Dynamic copula Bayesian network predictive model for assessing the impact of initiative programs on child...

Child undernutrition remains a major public health concern in Ethiopia, influenced by multiple and interacting household and community factors. Despite large-scale initiatives such as the Productive Safety Net Program, Emergency Aid Program, and Health Extension Program, evidence is still needed on how these interventions affect the determinants of child nutritional status over time.

We applied a Dynamic Copula Bayesian Network (DCBN) to model time-varying associations between program participation and key determinants of child undernutrition: food security (FS), household wealth (WQ), and mother subjective well-being (MSW). Data were drawn from the Young Lives-Ethiopia surveys (waves 2009, 2013, 2016) with baseline information from 2002 and 2006. The DCBN framework incorporated 26 copula families, Kendall's τ for dependence measures, and Markov Chain Monte Carlo (MCMC) for parameter estimation. Model performance was evaluated using root mean square error (RMSE) and Nash-Sutcliffe efficiency (NSE). We further accounted for program spillovers through a community program intensity proxy and assessed robustness with baseline conditioning and inverse probability weighting (IPW).

Program participation was positively associated with household food security and wealth. Both FS → CUS and WQ → CUS edges showed negative and strengthening dependencies across waves, indicating that improvements in food security and wealth are associated with reductions in child undernutrition. These associations were robust to baseline conditioning, spillover adjustments, IPW weighting, and estimation method (MCMC vs. local optimization).

The study demonstrates the utility of DCBNs for mapping dynamic, nonlinear associations between social protection and health programs and child undernutrition determinants. The results highlight that strengthening household food security and wealth plays a central role in reducing child undernutrition. Although findings are associational, the transferable dependence map can be re-estimated with contemporary data to guide program targeting, monitoring, and policy decisions in Ethiopia.