Policy Research Working Paper 10844 Can Targeted Allocation of Teachers Improve Student Learning Outcomes? Evidence from Malawi Salman Asim Ravinder Casley Gera Martin Moreno Kerry Wong Education Global Practice July 2024 Policy Research Working Paper 10844 Abstract Teachers are one of the most important inputs for learn- successful just 22 percent. Using administrative data, the ing, but in many low-income countries they are poorly paper identifies the impacts on student repetition rates of distributed between schools. This paper discusses the case of reductions in pupil–qualified teacher ratios as a result of the Malawi, which has introduced new evidence-based policies new teachers. The findings show that schools that moved and procedures to improve the equity and efficiency of the from having more than 90 pupils per qualified teacher to a allocation of teachers to schools. The analysis finds that lower ratio experienced reductions in lower primary school adherence to these policies has been highly variable between repetition rates of 2–3 percentage points. However, similar the country’s districts, with the most successful deploying impacts on dropout are not observed. 75 percent of teachers according to the rules and the least This paper is a product of the Education Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at sasim@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Can Targeted Allocation of Teachers Improve Student Learning Outcomes? Evidence from Malawi Salman Asim*, Ravinder Casley Gera±, Martin Moreno∞, and Kerry Wong≠ JEL classification: C1; I21; I28; D73; D78 Keywords: Econometrics; Education Policy, Education Quality, Teacher, Bureaucracy; Policy implementation *Senior Economist, World Bank; sasim@worldbank.org ± Education Specialist, World Bank ∞ Economist, World Bank ≠ Assistant Professor, London School of Hygiene and Tropical Medicine Table of Contents 1. Introduction ................................................................................................................. 3 2. Rules-based targeting of new teachers to schools.......................................................... 5 Maximizing the efficiency of allocation rules ............................................................................... 6 T1: One teacher per grade................................................................................................................................ 6 T2: Allocation to schools with high PTRs .......................................................................................................... 6 Estimating need under the revised rules...................................................................................... 7 3. How well did teacher allocation match the modified rules? ......................................... 10 Total deployment at the national level ...................................................................................... 10 Allocation of teachers to districts .............................................................................................. 10 Allocation of teachers to schools ............................................................................................... 11 Share of need met at the school level ........................................................................................ 13 4. Impact of teacher allocation on school staffing ........................................................... 14 Teacher headcounts ................................................................................................................. 14 PTR change............................................................................................................................... 15 5. Impact of improved staffing on student outcomes ....................................................... 16 Repetition ................................................................................................................................ 18 Dropout ................................................................................................................................... 20 Robustness checks ......................................................................................................................................... 21 6. Conclusion ................................................................................................................. 21 7. References ................................................................................................................. 23 8. Tables ........................................................................................................................ 25 2 1. Introduction Low-income countries, particularly in Sub-Saharan Africa, have typically struggled to deploy teachers to schools in remote areas, resulting in large variations in staffing levels between schools (Mulkeen, 2010; Majgaard and Mingat, 2012; Bashir et al., 2018). These variations stem primarily from teachers’ preferences, typically for placements near to towns and larger villages, known as trading centers (Asim at al., 2019). Countries employ a range of rules and procedures to attempt to address these disparities, ranging from administrative policies to legal standards. The impact of these efforts, however, is highly variable. In India for example, maximum school PTRs are legally mandated by the Right to Education Act (2009); however, as of 2013 three provinces had an overall PTR above the legal maximum (Azim Premji, 2014). Teachers employ a range of formal and informal means to apply pressure to obtain desirable postings, and in many countries officials lack adequate incentives to enforce the rules (Cummings et al., 2016; Asim et al., 2019). In some countries, including some provinces of India, elected representatives have formal right of veto over teacher movements which can interfere with efforts to enforce PTR rules (Ramachandran et al., 2018). Additional incentives for teachers to remain in posting in remote areas, such as hardship allowances, are often required in order to achieve rationalization of teacher distributions, such as successfully tried in The Gambia (Pugatch and Schroeder, 2014). In Malawi, one of the poorest countries in Sub-Saharan Africa, inequities in teacher distribution are particularly acute. While the national pupil-teacher ratio (PTR) is 62, 10 percent of schools have a PTR of 93 or more, while a more fortunate 10 percent have a PTR of 36 or lower. These inequities occur between districts and sub-district areas, but primarily occur within a single sub-district area (zone): school PTRs within a single zone can vary by a factor of ten or more (Ministry of Education, 2021a). This inequitable distribution of teachers contributes to Malawi’s poor levels of student retention and progression. More than 5 percent of students drop out in Grade 1,1 with dropout rates rising to almost 7 percent in upper grades (Malawi Longitudinal School Survey, 2021). Fewer than two-thirds of students entering Grade 1 are still in school by Grade 5. Those students who remain in school frequently repeat grades:2 more than one-third of Grade 1 students repeat, and repetition rates are above 18 percent in all grades (Ministry of Education, 2021b). Those students who do progress in school nevertheless achieve low learning outcomes: in Grade 2, fewer than 25 percent of students achieve minimum proficiency levels in the Early Grade Reading Assessment (EGRA) conducted by the United States Agency for International Development (USAID). At Grade 6, fewer than 25 percent of students achieve minimum proficiency levels in the Southern African Consortium for Monitoring Educational Quality (SACMEQ) assessment in Mathematics, placing Malawi near the bottom in the region. The persistence of this poor distribution of teachers reflects weaknesses in Malawi’s systems of national- and district-level resource allocation. Until recently, the rules governing allocations were excessively broad, requiring only that teachers be sent to schools with a PTR above 60. With more than half of schools above this level, this introduced a high degree of discretion in allocations.3 Teachers exploit this discretion, using formal and informal channels of influence to resist placements 1 Grades are known as Standards in Malawi, but will be referenced as Grades throughout this paper. Grades range from Grade 1-12. Primary Education is from Grade 1-8, and Secondary Education from Grade 9-12. 