At the 14th Africa Evidence Summit, held on July 23–24, 2026, Nneka Esther Osadolor of the African Economic Policy Forum (AFEP) presented research examining whether reducing class sizes improves learning outcomes in Rwanda's public primary schools. The study, titled “Does Class Size Matter for Learning? For Whom and How? Evidence from Public Primary Schools in Rwanda,” investigated not only whether smaller classes improve learning, but also for which students the effects are strongest, under what conditions class-size reductions become effective, and how substantial the reduction needs to be before meaningful learning gains emerge.
The presentation positioned the study within the broader learning crisis facing many African education systems. The presenter explained that although Rwanda had achieved substantial progress in school enrolment, increased access had also generated significant pressure on the capacity of schools, classrooms, and teachers. The central concern, therefore, was that increasing enrolment alone did not necessarily translate into improved learning outcomes.
Osadolor explained that Rwanda had experienced significant growth in primary-school enrolment, but that this expansion had been accompanied by overcrowded classrooms. According to the figures presented, public primary schools in Rwanda had a pupil-classroom ratio of approximately 60 students per classroom in 2023/24. The pupil-teacher ratio was also reported to be around 48 students per teacher, considerably higher than global averages and continuing to rise as enrolment increased by approximately 6 percent annually.
The presenter noted that these pressures were occurring despite education reforms aimed at improving learning. She highlighted evidence indicating that approximately 80 percent of lower-primary pupils remained below expected reading levels, illustrating the scale of Rwanda's learning challenge. The presentation also placed Rwanda's experience within the wider Sub-Saharan African context, where approximately 86 percent of children were described as experiencing learning poverty, meaning that they were unable to read and understand a simple text by around age ten before the COVID-19 pandemic.
Osadolor argued that these figures demonstrated that the education challenge was no longer simply one of ensuring that children entered school. Instead, the policy focus increasingly needed to shift toward ensuring that children actually learned while they were in school.
The presenter explained that reducing class sizes could potentially improve learning because teachers would have fewer pupils to manage and could provide more individualized attention. However, she emphasized that class-size reduction was an expensive education intervention.
Reducing the number of pupils per class would require governments to recruit, train, deploy, and retain additional teachers. It could also require additional classrooms and other educational infrastructure. This was particularly important in Rwanda because the country was already operating with a relatively high pupil-teacher ratio while simultaneously experiencing continued growth in enrolment.
The presentation therefore framed the issue as a policy optimization problem. Rather than asking simply whether smaller classes were beneficial, policymakers needed evidence on who benefited, under what circumstances, and how small classes needed to become before learning improvements appeared.
Osadolor explained that the Rwanda study provided an important contribution because the country's starting pupil-teacher and class-size ratios were unusually high compared with many previous studies of class-size reduction.
She noted that the intervention examined in the study reduced class sizes from approximately 52 pupils to around 36 pupils, representing a substantially larger reduction than those examined in some influential earlier studies. For comparison, she referred to the work of Angrist and Lavy, which examined a reduction of approximately eight pupils, and the well-known Tennessee Student/Teacher Achievement Ratio (STAR) experiment, in which class sizes were reduced from approximately 22 to 15 pupils.
According to the presenter, Rwanda's much higher starting class sizes raised a different set of policy questions. Evidence from systems where classes were already relatively small might not necessarily apply to education systems operating with very high pupil-teacher ratios. The Rwanda case therefore offered an opportunity to investigate whether substantial reductions in class size could generate measurable learning benefits in a high-ratio environment.
The presenter then provided an overview of Rwanda's education system and the policy programme through which the class-size threshold was implemented. She explained that Rwanda's education system followed a 3-6-3-3 structure, consisting of pre-primary, primary, lower secondary, and upper secondary education. Basic education had been free and compulsory since 2003.
She also noted that many public primary schools operated on a double-shift system, meaning that different groups of pupils attended school at different times of the day. This arrangement was part of the system's response to limited infrastructure and high enrolment pressures.
The study was closely linked to RwandaEQUIP, an education reform programme launched in January 2022 and designed to improve the quality of teaching and learning in public primary schools across the country.
According to the presentation, RwandaEQUIP incorporated several major interventions, including scripted lesson guides, teacher tablets, ten-day teacher training, weekly coaching, and real-time data monitoring. An important feature of the programme was the introduction of a 69-pupil enrolment threshold, which created a discontinuity in how schools organized pupils into streams or classes.
Osadolor explained that RwandaEQUIP had been implemented in phases. The first phase began in February 2022, covering 100 schools in 13 districts across five provinces. The second phase began in September 2022, expanding the programme to 250 schools. A third phase started in September 2024, expanding coverage to 761 schools.
