was set up (an increase from 65) and the District Assemblies created. Article 35 o f the 1992 constitution embodies the principle o f decentralization, with the parameters elaborated in Chapter 20. Article 252, states that through the District Assemblies Common Fund (DACF) not less than 5 percent o f govemment revenue should be provided to districts. The Common Fund, formally created by the DACF A c t in 1993, i s allocated by a formula 3. Dr. Samer AI-Samarri, Institute o f Development Studies, University o f Sussex (personal communication). 61 Annex B including a needs-based component to ensure poverty targeting. O f the seven indicators used in the district allocation rule, two relate to education. A parliamentary-approved formula also provides guidelines o n the use o f funds, all o f which should be used for investment purposes. 18. n addition to the Common Fund, districts can raise their own funds from taxes, fees, I and levies, as w e l l as receiving ceded revenue from central government by which the income npractice, the Common Fund i s the from certain taxes i s meant to be given to the districts. I main source o f district revenue, accounting for two-thirds o f district income o n average and more in poorer and rural d i s t r i ~ t s . ~ 19. While the constitution allowed for the handing over o f all responsibility for local services to the district this i s not what has happened in practice. District-level offices o f the various ministries and agencies have been created, such as the Ghana Education Service (GES), which are accountable to their central body rather than the District Assemblies. Hence GES i s responsible for the distribution o f recurrent inputs to schools, and teachers’ salaries are paid ~ e n t r a l l ywith , ~ the districts being responsible for infrastructure such as school rehabilitation, although they also provide school hrniture. 20. Expenditure data show that total district expenditure i s just over 4 percent o f central government expenditure (Appiah et al. 2000), but between 20-25 percent o f the government’s development budget (Boko 2002). O f this amount, on average a little over one-third i s spent o n education. Hence the creation o f the D A C F has increased the share o f government spending on education by just over 1percent over and above the increase seen in Figure B.1. The district spending may safely be assumed to be virtually entirely devoted to basic schools, so that the share received by basic education has also been increased by the introduction o f the DACF. However, the magnitudes involved are not suficient to make any substantial difference to the trends noted above. This does not mean that the spending does not matter at school level. Since central resources are largely financing salaries (see next section) additional financing at local level has the potential to create notable variations in school quality. Functional Distribution of Expenditure and Relative Importance of Donor Finance 21. The M o E Y S spreadsheets show the breakdown o f spending for each budget line under the following headings: (i) personal emoluments (PE), ( ii) traveling and transport, (iii) general expenditures, (iv) maintenance and repairs, (v) supplies and stores, (vi) investment costs, and (vii) subventions. Since 1999, headings (ii) to (v) have been replaced by two headings: (i) administrative costs and (ii)service costs, which are listed under items (ii i) and (iv) in Table B.1. 22. From 1989-98, PE accounted for between 70 and 80 percent o f total expenditure. The share jumped after that because o f a change in the system o f financial administration. 4. Commonwealth Local Government Forum (2002), Bok0 (2002), and Appiah et al. (2000). 5. Salaries are paid direct to the banking system by the Controller Accountant General (CAG) using funds received directly from the Ministry o f Finance (Canagarajah and Y e 2001). 62 Annex B Support to tertiary institutions was made entirely through subventions until 1998; so that tertiary spending o n other categories i s not reflected in the table. But from 1999 the subventions stopped, so that wages and salaries previously covered by the subventions n o w appear as PE. To ensure comparability, in the data shown here spending o n tertiary education in the period since 1999 has been treated as though it were a subvention. 23. Inthe basic education sectors that are the focus o f this study, an even greater proportion o f spending has been for PE -staying at 99 percent in primary until 1997 and never falling below 96 percent. I nprimary, virtually nothing was spent o n the three areas that have benefited most fiom Bank support -maintenance, repairs and renewals, supplies and stores, and investment costs. The figures are little different for JSS, though with a small (but still negligible) amount o f spending on maintenance. The PE share i s lower inteacher training and secondary, but inthe former case, the difference i s largely consumed by traveling and transport. A slightly larger amount is spent o n supplies and stores for secondary, but this presumably includes the costs related to boarders. From this discussion it i s clear that the larger percentage o f supplies and stores in total expenditure reflects supplies and stores purchased for administrators i nMoEYS HQand GES, not for schools. Table B.1: Functional classification o f education spending (percent) Primary 1989 100.0 99.3 0.5 0.2 0.0 0.0 0.0 0.0 1990 100.0 99.3 0.5 0.2 0.0 0.1 0.0 0.0 1991 100.0 99.4 0.3 0.2 0.0 0.0 0.0 0.0 1992 100.0 99.6 0.2 0.2 0.0 0.0 0.0 0.0 1994 100.0 98.8 0.5 0.6 0.0 0.1 0.0 0.0 1995 100.0 98.8 0.5 0.6 0.0 0.1 0.0 0.0 1996 100.0 99.3 0.3 0.2 0.0 0.2 0.0 0.0 1997 100.0 98.4 0.3 0.2 0.0 1. I 0.0 0.0 1998 100.0 97.4 0.7 0.5 0.5 0.8 0.0 0.0 1999 100.0 95.9 3.8 0.2 0.0 0.0 0.0 0.0 2000 100.0 97.5 0.0 2.5 0.0 0.0 0.0 0.0 2001 100.0 98.6 0.0 1.4 0.0 0.0 0.0 0.0 JSS 1989 100.0 97.5 1.5 0.5 0.5 0.0 0.0 0.0 1990 100.0 98.0 1.o 0.4 0.6 0.1 0.0 0.0 1991 100.0 98.7 0.6 0.3 0.4 0.1 0.0 0.0 1992 100.0 99.2 0.3 0.2 0.3 0.1 0.0 0.0 ~~ 6. Tertiary accounted for the bulk o f subventions in each year. The other line receiving subvention payments was the general administration. These subventions also stopped in 1999. 63 Annex B 1994 100.0 98.9 0.4 0.3 0.3 0.1 0.0 0.0 1995 100.0 98.9 0.5 0.5 0.1 0.1 0.0 0.0 1996 100.0 99.3 0.3 0.2 0.1 0.1 0.0 0.0 1997 100.0 97.8 0.3 0.2 0.1 1.7 0.0 0.0 1998 100.0 98.1 0.5 0.6 0.3 0.6 0.0 0.0 1999 100.0 97.0 2.8 0.2 0.0 0.0 0.0 0.0 2000 100.0 97.6 0.0 2.4 0.0 0.0 0.0 0.0 2001 100.0 98.7 0.0 1.3 0.0 0.0 0.0 0.0 Secondarv 1989 100.0 90.2 5.3 3.5 1.o 0.0 0.0 0.0 1990 100.0 88.9 4.3 3.2 1.2 2.4 0.0 0.0 1991 100.0 92.1 3.1 2.3 0.8 1.8 0.0 0.0 1992 100.0 94.4 1.8 1.6 0.5 1.7 0.0 0.0 1994 100.0 93.1 2.8 2.4 0.8 0.9 0.0 0.0 1995 100.0 85.3 3.1 8.4 0.9 2.3 0.0 0.0 1996 100.0 84.4 3.8 8.0 0.8 3.0 0.0 0.0 1997 100.0 85.5 4.1 7.8 0.1 2.4 0.0 0.0 1998 100.0 85.6 3.2 6.0 1.5 3.8 0.0 0.0 1999 100.0 84.4 6.7 0.2 8.7 0.0 0.0 0.0 2000 100.0 79.5 0.0 5.3 0.0 0.0 15.2 0.0 2001 100.0 95.8 0.0 2.7 0.5 0.0 1.o 0.0 Teacher training 1989 100.0 83.2 9.9 3.7 1.2 2.0 0.0 0.0 1990 100.0 84.6 8.5 3.5 0.9 2.5 0.0 0.0 1991 100.0 86.1 7.0 3.5 0.9 2.5 0.0 0.0 1992 100.0 93.7 2.6 1.8 0.5 1.5 0.0 0.0 1994 100.0 94.1 2.5 1.7 0.5 1.2 0.0 0.0 1995 100.0 92.5 2.0 4.2 0.2 1. I 0.0 0.0 1996 100.0 93.6 2.3 2.7 0.5 1.o 0.0 0.0 1997 100.0 94.5 2.0 2.3 0.3 0.9 0.0 0.0 1998 100.0 95.3 1.3 1.5 0.6 1.3 0.0 0.0 1999 100.0 97.7 1.4 0.1 0.8 0.0 0.0 0.0 2000 100.0 95.5 0.0 3.9 0.0 0.0 0.7 0.0 2001 100.0 96.8 0.0 2.6 0.2 0.0 0.4 0.0 64 Annex B Total expenditure 1989 100.0 74.8 1.7 1.3 0.5 5.0 0.0 16.8 1990 100.0 71.O 1.4 1.5 0.8 7.6 0.0 17.7 1991 100.0 76.8 1. I 1.o 0.7 3.8 0.0 16.5 1992 100.0 78.9 0.7 1. I 1.o 5.2 0.0 13.1 1994 100.0 75.3 1.I 1.5 2.2 2.8 0.0 17.2 1995 100.0 78.2 1.3 2.4 1.2 3.4 0.0 13.4 1996 100.0 77.4 1.4 1.9 1.o 3.0 0.0 15.2 1997 100.0 77.4 1.5 2.1 1. I 2.7 0.0 15.2 1998 100.0 76.1 2.0 2.7 1.5 3.0 0.0 14.6 1999 100.0 85.5 0.0 8.3 4.4 0.0 1.8 0.0 2000 100.0 88.7 0.0 5.7 3.9 0.0 1.7 0.0 2001 100.0 95.0 0.0 3.7 0.8 0.0 0.5 0.0 Source: MoEYS expenditure spreadsheets The role o f donor finance’ 24. As a first approximation, i t can be said that in basic education government i s financing “nothing but salaries.” There i s a small amount for supplies such as chalk and some infrastructure improvements and school furniture are financed by the districts. Communities also finance some inputs, but these amounts will be small compared to official finding, especially since responsibility for construction was taken away from communities in the 1990s. I t i s clear that most of what has been done in upgrading school infrastructure as well as textbook supply has come from elsewhere. The relative importance o f donor finance illustrates this point. 25. Data are available for Bank disbursements o n an annual basis from Implementation or Project Completion Reports (ICRs and PCRs). These data are reported in Table B.2(a). These totals were converted to cedis using the average exchange rate. The disbursement figures can then be expressed as a percentage o f total expenditure and o f total non-PE expenditure, see Table B.2(b). Figure B.3 shows total Bank disbursements on an annual basis and the ratio o f these to non-PE expenditure. If the Bank funds are entirely outside o f the budget then they have provided up to an additional 40 percent o f resources compared to government’s non- wage spending. But if Bank finds pass through the budget they have accounted for up to 70 percent o f non-wage 7. This section draws on Mettle-Nunoo and Hilditch (2000). 65 Annex B oooco 9 - " " ! Y O 7 7 v?zz O O O F 2 2 .. .. a, I 66 Annex B expenditure. I t i s likely to be the norm that the funds are off-budget. I t i s most likely that some EdSAC resources passed through the budget. The figures are serious under-estimates for the Bank’s contribution to the non-wage component o f basic education spending, since it has been shown that government finances little other salaries in that sub-sector. The value o f the Bank’s resources for non-salary spending in basic education i s many times that o f the government. Figure B.3: Total Bank disbursements 45.0 , 80 40.0 70 35.0 60 30.0 50 a .- -‘0 25.0 .- * 2 a E 40 2 v) 20.0 2 3 30 15.0 20 10.0 5.0 10 0.0 + i + 0 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Education credit disbursements (left axis) +As a percent of non-salary education expenditure (right axis) 26. Data o n bilateral flows are more difficult to come by. The Development Assistance Committee (DAC) online database provides annual commitment data to each recipient by country and sector. The coverage o f these data i s somewhat uneven, so there i s a danger o f under-reporting. In addition, the data refer to commitments rather than disbursements. For what they are worth, the data show that total bilateral support to education has been about $3 18 million, compared t o $257 m i l l i o n from the Bank. There was very little bilateral fimding before 1990, when the first large U S A I D project began, which was in fact budget support, as was EdSAC. In the early 1 9 9 0 Bank ~ ~ and bilateral fknding were o n a par. Inthe later part o f the 1 9 9 0 bilateral ~~ funds have exceeded those from the Bank. Over the period as whole, the Bank has provided about 45 percent o f external support to the education ~ e c t o r . ~ 9. This figure excludes direct support from NGOs (as opposed to NGO-implementation o f officially financed projects). The scale of N G O activities i s too low to substantially affect the figures reported here. 67 Annex B 27. The main donors involved in basic education have been the World Bank, USAID, DFID (each o f which has put a comparable amount into the sector), and the EC through the micro-projects program. These activities are summarized in Table B.3. Table B.3: Main bilateral (and EC) supported activities in basic education ProjecffProgram Period Budget Activities DFID Whole School 1988-2005 UKf50 million Support to 2 pilot schools in Development each district. Construction of 125 classroom blocks. EC Micro-projects 1990-91 ECUI .Imillion District allocated fund for 1991-94 ECU6.O million community activities with allocation guideline of 20% for 1994-96 ECU7.0 million education. In practice about 1996- ? ECU9.0 million 30% used for education (classroom rehabilitation,VIP construction etc.). 1,855 projects financed over period covered by these data. Japan School block 2000-2002 US$0.5 million Approximately 80 classroom construction through blocks. Grant Assistance for Grassroots Projects USAID Primary Education 1990-1995 US$ 35 million U S 3 2 million budget support Project (PREP) plus US$3 million TA Quality Improvements 1997-2004 US$ 53 million US$14 million budget support in Primary Schools U S 3 9 million for (QUIPS) improvements in 330 schools (includes demand-driven infrastructure component for program schools only). 68 Annex C Annex C : School Costs 28. GLSS2 and the GSS/OED collected data on school costs from three separate survey instruments: 1. The school survey contained a question o n school expenses (see Box C.1). 2. The household survey collected data on various categories o f educational expenditure for each child. 3. The price questionnaire collected data on the prices o f school supplies. This annex presents a summary o f these data for 2003. Box C.1: School questionnairequestions on school expenses Please tell me the amount in cedis that students have to pay for the following items. If the amount varies by grade, please tell m e the average for all grades. I I Amount I Comments I 1. Enrolment fee 2. School fee 3 . Sports and culture fee 4. PTA levy 5. Other fees (e.g., District Assembly levy) 6. Value o f materials for nracticals PRESENTATIONOF TABLES 29. Tables C.1 and C.2 show the costs o f fees and materials for primary and junior secondary schools, respectively, from the school survey. These costs are calculated as averages for urban and rural areas, and for three ecological zones separately. The information on the amount in cedis paid by households for pupils’ education i s obtained through interviews with the head-teachers o f each o f the surveyed schools. These figures are not the amounts actually paid by households, but averages estimated by the head-teachers interviewed. Since fees can vary across different grades within the same school, the head- teachers were asked to provide a figure that was the most representative o f what was paid by children o f all grades attending the school. 69 Annex C T a b l e C.1: School costs: Primary schools (cedis) Enrol- School Sports & PTA Other Materials Total Total fees ment fee culture fee levy fees (sum of fee cols. I, 2, 3, (3) (1) (2) (4) (5) (6) (7) and 5) Coast Urban 13341 42813 5019 8374 4629 1319 75494 65802 Rural 1400 6640 6436 2112 3240 2120 21948 17716 Forest Urban 6801 33820 4998 9050 8633 3748 67051 54252 Rural 1835 15608 3912 3519 8589 2330 35793 29944 Savannah Urban 1615 16308 3258 5538 4850 0 31569 26031 Rural 343 5829 3834 2040 1100 3 13149 11106 Average 5884 26476 4630 6393 6558 2242 52182 43548 Table C.2: School costs: Junior secondary schools (cedis) Enrol- School Sports & PTA Other Materials Total Total fees ment fee culture fee levy fees (sum of fee cols. 1, 2, 3, (3) (1) (2) (4) (5) (6) (7) and 5) Coast Urban 10071 39869 7316 9136 6971 3107 76470 64227 Rural 0 0 7574 2553 5095 6526 21747 12669 Forest Urban 7361 29458 5244 8760 10638 6369 67830 52701 Rural 850 7150 4378 3917 4247 12608 33150 16625 Savannah Urban 1588 12882 3994 4388 8000 0 30853 26464 Rural 0 0 4126 2079 527 3211 9942 4653 Average 5358 22500 5572 6741 7239 6302 53712 40669 30. Tables C.3 and C.4 show household educational expenses for children attending primary and junior secondary schools. The information on these expenses was obtained from the household questionnaire. Parents were asked how much they had spent during the 12 months preceding the interview on the items presented in the table. The expenses were reported in the questionnaire for each child separately, thus allowing the distinction between primary and JSS (and higher levels o f education). The figures are average expenses per pupil calculated across ecological zones and for urban and rural areas separately. 3 1. Comparing table C.1 with C.3, and C.2 with C.4, shows some differences between reported school and household costs. In the case o f primary, the total fees o f cedis 43,500 i s close to the tuition and registration fees households say they pay (cedis 46,200). In the case of JSS the households claim to pay rather more than the schools report. This may reflect the differing composition o f attendance, but m a y also reflect under-reporting o f fees, especially unofficial ones, by schools. Government policy i s that there are no fees, such as enrolment fees, in public basic schools. Hence, in the government’s view the non-zero values shown in the public school section o f tables C.7 and C.8 reflect either respondent or data entry error. 70 Annex C Table C.3: Households’ costs: Primary schools (cedis) PTA Uniforms Books & Transport Cafeteria Tuition & Other Total levy & supplies & registration clothes lodging fees (1) (2) (3) (4) (5) (6) (7) (8) Coast Urban 8074 42946 53320 59641 244890 89565 45653 909793 Rural 2594 23308 15364 10573 121094 23099 14549 314399 Forest Urban 10195 39693 42839 33733 237249 66674 45695 591333 Rural 2833 30845 18640 12090 146202 29468 18337 305793 Savannah Urban 9658 26216 9868 180 64216 24641 22018 207774 Rural 2641 22122 6891 618 44066 7519 6786 112898 Average 6008 33079 28729 24311 164012 46272 28400 462024 Table C.4: Households’ costs: Junior secondary schools (cedis) PTA Uniforms Books & Transport Cafeteria Tuition & Other Total levy & supplies & registration clothes lodging fees (1) (2) (3) (4) (5) (6) (7) (8) Coast Urban 10378 64920 88480 116227 246908 75239 54513 1418049 Rural 4386 47500 88273 101409 157500 38068 40955 785227 Forest Urban 24496 53708 79375 103843 339843 74915 87407 891814 Rural 4774 31870 47496 44521 223432 40449 30615 531570 Savannah Urban 4938 19438 29250 0 82813 41813 25188 245475 Rural 4557 43273 25309 2045 64977 18109 21259 224961 Average 12055 50033 68305 80149 235719 58378 53407 876883 32. Table C.5 shows the costs o f selected school items. These costs were obtained through market surveys carried out in each o f the selected clusters. In every locality, three different shops selling stationery and clothes were interviewed, and the average prices o f each item per locality were calculated. The figures shown are again averages across ecological region and rurayurban areas. 71 Annex C T a b l e C.5: Prices of education items Exercise book Pencil Eraser School School uniform uniform (girls) (1) fbOYS) (51 (2) (3) (4) Coast Urban 1111 270 253 41674 41493 Rural 1033 280 217 26875 27167 Forest Urban 977 232 239 22077 22205 Rural 1072 255 228 22800 21909 Savannah Urban 1050 267 242 24250 23417 Rural 1153 331 220 24909 24091 Average 1069 267 236 29848 29429 33. Table C.6 shows the number o f school o f each type (private and public) and level (primary and JSS) that were covered by the survey. The table displays their geographical distribution. Table C.6: Distribution o f schools by level, type, and location Primary schools 1 Junior Secondary schools Public Private Total Public Private Total Coast Urban 62 29 91 57 13 70 Rural 22 3 25 18 1 19 Forest Urban 108 35 143 86 18 104 Rural 75 22 97 54 6 60 Savannah Urban 21 5 26 14 3 17 Rural 32 3 35 19 0 19 Average 320 97 417 248 41 289 34. Table C.7 and C.8 are disaggregations o f Tables C.l and C.2 for primary and private schools. Similarly, tables C.9 and C.10 are disaggregations o f tables C.3 and C.4. 72 Annex C Table C.7: School costs in public and private schools: Primary (cedis) Enrol- School Sports & PTA Other Materials Total Total fees ment fee culture levy fees (sum of fee fee cols. 1, 2, (1) (2) (3) (4) (5) (6) (7) 3, and 5) Public schools Coast Urban 262 246 6504 10459 6675 1869 26016 13688 Rural 0 500 7223 2082 3682 2045 15532 11405 Forest Urban 792 3190 5856 9178 4055 2426 25495 13892 Rural 107 2547 4727 4124 3388 2887 17779 10768 Savannah Urban 0 952 3581 6143 52 0 10729 4586 Rural 94 0 4100 1763 1203 3 7163 5397 Average 353 1823 5483 6802 3824 1999 20282 11482 Private schools Coast Urban 39933 129367 1998 4467 468 200 176433 171767 Rural 11667 51667 667 2333 0 2667 69000 64000 Forest Urban 25343 136909 2351 8657 22761 7829 203850 187364 Rural 7727 60136 1136 1455 26318 432 97205 95318 Savannah Urban 8400 80800 1900 3000 25000 0 I19100 116100 Rural 3000 68000 1000 5000 0 0 77000 72000 Average 23888 109784 1855 5163 15456 3036 159181 150982 Table C.8: School costs in public and private schools: JSS (cedis) Enrol- School Sports 8 PTA Other Materials Total Total fees ment fee culture levy fees (sum of fee fee cols. 1, 2, (1) (2) (3) (4) (5) (6) (7) 3, and 5) Public schools Coast Urban 0 687 5283 10736 8791 3355 28852 14761 Rural 0 0 7550 2583 5378 6889 22400 12928 Forest Urban 4541 5905 5920 9581 5031 7057 38035 21397 Rural 37 2194 4559 4278 4459 6972 22500 11250 Savannah Urban 0 0 4171 4829 786 0 9786 4957 Rural 0 0 4126 2079 526 3211 9942 4653 Average 1596 2700 5360 7313 5182 5500 27650 14837 Private schools ~~ ~ Coast Urban 47000 212867 2500 3267 300 2200 268133 262667 Rural 0 0 8000 2000 0 0 10000 8000 Forest Urban 20833 202544 2794 7833 4094 3083 241183 230267 Rural 8333 51750 2750 667 2333 0 65833 65167 Savannah Urban 9000 122333 3167 3667 1667 0 139833 136167 Rural ... ... ... ... ... ... ... ... Average 26907 174798 2833 4814 2260 2058 213670 206798 73 Annex C Table (2.9: Households’ costs in public and private schools: Primary (cedis) PTA Uniforms Books & Transport Cafeteria Tuition & Other Total levy & supplies & registration clothes lodging fees (1) (2) (3) (4) (5) (6) (7) (8) Public schools Coast Urban 4401 20664 23627 13503 132379 13222 26460 681220 Rural 2260 21122 13450 7481 109012 13055 10302 283901 Forest Urban 6557 33019 26926 12573 156941 16479 21793 356998 Rural 2344 27054 17656 4079 139512 13095 13658 261147 Savannah Urban 4914 16978 8175 0 46860 9054 16032 160170 Rural 2425 21840 6733 656 34627 3707 7203 100814 Average 3638 24893 16987 6419 111212 11617 15588 302521 Private schools Coast Urban 11810 65612 83526 106575 359342 167224 65177 1142306 Rural 6250 47167 36250 44333 253000 132750 60917 647333 Forest Urban 16563 51371 70688 70763 377787 154515 87522 1001419 Rural 6023 55554 25052 64304 189804 136189 48830 596791 Savannah Urban 34167 73944 18611 1111 153889 105172 52944 453728 Rural 6167 26700 9467 0 197600 69533 0 309467 Average 13205 57936 64384 78640 324339 151501 67305 946354 Table C.10: Households’ costs in public and private schools: JSS (cedis) PTA Uniforms Books & Transport Cafeteria Tuition & Other Total levy & supplies & registration clothes lodging fees (1) (2) (3) (4) (5) (6) (7) (8) Public schools Coast Urban 4173 29407 47619 67272 187247 22805 39420 1452985 Rural 4325 50000 92950 93550 135750 32125 39650 754450 Forest Urban 19554 47873 66889 88542 329446 33936 65229 724693 Rural 3562 29315 41557 26215 221315 231 11 33122 485705 Savannah Urban 3933 20733 28200 0 52333 14467 23867 188373 Rural 4663 39628 23828 2093 66488 8088 21753 213030 Average 8269 36730 50217 53625 206044 24018 41845 770316 Private schools Coast Urban 23605 140618 175579 220579 374079 187008 86684 1343579 Rural 5000 22500 41500 180000 375000 97500 54000 1093000 Forest Urban 40904 73080 120826 154640 374360 210965 161040 1446655 Rural 14625 52625 95750 193250 240625 181325 10250 904225 Savannah Urban 20000 0 45000 0 540000 452000 45000 1102000 Rural 0 200000 89000 0 0 449000 0 738000 Average 27555 104487 142342 188720 357187 199027 100733 1313096 74 Annex C DISCUSSION Regional Disparities 35. The regional disparities in the above data are very marked. Education costs are usually highest in the urban coastal areas and lowest in the Figure C.l Fees collected at public schools rural savannah. The second retained by school (including PTA levy) most expensive area i s urban forest. N e x t i s urban savannah. That is, costs are higher in all 30 three urban areas than all three .- In 'CI rural areas. This result partly 8 20 0 reflects the concentration o f 0 private schools in urban areas. 10 Looking at public schools only, the same pattern holds for JSS, 0 but not primary, for which Primary JSS costs are higher in rural forest Coast (Urban) ia Coast (Rural) Forest (Urban) than either savannah region and Forest (Rural) 0 Savannah (Urban) E Savannah (Rural) for JSS quite above those o f rural coast (Tables C.7-C. lo). 36. For primary schools in aggregate, total fees in the coastal region are six times those in the savannah (Table C.1). The difference i s even greater for JSS (Table C.2). The household data show an even greater divergence, with costs being eight times as much for primary and nearly seven for JSS (Tables C.3. and C.4). 37. These disparities are not much explained by differences in the cost o f school supplies, which do not vary too greatly other than the greater cost o f uniforms in urban coastal (Table C.5). Levels o f Schooling 38. As is to be expected, JSS costs more than primary, though these differences are not particularly fee related (Figure C. 1). The aggregate data fiom the school survey show similar levels o f both fees and total costs (Tables C.l and C.2). But the household data show that parents spend nearly twice as much to send a child to JSS as they do to primary. Expenditures are greater for JSS for all the expenses shown, but the difference i s greatest for transport and books and supplies. Comparing public schools only, JSS appears more expensive according to both school-level data and household data, with the gap being large for the latter (Tables C.7- c.10). Public versusprivate 39. Private schooling o f course costs more than public (Tables C.7-C.10), with the differential appearing larger from the school data than the household data. This i s since items 75 Annex C not included in the school data -uniforms, transport, and cafeteridlodging -have a much lower differential. Funds available to schools 40. The sports and culture levy Table C.ll: Zero and maximum responses i s the only official school fee set by in primary school cost data government. In addition, the PTA No. of schools Maximum value levy i s intended to benefit the for which zero (cedis) school. Figure C.1 shows the 1. Enrolment fee 233 (253) 40,000 average o f fees other than the 2. School fee 223 (254) 190,000 sports and culture levy, these being Y 3. and culturefee 9 (317) 40,000 4. PTA levy 57 (312) 150,000 amounts likely to be retained by 5. Other fees 79 (306) 120,000 the From this figure, the (e.g,, District Assembly levy) differentials seem quite large. In fact, they are even larger. For example, o f 317 public primary schools 233 reported charging n o enrolment fee compared to a maximum charged in one school o f 40,000 cedis (Table C.11). There are 57 schools not even charging a PTA levy, compared to a maximum o f 150,000. The table shows in brackets the number o f responses to each question. I t cannot be assumed that non-responses are zero, since zero was an accepted response. I t i s just as likely that respondents were reluctant to provide the information since such fees are not meant to be charged. This would mean that the school costs under-estimate charges. The household data show higher fees than the school data for JSS, though they are similar for primary. 76 Annex D Annex D: School-Level Changes in Inputs, Management, and Methods INTRODUCTION 1. This annex reports the data from the school and teacher surveys o n school quality. Tabulations are presented o n variations o f school quality by zone (coastal, forest, and savannah), ruralhrban, and the economic well-being o f the community in which the school i s situated. For some variables regression results are presented to explore the determinants o f different aspects o f school quality. Insofar as the data Table D.l: Sample sizes permit, comparisons are made with 1988. These comparisons are reported in two ways: (1) comparisons o f 1988 the whole sample o f 519 schools in 1988 with the 706 Primary schools 286 surveyed in 2003, and (2) comparison based only o n the Middle/JSS 233 panel o f 196 schools that could be matched between the Total 519 two survey rounds.' A teacher survey was not carried out 2003 in 1988, although limited information was collected from Primary schools 417 the teacher roster in the school questionnaire. JSS 289 Total 706 2. The next section discusses the variables to be used Teachers 3,129 in more detail, with subsequent sections discussing in turn monetary (capital and recurrent) inputs, teacher quality, Memo item: matched schools methods, morale, and school management. Primary schools 128 JSS 68 Total 196 FOUR DIMENSIONS OF SCHOOL QUALITY: VARIABLE SELECTION 3. The school survey from 1988 and the school and teacher surveys provide a number o f variables that can be used to measure school-level inputs to the education process. Four dimensions o f school-level inputs can be identified: 0 Physical 0 Material 0 Teachers 0 School management Table D.2 identifies the variables that can be used t o measure the quantity and, where possible, quality, o f these various inputs. Few data o n the last dimension, school management, were collected in 1988, so that comparisons across time are largely restricted to the first three dimensions. The subsequent sections o f this annex present the main findings 1. Since the surveys took place in the same clusters in the two rounds, all schools surveyed in the f i r s t round should also have been surveyed i n the second round, unless they closed. However, school name information was not entered with the data in the f i r s t round and could not be recovered for a l l schools. Matching was attempted during field work and checked against location, year o f establishment, and whether the school was public or private. 77 Annex D with respect t o each dimension. This section provides an overview o f the variables to be used. Table D.2: Survey-based measures o f school quality Quantitv Qualitv - - Physical Classrooms Adequate number of classrooms % of classrooms that can be used when raining Classes held in shared classrooms* Height of internal walls* Noise disruption* Chalkboard % of classrooms with chalkboard Board quality Water % of schools with own water Type of WaterIStorage supply Library % of schools with library Material Chalk Availability English textbooks English Textbooks - Pupil Ratio Textbook usage* Mathematics textbooks Mathematics Textbooks - Pupil Textbook usage* Ratio Desks Writing Places - Pupil Ratio Chairs Seating places-pupil ratio* Teachers and teaching methods Teachers Adequate number of teachers Teacher morale (subjective)* % trained teachers Absenteeism Teacher test scores Head’s assessment* Teaching methods* Frequency of homework* School management Community involvement Existence SMC and PTA* Active SMC and PTA* SPAM* Role of SPAM* Circuit supervisor Frequency of visits Activities of circuit supervisor* Head teacher Activities of head teacher* Note: not collected in 1988. 4. Most o f the variables are self-explanatory. Explanation i s provided here for those that are not (Annex F provides a variable list with definitions): (1) Adequate classrooms and adequate number o f teachers are both based o n dividing the actual number available by the required number. The required number i s the number o f classes taught in the school, taking into account multiple streams and shifts. If a primary school teaches grades 1 to 6 and has two streams without shifts it needs 12 classrooms and 12 teachers.* But if that school operated a shift system it would need only 6 classrooms but still require 12 teachers. Classrooms in such 2. Teachers teach for just one o f the two shifts. The head teacher i s meant to be present for both shifts. 78 Annex D poor state o f repair that they cannot be used are excluded from the number available (the numerator). The 2003 data contain additional variables o n classroom quality. Detached head teachers are excluded from the numerator. The percentage o f classrooms that can be used when i t i s raining i s 100 less those that cannot be used when it i s raining plus those that cannot be used at all divided by total number o f classrooms (including those that cannot be used at all). Board quality is a subjective assessment. In 1988 this assessment was made by the respondent (usually the head) for the whole school. The same question was asked in 2003 and i s used for the purposes o f comparisons. However, in 2003 teacher- level data o n board quality (and more specific questions o n that quality) are also available and used for analysis specific to 2003. Textbook availability. The number o f books at each grade i s summed across grades and divided by total enrolments, which i s equivalent to an enrolment- weighted average for the scho01.