2 Most Malawian schools require students to repeat grades if they fail to pass an end-of-year examination. 3 In 2021, 61 percent of public primary schools had a PTR above 60. 3 in remote schools, resulting in continued disparities. Even where teachers are successfully placed in schools in remote areas, they may not remain for a significant period of time, exploiting these same formal and informal channels to obtain transfers to more desirable postings. A ‘rural allowance’ scheme is in place to motivate teachers to remain in remote postings, but it too is inadequately targeted with minimal incentive effect (see Asim et al., 2019, for a detailed exploration of these dynamics). In recent years, a number of steps have been taken by the Ministry of Education (MoE) of Malawi, with support from the World Bank, to introduce more well-defined rules for allocation of teachers. Beginning in 2017, districts were advised to prioritize schools with the highest PTRs, not only those with PTR above 60. At the same time, districts were instructed to prioritize schools with fewer teachers than grades, in an attempt to address the common practice of multi-grade teaching in understaffed schools.4 In this paper, we assess the level of adherence to rules-based allocation of teachers in Malawi’s schools over the period 2017-2019; and the impact of improvements in allocations on school PTRs and on outcomes. We draw data from multiple rounds of administrative data from the Government of Malawi’s Educational Management Information System (EMIS) database. EMIS data is collected via an Annual School Census (ASC) and includes a wide range of data on school size, conditions, staffing, finances, infrastructure and equipment. We employ data on grades offered, staffing and enrollment to establish the schools which meet each of the criteria established by the government. In addition, we employ administrative data on the allocation of new teachers to schools provided by district-level officials via the MoE. Our ultimate outcomes of interest are student dropout and promotion rates. These outcomes are informed by the international evidence on the impacts of improvements in school teacher staffing and reductions in PTRs. We anticipate reductions in class sizes at the most overcrowded schools as a result of improved staffing, and our selection is also informed by the literature on the impacts of reductions in class sizes. There is substantial evidence for a relationship between reduced PTRs and improved test scores (Angrist and Lavy, 1999; Kreuger, 1999; Muralidharan & Sundararaman, 2013), although other studies find no significant effects (Hoxby, 2000; Duflo et al., 2015; Angrist et al., 2017). Reduced PTRs can lead to improvements in learning through two channels. First, an adequate supply of teachers can lead to reduced class sizes, with benefits for learning.5 Second, having an adequate level of staff can reduce schools’ reliance on multi-grade teaching where multiple classes are combined under a single teacher. Although the existing evidence-base is still limited, available research on the effects of multi-grade teaching suggests that it is harmful to student performance (Checci & De Paola 2017, Jacob et al. 2008). Using administrative data, we are unable to present direct analysis of test scores. However, in Malawi, student promotion is closely linked to test performance, with mandatory repetition in most schools for students who do not pass a year-end assessment; we therefore measure repetition rates as a proxy for learning. Walter (2018), assessing evidence from a panel of 20 high-, middle- and low-income 4 Although in most schools achieving this standard require eight teachers, Malawi includes a number of ‘junior primary’ schools which only teach Grades 1-4, as well as other schools without a complete set of eight grades. 5 As is the case in Angrist and Lavy, 1999 and Muralidharan & Sundararaman, 2013. 4 countries, estimates the gains in student promotion from optimal allocation of existing teachers between schools at between 0.1 and 4.2 percentage points. • In Chapter 2, we present a stylized algorithm, based on the MoE’s rules, to identify target schools in need of more teachers. Drawing on administrative data collected in 2017, 2018 and 2019, we apply this algorithm to identify all target schools in each year. We present estimated needs for teachers according to this algorithm, at national, district and school levels. • In Chapter 3, we compare our targeting algorithm to the actual allocations of new teachers in Malawi in each of the four years, and assess the extent to which allocation met with need, both nationally and at the district level. • In Chapter 4, we assess the impact of the allocation of new teachers on school PTRs. • In Chapter 5, we assess the impact of reduced PTRs on student repetition rates and dropout rates. • Chapter 6 discusses the results. 2. Rules-based targeting of new teachers to schools The allocation of newly deployed teachers – typically 3,000 to 5,000 per year – is a multi-stage process across multiple levels of government (see Asim et al., 2019, for a detailed discussion). Teachers enter the workforce having completed pre-service training under the auspices of the Department for Teacher Education and Development (DTED) of MoE, primarily through a program known as the Initial Primary Teacher Education (IPTE). Through the Directorate of Basic Education (DBE) and the Local Government Service Commission (LASCOM), MoE allocates teachers to each of Malawi’s 34 education districts;6 District Education Offices (DEOs) in each district then allocate teachers to specific schools. Allocating new teachers equitably to schools may seem a simple matter of targeting the schools with the highest PTRs, but this has not historically been the case. Malawi has an official target for school PTRs of 60 and, prior to 2017, teacher allocations were guided only by this target. However, with more than half of schools above this ratio, this approach did not significantly guide allocations. The result was that allocations of teachers to districts and, in particular, to schools was conducted with a high degree of discretion that left district officials subject to pressure from teachers to avoid postings in remote schools (see Asim et al., 2019, for a detailed description of Malawi’s teacher allocation system and the history of attempts to rationalize deployments). Beginning in 2017, a new set of targeting rules was introduced to support more equitable allocations of teachers to schools. In 2017, national and district officials were instructed to allocate teachers to schools with PTR below 60 if they did not have at least one teacher per grade offered. This was intended to prevent multi-grade teaching. Beginning in 2018, officials were instructed that, having fulfilled this first condition, they should target the remaining teachers to the schools with the highest PTRs. A revised Primary Teacher Management Strategy codified the new guidance (Ministry of 6 Malawi has 28 local government authorities, including districts and municipalities. Some of these are subdivided into two or three components for the purposes of education management, producing a total of 34 education districts. In this paper, we use ’district’ to refer to these education districts. The country’s four urban districts, Lilongwe City, Blantyre City, Mzuzu and Zomba Urban, are typically excluded from receiving newly deployed teachers owing to high existing levels of staffing. 