The presenter explained that this phased national expansion created a valuable opportunity for researchers because administrative education data could be combined with the programme's enrolment threshold to examine the relationship between class size and student learning.
The study relied on administrative data covering 761 public primary schools and approximately 4,471 school-by-grade observations. The researchers combined several sources of information to construct a comprehensive dataset.
The data included Foundational Literacy and Numeracy (FLN) baseline assessments, as well as end-of-term examination scores from Terms 1 and 3 of the 2023/24 academic year.
The analysis covered five subjects: English, Kinyarwanda, Mathematics, Science and Elementary Technology (SET), and Social and Religious Studies (SRS). The researchers also incorporated contextual variables such as enrolment, number of streams, school-shift status, province, and RwandaEQUIP implementation phase.
The presenter provided descriptive statistics for Grade 1 and Grade 6. She reported that the average pupil-to-stream ratio was approximately 51.5 pupils in Grade 1 and 39.9 pupils in Grade 6. Average English scores were about 50.9 percent in Grade 1 and 50.2 percent in Grade 6, while Kinyarwanda scores were approximately 46.6 percent and 50.8 percent, respectively.
Mathematics performance showed a wider difference between the grades, with an average score of approximately 42.9 percent in Grade 1 compared with 77.7 percent in Grade 6. SET scores were approximately 54.0 percent in Grade 1 and 33.4 percent in Grade 6, while SRS scores were approximately 55.1 percent and 32.9 percent.
The data also showed that the average age of pupils was approximately 7.9 years in Grade 1 and 14.4 years in Grade 6. Average enrolment per grade was approximately 163 pupils in Grade 1 and 66 pupils in Grade 6. Female pupils accounted for approximately 46 percent of Grade 1 enrolment and 56 percent of Grade 6 enrolment.
A major methodological component of the presentation was the study's use of a Fuzzy Regression Discontinuity Design (RDD).
Osadolor explained that the RwandaEQUIP policy created a natural threshold at 69 pupils. When enrolment reached or exceeded this threshold, schools became substantially more likely to divide pupils into at least two streams. However, the threshold did not perfectly determine whether schools actually divided their pupils into multiple classes. Because there was therefore only partial compliance with the threshold rule, the researchers used a fuzzy rather than a sharp regression discontinuity design.
The presenter explained that schools just below and just above the 69-pupil threshold were likely to be relatively similar in many respects. However, crossing the threshold substantially increased the probability that the school would create at least two streams.
According to the estimates presented, crossing the threshold increased the probability of having at least two streams by approximately 49 percentage points. The average number of streams also increased from approximately 1.3 to 1.9 at the threshold.
This discontinuity provided the researchers with a credible source of quasi-experimental variation in class size. Rather than simply comparing schools with large and small classes—which could produce biased results because the schools might differ in many other ways—the approach allowed the researchers to compare schools located around the policy threshold.
The presenter explained that the researchers estimated the effects using a two-stage least squares (2SLS) framework.
In the first stage, the analysis estimated how crossing the 69-pupil threshold affected the number of streams or class organization. This stage established the strength of the relationship between the enrolment threshold and actual class-size changes.
In the second stage, the predicted class-size or stream changes from the first stage were used to estimate their effects on student learning outcomes. The approach therefore allowed the researchers to isolate the portion of class-size variation generated by the policy threshold rather than relying on simple observational correlations.
The presenter emphasized that this methodology was particularly valuable because schools did not necessarily comply perfectly with the threshold rule. The fuzzy RDD therefore accounted for the fact that crossing the threshold increased the probability of creating additional streams without making such a change automatic.
The presentation highlighted the broader policy importance of understanding class-size effects in high-enrolment African education systems. The presenter suggested that evidence from Rwanda could help policymakers determine whether investments in additional teachers and classrooms would translate into meaningful learning improvements.
If reducing very large classes generated substantial learning gains, policymakers could justify additional investments in teachers, classrooms, and school infrastructure. However, if the effects were small or concentrated among particular groups of students, governments might need to consider whether alternative investments—such as teacher training, instructional materials, coaching, or targeted interventions—could provide greater educational returns.
The study was therefore presented as moving beyond the simple question of whether “smaller classes are better”. Instead, it sought to identify the conditions under which class-size reductions were most effective and the students who benefited most from such changes.
In conclusion, Nneka Esther Osadolor's presentation highlighted the importance of generating rigorous evidence to guide education policy in countries facing rapid enrolment growth and persistent learning challenges. The Rwanda study examined a particularly important policy question because the country had achieved significant progress in expanding access to education while continuing to face overcrowded classrooms and weak learning outcomes.
By exploiting the 69-pupil enrolment threshold under RwandaEQUIP and applying a fuzzy regression discontinuity design, the study sought to estimate the causal effects of reducing class sizes in a high-pupil-teacher-ratio environment.
