Better measures of progress: Developing reliable estimates of educational access and quality in Francophone sub-Saharan Africa

When it comes to development goals, an estimate is only as good as its measurement. There is a long history of developmental goal setting by national governments and international organisations, but far less emphasis on how those goals are measured accurately, especially over time. The measurement of new goals, such as learning, needs to be carefully thought-through and published estimates should reflect this process. This research tackles one prominent source of measurement error in large-scale cross-national cognitive assessment data: sample selection bias. Sample selection bias is a problem in assessment data wherever assessments are conducted within schools and there is below universal access to schooling. Francophone sub-Saharan Africa has some of the lowest schooling rates worldwide and therefore some of the largest bias in its regional assessment data. This paper follows and updates a methodology first conceptualized by Spaull and Taylor in 2015. The new aspects of the methodology allow estimates adjusted for sample selection to be calculated immediately on the release of assessment data, rather than many years hence. After adjusting for sample selection, this paper finds that published learning estimates in Francophone sub-Saharan Africa are vast overestimates of the true rates of literacy and numeracy in the region.