5 Education, Science and Technology, 2018) and a spreadsheet-based tool was developed to guide districts in completing allocations according to the rules. In this section, we employ administrative data to create lists of schools that would have been eligible to receive new teachers in each of the years 2017-2019 according to the new rules. Maximizing the efficiency of allocation rules T1: One teacher per grade The first stipulation of the revised rules – to provide adequate teachers to schools to ensure that they have at least one teacher per grade offered – is intended to eliminate multi-grade teaching. This rule is most likely to be relevant in smaller schools, often in remote areas, which are the most likely to lack one teacher per grade owing to small enrollments.7 Such schools are large in number and concentrated in particular districts. In 2017, one district, Mzimba South, required 423 teachers to meet this rule, and the top five districts combined required 1,039 new teachers – more than 40 percent of all the new teachers needed to fully attain T1 in the whole country. These districts, unless experiencing severe overall shortages of staff, are poorly allocating their existing teacher resources and may not be expected to appropriately allocate additional teachers. In addition, meeting this need entirely would be likely to severely reduce the potential for PTR reduction at more understaffed schools, and would entail allocating a significant number of teachers to schools with PTR below 60 which are otherwise adequately staffed. This may not be an effective use of limited resources, given overall constraints on teacher availability. We therefore apply the following modifications to T1: • Schools with a PTR of 60 or below are excluded from receiving new teachers. • Schools whose PTR would drop to 60 or below on receipt of a new teacher, are excluded from receiving new teachers. • Districts where more than 50 percent of schools meet T1 are excluded from receiving teachers based on T1, unless their overall district PTR is above 75:1. T2: Allocation to schools with high PTRs The second stipulation, T2, is expected to be applied following fulfilment of the modified T1. In a context of perfect information, the preferred approach to allocating teachers in order to reduce PTRs would be the “smallest achievable maximum PTR rule” (Walter, 2018), in which teachers are allocated in order to reduce the largest school-level PTR, nationally or in a district, as much as possible. However, this approach, while the most efficient way to reduce PTRs in theory, is difficult to implement in practice as it requires a complete and accurate ranking of schools by PTR. As this may be complicated to implement, in practice, many countries adopt a simpler approach of identifying a target level of PTR above which schools are considered understaffed and providing teachers to bring 7 Of schools in the most remote areas, known as ‘Category A’, 38 percent have fewer than eight teachers, versus only 18 percent of the least remote ‘Category C’ schools (Ministry of Education, 2022; see Asim et al., 2019, for categorization). 6 schools below that level. For example, India establishes a target of 30 for each class (Azim Premji, 2014). In Malawi, as described above, the qualifying PTR level has historically been 60, but with more than half of schools having PTRs above this level, this targets teachers with inadequate specificity to bring down the maximum PTR effectively. Evidence from Malawi suggests that schools with PTRs above 90 typically achieve lower learning outcomes (Asim and Casley Gera, 2024). We therefore adopt a PTR of 90 as the threshold to consider schools highly understaffed and eligible to receive teachers under T2, for evaluation of the 2017 allocation of teachers. However, as the allocation of teachers improves, we would expect the share of schools with PTR above 90 to reduce, making adherence to this rule more difficult. Indeed, the share of schools with PTR above 90 reduced from 32 percent in 2017 to just 16 percent in 2019. Applying T2 only to these schools would not identify enough beneficiary schools to utilize the full supply of teachers. Therefore, for 2018 and 2019, as the share of needy schools declines in response to optimal teacher allocations, we lower the threshold to 80, maintaining a similar share of schools identified as needy. Over time, with continued need-based allocations, we would expect to be able to lower the threshold further and eventually to the government’s official target of 60. Box 2.1 summarizes the modified rules. Box 2.1. New rules for teacher allocation Target 1 (T1): deploys the least number of teachers needed such that schools would have at least one teacher per grade. • Except schools with PTR of 60 or below or whose PTR would be reduced to 60 or below with addition of new teachers • Except districts where more than 50 percent of schools meet T1 are excluded from receiving teachers based on T1, unless their overall district PTR is above 75:1. Target 2 (T2): deploys the least number of teachers needed so that PTR reduces to below 90 (in 2017) or 80 (in 2018 and 2019). Estimating need under the revised rules Employing these revised rules, we then estimate the need for teachers to fulfill these targets in each year from 2017 to 2019. Table 2.1 shows the number of teachers required in each district in each year following the application of both rules, along with the number of schools needing teachers. In each year, the data on school-level PTRs is drawn from EMIS data, reflecting the previous year’s allocations of new teachers as well as other movements, deaths and retirements of teachers in the system. 