Experimental Design
Participant Recruitment and Preference Elicitation
Study participants were recruited in village meetings taking place across 125 villages in the Kailali and Kanchanpur districts of Nepal. Those who were interested in migrating for a new job in other parts of Nepal or India within the next 12 months, if they have someone who has migrated and worked in other parts of Nepal or India to help, were eligible to enroll as participants. We targeted enrollment of 2,000 participants at this stage, but ended up with a final sample of only 1149 interested participants. This group constitutes the pool of eligible mentees among which we randomize for evaluation.
After enrolling mentees into the study, we conducted a pre-trial survey that included questions on past migration experience, current migration plans, and preferences over future migration. Preference questions included desired destination, sector of employment, timing of migration, and social proximity of prospective migration mentors. These preferences served as the basis for the assignment algorithm described below.
Prior to preference elicitation, we provided information on destination earnings by sector to two-thirds of mentees, selected at random. This information took the form of a set of sector-specific pamphlets with information on the average, 20th percentile, and 80th percentile of earnings among past migrants from the region to four major destinations for that sector, including the most lucrative destination, and at a nearby city in Nepal for comparison. This information provision was used as a stratification variable in our mentorship randomization.
Mentor Recruitment
People from the region are eligible to serve as migration mentors if they meet the following three criteria: a) they have migrated for work in the past two years, b) they are planning to migrate for work again within the next year, and c) they believe they can help someone else find a job during their next migration episode.
Mentors were initially recruited at village meetings alongside mentees. Those present were also encouraged to refer household members or other acquaintances with more migration experience, including people who may already be at their migration destination. Upon expressing interest, mentors are screened for eligibility and then asked about their migration plans---intended destination, sector, and timing of migration---for purposes of assignment to mentees.
Initial recruitment resulted in 492 interested mentors. Of these, 329 were assigned to mentees, and a random subset of 163 prospective mentors remained unassigned for monitoring and evaluation purposes. Enrollment of new mentors remains open throughout the study period, and even study participants who initially enrolled as mentees may act as mentors if they subsequently develop their own migration plans.
Randomization and Mentor Assignment
The 1149 prospective mentees were randomized into a treatment group of 578 mentees to be matched with mentors and a control group of 571 participants to remain unmatched. Randomization was stratified by ward, randomization date, and information provision described above. This treatment assignment will remain fixed for the duration of the study.
Within the treatment group, mentees are assigned mentors according to their stated migration preferences. The assignment followed a Gayle-Shapley stable matching algorithm in which match quality was defined by the goodness of fit between mentee preferences and mentor plans. Specifically, mentee preferences over mentors followed the migration preferences they articulated on the baseline survey. Mentor preferences over mentees were based on quality of fit. Matching took place within geographic pools so that mentees and their assigned mentors originated from nearby locations. The algorithm assigned 1-3 mentors to each mentee, and each mentor was assigned 1-7 prospective mentees.
Prior to the trial, we ran the matching algorithm a second time to construct a counterfactual mentor assignment for control mentees. To do so, we used the same mentor pool as in the initial matching and control mentees in place of treatment mentees, then ran the same algorithm as described above. Control mentees' counterfactual mentor assignments were recorded in data but not shared with either mentees or mentors.
Late Enrollment and Reassignment
Of the 578 mentees assigned to treatment, 50 joined after the initial mentor assignments had already been made in their locality. An additional 143 requested reassignment of at least one of their mentors within the first month because their initial mentors were unresponsive or they did not like the match. To accommodate these requests, we did the rematching by removing the already matched pairs and adding new enrollees to the pool. Rematching will take place again at the six-month mark for mentees whose preferences have changed.
Mentor enrollment will remain open throughout the study period as migration plans solidify and new potential mentors emerge. When a new mentor is enrolled, they will be assigned mentees according to quality of fit, with a preference for those who have fewer existing mentor matches.
Mentorship Treatment
The mentorship treatment consists of facilitating communication between mentees and their assigned mentors by sharing contact information between both parties and encouraging communication. To encourage mentorship, we offer mentors an incentive payment of 4,000 NPR (roughly 30 USD) if the mentee finds work in the mentor's destination and sector at some point during the trial and confirms the mentor's assistance was informative or useful in securing employment. Incentive payments will be made either through mobile money or delivery in cash to the mentor's home village, according to their preference. Payment will not be made if a mentee reports the mentor was at all dishonest or misleading.