Skills trainings and Bayesian learning: A multisite randomized controlled trial in Ghana
Despite substantial investment in skills training worldwide, evidence for the effectiveness of such interventions in sub-Saharan Africa is still relatively sparse. We contribute to this literature by implementing a multisite randomized controlled trial (RCT) of the Professionalization of Artisans (ProfArts) program in Ghana, a state-of-the art training program designed to improve employment quality through skilled trades training. The research design enables the assessment of impacts on labor market outcomes across multiple sites (four different urban labor markets, six providers). We find limited overall impacts, with variation across cities, including notable employment and income effects in Tamale in the less developed north and some benefits in job quality in the more developed central city of Kumasi. We present evidence that these differences are not well explained by variation in program implementation, hinting at the importance of local labor market conditions. We also document how Bayesian implementers could learn from this evaluation and how real-world stakeholders actually learn. We collect data on implementers' expectations regarding the program's effectiveness, both before and after the presentation of RCT results. Stakeholders' beliefs about the program's impact adjust in response to empirical findings, but more optimistically than Bayesian learning would suggest.