7 Table 2.1. Summary of the number of teachers needed in 29 primarily rural districts+ 2017 2018 2019 Total number of schools 5,368 5,345 5,527 Total number of students 4,567,817 4,573,924 4,775,675 Total number of teachers 59,233 64,865 71,471 National PTR 77.1 70.5 66.8 Total number of teachers needed in 29 primarily rural districts 5904 5676 3789 % schools that needed at least one new teacher 36% 36% 29% % schools that needed at least one new teacher to attain T1 17% 6% 5% Teachers needed among schools with need Number of teachers needed – median 2 2 2 Number of teachers needed – 10th percentile 1 1 1 Number of teachers needed – 90th percentile 6 6 5 + Blantyre City, Lilongwe City, Mzuzu City, Zomba Urban, and Likoma are excluded. The total need for teachers falls gradually, partly as a result of the improvement in allocations that took place during this period. This decline occurs despite the modification of T2 to require PTR above 80 in 2018 and 2019. The proportion of schools that needed at least one new teacher also drops over time. Despite this overall trend of reducing need, however, district-level needs vary more widely over time. This reflects the fact that the extent to which allocations to districts is aligned to their needs also varies (see next section). Figure 2.1 shows the district-level need in 2017-19 according to the modified rules. The mean number of teachers needed across districts in 2017 was 204, dropping to 196 in 2018 and 132 in 2019 (Table 3.2). The number of teachers needed across the districts varied between 25 (Rumphi) to over 550 (Mangochi) in 2017. This range narrowed in 2018 and in 2019. In terms of overall numbers (Panel A), the majority of districts reduced their need substantially, but some districts saw increasing need over time, notably Ntcheu and Mzimba North, suggesting allocations during this period were not adequate to meet these districts’ needs. Section 3 provides further analysis of this. 8 Figure 2.1. Number and share of teachers needed to fulfil modified rules in 2017, 2018 and 2019: by district A. Number of teachers required by district B. Share of total teachers required by district 9 3. How well did teacher allocation match the modified rules? In this section, we compare the actual allocation of teachers to schools with the predicted need according to the modified rules. We employ administrative data on annual IPTE allocations, linked to EMIS school records using a unique school ID number. We focus our analysis on 29 districts which are rural and in mainland Malawi. We exclude four urban districts (Blantyre, Lilongwe and Zomba Urban districts, and Mzuzu) as these districts are excluded by custom from allocation of newly qualified IPTE teachers (as a result of generally lower PTRs). We also exclude Likoma, an island district in Lake Malawi, which has a small number of schools and experiences unusual constraints in receiving adequate allocations of teachers. Figure 3.1 shows the timeline of allocations along with ASC data collection which informs the EMIS data. Teacher allocations typically take place in August, prior to the start of the school year, while ASC data collection takes place early in the school year in October-November. Figure 3.1. Schematic timeline of EMIS and IPTE 9, 10-11 and 12 allocations, 2017-19 Total deployment at the national level The total numbers of teachers deployed in 2017, 2018 and 2019 were 4,453, 5,511 and 3,387, respectively (Table 3.1). In 2018, the allocation was unusually large as it included two rounds of IPTE teachers.8 Nevertheless, the allocation was not adequate to fill the national need according to the modified rules. Table 3.1. Total number of teachers needed and allocated, and share of need met (national) 2017 2018 2019 Average Total number of teachers needed in 29 primarily rural districts 5,904 5,676 3,789 5,123 Total number of teachers allocated to the 29 primarily rural districts 4,453 5,511 3,387 4939 Share of need for teachers met at national level 75% 97% 89% 72% Allocation of teachers to districts The allocation of teachers to districts, conducted by the MoE, varied considerably over the three years as a proportion of districts’ needs. Table 3.2 (in Tables & Figures) shows the number of teachers 8 In 2017, graduates of the seventh cohort of IPTE trainees, known as IPTE9, were deployed to schools two years after their graduation in 2015. In 2018, in order to reduce the waiting period for new graduates before deployment, both IPTE10 and IPTE11 graduates, who graduated in 2016 and 2017, were deployed to schools. In 2019, IPTE12 graduates, having graduated in 2018, were deployed to schools. 10 needed, the number allocated, and the share of district-level need met, for each rural district for each year 2017-19 and on average across the three years. In 2017, while the average district received 70 percent of their need (in line with the national picture as seen in Table 3.1), the share of need met at district level was as low as 6 percent (in Blantyre Rural district, which received only seven teachers against a need of 126); and as high as 162 percent (in Thyolo district, which received 170 teachers against a need of 105). In total, 10 districts received fewer than half the teachers needed to meet the modified rules. In 2018, despite the overall size of the allocation being larger, three districts still received fewer than half the teachers they needed (receiving fewer than five teachers each). A total of 14 districts received more teachers than were needed to meet the modified rules, suggesting that the rules were not being consistently followed within MoE where the allocation of teachers to districts was carried out. In 2019, as the share of total need met fell to 94 percent, the allocation to districts remained only loosely aligned with need, with the result that ten districts received fewer than half the teachers needed while three received more than double the required number. For a given district, the share of needs met was highly variable over time, with a single district, Chikwawa, receiving between one and 175 percent of the required teachers depending on the year. In total, 13 districts received fewer teachers than required across the three years while 16 received more than required. Only one district received fewer than half the required teachers across the three years, and one received more than double the required teachers. Allocation of teachers to schools How well did districts allocate the teachers they received? Table 3.3 summarizes the total performance of the 29 districts in allocating teachers to needy schools according to the modified rules. Panel A shows the share of the total share of all schools in these districts which were allocated a teacher along with the average number of teachers received by these schools. Panel B shows the share of schools which did not need a teacher according to the modified rules, which nevertheless received them: between 24 and 32 percent of teachers across the three years were deployed to schools which did not qualify for one according to the modified rules. Panel C shows the share of schools which did need at least one new teacher which received teachers, rising to a high of 61 percent in 2018 before falling to 45 percent in 2019. Strikingly, in the larger allocation of 2018, the share of both needy and non-needy schools receiving teachers increased, suggesting that an opportunity was missed to target this large one-off double allocation to achieve a large reduction in school PTRs. 11 Table 3.3 Allocation of teachers to schools (national) 2017 2018 2019 A. All schools % schools allocated new teachers 38% 43% 33% Number of teachers allocated – mean* 2.2 2.4 1.9 B. All schools that needed 0 new teacher % schools allocated new teachers 25% 32% 24% Number of teachers allocated – mean* 1.7 1.9 1.6 C. All schools that needed at least one new teacher(s) % schools allocated new teachers 59% 61% 45% Number of teachers allocated – mean* 2.6 3.0 2.2 *Denotes number of teachers allocated to schools which were allocated a teacher. Table 3.4 (in Section 8, Tables) shows the performance of individual districts in allocating teachers to the correct schools, as well as the national total performance. At national level, the share of deployed teachers who