Edutainment to Mitigate Inter-Group Violence: Experimental Evidence from Pastoral Kenya

The intervention consists of the public screening, through a mobile cinema set-up, of a 14-minute animation video followed by a 1 hour structured and standardized discussion. The video depicts events leading up to a violent pastoral raid, its consequences for all members of the community including women, and alternative solutions to raiding for sustainable income provision: skills-based jobs during rainy seasons when opportunity costs of conflict are lowest, and grazing- and herding agreements between communities to preserve pasture and reduce the profitability of raiding. The post-video discussion facilitated by the International Livestock Research Institute and peace-building NGO International Alert uses a fixed set of questions that are addressed to the community, which the community then discusses among themselves.

2025-06-30

2025-07-30

1. Likelihood of livestock raids
2. Animal losses, human injury and loss of human life due to raids

Likelihood of livestock raids: both the perceived likelihood and observed occurrence.
Animal losses will be measured in TLU, human injury and loss of human life due to raids both at the intensive and extensive margins.

1. Likelihood of livestock theft
2. Animal losses (in TLU), human injury and loss of human life due to theft
3. Share of young men engaging in alternative livelihood strategies out of total young men
4. Number of grazing- and restitution-agreements
5. First and second order beliefs about the costs of raiding and theft
6. First and second order beliefs about alternative sources of income to herding
7. First and second order beliefs towards grazing- and restitution agreements

Theft measurements analogous to the raid measurements above.
First order beliefs are answers to attitude questions measured on a five-point Likert scale. We estimate the answers to each of the these questions as separate outcome variables. Per overarching attitude-theme per paragraph below, we also construct a standardized index, using the control group mean and standard deviation at endline after aggregating all answers and dividing them by the number of answered questions per theme. A "don't know" answer is treated as a missing observation. If needed, separate items are re-coded prior to indexing so that higher values uniformly reflect attitudes, with a higher score meaning an "anti-conflict" attitude, and a lower score referring to a "pro-conflict" attitude.
Second order beliefs are answers to a rephrasing of the belief statements in the form of "out of 10 [elders, young men, women] in your community, how many would agree that ...." .

We conduct a cluster-randomized control trial (RCT) in 246 sublocations in Isiolo, Marsabit, Samburu, Turkana, Tana River, and Wajir counties in Kenya. It has two arms: Treatment (T1), where the video is screened through a public mobile cinema screening and discussed in a structured and standardized post-video discussion; and Control (T0), where no intervention is received.

Not available

Randomization by computer.

Given that there is significant variation in conflict across sublocations, we created clusters of subcounties for comparison based on geographical closeness and similar conflict experiences in prior panel surveys. In this first step, through computer randomization, these subcounty clusters were randomly assigned to the high or low intensity treatment. In the high intensity treatment, 70% of sublocations were treated and 30% of sublocations functioned as control. The low intensity treatment has 30% of sublocations treated and 70% as control.

After randomizing the subcounty clusters into high and low treatment intensities, we allocated the sublocations within the subcounty clusters into treatment and control through computer randomization. This is stratified by an insurance encouragement intervention, which is an exogenous determinant of insurance. Hence, for high intensity subcounty clusters, 70% of the standard sublocations and 70% of the sublocations with the insurance encouragement intervention are randomized into treatment for edutainment, and the rest to control. For low intensity treatment subcounty clusters, 30% of sublocation without the insurance encouragement and 30% of sublocations with insurance encouragement are randomized into treatment, and the rest to control. The exception here are the 20 added sublocations, as these were not a part of the insurance encouragement study. Hence, these were not stratified but followed the other randomization steps. We have checked whether household and sublocation characteristics were balanced by treatment assignment based on the following characteristics: At the household level, whether the household as experienced conflict in the last year (binary), household head has attended school (binary), herding as a primary livelihood strategy (binary), petty trading as a primary livelihood strategy (binary), casual labor as a primary livelihood strategy (binary), wage employment as a primary livelihood strategy (binary), owning camels (binary), owning cattle (binary), owning goats (binary), owning sheep (binary), number of camels, cattle, goats and sheep. At the sublocation level: conflict experienced in the last year (binary), raids experienced in the last year (binary), grazing agreements (binary), restitution agreements (binary), women groups (binary), number of households with as a primary, secondary and third livelihood strategy herding, livestock production, shop or business, petty trading and farming.

Yes

9 clusters of sublocations

246 sublocations

116 sublocations control, 130 sublocations treatment