The unintended consequence of COVID-19 containment measures in Africa: A double machine learning perspective

This paper examines how COVID-19 containment measures affected multidimensional welfare across seven sub-Saharan African countries. We collected novel survey data from 840 respondents and applied Double Machine Learning (DML) with dual treatments to identify causal effects while addressing confounding and selection bias. Stay-home orders reduced quality of life by 0.184 to 0.489 standard deviations. These effects operated through increased financial difficulties (0.074 SD) and social disruption (0.095 SD). Curfews generated smaller effects, primarily affecting mental health (0.066 SD). Tanzania experienced improved quality of life (0.374 SD) under limited restrictions, while Zimbabwe faced severe negative impacts (-0.489 SD). These contrasts reveal that institutional capacity critically moderates policy effects. Neither intervention significantly affected insecurity or violence. Our findings demonstrate that containment measures impose substantial, persistent welfare costs 24–30 months post-implementation, with effects varying dramatically by policy design and institutional context. Our results advance pandemic policy evaluation by providing the first rigorous causal evidence across multiple African countries using machine learning methods. They reveal that optimal policy design requires context-specific approaches rather than uniform strategies. These insights inform pandemic preparedness and social protection design in developing economies.