This study by SDSN member Ken Giller and colleagues shows that there is as much, if not more, variability in livelihood activities within a region (i.e. among households) as there is among regions. Hence, interventions that aim to target the poor must recognise the large diversity in livelihood activities within any given area, which often overrides the variability between regions.

The authors attempted to identify country-wide patterns in food availability and contributing livelihood activities using spatially explicit environmental and socio-economic data and household resource characteristics as explanatory variables. With few exceptions, predictions of the indicators were weak, highlighting the difficulty in capturing variability at larger scale. Also household explanatory variables identified little additional variation compared to environmental explanatory variables alone. Spatial predictability was strongest for indicators for which environmental gradients determined their distribution, such as highland banana contribution to the crop part of food availability. In contrast, indicators of crops that were more ubiquitously present across agroecological zones showed large local variation, which often overruled large-scale patterns (e.g. cassava and maize contributions to the crop part of food availability, and livestock and off-farm income contributions to food availability).