The presentation ultimately emphasized that the policy debate should not be limited to whether class-size reduction works in general. Rather, policymakers need to understand for whom smaller classes matter, how large the reduction needs to be, through which mechanisms learning improves, and whether the benefits justify the substantial financial and institutional costs of hiring additional teachers and expanding infrastructure.
The study was thus positioned as an important contribution to the growing African evidence base on education quality, learning outcomes, and the effective allocation of scarce public resources.
Presenting the main findings of the study, Nneka Esther Osadolor explained that reducing class size produced meaningful short-term improvements in student learning, although the magnitude of the effects differed considerably across subjects.
The study found that reducing class size by approximately 16 pupils per stream—from around 52 to 36 pupils—generated significant improvements in several learning outcomes during the initial period of the intervention. English scores increased by approximately 0.23 standard deviations, while Science and Elementary Technology (SET) improved by approximately 0.22 standard deviations. Social and Religious Studies (SRS) recorded an improvement of approximately 0.20 standard deviations. The effect on Kinyarwanda was smaller, at around 0.13 standard deviations, and was described as marginally significant. Mathematics, however, showed an effect close to zero and was not statistically significant.
The presenter noted that the magnitude of the effects was broadly comparable to findings from earlier influential studies of class-size reduction. She explained that Angrist and Lavy's 1999 study, which examined a reduction of approximately eight pupils, had found an effect of around 0.20 standard deviations, while the STAR experiment examined a reduction from approximately 22 to 15 pupils and reported an effect of around 0.25 standard deviations. In comparison, the Rwanda study found an effect of approximately 0.23 standard deviations from a much larger reduction, from approximately 52 to 36 pupils. This comparison suggested that meaningful learning gains could also be achieved in education systems where starting class sizes were substantially larger.
The presenter then highlighted an important qualification to the initial findings. Although class-size reduction generated substantial improvements during Term 1, these gains were considerably smaller by Term 3, suggesting that simply reducing the number of pupils in a classroom was not sufficient to guarantee sustained improvements in learning.
For English, the initial effect of approximately 0.23 standard deviations declined to only about 0.04 standard deviations by Term 3. Similarly, the SET effect declined from approximately 0.22 standard deviations to 0.04, while the SRS effect declined from approximately 0.20 to 0.02 standard deviations. The Kinyarwanda effect declined from approximately 0.13 to 0.07 standard deviations. Interestingly, Mathematics, which had shown virtually no initial effect, recorded a small positive effect of approximately 0.06 standard deviations by Term 3.
Based on these findings, Osadolor argued that class-size reduction alone was not sufficient to sustain learning gains over time. She suggested that maintaining improvements would require complementary investments in teachers and instructional systems, including coaching, structured lesson guides, continuous professional development, and other forms of pedagogical support.
The findings therefore pointed to an important policy lesson: smaller classes could create better conditions for learning, but the benefits could diminish if teachers did not receive continued support to take advantage of those conditions.
The study also found substantial differences in the effects of class-size reduction across grade levels. According to the presenter, the largest and most consistent gains were observed among pupils in Grades 1–3, corresponding to the lower-primary years.
In these early grades, English scores increased by as much as 0.35 standard deviations, while Kinyarwanda showed an effect of approximately 0.37 standard deviations. The effects were particularly strong for SET, where improvements reached as high as approximately 0.60 standard deviations.
Importantly, the presenter noted that the positive effects observed in the lower-primary grades were more likely to persist into Term 3. This contrasted with the results for Grades 4–6, where the effects were weaker and less consistent. In many cases, the effects among upper-primary students became negative by Term 3.
These findings suggested that the timing of class-size interventions matters considerably. The presenter argued that reducing class size in the early years of primary education may generate greater returns because foundational literacy, numeracy, language, and learning behaviours are established during these formative years.
Another important finding concerned the organization of the school day. The researchers examined whether the effects of class-size reduction differed between schools operating on a single-shift system and those operating on a double-shift system.
The results indicated that single-shift schools experienced the strongest benefits from class-size reduction. In these schools, English scores improved by approximately 0.18 standard deviations, SET by approximately 0.25 standard deviations, and SRS by approximately 0.20 standard deviations. The presenter noted that approximately 91 percent of RwandaEQUIP schools were single-shift schools.
In contrast, the effects were close to zero across subjects in double-shift schools. The presenter explained that one possible mechanism was related to teacher workload and fatigue. When schools divide pupils into smaller classes but operate multiple shifts, teachers may have to teach additional groups during the same day. Consequently, the benefits associated with having fewer pupils in each classroom may be offset by the increased workload placed on teachers.
This finding suggested that class-size policies could not be evaluated independently of the broader organization of schooling. Reducing class size without considering teacher workload and school scheduling could limit or even eliminate the expected benefits.