~ Writing places. For 2003 desks per pupil i s calculated in an analogous way to textbook availability, but using adjusted enrolments where the adjustment takes account o f a split shift (i.e., the same desk can be used by different pupils in morning and afternoon). However, for 1988 there is a categorical variable o n whether there are enough, some, or no desks. The 2003 data are categorized for comparability (see Annex F for cut-offs). Teacher morale i s a subjective measure based o n two questions, whether the respondent enjoys being a teacher, and if he or she plans to remain a teacher for their whole career. The head’s assessment i s a categorical classification o f all teachers as very good, good, average, poor, or very poor. None o f these data were collected in 1988. Teacher morale i s analyzed in para. 84 ff. Teaching methods were assessed through three questions in the teacher questionnaire designed to assess the extent t o which the teacher claims to use “improved methods,” including a check o n their knowledge o f these methods (more detail i s provided below). PHYSICAL AND MATERIAL SCHOOL-LEVEL INPUTS The M a i n Message: School Quality Has Improved 5. The main message from the school survey i s the overwhelming improvement in school quality. For example: 3. Ifi t were the case that some grades had books while others did not then grade-specific textbook indicators would be required. However, analysis o f the data shows t h i s not to be the case, so that the school-wide average will suffice. 79 Annex D 0 In 1988 less than half o f schools could use all their classrooms when i t was raining, but in 2003 over two-thirds can do so. 0 Today 94 percent o f schools have a blackboard in every classroom compared to 78 percent 15 years ago 0 Fifteen years ago over two-thirds o f primary schools reported occasional shortages o f chalk, only one in twenty do so today, with 86 percent saying there i s always enough 0 The percentage o f primary schools having at least one English textbook per pupil has r i s e n from 21 percent in 1988 to 72 percent today and for math books in JSS these figures are 13 and 71 percent, respectively. 6. Despite the greatly improved school quality, variation remains across the country, with some “biases” in the allocation o f school resources. Analysis reported below shows that the strongest bias comes from the ability o f better-off communities to better support schools in their locality. I t i s also shown that the Bank’s Primary School Development Project made a significant contribution to aspects o f school quality. Material Inputs 7. The material inputs for which data can be compared between 1988 and 2003 are the availability o f chalk, math and English books, and desks. For each o f these four variables there has been a strongly significant (all significant at 1percent level) improvement in the level o f inputs at both primary and JSS level (see Table D.3). This statement i s also true for the panel o f 196 schools. Table D.3: Significant changes in the availability o f material inputs Primary Middle/JSS Total Chalk *** *** *** English books *** *** *** Math books *** *** *** Desks *** *** *** Note: *** significant at I % ,** significant at 5%, significant at 10%. Significance is based on chi-squared statistic based on cross-tabulationof categorical version of variable against year (1988 and 2003). See Annex E for cross-tabulations and Annex F for variable definitions. 8. The four variables were combined into a simple index o f material inputs. The resulting figures are shown in Figure D.1.4 4. The index was constructed by scaling each o f the four variables over the range 0-1 and then taking a simple average. The resulting index inprinciple ranges from zero to one, though no school has the minimum score o f zero. A principal components analysis o f the four variables was also conducted. Each o f the four variables entered the first component, which accounted for about half o f the variation, with approximately equal weights. 80 Annex D Figure D.1: Schools in nearly all areas have more material inputs than before: cluster-level material inputs to school quality (a) Material Primary (b) Material Middle/JSS 1 1.o 1 1.0 0.8 0.8 X X a, a, c U S ._ .- U 0.6 5 0.6 3 Q n .- c .- c c c E 0.4 3 2 0.2 0.0 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 - Cluster Cluster _ _ _ _ -1988 -2003 -2003 (public only) _ - _ - - 1988 2003 -2003 (public only) Source: GLSS2 and GSS/OED school survey 9. The figures show the cluster level average o f the material input index for 1988 and 2003, calculated separately for primary and middle/JSS. In each graph, the clusters have been ranked according to the value o f the index in 1988, so that the clusters with the schools with the fewest material inputs appear to the left o f the scale. Where the line for 2003 lies above that for 1988 there has been an increase in the material input index for that cluster. T w o points jump out from these graphs: 0 There has been a substantial increase in the level o f material inputs across the country, especially in primary schools. In only two clusters (which had the maximum value o f 1 in 1988) has the level o f material inputs declined at the primary level (Table D.4). For middle/JSS there have been, mostly small, declines in nine o f the 76 clu~ters.~ 5. There were 85 clusters in the 1988 survey. One o f these was no longer inhabited so that the 2003 survey covered 84 clusters. A fiuther six clusters with l o w populations in 1988 were also skipped in 2003 as a result o f the self-weighting sample design. School data were not available for two clusters from 1988. Hence comparisons across time may be made using data from 76 clusters. 81 Annex D 0 The improvement has been greatest the lower the initial level o f the index, meaning that the clusters in which schools that were the most deprived have seen the largest improvements in material inputs (Table D.5).6 Table D.4: Summary o f observations in Figures D.1 and D.2 Primary Middle/JSS Number of clusters experiencing deterioration Material 2 9 Physical 22 31 Number of clusters in which public schools have lower quality than private schools Material 19 13 Physical 29 17 Memo: Number of clusters with private schools in 2003 41 24 10. The panel data for chalk availability Table D.5: Correlation coefficient between in primary schools provides a clear example cluster-level change in index and of h o w the improvement o f school quality the initial (1988) value has been concentrated in the most disadvantaged schools. In the general sample Material Physical i t has been seen that 86 percent o f schools Primary -0.88 -0.65 today say there i s enough chalk compared to Middle/JSS -0.71 -0.65 67 percent 15 years ago. In Table D.6 most the observations l i e o n the upward sloping diagonal from the bottom left. All schools that said there was never enough chalk in 1988 today have enough. O f the 102 suffering occasional shortages 15 years ago, 88 (86 percent) n o w always have sufficient. And a l l but 3 of those which had sufficient in 1988 s t i l l do so today. Simply put, the 28 schools already having sufficient chalk had n o room for improvement. But the 16 who never had enough could at worst stay the same -though in the event all those 16 n o w report having sufficient supply. ~~ 6. This result may be partly explained from measurement error, though t h i s would require substantial systematic under-reporting o f basic school quality variables. 82 Annex D 11. The share o f private Table D.6: Change in chalk availability against schools in the sample increased initial availability for primary school panel from 5 to 20 percent between Chalk availability in 1988 1988 and 2003. Hence i t might be Change in argued that the observed increase rating Never Occasional Always in school quality simply results (2003-1988) enough (=I) shortages (=2) enough (=3) Total from the better quality o f private -2 0 0 2 2 schools. This i s n o t so. Figure D.1 -1 0 7 1 8 also shows the material input 0 0 7 28 35 index for 2003 calculated for 1 0 88 0 88 public schools alone. In general 2 16 0 0 16 this line is not far removed from Total 16 102 31 149 the overall cluster average. Indeed it i s below it, indicating That public schools have a superior level o f material inputs than do private ones in 22 o f the 41 clusters that have public schools (Table D.4). When the changes in the index and i t s components are calculated for public schools only these changes all remain significant at the 1 percent level (Annex E, Tables E.6-E10). 12. Private schools do have a higher level o f material inputs in some respects. Although there i s no overall significant difference in the material input index for public and private schools in 2003 this result conceals that private schools do significantly better o n two o f the four components of the index (English books and desks), but significantly worse on two (chalk and math books) (Annex E, Tables E.15-E. 19). Private schools on average have higher levels o f some inputs than do public ones, but this i s not the main reason for the observed improvement in school quality between 1988 and 2003. Physical Inputs 13. The index for physical inputs comprises the adequacy o f the number o f classrooms, the proportion that can be used when raining, the proportion with a blackboard and the quality o f those boards, the presence o f a library and o w n water supply. Two o f these have not improved (sufficient number o f classrooms and library) for either type o f school, one (library) has not for primary schools, and another (classrooms that cannot be used when raining) for middle/JSS (Table D.7). The lack o f change with respect to there being sufficient classrooms shows that classroom building has kept pace with growing student numbers. So the number o f classrooms has increased, but been matched by more students. For many schools there has been no shortage o f physical facilities (although their quality i s a different matter), so that no improvement in this measure is expected or required. Overall, there has been a significant increase in the index o f physical inputs. 83 Annex D Table D.7: Significant changes in physical inputs to school quality Primary Middle/JSS Total Adequate classrooms Classrooms which cannot be used when raining *** *** Percentage of classrooms with a chalkboard *** *** *** Chalkboard quality *** *** *** Own water supply *** *** *** Libraw ** ** ** significant at 5%, * significant at IO%, and - indicates no significant change. Note: *** significant at I%, Significance is based on chi-squared statistic based on cross-tabulationof categorical version of variable against year (1988 and 2003). See Appendix E for cross-tabulations and appendix F for variable definitions. Figure D.2: The quality o f school infrastructure has improved in most areas: cluster-level physical inputs to school quality (a) Physical Primary (b) Physical Middle/JSS 1.o 0.8 x Q1 U .- C 3 0.6 - P .- C 8 ‘9 0.4 c a 0.2 0.0 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 Cluster _____ Cluster - - ,1988 -2003 -2003 (public only) 1988 -2003 -2003 (public only) 1 J Source: GLSS2 and GSS/OED school survey 14. Figure D.2 shows the change in physical inputs in the same w a y as Figure D.l showed material inputs.’ There are n o w many more clusters, though s t i l l w e l l less than half (Table D.4), which have not experienced an improvement. Once again, although private schools do perform better in some respects, their increase does not account for the improvement in school quality that has taken place. 7. The index was constructed inthe same way as that for material inputs. Int h i s case principal components analysis suggested that water and library should enter with a slightly lower weight than the other four variables. 84 Annex D 15. n2003, private schools had superior inputs with respect to the percentage o f I classrooms that could be used when raining and having their own water supply. They also had slightly better average quality chalkboards, although the difference i s not quite statistically significant @rob value=0.1 1). There i s no difference with respect to having sufficient classrooms, chalkboards, or a library. Allocation of School Inputs 16. Tables D.8 and D.9 summarize some distortions in the allocation process. Looking first at income (Table D.S), a difference can be seen with respect to material and physical inputs. In the former case, allocation biases that existed in 1988 have been reduced or eliminated except for the number o f desks per student. However, in the case o f physical inputs the bias has continued for three o f the six measures, and emerged in one case where i t was not previously present. Only in one case has it been reduced and in one eliminated. Based o n the chi-squared statistic, the bias in allocation has increased for physical inputs, while it has declined for material ones. While these biases should be a source o f concern t o policymakers, they must be seen in the context o f the overall rise in the level o f inputs that has taken place across virtually the whole country. Table D.8: Allocation o f education resources b y expenditure quintile (public schools only) 1988 2003 Comment Physical Adequate number of No bias Significant (5%)’ bias Allocation bias emerged Classrooms against poorer clusters Classrooms cannot Significant (1%) bias Significant (10%) bias Reduced bias against be used when raining against poorer clusters against poorer clusters poorer clusters Chalkboard Significant (1%) bias No bias Allocation bias eliminated against poorer clusters Board quality Significant (10%) bias Significant (10%) bias Continued bias against against poorer clusters against poorer clusters poorer clusters Library Significant (10%) bias Significant (5%) bias Continued bias against against poorer clusters against poorer clusters poorer clusters Water Significant (1%) bias Significant (1%) bias Continued bias against against poorer clusters against poorer clusters poorer clusters Material Chalk Significant (1%) bias Significant bias (10%) in Allocation bias eliminated against poorer clusters favor of poorer clusters English books Significant (5%) bias in Significant (10%) bias in (Progressive) allocation favor of poorer clusters favor of poorer clusters bias reduced Math books No bias No bias (Progressive& allocation bias reduced Desks Significant (1%) bias Significant (1%) bias Continued bias against against poorer clusters against poorer clusters poorer clusters Adequate number of Significant (5%) bias Significant (5%) bias Continued bias against Teachers against poorer clusters against poorer clusters poorer clusters Notes: I/ The percentage in each cell, where shown, is the level of significance of the chi-squared statistic in the cross-tabulationagainst income quintile. 2/ Although there is no significant relationship in either period the prob value is 0.102 in 1988, with the allocation favoring poorer clusters, and 0.654 in 2003. 85 Annex D 17. Table D.9 incorporates table D.8 and presents biases in allocation by zone that may underlie these income biases (since non-coastal zones, particular savannah, are poorer, as are rural ones). Considering first biases against rural areas, these are much less today than they were 15 years ago. The only significant biases against rural areas are in water supply, which i s clearly linked to their location, and their relatively lower likelihood o f having a library. Previous biases, notably in school infrastructure, have been rectified. 18. However, in the case o f regional allocation the opposite appears to be the case, biases having appeared in the allocation o f both physical and material resources. In general, where such bias exists, then the coastal region i s most well provided for. The exception i s for math books, for which the forest region has the greatest availability. In all cases, the savannah zone i s the least w e l l resourced, except for the mild exception o f the presence o f a library. But in general, the forest and coastal regions are relatively close in resource availability, with schools in the savannah region trailing behind. Table D.9: Biases in the allocation o f educational resources, 1988 and 2003 (bias i s against poorer households and rural areas unless otherwise noted) Income - Region Ruralhrban 1988 2003 1988 2003 1988 2003 Physical Adequate number of ** *** *** Classrooms Classrooms cannot *** *** be used when raining Chalkboard *** ** *** Board quality * * *** * Library * ** *** *** * Water *** *** *** *** *** *** Material Chalk *** *I *** **2 English books ** * ***2 Math books * *2 Desks *** *** *** *** * Adequate number of ** ** *** *** *** Teachers -~ Explanatory note: The more stars the more significant the bias in resource allocation. A - indicates no significant bias. The bias is against poorer, non-coastal and rural communities unless otherwise stated. Significance is based on the chi-squared statistic calculated from the bivariate cross-tabulation of the school measure against each characteristic (income quintile, region, ruralhrban) in turn. Notes: 11in favor of poorer communities; 21 in favor of rural communities. More on variations in monetary school inputs 19. According to broad aggregate measures, biases in the allocation o f resources have lessened. Yet Figures D.1 and D.2 show considerable variation in the community-level averages. This suggests that the differences in the levels o f school inputs are n o t w e l l explained by broad aggregate categories such as ruralhrban or zone. This view i s supported by an analysis of variance, which finds that the variation within zones i s significantly greater 86 Annex D than that between them for both indices, and all their component parts with just two exceptions (English books and classrooms that can be used when it i s raining). Two possible explanations are pursued here for the large within-area variations: the role o f projects and community support. 20. There have been three major project initiatives providing direct support to primary schools for infrastructure and supplies: the World Bank’s Primary School Development project (PSD), USAID’s QUIPS, and the Whole School Development (WSD) program supported by DFID. Data on support fi-om WSD and QUIPS, which are ongoing, were collected in the school questionnaire. PSD beneficiary schools within the GSS/OED sample were identified fi-oma l i s t o f all beneficiary schools. There are 19 o f the latter in the panel o f 196 schools, 14 W S D schools andjust 5 who have benefited from QUIPS. Table D.10: Impact of projects on change in school inputs (panel data) PSD QUIPS WSD Material inmts Chalk English books Math books Desks +ve) ,, ( -ve (*) Index Physical inputs Sufficient classrooms Use classrooms when raining ,, +ve)( -ve (*) Chalkboard Chalkboard quality ,, +ve)( -ve (**) Library Water Index +ve(**) -ve (**) Note: *** significant at 1% level, ** 5% level and * 10 % level. 2 1. Bivariate analysis o f the panel data shows that PSD (Table D.10) i s associated with significantly larger improvements in the proportion o f classrooms that can be used when it is raining, availability o f desks, and quality o f blackboards - as well as with the physical input index. Since PSD’s main activity was the financing o f new classroom blocks (which don’t leak and have good blackboards), usually to replace old ones, and to provide desks for these blocks, these results make sense. Many classrooms in Ghana are in pavilions, that is, a r o o f o n supports but with low or no wall. These cannot be used during heavy rain. In the 1980s, these were often community-made structures from tree trunks/large branches and thatch. PSD replaced these with concrete and corrugated iron pavilions. The construction o f walls (cladding) was l e f t to the community as their contribution, though this was not always done. 22. By contrast, QUIPS appears to have had a perverse impact on the change in some school inputs, though the sample size i s very small and endogeneity i s the likely cause o f . 87 Annex D these results.* N o impact appears from WSD. These findings are not surprising in light o f the fact that neither project has focused on hardware and are both o f more recent vintage and s t i l l ongoing. 23. A second approach to the analysis i s to estimate regressions for the school input variables using the 2003 data only. In addition to the project dummy variables, measures are also entered o f community well-being (per capita expenditure), the level o f fees, if support has been provided by the PTA or SMC, and the value o f the P T A levy. Dummies are entered for zone, rurallurban, and private schools. The results are summarized i n Table D.11. 24. The main points to emerge are as follows: 0 The regression models do not explain the distribution o f textbooks; none o f the variables are significant in the model for math books and only the primary dummy in that for Englishbooks. On the other hand, financial resources do matter for desks and, t o a lesser extent, chalk availability. 0 Financial variables matter for physical inputs, being highly significant for the index as a whole and some o f them at least being significant for classrooms that can be used when it is raining, chalkboard quality, library, and water supply. The PTA levy i s never significant, but is correlated with other financial variables. Moreover, the school-level data reflect the levy set, not the amount actually collected in additional contributions. 0 The positive impact o f the PSD project on classrooms that can be used when it i s raining i s supported by this analysis, as i s i t s impact on the number o f classrooms. 25. These findings indicate that the level o f inputs to schools reflects the economic well- being o f the surrounding community -directly through the level o f fees they can afford, the level o f the PTA levy, and the likelihood o f help from the PTA or SMC. On top o f these, the community’s level o f expenditure matters, presumably picking up other channels through which support i s provided. However, the wealth o f the community does not matter for textbooks, and matters less for chalk, since these are things provided centrally through GES. I t does matter for desks, which are increasingly likely to be provided by the district, whose resource availability depends on that o f the population’s income. 8. The levels analysis suffers fiom the potential bias that beneficiary schools are worse o f f than the average when selected for program participation, which will be picked up as a negative program effect, especially in schools new to the program. This problem is not so evident for PSD, which was completed some time ago, but may explain the chalkboard result. It is, however, a plausible explanation for QUIPS. Ideally, a program selection equation could be estimated and a two-step estimation procedure applied. This only makes sense for PSD for which the 1988 data are not too far removed ftom the date o f selection and sample size i s reasonable (the panel data have to be used since it i s only inthose data that the schools can be identified inthe earlier data set). A selection equation was estimated for PSD. The main determinants o f inclusion are: (1) being in a poor community, (2) shortage o f desks, (3) urban. Lack of chalkboards had the expected sign but was not sigruficant. 9. The PTA levy recorded in the questionnaire i s that set by the P T A as a minimum, with better o f f parents expected to pay more. More i s said o n the P T A contribution below. 88 Annex D * ;* ; * * : + . ** . . . . . . . . . . . . * 2 :* : : * * . . . : : * * : : E : . . . . . . . . . . . . . . . . f ; : : : : E : . . . . * : : h a, . . . . . . . . . . . . * * . . . . F . . * Y . . * h a, : * : : : . . . . . . . . $ 2 * * i . . . . . . .. .. . * . * . ' . . . . * * . . . . . . . . : * * E : . . . . . . . . * E : : : . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . * . . h a, * : : : * E ; : . ? . . . . * . Y * 89 Annex D SCHOOL MANAGEMENT Supervision by H e a d Teacher and Circuit Supervisor 26. The focus o n software means an increased focus o n issues o f school management. Data o n two aspects o f school management are available for both 1988 and 2003: the frequency o f visits o f the circuit supervisor (formerly school inspector) and the presence o f the PTA.'' The former, which has increased over time, is shown later to be a significant determinant o f the quality o f teaching methods. The latter variable i s not very revealing, since virtually all schools have a PTA, although some private ones do not. What matters is the extent to which the PTA supports the school. In 2003, considerably more data were collected o n the activities o f the circuit supervisor and the head teacher in supervising the work o f teachers. Data were also collected o n the support provided by the PTA and the workings o f the SMC and SPAM. Frequency of Supervision Visits 27. In 1988, circuit supervisors visited schools just over once every two months o n average (Table D.12). But there was considerable variation around this average, with over a fifth o f middle schools/JSS and a quarter o f primary schools receiving only two visits or less a year. By 2003 the mean number o f visits rose, from 6 to 9 a year for primary schools and a bit less for JSS. And the proportion o f schools receiving infrequent visits fell to 11percent for JSS and 16 for primary. Table D.12: Frequency o f school visits by circuit supervisor/school inspector, 1988 and 2003 (percent) 1988 I 2003 Primary Middle Primary JSS School school/JSS School Twice a year or less 25.2 21.o 16.3 11.1 Between three and six times a year 37.4 44.2 39.6 41.2 Between every one to two months 21.3 18.9 21.6 31.8 Once a month or more 16.1 15.9 22.5 15.9 Total 100.0 100.0 100.0 100.0 Memo items Sample size 282 230 417 289 Mean number of visits 6.40 6.30 9.07 7.89 28. Perhaps surprisingly, bivariate analysis does not reveal any significant difference between rural and urban schools (Table D.13). Rural schools were less likely to be among either the least visited or the most visited, more o f them falling into the category "between 10. Data o n both o f these was collected from the school questionnaire for which the respondent was usually the head teacher or the proprietor in the case o f private schools. 90 Annex D three and six times a year” than i s the case for urban schools. However, there i s a pattern regarding ecological zones. Schools in the forest zone are more likely to be visited by circuit supervisors. This was the case in both 1988 and 2003. In addition, in 1988 schools in the coastal zone were visited less frequently than those in the savanna area, whereas in 2003 the situation was reversed. Finally, there were few private schools in 1988 and the difference in supervision rates was not significant. But by 2003 there i s a large gap, with 45 percent o f private schools receiving infrequent supervision visits. Table D.13: Bivariate analysis o f frequency o f circuit supervisor/school inspector visits, 1988 and 2003 (percent) Urban Rural Coastal Forest Savannah Public Private All 1988 Twice a year or less 27.4 19.9 32.7 13.4 26.9 23.0 29.6 23.4 Between 3-6 times a year 37.1 43.3 39.6 43.8 34.4 40.3 44.4 40.4 Between 7-11 times a 18.1 22.0 14.4 21.9 29.0 20.2 18.5 20.0 year Once a month or more 17.3 14.9 13.4 21.0 9.7 16.5 7.4 16.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 237 282 202 224 93 491 27 519 2003 Twice a year or less 14.4 13.7 14.6 14.4 12.4 6.7 44.9 14.0 Between 3-6 times a year 36.4 47.1 35.1 39.6 53.6 41.2 36.2 41.5 Between 7-11 times a 27.9 22.0 32.7 23.0 22.7 29.2 11.6 24.8 year Once a month or more 21.3 17.3 17.6 23.0 11.3 22.9 7.2 19.7 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 451 255 205 404 97 568 138 706 29. These differences are partly supported by multivariate analysis (Table D.14). Since there are 49 schools receiving no visits at all the estimated model i s a two-part estimation model to allow for sample selection (Heckman).” The selection equation i s a probit model o f the factors affecting whether a school i s visited at all. Private schools are less likely to receive any visits at all, and those that do have fewer o f them. The same is true o f rural schools. Conversely, large schools are more likely t o be visited and to receive more visits. Finally, there has been an “autonomous shift” with schools more likely to be visited, and t o have more visits, in 2003 than in 1988. The zone dummies are not significant. 11. A tobit regression for a censored dependent variable i s not appropriate here, since tobit should only be used if the latent variable (desired number o f visits in t h i s case) can in principle take the censored values. Since the number o f visits cannot be negative, this condition i s not met. 91 Annex D Table D.14: Multivariate analysis of the number of visits from the circuit supervisor Coefficient z-stat Number of visits ** ~ Rural -0.95 -1.86 Primary 0.64 1.33 Private -5.65 -7.75 *** 2003 3.01 6.09 *** Number of teachers 0.13 2.36 ** Intercept 8.07 5.94 *** Selection equation Rural -1.86 * Forest 0.00 0.00 Savannah 0.00 0.00 Number of teachers 0.02 2.36 ** Private -0.70 -7.55 *** 2003 0.37 6.00 *** Intercept 0.96 5.55 *** lathrho 13.38 0.89 Ansigma 2.09 101.82 Number of obs 1218 Censored obs 49 Uncensored obs 1169 Log likelihood = -4144 Note: *** significant at I%, ** 5% and 10% Activities of the Head Teacher and Circuit Supervisor 30. Table D.15 reports results o n the engagement o f the head teacher and circuit supervisor with teachers in five supervision activities. The majority o f head teachers are, according to the teachers in their schools, actively involved in the different types o f supervision. Notably, less than 5 percent o f teachers say that the head teacher does not look at their lesson plans o n a regular basis (meaning at least once a week, which i s the frequency with which it is expected to be done). However, less than h a l f say that the head actually discusses the lesson plan with them. Somewhat higher percentages look at samples o f students’ work and sits in class at least once a week. 3 1. Whde the large majority o f schools have visits f i o m the circuit supervisor, 44 percent o f teachers have no direct contact with him o r her (Table D.15). The reported figures suggest that those that do have contact with teachers carry out the f i l l range o f activities shown in the table, though not much more than half discuss career development. 92 Annex D Table D.15: Head teacher and circuit supervisor supervisions Head teacher I Circuit supervisor QUIPS WSD %inall QUIPS WSD % i n all schools schools doing so on doing so a regular at all basis‘ Sits in on class 61.3 * ** 41.4 Looks at a sample of students’ work ** 52.4 ** 43.6 Looks at lesson plans ** 95.4 *** 51.3 Discusses lesson plans 45.7 ** ** 41.7 Discusses career development 57.2 *** 27.4 Memo item: Percent of teachers responding “Yes” to n.a. n.a. 70.3 62.5 55.7 question whether they had had direct contact with the circuit suDervisor Notes: *** significant at 1% level, ** 5% level and * 10 % level. I / Regular is at least once a week for all categories other than career development, which is if head teacher ever does so. 32. Table D.15 also reports tests o f significance for these variables for teachers in schools supported by QUPSand WSD (the fdl results are given in Annex E, Tables E.47 and E.48). There are only 2, out o f a possible 10, significant results for schoolteachers. However, it i s worth remarking that no teachers at a l l in QUIPS and WSD schools reported that head teachers do not look at their lesson plans. By contrast with the results for head teachers, there are seven significant results for circuit supervisors - four out o f five for WSD, and the case that is not significant is only marginally not so. However, legitimate questions can be raised about the direction o f causation since both projects are have begun work first in districts considered to have the necessary capacity to administer the project. Involvement of the Community: the PTA, SMC, and SPAM 33. The Parent Teacher Association (PTA) provides a means by which parents can support the schools attended by their children usually financially but also by providing help in kind. Virtually all schools have a PTA. Over 99 percent o f public basic schools did so in 2003, as do 94 percent o f private schools (Table D.16). Since PTAs are so widespread, statistical analysis will not be able t o pick up any effect they m a y have o n school processes and outputs. However, it is not the mere presence o f a PTA that will make the difference, but the extent to which i t provides support to the school. There i s considerable variation in the extent to which PTAs have provided support to schools and in the value o f parents’ monthly contributions (see below). 