Randomization Method
We start with a list of all the villages in Murshidabad (over 2000). After discarding the smallest villages (by Census population), we then order the remaining villages by population into four-village tiers, starting with the largest. Within each tier, the four villages are then randomly assigned to four treatment statuses, as follows:
Treated Individuals to be noted as (A) and the non-treated Individuals to be noted as (B)
Categories of the treatment is represented as: 1,2, 3 and 4
1. Fully Treated Villages– 330 villages: 1A. 1980 persons (6 per village), 1B. 0 persons
2. Partly Treated Villages – 330 villages: 2A. 990 persons (3 per village), 2B. 990 persons (3 per village)
3. Non-Treated Villages (3) – 330 villages: 3A. 0 persons, 3B. 1980 persons (6 per village)
4. Pure Control Villages (4) – 330 villages: 4A. 0 persons, 4B 1980 persons (6 per village)
Note: per-village counts are means; actual numbers will vary substantially with village size.
Depending on response rates to our invitation to the entry survey, we will either use all the villages in Murshidabad or a subsample of village tiers for our study. Based on very rough estimates (an average of 6 respondents per village), we expect to have about 1320 participating villages and 7920 participating individuals.
Participants who are not selected for treatment are free to sign up to the platform on their own (without assistance from Overseas staff) unless they live in the Pure Control villages. All persons living in those villages will be excluded from platform access using geo-coded location information at the sign-in process.
Approximately six months or one year later, we will again contact all participants (regardless of treatment status) and conduct the end-line telephone survey. This survey asks about their current earnings, location, and the quality of their migration experience.
For any outcome, Y (measured in our end-line survey or from Overseas’s activity records), we can then make the following comparisons:
Comparison 1: 1A – 4B. This gives the effect of a “saturated” Facebook-based platform rollout (offering personalized onboarding assistance to all campaign respondents) on target village residents, relative to living in a village that does not have platform access. It combines the effect of having platform access and receiving personalized onboarding assistance.
Comparison 2: 1A – 3B. This gives the effect of a “saturated” Facebook-based platform rollout (offering personalized onboarding assistance to all campaign respondents) on target village residents, relative to living in a village that has platform access, but no onboarding assistance. In other words, once Overseas becomes available in a region, this gives the effect of offering personalized onboarding assistance to all persons who attempt to sign on to the platform.
Comparison 3: [2A + 2B] – 3B. This gives the effect of a partial Facebook-based platform rollout (offering personalized onboarding assistance to 50% of campaign respondents) on target village residents, relative to living in a village that has platform access but no onboarding assistance. In other words, once Overseas becomes available in a region, this gives the effect of offering personalized onboarding assistance to half the people who attempt to sign on to the platform.
Comparison 4: 2A – 2B. This gives the effect of receiving the treatment in a village where half the people are treated, relative to the non-treated people in your village: It includes the direct effect of the treatment on the treated, minus the (within-village) spillover effect of the treated on the non-treated.
Comparison 5: 2B – 3B. For a non-treated person, this gives the effect of living in a village where others are treated (compared to living in a village in which no one is treated). In other words, this is the spillover effect of treatment on the non-treated. Combining this estimate with Comparison 4 should give us the direct effect of the treatment on the treated.
Comparison 6: 1A – 1B. For a treated person, this gives the effect of living in a village where all others are treated (compared to living in a village where only half of others are treated): In other words, this gives the spillover of treatment on the treated.
Notes:
1. In all the above cases, ‘treatment’ refers to an offer of onboarding assistance to Overseas. To estimate the effect of using Overseas, we will explore instrumenting Overseas use measures (both from our endline survey and from internal portal data) with our random treatment assignment.
2. Consistent with a small pilot survey about migration information flows, the preceding interpretations of our estimates assume there are no cross-village spillovers within our treatment area (Murshidabad). We will explore using random variation in the treatment status of nearby villages to measure the importance of such spillovers.