The study also examined an important practical policy question: How small do classes need to be before additional reductions stop generating meaningful learning gains?
The findings suggested that the relationship between class size and learning was not linear. The presenter reported that learning outcomes suffered when classes contained more than 45 pupils per stream. The largest gains appeared when classes were reduced to approximately 40–45 pupils per stream. Below 40 pupils, the gains began to plateau, while reductions below approximately 35 pupils generated diminishing returns.
For English, SET, and SRS, the study found that pupil-teacher ratios of 40 or fewer pupils were associated with effects ranging from approximately 0.40 to 0.42 standard deviations. At ratios of 45 or fewer pupils, the effects were approximately 0.31 to 0.34 standard deviations. At ratios below 35 pupils, the effects remained positive but were largely statistically insignificant.
The presenter therefore argued that the evidence did not support an unlimited policy of progressively reducing class sizes. Instead, there appeared to be a threshold beyond which the additional benefits of smaller classes became limited relative to their financial cost.
The researchers identified personalized instruction and increased teacher attention as an important mechanism through which smaller classes could improve learning.
The presenter explained that when teachers have fewer pupils in their classrooms, they have more opportunities to engage with individual students. They can spend more time providing feedback, checking whether pupils understand lessons, identifying learning difficulties, answering questions, and adapting instruction to individual needs.
This mechanism was particularly important because the benefits of smaller classes appeared to depend on classes becoming sufficiently small to allow meaningful teacher-pupil engagement. Merely reducing a very large class by a few pupils may not necessarily create the conditions required for individualized instruction. However, bringing class sizes down to an appropriate range could provide teachers with substantially more opportunities to interact with individual learners.
The study therefore suggested that class-size reduction works not simply because there are fewer children physically present in a classroom, but because it can change the nature and intensity of teacher-student interaction.
Based on the findings, Osadolor outlined several policy priorities for Rwanda's Ministry of Education (MINEDUC) and its education partners.
Prioritize Grades 1–3
The presenter recommended that class-size reduction should initially focus on Grades 1–3, where the marginal returns appeared to be highest. Because these grades establish foundational literacy, numeracy, language, and learning skills, investments in reducing class sizes at this stage could generate greater educational benefits than blanket reductions across all primary grades.
Establish 40–45 Pupils per Stream as a Practical Benchmark
The study suggested that 40–45 pupils per stream could represent a practical national benchmark. The presenter argued that reducing classes below approximately 40 pupils did not consistently generate additional significant gains, while doing so would require substantial additional financial resources.
The implication was that policymakers should seek an optimal rather than a minimum class size—one that balances learning benefits against the fiscal cost of additional teachers and classrooms.
Accelerate the Transition to Single-Shift Schooling
The researchers recommended accelerating the transition toward single-shift schooling, given the stronger effects observed in single-shift schools.
The evidence suggested that the benefits of smaller classes could be undermined when teachers were required to teach multiple shifts. Reducing class size should therefore be accompanied by efforts to manage teacher workload and ensure that teachers have sufficient time and energy to provide individualized attention.
Combine Class-Size Reduction with Teacher Support
Another major recommendation was that class-size reduction should not be implemented as a stand-alone intervention. The presenter emphasized the importance of combining smaller classes with structured lesson guides, teacher coaching, and continuous professional development (CPD).
This recommendation was particularly important given the decline in learning effects between Term 1 and Term 3. The findings indicated that smaller classes could initially improve learning but that these gains were difficult to sustain without continued pedagogical support.
Avoid Blanket Class-Size Reduction
Finally, the presenter cautioned against implementing uniform class-size reductions across all primary grades. The evidence showed that the benefits were strongest in the early grades and weaker and less consistent in Grades 4–6.
A more targeted approach would therefore prioritize younger learners, schools where class sizes are particularly large, and contexts where teachers can effectively use the additional instructional time created by smaller classes.
The central message of the presentation was that class size does matter for learning—but its impact depends strongly on where, when, and how class-size reduction is implemented.
The Rwanda evidence suggested that reducing very large classes from approximately 52 to 36 pupils could produce substantial short-term learning improvements, particularly in English, SET, and SRS. However, the disappearance of much of these gains by the end of the academic year demonstrated that smaller classes alone could not solve the learning crisis.
The strongest effects were concentrated among early-grade pupils and single-shift schools, while the largest gains appeared when classes were brought into the 40–45 pupil range. Further reductions below this level produced diminishing returns.
The findings consequently pointed toward a more strategic education policy: rather than attempting to make every class as small as possible, Rwanda could achieve greater value from its education investments by prioritizing early grades, targeting overcrowded classrooms, aiming for approximately 40–45 pupils per stream, strengthening single-shift schooling, and combining class-size reduction with sustained teacher coaching, structured instructional support, and professional development.