93 Annex D Table D.16: Presence o f PTA at public and private schools, 1988 and 2003 1988 I 2003 Public Private Total I Public Private Total Yes 96.3 96.2 96.3 1 99.1 95.7 98.4 No 3.7 3.8 3.7 0.9 4.3 1.6 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 488 26 514 568 138 706 34. SMCs are also widespread, being present in over 80 percent o f the schools surveyed (Table D.17). However, in only h a l f o f schools had SMCs met in the preceding month or provided support in the past year, and in even fewer helped the school in dealings with outside agencies. Where the school had not asked for support ftom the SMC this was often because i t was felt that the SMC would be either unwilling or unable to help (48 percent o f cases for primary and 39 percent for JSS). The lower prevalence o f SMCs than P T A s i s largely explained by the fact that they are not required at private schools: over 90 percent o f public schools have SMCs. But the facts remain that these SMCs are not active in a large number o f schools. For most o f the questions asked, the PTA was seen as a more supportive organization. Table D.17: School management organizations, 2003 PTA SMC Primary JSS Primary JSS Organization associated with school 97.8 99.3 81.06 85.8 Organization met in the last month 53.4 51 .I 50.3 42.3 School asked for or got support from 87.3 88.9 38.1 61.6 organization in the last year Organization provided support in the last year 63.7 66.6 50.0 46.8 Organization helped with dealings with district 40.7 45.3 44.1 44.0 or outside agencies 35. Virtually all public primary schools (92 percent) have had a SPAM, at 98 percent o f which an action plan was agreed. The most common actions agreed at the S P A M were (remembering that it is a teacher replying t o the survey) that parents should ensure children attend school (41 percent) and parents should provide pencils and exercise books (38 percent). The most common actions for teachers were to provide extra classes (33 percent) and to be punctual (17 percent). Problems o f absenteeism were mentioned in less than 10 percent o f cases.' Responsibility for implementation o f the action plan was seen to rest with the head teacher (47 percent o f cases) or the circuit supervisor (24 percent). In only 20 percent o f cases were the PTA or SMC said to be responsible. Finally, in only 6 percent o f cases was it said that the planned actions were not being carried out at all, and in 42 percent they were claimed to be being carried out completely. 1 It should be recalled, however, that the respondent for these questions was the headteacher. 94 Annex D 36. The school survey suggests there is little difference between rural and urban areas with respect to any o f the variables shown in Table D.17. If anything, school respondents reported PTAs t o be more active in urban areas, though there was no difference for SMCs. But the data collected from households give a different picture. Households with children in basic school were asked if there was a PTA and SMC at the child’s school and if any household member was a member o f the organization. They were also asked if there had been a S P A M at the child’s school and if any household member had attended. The rates for rural households are significantly higher than those for urban households for all six questions (Table D.18). Knowledge and participation in PTAs i s widespread. However, knowledge o f SMCs and the S P A M i s f ar less common that the school-level data suggests it should be, and participation rates correspondingly low. Only 6 percent o f households say that someone attended a S P A M at their child’s school. Urban Rural Total Exists Member/ Exists Member/ Exists Member/ Attend attend Attend *** PTA SMC 92.7 42.5 96.4 5.0 97.5 59.3 *. ** 96.7 11.3 *** 95.1 51.O 96.5 8.2 *** *.. SPAM 19.5 3.2 27.6 9.1 23.6 6.2 37. But while rural communities may be easier to mobilize in support o f schools, they also tend to be less well off, reducing their ability to provide financial support. Table D.19 reports regression results from the analysis o f average PTA contributions per pupil at the community-level.” The elasticity o f P T A contributions with respect to community income i s close to two. That means that doubling community income increases the value o f contributions to schools nearly threefold. In 2003, the richest community in the sample was more than five times richer than the poorest, suggesting that schools in the former will receive 15 times as much money through PTA contributions than schools in the latter. The actual range i s far higher, since some schools collect n o contribution compared with a maximum o f 150,000 cedis (see Annex C). On top o f that, the rural dummy i s significantly negative: rural communities give less cash support to schools through PTA contributions, most probably reflecting the fact that there i s less cash around in rural areas (which rely far more on o w n production and barter than do urban areas). 12. The average was calculated only with respect to children in basic school. In four communities the average was 0. Since the dependent variable i s logged these observations were assigned a value o f ln(lOO), compared to the observed non-zero minimum o f 400. Excluding these four observations does not make a substantive difference to the results. 95 Annex D Table D.19: Community-levelregression analysis o f determinants o f (logged) PTA expenditure per pupil Model 1 Model 2 Coeff. t-stat Coeff. t-stat ~~ Community variables Average community income 1.86 3.43 *** 1.95 3.98 *** (Qmd) Forest -0.17 -0.45 Savannah -0.14 -0.23 Rural -0.85 -2.16 ** -0.86 -2.40 ** School variables Head teacher supervision 4.37 2.39 ** 3.59 2.61 ** Activities Teacher social relations -0.01 -0.38 PTA -1.15 -0.57 SMC 0.80 0.78 SPAM -2.13 -1.71 * SMC participation 2.57 1.66 2.97 2.30 ** PTA participation -0.26 -0.25 SPAM participation 0.77 0.40 Dummy (cluster 40)’ 2.34 1.75 * 2.40 1.90 * Intercept -19.78 -2.06 ** -22.46 -2.72 *** R2 0.44 0.43 N 80 80 Notes: *** significant at I%, ** significant at 5% and * significant at 10%. I / Exceptionally high PTA contributions are probably explained by some PTA-managed investment, such as constructing a classroom block. 38. School-level variables also affect the level o f PTA contributions. Contributions are higher when head teachers are active in monitoring teachers’ work (this variable i s discussed more below). This result may be picking up one o f three things, o r a combination o f them: (1) parents appreciate a good headmaster; (2) a good headmaster i s also one who is active in soliciting support from the community (Le., the variable proxies for an unobserved variable of head teacher-community interaction); or (3) it i s indirectly picking up the effect o f teacher quality, which i s improved by a good head teacher. The cluster o f school management variables i s highly inter-correlated. In the full model the two S M C variables are both positive but insignificant (Model 1) -but either one alone i s significant, with S M C participation being the stronger o f the two. The S P A M variables are particularly highly correlated with the S M C variables. Including just one S P A M variable, and n o S M C variable, makes the former significantly positive. Given the extent o f this inter-correlation no weight should be given t o the negative coefficient on the S P A M variable in M o d e l 1. While the PTA variables have 96 Annex D negative coefficients, neither i s significant (and do not become so in any model specification). Teacher social relations with the community were not ~ i g n i f i c a n t . ' ~ 39. Similar results were found from a household-level analysis o f the determinant o f PTA contributions, but also some differences (Table D.20). The dependent variable in this case i s the l o g o f PTA contributions with respect to each child in basic school (so the model i s only estimated for households that have a child currently enrolled in basic school). Many such households make a zero contribution, so that OLS estimation would be biased. Instead the Heckman model i s used, which i s a two-part estimation procedure. First a probit model i s estimated o f whether or not the household will pay any contribution and second the determinants o f the level o f that contribution estimated for those households that are contributing. The lower part o f Table D.20 reports the results o f the selection equation. An obvious omission i s whether the child i s at a private school with no PTA (recall that applies to 4 percent o f private schools). However, the variable o f whether the respondent states that there i s a PTA at the child's school is a good proxy for this (as well as picking up the small number of public schools with no functioning PTA), and appears as the most significant determinant o f whether a PTA contribution i s paid or not. 40. B o t h community and household income (expenditure) matter for h o w m u c h i s paid to the PTA. The higher a household's income the more likely it is to make a PTA contribution, although the average community income does not matter for this decision, which is a sensible result. These results make sense since the PTNschool set the PTA levy as a minimum amount, which will be done with reference to community income. But whether a household actually pays depends o n its own resources, not those o f the surrounding community. Average community income i s the stronger (larger and more significant) determinant o f the level o f PTA contribution, although household income also matters. The elasticity o f PTA expenditure with respect to community expenditure appears as 1 in this model (ranging f r o m 0.90 to 1.05 in the various model specifications estimated). This i s lower than that estimated in the community-level model, but i s an under-estimate o f the effect o f income. Doubling community income means doubling the income o f every household, so the income elasticity i s the sum o f the coefficients on the two income terms, which i s 1.3. This is still an underestimate, since the doubling o f income will, through the selection equation, make households more likely t o contribute at all. The other household characteristic included - education o f the household head - matters for the level o f the contribution but not whether it i s made or not. 13. This i s a composite variable based o n if the teacher i s the member o f any community-based group and a 1-4 scale of how cordial they judge their relations with the community to be. 97 Annex D Table D.20: Determinants of household PTA contributions (Heckman maximum likelihood estimation) Fuii model Parsimonious model CoeR z-stat Coeff, z-stat Dependent variable: PTA expenditure (logged) Household expenditure (logged) 0.29 2.70 *** 0.25 2.69 *** Average community expenditure (logged) 1.01 6.73 *** 0.99 6.74 *** Education of household head 0.12 3.95 *** 0.13 4.19 *** Average community knowledge of SMC 0.51 2.24 ** 0.54 2.37 *** Household participation in PTA 0.71 3.24 *** 0.74 3.48 *** Average community participation in PTA 0.23 0.78 Knowledge of SPAM 0.12 1.40 0.13 1.43 Participation in SPAM 0.05 0.32 Teacher social index 0.00 -0.56 Supervision activities of head teacher 1.47 3.08 *** 1.48 3.28 *** Dummy for high observations 2.20 4.93 *** 2.19 4.94 *** Rural -0.61 -5.53 *** -0.58 -5.61 *** Forest -0.17 -1.68 * -0.20 -2.07 ** Savannah -0.38 -2.52 ** -0.41 -2.89 *** Intercept -12.17 -4.40 *** -11.49 -4.47 *** Selection equation Household expenditure (logged) 0.44 5.04 *** 0.44 5.59 *** Average community expenditure (logged) 0.02 0.13 Education of household head -0.02 -0.49 Average community knowledge of PTA 0.73 2.23 ** 0.62 2.01 ** Household knowledge of PTA 2.33 8.57 *** 2.35 8.80 *** Average community participation in PTA -0.20 -0.64 Participation in SMC 0.57 3.29 *** 0.59 3.47 *** Supervision activities of head teacher 1.11 2.37 ** 1.07 2.35 ** Teacher social index -0.02 -2.69 *** -0.01 -2.67 *** Rural -0.40 -3.65 *** -0.40 -4.33 *** Forest 0.19 1.68 * 0.20 1.76 * Savannah -0.10 -0.57 -0.09 -0.57 * Intercept -8.38 -3.08 *** -8.09 -5.68 *** lathrho 0.33 0.86 0.17 0.61 * /Insigma 0.08 1.62 0.07 2.31 ** No of observations 1348 1348 Of which censored 365 365 Log likelihood -2116 -2117 41. School management organization variables are important for P T A contributions. For reasons already given, the household stating that there i s a P T A matters for if a contribution i s made. But average community knowledge o f the P T A matters as does whether the household has been actively involved with the SMC. Community knowledge o f the P T A and 98 Annex D SMC both matter for the level o f the contribution. The implication i s that where these school management organizations are active in the community that each household feels more inclined to make a contribution and that contribution is larger. 42. A s with the community results, the presence o f a head teacher who i s actively involved in supervising teachers increases both the likelihood that a payment i s made and the level o f that payment. The only possibly perverse result in the regression i s that the better teacher-community relations then the less likely are households to pay PTA fees. This finding m a y reflect either that teachers with good social relations are less w e l l placed to enforce payment, or that relations are good precisely because they do not do so. 43. Finally, and unlike in the community results, the location dummies are significant. Forest region residents are more likely, and savanna ones less likely, to pay PTA fees than those in coastal region. But both regions pay a smaller amount than do coastal regions. Rural residents are both less likely to pay, and to pay less if they pay at all, than urban residents. This finding i s consistent with the community-level results. These location variables may reflect variations in the availability o f cash in the local economy, which is required if fees are to be paid. TEACHING CONDITIONS AND METHODS Teacher Training and Test Scores 44. Table D.21 shows the teacher test scores T a b l e D.21: Teacher test scores, for 1988 and 2003. N o change i s expected in the 1988 and 2003 Raven’s test unless teachers are n o w being drawn 1988 2003 t-stat f i o m a different segment o f the population. In fact, Raven’s test 29.6 29.0 -3.23 *** there i s a small but significant drop. However, mirroring progress in the rest o f the population, the Standard error (3.2) (6.5) math score has risen significantly, though English No. of observations 430 3,061 has not. These results thus do not give any clear English 22.6 22.5 -0.35 message regarding the academic ability o f Standard error (2.5) (4.0) teachers. No. of observations 436 3,051 Maths 19.9 21.4 8.01 *** 45. The level o f schooling among teachers has Standard error (3.4) (5.5) risen. In 1988, only 40 percent were secondary No. of observations 435 3,050 school graduates, compared to three-quarters today Local language n.a. 24.9 (Table D.22). Two factors lie beyond changes in Standard error 3.9 the education levels o f teachers. One is the rise o f N ~of. observations 1,793 private schools. These schools typically do no Note: *** significant at 1%. Figures for 1988 require recruits to have teacher training but use are mean of school-level average. secondary school graduates. In 2003,94 percent o f private school teachers have at least secondary education, compared to 72 percent o f teachers in public schools. But less than 15 percent o f private school teachers had teacher training, compared to 88 percent in public schools. 99 Annex D Table D.22: Teacher schooling (percent) 1988 2003 Primary 1.7 0.0 Middle/JSS 57.2 19.5 Senior Secondary 40.8 75.6 Tertiary 0.4 5.0 Total 100.0 100.0 No. of observations 4547 3129 Table D.23: Teacher education and training b y public vs. private, 2003 Public Private Total Schooling MiddlelJSS 22.4 6.2 19.5 Senior Secondary 72.3 90.1 75.6 Tertiary 5.3 3.7 5.0 Total 100.0 100.0 100.0 Teacher training Yes 87.5 14.7 74.4 No 12.5 85.3 25.6 Total 100.0 100.0 100.0 No. of observations 2564 565 3129 46. The more general trend in trained teachers i s shown in Figure D.3.I4 In Figure D.3: Fraction o f trained teachers, primary and JSS public schools untrained teachers are called “pupil- 100 , 90 J teachers.” These are teachers 80 - o n the teaching staff, and should be distinguished from “ii parental volunteers who may E also help out, especially when class sizes are large or a n 30 20 school short o f teachers. Official policy has been t o lo0 eliminate pupil-teachers, and 84/85 88/89 92193 96197 00101 as part o f the r e f o m s in the -Primary JSS 1990s they were given a Source: MoE data period to acquire training. If they failed to do so they would lose their jobs. This policy resulted in the rise in 14. The data cover all teachers other than 1991-2 which are for public schools only. 100 Annex D the proportion o f teachers who are trained, which i s clearly shown in the figure. Since the late 1990s the upward trend has reversed itself especially in primary schools, though this is partly the result o f the rise in private schools. 47. The government’s policy i s to increase reliance on in-service training (INSET) to develop teaching skills. This has not happened. Only 3 percent o f teachers receive such training o n a regular basis. Teaching Methods 48. Teachers were asked three questions to test their familiarity with improved teaching methods, and the extent to which they claim to use them in the classroom. The questions, described in the titles to Tables D.24-D.26, asked if children are encouraged to explore material by themselves, the use o f simulations (role play), and the use o f cues in explaining a word. In the second case, in which they were asked directly if they use simulations, those claiming to use them were asked to explain the approach. The results show that about a third of teachers use a student-centered learning approach and use simulations o n a regular basis, though about a fifth of the latter could not explain them properly. And about one-fifth use cues to help explain difficult words. In summary, improved methods are far from unknown, but their use cannot be described as widespread, being used by a minority o f teachers. Table D.24: Which of the following describes your approach to teaching? ~~ Number Percent Allow children to explore material on their own 1,141 36 Present material to children which you 1,988 64 have prepared in advance Total 3,129 100 Table D.25: H o w often do you use simulations as an instructional approach? Number Percent Of which percentage able to give a correct explanation Often 1,007 32 80 Sometimes 1,229 39 76 Rarely 357 11 44 Never 536 17 ma. Total 3,129 I00 73 101 Annex D Table D.26: One o f your pupils has difficulty in pronouncing a word in a group reading. How will you go about helping the other pupils to understand the word? Number Percent Would not do so 18 1 Tell the student to read the word again 661 21 Explain or define the word 1,286 41 Use cues in the story to explain the word 653 21 Other 51 I 16 Total 3,129 100 49. The data from these questions were used to construct a single composite variable on teaching methods (TMETHODS). This variable i s a simple average o f the three responses (multiplied by 100/4), each response re-scaled as necessary to range from 1 to 4. Teachers unable to correctly describe simulations were re-coded as 1 (‘Never’ use) for question two. The average value o f this variable for all 3,129 teachers is 62.5 (Table D.27), but with a reasonable degree o f variation (the coefficient o f variation i s equal to 0.28). Trained teachers are significantly more likely to use improved methods than untrained ones, although there i s not a significant difference between teachers who have received university-level teacher training and those trained by TTCs. There i s also significant variation across the country, with both the forest and savannah zones showing significantly less use o f improved methods than the coastal region. Table D.27: Bivariate tabulation o f teaching methods against teacher training and geographical zone (percent) Teaching Standard Number of methods deviation observations &stat* Teacher training None 57.3 16.6 803 Certificate 64.1 17.7 2247 9.84 Tertiary 66.3 18.5 80 4.16 - Geographical zone Coastal 65.2 18.3 933 Forest 61.2 17.6 1835 -5.49 Savannah 61.4 15.9 361 -3.71 Total 62.5 17.7 3129 Note: *for teacher training the t-stat compares with the row above, for zone both t-stats compare with the coastal region. 50. These findings remain valid for multivariate analysis, shown in Table D.28. The table reports results both a hllmodel (Model 1) and a more parsimonious one (Model 2) including only variables which are significant in the full model. Both teacher training and in-service training are significant, the former markedly so. Also significant are the teacher’s level o f schooling and h i s or her ability as measured by the Raven’s test. However, neither o f the teacher living and working conditions are significant and neither is morale. None o f these three 102 Annex D variables are individually significant if the other two are dropped. Head teacher supervision o f teachers has a significantly positive impact. Introducing the variables that make up this composite separately into the regression shows the strongest effect to come from the head teacher “sitting in” on the class (this question was interpreted as appearing in the class rather than necessarily sitting down for any length o f time). The teacher having had direct contact with the circuit supervisor also has a positive effect, although the variables capturing the activities o f the circuit supervisor are insignificant. Table D.28: Regression analysis of determinants o f teaching methods Model 1 Model 2 Model 3 CoeR t-stat CoeR t-stat 2oeK t-stat Teacher training and ability Teacher training 6.31 8.77 *** 6.31 8.93 *** 5.24 6.31 *** In-service training 3.36 2.60 *** 3.21 2.50 ** 1.74 1.16 Teacher schooling 2.58 3.37 *** 2.67 3.81 *** 2.59 3.70 *** Teacher Raven’s score 0.10 2.10 ** 0.11 2.24 ** 0.10 2.15 ** Teaching conditions Teacher morale 0.90 0.94 Teacher working conditions -0.15 -0.31 Teacher living conditions 0.03 0.11 Supervision Head teacher supervision 2.76 2.19 ** 2.33 1.88 * 2.33 1.89 * Visit by circuit supervisor 4.32 5.19 *** 3.60 5.51 *** 3.53 5.40 *** Circuit supervisor supervision -4.31 -1.41 Other variables Teacher’s discipline 1.18 3.30 *** 1.15 3.23 *** 1.15 3.23 *** Teacher’s perception of materials 0.77 3.03 *** 0.75 2.97 *** 0.70 2.76 *** Primary teacher -0.27 -0.38 Combined school 1.37 1.79 * 1.41 1.85 1.44 1.90 * Forest -3.55 -4.73 *** -3.52 -4.75 *** -0.49 -0.37 Savannah -5.34 -4.50 *** -5.49 -4.69 *** -2.45 -1.54 Constant 47.38 11.34 *** 45.60 13.79 *** 43.87 13.01 *** Interactive terms Teacher training in coastal region 3.49 2.46 *** In-service training in coastal region 5.55 1.92 ** N 2,939 2,953 2,953 R2 0.068 0.067 0.070 Note: *** significant at I % , ** significant at 5% and * significant at 10%. 5 1. Two teacher perception variables play a significant role. One is that teaching methods are better the worse the teacher perceives the supply o f materials to be. There are two possible explanations for this finding. The first i s that the teacher compensates for inadequate materials by using more innovative methods. A second explanation i s that teachers wishing to use more innovative methods are more likely to perceive materials as being inadequate as 103 Annex D those teachers who invest less heavily in method. Support for the second explanation is given by the fact that teachers at the same school (i.e., with objectively the same materials available) are more likely to perceive materials as inadequate the higher their value o f TMETHODS. Second, the variable DISCIPLINE, which captures the severity with which a teacher believes various offences committed by teachers should be punished, i s significantly positive. This variable should probably not be interpreted as an explanatory variable but rather as the correlation in different aspects o f professionalism among teachers. 52. The primary school dummy i s not significant, so primary teachers are neither more nor less likely to use improved methods than those in JSS. However, the combined school dummy is significant at 10 percent. This finding may reflect a spillover effect, which is more likely in larger schools (although the number o f teachers i s not significant when included in the model), or may reflect the concentration o f better teachers in those schools. 53. Finally, the zonal dummies are significant, implying that there are factors common to non-coastal areas that are not included in the model, which encourage the lesser use o f improved teaching methods in these areas. One explanation may be that teacher training in the coastal region introduces teachers to these methods more successfully than does teacher training elsewhere. To test this hypothesis interactive terms were introduced o f the coastal dummy multiplied by each o f the teacher training variables.” The results (Model 3) strongly support the hypothesis: the two interactive terms are significantly positive (with the result that coefficients o f the training variables and their significance is reduced, becoming insignificant in the case o f in-service training),I6 and the zonal dummies become insignificant. 54. Dummies were not included in the regressions for the WSD and QUIPS projects since these projects are restricted to primary schools and so their inclusion would have reduced the sample size. When they were introduced into the regression neither variable was significant, suggesting the projects have n o direct effect over and above the indirect effect through their influence o n any o f the variables already included in the model. However, bivariate analysis of the difference in means o f teachers in project schools and other teachers showed n o significant impact o f either project on teaching methods as captured in our data, suggesting that they have n o indirect effects. 55. Data were collected o n the frequency with which teachers set homework, look at and assess students’ work for both math and English. Table D.29 summarizes results for two o f these. Homework is set at least once a week by over 95 percent o f teachers for math and English, and work assessed with the same frequency by half, The most striking result i s the lesser attention paid to local languages, for which homework i s set infrequently by nearly 20 percent o f teachers and work rarely assessed in over h a l f the total number o f cases. 15. This test i s problematic to the extent that teachers may have been trained in a different zone from that in which they are teaching, though t h i s problem does not apply to in-service training. 16. This does not mean that in-service training i s ineffective, but that it i s only effective at improving teaching methods in the coastal zone. 104 Annex D Table D.29: Monitoring of student performance by teachers Homework I Assess work English Math Local English Math Local language language Once a month or less 3.8 3.0 17.3 39.5 38.6 52.1 Once a week 78.8 76.7 79.2 50.2 51.2 44.9 Daily 17.4 20.2 3.5 10.3 10.2 2.9 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 2055 204 1 1552 2054 2041 1561 56. These data were used to construct an index o f student monitoring by teachers.” A regression model was used to examine the determinants o f this behavior (Table D.30). B o t h in- service training and visits by the circuit supervisor were found to exert a significant positive impact o n student monitoring. School quality variables appear not to matter, although the availability o f materials has a positive effect, which may be picking up particularly good schools. Teachers who think that school management i s a problem are more likely to undertake monitoring (perhaps because better teachers more readily perceive such problems), whereas teachers who think morale is a problem are less likely to monitor student performance (perhaps as bad teachers are more likely to complain about morale). Table D.30: Determinants of student monitoring by teachers Full model Parsimonious model Coeff. t-stat Coeff. t-stat Teacher training, ability and attitude Teacher training -4.12 -0.75 -6.99 -1.38 In-service training 15.61 2.01 ** 15.58 2.03 ** Teacher schooling 3.24 0.65 Teacher’s Ravens score 0.04 0.1 1 Teacher’s English score -0.14 -0.22 Teacher’s Math score 0.22 0.49 Teaching method 0.56 4.52 *** 0.56 4.72 *** Discipline 3.72 1.54 3.73 1.6 Teaching conditions Teacher’s morale -6.28 -0.98 Teacher working conditions 5.20 1.55 2.92 0.93 Board quality: size -1.80 -0.25 Board quality: easy to Clean -0.65 -0.10 17. For example, the frequency (number o f times a week) with which a teacher set homework for the three subjects was calculated and averaged over the subjects taught by that teacher to get an average homework frequency. The same was done for looking at and assessing students’ work. A simple average was taken o f these three averages to arrive at the student monitoring composite variable. The variable can range from 0 (for a teacher who never does any o f these things in any subject, there are 12 such teachers in our sample) to 5 (for one who does them all in all subjects they teach once a day, which 2 percent o f our sample claim to do). 105 Annex D Full model Parsimonious model Coeff. t-stat Coeff t-stat Materials to display 6.86 2.14 ** 6.89 2.33 *** Shared classroom 11.56 1.52 10.41 1.42 Full-sized internal walls -8.63 -1.17 Class disturbed by noise -5.48 -1.19 Living conditions 2.71 1.30 2.83 1.4 Supervision Head teacher supervision 2.55 0.30 0.48 0.06 Circuit supervisor supervision 6.42 0.31 3.64 0.18 Visit by circuit supervisor 13.11 2.32 ** 12.29 2.23 ** Teacher perceptions of Student discipline 0.28 0.14 Student ability -3.32 -1.63 -3.77 -1.98 ** Availability of materials -0.24 -0.14 Teacher morale -4.04 -1.85 * -3.51 -1.67 * School management 4.09 1.76 * 4.26 1.9 * Other variables Primary school 28.70 5.37 *** 27.03 5.64 *** Combined School -0.33 -0.06 Urban/rural -0.83 -0.18 Forest -7.19 -1.40 Savannah -8.40 -1 .oo Constant 57.42 1.58 50.04 1.83 * N 2323 2381 R2 0.042 0.040 Note: *** significant at I % , ** significant at 5% and * significant at 10%. Textbook U s e 57. The data show a substantial rise in textbook availability. However, it i s not the availability o f textbooks that matters but whether they are used or not. The teacher questionnaire asked whether textbooks had been used in the last class taught o n each subject. For math and English, nearly all teachers h ad textbooks available and over 90 percent used them (Table D.31). For local language nearly h a l f did not have textbooks, and a lower percentage o f those who have them used them. These findings are consistent with those o f Okyere et al. (1997) who found that textbooks were used when there are sufficient o f them. 106 Annex D Table D.31: Most teachers use textbooks when they are available English Math Local language No textbooks available 1.9 2.4 44.2 Yes 94.5 90.7 47.6 No textbooks available 3.6 6.9 8.2 Total 100.0 100.0 100.0 Memo: % of those having textbooks who use them 96.3 93.0 85.3 Time on Task 58. The teacher questionnaire included questions about classroom activity, specifically the amount o f time in a typical class spent: e Preparing for class, handing out materials, writing on the board material for exercises or copying e Disciplining students All students engaged in copying, reading, or other forms o f exercise e Dealing with students on a one-to-one basis e Addressing the whole class 59. Enumerators reported difficulties in administering this question. The total implied class time from summing the answers to the above questions ranged from 5 minutes to four hours. However, these extremes were limited to a handful o f observations. O f the 3,128 observations with complete data, 3,107 are retained in the sample if those with a total class time o f less than 25 minutes or more than two hours are dropped. The results, shown in Table D.32, show that on average 28 percent o f time is spent on getting ready for class or disciplining students. By a broad definition o f time on task therefore 72 percent o f class time i s spent on task, though t h i s varies from 29 to 100 percent. A narrow definition o f time on task takes into account only the activities that actively engage all students, in which case the average is 33 percent, ranging from 0 to 73 percent. 107 Annex D Table D.32: Time on task: classroom activities Mean Minimum Maximum Coefficient of variation Total class time (minutes) 51 25 120 Of which (percent) Preparing for class 21 0 68 0.55 Disciplining students 7 0 48 0.69 All students engaged copying etc 33 0 73 0.30 Dealing with student I-to-I 13 0 68 0.62 Addressing whole class 26 0 80 0.51 Memo: time on task Broad definition 72 29 100 0.17 Narrow definition 33 0 73 0.30 Table D.33: Time on task: classroom activities (only those with class time of an hour or less) Mean Minimum Maximum Coefficient of variation Preparing for class 22 0 65 0.51 Disciplining students 7 0 33 0.65 All students engaged copying etc 34 0 73 0.30 Dealing with student I-to-I 12 0 63 0.62 Addressing whole class 25 0 80 0.54 60. The robustness o f these results was checked by considering the means only for those reporting a total class time o f one hour or less, a sample o f 2,215 teachers (Table D.33). The percentage distribution o f activities i s hardly changed. 61. Bivariate analysis shows n o relationship between the sex or location o f the teacher but a positive association with teacher training. These results are supported by multivariate analysis for the determinants o f the broad definition o f time o n task (the narrow definition regression does not yield good results); see Table D.34. The regression shows also that teacher training matters, as does overall teacher morale at the school (the individual teacher’s morale i s significant if mean morale i s not included). In-service training i s n o t significant, but it should be recalled that less than 5 percent o f teachers receive such training on a frequent basis. 