The problem
Rwanda's early childhood classrooms are teacher-led by default. Play-based learning is national policy ambition, but at baseline, teachers averaged around four playful practices per 20-minute lesson, most of them low quality. Twigire Mumikino Rwanda ("Let's Learn through Play") set out to shift teacher practice at scale through a tablet-based Teacher App, workshops for head teachers and sector inspectors, and volunteer-supported mentoring and communities of practice.
The programme made an unusual commitment: test the model properly. Not a pre-post survey, but a three-arm randomized controlled trial (control, full intervention, and an app-only arm) across 372 schools in all 30 districts, with IPA leading the evaluation.
What I did
As EdTech Advisor, I designed and strategically implemented the two digital components of the intervention: the VSO Teacher App, a tablet-based application taking teachers through 15 modules on the rationale and practice of Learning through Play, and the accompanying lesson planning template that translated app content into classroom routines. Both had to work in low-connectivity rural schools, on shared devices, for teachers with limited prior digital experience.
Beyond the product, I contributed to the research itself: supporting the design and conduct of the evaluation, the baseline study, and the data analytics behind the endline findings. The evaluation used the PLAY observation toolkit (developed by RTI International and NYU Global TIES for Children, funded by the LEGO Foundation), with over 700 classroom videos coded blind to treatment status at an inter-rater reliability of 0.91.
What the RCT found
Teachers in the full intervention showed a statistically significant improvement in overall play-based instruction compared to control (p = 0.002), driven by gains in exploration and learner agency: teachers bringing objects into lessons for children to handle, and shifting activity from teacher-led to child-led. Regression analysis found a dose-response relationship with the app itself: each additional module completed was positively associated with the level of play-based instruction observed.
The app-only arm was inconclusive. A platform migration delayed delivery, so teachers completed only 2 to 3 of 15 modules before endline. That is itself a finding worth publishing: implementation fidelity, not product quality, is where lightweight EdTech models live or die. The study was extended into the following school year to test the app-only model properly.
Why it matters
Most teacher-facing EdTech ships without evidence. This programme built the counterfactual in from the start and let an independent evaluator report the results, including the uncomfortable ones. For funders and ministries deciding what to scale, that distinction, evaluated versus asserted, is the whole game.
Start a conversation