were allocated correctly fell from 56 percent in 2017 to 49 percent in 2018, reflecting the fact that the extra-large allocation was not well targeted as noted above. The share correctly allocated fell further to 43 percent in 2019, showing a further deterioration in adherence to the new rules following their introduction. This aggregate picture, however, masks a high degree of variation between districts. On average across the three years, the share of teachers received which were allocated correctly varies from as low as 22-23 percent (Rumphi and Mzimba North districts) to 75 percent (Dowa district). Strikingly, each district varies considerably year-on-year in its adherence to the guidance, with only four districts (Mwanza, Kasungu, Mangochi and Machinga) allocating more than half their teachers correctly in all three years. To provide a deeper exploration of the dynamics of district allocations over time, we focus on 13 districts which received at least 50 teachers in all three of the years 2017-19. Figure 3.2 shows the performance of these districts in allocating teachers to schools across the three years. In a number of cases, the quality of allocations fell in 2018 in response to the larger-than-normal allocation before improving in 2019. However, in other cases the quality of allocations increased in 2019 before falling, or declined consistently. None of the 15 districts achieved a consistent improvement in the share of teachers which were allocated according to the modified rules. 12 Figure 3.2. % teachers correctly allocated to schools, 2017-19 (13 most allocated districts) Share of need met at the school level To what extent did the allocation of new teachers, flawed as it was, meet the needs of schools? The fourth column in each year in Table 3.4 compares the number of new teachers correctly allocated to needy schools to the district-level need (as seen in Table 3.2) to identify the share of need which was met each year at district level by teachers being placed in needy schools. The share of needs met at the district level reflects both the adequacy of the number of teachers allocated to a district, and the quality of their allocation to schools. The share of need met varied at national level, from 31 percent in 2017 to 44 in 2018 and falling back to 37 percent in 2019. The fact that only 44 percent of need was met in 2018, despite the total number of teachers deployed being 97 percent of the need, demonstrates the inefficiency of allocations. Figure 3.3 shows the relationship across districts between the number of teachers needed in schools to meet the modified rules, and the number of teachers allocated. In each year, at least 20 percent of even the neediest schools, in need of five or more teachers, received zero teachers, while a similar share of schools which needed zero teachers received them. Overall, the analysis suggests that adherence to the revised targeting rules was generally weak and highly inconsistent, between districts and over time. But where teachers were allocated to schools correctly, what were the impacts on overall staffing levels? Section 4 explores this question. 13 Figure 3.3. Distribution of allocation outcome by level of need 4. Impact of teacher allocation on school staffing Although allocation of newly deployed teachers to schools is expected to lead to an increase in overall school staffing levels, there are a number of reasons why this relationship may not be consistent. First, teachers may leave a school at the same time that the school receives a new teacher, either through transferring to another school, movement out of teaching, death or retirement. New teachers allocated to schools may have been so allocated in response to an expected departure, leaving the school with no net gain in teachers. Teacher headcounts To analyze trends in the number of teachers employed at schools, we compare staffing levels year-on- year using EMIS data. Figure 4.1 shows the distribution of schools which needed at least one teacher each year, and the change in the number of teachers employed at schools, both those which received new teachers as part of the annual allocation and those which did not. Among schools that needed teachers and did not receive new teachers from IPTE, many still had more teachers in the next year (68 percent in 2017/18, 58 percent in 2018/19, 35 percent in 2019/20); and some schools that received teachers from IPTE had a lower recorded number of teachers in the following year (3 percent in 2017/18, 9 percent in 2018/19, 21 percent in 2019/20). 14 Figure 4.1. Changes in the number of teachers among schools that needed new teachers 2017-2018 2018-2019 2019-2020 Numebr of schools ∎ Received new teachers form IPTE or ODL ∎ Did not receive new teachers from IPTE or PTR change Even in a situation where the total number of teachers employed at a school increases, changes in enrollment can mean that PTR does not improve. We postulate that in remote areas in Malawi, where shortages of teachers are most common, additional teachers may be met by an increase in demand by parents and subsequently in school enrollment, effectively ‘washing out’ gains in PTR from the additional teacher. To what extent did the correct allocation of teachers, where achieved, reduce school PTRs? Figure 4.2 shows the distribution of the change in school PTRs between 2017-2018 (Panel A), reflecting the impact of the 2017 allocation; 2018-19 (panel B), reflecting the impact of the 2018 allocation; and 2019-20 (Panel C), reflecting the impact of the 2019 allocation; for schools which needed teachers and (i) did not and (ii) did receive them. The green bars show schools which reduced PTR, while the red bars show schools where PTR increased. As expected, the share of schools with PTRs reducing is higher among those schools that received new teachers, as is the mean reduction in PTR. Nevertheless, PTRs did reduce on average in schools which did not receive new teachers, which may reflect teachers joining these schools through transfers, or declining enrollment. In addition, a number of schools receiving new teachers nevertheless experienced increases in PTR, reflecting a net decrease in teacher numbers (as a result of teachers moving away as described above) and/or increasing enrollments. Overall, the analysis suggests that the correct allocation of teachers to needy schools did reduce PTRs, but these effects were blunted by ongoing movement of teachers within the system. Section 5 explores whether, in those cases where PTRs were reduced, this led to improvements in student outcomes. 15 Figure 4.2. Annual change in PTR among schools that needed new teachers (i) Did not receive new teachers (ii) Received new teachers 1. 2017-2018 Mean PTR change: -11 (-14, -9) -29 (-31, -27) 2. 2018-2019 Mean PTR change: -11 (-14, -9) -21 (-22, -19) 3. 