Bayesian Analysis of Conflict-Induced Displacement and Mortality among Vulnerable Populations in Southern Ethiopia

Abstract Background Humanitarian consequences of intergroup conflict have been reported in Southern Ethiopia, mainly among the most vulnerable group: children under five years, pregnant women, and nursing mothers. No reports have been recorded at the district level in terms of persons displaced and deaths due to conflict. Methods We applied Bayesian Negative Binomial regression for the analysis of conflict-displacement and conflict-caused mortality rates for the periods between July 2020 and January 2024 for the five adjacent districts: Konso, Ale, Burji, Koore, and Gardula. The incidence rate ratio, 95% credible interval, and posterior probability for the posterior quantities were calculated for both models developed. Model adequacy was assessed by Deviance Information Criteria, Watanabe-Akaike Information Criteria, Leave-One-Out cross-validation, and posterior predictive checking. Results There were 116,327 displaced people, with the largest numbers in the Konso region (44,337) and the Gardula region (22,263). Vulnerable groups had a significant proportion of displaced persons, including 5,536 children less than five years of age, 3,780 pregnant women, and 1,370 breastfeeding mothers. The posterior estimations in the Bayesian analysis revealed an increased risk of displacement in the Konso region (IRR 2.10, 95% CrI: 1.76–2.52) and the Gardula region (IRR 1.86, 95% CrI: 1.48–2.33) compared to other regions, as well as an increased risk of death in the Konso region (IRR 1.79, 95% CrI: 1.46–2.21) and the Gardula region (IRR Conclusion IDP risks and deaths are geographically patterned and influence maternal and child groups. District-level strategies tailored to vulnerability are essential to mitigate risks. Bayesian approaches offer an effective methodology for predicting IDPs and deaths, and must form the basis of early warning systems and strategies.