In this way, the study demonstrated that the question facing education policymakers was not simply whether to reduce class size, but how to design class-size policies that maximize learning gains while remaining financially and institutionally sustainable.
The presentation positioned the study within the broader learning crisis facing many African education systems. The presenter explained that although Rwanda had achieved substantial progress in school enrolment, increased access had also generated significant pressure on the capacity of schools, classrooms, and teachers. The central concern, therefore, was that increasing enrolment alone did not necessarily translate into improved learning outcomes.
Osadolor explained that Rwanda had experienced significant growth in primary-school enrolment, but that this expansion had been accompanied by overcrowded classrooms. According to the figures presented, public primary schools in Rwanda had a pupil-classroom ratio of approximately 60 students per classroom in 2023/24. The pupil-teacher ratio was also reported to be around 48 students per teacher, considerably higher than global averages and continuing to rise as enrolment increased by approximately 6 percent annually.
The presenter noted that these pressures were occurring despite education reforms aimed at improving learning. She highlighted evidence indicating that approximately 80 percent of lower-primary pupils remained below expected reading levels, illustrating the scale of Rwanda's learning challenge. The presentation also placed Rwanda's experience within the wider Sub-Saharan African context, where approximately 86 percent of children were described as experiencing learning poverty, meaning that they were unable to read and understand a simple text by around age ten before the COVID-19 pandemic.
Osadolor argued that these figures demonstrated that the education challenge was no longer simply one of ensuring that children entered school. Instead, the policy focus increasingly needed to shift toward ensuring that children actually learned while they were in school.
The presenter explained that reducing class sizes could potentially improve learning because teachers would have fewer pupils to manage and could provide more individualized attention. However, she emphasized that class-size reduction was an expensive education intervention.
Reducing the number of pupils per class would require governments to recruit, train, deploy, and retain additional teachers. It could also require additional classrooms and other educational infrastructure. This was particularly important in Rwanda because the country was already operating with a relatively high pupil-teacher ratio while simultaneously experiencing continued growth in enrolment.
The presentation therefore framed the issue as a policy optimization problem. Rather than asking simply whether smaller classes were beneficial, policymakers needed evidence on who benefited, under what circumstances, and how small classes needed to become before learning improvements appeared.
Osadolor explained that the Rwanda study provided an important contribution because the country's starting pupil-teacher and class-size ratios were unusually high compared with many previous studies of class-size reduction.
She noted that the intervention examined in the study reduced class sizes from approximately 52 pupils to around 36 pupils, representing a substantially larger reduction than those examined in some influential earlier studies. For comparison, she referred to the work of Angrist and Lavy, which examined a reduction of approximately eight pupils, and the well-known Tennessee Student/Teacher Achievement Ratio (STAR) experiment, in which class sizes were reduced from approximately 22 to 15 pupils.
According to the presenter, Rwanda's much higher starting class sizes raised a different set of policy questions. Evidence from systems where classes were already relatively small might not necessarily apply to education systems operating with very high pupil-teacher ratios. The Rwanda case therefore offered an opportunity to investigate whether substantial reductions in class size could generate measurable learning benefits in a high-ratio environment.
The presenter then provided an overview of Rwanda's education system and the policy programme through which the class-size threshold was implemented. She explained that Rwanda's education system followed a 3-6-3-3 structure, consisting of pre-primary, primary, lower secondary, and upper secondary education. Basic education had been free and compulsory since 2003.
She also noted that many public primary schools operated on a double-shift system, meaning that different groups of pupils attended school at different times of the day. This arrangement was part of the system's response to limited infrastructure and high enrolment pressures.
The study was closely linked to RwandaEQUIP, an education reform programme launched in January 2022 and designed to improve the quality of teaching and learning in public primary schools across the country.
According to the presentation, RwandaEQUIP incorporated several major interventions, including scripted lesson guides, teacher tablets, ten-day teacher training, weekly coaching, and real-time data monitoring. An important feature of the programme was the introduction of a 69-pupil enrolment threshold, which created a discontinuity in how schools organized pupils into streams or classes.
Osadolor explained that RwandaEQUIP had been implemented in phases. The first phase began in February 2022, covering 100 schools in 13 districts across five provinces. The second phase began in September 2022, expanding the programme to 250 schools. A third phase started in September 2024, expanding coverage to 761 schools.
The presenter explained that this phased national expansion created a valuable opportunity for researchers because administrative education data could be combined with the programme's enrolment threshold to examine the relationship between class size and student learning.
The study relied on administrative data covering 761 public primary schools and approximately 4,471 school-by-grade observations. The researchers combined several sources of information to construct a comprehensive dataset.