2019-2020 Mean PTR change: -5 (-7, -4) -12 (-14, -10) 5. Impact of improved staffing on student outcomes In this section, we evaluate the extent to which improvements in staffing, primarily as a result of the correct allocation of newly deployed teachers to schools where implemented, led to improvements in student outcomes. Our outcome of interest is student repetition rates, as a proxy for learning levels (see Introduction). We derive these rates from EMIS data. 16 Although the assignation rules used in Malawi allocate teachers both to schools without one teacher per grade (T1) and schools with PTR above 90 or 80 (T2), in this analysis we focus primarily on T2 and the reduction of overall school PTRs. This is done in response to evidence from Malawi which suggests that schools with PTRs above 90 have lower overall learning outcomes (Asim and Casley Gera, 2024). In addition, in this analysis, we exclude non-qualified and student teachers and focus on school pupil-qualified teacher ratios (PqTRs). In recognition of the fact that the dynamics of school staffing vary significantly regardless of the allocation of new teachers (see Section 4), we focus our analysis on schools which an increase in overall teacher numbers and experienced reduction in PqTR, regardless of whether they were allocated a new teacher. In other words, a school which gained an additional teacher through transfer is treated the same as one which received a newly deployed teacher. We define a school has having been ‘treated’ with additional teachers in a given year if it meets the following conditions: 1. The number of teachers employed in the school is higher than in the previous year; 2. The school’s overall PqTR was above 90 in the previous year and is now below 90. We define a control group of schools which had PqTR above 90 in the previous year – meaning they needed a new teacher – but did not experience a net gain in the number of teachers. As a result of the dynamic nature of teacher allocations, the treated and control schools therefore vary across the various year comparisons. Table 5.1 summarizes the sample. Table 5.1. Impact analysis sample 2017-18 2018-19 2019-20 Treatment 902 710 468 Control 412 296 330 Both control and treatment schools experienced changes in PqTR, repetition rate, and dropout rate year-on-year. To capture the differential dynamics between treated and control schools, we employ difference-in-difference (DiD) analysis. Table 5.2 (in Section 8, Tables) presents DiD estimates for the impact of treatment on PqTRs. As expected given the definitions of treatment and control, treated schools experienced decline in PqTR compared to control schools in each year – an substantial decline of 41 pupils per qualified teacher. Lagged effects for dropout and repetition. Because dropout and repetition rates are determined by schools at the end of a school year, there is an expected ‘lag’ in impacts for an improvement in school staffing. Recall that new teachers are typically deployed to schools at the start of the new school year in September, with the EMIS data collection taking place around two months later in October- November. In order to allow time for the impacts of new teachers to be felt and measured, we report lagged effects for these indicators from the following year’s EMIS data. For example, to evaluate the impact of teachers allocated in August/September 2017, we compare: 17 • EMIS 2018 (collected in October-November 2017, and reflecting the dropout and repetition status at the end of the 2016/17 school year, prior to the allocation of teachers) with • EMIS 2019 (collected in October-November 2018, and reflecting the dropout and repetition status at the end of the 2017/18 school year, following the first full year of school with the increased level of staffing and reduced PqTR). The COVID-19 pandemic led to the closure of all schools in Malawi for seven months during 2020 and appears to have led to significant dropout.9 As the EMIS data collected in October-November 2019 is the most recent available prior to the onset of the pandemic, we restrict our analysis to the 2017 and 2018 allocations of teachers for which lagged information is available prior to the pandemic. Tables 5.3-5.6 show DiD results. In each table: • The first column defines treated schools as those which received at least one additional teacher in 2017 and as a result whose PqTR was brought below 90 in that year.10 The regression compares the repetition or dropout rate in schools at the end of the 2016/17 year to those at the end of the 2017/18 school year, reflecting a full year of the increased staffing. • The second column defines treated schools as those which received at least one additional teacher in 2018 and as a result whose PqTR was brought below 90 in that year.11 The regression compares the repetition or dropout rate in schools at the end of the 2017/18 year to those at the end of the 2018/19 school year, reflecting a full year of the increased staffing. Repetition Tables 5.3 and 5.4 show DiD results for repetition rates. Table 5.3 focuses on overall repetition across grades. We find that in schools treated in 2017, repetition rates were significantly reduced by 0.28 percentage points. We do not observe significant impacts on overall repetition rates from treatment in 2018. 9 The total enrollment in Malawi’s primary schools fell for the first time in over a decade following the closure of schools, with 4,815,286 students enrolled in public primary schools in 2020/21 versus 5,274,819 in 2019/20. 10 To identify treated schools, we compare data from EMIS 2016 and EMIS 2017. As repetition and dropout rates are decided at the end of the school year and reported in the following year’s EMIS, for repetition and dropout rates, we compare data from EMIS 2017 and 2018. 11 To identify treated schools, we compare data from EMIS 2017 and EMIS 2018. As repetition and dropout rates are decided at the end of the school year and reported in the following year’s EMIS, for repetition and dropout rates, we compare data from EMIS 2018 and 2019. 18 Table 5.3. Impact of PqTR reduction on repetition rates 2017-18 2018-19 Control 0.266*** 0.253*** (0.010) (0.006) Time -0.014 -0.001 (0.014) (0.009) Treatment 0.007 -0.016 (0.011) (0.012) DiD (Treatment and Time) -0.028* -0.021 (0.016) (0.015) Control (N) 118 292 Treatment (N) 379 134 Data Source: EMIS 2016-2019 Note: Robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 2017/18 compares Repetition rates at end 2016/17 and end 2017/18. Treated schools are those where PqTR brought below 90 in 2017. 2018/19 compares Repetition rates at end 2017/18 and end 2018/19. Treated schools are those where PqTR brought below 90 in 2018. Table 5.4 focuses on repetition rates in lower primary (Grades 1-4). Again, we find significant impacts from the 2017 allocation of teachers, with lower primary repetition rates reduced by 3.5 percentage points, but we do not observe similar impacts from the 2018 deployment. Table 5.4. Impact of PqTR reduction on repetition rates in lower primary 2017-18 2018-19 Control 0.276*** 0.269*** (0.011) (0.007) Time -0.014 -0.005 (0.015) (0.010) Treatment 0.016 -0.026** (0.013) (0.013) DiD (Treatment and Time) -0.035** -0.017 (0.017) (0.016) Control (N) 118 292 Treatment (N) 379 134 Data Source: EMIS 2016-2019 Note: Robust standard errors in parentheses. *p<0.10, ** p<0.05, *** p<0.01 2017/18 compares lower primary Repetition rates at end 2016/17 and end 2017/18. Treated schools are those where PqTR brought below 90 in 2017. 2018/19 compares lower primary Repetition rates at end 2017/18 and end 2018/19. Treated schools are those where PqTR brought below 90 in 2018. 19 Dropout Table 5.5 and 5.6 show DiD results for dropout rates. Table 5.5 focuses on overall dropout across grades; Table 5.6 focuses on lower primary dropout rates. In both cases, we do not observe significant impacts in schools treated in either 2017 or 2018. Table 5.5. Impact of PqTR reduction on dropout rates 2017-18 2018-19 Control 0.031*** 0.040*** (0.004) (0.003) Time 0.001 -0.003 (0.006) (0.004) Treatment 0.020*** 0.003 (0.006) (0.006) DiD (Treatment and Time) -0.006 0.005 (0.008) (0.008) Control (N) 118 292 Treatment (N) 379 134 Data Source: EMIS 2016-2019 Note: Robust standard errors in parentheses. *p<0.10, ** p<0.05, *** p<0.01 2017/18 compares dropout rates at end 2016/17 and end 2017/18. Treated schools are those where PqTR brought below 90 in 2017. 