Kenya’s Financial Lives Mapped: Food Vulnerability in FinAccess 2024
Data & Evidence
Institutions
This post uses the FinAccess 2024 public data to explore food vulnerability in Kenya.It builds a decision tree model to identify subgroups where food vulnerability is most pronounced.
Recently, I learned that nearly four in ten Kenyan adults in the FinAccess 2024 survey reported going without enough food in the previous month. I had expected food pressure to appear in the data, but not at that scale. The national number is high, but it is also useful to look at which subgroups in the country are more disadvantaged: food vulnerability is not spread evenly. It is concentrated among particular households, in particular counties, and around particular life conditions: casual work, weak credit options, limited education, and more dependents. This post uses the public FinAccess data to ask a more practical question: who is most exposed to food vulnerability, and where are those profiles concentrated?
County food vulnerability ranges from about 12% in Bomet to 87% in Turkana. Turkana, Tana River, and West Pokot sit at the high end, while Bomet, Trans Nzoia, and Kitui sit at the low end. Most counties fall between 30% and 40%, but a sizeable group is above 50%, so the 39.2% national figure is not experienced evenly across Kenya.
County share of adults reporting food vulnerability in the past month
But geography is only part of the story, which is representing underlying household characteristics. A county map can show where food vulnerability is high; it does not show the household profiles behind it. For that, I fitted a simple decision tree to group adults into recognizable risk profiles.
The tree shows that food vulnerability is not only about where someone lives, but also about how they survive economically. The first split is livelihood: adults relying on casual or seasonal work are immediately separated from everyone else. The tree then picks up combinations of credit exposure, partnership status, education, subsistence farming, and own-business activity. I read these combinations as risk profiles, not as causes by themselves.
To map the tree, I treated its highest-risk terminal leaves as risk profiles. A profile was counted as high-risk if its food vulnerability rate was at least 25% higher than the national average and it represented at least 2% of the weighted adult population. Each adult was then assigned back to their tree leaf, and each county was summarised by the share of adults who matched one of these high-risk profiles.
Risk profiles are terminal leaves from the decision tree with lift >= 1.25 and weighted population share >= 2%.
This map does not simply repeat the food vulnerability map. It shows where the risk profiles identified by the tree are more common. In other words, it moves from “where are people food vulnerable?” to “where are the conditions linked to higher food vulnerability concentrated?”