The data included Foundational Literacy and Numeracy (FLN) baseline assessments, as well as end-of-term examination scores from Terms 1 and 3 of the 2023/24 academic year.
The analysis covered five subjects: English, Kinyarwanda, Mathematics, Science and Elementary Technology (SET), and Social and Religious Studies (SRS). The researchers also incorporated contextual variables such as enrolment, number of streams, school-shift status, province, and RwandaEQUIP implementation phase.
The presenter provided descriptive statistics for Grade 1 and Grade 6. She reported that the average pupil-to-stream ratio was approximately 51.5 pupils in Grade 1 and 39.9 pupils in Grade 6. Average English scores were about 50.9 percent in Grade 1 and 50.2 percent in Grade 6, while Kinyarwanda scores were approximately 46.6 percent and 50.8 percent, respectively.
Mathematics performance showed a wider difference between the grades, with an average score of approximately 42.9 percent in Grade 1 compared with 77.7 percent in Grade 6. SET scores were approximately 54.0 percent in Grade 1 and 33.4 percent in Grade 6, while SRS scores were approximately 55.1 percent and 32.9 percent.
The data also showed that the average age of pupils was approximately 7.9 years in Grade 1 and 14.4 years in Grade 6. Average enrolment per grade was approximately 163 pupils in Grade 1 and 66 pupils in Grade 6. Female pupils accounted for approximately 46 percent of Grade 1 enrolment and 56 percent of Grade 6 enrolment.
A major methodological component of the presentation was the study's use of a Fuzzy Regression Discontinuity Design (RDD).
Osadolor explained that the RwandaEQUIP policy created a natural threshold at 69 pupils. When enrolment reached or exceeded this threshold, schools became substantially more likely to divide pupils into at least two streams. However, the threshold did not perfectly determine whether schools actually divided their pupils into multiple classes. Because there was therefore only partial compliance with the threshold rule, the researchers used a fuzzy rather than a sharp regression discontinuity design.
The presenter explained that schools just below and just above the 69-pupil threshold were likely to be relatively similar in many respects. However, crossing the threshold substantially increased the probability that the school would create at least two streams.
According to the estimates presented, crossing the threshold increased the probability of having at least two streams by approximately 49 percentage points. The average number of streams also increased from approximately 1.3 to 1.9 at the threshold.
This discontinuity provided the researchers with a credible source of quasi-experimental variation in class size. Rather than simply comparing schools with large and small classes—which could produce biased results because the schools might differ in many other ways—the approach allowed the researchers to compare schools located around the policy threshold.
The presenter explained that the researchers estimated the effects using a two-stage least squares (2SLS) framework.
In the first stage, the analysis estimated how crossing the 69-pupil threshold affected the number of streams or class organization. This stage established the strength of the relationship between the enrolment threshold and actual class-size changes.
In the second stage, the predicted class-size or stream changes from the first stage were used to estimate their effects on student learning outcomes. The approach therefore allowed the researchers to isolate the portion of class-size variation generated by the policy threshold rather than relying on simple observational correlations.
The presenter emphasized that this methodology was particularly valuable because schools did not necessarily comply perfectly with the threshold rule. The fuzzy RDD therefore accounted for the fact that crossing the threshold increased the probability of creating additional streams without making such a change automatic.
The presentation highlighted the broader policy importance of understanding class-size effects in high-enrolment African education systems. The presenter suggested that evidence from Rwanda could help policymakers determine whether investments in additional teachers and classrooms would translate into meaningful learning improvements.
If reducing very large classes generated substantial learning gains, policymakers could justify additional investments in teachers, classrooms, and school infrastructure. However, if the effects were small or concentrated among particular groups of students, governments might need to consider whether alternative investments—such as teacher training, instructional materials, coaching, or targeted interventions—could provide greater educational returns.
The study was therefore presented as moving beyond the simple question of whether “smaller classes are better”. Instead, it sought to identify the conditions under which class-size reductions were most effective and the students who benefited most from such changes.
In conclusion, Nneka Esther Osadolor's presentation highlighted the importance of generating rigorous evidence to guide education policy in countries facing rapid enrolment growth and persistent learning challenges. The Rwanda study examined a particularly important policy question because the country had achieved significant progress in expanding access to education while continuing to face overcrowded classrooms and weak learning outcomes.
By exploiting the 69-pupil enrolment threshold under RwandaEQUIP and applying a fuzzy regression discontinuity design, the study sought to estimate the causal effects of reducing class sizes in a high-pupil-teacher-ratio environment.
The presentation ultimately emphasized that the policy debate should not be limited to whether class-size reduction works in general. Rather, policymakers need to understand for whom smaller classes matter, how large the reduction needs to be, through which mechanisms learning improves, and whether the benefits justify the substantial financial and institutional costs of hiring additional teachers and expanding infrastructure.
