📈 economics

Headteacher experience, management quality and pupils’ learning in Sub-Saharan African schools: evidence from PASEC 2019

Analyzing 2019 PASEC data from 14 Sub-Saharan African countries, this study finds that while flexible machine learning models do not outperform linear regressions in predicting student learning, only a headteacher's teaching experience and, in public schools, management quality are robustly associated with higher pupil scores, with no significant governance indicators surviving correction in private schools.

Original authors: Kouassi Sylvain Assienin, Christian K. N’da, Auguste K. Kouakou

Published 2026-11-03

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Kouassi Sylvain Assienin, Christian K. N’da, Auguste K. Kouakou

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

In the landscape of human development, education stands as a fundamental pillar, yet in many parts of Sub-Saharan Africa, the promise of schooling often remains unfulfilled. Children are enrolled in classrooms, but too many leave without acquiring the basic skills needed to read a simple text or solve a math problem. This gap between attendance and learning has become a critical challenge for the region. For years, policymakers have debated where to focus their efforts to fix this broken system. Some argue that more money is the answer, pointing to the fact that many nations already spend a significant portion of their national budgets on education. Others suggest that the volume of resources matters less than how those resources are managed. The school itself is the place where this management happens, and the headteacher is the person in charge. The question driving this research is simple but difficult to answer: which specific actions taken by school leaders actually help children learn better?

To find the answer, a team of researchers turned to a massive dataset known as PASEC 2019. This survey tested more than 25,000 students across 14 French-speaking countries in Sub-Saharan Africa, measuring their performance in reading and mathematics. The researchers did not just look at the test scores; they also gathered detailed information about the schools, including sixteen different indicators of how the schools were run. These indicators covered everything from the quality of administrative management and the frequency of teacher meetings to the nature of relationships between the school and parents. The goal was to see which of these factors were truly linked to higher student scores, while being careful to account for the fact that students in the same school share the same environment. To ensure their findings were solid, the team used advanced statistical methods that treated the schools as distinct groups, preventing them from drawing false conclusions based on the sheer number of students involved. They also compared their results against complex computer models to see if simple, linear relationships could explain the data just as well as more complicated ones.

The study produced a clear and somewhat surprising result. After running the numbers through rigorous tests that corrected for the many different factors being examined, only one governance indicator stood out as robustly linked to better learning outcomes across all schools: the number of years a headteacher had spent teaching in a classroom. This finding held true whether the school was public or private, and the association was positive at every level of student performance. A headteacher with more classroom experience was consistently associated with higher scores for their students. In public schools, a second factor emerged as a possible contributor, though with less certainty: the overall quality of school management. However, in private schools, no governance indicator survived the rigorous testing, suggesting that the factors driving success there might be different or that the data was not precise enough to detect them.

The researchers also investigated whether the relationship between these factors and learning was straightforward or if it changed depending on how well a student was already performing. They found that the benefit of a headteacher's teaching experience was present for students at the bottom of the score distribution, in the middle, and at the top. The connection was strongest near the middle of the pack. This suggests that experienced leaders help all students, not just the struggling ones or the high achievers. In contrast, other common practices, such as the frequency of meetings organized by the headteacher or the way schools rewarded top students, did not show a reliable link to learning once the statistical corrections were applied. In fact, some of these factors appeared significant only before the researchers applied the strict corrections needed to avoid false alarms, and they faded away when the analysis was tightened.

Perhaps the most significant finding of the study was what it ruled out. The researchers tested whether complex, non-linear patterns in the data could explain student success better than simple, straight-line relationships. They used sophisticated machine learning tools, which are often touted for their ability to find hidden patterns, and compared them against standard statistical models. The result was that the complex models did not perform any better than the simple ones when tested on schools they had never seen before. In fact, the simple models were often more accurate. This implies that the relationship between school governance and student learning is not a tangled web of hidden interactions, but rather a set of direct, linear associations that can be understood without needing a computer to find the answer. The data suggests that the observed factors explain about 31% of the differences in student scores across all schools, and about 25% in public schools, but less than 7% in private schools, where the results were too uncertain to draw firm conclusions.

The implications of these findings are specific and grounded in the evidence. The study suggests that when selecting or training school leaders, priority should be given to those with substantial classroom teaching experience, as this is the single factor most reliably associated with student learning. It also points to the potential value of improving the general quality of management in public schools, though this evidence is slightly less certain than the case for experience. Conversely, the study advises against treating the frequency of contact with parents as a primary goal for improvement, as the data showed no consistent link between how often schools contacted parents and how well students learned. The researchers caution that these findings show associations, not cause-and-effect, and that more targeted studies would be needed to confirm that changing these factors will directly improve learning. Nevertheless, the work provides a clear, data-driven map of what matters in the complex world of African education, highlighting the enduring value of the classroom experience of the person leading the school.

