Predicting the issuance of COVID-19 stay-at-home orders in Africa: Using machine learning to develop insight for...

During the COVID-19 pandemic, many countries have issued stay-at-home orders (SAHOs) to reduce viral transmission. Because of their social and economic consequences, SAHOs are a politically risky decision for governments. Researchers typically attribute public health policymaking to five theoretically significant factors: political, scientific, social, economic, and external. However, a narrow focus on extant theory runs the risk of biasing findings and missing novel insights. This research employs machine learning to shift the focus from theory to data to generate hypotheses and insights "born from the data" and unconstrained by current knowledge. Beneficially, this approach can also confirm the extant theory. We apply machine learning in the form of a random forest classifier to a novel and multiple-domain data set of 88 variables to identify the most significant predictors of the issuance of a COVID-19-related SAHO in African countries (n = 54). Our data set includes a wide range of variables from sources such as the World Health Organization that cover the five principal theoretical factors and previously ignored domains. Generated using 1000 simulations, our model identifies a combination of theoretically significant and novel variables as the most important to the issuance of a SAHO and has a predictive accuracy using 10 variables of 78%, which represents a 56% increase in accuracy compared to simply predicting the modal outcome.