What AFLEARN Taught Us About Doing Better Education Research

Education research in Africa has generated a growing body of observational data. However, most of this data is analysed using designs that cannot support credible causal claims. While not all research seeks to establish causal evidence, where such is intended, getting the methods right is how we ensure this work produces evidence that policy makers can act on.

During the AFLEARN Residency at the University of Cape Town, we spent ten days engaging deeply with questions of research and evidence. We first participated in a one-week residential Education Data Workshop designed to deepen researchers’ ability to work with complex education datasets and generate rigorous, policy-relevant insights. The workshop focused on research writing, data landscape, regression, and causal inference. This was then followed by five days of applying those lessons to our own work. 

What follows are our reflections on what stayed with us.

Good writing is harder than it looks

Dr Linda Zuze opened the workshop with a session on applied research writing. Her challenge was deceptively simple: write your research problem in one paragraph, and describe your data in another. It sounds manageable — until you sit down and try. Every word feels load-bearing, and cutting feels like loss. 

The standard she pointed us to was high; that a well-written article should pull you in so completely that you forget you are reading a research paper¹. It is a high bar — but one worth striving for.

Landscape of education data in Africa

Professor Martin Gustafsson’s session was quietly humbling. Harmonising SACMEQ, PASEC, MICS, TIMSS, PIRLS, and other datasets sounds like a well-defined exercise in merging and appending — until you see what it actually entails: fragile comparability across assessments, uneven country participation, data quality losses, and contested decisions about which data even counts.

What looked effortless in his presentation represented years of painstaking work and difficult judgment calls. Martin connected all of it to SDG 4, but the deeper message was reassuring: measuring foundational learning progress across Africa is genuinely hard, and the politics of data selection are inseparable from the research. Anyone working with these datasets casually hasn't earned that confidence.

Press regress — but not yet

"Getting to know your data" sounds obvious. It isn’t. Dr Amy Thornton’s two-day session on research pipelines made explicit what many analysts overlook: the process is iterative, not linear, and rushing to estimation before understanding your data produces unreliable results, which are presented with greater confidence than they deserve. 

Who is in the data? How was it generated? Do the variables measure what you think they measure? These questions can be considered as preliminary, but they represent the core work. 

Dr Emma Whitelaw's regression lab reinforced the point. Before adding controls, understand why you are adding them. Theory and your empirical question should drive variable selection. Without that discipline, you risk bad controls involving the variables that are themselves outcomes of your treatment, and consequently, which bias rather than improve the precision of your estimates.

Causal inference: moving beyond before-and-after

This was the session we needed most. Working in education, we have relied on pre-and-post comparisons longer and consciously understood the downside. They are useful for a general picture, but they almost certainly overestimate effects by failing to account for selection bias. 

Professor Cally Ardington's sessions on quasi-experimental designs reframed what rigorous evaluation should look like in contexts where randomised controlled trials face ethical and practical constraints — which, in African education research, is often the case. The goal of these designs is precise: isolate the causal effect of an intervention by constructing a credible counterfactual. 

Regression discontinuity, difference-in-differences, and instrumental variables are not just jargon but methods that are both achievable and necessary. They bring us closer to an honest answer to the question of whether an intervention worked. 

We left Cape Town more convinced that the field needs to raise this standard, and more equipped to push for it in our own work.

Applying the training to our own work

The final five days provided an opportunity to apply these lessons to our own work — and, in doing so, revealed some uncomfortable gaps. In several cases, we simply did not have the right data for the questions we wanted to ask.

We didn’t always find answers, but we learned to ask better questions. That feedback, though difficult, was ultimately the most valuable part.

Beyond the sessions

Beyond the learning experience, Cape Town itself was a highlight — truly a beautiful place and well worth a visit. From Table Mountain and the colourful streets of Bo-Kaap to the views over Camps Bay, the city offered plenty to take in.

We even tried pap — “ugali” for Kenyans — and later found a plate of chapati served with peanut beef at a Swahili restaurant in the city. Far from East Africa, but for a moment, it didn’t feel like it.

^[1] Marc F. Bellemare, How to Write Applied Papers in Economics, September 7, 2020