108 Annex D Table D.34: Determinants o f time on task Broad definition 1 Narrow definition CoeH t-stat 1 CoeH t-stat Teacher characteristics Male -0.34 -0.67 -0.23 -0.55 Teacher training 2.96 4.63 *** -0.02 -0.03 In-service training -0.37 -0.40 0.31 0.41 Old dummy 0.26 0.34 0.97 1.55 Teaching methods -1.04 -0.80 -1.31 -1.23 School characteristics Primary -0.29 -0.55 0.95 2.20 ** Private 0.42 0.52 -0.70 -1.06 Board quality 1.41 0.33 2.31 0.65 Class have internal walls 0.23 0.30 0.35 0.56 Class disturbed by noise -1.00 -2.08 ** 0.27 0.69 Display material available 1.83 5.36 *** 0.29 1.05 Head teacher supervision 3.70 4.11 *** 0.92 1.25 Circuit supervisor supervision 5.42 3.12 *** 1.11 0.78 Textbook availability 1.11 3.45 *** 0.48 1.83 Chalk availability 1.33 3.44 *** 0.02 0.07 School mean teacher morale 2.03 2.84 *** 2.22 3.79 *** School mean discipline 1.84 4.02 *** 0.51 1.35 School mean teacher perception of management -1.30 -3.40 *** -0.37 -1.19 Community Characteristics Savannah -0.03 -0.04 -1.93 -2.58 *** Forest 2.76 5.23 *** 0.38 0.88 Rural -0.06 -0.12 0.31 0.74 Intercept 44.65 7.95 **** 23.52 5.12 *** R squared 0.08 0.02 No. of observations 2,919 2,919 62. School quality matters to the time spent o n task: materials matter (display material, textbooks, and chalk, and the average o f teachers’ perceptions o f material availability), as does the quality o f infi-astructure (class disturbed by external noise). School management also matters: supervision o f the teacher by both the head teacher and the circuit supervisor improve time o n task, as does teachers’ perception o f school management (the worse i s the perception then the less the time o n task). CONDITIONS, MOTIVATION, TEACHER AND MORALE 63. This section considers three related variables: teacher morale, teacher working conditions, and teacher living conditions. Few data were collected o n these issues in 1988 - though those available mostly show a clear improvement - so the discussion i s mostly restricted to an analysis o f the 2003 data. 109 Annex D Descriptive Analysis o f Teacher Morale and Conditions 64. The teacher morale variable is constructed as the simple s u m to two questions: do you enjoy being a teacher (no=O, yes=l) and do you intend to remain as a teacher (no=O, yes=l)? The resulting variable, shown in Table D.35, i s categorical fiom 0 to 2.'* The variable suggests reasonably highmorale, with two-thirds o f teachers being in the top category. 65. Questions were also asked o n teachers' Table D.35: Measures of teacher morale subjective perceptions o f living and working Enjoy Remain Morale conditions, based o n a 5- and 4-point scale, 0 13.3 31.0 10.2 respectively. The former o f these was combined 1 86.7 69.0 24.0 with the results o f a question regarding the cordiality o f relations with the local community 2 65.9 to make a subjective livings conditions index Total 100.0 100.0 100.0 (Table D.36). No. of observations 3,129 3,129 3,129 Table D.36: Subjective perceptions o f working and living conditions How would you describe How would you describe the your working conditions? conditions of your accommodation? Urban Rural Total Urban Rural Total Very poor 8.0 4.9 7.0 Very poor 6.3 5.6 6.1 Poor 33.7 30.6 32.7 Poor 19.4 20.0 19.6 Good 54.2 59.2 55.8 Adequate 27.2 29.5 28.0 Very good 4.2 5.2 4.5 Good 38.5 38.4 38.5 Total 100.0 100.0 100.0 Very good 8.5 6.4 7.8 No. of obs. 2096 1033 3129 Total 100.0 100.0 100.0 Chi-squared 16.39 (0.001) Chi-squared 6.05 (0.196) (Prob.) (Prob.) 66. Teachers appear more satisfied with their working conditions (60 percent replying they are good or very good) compared to their living conditions (46 percent describing as good or very good). Urban teachers are less satisfied with their working conditions than those in rural areas, though there i s n o difference in the perception o f living conditions (although the objective measures given below suggest they are worse in rural areas). 67. Teacher morale i s related to all three o f the subjective perception variables mentioned above, in particular working conditions (Table D.37) 18. The components o f teacher morale are unsurprisingly correlated. Over three quarters o f those who do not enjoy being a teacher plan to leave teaching, compared to under a third o f those who do enjoy it. However, the creation o f the composite gives a b i t more variation in the dependent variable. 110 Annex D Table D.37: Relationship between teacher morale and teacher conditions Teacher morale Chi- squared Low Medium High Tofa/ ~ ~~ Subjective living conditions Very poor 10.1 6.0 5.5 6.1 15.6 Poor 21.7 20.4 19.0 19.6 (0.049) Adequate 27.4 26.4 28.7 28.0 Good 32.4 39.2 39.2 38.5 Very good 8.5 8.0 7.7 7.8 Total 100.0 100.0 100.0 100.0 Subjective living index Low 30.8 25.2 21.9 23.6 16.1 Medium 25.2 22.8 25.7 24.9 (0.003) High 44.0 52.0 52.5 51.5 Total 100.0 100.0 100.0 100.0 Subjective working conditions Very poor 15.4 9.6 4.8 7.0 152.2 Poor 49.1 37.6 28.3 32.7 (0.000) Good 33.3 48.7 61.9 55.8 Very good 2.2 4.1 5.0 4.5 Total 100.0 100.0 100.0 100.0 No. of observations 318 750 2061 3129 68. Objective data were also collected o n teacher conditions. Regarding living conditions data are available on whether pay i s received o n time, if housing i s provided for the teacher, the type o f water supply at the teachers’ housing, if teacher housing has electricity, and if the teacher i s a member of a group within the community (e.g., church, cultural organization, or sports club). Data on the first o f these were also collected in 1988, although at the school rather than teacher level and for a smaller number o n water and electricity.” All variables that can be compared across time show an improvement in the living conditions o f teachers. 69. n 1988,95 percent o f schools did not provide lodging for any o f their teachers. By 2003 I this figure fell to 70 percent. Today, 60 percent o f teachers have access to pipe-borne water compared to only 20 percent 15 years ago. I nonly 13 percent o f schools do no teachers have ntheir home in 2003 compared to half in 1988. I electricity i n2003,72 percent o f teachers reported that they always or in most months received their salary o n time, compared to the only 25 percent o f schools for which it was reported that salaries were almost never late in 1988. The problem o f late pay is greater for new teachers. Teachers with less than a year’s experience are significantly more likely to say that they never receive their pay o n time. But there remains a not insignificant group of older teachers who get their pay late, and this is found to be a critical factor inteacher morale (see below). 19. In 1988, data were only collected o n whether teachers’ lodgings had water and electricity if those lodging were provided by the school, which was a very l o w percentage. 111 Annex D 70. O f the five objective Table D.38: Objective index o f living conditions living conditions four show a Urban Rural Total bias in favor o f teachers living Low 22.9 53.1 32.9 in urban areas, two strongly so Medium 42.9 31.9 39.3 (water and electricity). Just one indicator i s more High 34.2 14.9 27.8 favorable in rural areas. That Total 100.0 100.0 100.0 is having lodging provided, No. of observations 2096 1033 31 29 though this i s not common Chi-squared (Prob.) 306.2 (0.000) anywhere (21 percent o f teachers in rural areas and 6 percent in urban). I nconsequence, the objective index o f living conditions, calculated as the average o f the scaled values o f these five variables, i s better in urban areas than rural (Table D.38) -posing a puzzle as to why subjective perceptions o f living conditions do not vary between rural and urban areas. 71. Data o n working conditions covered both teacher-level data o n the various dimensions o f school quality and school-level data. The analysis o f material and physical inputs earlier @ara. 5 ff) has already shown h o w these have improved in nearly all schools between 1988 and 2003. Determinants of Teacher Morale and Conditions 72. Each o f teacher morale, and teacher perceptions o f living and working conditions was modeled as an ordered probit. Inthe first and most general equation, both subjective and objective measures were included. However, it seems likely that the three dependent variables are determined simultaneously so that there is a problem in including the subjective perceptions as regressors. Hence model 2 in each case drops these variables, which m a y be considered as reduced form estimate. Since a more parsimonious model i s identified in each case with some differing regressors the model can be considered to be identified, with model structure determined by a data analytic approach. 73. Table D.39 summarizes the results, from which the following main points emerge: 0 Teacher characteristics: young teachers and especially males, especially those in rural areas, are more generally dissatisfied. Better-qualified and -educated teachers also tend to be less satisfied. Living in the home district and with one’s spouse both have a positive effect. a Teacher living conditions: a resoundingly robust result i s the importance o f receiving pay on time, which has a significantly positive effect o n all three measures. Having good social relations with the community are also important. Other aspects o f living conditions affect the subjective perception o f living conditions, but not the other two variables. a Teacher-level school variables: several o f these are significant, virtually all with the expected sign. Both board (easy to clean) and classroom (not disrupted by noise) quality affect teacher perceptions o f both working and living conditions. While some aspects o f school quality thus seem to spill over into perceptions o f living conditions (but not vice 112 Annex D versa, see previous bullet), two perverse results appear with respect to in-service training and visits o f the circuit supervisor. Also surprising is the negative impact o f most variables measuring “quality” o f other teachers at the school. 0 School management and projects: the school management variables send mixed signals. This result partly follows from their inter-correlation. Both PTA variables have a positive effect on subjective working conditions, and PTA is positive in the subjective living conditions equation. Despite the small number o f observations, the W S D dummy is significant in one case (living conditions). 0 Community characteristics: taking into account all these factors, the indices are systematically higher in forest and savannah zones and i nrural areas. However, there is a negative effect from community income and education (which are positively correlated with the objective measure o f living conditions). The likely explanation (arising from field observation) is that teachers compare themselves with their peers in the neighboring community -they are much lower down the scale in well o f f communities than poor ones, and so will be less satisfied with their lot. Table D.39: Determinants of morale and conditions (results from regression analysis) Subjective working Subjective living Teacher conditions conditions morale Teacher characteristics Male -ve)( ,, ,, -ve )( -ve (***) Young ( ~ 3 0 ) ,, -ve)( Young in rural area -ve (*) -ve (*) Young male in rural savannah ) , +ve ( Old p 5 0 ) ,, +ve)( Living with spouse +ve ( ,*) Living in home district ) , +ve ( Raven’s score , ) +ve ( English score , ) -ve ( , ) +ve ( Years of schooling -ve (*) -ve ( ,*) Level of teacher training -ve (**) Subjective indices Morale +ve C**) n.a. Subjective living conditions ,, +ve)( n.a. Subjective working conditions n.a. , ( +ve ) , ( +ve ) Teacher living conditions Pay received on time ,* +ve ( ) +ve) ,, ( ,, +ve)( Water in residence ,, +ve)( Electricity in residence +ve ( ,*) Lodging provided +ve (,**) Member of social organization +ve ) ( , Perception of social relations +ve ) , ( +ve (*) Teacher-level school data Frequency of in-service training -ve (,) 113 Annex D Subjective working Subjective living Teacher conditions conditions morale Teach extra classes +ve (**) +ve (**) Have to share classroom -ve (*) Noise disrupts classes ,, -ve )( -ve (*) Head teacher visits classes +ve (*) Display material available +ve r**) Board easy to clean +ve Y**) ,, +ve)( Teacher meets with circuit supervisor +ve (**) Circuit supervisor monitoring of +ve (*) -ve (*) teacher activities School-level data Own water supply +ve (**) Average level of teacher training -ve),,( +ve (*) Average level of teacher schooling -ve (*) Average teacher test score -ve (*) School management PTA +ve (*) +ve (*) SMC -ve (*) PTA met in last month ,, +ve)( SMC helped in the school in past year -ve (*) Plan from SPAM being implemented ,, +ve)( Project dummies Whole School Development +ve (**) QUIPS Primary School Development Community variables Average income ,, -ve )( Average education of household -ve r**) heads Savannah +ve Y**) +ve ) ( , Forest ,, +ve)( ,, +ve)( Rural +ve (***\ +ve I***) Absenteeism 74. In 1988, data were collected on absenteeism at the school level, asking how many teachers had been absent for reasons other than sickness during the last 12 months. I n2003, this question was included in the teacher roster o f the school questionnaire, asking if the teacher had been absent in the past four weeks for reasons other than sickness. The results o f the two surveys should not be comparable because the longer reference period used in 1988 will bias results toward finding a greater degree o f absenteeism in that year. 114 Annex D 75. However, despite this bias, the data clearly Table D.40: Percent teachers absent show that absenteeism has increased over the past by year and type o f school 15 years. In 2003, nearly 13 percent o f teachers had Primary Middle/ Total been absent in the past month, compared to just JSS over 4 percent in 1988 (Table D.40). 1988 4.7 3.7 4.3 2003 12.8 12.8 12.8 76. Correspondingly, more schools are affected by absenteeism. In 1988, 85 percent o f schools did not suffer at all; whereas this figure has n o w fallen to 61 percent, with 13 percent o f schools have over one-third o f the teachers being absent for reasons other than sickness in the past month (Table D.41). 77. Table D.42 reports bivariate analysis o f Table D.41: Absenteeism rates by year school-level absenteeism rates using 2003 data. (percent of schools in each category) The strongest difference i s between public and private schools: 80 percent o f private schools 1988 2003 have no problem with absenteeism, compared to None 85 61 not much more than h a l f o f public schools. There Up to a third 11 26 i s also a significant relationship with rural versus Between one to two thirds 3 9 urban schools, 7 percent o f rural schools More than two thirds 1 4 suffering absenteeism rates o f over two-thirds. Total 100 100 L i k e l y reasons for greater absenteeism in rural No. of observations 518 706 areas are that: (1) teachers may live in town some distance from the school and suffer transport problems, (2) they have to travel to town once a month to collect their pay, which they may find is not yet there, and (3) rural teachers attend t o their farming activities." Finally, absenteeism i s worst in the forest zone perhaps because o f the greater scope for profitable fanning in the zone. Table D.42: Cross-tabulations for absenteeism rates, 2003 Coastal Forest Savannah Urban Rural Public Private None 66 56 69 61 61 56 80 Up to a third 25 28 19 27 23 29 14 Between one to two thirds 5 12 10 9 10 11 4 More than two thirds 4 4 2 2 7 5 1 Total 100 100 100 100 100 100 100 No. of observations 205 404 97 451 255 568 138 Chi-squared 13.9" 9.9" 28.3." Note: ** significant at 5% level, *** significant at 1% level, 78. Multivariate analysis of teacher-level data also shows that private schools are less likely to suffer from absenteeism (Table D.43). I t also shows that poor working conditions are associated with a greater likelihood of absenteeism. The subjective working condition index i s significantly negative. The most important component o f that index - receiving pay on time -i s so important it i s also significant in its o w n right when entered alongside the 20. A main source o f income for urban teachers i s extra classes, w h c h necessarily do not take place during school hours. Rural communities, which are more cash constrained, offer fewer opportunities for extra classes. 115 Annex D index. There m a y be a direct effect o f time taken by teachers in going to inquire about their pay. L o w morale is also associated with absenteeism. 79. Some direct measures o f school conditions also matter. A high pupil-teacher ratio encourages absenteeism, as does poor facilities as measured by lack o f desks. The effect o f the head teacher discussing lesson plans is a perverse result. 80. On the other hand there are a cluster o f living condition variables, such as living with spouse, being in the home district and having good social relations, that appear conducive to absenteeism.” Presumably such circumstances provide distractions from work. Table D.43: Determinants o f teacher absenteeism Coeficient z-stat Teacher characteristics Male 0.110 1.15 Age -0.006 -1.24 In-service training 0.030 0.22 Teacher training 0.189 1.28 Teacher perception of morale 0.068 1.69 * Teacher conditions Teacher morale -0.133 -1.93 * Subjective working conditions -0.112 -1.69 * Pay on time -0.088 -2.05 ** Subjective living conditions 0.073 1.69 * Social relations 0.124 1.52 Home district 0.122 1.31 Living with spouse 0.094 1.27 Objective living conditions -0.515 -1.73 School characteristics Primary -1.505 -1.67 Private school -0.299 -1.75 PTA helped in last month -0.1 13 -1.21 QUIPS -0.359 -1.37 Desks -0.179 -2.10 ** Pupil teacher ratio 0.01 1 3.45 *** Head teacher discuss lesson plans 0.495 2.18 ** Community characteristics Per capita expenditure 0.130 1.oo Forest 0.240 2.55 ** Intercept 0.725 Number of obs 1606 Pseudo R2 0.074 Note: *** significant at I%, ** 5% and * 10% ~ 2 1. These three variables are not individually significant but are so jointly. 116 Annex E Annex E: Tables Of School Quality Variables INPUTS RECURRENT (a) Full sample Table E.l: Chalk (full sample) Primary I Midd/e/JSS I Total 1988 2003 1 1988 2003 I 1988 2003 Never enough 12.7 8.4 8.7 9.3 10.9 8.8 Occasional shortages 66.8 5.5 72.7 10.7 69.5 7.6 Always enough 20.5 86.1 18.6 79.9 19.6 83.6 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 283 417 231 289 514 706 Chi-squared (Prob) 388.9 (0.000) 220.6 (0.000) 553.2 (0.000) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Less than one book between two 58.4 11.3 42.1 22.1 51 .I 15.7 At least one book between two 20.6 16.3 37.3 42.6 28.1 27.1 At least one book per student 21.o 72.4 20.6 35.3 20.8 57.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 286 417 233 289 519 706 Chi-squared (Prob) 214.13 (0.000) I 27.4 (0.000) I 218.1 (0.000) Table E.3: Math books (full sample) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Less than one book between two 35.7 11.o 34.3 5.5 35.1 8.8 At least one book between two 32.9 35.7 52.4 23.2 41.6 30.6 At least one book per student 31.5 53.2 13.3 71.3 23.3 60.6 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 286 41 7 233 289 519 706 Chi-squared (Prob) 69.5 (0.000) 184.8 (0.000) 211.0 (0.000) 117 Annex E Table E.4: Desks (full sample) Primary I Middle/JSS 1 Total 1988 2003 1988 2003 1988 2003 No desks or tables at all 18.2 1.2 9.0 1.o 14.1 1.I Some, but not enough 70.3 31.4 77.7 31.I 73.6 31.3 Enough for everyone 11.5 67.4 13.3 67.8 12.3 67.6 Total 100 100 100 100 100 100 No. of observations 286 417 233 289 519 706 Chi-squared (Prob) 288.1 (0.000) 159.8 (0.000) 391.O (0.000) Table E.5: Recurrent input index (full sample) Primary 1 Middle/JSS I Total 1988 2003 I 1988 2003 I 1988 2003 Low ( ~ 0 . 5 ) 325.7 1.o 27.7 0.7 32.1 0.8 Medium (0.5-0.75) 49.8 19.2 60.2 28.4 54.5 22.9 High (>0.75) 14.5 79.9 12.1 70.9 13.4 76.2 Total 100 100 100 100 100 100 No. of observations 283 417 231 289 514 706 Chi-squared (Prob) 320.5 (0.000) 203.5 (0.000) 524.5 (0.000) (b) Public schools only Table E.6: Chalk (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Never enough 13.4 7.2 8.8 10.5 11.3 8.6 Occasional shortages 70.6 3.4 72.6 9.7 71.5 6.2 Always enough 16.0 89.4 18.6 79.8 17.2 85.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 262 320 226 248 488 568 Chi-squared (Prob) 336.7 (0.000) 205.9 (0.000) 536.7 (0.000) Table E.7: English books (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Less than one book between two 61 .O 13.1 I 42.1 21.8 1 52.2 16.9 At least one book between two 22.0 17.2 37.3 44.4 29.1 29.0 At least one book per student 17.0 69.7 20.6 33.9 18.7 54.0 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 264 320 228 248 492 568 Chi-squared (Prob) 185.6 (0.000) 24.8 (0.000) 187.3 (0.000) 118 Annex E Table E.8: Math books (public schools only) Primary 1 Middle/JSS Total 1988 2003 1988 2003 1988 2003 Less than one book between two 37.1 8.8 34.6 4.0 36.0 6.7 At least one book between two 34.8 40.6 52.2 23.0 42.9 32.9 At least one book per student 28.0 50.6 13.2 73.0 21.I 60.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 264 320 228 248 492 568 Chi-squared (Prob) 74.3 (0.000) 183.0 (0.000) 216.0 fO.OOO\ Table E.9: Desks (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 No desks or tables at all 18.9 1.6 9.2 0.8 14.4 1.2 Some, but not enough 73.1 35.3 78.5 34.3 75.6 34.9 Enough for everyone 8.0 63.1 12.3 64.9 10.0 63.9 Total 100 100 100 100 100 100 No. of observations 264 320 228 248 492 568 Chi-squared (Prob) 201.I (0.000) 142.2 (0.000) 341.2 (0.000) Table E.10: Recurrent input index (public schools only) Primary Middle/JG Total 1988 2003 1988 2003 1988 2003 Low (<0.5) 33.6 1.6 27.4 4.0 30.7 2.6 Medium (0.5-0.75) 51 .I 17.8 55.3 25.0 53.1 21.o High (>0.75) 15.3 80.6 17.3 71 .O 16.2 76.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 262 320 226 248 488 568 Chi-squared (Prob) 261.4 (0.000) 145.4 10.000) 404.2 (0.000) (c) Panel data Table E.11: Chalk 1988 2003 Never enough 10.8 7.2 Occasional shortages 69.6 8.2 Always enough 19.6 84.5 Total 100 100 Observations 194 194 119 Annex E Table E.12: English books 1988 2003 Less than one book between two 46.9 17.3 At least one book between two 30.6 28.1 At least one book per student 22.4 54.6 Total 100 100 Observations 196 196 Table E.13: M a t h books 1988 2003 ~ Less than one book between two 30.6 7.1 At least one book between two 40.3 30.1 At least one book per student 29.1 62.8 Total 100 100 Observations 196 196 Table E.14: Desks 1988 2003 No desks or tables at all 15.8 1.5 Some, but not enough 76.5 34.7 Enough for everyone 7.7 63.8 Total 100 100 Observations 196 196 (d) Public versus private, 2003 Table E.15: Chalk Public Private Total Never enough 8.6 9.4 8.8 Occasional shortages 6.2 13.8 7.6 Always enough 85.2 76.8 83.6 Total 100.0 100.0 100.0 Total 568 138 706 Chi-squared (Prob) 9.4 (0,009) 120 Annex E Table E.16: English books Public Private Total Less than one book between two 16.9 10.9 15.7 At least one book between two 29.0 18.8 27.1 At least one book per student 54.0 70.3 57.2 Total 100.0 100.0 100.0 Total 568 138 706 Chi-squared (Prob) 12.2 (0.007) Table E.17: Math books Public Private Total Less than one book between two 6.7 17.4 8.8 At least one book between two 32.9 21 .o 30.6 At least one book per student 60.4 61.6 60.6 Total 100.0 100.0 100.0 Total 568 138 706 Chi-squared (Prob) 20.6 (0.000) Table E.18: Desks Public Private Total No desks or tables at all 1.2 0.7 1.I Some, bat not enough 34.9 16.7 31.3 Enough for everyone 63.9 82.6 67.6 Total 100.0 100.0 100.0 Total 568 138 706 Chi-squared (Prob) 17.7378 (0.000) Table E.19: Recurrent index Public Private Total Low ( ~ 0 . 5 ) 2.6 3.6 2.8 Medium (0.5-0.75) 21.o 21.7 21.I High (>0.75) 76.4 74.6 76.1 Total 100.0 100.0 100.0 Total 568 138 706 Chi-squared (Prob) 0.4567 (0.796) 121 Annex E PHYSICALINPUTS (a) Full sample Table E.20: Adequate classrooms (full sample) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Less then half necessary 0.0 0.0 0.0 0.0 0.0 0.0 More than half necessary 22.7 21.3 19.7 17.0 21.4 19.5 Required amount 77.3 78.7 80.3 83.0 78.6 80.5 Total 100.0 100.0 100.0 100.0 100.0 100.0 Total 286 417 233 289 519 706 Chi-squared (prob) 0.2 (0.663) 0.7 (0.412) 0.6 (0.429) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 More than half 28.7 19.9 20.2 19.7 24.9 19.8 Less than half 23.8 12.0 18.9 13.8 21.6 12.7 None 47.6 68.1 60.9 66.4 53.6 67.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 286 41 7 233 289 519 706 Chi-squared (prob) 31.6 (0.000) 2.7 (0.264) 26.9 (0.000) Table E.22: Percentage of classrooms with a chalkboard (full sample) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 None 3.1 0.5 I 2.1 0.7 I 2.7 0.6 Less than half 6.6 1.9 4.3 0.7 5.6 1.4 More than half 20.6 3.6 25.8 3.8 22.9 3.7 All 69.6 94.0 67.8 94.8 68.8 94.3 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 286 417 233 289 519 706 Chi-squared (prob) 76.4 (0.000) 66.3 (0.000) 142.6 (0.000) 122 Annex E Table E.23: Board quality (full sample) Primary 1 Middle/JSS I Total 1988 2003 1988 2003 1988 2003 Poor 9.6 8.2 5.3 9.3 7.7 8.7 Fair 33.0 17.4 38.5 10.7 35.4 14.7 Good 52.8 69.1 49.6 79.9 51.4 73.5 Excellent 4.6 5.3 6.6 0.0 5.5 3.1 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 282 414 226 289 508 703 Chi-squared (prob) 24.8 (0.000) 82.2 (0.000) 81.5 (0.000) Table E.24: Own water supply (full sample) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Yes 16.1 28.3 17.2 31 .I 16.6 29.5 No 83.9 71.7 82.8 68.9 83.4 70.5 Total 100 ' 100 100 100 100 100 No. of observations 286 417 233 289 519 706 Chi-squared (prob) 14.1 (0.000) 13.5 (0.000) 26.4 (0.000) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Yes 7.7 9.8 9.9 16.6 8.7 12.6 No 92.3 90.2 90.1 83.4 91.3 87.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 286 417 233 289 519 706 Chi-squared (prob) 1.o (0.329) 5.0 (0.026) 4.8 (0.029) Table E.26: Physicalindex (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Low (~0.5) 44.7 26.3 35.4 27.0 40.6 26.6 Medium (0.5-0.75) 46.5 53.9 57.5 47.4 51.4 51.2 High (>0.75) 8.9 19.8 7.1 25.6 8.1 22.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 Total 282 414 226 289 508 703 Chi-squared (prob) 31.6 (0.000) 30.3 (0.000) 53.8 (0.000) 123 Annex E (b) Public schools only Table E.27: Adequate classrooms (public schools only) Primary 1 MiddleIJSS I Total 1988 2003 1 1988 2003 I 1988 2003 Less then half necessary 0.0 0.0 0.0 0.0 0.0 0.0 More than half necessary 22.7 21.6 19.7 18.5 21.3 20.2 Required amount 77.3 78.4 80.3 81.5 78.7 79.8 Total 100.0 100.0 100.0 100.0 100.0 0.0 No. of observations 264 320 228 248 492 568 Chi-squared (prob) 0.1 (0.736) 0.1 (0.742) 0.2 (0.661) Table E.28: Classrooms that cannot be used when raining (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 More than half 29.9 22.5 20.2 21 .o 25.4 21.8 Less than half 22.3 11.6 18.4 15.3 20.5 13.2 None 47.7 65.9 61.4 63.7 54.1 65.0 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 264 320 228 248 492 568 Chi-squared (prob) 21.6 (0.000) 0.8 (0.665) 15.2 (0.001) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 None 2.7 0.6 1.8 0.4 2.2 0.5 Less than half 7.2 2.2 4.4 0.4 5.9 1.4 More than half 20.8 3.4 26.3 4.0 23.4 3.7 All 69.3 93.8 67.5 95.2 68.5 94.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 264 320 228 248 492 568 Chi-squared (prob) 61.2 (0.000) 61.4 (0.000) 122.0 (0.000) 124 Annex E Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Poor 10.0 9.1 5.4 10.5 7.9 9.7 Fair 34.9 18.3 38.9 9.7 36.7 14.5 Good 51.3 67.8 49.3 79.8 50.4 73.1 Excellent 3.8 4.7 6.3 0.0 5.0 2.7 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 261 317 221 248 482 565 Chi-squared (prob) 22.1 (0.000) 78.6 (0.000) 78.0 (0.000) Table E.31: Own water supply (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Yes 13.3 21.3 16.7 26.6 14.8 23.6 No 86.7 78.8 83.3 73.4 85.2 76.4 Total 100 100 100 100 100 100 No. of observations 264 320 228 248 492 568 Chi-squared (prob) 6.4 (0.012) 6.0 (0,009) 12.9 (0.000) Table E.32: Library (public schools only) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Yes 5.7 9.7 10.1 16.1 7.7 12.5 No 94.3 90.3 89.9 83.9 92.3 87.5 Total 100.0 100.0 100.0 100.0 100.0 100.0 No. of observations 264 320 228 248 492 568 Chi-squared h o b ) 3.2 (0.074) 3.8 (0.052) 6.5 (0.01 1) Primary Middle/JSS Total 1988 2003 1988 2003 1988 2003 Low (<0.5) 46.7 30.0 35.3 28.6 41.5 29.4 Medium (0.5-0.75) 46.4 53.9 58.4 50.0 51.9 52.2 High (>0.75) 6.9 16.1 6.3 21.4 6.6 18.4 Total 100 100 100 100 100 100 Total 261 317 221 248 482 565 Chi-squared (prob) 31.6 (0.000) 30.3 (0.000) 53.8 (0.000) 125 Annex E (c) Panel data Table E.34: Adequate number of classrooms 1988 2003 Less then half necessary 0.0 0.0 More than half necessary 14.8 18.4 Required amount 85.2 81.6 Total 100.0 100.0 No. of observations 196 196 Table E.35: Classrooms that cannot be used when raining 1988 2003 More than half 29.6 22.4 Less than half 24.5 19.4 None 45.9 58.2 Total 100.0 100.0 No. of observations 196 196 Table E.36: Percentage o f classrooms whit a chalkboard 1988 2003 None 1.o 0.5 Less than half 4.1 2.0 More than half 23.5 3.6 All 71.4 93.9 Total 100.0 100.0 No. of observations 196 196 Table E.37: Chalkboard quality 1988 2003 Poor 8.9 9.9 Fair 35.9 19.3 Good 47.9 66.7 Excellent 7.3 4.2 Total 100.0 100.0 No. of observations 192 192 126 Annex E Table E.38: Water 1988 2003 No 85.7 81.I Yes 14.3 18.9 Total 100.0 100.0 No. of observations 196 196 Table E.39: Library 1988 2003 No 92.9 86.7 Yes 7.1 13.3 Total 100.0 100.0 No. of observations 196 196 (d) Public versus private, 2003 Table E.40: Adequate classrooms Public Private Total Less then half necessary 0.0 0.0 0.0 More than half necessary 20.2 18.1 19.8 Required amount 79.8 81.9 80.2 Total 100.0 100.0 100.0 No. of observations 568 138 706 Chi-squared (prob) 0.3 (0.573) Table E.41: Classrooms that cannot be used when raining Public Private Total More than half 21.8 11.6 19.8 Less than half 13.2 10.9 12.7 None 65.0 77.5 67.4 Total 100.0 100.0 100.0 No. of observations 568 138 706 Chi-squared (prob) 8.9 (0.01 1) 127 Annex E Table E.42: Share o f classrooms with a chalkboard Public Private Total None 0.5 0.7 0.6 Less than half 1.4 1.4 1.4 More than half 3.7 3.6 3.7 All 94.4 94.2 94.3 Total 100.0 100.0 100.0 No. of observations 568 138 706 Chi-squared (prob) 0.1 (0.994) Table E.43: Board quality Public Private Total Poor 9.7 4.3 8.7 Fair 14.5 15.2 14.7 Good 73.1 75.4 73.5 Excellent 2.7 5.1 3.1 Total 100.0 100.0 100.0 No. of observations 565 138 703 Chi-squared (prob) 5.9 (0.1 17) Table E.44: Own water supply Public Private Total Yes 76.4 46.4 70.5 No 23.6 53.6 29.5 Total 100.0 100.0 100.0 No. of observations 568 138 706 Chi-squared (prob) 48.2 (0.000) Table E.45: Library Public Private Total Yes 87.5 87.0 87.4 No 12.5 13.0 12.6 Total 100.0 100.0 100.0 No. of observations 568 138 706 Chi-squared (prob) 0.0298 (0.863) 128 Annex E Table E.46: Physical index Public Private Total Low (<0.5) 29.4 15.2 26.6 Medium (0.5-0.75) 52.2 47.1 51.2 High (>0.75) 18.4 37.7 22.2 Total 100.0 100.0 100.0 Total 565 138 703 Chi-squared (prob) 27.5 (0.000) Table E.47: Frequency o f head-teacher and circuit supervisor activities QUIPS Head-teacher WSD Other schools Total 1 I QUIPS Circuit supervisor WSD Other Total schools Sits in on class Never 4 12 303 319 30 50 1749 1829 Less than once a week 25 20 802 847 40 65 1094 1199 At least once a week 35 65 1388 1488 4 5 70 79 Daily 8 18 333 359 0 0 13 13 Total 72 115 2826 3013 74 120 2926 3120 Never 3 11 233 247 34 53 1673 1760 Less than once a week 37 33 1116 1186 38 62 1199 1299 At least once a week 31 62 1303 1396 2 4 47 53 Daily 1 8 171 180 0 0 7 7 Total 72 114 2823 3009 74 119 2926 3119 Never 0 0 68 68 22 48 1449 1519 Less than weekly 0 0 72 72 47 69 1418 1534 At least once a week 72 115 2684 2871 5 3 60 68 Total 72 115 2824 3011 74 120 2927 3121 Never 9 10 310 329 31 57 1717 1805 Less than weekly 36 54 1214 1304 38 60 1149 1247 At least once a week 27 51 1296 1374 0 3 42 45 Total 72 115 2820 3007 69 120 2908 3097 Discusses career development Never 26 52 1207 1285 48 78 2134 2260 Less than once a month 30 36 1089 1464 23 28 633 684 At least once a month 16 27 522 256 3 14 153 170 Total 72 115 2818 3005 74 120 2920 3114 129 Annex E Table E.48: Frequency o f head-teacher and circuit supervisor activities (percent) QUIPS Head-teacher WSD Other schools Total 1 I QUIPS Circuit supervisor WSD Other schools Total Sits in on class Never 5.6 10.4 10.7 10.6 40.5 41.7 59.8 58.6 Less than once a week 34.7 17.4 28.4 28.1 54.1 54.2 37.4 38.4 At least once a week 48.6 56.5 49.1 49.4 5.4 4.2 2.4 2.5 Daily 11.1 15.7 11.8 11.9 0.0 0.0 0.4 0.4 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Memo: absolute total 72 115 2826 3013 74 120 2926 3120 Prob. value 0.251 0.147 0.090 0.020 Never 4.2 9.6 8.3 8.2 45.9 44.5 57.2 56.4 Less than once a week 51.4 28.9 39.5 39.4 51.4 52.1 41.0 41.6 At least once a week 43.1 54.4 46.2 46.4 2.7 3.4 1.6 1.7 Daily 1.4 7.0 6.1 6.0 0.0 0.0 0.2 0.