Reliable knowledge claims on the recruitment and use of children: An empirical perspective

Abstract The risks of child recruitment by non-state armed groups are geographically, temporally and contextually situated. There are multilayered, multivariate arrays of risk factors associated with non-state armed groups, with conflicts, and with contexts. Using Bayesian network modelling with a global dataset of non-state armed group child recruitment practices between 2010 and 2022, we demonstrate the theoretical and practical importance of adopting a situational perspective to understand child recruitment risks. Methodologically, we demonstrate a robust model-checking process that checks the adequacy of our data, the magnitude and direction of estimated effects, and shows greater than 80% accuracy in predicting child recruitment by non-state armed groups. We review and contrast our approach with standard general linear modelling used in quantitative child recruitment research over the past two decades. Through adopting a situated orientation, and applying analytical tools appropriate to that orientation, we challenge and extend existing theory and propose new theoretical insights on child recruitment risks. We show how important violence is as a predictor of child recruitment risks and, using a new measure of fighting force efficacy, show that, contrary to published theory, less effective non-state armed groups were more likely to recruit children than more effective ones. But even these most notable results we show to vary markedly across situations.