2018/19 compares dropout rates at end 2017/18 and end 2018/19. Treated schools are those where PqTR brought below 90 in 2018. Table 5.6. Impact of PqTR reduction on dropout rates in lower primary 2017-18 2018-19 Control 0.029*** 0.037*** (0.005) (0.004) Time -0.001 -0.003 (0.007) (0.005) Treatment 0.020*** 0.002 (0.006) (0.006) DiD (Treatment and Time) -0.004 0.009 (0.009) (0.009) Control (N) 118 292 Treatment (N) 379 134 Data Source: EMIS 2016-2019 Note: Robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 2017/18 compares lower primary dropout rates at end 2016/17 and end 2017/18. Treated schools are those where PqTR brought below 90 in 2017. 2018/19 compares lower primary dropout rates at end 2017/18 and end 2018/19. Treated schools are those where PqTR brought below 90 in 2018. 20 Robustness checks In order to ensure that our particular treatment definition, of schools gaining additional teachers to bring PqTR below 90, is not creating a false appearance of systemic results with regard to repetition, we conduct additional analysis employing two alternative definitions of treated schools: 1. The number of teachers employed in the school is higher than in the previous year, and the school’s overall PqTR was above 80 in the previous year and is now below 80. 2. The number of teachers employed in the school is higher than in the previous year, and the school’s overall PqTR was above 100 in the previous year and is now below 100. For this analysis, we focus on the 2017 allocation of teachers. Using lagged effects, we compare the repetition rates in schools at the end of the 2016/17 year to those at the end of the 2017/18 school year, reflecting a full year of the increased staffing according to the adjusted thresholds.12 Tables 5.7 and 5.8 (in Section 8, Tables), show the findings. Adopting PqTR reduction below 80 as the threshold, we observe impacts on repetition rates in lower primary and overall of a slightly larger scope than with the original treatment definition of PqTR reduction below 90. However, adopting PqTR reduction below 100 as the threshold, although we still observe a substantial reduction in lower primary repetition rates, it does not obtain statistical significance. The findings suggest that reducing PqTRs below 90 is the ‘minimum’ treatment to achieve impacts on student learning. 6. Conclusion Our findings suggest that improvements in allocation of newly deployed teachers to schools, enabled by improved rules for allocation, can lead to improvements in student learning. Adopting repetition as a proxy for learning, we find that schools which gained additional teachers and brought PqTRs below 90 achieved significant improvements in lower primary repetition rates in comparison to schools with PqTR above 90 which did not gain additional teachers. The findings mirror others from Israel, the United States and India (Angrist and Lavy, 1999; Kreuger, 1999; Muralidharan & Sundararaman, 2013) which suggest that improvements in PTR are associated with improved learning outcomes, and demonstrate that similar dynamics persist in low-income countries. However, we do not observe impacts on dropout rates. The descriptive analysis also reveals the extent to which rules-based approaches to teacher allocations may face challenges in implementation on the ground. Despite the clarification of allocation rules and provision of software tools to support allocations, the adherence to guidance for allocation of new teachers appears to have been weak and highly variable. At the national level, the allocation of new teachers to districts was not fully aligned with the guidance, with the share of districts’ need for new teachers met through allocations varying from a low of 32 percent to a high of 224 percent across three years. At the district level, too, the quality of allocations was highly variable, with the most successful deploying 75 percent of teachers in accordance to the rules and the least successful just 22 percent. Had the teachers deployed during this period all been allocated according to the guidance, it is likely that the number of schools achieving reduction in PqTR to below 90 would have been larger, 12 The comparison of staffing to identify treatment and control schools remains non-lagged, e.g. comparing EMIS data from 2016 and 2017. 21 with the result that more students would benefit from increased learning and reduced repetition rates. In addition, even where schools were correctly allocated new teachers, the effects on PTR were blunted by other movement of teachers away from these schools. A new Hardship Support Scheme, expected to be rolled out during 2024, is intended to provide additional incentives to teachers in remote schools to remain in post (see Asim et al., 2019, for background). Future research will explore the impact of this scheme on teacher behavior, school PTRs, and student outcomes. 22 7. References Angrist, Joshua D.; Lavy, Victor. 1999 “Using Maimonides’ Rule to Estimate the Effect of Class Size on Scholastic Achievement.” The Quarterly Journal of Economics 114 (2), 533-575. Angrist, Joshua D.; Lavy, Victor; Leder-Luis, Jetson; Shany, Adi. 2017. “Maimonides’ Rule Redux.” NBER Working Paper No. 23486. Asim, S., J. Chimombo, D. Chugunov, and R. Casley Gera (2019). “Moving teachers to Malawi’s remote communities: A data-driven approach to teacher deployment.” International Journal of Educational Development 65 (2019) 26–43 Asim, S. and Casley Gera, R. 2024. What Matters for Learning in Malawi? Insights from a Longitudinal School Survey. Washington, D.C.: World Bank Publications. Azim Premji Foundation. 2014. “Pupil-Teacher Ratios in Schools and their Implications.” Available at: https://www.issuelab.org/resources/23973/23973.pdf [1.27.22] Bashir, S., Lockheed, M., Ninan, E., Tan, J., 2018. Facing Forward: Schooling for Learning in Africa. Washington, D.C.: World Bank Publications. Checchi, D. and De Paola, M. 2017. “The Effect of Multigrade Classes on Cognitive and Non- Cognitive Skills: Causal Evidence Exploiting Minimum Class Size Rules in Italy.” IZA DP No. 11211. Bonn: Institute of Labor Economics. Chin, Aimee (2005). Can redistributing teachers across schools raise educational attainment? Evidence from Operation Blackboard in India. Journal of Development Economics 78(2), 0–405. Cummings, Clare and Ali Bako M. Tahirou with Hamissou Rhissa, Falmata Hamed, Hamadou Goumey, and Idi Mahamadou Mamane Noura. 2016. “Collective action and the deployment of teachers in Niger: A political economy analysis.” Overseas Development Institute. Available at: https://cdn.odi.org/media/documents/10303.pdf [2.3.22] Duflo, Esther; Dupas, Pascaline; Kremer, Michael. 2015. “School governance, teacher incentives, and pupil-teacher ratios: Experimental evidence from Kenyan primary schools.” Journal of Public Economics (123), 92-110. Hoxby, Caroline M. 2000. “The Effects of Class Size on Student Achievement: New Evidence from Population Variation.” The Quarterly Journal of Economics 115 (4), 1239-1285. Jacob, Verghese; Kochar, Anjini; Reddy, Suresh. 2008. “School Size and Schooling Inequalities.” Stanford Center on Global Poverty and Development Working Paper 354. Krueger, Alan B. 1999. “Experimental Estimates of Education Production Functions.” The Quarterly Journal of Economics 114 (2), 497-532. Majgaard, K., and A. Mingat. 