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Memo: absolute total 72 114 2823 3009 74 119 2926 3119 Prob. value 0.029 0.255 0.214 0.038 Never 0.0 0.0 2.4 2.3 29.7 40.0 49.5 48.7 Less than weekly 0.0 0.0 2.5 2.4 63.5 57.5 48.4 49.2 At least once a week 100.0 100.0 95.0 95.4 6.8 2.5 2.0 2.2 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Memo: absolute total 72 115 2824 187 74 120 2927 3121 Prob value 0.113 0.036 0.000 0.109 Never 12.5 8.7 11.0 10.9 44.9 47.5 59.0 58.3 Less than weekly 50.0 47.0 43.0 43.4 55.1 50.0 39.5 40.3 At least once a week 37.5 44.3 46.0 45.7 0.0 2.5 1.4 1.5 Total 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Memo: absolute total 72 115 2820 3007 69 120 2908 3097 Prob. value 0.114 0.376 I 0.014 0.039 Never 36.1 45.2 42.8 42.8 64.9 65.0 73.1 72.6 Less than once a month 41.7 31.3 38.6 48.7 31.1 23.3 21.7 22.0 At least once a month 22.2 23.5 18.5 8.5 4.1 11.7 5.2 5.5 Total 100.1 100.4 100.0 100.0 100.0 100.0 100.0 100.0 Memo: absolute total 72 115 2818 3005 74 120 2920 3114 Prob. value 0.565 0.331 0.139 0.008 130 Annex F Annex F: Variable Definition Variable 1 Name I Data and construction 1 Physical: quantify Total number of classrooms Required classrooms I CROOMS REQCROOM i All classrooms at the school (including unusable) Sum of number of classes in each grade, where the number is divided by 2 if that grade is split shift I Adequate number of ADQCROOM CROOMS (less those which cannot be used at any time) classrooms divided by REQCROOM ADQCROOMC Categorical version of ADQCROOM: 1 (0-.49), 2 (0.50- % of classrooms with chalkboard I I .89), 3 (>= 0.9) No. of classrooms with chalkboardlCROOMS Own water supply WATER If the school has own water supply No=l, Yes=l Librarv LIB If the school has a library No=l, Yes=l Physical: quality % of classrooms that cannot RCROOMS Classrooms that cannot be used at all plus those that be used when raining i RCROOMSC 1 cannot be used when raining divided by CROOMS Categorical version of RCROOM (reversed): 1 (>= 0.50), 2 (0.01-.049), 3 (0) Classes held in shared CLASSSHA School average (from teacher questionnaire) of if have to classrooms* teach in a shared classroom Size of internal walls* CLASSWLA School average (from teacher questionnaire) of if have fall size internal walls Noise disruption* CLASSNSA School average (from teacher questionnaire) of if class is Board quality 1 BOARDQUAL I disturbed by external noise School survey respondent assessment of overall chalkboard quality School average of teacher BQUALTA Average of teacher responses as to quality of chalkboard assessment of board quality* in the class in which they teach School average of teacher BQUALSZA Average of teacher responses as to the size of assessment of if adequate chalkboard in the class in which they teach board size School average of teacher BQUALCLA Average of teacher responses as to how easily the assessment of if board can chalkboard in the class in which they teach can be be cleaned cleaned Type of Wateristorage WATTYPE Type of water supply from piped water to reservoir/other. Recurrent: quantity Chalk availability CHALK School survey assessment of chalk availability English textbooks-pupil ratio EBOOKR Sum of English books available in each grade divided by total enrolments (ENR) EBOOKC Categorical version of EBOOKR: 1 (0-0.49), 2 (0.50- 0.99), 3 (>=I) Mathematics textbooks-pupil MBOOKR Sum of math books available in each grade divided by ratio total enrolments (ENR) MBOOKC Categorical version of MBOOKR: 1 (0-0.49), 2 (0.50- 0.99), 3 (>=I) Writing places-pupil ratio I DESKS Total writing places (sum across ps6q12bl-b6) divided to adjusted enrolments (ENRA) 131 Annex F Variable Name Data and construction DESKSC Categorical version of DESKS: 1 (0-.09), 2 (0.10-0.89), 3 (>= 0.90) Seating places-pupil ratio* SEATS Total writing places (sum across ps6ql1a1-b6) divided to adjusted enrolments (ENRA) SEATSC Categorical version of SEATS: 1 (0-.09), 2 (0.10-0.89), 3 I>= 0.90) Teachers and teaching methods Number of teachers NOTEACH This variable is problematic since we do not want to include detached heads. They should not be included in the roster, so the variable is calculated by summing the number of teachers in the roster. However, in some cases it appears the head was included though detached. These cases have been adjusted by inspection. ~~~ Required number of teachers REQTEACH Calculated in the same way as REQCROOMS but without the adjustment for split shift. Adequate number of ADQTEACH NOTEACH/REQTEACH teachers % trained teachers TEATRAIN Proportion of teachers who have teacher traininq. Absenteeism ABSENT Proportion of teachers absent for reasons other than sickness (note reference period longer in 1988). Teacher test scores TSCORE Simple average of teacher‘s scores on English, math and Raven’s tests. Head’s assessment of if a GOODTEACH Head’s subjective assessment good teacher* GOODTEACHA School average of GOODTEACH Teaching methods* Teacher morale (subjective)* TMORALE If teacher enjoys being a teacher (No=l, Yes=l); ts6q5 TMORALEA School average of TMORALE TWCOND Teacher assessment of working conditions (1 Very Poor, 2 Poor, 3 Good, 4 Verv Good) TWCONDA School average of TWCOND Frequency of homework* AVEHOME How often homework is set per week, averaged over all three subjects (or number for which data available) Frequency student work AVEINSPECT How often students’ work is inspected per week, inspected* averaged over all three subjects (or number for which data available) Frequency student work AVEASSESS How often students’ work is assessed per week, assessed* averaged over all three subjects (or number for which data available) Monitoring of student STMONITOR Simple average of AVEHOME, AVEINSPECT and performance* AVEASSESS STMONITOR School average of STMONITOR School management SMC* SMC If there is a SMC at the school SMCMEET If the SMC has met in last month (No: 0; Yes: 1) I SMCHELP 1 If the SMC has provided help to the school during the last year (No: 0; Yes: 1) 132 Annex F Variable I Name Data and construction PTA I PTA If there is a PTA at the school PTAMEET If the PTA has met in last month (No: 0; Yes: 1) SMCHELP If the PTA has provided help to the school during the last year (No: 0; Yes: 1) SPAM* I SPAM SPAMPLAN If the school has had a SPAM in the previous year (No: 0; Yes: 1) Frequency of visits by circuit CSVlSlT How often the circuit supervisor (school inspector in supervisor 1988) has visited the school in the last 12 months Activities of head teacher I CSVISITC HTCLASS Categorical version of CSVIST: 0 (Never), 1 (1-5 times); 2 (6-11 times); 3 (12 or more times) Frequency with which head teacher (circuit supervisor) and circuit supervisor* CSCLASS sat in on class (per day). HTCLASSA and CSCLASSA are school averages. HTCLASSA CSCLASSA HTSMP, Frequency with which head teacher (circuit supervisor) CSSMP inspected students’ work (per day). HTSMPA and HTSMPA, CSAMPA are school averages. CSSMPA HTLLESS, Frequency with which head teacher (circuit supervisor) CSLLESS looked at lesson plans (per day). HTLLESSA and HTLLESSA, CSLLESSA are school averages. CSLLESSA HTDLESS, Frequency with which head teacher (circuit supervisor) CSDLESS discussed lesson plans. HTDLESSA and CSDLESSA are HTDLESS, school averages. CSDLESS HTDCD, Frequency with which head teacher (circuit supervisor) CSDCD discussed career development. HTDCDA and CSDCDA HTDCDA, are school averages. CSDCDA Pupil data Total enrolments ENR Total enrolments Adjusted enrolments I ENRA Enrolments adjusted for split shifts. Note: * not available from 1988 data 133 Annex G Annex G: Analysis o f Test Scores INTRODUCTION 1. In 1988/89 cognitive tests were administered to a sub-sample (1,594 households) o f the 3,200 households interviewed by the second round o f the Ghana Living Standard Survey (GLSS2).’ The 2003 survey used exactly the same tests as those used in 1988/89, and included also a local language test. The second part o f this annex describes the tests, discusses their limitations, and evaluates their reliability. The third part presents a descriptive analysis o f the tests results and the fourth part builds an econometric model o f the determinants test outcomes. DESCRI P TO I N OF THE COGNITIVE TESTS USED 2. A total o f seven tests were administered to members o f the 1,740 households interviewed in 2003. These tests are: 0 Raven’s Colored Progressive Matrices test 0 Short local language test 0 Short English test 0 Short math test 0 Advanced local language test 0 Advanced English test 0 Advanced math test 3. The Raven’s Progressive Matrices test is a measure intellectual ability intended to be independent o f education and experience.’ The test consists o f a puzzle with a missing piece that the person taking the test has to find among a choice o f 6 possible pieces. The test comprises 36 questions divided into three parts o f 12 questions each. The short English and math test were originally designed as a screening device to prevent people with very l o w skills fiom attempting the more advanced tests (Glewwe 1999). In 1988, given the l o w performance in the advanced test, the short test results became part o f the tests score analysis, and the same procedure was followed for this study. The short English reading test consists o f a few English sentences that make a short story. The person is required to read the sentences and then answer eight multiple-choice questions to measure the person’s understanding o f the story. The Short Mathematics test consists o f eight arithmetic operations (addition, subtraction, multiplication, and division), o f increasing difficulty. The advanced English and mathematics test are the same tests used by a study o n educational achievements conducted in Kenya and Tanzania in 1980. These tests were designed by the Educational Testing Service, based in 1. Specifically, the tests were conducted in half o f the 170 clusters covered by the survey. 2. In fact, Raven’s test results are likely to be influenced by levels o f schooling and household environment. In the second part o f the annex we w i l l show how we can isolate an innate ability component from individual scores using the same methodology used in Glewwe (1999). 3. The short English and mathematics test are reproduced in Annex A. 134 Annex G Princeton, N e w Jersey, based on school-leaving examinations o f primary and secondary Kenyan and Tanzanian students (Knight and Sabot 1990). The advanced Englishtest contains 29 multiple-choice questions. Some o f the questions, as in the short English test, are based o n the understanding o f a short story. In other cases, the person taking the test has to select the correct word from a choice o f four in a sequence o f sentence^.^ The advanced mathematics test consists of 36 questions o f increasing difficulty. The questions cover the knowledge o f all basic mathematics, including arithmetic, fractions and proportions, decimal numbers, real problem solving, geometry, equations, and algebra. The local language tests, short and advanced, are a new feature o f the 2003 survey. They were designed by the Department o f Linguistics and Ghanaian Languages at the University o f Ghana in association with the Department o f Linguistics at the University College o f Education at Winneba. They consist o f a translatiodadaptation o f the English tests to the most commonly spoken Ghanaian language^.^ 4. All household members aged between 9 and 55 were eligible to take the tests.6A maximum time for the completion of the tests, ranging from 10 minutes for the short English to 30 minutes for the advanced mathematics test, was applied. Despite the large number o f tests and the length o f time required for completion, the rate o f reported refusal on the short tests was only 5 percent. On average, however, only 50 percent o f individuals taking the short test were able to take the advanced ones, and many eligible individuals did not take the short tests because they found them too difficult. The 2003 survey screened eligible individuals with the use o f flash cards, thus reducing the number o f people taking the test.7 This procedure has some implications for the measurement o f mean test scores that are discussed below. Details o f the procedure adopted for the administration o f the tests can be found in the Test Administrator’s Instruction Manual designed for the survey which i s available on the study website. Limitations and Reliability o f the Tests 5. There are three limitations to the use o f the set o,ftests used in this study. First, the use o f numeracy and literacy tests focuses o n a narrow range o f child development, ignoring other aspects covered in the school A second limitation was the use o f a language (English) for the literacy test, which is the first language for only a very small minority o f Ghanaian children. This problem has been partially corrected by the introduction o f a test in Ghanaian languages, that allowed children to be tested in both languages. This fact will nAppendix A. 4. A sample o f the questions of the advanced Englishand mathematics tests can be found i 5. The English tests were translated into 12 Ghanaian languages, that together cover 80-90 percent o f all languages presently spoken in Ghana. These languages are: Akuapem, Asante-Twi, Dagaare, Dangme, Ewe, Fante, Ga, Gonja, Kasem, Kusaal, Nzema, and Wale. 6. Teachers from the surveyed schools were also administered the tests, but only the advanced math and English tests, the Raven’s test and the local language test using the local language officially designated for that school if they said they were competent in it. 7. The flash card for the language tests was a sentence taken from the text used inthe test itself. The flash card for the math test was a simple addition or subtraction. 8. This fact may give an advantage to private schools which some say focus o n these core slulls to the detriment o f other subjects. 135 Annex G particularly affect the test scores o f younger children since until recently local language has been the medium o f instruction until the end o f third grade. A third problem is that the mathematics test was designed in English, thus giving an advantage to those individuals that are proficient in this language. Test administrators in the field were advised to translate the mathematics questions whenever necessary. However, we performed regression analysis, not reported here, which shows that mathematics test scores were to some extent dependent o n English proficiency. The possibility o f re-designing the math test in local languages was rejected for the same reason that a broader range o f test instruments was not used. That i s that comparability was needed with the 1988 tests. This comparability is central to the study design, and was preserved at the expense o f using test instruments known to be imperfect. 6. In order to assess the reliability o f the tests ‘Cronbach’s alpha’ was calculated. Cronbach’s alpha is a commonly used measure o f the quality o f test instruments, which captures their intemal consistency. The statistic i s calculated using the responses to individual questions (i) for different individuals 6). If a test is internally consistent, then the scores across questions o f two individuals should be reasonably correlated. People doing w e l l will do w e l l o n the same questions and badly o n the same ones. People who do badly should get the same questions wrong as other people doing badly. Cronbach’s alpha is based o n the correlation coefficient between the test scores o f all possible pairs o f test takers. T o calculate the statistic a random sample o f 100 people was used for each test. A value o f the alpha statistic o f 0.7 and above is considered to be an indication o f a reasonable degree o f consistency. The Raven’s test scores tumed to be very high, and all other tests also have coefficients equal to Table G.1: Values of the Cronbach’s alpha test or in excess o f 0.7 other than the short Short Advanced local language test, which at 0.64 falls English 0.72 0.79 not far below the threshold (Table G.1). Mathematics 0.75 0.82 Local language 0.64 0.70 DESCRIPTIVEANALYSISOF THE TESTS Raven 0.94 n.a. 7. Before describing test performance and comparing scores o f 1988 with those o f 2003, the computation o f test scores used for the analysis is discussed. Short tests were composed of eight questions, and only individuals scoring five or more were allowed to take the advanced tests o f any type. Additionally, a screening mechanism was used, whereby the person was invited to take the short test only if able to read very short sentences or able to solve a simple arithmetic operation displayed o n a flash card. As a result, not all individuals were administered the entire set o f tests. This procedure results in a problem in the computation o f the tests scores that i s resolved in the following section. Censoring and screening 8. As in 1988, the advanced tests proved to be very difficult, and the majority o f the persons involved in the exercise did not score more than five o n the short tests. On the other hand, for people with higher level o f education, the short tests were very easy, and large number o f persons scored eight, which i s the maximum score. The latter can be seen as a problem o f censoring from above. T o clarify this point, assume that the English test scores are a measure o f the latent variable English language ability. However, the test has eight questions so that there i s a maximum score o f eight. If the data had not been censored at 136 Annex G eight, then those with higher ability would have scored more than eight. But they could not do so, so that scores are bunched (censored) at eight. This bunching constitutes a problem when w e want t o compare mean scores o f different groups or surveys, because simple means o f test scores will under-estimate the real difference in ability between the groups. A formula can be applied that adjusts the simple means for censoring.’ 9. As a result of applying the screening process described above, many individuals were reported as not having taken the test in 2003 because it was “too difficult.” On the other hand, in 1988, when n o screening was applied, few people said the test was too difficult, but many scored zero o n the short tests (especially in Englishwhen 14 percent scored zero compared to only 1.3 percent in 2003; for math these figures are 2.5 and 0.3 percent, respectively) -these people would have been screened out using the 2003 procedure. Ignoring them will overestimate mean scores o f 2003 respect to those o f 1988, since those who were scoring zero in 1988 are likely t o having been screened out o f the sample in 2003. The data suggest that this problem is more serious for the short English scores than for the short math. In order not to lose information, the option o f dropping all zero scores from the 1988 or both data sets was rejected (a solution that would have introduced other problems o f sample selection). Instead, w e assigned a score of zero to those individuals who did not take the English test because i t was too difficult, and a random score between 0 and 4 to those who were not able to take the mathematics test. lo Combined test scores 10. In the next section a multivariate model i s presented explaining test scores by individual, household, and school characteristics. The test scores used in the regressions are a combination o f the short and the advanced tests scores. Since short and advanced data were missing for many people it was necessary to impute the missing test scores. 11. The short test scores, ranging from 0 to 8, show little variability across the sample, and an obvious solution i s to add the short test scores to the advanced ones to get a combined 9. If observation are normally distributed, the observed censored mean i s (Greene, 2000): E [ x ] = @a+ (1- @)(p+ o&)) where x i s the variable of interest, CY i s the truncation point (truncation i s ‘from above’ in this case), p i s the uncensored mean and fl i s the uncensored variance. CY) i s the inverse Mill’s ratio and (for the censoring fi-om above case) i s defined as: A(a) = -q(a)/@(a) where r $ and are the density and the cumulated density o f the normal distribution and CY is: a =(a - p ) / o This set o f equations can be solved for p and a, thus producing the uncensored mean and variance. 10. The data show that a score o f zero was far more common for English than math. This make sense since people unable to read the text would simply give up and not answer any questions. In math, on the other hand, especially if the questions are read to them, people are likely to attempt all questions, at least guessing the more difficult ones. 137 Annex G test score. However, not all the children took the advanced test, because they scored less than 5 o r found the test too difficult. Also, some people who took the advanced test skipped the short test as being too easy. Various options were examined in order to impute scores, and the following were adopted. To impute advanced test scores, the advanced scores were regressed o n the short scores o f reading, math, and the Raven's test. The resulting predicted values were used to calculate scores for who did not take the advanced test." Similarly, regressions were estimated o f short tests scores o n the advanced test scores o f reading and math and w e calculated the predicted values in order to impute values for the missing short tests scores. l2For the latter model a tobit specification was used correcting for both right and left censoring. Overview o f test score outcomes 12. Tables G.2 and G.3 show the average test scores for all sample individuals and for primary graduates o f 2003 and 1988. Tests marked by a star have been corrected for right censoring as described above. The last columns o f the table report t -statistics and p-values o f the difference in the means between the two surveys. The data show a significant improvement in a l l test scores for both groups other than the Raven's test, which has increased for primary graduates only. Table G.2: Average tests scores: whole sample 1988 2003 t-stat p-value Raven's 19.4 19.4 0.1 1 0.914 Short English* 6.2 6.6 3.75 0.000 Short math* 5.5 5.9 8.16 0.000 Short local* ... 6.4 Advanced English 12.3 13.2 4.16 0.000 Advanced math 8.7 10.1 6.93 0.000 Advanced local ... 15.5 Combined English 17.7 19.2 5.28 0.000 Combined math 14.5 16.2 6.26 0.000 Combined local ... 21.I * Corrected for right censoring. 11. Negative predicted values were set to 0. 12. Predicted values larger than 8 were set to 8. 138 Annex G Table G.3: Average tests scores: primary graduates under 15 1988 2003 f-stat p-value Raven’s 20.2 22.3 2.87 0.004 Short English* 4.7 7.0 5.87 0.000 Short math* 5.2 6.3 3.42 0.000 Short local* ... 6.2 Advanced English 10.2 14.5 6.88 0.000 Advanced math 7.2 9.7 5.04 0.000 Advanced local ... 14.1 Combined English 14.5 20.8 7.72 0.000 Combined math 12.7 15.9 5.44 0.000 Combined local ... 19.8 Corrected for right censoring. Test score improvement by income group 13. The improvement in test scores can be observed for children from all income goups (Table G.4). For primary students the improvement has leveled the performance between children fiom different backgrounds in English, but with the opposite effect for math. The latter is also true for JSS, whereas for English scores for JSS students the benefit has been uniform. Table G.4: Test scores by schooling and income tercile 1988 2003 I 11 Ill I 11 Ill Primary school students I Raven’s 15.7 15.8 16.9 15.9 15.6 17.4 Short English 0.8 1.7 3.7 4.3 4.8 5.5 Short math 3.4 3.7 4.0 4.0 3.8 4.5 Advanced English 3.4 8.6 10.2 12.3 12.3 14.0 Advanced math 5.2 4.9 6.0 14.0 11.7 15.3 JSS students Raven’s 20.2 19.6 21.8 20.1 21 .I 24.3 Short English 4.0 4.0 4.8 6.1 6.3 6.6 Short math 5.1 4.9 5.3 5.8 5.7 6.0 Advanced English 10.5 12.0 12.3 15.0 15.4 17.2 Advanced math 7.2 7.5 8.4 13.9 16.5 20.0 MODELING TEST SCORES 14. In this section a model is defined and estimated o f children’s test scores. Information from both surveys is used in order to detect determinants o f changes in test scores over time. The interest i s to assess the impact o f school quality o n students’ achievements using household survey data, in order to control for the effects o f individual and household 139 Annex G characteristics. The dependent variables used in the model are the combined results o f the mathematics, English, and local language tests described in the previous section. The following section describes the sample and discusses the selectivity problem from the choice o f the sample. A listing o f variables determining test scores follows, and finally the regression results are presented. Sample selection 15. The sample used in the regressions consists o f all children aged between 9 and 15 who have recently attended or are currently attending school in the locality o f residence. Only children with at least three years o f schooling are included. This sample i s not a random sample, because only children either with at least three years o f schooling (1988) or able to read the flash cards (2003) took the tests. Other children did not take the test as a result o f refusal or absence. Children that were purposely or incidentally excluded from the sample can be grouped into the following three categories. 16. Eligible Children N o t Tested. Some 25 percent o f the eligible children did not take the tests. About one-fifth o f these did not do so because they found the test too difficult (mostly in 2003). This problem has been corrected by randomly assigning a l o w grade (math) or a zero grade (English) to those children with at least three years schooling. Another 20 percent o f these children were in school at the time o f the exercise, 20 percent were traveling, 5 percent were i ll,5 percent refused to take the test, and 30 percent did not take the test for other unspecified reasons. In general, the exclusion o f eligible children from the exercise seems to have operated randomly. There might be some concern, however, that children “traveling” and those not taking the test for “other” reasons could share common and distinctive characteristics. For example, they could be workers, and thus have less education. 17. Children Living Elsewhere. Child fostering is very common in Western Africa. Nearly 50 percent o f the sample children are “foster” children, in the sense they are not living in their household o f origin. Parents can use fostering for reasons as different as sending children to better schools, reduce the burden o n household resources, exploit opportunities in more developed areas, and strengthening kinship ties (Lloyd and Gage-Brandon 1994). Though households o f origin o f foster children may w e l l be poorer o n average than other households, foster children are not necessarily sharing common characteristics, precisely because the reasons at the origin o f the “fostering” choice are so different. Additionally, while 25 percent o f children could not be interviewed because they were “fostered out,” another 25 percent were interviewed because they were “fostered in.” Foster children are therefore largely represented in our sample. 18. A more serious problem is whether it i s possible to relate foster children behavior to the characteristics o f the fostering household. I t could be argued that the household o f origin makes choices for these children, rather than the household o f residence (Glewwe 1999) so that the relevant household characteristics are missing for children who are fostered in. A priori this hypothesis appears most appropriate for the schooling decision, whereas for test scores the household o f residence may be the most relevant. The hypothesis that it i s true parental characteristics can be tested by interacting a dummy for foster children (F=1 if fostered in, 0 otherwise) with household characteristics (X). The coefficient o n FX will be insignificant if 140 Annex G household characteristics affect fostered children in the same way as they do non-fostered children. If i t is the characteristics o f the “true household” that matter then the coefficient will be equal and opposite to that o n X. 19. Children Attending School. The sample chosen for the estimation implies that children who never attended school, or who left school before completing primary (or the third year of primary depending on the sample), are not considered. In 2003, 11 percent o f children aged between 10 and 20 had never attended schools, and 10 percent o f children aged over 20 starting primary school never achieved grade six. The same percentages are considerably higher for the children surveyed in 1988. These children are likely to have common characteristics that prevented them form attending or completing school. However, the quality of the schools available and the j o b opportunities also matter. In Appendix I,using a complete set o f household and locality explanatory variables, a model o f school achievement i s estimated whose results are used here to correct for the selectivity bias caused by the sample being dependent upon school attainment. Selectivity Adjustments 20. The selection o f a non-random sample can bias the regression results. The presence o f this bias can be tested for when necessary by including a sample selection term in the test score regression. T o clarify the problem, suppose that cognitive tests are administered to the sample of children graduated from primary school. In order to assess the importance o f school quality, a regression is estimated o f test scores using a school quality index as a regressor, resulting in a positive coefficient on this variable. But suppose that only wealthier families can afford for their children to complete primary school. Also suppose that children from wealthier families are better nourished, do not work, and that this improves their mental development and their performance at cognitive tests. A s a result, w e might erroneously attribute good test scores t o school characteristics, while they are, at least in part, determined by factors related to household wealth. 21. Since selection bias in the description given above can be seen as an omitted variable problem, the presence o f the bias should be detected by a test o n omitted variables such as the Ramsey test. If a selectivity bias i s found a correction can be made by modeling the selection o f children to be in the sample. 22. Algebraically, the equation o f interest is: yi = pxi + Ei (1) where y j i s the test score for child i, xi is the quality index o f the school attended by the child and /3 i s the parameter to be estimated. But only a selected number o f children enter equation (l),since many children never go to school or drop out before reaching a given grade. We can use the entire sample o f children and model school attendance based o n a series o f child and household characteristics. This i s called the ‘selection equation’ : z; = yjc.wji + u i I 141 Annex G where z: i s a variable defining whether the child is attending school or not and the wji are a set of explanatory variables. The variable yi i s observed only when : z i s larger than zero. Equation (1) corrected for selectivity i s thus the expectations o f y i conditional o n z: being larger than zero: '[Y, ; '1 )o] PXi + P n l i Qu )+ vi (3) where X. (04$ i s the inverse M i l l ' s ratio obtained from the selection equation (Greene 2000). This ratio i s usually derived after running a probit o f the selection equation and takes the form: where (b and 9 are the density and the cumulated density o f the normal distribution, and the Xfl are the predicted values o f the selection equation. A significant ,f3~ i s a test for the presence o f selection bias, and at the same time a correction o f the estimates o f the equation (1) for that bias. In the case o f the ordered probit achievement regression (used in Annex I ) the inverse M i l l ' s ratio for the children having attained at least a given grade is: 1- q(cutofl2 - xp) A= 1- @(cutoff2 - xp) where the cutoff i s the threshold used for the sample selection, for example the achievement o f at least grade three. Explanatory Variables 23. The explanatory variables used in the model are the characteristics o f the child, o f the household, and o f the school attended. The variables are listed in Annex F with their description (variable with an * are not available for 1988). Individual variables Sex: (dummy) child sex Age: (continuous) age o f the child in completed years Schooling: (continuous) number o f completed years o f s c h ~ o l i n g ' ~ Order: (continuous) child's birth order Siblings: (continuous) number o f alive siblings Ability: (continuous) estimated innate ability Ability missing: (dummy) children whose innate ability is missing 13. Years o f schooling are derived from the data o n highest completed grade, as the number o f years necessary to achieve a given grade. The transformation o f grades into years followed the GLSS2 Basic Information Document. 142 Annex G 24. The ability variable was estimated Table G.5: Innate ability regression (1988) using the model formulated by Glewwe Coefficient t-sfatistic (1999). This model regresses Raven's test Age 0.36 6.19*** scores o n the age, sex, and years o f schooling Age squared 0.00 -4.81*** o f the child and o f the parents. The model Education 0.69 5.19*** predicts Raven's scores for children and their Education squared 0.01 2.19** parents at the same time, calculating household fixed effects. The ability variable Father's education -0.01 -1.21 is obtained as the s u m o f the estimated Mother's education -0.12 -0.98 household fixed effect and the error term, Sex 0.09 0.78 thus assuming that children inherit their Sex*age 0.31 0.44 innate ability from their parents. Tables G.5 Sex*education -0.06 -2.49** and G.6 present the results o f the regression Constant -0.29 -3.64*** run o n the 1988 and 2003 samples. The Observations 1732 results are very similar, and the only F-statistic 58.41 differences are the significance o f the age R square 0.32 square term and the coefficient estimates o f years o f schooling and gender. It seems that Table G.6: Innate ability regression (2003) gender and older age have become less Coefficient t-statistic important in explaining poor performance, Age 0.25 3.84*** while the effect o f additional years o f Age squared 0.00 -2.52** schooling has increased. Education 1.14 7.54*** Household variables Education squared -0.01 -1.06 Father's education -0.01 -1.85* Coast Forest and Savannah: (dummies) Mother's education -0.02 -0.14 three main agro-ecological zones o f the Sex -0.17 -1.22 country excluding Accra Sex*age 0.21 0.28 Rural: (dummy) residence in rural areas as Sex*education -0.01 -0.49 defined by the 1984 and 2000 demographic Constant -0.21 -2.23*** census respectively Observations 1716 Mother's education: (continuous) completed F-statistic 35.45 years o f schooling of the mother R square 0.23 Father's education: (continuous) completed years o f schooling o f the father Per capita expenditure: (continuous) logarithm o f household per-capita expenditure. Expenditure values o f 1988 were actualised to 2003 using the consumer price index "Parents meeting the teacher: (dummy) the parents are regularly meeting the child's teacher in order to discuss progress in school *PTA: (dummy) membership o f any o f the household member o f local PTA (Parents and Teachers Association) School variables Index o f recurrent inputs: this an index o f recurrent inputs described in Annex D. It includes school availability o f books, writing places and chalk. 