The study was thus positioned as an important contribution to the growing African evidence base on education quality, learning outcomes, and the effective allocation of scarce public resources.
Presenting the main findings of the study, Nneka Esther Osadolor explained that reducing class size produced meaningful short-term improvements in student learning, although the magnitude of the effects differed considerably across subjects.
The study found that reducing class size by approximately 16 pupils per stream—from around 52 to 36 pupils—generated significant improvements in several learning outcomes during the initial period of the intervention. English scores increased by approximately 0.23 standard deviations, while Science and Elementary Technology (SET) improved by approximately 0.22 standard deviations. Social and Religious Studies (SRS) recorded an improvement of approximately 0.20 standard deviations. The effect on Kinyarwanda was smaller, at around 0.13 standard deviations, and was described as marginally significant. Mathematics, however, showed an effect close to zero and was not statistically significant.
The presenter noted that the magnitude of the effects was broadly comparable to findings from earlier influential studies of class-size reduction. She explained that Angrist and Lavy's 1999 study, which examined a reduction of approximately eight pupils, had found an effect of around 0.20 standard deviations, while the STAR experiment examined a reduction from approximately 22 to 15 pupils and reported an effect of around 0.25 standard deviations. In comparison, the Rwanda study found an effect of approximately 0.23 standard deviations from a much larger reduction, from approximately 52 to 36 pupils. This comparison suggested that meaningful learning gains could also be achieved in education systems where starting class sizes were substantially larger.
The presenter then highlighted an important qualification to the initial findings. Although class-size reduction generated substantial improvements during Term 1, these gains were considerably smaller by Term 3, suggesting that simply reducing the number of pupils in a classroom was not sufficient to guarantee sustained improvements in learning.
For English, the initial effect of approximately 0.23 standard deviations declined to only about 0.04 standard deviations by Term 3. Similarly, the SET effect declined from approximately 0.22 standard deviations to 0.04, while the SRS effect declined from approximately 0.20 to 0.02 standard deviations. The Kinyarwanda effect declined from approximately 0.13 to 0.07 standard deviations. Interestingly, Mathematics, which had shown virtually no initial effect, recorded a small positive effect of approximately 0.06 standard deviations by Term 3.
Based on these findings, Osadolor argued that class-size reduction alone was not sufficient to sustain learning gains over time. She suggested that maintaining improvements would require complementary investments in teachers and instructional systems, including coaching, structured lesson guides, continuous professional development, and other forms of pedagogical support.
The findings therefore pointed to an important policy lesson: smaller classes could create better conditions for learning, but the benefits could diminish if teachers did not receive continued support to take advantage of those conditions.
The study also found substantial differences in the effects of class-size reduction across grade levels. According to the presenter, the largest and most consistent gains were observed among pupils in Grades 1–3, corresponding to the lower-primary years.
In these early grades, English scores increased by as much as 0.35 standard deviations, while Kinyarwanda showed an effect of approximately 0.37 standard deviations. The effects were particularly strong for SET, where improvements reached as high as approximately 0.60 standard deviations.
Importantly, the presenter noted that the positive effects observed in the lower-primary grades were more likely to persist into Term 3. This contrasted with the results for Grades 4–6, where the effects were weaker and less consistent. In many cases, the effects among upper-primary students became negative by Term 3.
These findings suggested that the timing of class-size interventions matters considerably. The presenter argued that reducing class size in the early years of primary education may generate greater returns because foundational literacy, numeracy, language, and learning behaviours are established during these formative years.
Another important finding concerned the organization of the school day. The researchers examined whether the effects of class-size reduction differed between schools operating on a single-shift system and those operating on a double-shift system.
The results indicated that single-shift schools experienced the strongest benefits from class-size reduction. In these schools, English scores improved by approximately 0.18 standard deviations, SET by approximately 0.25 standard deviations, and SRS by approximately 0.20 standard deviations. The presenter noted that approximately 91 percent of RwandaEQUIP schools were single-shift schools.
In contrast, the effects were close to zero across subjects in double-shift schools. The presenter explained that one possible mechanism was related to teacher workload and fatigue. When schools divide pupils into smaller classes but operate multiple shifts, teachers may have to teach additional groups during the same day. Consequently, the benefits associated with having fewer pupils in each classroom may be offset by the increased workload placed on teachers.
This finding suggested that class-size policies could not be evaluated independently of the broader organization of schooling. Reducing class size without considering teacher workload and school scheduling could limit or even eliminate the expected benefits.
The study also examined an important practical policy question: How small do classes need to be before additional reductions stop generating meaningful learning gains?
