Refugees Vulnerability Study, Kakuma, Kenya


Kakuma Refugee Camp is one of the longest lasting humanitarian settlements in sub-Saharan Africa and one of the largest refugee camps in the world. In response to recent reductions in funding for the Kenyan refugee operation, increased global competition for funds, and a common belief that not all refugees in such protracted situations have the same humanitarian assistance needs, the World Food Programme (WFP) and United Nations High Commissioner for Refugees (UNHCR) commissioned Kimetrica to undertake a vulnerability study among refugees in Kakuma Camp. The main aims were to fill knowledge gaps regarding refugee livelihoods and the level and differences of vulnerability in refugee households, as well as to explore the feasibility of delivering more differentiated assistance and to identify the mechanisms that would need to be put in place to do so. Specifically, the study was meant to determine whether a permanent targeting mechanism could be put in place or whether obvious ways of prioritising assistance exist when resources are tight.

To complete the study, Kimetrica carried out three phases of fieldwork: an initial scoping study, a 2,000 household survey and a follow-up mission to explore the feasibility of various targeting mechanisms. The household survey covered 13,378 refugees distributed across each of the 126 administrative blocks in the camp. Designed to be statistically representative at the subcamp level, with 500 households sampled in each of the four sub-camps (Kakuma 1, 2, 3 and 4), it is one of the most comprehensive studies on the livelihoods, wealth and vulnerability of the Kakuma refugee population.


Many households either farmed or reared livestock before arriving in the camp. Unlike other refugee camps where agriculture and livestock rearing can be major livelihoods, the restrictions on livestock ownership and the scarcity of water limit these activities in Kakuma. Legal constraints on other livelihood opportunities, a small customer base (mostly refugees themselves) and the remoteness of the camp result in a challenging environment for business operations. The only major external cash flows into the camp are via remittances and incentive work. Although 10 percent of households reported receiving cash income from employment, 8.2 percent from a business and 6.0 percent from remittances, the value of earnings are low. Very few households (only 2.9 percent) earn more than the Kenyan minimum wage of 10,000 Ksh per month.


The socio-economic vulnerability of households was measured as the non-gifted household cash equivalent consumption expenditure on food and non-food items (NFIs) per capita per day, following the global Living Standard and Measurement Surveys (LSMS). This measure reflects effective purchasing power and corresponds with the units of estimation of the minimum consumption basket. The median cash equivalent consumption expenditure was 7.4 Ksh per capita per day. This was below 4 Ksh for households in Kakuma 4, South Sudanese and new arrivals, and 16 Ksh for Somalis and 18 Ksh for Ethiopian refugees. Although not all households in the camp have the same level of vulnerability, only a small proportion (4.2 percent) would be able to sustain themselves without any assistance (valued at 77 Ksh/capita/day for a healthy food basket and essential NFIs). In addition to completely eliminating assistance, four scenarios were explored to identify whether households would be able to support themselves with some level of reduced assistance. Given current income and expenditure patterns, only 5.7 percent could cover all their food needs, 9.1 percent half food and all NFIs, 15 percent half food and 31 percent all NFI needs from their own resources.


From focus group discussions with community leaders, there was a general sense that “refugees are all the same,” and any form of reduced assistance to certain groups (herein referred to as targeting out) would be generally opposed by the majority of the camp population.
Rather, community leaders suggested that the general population should continue to receive the same amount of assistance, and vulnerable groups should receive more (hereafter referred to as targeting in). Four targeting mechanisms — community-based targeting, categorical targeting, proxy means testing and self-targeting — were explored for both targeting in and targeting out options and compared to the status quo of blanket assistance (inclusion error of 4.2 percent (proportion of households selected for assistance that do not require it) and exclusion error of 0 percent (proportion of households that need assistance that are excluded)).

Community-based targeting relies on the knowledge of community leaders to identify households that do not need assistance. However, in Kakuma Camp, leaders’ knowledge of the households in their blocks is far below the level needed for this targeting strategy to work properly. Even for households they knew, community leaders were not able to accurately distinguish between households that do and do not need assistance. As such, communitybased targeting would not be an effective targeting technique in Kakuma Camp.

The simplest and most common method of targeting, categorical targeting relies on using a household characteristic to identify a group for targeting in or out. Of the categories tested, only targeting out of households with a business from all or part of the assistance resulted in errors considered acceptable by WFP standards (inclusion error of 2.9 percent and exclusion error of 6.9 percent for full reduction in assistance).

Proxy means testing can be seen as an extension to categorical targeting and involves using regression analysis on several household characteristics to identify vulnerables and nonvulnerables.
After testing several models on multiple datasets and against varying thresholds, the Extremely Randomised Trees model, a machine learning model, produced the best results (inclusion error of 1.6 percent and exclusion error of 4.3 percent using a set of 12 household characteristics against a threshold of a full reduction in assistance). Although both mechanisms offer errors considered acceptable by WFP, the continuation of blanket coverage is preferable, as it complies with the “do no harm” principle, tends to minimise overall errors and has no costs of implementation.

Together with continuation of blanket coverage, allowing refugees the option to self-target out of assistance by offering incentives to business owners or increasing incentive pay could be explored further.


In addition to addressing the primary objectives of the study, this research revealed several unexpected findings. First, unless a household updates its status on the UNHCR database voluntarily, the ration card received upon arrival to the camp remains the unit for which all assistance is delivered and all statistics are reported. However, the reality is that after arrival, many ration cards join together to form larger family units. This has implications for the UNHCR demographic statistics, and a household census should be conducted to update the database.
Second, several groups that are typically perceived as vulnerable are not necessarily as vulnerable as previously expected, or vice versa. For instance, while households with a disabled member or head are traditionally considered to be more vulnerable than the average population, their median cash equivalent consumption expenditure is actually higher than the sample median. Household size 1s are also typically perceived as a vulnerable group. However, the median cash equivalent consumption expenditure for household size 1 is more than twice the average. Similarly, three times more household size 1s would be able to support themselves in the absence of food and NFI assistance than the average household. A more detailed analysis of household size suggests that vulnerability actually increases as households get larger, such that households with more than 10 members are most vulnerable. Finally, there is a common conception that there is a negative linear relationship between vulnerability and year of arrival; that is, households that have arrived recently are most vulnerable, while households that have been resident in the camp longest should be able to support themselves. However, the results suggest that vulnerability may follow more of a U shaped curve; although recent arrivals demonstrate the highest levels of vulnerability, households that have been resident in the camp for a medium term (between 6—20 years) appear to have lower levels of vulnerability than the most long-term residents.


The findings from this study support the continuation of blanket coverage of assistance in Kakuma Refugee Camp. Furthermore, unless there is a major change in policy that would give refugees greater freedom to own livestock or to move freely to establish businesses and livelihoods outside of the camp, these refugees will continue to need high levels of assistance.
Given the unique context of the camp (restrictions mentioned above, the harsh climate of Turkana, the remote location of the camp and the protracted nature of the settlement), this recommendation does not necessarily preclude the possibility that targeting may be an effective strategy in other refugee settings.

Although both the household survey and the community-based targeting exercise included questions on remittances, still little is known about this sensitive, but important, income source that contributes roughly one-third of the cash income to the camp’s economy. We suggest an updated and focused review of the sizes, sources, uses and mechanisms for transfer of remittances in the context of refugee camps as an expansion of, and update to, Professor Oka’s previous work in this area.