Measuring Cognitive Skills in African Labor Markets: A Bounding Approach to Non‐Completion in the STEP Surveys

ABSTRACT

This paper addresses the problem of assessment non‐completion in surveys within the context of cognitive skills measurement in sub‐Saharan Africa. The World Bank's STEP Skills Measurement Program represented a major initiative to measure skills in low‐ and middle‐income countries, with surveys conducted in two African countries, Ghana and Kenya. While STEP aimed at measuring cognitive skills by producing literacy skill scores comparable to those of the PIAAC assessment of skills in OECD countries, I show that participation in the standardized literacy assessment suffered from high rates of non‐completion, amounting to 60% in Ghana and 33% in Kenya. I then highlight STEP's methodology to address non‐completion, documenting that STEP's treatment of non‐completers depends on the reason for non‐completion rather than the assessment stage reached. I further point out a divergence in the scoring procedure between Ghana and Kenya that compromises STEP's cross‐country comparability. In response, I propose a decomposition that sequentially incorporates the non‐completers using a harmonized imputation model for both countries, complemented by nonparametric bounds. I find that under varying assumptions about the non‐completers, the imputation produces a narrower bound on the Ghanaian and Kenyan skill distributions than the nonparametric approach, while still implying sizable skill gaps relative to literacy skills measured in OECD countries. I also show that the treatment of non‐completion can materially shift estimated returns to skills. The findings provide guidance from a partial identification perspective for handling assessment non‐completion problems in applied research and for the design of future skill assessments in low‐ and middle‐income countries.