143 Annex G Index o f physical inputs: this an index o f physical inputs described in Annex D. I t includes quantity and quality o f classrooms and blackboards, availability o f water and library. *Classrooms with internal walls: (continuous) school average o f classrooms with fbll size internal walls *Noise disruption: (continuous) school average o f outside noise disturbing classes *Private: (dummy) whether the school attended is private Teacher education: (continuous/categorical) average number o f teacher s years o f schooling School management *SPAM plan: (dummy) school had a SPAM in the last year whose plan was actually carried out Visits o f circuit supervisor: (continuous/categorical) number o f inspection visits by the circuit supervisor Other Lambda: (continuous) i s the inverse Mill’s ratio obtained form the attainment regression reported in Annex I. 25. The set o f test score equations -English, math, and local language -might appear to comprise a system o f seemingly unrelated regression equations (SURE), meaning that the error terms between each equation are correlated with one another. Using S U R estimation rather than OLS improves the efficiency o f the estimates. However, S U R estimation requires the sample to be the same for each equation, so including the local language test reduces same size quite considerably. Moreover, there is n o gain if the regressors are the same in each case. Whilst it might be thought that there could be some variation in the regressors (math books for math scores etc.), these distinctions did not prove good ones to keep. Results are first presented (Table G.7) for English and math. The local language estimates, which gave rather different results, are then discussed. 26. I t proves quiet difficult to get good results from the pooled data. I t is always the case that schooling is positively and significantly related with higher test scores - and t h i s remains so even i f j u s t one or two years o f schooling are included in the model. However, interacting school quality variables with years o f schooling does not yield good results. A few of the school qualities have a “shift effect” o n test score outcomes, notably math textbook availability is significant in some, but by n o means all, model specification^.'^ A high pupil-teacher ratio i s detrimental to English test scores (though appears good for maths scores in JSS), and being a beneficiary o f the WSD program improves them.15 Private school students perform significantly better in English though not in math, which i s an unsurprising 14. The availability o f math and English books i s highly correlated. However, if Englishbooks are included and math books dropped the coefficient i s not significant. 15. The pupil teacher ratio i s entered as two dummy variables (low and high) rather than a continuous variable since evidence from other studies suggests that the ratio has no impact over a large range, but very small classes can be good and very large ones detrimental. 144 Annex G result as English i s the medium o f instruction in private schools. Having own water supply i s robustly significant. Table G.7 (a): Test score determinants: pooled data children in primary school (OLS) Math English Coefficient z-stat Coefficient 2-stat School variables Years schooling 0.92 2.39 ** 1.21 2.67 *** Math books 0.46 1.22 0.11 0.24 Classrooms can be used when raining 0.01 0.66 -0.01 -0.84 Water supply 1.28 1.44 2.46 2.48 ** Board quality 0.59 1.30 -0.67 -1.28 Teacher test score -0.15 -1.26 -0.10 -0.78 Private school 0.42 0.43 2.89 2.62 *** High PTR -1 .oo -0.95 -4.37 -3.68 *** PSD dummy -0.62 -0.46 2.89 1.98 ** WSD dummy 0.83 0.50 3.14 1.67 * QUIPS dummy -2.17 -1.48 -0.98 -0.61 Community variables Forest -1.27 -1.90 * -1.73 -2.21 ** Savannah -0.29 -0.25 -2.30 -1.73 Rural -0.21 -0.34 -1.oo -1.39 Child characteristics Age 0.29 1.44 0.26 1.10 Sex -0.29 -0.52 -0.20 -0.30 Ability 0.35 5.95 *** 0.49 7.34 *** Ability missing 4.09 4.45 *** 7.96 7.50 *** Household variables Mother’s education 0.13 2.00 ** 0.23 3.26 *** Income 0.07 0.12 1.41 2.11 ** Other Survey dummy 3.56 4.12 *** 8.32 8.08 *** Selectivity correction -1.25 -1.02 0.41 0.28 Number of obs. 331 298 R2 0.29 0.57 145 Annex G Table G.7 (b): Test score determinants: pooled data children in middle/JS school (OLS) Math English Coefficient z-s fat Coefficient z-stat School variables Years schooling 2.13 5.29 *** 3.25 5.95 *** English books 0.16 0.32 0.29 0.43 Math books 0.55 1.06 -0.47 -0.67 Classrooms can be used when raining -0.01 -0.91 -0.01 -0.61 Water supply 0.39 0.48 1.03 0.97 Board quality -0.21 -0.44 0.51 0.81 Teacher test score 0.10 0.82 0.08 0.48 Private school -0.83 -0.60 1.29 0.69 High PTR 2.99 1.90 * -0.33 -0.16 WSD dummy -1.81 -0.99 3.57 1.50 QUIPS dummy 2.41 1.06 2.86 0.97 Community variables Forest -1.26 -1.83 * -0.54 -0.59 Savannah -0.28 -0.23 -0.32 -0.20 Rural -1.41 -2.07 ** -2.17 -2.32 ** Child characteristics -0.99 -3.46 *** -1.16 -3.02 *** Age Sex -2.39 -4.08 *** -1.31 -1.64 * Ability 0.24 4.20 *** 0.42 5.55 *** Ability missing 2.39 2.62 *** 4.21 3.36 *** Household variables Mother's education 0.17 2.73 *** 0.16 1.95 ** Income -0.72 -1.23 -0.18 -0.22 Other Survey dummy 3.82 4.29 *** 8.44 6.99 *** Selectivity correction 1.30 0.60 0.68 0.19 Number of obs. 272 250 R2 0.45 0.58 146 Annex G Table G.8 (a): Test score determinants, primary schools 2003 (OLS) Math score English score Coeff z Coeff z School characteristics English books -0.35 -0.72 -0.67 -1.04 Math books 1.04 1.81 * 1.27 1.78 * Physical Index 4.96 1.92 * 4.79 1.61 Low PTR 2.22 2.05 * -0.57 -0.39 High PTR -0.51 -0.45 -4.15 -3.05 *** Teachers speak local language 2.65 1.90 * -0.21 -0.13 Teachers’ discipline 1.27 2.1 1 ** 0.56 0.91 Teaching methods -5.77 -1.73 * -4.54 -1.13 Time on task 0.07 1.66 * 0.03 0.60 Board easy to clean -0.14 -0.09 -2.34 -0.93 Classrooms have internal walls 1.94 1.45 1.22 0.61 Class disrupted by noise -1.22 -0.77 -1.60 -1.00 Private school 0.53 0.33 4.05 1.90 * PSD dummy -1.23 -1 .I 1 2.95 2.52 ** WSD dummy 0.61 0.38 3.51 1.95 * Circuit supervisor discuss lesson plans -3.17 -0.17 15.16 0.68 Head teacher sits in on lessons 3.29 I .87 * 3.92 1.69 * Student monitoring 0.41 0.58 -0.89 -1.20 School had a SPAM 0.00 0.00 0.52 0.78 Teachers morale 1.26 1. I O 0.38 0.25 Students indiscipline -0.80 -1 5 5 -0.91 -1.85 * Community characteristics Forest -2.49 -2.78 *** -1.64 -1.45 Savannah -0.70 -0.59 -1.97 -1.50 Rural -1.96 -2.52 ** -2.70 -2.79 *** Child characteristics 0.52 1.87 * 0.26 0.77 Age Sex 0.05 0.08 0.10 0.12 Years of schooling 1.I9 2.56 ** 1.58 2.78 *** Ability 0.38 4.40 *** 0.50 6.95 *** Ability missing 5.97 3.34 *** 7.22 3.34 *** Fostered in 14.89 0.49 23.30 0.73 Fostered*income -1.12 -0.56 -1.50 -0.70 Household characteristics Mother’s education 0.17 1.94 * 0.31 2.67 ** Income 0.29 0.36 2.03 2.24 ** Parent in PTA 1.33 1.13 -0.88 -0.64 Parent met with teacher -1.80 -2.40 ** -0.72 -0.64 Other variables Selectivity correction 0.97 0.54 3.48 1.74 * Constant -18.33 -1.27 -37.97 -2.24 ** No. of obs. 206 204 R2 0.38 0.45 147 Annex G Table G.8 (b): Test score determinants, 2003 JSS (OLS) Math score English score CoeK Z CoeK z School characteristics English books 1.44 2.34 ** 0.04 0.05 Math books 2.38 3.52 *** 2.20 3.20 *** Physical Index 0.19 0.08 4.48 1.57 Low PTR -2.62 -3.18 *** -0.81 -0.73 High PTR 3.19 1.84 * 2.65 1.56 Teachers speak local language 2.26 1.96 * 0.34 0.17 Display material available 0.86 1.01 1.44 1.03 Teachers’ discipline 0.85 1.29 -0.11 -0.12 Teaching methods 14.25 2.93 ** 9.24 1.37 Time on task -0.03 -0.38 0.19 3.13 *** Board easy to clean 3.44 1.67 1.72 0.50 Classrooms have internal walls 3.42 1.90 * 3.77 1.76 * Class disrupted by noise -3.95 -3.61 *** -3.69 -2.36 ** Private school 1. I 4 0.69 8.45 4.18 *** WSD dummy 3.55 2.48 ** 3.67 1.67 QUIPS dummy 1.69 0.89 2.94 2.34 ** Circuit supervisor discuss lesson plans 22.43 1.66 * 63.34 2.55 ** Head teacher sits in on lessons -3.58 -2.09 ** -5.35 -1.97 Student monitoring 0.1 1 0.21 -0.11 -0.19 Community characteristics Forest 0.51 0.53 2.90 2.23 ** Savannah 4.83 3.49 *** 4.16 2.61 ** Rural -3.12 -4.30 *** -3.05 -2.88 *** Child characteristics -1.22 -2.91 *** -1.61 -3.65 *** Age Sex -2.32 -2.81 *** -0.06 -0.06 Years of schooling 2.93 5.25 *** 3.39 4.63 *** Ability 0.13 1.82 * 0.23 2.11 ** Ability missing -0.02 -0.01 1.45 0.42 Fostered in 53.25 2.80 *** 62.79 2.00 ** Fostered*income -3.42 -2.68 *** -4.15 -2.00 ** Total hours worked 0.01 2.17 ** 0.02 2.11 ** Household characteristics Father’s education 0.25 3.67 *** -0.02 -0.18 Mother’s education 0.04 0.47 0.23 1.88 * Income 1.92 2.01 ** 1.73 1.17 Parent in PTA 3.01 1.89 * 4.07 2.24 ** Parent met with teacher 2.75 3.28 *** 0.47 0.49 Other variables Selectivity correction -19.30 -3.77 *** -12.63 -1.47 Constant -35.36 -1.91 * -31.89 -1.24 No. of observations 137 137 R* * . 0.67 0.59 148 Annex G 27. The data f o r 2003 allow the inclusion o f rather more school characteristics and doing so leads t o more satisfactory results, especially for JSS, as shown in Table G.8. Years o f schooling are significant and the coefficient rather higher. Textbook availability has the right s i g n in all four cases for JSS and is always significant. This i s the case only for math books at primary level (English books are negative but insignificant). Teaching methods significantly improve test scores for math in JSS but have a perverse effect in primary. Time o n task i s always positive and significantly so in two cases. And there i s substantial evidence that the quality o f infrastructure matters: test scores are significantly higher in schools in which classrooms have full-sized internal walls and lower in ones where noise disrupts teaching. Participation in WSD positively affects both English and math scores in JSS, and having been a PSD beneficiary i s good for math scores. As before, English test scores are significantly better in private schools, but this i s not so for math (primary schools only, there i s n o difference in either for JSS). For both primary and JSS teachers being able to speak the local language improves student math scores.16 Finally, teachers’ perceptions o f student discipline show that indiscipline significantly worsens English scores. 28. Tuming t o other characteristics: rural children perform worse, as do girls in math. The dummy for fostered-in children has a staggeringly large coefficient. The parental education interactive dummies are not significant, suggesting that it i s the actual residence that matters -but in the case o f income the term i s significant, more than offsetting the beneficial impact o f household income o n test scores. Variables measuring parental involvement also matter: having met with the student’s teacher improves math scores and being in the PTA improves English scores. 29. Finally, when local language test scores are included the most striking result i s the lower explanatory power o f most the variables (results not shown). Schooling s t i l l matters, as does parental education. But virtually none o f the school variables contribute: none o f the project dummies are significant, nor are any o f the school input variables. However, a variable measuring if teachers took the local language test is significant. Interpretation 30. Interpretation o f the results is made by analysing the change in test scores attributable to the change in each o f the determinants. The more complete model estimated using the 2003 data is used for this purpose. For variables collected in each o f the two rounds the comparison i s made using the change in sample means (Table G.9a). For other variables (which are mostly 0-1 dummies) the minimum and maximum values are used (Table G.9b). The exception is schooling, which saw a small drop in the sample, but the sample i s known not t o be representative. 3 1. Schooling improves test scores, each additional year increasing the combined English score by 3.6 points and math by 4.9 points. Enrolments have risen and dropout rates are low: in 2003,95 percent o f those beginning primary complete i t and 86 o f them complete JSS 16. This variable i s measured as the percentage o f teachers taking the test for the designated Ghanaian language for that school. Even though English i s officially the medium o f instruction (though only recently so for the f x s t three grades) teachers may resort to local language to get ideas across. Their being able to do so helps in math. 149 Annex G (Annex H). The 10 percent o f the age group attending school who would not have done so 15 years group can expect to have an increase in their English score o f 20 if they complete primary and 27 if they go o n to complete JSS. For math these figures are 16 and 2 1, respectively. Table G.9 (a): Change in test scores attributable to between sample changes in explanatory variables Samde mean Coefficient Chanae 1988 2003 English Math English Math Forest 0.60 0.40 3.51 0.86 -0.70 -0.17 Savannah 0.1 1 0.17 4.76 4.95 0.27 0.29 Rural 0.44 0.56 -3.45 -2.81 -0.40 -0.33 Age 12.48 12.54 -1.70 -1.30 -0.09 -0.07 Sex 0.45 0.47 0.13 -1.60 0.00 -0.04 Years of schooling 5.04 4.90 3.59 2.76 -0.51 -0.39 Ability 5.93 7.74 0.17 0.12 0.31 0.22 Missing ability 0.48 0.33 -0.91 -3.20 0.13 0.46 Father's education 6.24 7.81 -0.04 0.12 -0.06 0.19 Mother's education 3.81 5.14 0.16 0.02 0.21 0.03 Household expenditure 13.96 14.78 2.70 1.43 2.23 1.18 Fostered in 0.23 0.20 74.64 49.92 -2.23 -1.49 Fostered*expenditure 3.27 3.00 -4.82 -3.02 1.33 0.83 Hours worked 136.86 30.58 0.01 0.01 -1.06 -1.06 English books 1.69 2.47 0.17 1.70 0.13 1.32 Math books 2.00 2.44 1.86 1.53 0.80 0.66 Physical index 0.50 0.59 4.03 0.60 0.38 0.06 Low PTR 0.30 0.18 -1.83 -2.16 0.23 0.28 High PTR 0.03 0.1 1 0.88 1.04 0.08 0.09 Table G.9 (b): Change in test scores attributable to maximum possible changes in explanatory variables Assumed values Coefficient Change Low High English Math English Math Display material available 0.00 1.oo 1.32 1.03 1.32 1.03 Discipline 0.00 6.00 0.70 1. I 4 4.20 6.84 Teaching methods 0.00 1.oo 6.20 8.80 6.20 8.80 Board easy to clean 0.00 1.oo 1.94 2.38 1.94 2.38 Internal class walls 0.00 1.oo 5.08 3.87 5.08 3.87 Class not disrupted by noise 0.00 1.oo 4.28 3.81 4.28 3.81 Private 0.00 1.oo 7.61 0.67 7.61 0.67 PSD 0.00 0.07 3.11 5.40 0.22 0.38 WSD 0.00 0.05 6.62 3.80 0.36 0.21 QUIPS 0.00 0.06 3.40 2.42 0.21 0.15 Circuit supervisor discuss lesson plan 0.00 1.oo 22.40 -1.12 22.40 -1.12 Student monitoring (homework etc) 0.00 5.00 0.36 0.50 1.80 2.50 32. The increase in recurrent and physical items between the t w o rounds increased math scores by 1.6 and English by 2.0 points. This understates the gains in the most deprived areas. Ensuring that a school has one math and Englishbook per child compared t o the 150 Annex G situation in the mid-1980s o f one text per classroom will increase average English scores in children from that school by 6 points and math scores by close to 10 points. 33. Turning to the variables collected only in 2003, it is shown that both process and infrastructure matter. The most important single variable i s teaching methods. I f a l l teachers in the school used modern methods then, compared to a situation in which none do so, children’s English scores would be 6.2 higher and their math score 8.8. The three infrastructure variables combined can improve English scores by 11.3 points and math by 10.1. 34. Home factors also matters to student performance. The two measures o f parental involvement in a child’s education (membership o f PTA and meeting with a teacher) give a combined impact o f 3.5 and 3.9 o n math and English scores respectively. Income also matters; economic growth (the between sample rise in incomes) has increased average English scores by 2.2 and math scores by 1.2 points. 151 Annex H Annex H: Data on Educational Performance WHAT TO ENROLMENTS? HAPPENED HAS 1. Table H.l i s reproduced from the 2002 Education Sector Strategy. I t shows primary enrolments in 2000 down slightly from those in 1990, having behaved erratically during the course o f the decade. However, the report o f the Ghana Statistical Service (GSS) on the Ghana Living Standards Survey reported a school attendance rate for 6-1 1 year olds o f 73.2 percent for GLSS2 (1988/89), rising a full 10 percent to 83.1 percent for GLSS4 (1998/99), see Table H.2. Table H.l: Official data on primary enrolments ~~ Year School-age Primary school Gross Proportion Gender Population enrolment Enrolment enrolled in private parity (Public& private) Rati0 schools I986 2,173,oag 1,679,072 77.3 4.1 0.81 1990 2,453,146 1,945,422 79.3 7.3 0.82 1991 2,544,676 2,011,062 79.0 10.2 0.84 1992 2,63a,a31 2,047,293 77.6 9.7 0.85 1993 2,736,919 2,138,635 78.1 10.7 0.85 1994 2,838,678 2,154,676 75.9 10.9 0.87 1995 2,944,253 2,197,172 74.6 11.o 0.87 1996 3,048,161 2,333,347 76.5 13.1 0.88 1997 3,155,758 2,445,353 77.5 13.1 0.89 1998 3,267,002 2,562,229 78.4 13.1 0.90 1999 3,382,649 2,684,689 79.4 13.1 0.91 2000 3,154,152** 2,477,990 7a.6** Note: ** Data from the 200012001 MOEYS annual school census and population data from the 2000 national population census conducted by the Ghana Statistical Service. All other population figures are based on projections from the 1984 Population Census. Source: SRIMPR Division MOEYS. 2. To understand what i s going on here and Table H.2: Attendance rates reported get an accurate picture o f what i s happening to from Ghana Living Standards Survey enrolments it i s helpful to first clarify definitions. Age range GLSSZ GLSS4 ( I 988/89) (1998/99) Some Definitions 6-1 1 73.2 83.1 12-15 71.8 80.4 3. Enrolment data may be either net or gross, 16-18 54.1 47.0 for which the definitions are as follows, given 19-25 14.1 13.5 here for the case o f primary education: Source: GSS (1996 and 2000) No. of children of primary school age in primary school Net enrolment rate, NER = f children o No. o f primary school age 152 Annex H No. of children in primary school Gross enrolment rate, GER = No. of children of primary school age However, using household survey data that ask for the age o f each o f each household member and whether they are currently in school or not, i t i s most straightforward to work out an attendance rate (this i s the term used by GSS, i t does not refer to the proportion o f children enrolled actually attending school, which i s another meaning o f the term), given here for children aged 7-12: No. of children aged 7 - 12 in school Attendance rate, AR = No. of children aged 7 - 12 4. While i t must always be the case that NER p2 =@(e, -xp>-aqc, -xp> p3 =1-@(c, -xp) where c l and e2 are the thresholds andXp i s the product o f variables and coefficients to be estimated. The model can be estimated in this way for the uncensored observations, but these probabilities are not correct i n the case o f censored observations. For a censored child who has achieved level 1, the correct probability i s not P2, as it would be for an uncensored child, but 1- P2 - P3, since we assume the child would achieve at least level 2, and possibly level 3. In general, for a censored child having achieved the schooling level j, the probability to be estimated is: Pj = 1- @(Cj-, - xp) The likelihood function for the censored ordered probit i s therefore the sum o f the likelihood function for the censored and uncensored observations, where the probabilities for the censored and the uncensored children are estimated differently in the way described above. 175 Annex I Table 1.4: School attainment (censored ordered probit) 1988 2003 Pooled Coeff. z-stat Coeff. z-stat Coeff. z-stat Community characteristics Forest 0.03 0.08 -0.02 -0.06 0.17 0.70 Savannah -1.33 -4.74 *** -0.07 -0.22 -0.76 -2.69 *** Rural 0.19 1.80 * -0.09 -0.63 0.08 0.66 Female -0.39 -2.39 ** -0.26 -1.15 -0.34 -3.62 *** Female in Savannah -0.03 0.10 0.30 0.26 0.24 1.75 * Female in Forest -0.17 -0.84 0.45 0.27 0.09 0.71 Private school in locality -0.03 -0.25 0.80 2.67 *** 0.10 0.79 Average school fee -0.00 -2.15 ** -0.00 -0.39 -0.00 -0.34 Household characteristics No. of siblings -0.03 -1.61 * -0.09 -3.42 *** -0.05 -2.23 *** Father's education 0.07 5.60 *** 0.04 4.62 *** 0.06 7.00 *** Mother's education 0.05 3.42 *** 0.03 1.19 0.03 3.13 *** Ability 0.02 1.69 * 0.02 1.44 0.02 2.20 ** Missing ability -0.26 -2.56 *** -0.27 -1.02 * -0.23 -2.95 *** Log per capita expenditure -0.05 -0.59 0.32 3.86 *** 0.11 1.47 Child characteristics Age 0.08 2.37 *** 0.04 1.13 0.05 2.42 ** Birth order 0.13 2.87 *** 0.12 3.27 *** 0.10 3.00 *** School characteristics Distance (minutes) -0.01 -2.24 *** -0.01 -1.04 -0.01 -2.21 *** English books 0.01 0.12 -0.07 -0.71 -0.01 -0.24 Math books 0.06 0.62 -0.31 -2.39 0.02 0.12 Chalk 0.00 0.01 0.19 1.43 0.09 1.00 * Desks -0.04 -0.37 0.94 2.62 *** 0.09 0.86 Adequate classrooms 0.00 0.05 0.01 2.09 ** 0.00 0.81 Classrooms can be used when raining -0.12 -1.12 0.00 1.91 * 0.00 1.40 Library 0.27 0.88 0.51 2.25 ** 0.54 2.00 ** Water -0.28 -1.20 0.10 0.57 -0.08 -0.45 Classrooms with chalkboard 0.39 1.50 -0.31 -0.83 0.06 0.19 Board quality 0.21 2.50 ** 0.17 1.46 0.1 1 1.29 Low pupil teacher ratio 0.1 1 0.66 1.oo 2.88 ** 0.14 0.76 Hiah t)uDil teacher ratio " I , -0.13 -0.59 -0.41 -1.73 * -0.13 -0.59 Number of observations 1399 1334 2733 Log likelihood -74 1 -489 -1342 176 Annex I Table 1.5: School attainment: Cox regression 1988 2003 Pooled Hazard z-statistic Yazard z-statistic fazard z-statistic ra ti0 ratio ratio Community characteristics Forest 0.86 -0.31 1.35 0.70 0.73 -0.83 Savannah 3.92 2.81 *** 1.22 0.46 2.14 1.91 * Rural 0.79 -1.21 0.97 -0.14 0.88 -0.79 Female 1.79 2.92 *** 1.68 1.26 1.63 3.00 *** Female in Savannah 0.77 -1.03 0.53 -1.43 0.71 -1.87 * Female in Forest 1.17 0.56 0.46 -1.65 * 0.89 -0.57 Private school in locality 0.89 -0.62 0.19 -2.69 *** 0.83 -1.21 Averaae school fee 1.oo 2.88 *** 1.00 0.28 1.00 0.19 Household characteristics No. of sibling 1.04 1.82 * 1.09 3.07 *** 1.05 2.24 ** Father’s education 0.90 -5.20 *** 0.91 -3.36 *** 0.91 -5.26 *** Mother’s education 0.98 -1.01 0.96 -0.96 0.97 -1.58 Ability 0.98 -1.29 0.96 -2.59 ** 0.98 -2.24 ** Missing ability 1.25 1.62 1.19 0.68 1.24 1.92 * Log per capita expenditure 1.11 0.77 0.60 -2.62 *** 0.85 -1.44 Child characteristics ~~ Age 1.01 0.19 1.03 0.65 1.01 0.45 Birth order 0.87 -2.60 *** 0.89 -2.18 ** 0.89 -2.73 *** Fostered in 7.61 0.75 0.00 -1.51 0.67 -0.21 Fostered * father’s ed. 1.05 1.64 * 1.00 0.03 0.97 -1.13 Fostered * mother’s ed. 0.97 -1.07 0.91 -1.72 * 1.03 1.03 Fostered * income 0.86 -0.74 1.52 1.51 1.03 0.24 School characteristics Distance (minutes) 1.01 2.86 *** 1.01 1.34 1.01 2.65 *** English books 1.oo 0.01 1.04 0.30 1.01 0.07 Math books 0.89 -0.89 1.59 2.80 *** 0.96 -0.46 Chalk 0.96 -0.28 0.79 -1.44 0.84 -1.41 Desks 0.87 -0.65 0.33 -2.50 ** 0.79 -1.92 * Adequate classrooms 1.oo -0.61 0.99 -2.32 ** 1.00 -0.95 Classrooms can be used when raining 1.18 1.20 0.99 -1.88 1.00 -1.46 Library 0.67 -1.05 0.48 -2.33 ** 0.48 -2.32 ** Water 1.35 1.04 0.88 -0.52 1.13 0.54 Classrooms with chalkboard 0.74 -0.88 0.97 -0.05 0.91 -0.32 Board quality 0.80 -2.21 ** 0.85 -1.31 0.91 -1.07 Low pupil teacher ratio 0.87 -0.74 0.23 -2.90 *** 0.81 -1.24 High pupil teacher ratio 0.39 -1.92 * 1.94 2.37 ** 1.30 1.02 Number of observations 1399 1334 2733 Log likelihood -2274 -1 187 -3879 177 Annex I 8. The Cox regression results are shown in Table I.5.6 O f the 10 school quality variables four are significant with the expected sign: having an adequate number o f classrooms matters, as does having enough desks, good quality chalkboards, and a library.’ In addition, the distance to the nearest school has a significant impact o n the probability o f school attending and staying in school. There i s one perverse result, which i s that having more classrooms that can be used when it i s raining reduces school attainment.* I t i s possible that this result i s explained by the nature o f the school pavilions erected using W o r l d Bank financing. These metal, concrete-based, structures were undoubted improvements o n the mud-walled classrooms they frequently replaced. But unless clad, which many are not, they cannot be used when i t rains h e a ~ i l yHence, .~ these schools are improved but suffer this problem. Teacher numbers also have the expected effect: schools with high pupil-teacher ratios deter students, whereas those with l o w numbers encourage them.” 9. Turning to household and child characteristics, education o f the parents also has the expected sign; though in 2003 mothers’ education appear to have lost significance. The innate ability coefficient has the expected sign, but i s not significant in 2003. Possibly this i s a consequence o f the reduction in the number o f dropouts for which ability i s more relevant. Household income has become an important determinant o f a child’s education. Virtually a l l o f the fostering terms are insignificant, suggesting that the characteristics o f the household in which the child i s resident do matter for the educational choices relating to that child. 10. The presence o f a private school in the locality increase attendance, though the coefficient i s significant only in 2003, since in 1988 private schools were not very common. The average locality school fee has the expected negative effect in 1988, but none in 2003. 11. The Cox regression was also estimated up to attendance in senior secondary school. N o data were collected on SSS quality. However, it i s likely there i s less variation in this than there i s between basic schools, so that variation will not be a major determinant. But costs are considerably higher for senior secondary, so income may be expected to play a larger role. Table 1.6 bears this out. The odds ratio for household expenditure is (a bit) lower and its significance rather higher. Basic school variables also matter to whether a child makes it through to secondary: two o f the school quality indices are significant, as is the distance to school. Highpupil-teacher ratios also discourage attendance. 6. They are not greatly different from those in Table 1.4 in terms o f which variables are significant or not. 7. Few schools have a library. T h i s variable may be acting as a dummy for “very good” schools. 8. M a t h book availability has a perverse result in one case. 9. During field work the study team got stuck in one o f these pavilions in heavy rain. Staying dry requires huddling in the middle o f the “room” Goined by neighboring livestock also trying to stay dry), none o f which i s conducive to study. The Primary School Development (PSD) project significantly improved the percentage o f schools with classrooms that can be used when raining (Annex D). But the majority o f PSD pavilions have been clad (PSD ICR). I t i s possible that more recent structures have not. 10. 1988 i s an exception with respect to high PTR. 178 Annex I T a b l e 1.6: School attainment up to senior secondary school (Cox regression) Coefficient z-statistic Coast 1.02 0.14 Forest 0.93 -0.45 Savannah 1.57 2.62*** Rural 1.06 0.78 Sex 1.58 4.1 7*** Female savannah 0.80 -1.44 Female forest 0.95 -0.35 Age 1.03 2.26** Birth order 0.92 -3.47*** Number of siblings 1.03 2.25** Father's education 0.95 -8.41*** Mother's education 0.96 -4.85*** Innate ability 0.99 -1.42 Innate ability missing 1.03 0.50 Per capita expenditure 0.84 -3.62*** Distance to nearest primary 1.oo -0.98 Distance to nearest JSlmiddle school 1.01 7.31*** Index recurrent inputs (primary) 0.40 -4.30*** Index physical inputs (primary) 1.08 0.28 Index recurrent inputs (middle/JSS) 0.77 -0.95 Index physical inputs (middle/JSS) 0.45 -3.31*** Private 0.98 -0.18 School fee 1.oo -1. I 4 Low pupiVteacher ratio 0.96 -0.28 High pupil teacher ratio 2.31 6.98*** Observations 4002 Chi square 978.7 Log likelihood -9093.0 Interpretation 12. To interpret the relative importance o f the different factors affecting school attainment it i s necessary to combine the level and range o f the explanatory variables with their coefficients. For ease o f exposition, this analysis i s presented using the results from a probit model o f enrolments, which yields similar results to those in the other attainment regressions. The sample used here i s children aged 10- 15. If the sample 9- 15 i s used, the results are similar except that the age term i s significant, showing that some children aged nine have not yet started school but are likely to do so. 13. The results are shown as the marginal effects o f the probit model (Table 1.7), together with the sample means o f the explanatory variables for 1988 and 2003. I t i s therefore 179 Annex I possible to calculate the impact o n enrolments o f the observed changes in the different independent variables. Table 1.7: Marginal effects from probit model of enrolments and implied change in enrolments from different factors Variable Whole 1988 2003 Marginal Accountable sample impact change Mean (*loo) Forest 0.492 0.523 0.460 3.43 -0.22 Savannah 0.220 0.201 0.238 -21.11 -0.78 Rural 0.468 0.451 0.485 1.63 0.06 Sex 0.465 0.449 0.481 -20.78 -0.66 Sex*survey 0.701 0.449 0.962 8.35 4.28 Female*savannah 0.093 0.077 0.110 3.78 0.12 Female*forest 0.238 0.247 0.229 -12.33 0.22 Age 13.008 13.013 13.003 -0.13 0.00 Birth order 3.176 2.892 3.470 1.29 0.75 Father's schooling 6.487 5.744 7.255 1.08 1.63 Mother's schooling 3.657 2.894 4.447 0.31 0.48 Ab iIity 9.663 11.553 7.708 -0.07 0.27 Missing ability 0.422 0.493 0.349 -5.87 0.85 Household expenditure 14.329 13.967 14.704 3.48 2.56 Fostered in 0.240 0.248 0.232 24.38 -0.38 Foster*income 3.451 3.469 3.433 -2.99 0.1 1 Distance to school 12.475 13.696 11.21 1 -0.10 0.26 Chalk 2.405 2.092 2.730 2.70 1.72 Adequate classrooms 94.742 96.1 16 93.327 0.05 -0.14 Rain rooms 0.202 0.259 0.144 10.22 -1.18 Board quality 2.581 2.494 2.672 3.25 0.58 Library 0.092 0.067 0.117 6.31 0.32 High PTR 0.203 0.268 0.135 -1.75 0.23 Survey 1.491 1.ooo 2.000 -6.00 -6.00 14. I nthe sample, enrolments (the attendance rate) grew by 5.5 percent, from 81.6 to 87.1 percent. Some factors acted to lower enrolments, so that the cumulative effect o f all the positive factors exceeds 5.5. The negative factors are mostly demographic shifts and the largest i s a pure "survey round" effect (which i s not significant)." The results are as follows: 0 The largest single effect comes fiom the reduction of gender bias in enrolments, which raised enrolments by over 4 percent. 11. In line with standard modeling procedure, all variables are entered into the analysis whether or not they were significant. 180 Annex I e The improvements in the school quality variables accounted for an increase in enrolments o f over 3 percent, the largest single impact coming from the chalk variable, though this i s probably picking up the general availability o f resources in the school. This effect i s partially offset by the perverse impact o f classrooms that cannot be used when raining and the fact that, in this sample, the percentage o f schools with adequate classrooms f e l l slightly. e There is also a substantial impact (of 2.5 percent) from the increase in household expenditure between 1988 and 2003. 15. As presented here, it i s difficult to see the impact Figure 1.1 Travel time has been reduced considerably o f school building and for those farthest from school rehabilitation. But it can be seen in three ways. First, the reduction in travel time to school i s a result of school building. While the mean travel time has not fallen very much, those who were furthest from school (more remote, and typically more disadvantaged, groups) have benefited most. Figure loo I 1 shows the distribution o f . 0 20 40 eo Time taken to primary School 80 izo the sample over travel time. " = I988 -2003 For 80 percent o f the sample this number has not changed. But for those furthest from schools travel time has been reduced considerably. The maximum travel time has fallen from 2 hours to 90 minutes, and the average travel time for those more than 20 minutes away fallen from 48 to 36 minutes. By 2003, only 4 percent o f the sample was more than 30 minutes from a school compared t o nearly 10 percent in 1988. 