What matters for learning in East Africa? A comparison of education production functions between and within countries Preliminary draft – not for citation

East African education systems have undergone rapid transformation over recent decades. Nonetheless, little is known about what matters for learning across the region. Based on a unique micro-dataset incorporating test scores for almost 350,000 school-aged children in Kenya, Tanzania and Uganda (enrolled and not enrolled), this paper provides reducedform estimates of educational production functions at the regional, country and sub-group levels. The possibility of parameter heterogeneity is admitted at the outset and motivates development of a sample-weighted mean-group estimator that ensures consistent estimation of average marginal effects across heterogeneous groups, such as geographical units. The findings confirm that parameter heterogeneity is substantial. For instance, the average child in Uganda acquires basic skills much later than in Kenya and Tanzania, holding all other inputs constant. However, there are also regularities. The contribution of family background factors to learning is most crucial. School inputs are statistically significant but play a moderate role on average. Amongst these, teacher-pupil ratios and the size of schools are both associated with the largest positive gains in learning.