The findings suggested that the relationship between class size and learning was not linear. The presenter reported that learning outcomes suffered when classes contained more than 45 pupils per stream. The largest gains appeared when classes were reduced to approximately 40–45 pupils per stream. Below 40 pupils, the gains began to plateau, while reductions below approximately 35 pupils generated diminishing returns.
For English, SET, and SRS, the study found that pupil-teacher ratios of 40 or fewer pupils were associated with effects ranging from approximately 0.40 to 0.42 standard deviations. At ratios of 45 or fewer pupils, the effects were approximately 0.31 to 0.34 standard deviations. At ratios below 35 pupils, the effects remained positive but were largely statistically insignificant.
The presenter therefore argued that the evidence did not support an unlimited policy of progressively reducing class sizes. Instead, there appeared to be a threshold beyond which the additional benefits of smaller classes became limited relative to their financial cost.
The researchers identified personalized instruction and increased teacher attention as an important mechanism through which smaller classes could improve learning.
The presenter explained that when teachers have fewer pupils in their classrooms, they have more opportunities to engage with individual students. They can spend more time providing feedback, checking whether pupils understand lessons, identifying learning difficulties, answering questions, and adapting instruction to individual needs.
This mechanism was particularly important because the benefits of smaller classes appeared to depend on classes becoming sufficiently small to allow meaningful teacher-pupil engagement. Merely reducing a very large class by a few pupils may not necessarily create the conditions required for individualized instruction. However, bringing class sizes down to an appropriate range could provide teachers with substantially more opportunities to interact with individual learners.
The study therefore suggested that class-size reduction works not simply because there are fewer children physically present in a classroom, but because it can change the nature and intensity of teacher-student interaction.
Based on the findings, Osadolor outlined several policy priorities for Rwanda's Ministry of Education (MINEDUC) and its education partners.
Prioritize Grades 1–3
The presenter recommended that class-size reduction should initially focus on Grades 1–3, where the marginal returns appeared to be highest. Because these grades establish foundational literacy, numeracy, language, and learning skills, investments in reducing class sizes at this stage could generate greater educational benefits than blanket reductions across all primary grades.
Establish 40–45 Pupils per Stream as a Practical Benchmark
The study suggested that 40–45 pupils per stream could represent a practical national benchmark. The presenter argued that reducing classes below approximately 40 pupils did not consistently generate additional significant gains, while doing so would require substantial additional financial resources.
The implication was that policymakers should seek an optimal rather than a minimum class size—one that balances learning benefits against the fiscal cost of additional teachers and classrooms.
Accelerate the Transition to Single-Shift Schooling
The researchers recommended accelerating the transition toward single-shift schooling, given the stronger effects observed in single-shift schools.
The evidence suggested that the benefits of smaller classes could be undermined when teachers were required to teach multiple shifts. Reducing class size should therefore be accompanied by efforts to manage teacher workload and ensure that teachers have sufficient time and energy to provide individualized attention.
Combine Class-Size Reduction with Teacher Support
Another major recommendation was that class-size reduction should not be implemented as a stand-alone intervention. The presenter emphasized the importance of combining smaller classes with structured lesson guides, teacher coaching, and continuous professional development (CPD).
This recommendation was particularly important given the decline in learning effects between Term 1 and Term 3. The findings indicated that smaller classes could initially improve learning but that these gains were difficult to sustain without continued pedagogical support.
Avoid Blanket Class-Size Reduction
Finally, the presenter cautioned against implementing uniform class-size reductions across all primary grades. The evidence showed that the benefits were strongest in the early grades and weaker and less consistent in Grades 4–6.
A more targeted approach would therefore prioritize younger learners, schools where class sizes are particularly large, and contexts where teachers can effectively use the additional instructional time created by smaller classes.
The central message of the presentation was that class size does matter for learning—but its impact depends strongly on where, when, and how class-size reduction is implemented.
The Rwanda evidence suggested that reducing very large classes from approximately 52 to 36 pupils could produce substantial short-term learning improvements, particularly in English, SET, and SRS. However, the disappearance of much of these gains by the end of the academic year demonstrated that smaller classes alone could not solve the learning crisis.
The strongest effects were concentrated among early-grade pupils and single-shift schools, while the largest gains appeared when classes were brought into the 40–45 pupil range. Further reductions below this level produced diminishing returns.
The findings consequently pointed toward a more strategic education policy: rather than attempting to make every class as small as possible, Rwanda could achieve greater value from its education investments by prioritizing early grades, targeting overcrowded classrooms, aiming for approximately 40–45 pupils per stream, strengthening single-shift schooling, and combining class-size reduction with sustained teacher coaching, structured instructional support, and professional development.
In this way, the study demonstrated that the question facing education policymakers was not simply whether to reduce class size, but how to design class-size policies that maximize learning gains while remaining financially and institutionally sustainable.