In many regions of the world that are heavily depend on rainfed agriculture, the rural-to-urban wage gap peaks in the lean season. As a lean season strategy, seasonal rural-to-urban labor migration is deployed by some households as a coping strategy. We experimentally test constraints to seasonal migration by randomizing 775 eligible, poor households in rural West-Timor, Indonesia, to either an unconditional cash transfer (UCT) or a cash transfer conditional on migration (CCT). The CCT group received information about seasonal migration possibilities in the form of a leaflet. To impose the conditionality in the CCT group, the second half of the transfer for the CCT was paid out at the destination. The CCT group was further split into three groups: (1) a `High’ (CCT High) treatment of IDR. 150,000 upon accepting the offer, and IDR 150,000 at the destination; a (2) a `Low (CCT Low) treatment of IDR 75,000 at the origin and IDR 75,000 at the destination; and (3) a `Low + surprise’ (CCT Low+) treatment of IDR 75,000 at the origin and IDR 225,000 at the destination. At the offer stage, the CCT Low and CCT Low+ are told that their second disbursement is IDR 75,000, so the CCT Low+ are `surprised’ with the larger transfer at the destination. The UCT treatment arm receives IDR 150,000 in one disbursement at the origin.
The main outcome is household income at the height of the migration season. The CCT offer is expected to be accepted by a lower share of households than the UCT, but to increase migration compared to the UCT. Hence the intent-to-treat effects of CCT assignment provides evidence not only on the benefits of migration and the need for information and nudges to encourage it, but also on the cost-effectiveness of the CCT and UCT treatments, given their different take-up rates. Going by migration season household income, we find the CCT treatments to compare favorably to the UCT in terms of the ratio of benefits to costs. The comparison between CCT Low and CCT High provides evidence on whether size of the CCT Low transfer is large enough to relax the migration constraint, whereas the comparison between CCT High and CCT Low+ provides the best estimate of selection into migration. Specifically, the CCT High may induce additional households to send migrants with lower marginal gains to migration (lower comparative advantage in the migration sector as compared to the CCT Low+), which would indicate distortion. We indeed find that the CCT Low and CCT Low+ arms induce slightly fewer people to migrate than the CCT High, but to have higher intent-to-treat impacts on income than CCT High. This evidence of selection implies that there is an optimum transfer size beyond which distortionary effects outweigh marginal benefits to the recipients.