2012. Education in Sub-Saharan Africa: A Comparative Analysis. A World Bank Study. Washington, DC: World Bank. 23 Malawi Longitudinal School Survey, 2021. Malawi Longitudinal School Survey Endline data. Unpublished. Ministry of Education, Science and Technology (2018). Primary Teacher Management Strategy. Lilongwe: Government of Malawi. Ministry of Education (2021a). Education Management Information System (EMIS) data, 2019/20. Unpublished. Ministry of Education (2021b). Malawi Education Statistics 2019/20. Mimeo. Ministry of Education (2022). Education Management Information System (EMIS) data, 2020/21. Unpublished. Malawi Longitudinal School Survey (2021). Unpublished endline data. Mulkeen, A., 2010. Teachers in Anglophone Africa: Issues in Teacher Supply, Training, and Management. Washington, D.C.: World Bank Publications. Muralidharan, K.; Sundararaman, V. 2013. “Contract Teachers: Experimental Evidence from India.” NBER Working Paper No. 19440. Pugatch, T and Schroeder, E. 2014. “Incentives for teacher relocation: Evidence from the Gambian hardship allowance.” Economics of Education Review 41, 120-136. Ramachandran, Vimala, Tara Béteille, Toby Linden, Sangeeta Dey, Sangeeta Goyal, and Prerna Goel Chatterjee. 2018. Getting the Right Teachers into the Right Schools: Managing India’s Teacher Workforce. World Bank Studies. Washington, D.C.: World Bank Publications. Walter, T. F. 2018. “Misallocation of State Capacity?” PhD Thesis. London School of Economics and Political Science, London. Available at: http://etheses.lse.ac.uk/3852/1/Walter misallocation-of- state-capacity.pdf [9.1.22] 24 8. Tables Note: Tables 2.1, 3.1, 3.3, and 5.1-5.4 can be found in the main text. Table 3.2. Total number of teachers needed and allocated, and share of need met (by district) No. of teachers needed/allocated, district level, 2017-19 2017 2018 2019 Total District % % % % Allo’t Need need Need Allo. need Need Allo. need Need Allo. need ed met met met met Dowa 313 90 29 363 7 2 247 148 60 923 245 27 Blantyre Rural 126 8 6 99 35 35 80 48 60 305 91 30 Balaka 193 100 52 157 1 1 105 48 46 455 149 33 Neno 83 39 47 68 1 1 67 45 67 218 85 39 Ntcheu 167 67 40 117 88 75 162 58 36 446 213 48 Mwanza 58 25 43 44 31 70 33 13 39 135 69 51 Salima 130 51 39 185 159 86 114 51 45 429 261 61 Dedza 228 158 69 244 181 74 159 49 31 631 388 61 Ntchisi 99 25 25 83 61 73 87 80 92 269 166 62 Chiradzulu 80 20 25 89 104 117 40 7 18 209 131 63 Mzimba North 110 26 24 201 1 0 244 362 148 555 389 70 Lilongwe Rural East 253 118 47 247 279 113 128 79 62 628 476 76 Chikwawa 317 320 101 227 3 1 93 163 175 637 486 76 Lilongwe Rural West 133 91 68 192 139 72 124 143 115 449 373 83 Zomba Rural 291 211 73 337 329 98 195 191 98 823 731 89 Machinga 463 384 83 435 533 123 233 162 70 1131 1079 95 Nkhotakota 180 104 58 163 174 107 134 178 133 477 456 96 Nsanje 132 103 78 177 124 70 131 200 153 440 427 97 Mchinji 196 142 72 190 283 149 63 15 24 449 440 98 Mangochi 574 552 96 485 599 124 262 163 62 1321 1314 99 Kasungu 537 618 115 328 520 159 401 208 52 1266 1346 106 Phalombe 119 129 108 149 204 137 70 27 39 338 360 107 Thyolo 105 165 157 234 293 125 67 16 24 406 474 117 Karonga 178 150 84 159 294 185 73 35 48 410 479 117 Mzimba South 488 409 84 230 456 198 151 173 115 869 1038 119 Mulanje 206 200 97 225 318 141 109 142 130 540 660 122 Nkhata Bay 58 46 79 92 95 103 72 177 246 222 318 143 Chitipa 62 84 135 80 53 66 70 215 307 212 352 166 Rumphi 25 18 72 76 146 192 75 191 255 176 355 202 Mean 204 154 69 196 190 93 131 117 95 530 460 88 Median 167 103 72 185 146 98 109 142 62 449 388 89 25 Table 3.4 Allocation of teachers to schools (by district) 2017 2018 2019 Average Teachers Correctly % cor’ly % need Teachers Correctly % cor’ly % need Teachers Correctly % cor’ly % need Teachers Correctly % cor’ly % need allocated allocated allocated met allocated allocated allocated met allocated allocated allocated met allocated allocated allocated met Balaka 100 68 68 35 1 0 0 0 48 21 44 20 50 30 37 18 Blantyre Rural 8 4 50 3 35 16 46 16 48 26 54 33 30 15 50 17 Chikwawa 320 186 58 59 3 0 0 0 163 53 33 57 162 80 30 39 Chiradzulu 20 16 80 20 104 69 66 78 7 1 14 3 44 29 54 33 Chitipa 84 26 31 42 53 22 42 28 215 33 15 47 117 27 29 39 Dedza 158 81 51 36 181 122 67 50 49 26 53 16 129 76 57 34 Dowa 90 48 53 15 7 7 100 2 148 114 77 46 82 56 77 21 Karonga 150 60 40 34 294 97 33 61 35 10 29 14 160 56 34 36 Kasungu 618 391 63 73 520 279 54 85 208 140 67 35 449 270 61 64 Lilongwe Rural 118 84 71 33 279 163 58 66 79 36 46 28 159 94 58 42 East Lilongwe Rural 91 27 30 20 139 72 52 38 143 61 43 49 124 53 41 36 West Machinga 384 299 78 65 533 295 55 68 162 112 69 48 360 235 67 60 Mangochi 552 379 69 66 599 344 57 71 163 109 67 42 438 277 64 60 Mchinji 142 54 38 28 283 127 45 67 15 5 33 8 147 62 39 34 Mulanje 200 78 39 38 318 194 61 86 142 52 37 48 220 108 46 57 Mwanza 25 15 60 26 31 16 52 36 13 7 54 21 23 13 55 28 Mzimba N 26 5 19 5 1 0 0 0 362 183 51 75 130 63 23 27 Mzimba S 409 235 57 48 456 128 28 56 173 52 30 34 346 138 39 46 Neno 39 22 56 27 1 1 100 1 45 14 31 21 28 12 63 16 Nkhata Bay 46 13 28 22 95 33 35 36 177 39 22 54 106 28 28 37 Nkhotakota 104 73 70 41 174 81 47 50 178 74 42 55 152 76 53 48 Nsanje 103 42 41 32 124 119 96 67 200 116 58 89 142 92 65 63 Ntcheu 67 40 60 24 88 34 39 29 58 30 52 19 71 35 50 24 Ntchisi 25 11 44 11 61 36 59 43 80 23 29 26 55 23 44 27 Phalombe 129 60 47 50 204 100 49 67 27 13 48 19 120 58 48 45 Rumphi 18 3 17 12 146 34 23 45 191 49 26 65 118 29 22 41 Salima 51 40 78 31 159 58 36 31 51 18 35 16 87 39 50 26 Thyolo 165 27 16 26 293 155 53 66 16 9 56 13 158 64 42 35 Zomba Rural 211 110 52 38 329 112 34 33 191 39 20 20 244 87 36 30 Malawi 4453 2497 56 42 5511 2714 49 46 3387 1465 43 3 1 3810 1751 30 26 Table 5.2. Impact of treatment on PqTRs in treated schools 2016-17/ 2017- 2017-18/ 2018- 2018-19/ 2019- 18 19 20 Control 111.067 *** 108.804 *** 106.133 *** (1.066) (1.284) (0.860) Time 5.231 *** 3.585* 1.845 (1.789) (2.058) (1.508) Treatment 3.193** 1.433 2.392 (1.393) (1.535) (1.549) DiD (Treatment and -46.187 *** -40.099 *** -37.431 *** Time) (2.036) (2.263) (2.085) Control (N) 415 299 470 Treatment (N) 912 714 330 Data Source: EMIS 2016-2019 Note: Robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Table 5.7. Robustness check: Impact of PqTR reduction on repetition rate (PqTR 80) Overall Lower Primary Control in 2018 0.260*** 0.272*** (0.007) (0.008) Time (year=2019) -0.009 -0.010 (0.010) (0.011) Lagged Treatment 0.012 0.016 (0.009) (0.010) Lagged Treatment (DiD) -0.034*** -0.040*** (0.012) (0.014) Control (N) 216 216 Treatment (N) 424 424 r2 0.03 0.03 Data Source: EMIS 2018 and 2019 Note: Robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Compares rates at end 2017/18 and end 2018/19. Treated schools are those where PqTR brought below 90 in 2018. 27 Table 5.8. Robustness check: Impact of PqTR reduction on repetition rate (PqTR 100) Overall Lower Primary Control in 2018 0.268*** 0.277*** (0.013) (0.015) Time (year=2019) -0.017 -0.016 (0.018) (0.020) Lagged Treatment -0.002 0.005 (0.015) (0.017) Lagged Treatment (DiD) -0.026 -0.033 (0.020) (0.023) Control (N) 73 73 Treatment (N) 253 253 r2 0.03 0.04 Data Source: EMIS 2018 and 2019 Note: Robust standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 Compares rates at end 2017/18 and end 2018/19. Treated schools are those where PqTR brought below 100 in 2018. 28