Layered interventions for climate resilience: an analysis of household recovery and food security outcomes in Southern Malawi

This study assesses the implications of layered resilience-building interventions in climate-vulnerable regions of Malawi, with a focus on household recovery capacity and food security outcomes during the 2024 El Niño–induced drought in Malawi. Household resilience is proxied by self-reported recovery duration following climatic shocks, while food security is measured using Food Consumption Scores (FCS). The analysis draws on cross-sectional data from 309 randomly sampled households in the disaster-prone districts of Mangochi and Zomba, alongside a sample from Phalombe in Southern Malawi. To examine heterogeneity in program exposure, the study applies Machine Learning (ML) classification and clustering techniques to identify distinct intervention intensity profiles. Cluster validation using elbow and silhouette diagnostics reveals two dominant groups: households receiving a high-intensity intervention package (averaging 8 of 17 program interventions) and those exposed to a lower-intensity package (averaging 3 interventions). The effects of these intervention clusters on recovery duration and food security during the El Niño shock are subsequently estimated using Negative Binomial and Ordered Logit Regression models. Findings indicate that while layered interventions did not significantly reduce recovery duration, reflecting the severity of the climatic shock, household-level characteristics, mainly household size and maize productivity, were strong predictors of recovery time. On the other hand, quantity of maize, perceived recovery period, livestock holding and bicycle ownership were predictors of food security. Participating in interventions was not significant. The results suggest that mere participation in integrated programs might not enhance absorptive and consumption-smoothing capacities in times of extreme weather events. The study underscores the importance of absorptive social protection mechanisms (such as food aid) during severe climate events and demonstrates the value of integrating ML approaches into the design and evaluation of complex sustainable development interventions.