16. Imagine a community with the nearest school one hour away. Building a school in that community, giving an average travel time o f 10 minutes, will increase enrolments in that community by 5.2 percent. Table 1.8 shows the 10 clusters (out o f 79 for which the calculation can be made) with the largest change in average reported travel time t o the nearest primary school, and the change in enrolments expected from that change implied by the regression coefficient. In the cluster with the largest change, o f a 45-minute reduction, enrolments were expected to rise by 4.7 percent. On average enrolments are expected t o have risen by about 2.2 percent in these cluster as a result o f schools being closer. The fact that schools are closer will depend in part o n changing settlement patters, but the largest effects will result from school building. 18'1 Annex I Table 1.8: Changes in travel time to nearest primary school in 10 clusters with largest change between surveys Cluster no. Year of survey Rural (Yes=l) Change in 1988 2003 Travel Attributable time enrolments 81 72 26 1 -4 5 4.7 1 43 13 1 -30 3.1 68 31 6 0 -25 2.6 45 33 12 1 -2 1 2.2 54 24 6 0 -18 1.9 69 36 19 1 -18 1.8 74 23 5 1 -17 1.8 66 20 7 0 -13 1.3 20 16 6 1 -1 1 1. I 46 16 6 1 -10 1.o 17. The impact o f school building on travel times can be seen by looking in more detail are the data from those clusters with the largest reductions in travel time to the nearest primary school. Table 1.9 reports the distance to school reported by each household for four o f the clusters shown in Table 1.8. In the first cluster shown, which i s that with a reduction in travel time o f 45 minutes, in 1988, 19 o f 20 households reported that the nearest primary school was over 30 minutes away, but in 2003, nearly 80 percent said it was less than 30 minutes. The data seem clear that a school was established in the community closer to the majority o f the population.'* Enrolments in this community in fact increased from only 10 percent to 80 percent.13A similar pattem can be seen for the second cluster shown, where enrolments increased by over 20 percent. Table 1.9: Distribution o f travel time to nearest primary school in clusters with substantial reductions in travel time Rural Rural Rural Urban 1988 2003 1988 2003 1988 2003 1988 2003 10 minutes or less 0 5 0 9 14 16 0 32 11-30 minutes 1 10 2 9 11 13 14 1 31-45 minutes 2 2 8 0 10 0 0 0 Over 45 minutes 17 2 4 0 13 0 0 0 Total 20 19 14 18 48 29 14 33 12. The school survey contains data for four schools in t h i s cluster, two o f which were established in the mid- 1990s. In 1988, enumerators surveyed schools outside the cluster if there were none inside. 13. Sample sizes are rather s m a l l to rely o n community-level enrolment data. Nonetheless, the substantial rise in enrolments in the three communities with the largest reduction in school distance i s notable (the increase being over 20 percent in the other two clusters). 182 Annex I 18. In the third cluster shown, close to h a l f the children were already less than 30 minutes from school in 1988. But in 2003 they all were, suggesting that another school was built in the cluster. The school survey shows that a new school was built in this community in 2001. Finally, the table shows data for an urban cluster where another school appears to have been built: while all households were less than 30 minutes from a school in 1988 (but more than 10 minutes), in 2003 all but one are 10 minutes or less from the nearest school. A new school was built in this cluster in 1991. 19. In summary, new school building can have a substantial impact on enrolments in the community in which the school i s built, particularly if i t suffered from being a great distance to the existing school before the new construction. While these effects are great at the local level, the fact that the vast majority o f the population was already within 20 minutes o f a school in 1988 means that the aggregate effect o f school building at national level i s not that great, adding only about one quarter o f a percent to enrolments. 20. The second channel through which school building can have an effect on enrolments i s that having an adequate number o f classrooms can have a substantial impact. Although the large majority o f schools do have sufficient rooms, some do not -nearly 10 percent o f schools have only h a l f the required number or fewer. Suppose a primary school teaching all six grades has two classroom blocks but one is unusable, so that it only has h a l f the required number o f classrooms. Rehabilitating (or replacing) the unusable block will raise enrolments in the school’s catchment area by 2.4 percent. 21. Analysis o f the data from the five clusters with the largest increase in having sufficient classrooms in a school (dealing only with clusters which had far less than necessary in 1988). The increased availability o f classrooms appears to have increased enrolments in these clusters by, o n average, around 2 percent. However, these figures understate the impact o f classroom building through this channel since, with growing population, new classrooms have to be built just to maintain having sufficient classrooms. 22. The final issue i s classrooms that cannot be used when it i s raining. Considering the sample as a whole, this variable has a robustly significant positive impact on enrolments (the more classrooms that cannot be used when raining then the higher are enrolments). However, analysis o f the data shows that this result comes from the fact that schools with 100 percent o f classrooms that cannot be used when it i s raining have high levels o f enrolment. In bivariate analysis, there is a highly significant negative relationship between the two variables if individuals linked to schools with 100 percent o f classes that cannot be used are dropped. In the multivariate analysis, this i s so for 85 percent o f the sample (using a cut-off of 50 percent o f classrooms). For the large majority o f the sample the coefficient i s robust giving a marginal impact in the region o f -0.33, implying that rehabilitating a school so that all rather than none o f the classrooms can be used when raining increases the probability o f enrolment by one-third. The actual improvement in the percentage o f classrooms that can be used when raining accounts for a 3.5 percent increase in enrolments across the country. 23. Finally, there i s an “enrolment multiplier effect” since educated parents are more likely to send their children to school than are less educated ones. The increase in parental 183 Annex I education between the two rounds o f the survey contributed another 2 percent t o enrolments over the 15 years. Schooling and Child Labor 24. Schooling and child labor are inter-linked as the household decides o n the allocation o f the child’s time between one or the other. Bhalotra and Heady (2003) found that the most important determinants o f child labor for Ghanaian farm households are, besides the usual region, religion, and ethnicity dummies, the number o f farms operated (but not farm size), the absence o f the father (but only for girls work), the education o f the mother (reflecting preferences), the availability o f public transport in the community (reflecting distance t o school effect), rainfall (negative effect), electricity (positive effect) and “the dynamism o f the region as reflected in subjective assessments o f l i f e and work opportunities having got better in the last ten years.” N o relationship was found between child labor and household expenditure after instrumenting for expenditure in order to circumvent endogeneity (since child work may increase household consumption). Similarly, Canagarajah and Coulombe (1997) find that child labor i s poorly correlated with poverty. Father’s education has a negative influence o n child work (especially for girls), and child labor i s more common among family enterprises (farming or otherwise). School participation, o n the other hand, i s found negatively correlated with school costs (official and unofficial fees). 25. The opposite side o f the coin t o rising enrolments should be reduced child labor. Figure 1.2, which shows the proportions o f children working out o f the samples o f 1988 and 2003, demonstrates that this has been the case. The number o f children working has Figure 1.2 Proportions o f working children in 1988 and 2003 by age 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Age Source: GLSSP and GSS/OED household survey 184 Annex I decreased dramatically between the two survey periods. Similar factors seem to drive both trends: in particular more educated parents are more likely to send their children to school and less likely to require them to work. An additional factor may be the lengthening o f the school day, giving children less time for work, and (though we have n o evidence o f this) the increase in homework n o w that textbooks are available. 26. Note o n Figure 1.2: The sample includes all children surveyed, thus some children are full-time workers while others are working and studying at the same time. The definition o f working i s obtained from the household questionnaires. Children are considered workers if they have worked for some time during the past 12 months. A large number o f workers results from this definition since it includes people working only occasionally, but it seems relevant for this study because the working time is potentially lost studying o r school time. The definition o f work is broad, including wage work, family farm and family enterprise work. 186 Annex J a, K 0 nm 53 187 Annex J 188 Annex J 189 Annex J P E B cn a h Y Annex J Annex J 192 Annex J P 8 I)! & v) 3 Y 0 % I e h 193 Annex J Annex J 195 Annex J POLICY-RELATEDLEGAL FOR COVENANTS PRI M ARY SCHOOL DEVELOPMENT PROJECT Description Status Comments Implement a program to monitor instructional time, CDlCP Although standard statistics are being intakes, enrollments, dropouts, attendance, and gathered through EMIS, instructional learning time of primary school students. time is not monitored. Implement information campaign regarding impact of CP Periodic public announcements were student fees and levies on primary school made and information campaigns were enrollments. carried out, but fees and levies have continued to Droliferate. Implement recruitment procedures for primary CP Procedures were followed in most of school head-teachers; require applicants to meet the schools, but the role of minimum criteria and serve for four years, and communities often handled by a district establish selection panels that include local official residing outside the community. community Ieade rs. Prepare and implement training programs for C Training was not put to use primary school head-teachers Prepare and implement orientation programs for C None district-level officials and community leaders Prepare and implement training programs for circuit C Training has not been put to use. officers to increase capacity to support and monitor classroom construction activities and effective Drimarv school management ~ Conduct a detailed school mapping of approximately NC To be included for future work as part 1,500 schools to help identify school amalgamation of EMIS. options and need for rehabilitation of school facilities. Borrower to implement program for increasing actual CP Directives were sent but were not instructional time of primary school students. monitored and thus not enforced. Borrower to take action to ensure that no new fees CP MOE directives were issued but not or levies are imposed on primary school students. monitored or implemented. Borrower to implement program to eliminate fees CP MOE formally announced that all and levies imposed on primary school students other unapproved fees and levies were to be than those approved by Ministry of Education. abolished. This was not well monitored or enforced. Source: World Bank documents 196 Annex K Annex K: Education and Welfare Outcomes EARNINGS FUNCTIONS the household head, although these Mean earnings As percent lower level increases are less marked, other than for 1988 2003 1988 2003 higher education, in 2003 than 1988. None 2.05 4.45 Primary 2.54 4.99 23.9 11.9 2. Table K.2 reports the results fkom MiddlelJSS 2.93 5.09 15.3 2.1 the OLS regression o f loggedper capita ss 3.96 5.65 35.1 11.0 expenditure. These are augmented earnings Higher 5.46 8.25 38.0 45.9 Table IC2 (a): Earnings functions using school years (pooled data): education o f household head Model 1 Model 2 Model 3 Coeff t-stat Coeff t-stat Coeff t-stat Age of head (logged) -0.16 -2.44 0.02 0.63 .. -0.12 -1.83 ** School years 0.04 6.58 *** 0.03 16.00 *** Combined test score (logged) 0.13 3.58 *** 0.25 7.7 *** Female -0.24 -6.01 *** -0.15 -7.97 *** -0.23 -5.66 *** Dependency ratio -0.99 -16.38 *** -0.80 -26.06 *** -1.00 -16.23 *** Rural -0.19 -5.20 *** -0.19 -10.35 *** -0.21 -5.79 *** Forest -0.18 -4.90 *** -0.09 -4.77 *** -0.19 -5.1 *** Savannah -0.25 -4.79 *** -0.27 -10.72 *** -0.25 -4.72 *** Survey dummy 0.67 19.92 *** 0.68 38.10 *** 0.68 19.89 *** Intercept 15.21 60.74 *** 14.99 148.71 *** 15.09 59.29 *** R2 0.49 0.43 0.47 No. of observations 1113 4922 1113 ~ 1. The return here i s based o n expenditure per capita and so under-stated to the extent that a single person i s earning an income spread aver several people. 197 Annex K Table K.2 (b): Earnings functions using school years (pooled data): average education o f those aged 16 and over Model 1 Model 2 Model 3 Coeff f-stat Coeff f-stat Coeff f-stat Average age (16 and over) , 0.15 2.59 ** 0.33 11.49 *** 0.05 0.87 .. School years 0.05 12.16 *** 0.04 18.55 *** Combined test score (logged) 0.07 3.01 *** 0.19 7.79 *** Female -0.12 -4.05 *** -0.11 -5.74 *** -0.10 -3.4 *** Dependency ratio -0.85 -16.72 *** -0.84 -28.15 *** -0.97 -18.68 *** Rural -0.16 -5.53 *** -0.18 -9.75 *** -0.24 -7.98 *** Forest -0.14 -4.73 *** -0.10 -5.04 *** -0.15 -4.99 *** Savannah -0.22 -5.45 *** -0.26 -10.12 *** -0.27 -6.32 *** Survey dummy 0.65 24.62 *** 0.67 37.83 *** 0.65 23.86 *** Intercept 14.08 65.49 *** 13.80 128.00 *** 14.54 66.12 *** R2 0.48 0.43 0.43 No. of observations 1808 4922 1800 3. Such regressions have to be interpreted with caution. Using them t o estimate the growth effects o f educational expansion can fall into a trap o f the fallacy o f composition. Educating one person alters their l i f e chances given the current state o f affairs, so that they will likely enjoy a higher income. But educating many people changes the state o f affairs. I f the income gains o f education come from accessing a limited number o f employment opportunities, then the returns to education will fall as the number o f educated people rises. O n the other hand, if income gains are from genuine productivity increases - either for the self-employed or the employed if the wage reflects the marginal product -then educational expansion will indeed lead directly to growth. 4. Evidence o f the former, less happy, picture is give by looking at the Mincerean returns for the two periods (Table K.4). These returns are the coefficients o n education dummies in the earnings function where the (omitted) reference category i s n o education.’ The 1988 data show the expected pattern o f returns increasing for each category o f ed~cation, though ~ the return to primary education i s not significant. But by 2003 not only have all the returns fallen -the expected effect from having more educated people available - significant positive returns are only found for senior secondary and tertiary graduates. 5. Disaggregation into rural and urban areas shows returns to have fallen in both. There was a significant retum to primary education in 1988, but this is n o longer the case. In 2003, in rural areas the only significant return i s from post-secondary education. Plausibly, secondary graduates find employment in urban rather than rural areas, but there are a few professional positions in rural areas (teachers, health workers) for which people have received post-secondary education. 2. Those with incomplete primary are included in “no education.” There are few observations in t h i s category so their treatment does not alter the results. 3. Since all returns are with reference to the base category, not the preceding level as i s often done. 198 Annex K Table K.3: Earnings functions using level of education (education of household head) Pooled data 1988 2003 Coeff. t-stat CoeH t-stat Coeff. t-stat Age of head (logged) -0.18 -2.66 *** -0.10 -2.91 *** -0.07 -1.65 * Primary -0.18 -15 3 0.05 0.76 .. -0.04 -0.49 .. Middle/JSS -0.15 -3.56 *** 0.12 4.85 *** -0.06 -1.69 sss 0.02 0.52 *** 0.30 7.33 *** 0.06 1.74 * Tertiary 0.17 2.90 *** 0.40 4.58 *** 0.35 8.30 *** Combined test score (logged) 0.19 5.43 *** Female -0.23 -5.54 *** -0.06 -2.37 ** -0.25 -7.66 *** Dependency ratio -0.98 -15.88 *** -0.86 -22.03 *** -0.66 -12.87 *** Rural -0.18 -4.97 *** -0.19 -8.39 *** -0.27 -8.76 *** -0.18 -4.85 *** -0.09 -3.87 *** -0.08 -2.27 ** Forest Savannah -0.25 -4.79 *** -0.30 -9.88 *** -0.31 -6.96 *** Survey dummy 0.60 15.38 *** Intercept 15.58 57.40 *** 15.45 121.37 *** 16.18 100.45 *** R2 0.49 0.26 0.28 No. of observations 1113 3182 1740 1988 2003 Rural Urban Rural Urban Primary 0.16 * -0.06 0.04 -0.12 Middle/JSS 0.13 *** 0.1 1 *** -0.01 -0.08 * Senior Secondary 0.31 *** 0.28 *** 0.04 0.07 * Higher -0.10 0.49 *** 0.26 *** 0.37 *** 6. But there is an important caveat to place o n these findings, since the earnings equations also include test scores. The combined test score has a significantly positive impact o n average earr~ings.~ Education thus has a direct effect o n earnings and an indirect effect through higher test scores. Plausibly the direct channel picks up the screening function o f education, whereas the indirect channel reflects genuine productivity increases. Table K.5 reports the impact o f education o n earnings through the two channels based o n regression results for 2003 only.5 These show the indirect effect to be stronger than the direct effect in all cases, being sufficient to offset the apparent negative returns t o primary education. Hence those attending primary school and JSS, and attaining better test scores as a result, do indeed benefit from higher earnings. But children who do not make appreciable gains in cognitive achievement as a result o f school attendance are n o better o f f as a result o f their schooling. 4. The correlation coefficient between the combined math and English scores i s 0.78, so there two variables have been added to make a single variable. 5. The impact o f schooling o n test scores i s derived from substituting years o f schooling into the test score regressions in Annex G. The earnings regressions uses the d o g g e d combined test scores to facilitate the calculation. 199 Annex K Years of schooling Level of education Direct Indirect Total Direct Indirect Total Primary 15.4 16.0 33.8 -18.5 32.0 7.6 Middle/JSS 23.0 24.0 52.6 -15.1 48.0 25.7 Secondary 30.7 32.0 72.6 -7.4 64.0 51.9 Tertiary 38.4 40.0 93.8 7.4 80.0 93.3 AND CHILD N EDUCATION UTRI T O I N 7. The most studied welfare outcomes in the Ghanaian context are fertility, mortality, and nutrition. Existing studies demonstrate the beneficial impact o f education o n fertility and nutrition, and the GSS\OED survey did not include the variables necessary (a health module and mother’s birth history) to analyze these outcomes. However, i t i s possible to report data o n nutritional outcomes. These are o f particular interest since an earlier analysis by Alderman (1999) o f GLSSl data found no significant impact o f education o n nutritional outcomes (measured by height for age). Replication o f Alderman’s model using the GLSS2 data gave the same result. But recent papers using less representative data for Accra (Ruel et al. 1999 and Maxwell et al. 2000) have suggested that schooling does improve nutrition through its association with better childcare practices. 8. The question that arises is whether different inputs are substitutes or complements. For example, health education and clean water are usually argued to be complements in that the impact o f one is greater in the presence o f the other. By contrast, education has been suggested to be a substitute for income with respect to nutrition, meaning that well-educated but less well off women can achieve the same nutritional outcomes for their children than can better off, but less well-educated, women. Testing for complementarity or substitutability requires an interactive income and education term. If the coefficient i s positive, then the two inputs are complements, and if it i s negative then they are substitutes. 9. Two models are presented. One with a series o f education dummies, as used by Alderman, and one with the years o f education. The advantage o f the latter i s it allows some effects from incomplete primary without having to make a decision about h o w much primary constitutes “some.” In both models, expenditure i s instrumented with a set o f location variables and household characteristics, since a Hausman test shows it to be endogenous. The instruments include the education o f the household head, making it less likely that there will be any direct effect from father’s education in the nutrition education. Parents’ height i s omitted. Its inclusion does not greatly alter the coefficients, but the smaller sample size (less than h a l f o f that obtained if these two variables are not included) reduces significance o f the variables in some specifications. 10. The results, shown in Table K.6, are striking. The maternal education variables are not significant when the interactive t e r m i s not included. But with the interactive term a l l education variables are positive and significant. The interactive terms are significantly negative, indicating that education i s a substitute for income, as was also found by Ruel et al. 200 Annex K This means that the impact o f education o n nutrition falls with income, but i s positive over the range o f over 90 percent of the data.6A child in a household with a per capita expenditure o f cedis 0.16 m i l l i o n and mother who has completed JSS can expect a height for age z-score 4.5 points higher than a child in a household at the same income level but whose mother has no education. If the household’s expenditure i s cedis 3.2 million, then the “JSS premium” drops to 1.1points. 11. The substitutability point is illustrated in Figure K.1,which plots nutritional isoquants with education and income as the two inputs in the nutrition production fimction. These isoquants are convex to the origin, rather than concave, as i s usually the case. If they had been concave it would mean that after a certain point there i s n o nutritional return to higher income without increasing education. This is not the case -higher income will improve nutrition even if education remains low. Figure K.l: Nutrition isoquants 1 I HAZ=2 14 P .- 3 12 HAZ=O c 0 u) L 10- m HAZ=- s 01 \ , 10.5 11.0 11.5 12.0 12.5 13.0 13.5 Logged income 6. T h e coefficient of education + education x income becomes negative with a logged expenditure o f around 15. These negative effects are of course offset by the positive effect f i o m income. 20 1 Annex K T a b l e K.6: Determinants o f height for age z score, 2003 Model 1 Model 2 Model 3 Model 4 Household variables Per capita expenditure 1.33 ** 2.39 1.46 *** 2.14 *** (2.08) (1.92) (2.86) (2.90) 0.05 0.07 * 0.06 * 0.05 * Size (156) (1.81) (1.83) (1.70) Savannah -0.18 -0.05 -0.19 -0.24 (-0.67) (-0.15) (-1.12) (-1.40) Forest 0.08 0.10 (0.32) (0.36) Accra 0.13 0.28 (0.53) (1.14) Child characteristics Age in months -0.14 *** -0.14 *** -0.14 *** -0.14 *** (-8.23) (-7.87) (-8.09) (-8.12) Age squared 0.00 *** 0.00 *** 0.00 *** 0.00 *** (7.37) (7.06) (7.22) (7.35) Sex (Female = I ) -0.02 -0.06 -0.02 -0.07 (-0. f 6) 1-0.50) (-0.19) (-0.60) Mother’s education Years of schooling -0.02 2.78 (-0.55) (1.77) Complete primary -0.13 26.45 ** (-0.64) (2.36) Complete middle -0.1 1 31.85 *** (-0.41) (2.70) Secondary or higher -0.37 26.24 ** (-1.OO) (2.27) Father’s education Years of schooling 0.02 0.02 (1.16) (1.32) Complete primary 0.33 ** 0.14 (2.09) (0.72) Complete middle 0.29 0.15 (1.64) (0.77) Secondary or higher 0.13 0.17 (0.50) 10.73) Interactive terms Years x expenditure -0.19 * (-1.76) Primary x expenditure -1.82 ** (-2.36) Middle x expenditure -2.18 *** (-2.70) Secondary+ x expend. -1.79 ** (-2.27) intercept -18.83 -34.31 -20.82 *** -30.40 *** (-2.03) (-1.89) (-2.81) (-2.88) R squared 0.34 0.34 0.34 0.34 No. of observations 755 755 755 755 202 Annex L Annex L: Evaluation Approach Paper’ I. BACKGROUND AND RATIONALE Education and the international development agenda “All agree that the single most important key to development and to poverty alleviation i s education. This must start with universal primary education for girls and boys equally.. .” James Wolfensohn, January 1999l 1. Education i s central to the internationally adopted poverty reduction goals. This fact is recognized by the inclusion o f education in two o f the Millennium Development Goals (MDGs), namely those for universal primary education and gender equality in school enrolments. Support for education has also manifested itself in the Education for All (EFA) initiative. The Education for All declaration, made at Jomtien (Thailand) in 1990, gained international support through a partnership o f UNESCO, UNICEF, UNDP and the W o r l d Bank. I t was given a hrther boost by the Dakar W o r l d Education Forum in April 2000. The MDGs and EFA provide a basis for measuring progress on educational development. Each Millennium Development Goal has associated targets and indicators, and EFA has a set o f 18 core indicators. These indicators, listed in appendix 1, will be utilized in this study where practicable. 2. This evaluation will test some of the key assumptions behind the strategies being proposed to meet the MDG and EFA targets. Following Dakar, a framework for action was adopted based on 12 strategies, which embody the rationale behind the design o f much recent Bank lending to basic education, such as the need to engage c i v i l society at all levels o f educational development.’ The Bank’s o w n Education Sector Strategy can be considered as complementary to the Dakar li-amework, and incorporates elements such as curriculum reform and more accountable education management s y s t e m ~More .~ generally, the Bank has stressed the quality aspects o f EFA, stating that “many factors enter into the delivery o f adequate quality education, including interactive classroom pedagogies, effective multi-grade teaching techniques, the availability o f textbooks, instructional leadership from school principals, parental support, community involvement in school management, and the existence o f student assessments to make schools more accountable for learning pr~gress”.~ * This paper was produced by Howard White (Task Manager, OEDST) under the guidance of Alain Barbu and Roy Gilbert with inputs from Helen Abadzi, Martha Ainsworth, Soniya Carvalho, Osvaldo Feinstein, Nils Fostvedt, Patrick Grasso, Bill Hulbert, Greg Ingram, and Nalini Kumar. Inputs were also received from Benoit Millot, Rene Bonnel, Eunice Depaah, Xiao Ye at a review meeting held on September 24th, 2002. 1. Quoted in World Bank Education Sector Strategy, July 1999, p. iii. 2. UNESCO The Dakar Framework for Action, Paris, 2000. 3 The design paper shall elaborate upon EFA and the link with the Bank’s own strategy. 4. “Education for Dynamic Economies: Accelerating Progress Toward Educationfor All”, submission to Development Committee, September 2001, Education Sector, World Bank. 203 Annex L Education in Ghana 3. Ghana’s education sector, once one o f the most respected in Africa, has come to embody many o f the challenges faced by the sector across the continent. During the 1980s enrolments fell, with gross primary enrolment falling from 80 per cent in 1980 to just 69 per cent by 1987.5 The quantity as well as the quality o f education suffered, as non-salary recurrent expenditures were squeezed out, falling real wages and frequent late payments demoralized the teaching force. Meanwhile government spending was excessively oriented toward the tertiary sector. Over the last 15 years the government has been tackling these problems with considerable Bank support. 4. The government embarked o n an ambitious reform program in 1987, supported by two World Bank Sector Adjustment Credits (EdSAC I and 11, the first SECALs t o education), which reduced the length o f pre-university education from 17 to 12 years, introduced curriculum reform for a greater vocational element, placed a ceiling o n educational recruitment and eliminated untrained teachers. Whilst in principle there has always been free universal primary education in Ghana, fees charged at local level have been one factor in restraining enrolments. Free Compulsory Universal Basic Education (FCUBE), introduced in 1996, aimed at eliminating these fees. Since 1997 there has been decentralization o f the sector, including increased community management and accountability, through the introduction o f School Management Committees and School Performance Assessment Meetings. The Bank has supported increases in the quantity and quality o f primary education through two projects: Primary School Development Project (1994-1998) and Basic Education (1996-2002) which included components such as school-building, teacher training and interventions to improve school management. 5. Ghana’s education sector was chosen as the subject o f this impact evaluation for a number o f reasons. First and foremost are the range o f policy-relevant evaluation questions to be addressed regarding h o w government and the Bank can support improved educational outcomes - and hence a test o f the strategies being pursued to achieve EFA. A second factor i s the prominent role o f the Bank in supporting the sector over the last 15 years, with a sizeable portfolio to form the subject o f the evaluation (see paragraph 13 below), including the earliest example o f a sector program since several donors parallel co-financed the EdSA(k6 Third, are strong complementarities with other activities, notably Education for All’ (for which Ghana is one o f the pilot countries), WDR 2003, the planned OED education sector review’ and the on-going Joint Evaluation o f External Support for Basic Education in Developing Countries supported by the Netherlands.’ Fourth i s the availability o f suitable baseline data from the second round o f the Ghana Living Standards Survey (GLSS) in 1988, 5. Data from United States Statistical Information Service, which reports the most complete data series available. 6. A number of other donors have been actively involved in the sector, notably USAID and the UK (formerly ODA, now DFID). 7. Ghana i s a pilot country for the EFA “fast-track” initiative. 8. The OED review will take Uganda and Malawi as case studies. Hence Ghana adds a west African case to the l i s t of countries being studied. 9. Ghana i s one o f the cases in the Basic Education Evaluation. The documents for that study have been analysed to ensure that this study does not overlap with that evaluation. 204 Annex L which collected facility data from schools and carried out education tests o n all 9 to 55 year olds in a national sample o f households. 11. ADDRESSINGTHE KEY EVALUATION QUESTIONS T h e context for impact evaluation 6. Impact evaluation has taken various meanings at different times. The most common, which are not mutually exclusive, are: 0 A concem with the impact of an intervention o n welfare outcomes, meaning that it i s concemed with the final stage o f log-frame indicators. 0 Conducting a with versus without analysis, i.e. establishing the counterfactual. 0 Having a broader focus than merely a specific project, to examine the effect o f support t o a sector, or even country. 0 An analysis o f sustainability, by analyzing the lasting effects o f an intervention several years after it has been completed. 7. Over twenty years, OED produced over 70 Impact Evaluation Reports (IERs).” A preliminary review o f these reports shows that each o f the different meanings o f impact has been used. In addition to the work o f OED, the W o r l d Bank’s Research Department (Development Economics and Chief Economist, DEC) has been engaged in impact evaluation, including a research project entitled “Impact Evaluation o f Education Reforms”. Less recently, D E C sponsored the 1988 data collection and analysis o f educational achievement in Ghana.” DEC’s analyses are mostly concemed analyzing the welfare impact o f public policy. They do not share OED’s mandate o f focusing o n the impact o f specific Bank-supported interventions. 8. This evaluation will embrace all four meanings o f the term impact, though the key focus i s o n a counterfactual analysis o f project and welfare outcomes. Earlier OED studies often had difficulties in establishing a satisfactory counterfactual o n account o f the lack of baseline data. The design o f this study takes advantage o f a nation-wide survey conducted in 1988. Evaluation questions 9. This impact evaluation i s concerned with final outcomes and the role o f the World Bank in achieving those outcomes. This evaluation will focus o n four questions: (1) What are the determinants o f educational outcomes (that is, educational achievement)12 for children o f 10. Imagebank lists 72 separate IERs covering the period 1979-1999. Preparatory work for this evaluation will review previous education studies by OED, other education-related evaluation work at the Bank, and evaluations of the education sector undertaken by other agencies. For a review of these documents see Anju Gupta Kapoor “Review o f impact evaluation methodologies used by OED over the past 25 years”, OED Working Paper, 2002. 