Refugee and Host Integration through the Safety Net Impact Evaluation

This research project will examine the impacts of the Refugee and Host Integration through the Safety Net program (RHISN). RHISN is implemented by the Government of Ethiopia through the Ministry of Urban and Infrastructure and the Refugees and Returnees Service, is one of few attempts by a LMIC to integrate camp-based refugees into the national social safety net. Refugees and hosts alike will participate in remunerated public works activities (e.g. building water points, roads, irrigation ditches, etc.) for up to 3 years providing a stable source of income. In addition, beneficiaries will receive graduation-style livelihood support consisting of training, and a 600 USD business grant in the second year. Beneficiaries will also obtain a work permit to legally participate in the local host labor market. As such, RHISN represents a unique opportunity to study the consequences of integrating refugees into a graduation program through experimental variation at scale. Its large-scale nature will allow us to study not only direct impacts on recipients, but also impacts on non-recipients, on local economic activity, on social cohesion at the neighborhood and work-team level, and the channels that underlie them.

Together with the Ethiopian government, we have set up a large-scale Randomized Control Trial (RCT) among the four main refugee hosting areas in Ethiopia with over 20,000 beneficiaries to evaluate the scaled-up implementation of RHISN. Baseline social integration varies widely across these areas, ranging from places where hosts and refugees share a common language, religion and tribal affiliation to others where they do not share any, and where there is sporadic violent conflict. Experimental work in this space is limited because conducting RCTs in humanitarian settings is complicated by the high vulnerability and mobility of forcibly displaced populations. Even rarer are studies using randomization at-scale which is crucial to understanding both the implementation challenges of large programs, as well as the general equilibrium (or economy-wide) forces shaping economic and social integration dynamics, and for generalizing findings to other contexts. Our proposed project has both elements.

2023-10-01

2028-10-01

Primary research questions
1- What are the impacts RHISN on the wellbeing of the program’s beneficiaries?
Outcomes: household income, consumption, assets, labor outcomes, and mental health.

2- What are the impacts of team diversity and inter-group contact on productivity and social cohesion?
Outcomes: social cohesion between refugees and hosts, network formation, team productivity.

3- What are the spillover effects of the program on local economic aggregates and social cohesion?
Outcomes: prices, wages, product diversity, local economic activity, GDP; neighborhood-level social cohesion, out-group attitudes, economic integration.

Secondary research questions:
4- How do the impacts of diversity on team dynamics, task allocation, communication, and productivity vary by the share of outgroup members in the work team?
5- Does additional after-work socializing (teas, dinners, cultural festivals) amplify the impacts of outgroup contact?
6- How do social norms, attitudes towards the outgroup, and social cohesion propagate through existing social and economic networks?
7- How do public good improvements impacts social cohesion and well-being?

We will study the above-mentioned questions through a large-scale clustered RCT conducted among four refugee camps and nearby host populations across all major refugee hosting regions of Ethiopia with a total of 22,500 beneficiaries. All refugees who express interest are eligible. Among host populations, eligibility is determined through a baseline proxy-means test (PMT). Due to ethical considerations on the vulnerability of applicants, 37% of the beneficiary quota is assigned automatically to the poorest households according to the PMT. 63% of the quota is assigned randomly among the remaining vulnerable households.

Not available

Computer randomization

We will randomize the following aspects of the program:
T1 – Participation in RHISN: Eligible host community and refugee households are randomly selected into 3-year participation in remunerated public works, a legal work permit, livelihoods training and USD 600 business grants.
T2 – Contact through mixed refugee/host work teams: RHISN beneficiaries will be randomly assigned to public work groups. Work groups consist of ~30 individuals who work together on public projects for the first year and attend livelihood training together in the second year. There are 3 types of groups: ‘Refugee/host only groups’ are constituted only of the in-group. ‘Equal groups’ have 15 refugee and 15 host members; and ‘refugee minority groups’ are approximately 20 hosts and 10 refugees.
T3 – Social cohesion activities: A random subset of public work groups will be invited (by a partner NGO) to attend monthly teas and dinners, and an annual cultural festival promoting after-work socializing.
T4 – Local program saturation: We randomly vary the saturation of a) the program overall, and b) the extent of outgroup exposure within small geographic areas. Host communities are administratively organized into ketenas (neighborhoods) of approx. 200 households each. On the refugee side, an equivalent administrative unit is the block group (250 households each). There are 83 ketenas and 97 block groups, or a total of 180 clusters in our study area. We cross randomize at the cluster level in two dimensions.
A: We randomize the intensity of program saturation: In high-saturation clusters, ⅔ of eligible households will be treated, in low saturation clusters ⅓. This induces random variation in program intensity across neighborhoods/clusters and allows us to estimate economic spillover effects of the program on non-recipients, and on the aggregate economy.
B: We cross-randomize the intensity of outgroup contact: In high-exposure neighborhoods/clusters, 2/3 of RHISN beneficiaries are assigned to mixed groups, while in low-exposure clusters only 1/3 are mixed. We cross-randomize this with program saturation to be able to separately estimate social spillovers of outgroup contact.

Yes

180

12,000

Sample treatment: 8,000
Sample control: 4,000
We have multiple levels of randomization across groups to answer the pre-specified research questions.

For power calculations, we require estimates of ICC for the questions that are going to be randomized at the cluster or group level. We obtain these from two different sources: i) The Socio-economic Survey of Refugees in Ethiopia (SESRE), which completed its first round between January and February 2023 and targeted all our impact evaluation cites, with a total sample of approximately 2,000 refugees and hosts. ii) The administrative data collected as part of the RHISN program targeting (i.e. the proxy-means test and screening questions for determining program eligibility). We estimate ICC for key economic outcomes and social cohesion, as those will be our main primary outcomes. ICC estimates vary between 0.00 and 0.13. We thus use the highest and most conservative value (i.e., 0.13) for all cluster- and group-level randomizations for our power calculations. We are adequately powered for all main research questions. We are able to detect changes in outcomes of between 0.05 and 0.17 standard deviations for all questions and subgroups. For a program of this scale, and at this intensity, we expect effects to be large. In terms of economic outcomes, graduation programs similar to RHISN (or even just cash transfers of an amount similar to that entailed by RHISN) have been shown to have effects larger than our MDEs over the time horizons studied. One point worth noting relates to spillovers: since randomization of beneficiaries takes place within relatively small geographic units we expect economic spillovers to be relatively large. We have two strategies for addressing this: i) Our saturation design allows us to explicitly estimate such spillovers. With treatment intensities varying between 15 – 60\% of all households treated in an area, and an assumed multiplier of 2.5 (Egger et al. 2022), effects on the control group should amount to approximately 22\% - 90\% of the treatment group, and we should therefore be powered to detect such spillovers and estimate them relatively precisely. ii) We collect high-frequency short-run data on expenditures and attitudes at a time when spillovers are less likely to have showed up yet, mediating the concerns around control-group catch-up, and allowing us both to study dynamics of the initial effects, as well as the speed at which spillovers manifest