Machine learning methods to uncover mechanisms underlying the impacts of two long-term evaluations of youth skills training programmes in Uganda (7-year follow-up)
Programme of work
Enhancing evidence transferability
Innovations for Poverty Action Kenya
University of California, Berkeley
University of California, Los Angeles
February 2020 to January 2023 (TBC)
This study aims to shed light on the underlying mechanisms and compo-nents through which two innovative youth skills development and entre-preneurship interventions in Uganda operate. This will be done through collecting data in a 7-year follow-up to assess and compare the long-term impacts of three youth skills curricula that feature different combinations of soft and hard skills; namely, SEED-hard (25% focus on soft skills), SEED-soft (75% focus on soft skills) and Educate! (approximately 90% soft skills).
Follow-up instruments will be designed to shed light on the underlying mechanisms and components through which these interventions operate and yield impacts. Machine learning methods (i.e. regression trees) and causal mediation analysis will be combined to study how the programmes shape skills, how these are differentially rewarded in the labour market and their social spillovers (e.g. risky behaviours, partnership quality and intimate partner violence). This innovative methodology will go beyond the ‘effect of a cause’ (i.e. the treatment effect) and investigate the ‘cause of the effect‘ (i.e. the channels through which the effect on final outcomes is manifested).
The study will involve 7-year follow-up data collection on two youth skills-development and entrepreneurship interventions in Uganda, which were both evaluated at scale as randomised controlled trials.
This study aims to inform the debate on the optimal combination of soft and hard skills in the design of entrepreneurship training programmes. It will help develop a better understanding of which skills and underlying mechanisms are important for the development of entrepreneurship in communities.