11. See, for example, Paul Glewwe (1991) “Schooling, Skills and the Returns to Government Investment in Education: an explorationusing data from Ghana” Living Standards Measurement Survey 76, Washington D.C., World Bank. 12. In this study “educational achievement” refers to test scores and “educational attainment” the highest level o f education attained. 205 Annex L primary-school age in Ghana? (2) Which education interventions (“treatments” in evaluation terminology, drawing on the analogy o f medical research) have the greatest impact on the determinants o f educational outcomes?; (3) What has been the role o f the Bank in promoting education interventions which result in improved educational outcomes?; and (4) H o w do educational outcomes in Ghana promote improved welfare outcomes? 10. The following points should be observed with respect to the above: 0 The evaluation concerns primary education outcomes and will not in general be concerned with Bank support to secondary, tertiary or non-formal education. 0 The evaluation will judge the impact o f Bank projects and policy advice. This impact has been achieved through both (a) the creation o f school infrastructure, provision o f materials and teacher training and (b) institutional reform supported by a number o f agencies. In the latter case attribution will not always be possible. But it will be possible to say if the types o f reform supported by the Bank have been beneficial for education outcomes. 0 Impact here refers to both educational outcomes and the consequent improvements in socio-economic well-being (higher income, reduced mortality etc.). This evaluation will, to the extent possible, be concerned with both o f these. 0 OED evaluation is objective-based, so that the precise formulation o f the above evaluation questions will reflect the stated objectives o f the four projects under review, and the implicit strategy for achieving the international development goals for education as embodied in the Millennium Development Goals and EFA. 0 Some evaluations judge project impact by including a project dummy variable in a multiple regression to establish the determinants o f the relevant outcome variable. Such an approach i s unable to explain why particular project interventions have, o r have not, had the desired effects. Utilizing a theory-based approach built around a log-frame, this evaluation will combine a process-oriented approach with regression- based impact analysis, and hence “open the black box” o f what i s happening inside projects. This approach involves modeling the determinants o f the desired outcomes, and linking those determinants to the specific interventions supported by the Bank. The corresponding steps in the analysis are out-lined below. The Approach 11. Since 1986 there have been 10 Bank projects in support o f the Ghana education sector totaling U S $ 302 m i l l i o n in IDA credits (see Appendix 2),13 representing at least 20 per cent o f external support to the sector.14 The focus o f this evaluation will be on the four projects 13. T h i s figure includes only the education component o f the Health and Education Rehabilitation Project. 14. Calculation based on data from D A C on-line database, from which data are not very reliable. A project listing from the Ministry o f Education for the 199Os, which accurately records all Bank projects, puts the Bank‘s share o f extemal support as high as 63 per cent. 206 Annex L identified in paragraph 4 above which have supported primary e d ~ c a t i o n . ’However, ~ relevance shall b e addressed taking into account the whole o f the Bank’s education portfolio in Ghana: (i) examining the relative share o f the education portfolio in the light o f the country’s needs and priorities; (ii) analyzing the intra-sectoral composition o f Bank support against the priorities o f the Government o f Ghana and the Bank’s country strategies during the period; ( iii) comparing the objectives o f Bank projects with government’s o w n policy objectives and the most pressing policy issues o f the time. Box 1. Linking classroom building to increased enrolments School building supports higher enrolments. Although this seems obvious, it i s not at all obvious how to measure this effect. The f i r s t point i s that rehabilitation or expansion o f existing schools i s the norm, not the construction o f wholly new facilities. So if access (distance) i s the problem it i s not tackled by these projects. Second, even at existing schools, new classrooms may replace existing ones rather than be a net addition to the size o f the school. If this i s the case, parents may nonetheless be more willing to send children to school or there may be an indirect impact on enrolment through higher teacher motivation. Both effects seem probable if brick classrooms are replacing open, thatched structures, as are common in rural Ghana. When there i s a net addition in classroom size, the impact on enrolments i s not simply the net increase in classrooms time class size, since (1) demand i s needed to meet the supply, and (2) the increase in classrooms can be used to reduce class-size. The required approach i s econometric modeling o f the enrolment decision, with the determinants including variables affected by classroom building (such as average class size). 12. The evaluation criteria relate to different levels o f a log-frame. The log-frame provides the basis for a theory-based approach, since i t identifies the links f r o m activities to intended outputs and hence to outcomes. I t i s therefore particularly suited for an impact evaluation, which seeks not only to measure project impact but to identify the factors behind achieving that impact. Appendix 3 shows a log-frame for the support to formal basic education. The log-frame itself i s purely descriptive. The analytical challenge comes in testing the l i n k s from one box to another. This i s far from straightforward, as considering just one example demonstrates (see B o x 1).l6Identification o f indicators for each level o f the log- frame will take into account the EFA core indicators and the MDGs. For example, the analysis o f enrolments will be disaggregated by gender and measurement o f teacher quality will use EFA core indicators such as the percentage o f teachers who have attained the required academic qualifications and who are qualified to teach by national standards. 13. Following this log-frame, the key steps in the analysis are as follows: 0 Documenting the activities supported by the W o r l d Bank, which cover both improved supplies and facilities and institutional development.” This step identifies the 15. The Junior Secondary Schools in Ghana’s education systems cover grades usually counted as primary. 16. This log-frame i s based on an analysis o f the relevant project documents. The design paper for this evaluation will set the analysis in the context o f EFA-related strategies. 17. Institutional development refers to both activities and the outcomes f r o m both those activities and other activities with less direct institutional development effects. For primary education projects institutional development m a y b e addressed at three levels: (1) central government level, with a focus o n the Ministry o f Education and Govemment Education Service; (2) 207 Annex L interventions, or treatments, which are to be the subject o f the evaluation. These activities result in project outcomes. Insofar as physical outputs are concerned, these are determined from project documentation and project MISS. Determining the role o f the Bank in institutional development, including reforms, requires a qualitative approach. Substantial reforms took place during the period under review, with the Bank as a k e y player in supporting these reforms. 0 Quantifying the link from Bank-supported project outcomes to school-level outcome variables measuring the quantity and quality o f schooling.’* Specific questions concern the impact o f classroom building, changes in school management and teacher quality (skills and motivation) o n enrolments. The design paper will l a y out the scope o f the analysis more hlly. 0 Analyze the significant determinants o f educational outcomes (modeled as both levels and changes over time), as measured by achievement in individual-level test outcomes. These determinants include the school quality variables affected by project activities. 0 Examine the impact o f educational achievement and attainment o n socio-economic well-being. 14. A specific example o f the approach to attributing impact i s thus as follows. W o r l d Bank support resulted in the building o f x number o f classr~oms.’~The increase in the number o f classrooms reduced class size toy, which has a z impact o n school enrolments, and a change o f z leads to a w improvement in welfare. This is just one channel, as new classrooms can also affect the pupil-teacher ratio and teacher motivation. 111. METHODOLOGY 15. The range o f evaluation questions requires a mixed-methods approach. Issues o f institutional development are mostly dealt with through qualitative methods (document review and key informant interview) whereas measurement o f efficacy relies more o n quantitative methods. Table 1 summarizes the various approaches likely to be used to address different questions.” ~~ the capacity o f local government officials dealing with the education sector; and (3) school management (both headmasters and PTA members). 18. The OED evaluation criterion eficacy can be assessed against project outputs (e.g. numbers o f teachers trained and classrooms built), intermediate outcomes (higher enrolments, better classroom methods, improved school management), and final outcomes (improved learning outcomes and consequent socio-economic well-being). 19. This figure i s the actual number constructed adjusted for “replacement effects”, whereby n e w classroom replaced o l d ones. Where replacement occurs the possible impact o f the quality o f school infrastructure on school enrolments and achievement needs to be allowed for. 20. Subject to change during formulation o f precise evaluation questions. 208 Annex L T a b l e 1. Data collection methods Document Key Secondary School Household/ review informant data survey individual interviews analysis survey Institutional development and implementation of reform: X X Central government X X Local government X X X School management X X Teacher morale and methods X X Educational outcomes: Enrolments X X X X Learning outcomes X X Intermediate variables: School-building X X Teacher training X X 16. The initial document review will map out a time-line for the sector and the Bank’s involvement. This process will generate the objectives o f Bank support and the specific interventions (treatments) which have been applied. These objectives and interventions shall be set in the context o f EFA-related strategies. This analysis will lead to the development o f the related evaluation questions, and hence a toolkit to guide the qualitative fieldwork, which will examine the process o f reform and the role o f the Bank in that process.21This qualitative fieldwork will comprise interviews with k e y informants at both national and local level and visits to schools in both urban and rural areas. Organizations to be covered include relevant government agencies (MoE, GES, and local government officials), the teachers’ union, headmasters’ association, and PTAs at the local level. 17. Ghana i s rich in secondary data, including a computerized Education Management Information System (EMIS). The initial review will document what data are available and l i s t the existing studies made using these data.22Possible gaps relevant to this study will be identified and filled though commissioned The OED study will also utilize existing data rather than duplicate existing data collection. 18. The main quantitative data collection tool will be a household survey modeled o n the 1988 GLSS. Specifically, fifty o f the same communities will be re-surveyed (but not the same individuals), applying a reduced version o f the questionnaire used in 1988, including the 21. Toolkits were developed by the Public Sector Management anchor, and adapted for OED’s review o f social funds. Although analysis o f the reform process will be largely qualitative, quantitative indicators of reform, such as budget analysis, shall also be developed, partly to triangulate the different approaches. Where possible, these indicators will be based on EFA and MDG-related indicators, and on key performance indicators from the Bank credits. Recording progress on reform i s one thing, attributing responsibility i s another. I t i s not possible to prove attribution for policy reform. The evaluation seeks to establish “plausible significant influence” o f the Bank on policy outcomes. 22. A synthesis study was underway for the MoE during the preliminary field visit in May of this year. The results o f that study should be available for this evaluation. 23. For example, data are available on school-level exam results since 1987. It seems that no detailed analysis has been made of these data. 209 Annex L educational tests for math, EnglisWlocal language and a reasoning (Ravens) test.24The school survey (comprising a facility survey and separate teacher questionnaire) will also be repeated, in an expanded form to capture more aspects o f school management and quality o f s ~ h o o l i n g . ~ ~ These data will allow modeling at the individual, household and community levels to examine, for example, how school-building and changes in school management affect enrolments, and how classroom practices and teacher motivation affect pupil’s educational performance. The availability o f household data will make it possible to control for external factors. Community- level data will be linked to the 1988 data to examine the determinants in changes in community-level enrolment and community average educational scores over the fifteen-year period.26Using this approach, attribution to Bank-support is i n d i r e ~ t . ~ Specifically, ’ and as shown by the log-fi-ame, the analysis will establish which interventions, o f the sort supported by the Bank, have a significant impact on educational achievement. 19. The facility-level data will constitute a panel o f schools, allowing examination o f school-level changes for over 50 schools over the 15-year period. The purpose o f this analysis i s partly descriptive: how have schools fared over the last 15 years? The analysis will also help address issues o f sustainability. Questions include: What i s the current state o f Bank-supported Are teachers and government officials who have received Bank-financed training s t i l l working in relevant positions to utilize that training? Are Bank- supported changes in teaching methods and school management being applied? 20. The links between educational outcomes and socio-economic well-being are well documented. This study will apply established methods to the primary data. An analysis shall possibly be made o f rates o f retum to education, but examining the rate o f return to educational ) , ~ ~ than attainment (years o f schooling), and estimates achievement (education S C O ~ ~ Srather made of the impact o f achievement on nutrition and fertility. 24. The 1988 study used only a in English test. However, an important debate concerns the differential effects of literacy in a local language and English (or equivalent). 25. Questions of this sort are available from the Institute for Educational Quality, a Washington-based organization which has carried out extensive work for USAID, including in Ghana in the mid-90s. 26. The availability o f surveys at a fifteen year interval offers a unique opportunity to describe changes in educational inputs and outputs over a fifteen year period. Using these data for analysis of determinants has the advantage of removing community-level fixed effects (by differencing). However, a problem i s that observations of school quality are made at two points of time, whereas some many children covered in the survey may been educated under a different school regime in, say, the mid-90s. This problem can be tackled by restricting the sample to those in, or who have recently left, school. 27. The Primary School DevelopmentProject targeted support to 1,983 schools so a direct approach would be to isolate “Bank-supported schools” from other schools and conduct a control group analysis. I t i s only worthwhile to compare school- level variables for Bank-assisted schools and others if a suitable control can be established - but there are limited other data to construct such a control, which would be especially difficult since the 1,983 were chosen as “the most disadvantaged”. Moreover, a control of this sort cannot say what i s what about the intervention which “worked”. 28. Data from EMIS can (and may) be used to track changing scores in the 1,983 schools against a control of other schools, noting again the difficulties o f establishinga satisfactorycontrol group. 29. However, the survey will collect expenditure data rather than income. The pros and cons of such an approach are discussed in the design paper. 210 Annex L IV. DISSEMINATION 21. During the preliminary visit in M a y 2002 considerable enthusiasm was expressed by M o E officials for a launch workshop in Accra, which i s a good opportunity to define the scope for other necessary work. This workshop will be held as soon as possible, most likely late November 2002. A further workshop will be held in Accra to present preliminary findings to government, donors, NGOs and teachers’ representatives. Given the proposed collaborative nature o f the program o f impact evaluations, allowance i s also made for presentations to other major donors (e.g. DFID in London). v. COLLABORATION WITH OTHER AGENCIES 22. This evaluation seeks to build up capacity for such evaluations amongst both other agencies and borrower governments. They will do this by operating in a collaborative manner. This study i s being partly financed from DFID resources, and co-operation with DFID staff in Accra i s being sought.3o Discussions were held with relevant government agencies during the preliminary field visit in M a y 2002, and Ghana Statistical Services identified as the likely partner to conduct the survey. A firm basis for collaboration with the Ministry o f Education (MoE) and Government Education Service (GES) was established at that time. Collaboration will be sought with other donors active in the sector, such as USAID. The Ghanaian Evaluation Association will be contracted regarding possible collaborators. VI. MANAGEMENT SCHEDULE AND TASK 23. The inception phase o f this study, comprising the initial document review and compiling o f the questionnaires, has taken place in the period from August to October 2002. A design paper, including draft questionnaires, have been produced as a part o f this process. A preliminary field visit in October oriented the questionnaires to the current realities o f the Ghanaian education system. Data collection is to be undertaken by Ghana Statistical Services (GSS). The pre-test o f the survey instruments i s scheduled for November and the survey itself in January and February 2003. The researcher from the evaluation team will accompany the survey teams, with the task manager present for some o f the time. The second phase, data analysis, will begin in M a y 2003, with a first draft report for internal OED distribution by late July 2003, and a draft for management review by early October 2003. The report will b e sent to the Committee on Development Effectiveness (CODE) early November 2003. 24. The commissioned studies will be undertaken parallel with the above activities and December 2002. Currently envisaged studies are: (1) the political economy o f are due 3 lSt education reform and the role o f the World Bank, and (3) curriculum reform. 25. The evaluation will prepared by a team o f OED staff and consultants with the assistance o f Ghanaian government officials and consultants under the Task Management o f Howard White (OEDST). An advisory panel will be appointed to review the proposed evaluation design and drafi final report. 30. For example, participating in fieldwork or commissioning o f parallel studies. 21 1 Annex L VII. BUDGET 26. The program o f impact studies i s being supported by the DFID-OED partnership agreement. The total budget for this study i s $500,000. 212 Annex L Appendix 1. MDG and EFA Indicators Education-related MDGs, targets and indicators Goals and targets Indica tors Goal 2 -~ : Achieve universalprimary education Target 3: Ensure that, by 2015, children 6. Net enrolment ratio in primary education everywhere, boys and girls alike, will be 7. Proportion of pupils starting grade 1 who reach grade 5 able to complete a full course of primary schooling 8. Literacy rate of 15-24 year-olds Goal 3: Promote gender equality and empower women Target 4: Eliminate gender disparity in primary 9. Ratios of girls to boys in primary, secondary and tertiary and secondary education preferably by 2005 education and to all levels of education no later 10. Ratio of literate females to males of 15-24 year-olds than 2015 Core EFA Indicators ~ 1 Gross enrolment in early childhood development programs, including public, private, and community programs, expressed as a percentage of the official age-group concerned, if any, otherwise the age- group 3 to 5. 2 Percentage of new entrants to primary grade 1 who have attended some form of organized early childhood development program. 3 Apparent (gross) intake rate: new entrants in primary grade 1 as a percentage of the Population of official entry age. 4 Net intake rate: new entrants to primary grade 1 who are of the official primary school entrance age as percentage of the corresponding population. 5 Gross enrolment ratio. 6 Net enrolment ratio. 7 Public current expenditure on primary education a) as a percentage of GNP; and b) per pupil, as a percentage of GNP per capita. 8 Public expenditure on primary education as a percentage of total public expenditure on education. 9 Percentageof primary school teachers having attained the required academic qualifications. 10 Percentageof primary school teachers who are certified to teach according to national standards. 11 Pupil teacher ratio. 12 Repetition rates by grade. 13 Survival rate to grade 5 (percentage of a pupil cohort actually reaching grade 5). 14 Coefficient of efficiency (ideal number of pupil years needed for a cohort to complete the primary cycle, expressed as a percentage of the actual number of pupil-years). 15 Percentageof pupils having reached at least grade 4 of primary schooling who master a set of nationally defined basic learning competencies. 16 Literacy rate of 15-24 year olds. 17 Adult literacy rate: percentage of the population aged 15+ that is literate. 18 Literacy Gender Parity Index: ratio of female to male literacy rates. Source: Education for All Assessment: StatisticalDocumentation,World Education Forum, Dakar, April 2000, Awendix II 213 Annex L APPENDIX2. BANK EDUCATION SUPPORT TO GHANA SECTOR Project ID Budget Rating* Status IDA Total Health and education rehabilitation PO00876 18.0 18.1 n.a. Closed o/w education component 6.1 Education sector adjustment PO00891 38.3 45.5 S Closed Education sector adjustment II PO00896 53.2 S Closed Community and secondary school construction PO00954 14.7 19.6 S Closed Literacy and functional skills PO00917 27.8 S Closed Tertiary education P000933 44.8 51.O Marg. S Closed Primary school development PO00964 53.2 56.6 Marg. U Closed Basic education PO00975 34.7 S To close 12/02 Vocational skills and informal sector PO00948 5.8 U Closed National functional literacy program PO00974 23.7 S (impl: U) To close 12/04 Note: *ICR (or PSR for current projects). 214 Annex L APPENDIX3. FLOW FOR SUPPORT TO BASIC EDUCATION CHART LOG-FRAME Education budget allocations Teacher training + Teacher no. and quality Non-salary recurrent Classroom expenditure per pupil construction (equipment and materials) Stul :nts/classroom and access I k - lI 1 Eligible cohort of 1 I Teacher morale I I children. Household characteristics and Reforms in opportunity cost. school management ? I Enrolments t 4 b ~~ School quality 4 I Human capital Quantity (enrolments and completion) Quality (test scores) 1 Welfareoutcomes 1 215 Reference References Akyeampong, Kwame and David Stephens. 2002. “Exploring the backgrounds and shaping o f beginning teachers in Ghana: toward greater contextualisation o f teacher education.” International Journal of Educational Development 22 December 26 1-274. Alderman, Harold. 1990. “Nutritional status in Ghana and its determinants.” SDA in sub- Saharan Africa Working Paper No. 3. Washington D.C.: World Bank. Appiah, Francis et al. 2000. “Fiscal Decentralization and sub-National Government Finance in Relation to Infrastructure and Service Provision in Ghana” report for U S A I D and DANIDA. Barro, R. And J-W Lee. 2000. “International data o n educational attainment updates and implications” NBER Working Paper 79 11 [Cambridge, MA: NBER]. Benefo, K o f i Darkwa and T. Paul Schultz. 1994. “Determinants o f fertility and child mortality in CBte d’Ivoire and Ghana.” LSMS Working Paper 103. Washington D.C.: World Bank. Bennett, Nicholas (no date). “Ghana Education Portfolio Implementation: the advantage o f decentralized management from the field” (http://kms.worldbank,org/docs/study6.doc). Bhalotra, Sonia and Christopher Heady (2003) “Child Farm Labor: The Wealth Paradox” World Bank Economic Review, Vol. 17, No. 2, 197-227. Boko, Sylvian. 2002. Decentralization and Reform in Africa. Nonvell, MA: Kluwer Academic Publishers. Bonner, Margaret, Cameron Bonner, Joy du Plessis, Ash Hartwell and Celia Marshall assisted by Chris Akwesi, Jon Atta Quayson and Leslie Casley-Hayford. 200 1. QUIPS Mid-Term Assessment. Report prepared for USAID/Ghana. Boom, J.M. Van Den and W.K. Asenso-Okyere. 1996. “Nutrition, labor productivity and f Development Studies 32 80 1- labor supply o f men and women in Ghana.” Journal o 829. Canagarajah, Sudharshan and Harold Coulombe (1997) “Child Labor and Schoolina in Ghana,” Research Paper 1844. Washington D.C.: World Bank. Canagarajah, Sudharshan, and Xiao Ye. (2001). “Public Health and Education Spending in Ghana in 1992-98: Issues o f Equity and Efficiency,” Policy Research Working Paper 2579. Washington DC: World Bank. Chapman, D., E. Barcikowski, M. Sowah, E. Gyamera, and G. Woode. 2002. “Do communities know best? Testing a premise o f educational decentralization: community members’ perceptions o f their local schools in Ghana.” International Journal o f Educational Development 22 18 1-189. Cobbe, James. 1991. “The political economy o f education reform in Ghana” in Donald Rothchild (ed.) Ghana: the political economy of recovey . Boulder and London, L y y n e Rienner Publishers, pp.101-118. Commonwealth Local Government Forum. 2002. Commonwealth Local Government Handbook: Ghana profile. London: CLGF. 216 Reference Condy, Ann. 1998. “Improving the quality o f teaching and learning through community participation: achievements, limitations and risks. Early lessons from the schooling improvement fund in Ghana.” Social Development Working Paper, Social Development Division, DFID, London. DFID. 2003. Development Effectiveness Report. London: DFID. EARC. 2003. “Teacher time on task,” report for USAID, mimeo, Osu. Fobih, D., K. Akyeampong, K. and A. Koomson (1999) Ghana Primary School Development f Project Performance. Accra: Ministry o f Education Project: Final Evaluation o Glewwe, P. and J. Desai. 1999. “Child Health and Mother’s Education in Ghana” in P. Glewwe, The Economics o f School Quality Investment in Developing Countries: an empirical study o f Ghana. Basingstoke: MacMillan and N e w York: St. Martin’s Press, pp. 295-326. Greene, W.H. (2000) Econometric analysis. Upper Saddle River, NJ: Prentice Hall. Gupta, Neeru and Mary Mahy. 2003. “Adolescent childbearing in sub-Saharan Africa: can increased schooling alone raise ages at first birth?” Demographic Research 8(4) 93-106. Heady, Christopher. 2000. “What i s the effect o f child labour on learning achievement? Evidence from Ghana.” Innocenti Working Paper 79 Florence, UNICEF. Hedges, John. 2002. “The importance o f posting and interaction with the education bureaucracy in becoming a teacher in Ghana.” International Journal o f Educational Development 22 353-366. Heyneman, S.P. 2003. “The History and Problems o f the Making o f Education Policy at the World Bank 1960-2000.” International Journal o f Educational Development 23 December 315-337. Holmes, J. (2003). Measuring the determinants o f school completion in Palustan: analysis o f censoring and selection bias, Economics o f Education Review, 22(3), 249-264. Johnston, John and Irvine John DiNardo (1997) Econometric Methods, Columbus, Ohio: McGraw-Hill Jolliffe, D. 1999. “The Impact o f Cognitive Slulls on Income fi-om Farming” in P. Glewwe The Economics o f School Quality Investment in Developing Countries: an empirical study of Ghana. Basingstoke: MacMillan and N e w Y o r k St. Martin’s Press, pp. 253-293. King, Elizabeth and Lee Lillard (1987) “Education Policy and Schooling Attainment in Malaysia and the Philippines,” Economics of Education Review. 6(2): 167-18 1. Knight, John. and k c h a r d H. Sabot (1990) Education, Productivity, and Inequality The East African Natural Experiment. Oxford: OUP for the World Bank Kraft, Richard, B.K.Adipah, J.M. Dzinyela, K. ‘Anku, K. Gala, F. Godwyll, E. Goody, L. Larcom and B. Okyere (1995) A Tale o f Two Ghanas: The View from the Classroom, Accra: MOENSAID Lavy, Victor. 1995. “The impact o f the quality o f health care on children’s nutrition and survival in Ghana.” LSMS Working Paper 106. Washington D.C.: World Bank. Leach, Fiona, Vivian Fiscian, Esme Kadzamira, Eve Lemani and Pamela Machakanja. 2002. “An investigative study o f the abuse o f girls in African schools.” Education Department, DFID, London. 217 Reference Lloyd, C.B. and A.J. Gage-Brandon. 1994. “High fertility and children’s schooling in Ghana: Sex differences in parental contributions and educational outcomes,” Population Studies, Vol. 48 (2): 293-306. Maddala, G.S., (1983). Limited-dependent and qualitative variables in econometrics. New York: Cambridge University Press. Maxwell, D., C. E. Levin, M. Annar-Klemesu, M. T. Ruel, S. Morris, and C. Ahiadeke. 2000. “Urban livelihoods and food and nutrition security in Greater Accra, Ghana.” Research Report 112. Washington, DC: IFPRI. McWilliam, H.0.A and M.A. Kwamena-Poh. 1975. The Development o f Education in Ghana: an outline. London: Longman. Mettle-Nunoo and Louise Hilditch. 2000. “Donor participation in the education sector in Ghana.” www.elimu.org/_docs/old-site/countries/rpt-ghana.shtml. Millot, Benoit and Julia Lane. 2002. “The Efficient U s e o f Time in Education.” Education Economics lO(2) 209-228. MOEYS (2003) Education Sector Review (prepared by Kwasi Ansu-Kyeremeh, Leslie Casley-Hayford, J.S. Djangmah, James Nti and Francois Orivel) Accra: MOEYS. N’tchougan-Sonou, Christina. 200 1. “Automatic progression or large-scale repetition -which path to quality?” International Journal of Educational Development 2 1 149-162. Nugent, P. (1995), B i g Men, Small Boys and Politics in Ghana, Accra: Asempa Publishers. Okyere, Beatrice, Anthony Mensah, Harrison Kugbey and Abigail Harris. 1997. “What Happens to the Textbooks?’ American Institutes for Research for Improving Educational Quality Project (http://www.air,org/pubs/intemational/textbooks.pdf). Oliver, R. 1999. “Fertility and Women’s Schooling in Ghana” in P. Glewwe The Economics o f f Ghana. School Quality Investment in Developing Countries: an empirical study o Basingstoke: MacMillan and New Y o r k St. Martin’s Press, pp. 327-346. Peil, Margaret. 1995. “Ghanaian education as seen from an Accra suburb.” International Journal of Educational Development 15(3) 289-305. Ruel, M. T., C. E. Levin, M. Armar-Klemesu, and D. Maxwell. 1999. “Good care Practices can mitigate the negative effects o f poverty and l o w maternal schooling o n children’s nutritional status: evidence fkom Accra.” World Development, 27, 1993-2009. Sackey, H.A. 2003. “The Effects O f Education And Economic Activity On Demographic Pattern nGhana”, mimeo. Winnipeg: Department o f Economics, University o f Manitoba. I Sawyer, Harry. 1997. “Country-led aid coordination” (www.adeanet.org/vol9no3/ghana- eng.htm1) Sowa, Nii and Howard White (2003) “Politics and Partnership in Ghana in the 1980s” in Howard White and Geske Dijkstra Program Aid and Development. London: Routledge. Tsikata, Yvonne. 2001. “Aid and reform in Ghana” in David Dollar, Shantayanan Devarajan, Torgny Holmgren (eds.) Aid and Reform in Africa. Washington D.C.: World Bank. Teal, Francis (2001) ‘Education, incomes, poverty and inequality in Ghana in the 1990s’, Working paper WPS/2001.21. Oxford: CSAE, University o f Oxford. 218 Reference Van Donge. 2002. “Into the black box o f Ghanaian education: why do increased inputs not lead to better educational outputs?” Mimeo. The Hague: ISS. White, Howard and Geske Dijkstra. 2003. Program Aid and Development: Beyond Conditionality. London and New Y o r k Routledge.