Technical Summary: Headteacher Experience, Management Quality and Pupils' Learning in Sub-Saharan African Schools

Problem and Context
Despite significant public investment in education across Sub-Saharan Africa (averaging 5% of GDP), learning outcomes remain critically low, with 89% of ten-year-olds unable to read a simple text. The challenge has shifted from access to "enrolment without learning." While school governance is promoted as a low-cost lever for improvement, existing evidence often relies on in-sample predictions, ignores the hierarchical clustering of pupils within schools, or fails to separate public and private sectors. Furthermore, previous econometric work on PASEC data has rarely utilized methods capable of capturing non-linearities and interactions, nor has it rigorously corrected for multiple hypothesis testing when evaluating a broad set of governance indicators.

Methodology
The study utilizes the 2019 PASEC survey, covering 25,850 pupils across 14 French-speaking Sub-Saharan countries. The analysis relates sixteen governance indicators (derived from headteacher and teacher questionnaires) to a continuous learning score (average of reading and mathematics plausible values).

The methodological framework is distinguished by three core components:

School-Level Validation: To avoid optimistic error estimates common in hierarchical data, the authors employ a "school-split" validation strategy. Whole schools (20% of the sample) are held out as a test set, while the remaining 80% form the training set. This ensures models are evaluated on unseen schools rather than just unseen pupils.
Inference and Correction: The study combines Ordinary Least Squares (OLS) with Double Machine Learning (DML) for partial association estimation. DML uses LightGBM to estimate nuisance functions with school-grouped cross-fitting, allowing for flexible control of confounders while maintaining valid inference on the target parameters. Crucially, the study applies Holm's procedure to control the family-wise error rate across sixteen simultaneous tests, defining "robust" associations as those significant in both OLS and DML with an adjusted p-value below 10%.
Model Comparison and Heterogeneity: The authors compare the predictive accuracy of linear models (OLS, Median Quantile Regression) against flexible non-linear models (Generalized Additive Models, XGBoost, Random Forest, Extra Trees) on the held-out test set. Additionally, quantile regression and penalized-spline additive models (GAM) are used to examine whether associations vary across the distribution of scores or exhibit non-linear functional forms.

Key Results
The analysis yields three primary findings regarding the relationship between governance and learning:

Predictive Performance: Flexible, non-linear models (tree-based ensembles and GAMs) do not outperform linear models on unseen schools. In the pooled sample, OLS and median quantile regression achieve a test $R^2$ of 30.6%, while tree-based models range between 24.4% and 26.2%. In private schools, no model explains more than 7% of the variance on held-out data. This suggests that non-linearities and interactions among observed covariates do not carry significant predictive information beyond linear specifications.
Robust Governance Indicators: After applying Holm's correction for multiple testing:
- All Schools: Only the headteacher's years of teaching is robustly and positively associated with learning (+34.3 points across the observed range).
- Public Schools: Headteacher teaching experience remains robust (+62.9 points). Quality of school management is also associated with learning, though only at the 10% significance level (+10.0 points).
- Private Schools: No governance indicator survives the multiple-testing correction. The lack of significant findings here is attributed to smaller sample sizes (179 training schools) and wider confidence intervals rather than evidence of zero effect.
Heterogeneity: The association between headteacher teaching experience and learning is positive across every decile of the score distribution, peaking near the median. The equality of slopes across deciles is rejected for this variable in both the pooled and public school samples, indicating that the impact of experience varies depending on the student's performance level. Conversely, seniority in the headteacher post is not robustly associated with learning once teaching experience is controlled for.

Significance and Contributions
The paper makes three specific contributions to the literature on education production functions in Sub-Saharan Africa:

Methodological Rigor: It is the first study to evaluate flexible and linear models on held-out schools while testing a full set of governance indicators using clustered inference and multiple-testing correction. This addresses the gap where previous studies often ignored school clustering or failed to correct for the "garden of forking paths" in hypothesis testing.
Refined Understanding of Leadership: The study clarifies that it is the classroom teaching experience of the headteacher, rather than their seniority in the administrative post, that is robustly associated with student learning. This supports transformational leadership theories suggesting that professional credibility with teachers is derived from instructional experience.
Empirical Limits of Governance: The findings suggest that while school governance matters, the specific mechanisms are narrow. In the pooled sample, only teaching experience is a robust predictor. In private schools, the data does not support robust associations for any governance indicator, highlighting the need for caution in policy extrapolation.

Limitations and Cautions
The authors explicitly state that the estimates represent conditional associations, not causal effects, due to the cross-sectional nature of the data and potential unobserved confounding. They note that the 10% significance threshold for management quality in public schools is sensitive; at the 5% level, only teaching experience remains significant. Furthermore, the study treats the mean of plausible values as observed and does not utilize sampling weights, which may lead to underestimated standard errors. The results for private schools are constrained by limited statistical power. Consequently, the authors frame their findings as evidence to inform policy discussions—specifically regarding the selection of headteachers with prior teaching experience—rather than as definitive causal proof for scaling specific interventions.

Technical Summary: Headteacher Experience, Management Quality and Pupils' Learning in Sub-Saharan African Schools

More like this