by Sara Balestri, Andrea Crippa, Luca Pieroni[1]
Food insecurity in Sub-Saharan Africa is a persistent and multidimensional condition. It encompasses not only insufficient access to food, but also limited dietary quality, high vulnerability to shocks, and constrained capacity to recover from them — dimensions that interact with one another and evolve over time in ways that snapshot indicators rarely capture. Despite this complexity, most empirical work in the field relies on observable proxies — such as dietary diversity scores, caloric intake, food expenditures — that each illuminate one facet of the underlying condition while leaving the rest in shadow. The result is a measurement gap with real consequences: two households may look identical on any given indicator yet differ profoundly in their exposure to risk, their coping capacity, and their trajectory over time.
This paper takes that gap seriously. Using over 42,000 household-wave observations from Ethiopia, Malawi, and Nigeria — drawn from the World Bank's Living Standards Measurement Study (LSLM) across four survey waves between 2009 and 2019 — we model food security not as a static outcome but as a latent dynamic process, in which households move between more and less secure states in response to climate stress and structural disadvantage. The methodological framework is a Hidden Markov Model, which infers unobservable latent regimes from an observed signal — in our case, the Household Dietary Diversity Score — treated as an imperfect measure of an underlying condition that is never directly observed. The model selects two latent states: a low dietary diversity regime, in which households are concentrated around moderate food insecurity with a non-negligible risk of severe deprivation, and a high dietary diversity regime, in which households are almost certainly food secure.
A first important result concerns the dynamics of transitions between these two states. Food security is characterised by strong path dependence: households already in the high-diversity state remain there with a probability of 92%, while those in the low-diversity state have only a 65.6% chance of moving upward in any given period. The asymmetry is substantively important — once a more secure condition is reached, it is considerably easier to maintain than to regain — and it suggests that preventing deterioration, rather than simply responding to crises after the fact, should be a central objective of policy.
Against this backdrop, we examine the role of early climate stress. Our key climate variable is an early drought warning indicator based on the Standardised Precipitation Evapotranspiration Index (SPEI), set at a threshold that captures water deficits before conditions reach conventional drought severity. The focus on early-stage stress is deliberate: we find that even mild water deficits significantly reduce the probability of belonging to the food-secure state by 9.8 percentage points on average, a result that holds across alternative SPEI accumulation windows, alternative drought definitions, and different specifications of the outcome variable. The message is direct: the adverse effects of climate stress on household food security begin well before a drought is formally declared, which means that support programmes calibrated on conventional severity thresholds are, by construction, too late.
Perhaps the most striking finding, however, concerns the role of gender. Female-headed households — roughly 24.5% of the full sample — consistently report lower dietary diversity scores across all survey waves and all three countries. This is not an occasional pattern: it reflects a structural and persistent gap in access to the resources, markets, and safety nets that sustain dietary quality over time. In the HMM framework, female-headed households are 5.9 percentage points less likely to reach the high food security state compared to male-headed households, all else equal, with the penalty rising above 7.8 percentage points in the most affected regions. More importantly, the disadvantage compounds when climate stress and gender interact. Female-headed households start from a lower baseline and are disproportionately affected by early drought exposure, facing a substantially larger reduction in their probability of remaining food secure when a drought warning arrives. This compounded vulnerability is consistent with the structural constraints that female household heads typically face — more limited access to land, credit, agricultural inputs, and formal social protection — which narrow the space for adaptation precisely when conditions deteriorate.
Alongside gender, urban residence is associated with a 20.6 percentage point higher probability of belonging to the food-secure state, the largest single effect in the analysis. This reflects not only greater income diversification and market access in urban areas, but also the depth of the disadvantage facing rural households, who rely more heavily on rainfed agriculture, are more directly exposed to climate variability, and hold fewer buffers against the kinds of cumulative shocks that erode food security over time.
Taken together, these findings point to three policy priorities. Early-warning systems need to be connected to adaptive safety nets well before drought reaches conventional severity, since the erosion of food security begins at early stages of water stress. Interventions also need to be geographically targeted and gender-sensitive: subnational variation in both drought and gender effects is substantial, and uniform national responses will systematically miss the most exposed households; female-headed households in particular require dedicated support — through targeted transfers, improved access to agricultural inputs and technologies, or strengthened social protection — especially in regions where the compounded disadvantage is largest. Finally, the magnitude of the rural–urban gap points to a longer-term structural agenda: improving market access, rural infrastructure, and income diversification opportunities would have significant and durable effects on food security, independently of climate policy.
The broader implication runs through all three priorities. Food insecurity in the region is not a uniform or static condition: it is a dynamic process, differentiated across households and territories, shaped by the interaction of structural disadvantage and climate stress, and resistant to recovery once entrenched. Policies that treat it otherwise — as a temporary shortfall to be addressed after the fact, or as a condition distributed uniformly across the population — will fall short of what the evidence demands.
[1] University of Perugia
