Experimental Design
We plan to operate over the summer seasons 2025 and 2026, when farmers mostly grow maize and suffer increasingly from climate change and its effects on agriculture.
Information intervention: Extension agents in each village provide the list of names of all eligible farmers who live in pre-specified clusters within the village. Our team then visits the farmers and confirms eligibility and willingness to participate in a study on agriculture and youth employment. We use a clustered randomization design.
Job matching services: all individuals are offered the job matching service through a BDM mechanism.
Important aspects to note:
1) Impacts of job matching services: We will look at the effects of the job matching services on the outcome variables in domains B and C. To address endogeneity concerns, we will instrument take-up of the matching services by the random price assigned to each individual. For this analysis standard errors will be clustered at the individual level (the level of randomization).
2) Selection: In this project, we are particularly interested in studying how better information about climate change influences selection into the non-agricultural sector. We plan to conduct two types of analyses. First, among individuals who worked at least one hour outside agriculture, we will regress key individual characteristics on an indicator for the information treatment. Second, we will estimate the heterogeneous effects of job matching services depending on whether individuals received the information treatment. Specifically, we will interact the information treatment and job matching indicators, instrumenting them with the randomly assigned price and its interaction with the information treatment to address endogeneity concerns. For this analysis standard errors will be clustered at the cluster level (the level of randomization).
3) Heterogeneity: All heterogeneity dimensions are measured at baseline and coded as dummy variables. The most important heterogeneity dimensions we expect to study include:
- Beliefs about the feasibility of agriculture
- Beliefs about need for adaptation
4) Spillovers: As with many information interventions, the main threat is related to spillover effects. To determine the unit of randomization, we incorporated both conceptual and practical considerations. Conceptually, we decided to use a clustered RCT (instead of an individual RCT) to limit the potential for spillovers. We also aimed to include a large number of clusters to have enough statistical power to detect modest effects. Practically, because of budget constraints, and because villages in Egypt tend to be very large (the average rural villages have a population of 10,000 in Assiut), we favored clustering at the block (agglomeration) level. We are able to form 4 blocks on average per village. While clustering the intervention at the block level limits the scope for spillovers within blocks it does not tackle concerns over spillovers across blocks. We try to limit them by leaving buffer areas of at least of 180 meters between blocks. Additionally, to test whether this approach is effective, we randomize the share of treated blocks at the larger level (villages in our case) to measure these spillovers. Specifically, in half of the villages only one third of the blocks are randomized into treatment (low-saturation villages), while in the remaining half of the villages two thirds of the blocks are randomized into treatment (high-saturation villages). Comparing control individuals in low-saturation villages to control individuals in high-saturation villages gives an estimate of the spillovers across blocks.