Temporary Work Visas as Us-Haiti Development Cooperation: A Preliminary Impact Evaluation

We report a small-sample, preliminary evaluation of the economic impact of temporary overseas work by Haitian agricultural workers. This work occurs in the United States in the context of a pilot program designed as a form of post-disaster development assistance to Haiti. We find that the effects of matching new seasonal agricultural jobs in the US with Haitian workers differs markedly from the effects of more traditional forms of assistance to Haiti, in three ways: The economic benefits are shared roughly equally between Haiti and the United States; these benefits are very large, including raising the value of Haitian workers' labor by a multiple of fifteen; and the portion of the benefits accruing to Haiti is uncommonly well targeted for the direct benefit of poor Haitian households. We discuss implementation challenges faced by the program and the potential for policies of this kind to complement more traditional forms of development and humanitarian assistance.


Introduction
The world's poorest workers can experience life-changing increases in income by working abroad, even for a short time. This is because working in a rich country typically increases the economic productivity of low-income workers by 300-1,000 percent (Clemens et al. 2016). By comparison, even highly successful aid projects to ght poverty tend to raise incomes of the poor by 10-40 percent (e.g. Blattman et al. 2014;Banerjee et al. 2015b) while a large number of such projects have no detectable e ect on income (e.g. Banerjee et al. 2015a).
Income gains of this magnitude suggest important potential for a new form of international humanitarian and development policy: fostering temporary labor mobility (Luthria et al. 2006;Pritchett 2006). At this writing, New Zealand is the only country to experiment with building a seasonal labor mobility program explicitly as a form of development cooperation (Ramasamy et al. 2008;Gibson and McKenzie 2014b). The evaluation of that program, one of the only rigorous impact evaluations of temporary migration ows, found that its e ects on the low-income South Paci c islands home to the program participants ranked "among the most e ective development policies evaluated to date" McKenzie 2010, 2014a), as well as bene ting New Zealand's economy (Winters 2016). But in general the economic e ects of seasonal and temporary migration have been little studied (Dustmann and Görlach 2016). a In this paper we consider a pilot program to ease informational and bureaucratic barriers to temporary labor mobility between Haiti and the United States. We present new and preliminary evidence from a small-sample survey to evaluate the economic impact of temporary overseas work by Haitians under this program. We focus on three ways that temporary migration di ers from more traditional development assistance: the mutual economic bene t to both countries, the size of the income gains, and the direct bene t to poor families. We establish the counterfactual by surveying households that were willing and prepared to work in the United States but barred from doing so by administrative events unrelated to their individual characteristics. Limitations of the results include the small sample size and possible questions of external validity.
We nd that one worker-month of seasonal agricultural work by a male Haitian in the United States raises his current wage by approximately 1,400 percent, adds roughly US$3,000 to the 1 economy of Haiti, and-if it represents a marginal addition to the stock of US seasonal agricultural workers-adds more than US$4,000 to the economy of the United States. Nearly all of the resources transferred to Haiti directly bene t poor Haitian families, with the average e ect of doubling annual household income with 2-3 months of overseas work by one household member. This assesses the impact of migration to ll new agricultural jobs in the United States (the 'Treatment-on-Treated' e ect), not the impact of the project on all those it would wish to reach (the 'Intent-to-Treat' e ect). These results suggest unexplored potential in fostering temporary labor mobility as a policy to assist the poor-a policy alternative with much larger and more direct bene ts at the margin than traditional aid, with direct economic bene ts to both the rich and poor country involved.

A small pilot project in Haiti
In January 2010, a catastrophic earthquake in Haiti killed approximately 200,000-250,000 people, with economic costs exceeding US$8 billion (Cavallo et al. 2010). In response, the Center for Global Development commissioned a study of how US policies on Haitian labor migration could contribute to the relief and reconstruction e ort (Murray and Williamson 2011). This work highlighted the facts that regulations of the time made nearly all Haitians ineligible for employmentbased visas to the United States, while there have long been important ows of unlawful migration from Haiti. It recommended that policymakers consider allowing small numbers of Haitians to apply for seasonal work visas, both to contribute to the post-earthquake recovery and to o er an alternative, legal channel to unlawful migration.
This led a bipartisan group of eight US members of Congress and a coalition of humanitarian non-governmental organizations to petition the Administration to exercise its discretion under law to make Haitians eligible to apply for seasonal employment visas. b No change to statute was required; several months later, Haiti was granted eligibility (Napolitano 2012). This seasonal employment ('H-2') visa allows workers whose jobs do not require a university degree to enter the United States for up to 10 months each year to work at a single employer. c To sponsor a worker, an employer must receive certi cation from the state and federal Department of Labor that recruitment for quali ed US workers to ll each position was unsuccessful. The project 2 considered here used exclusively the agricultural work visa ('H-2A').
The impact of granting Haiti access to the visa was di cult to assess. Visa recipients and their families could not be interviewed, because the identities of US visa recipients are not made public. Moreover, the visa was little-used: between October 2012 and September 2013, 15 Haitians received a seasonal work visa. d US employers who had already been sponsoring the visa for workers of other nationalities were typically unfamiliar with Haiti and invested in the recruitment contacts and language skills needed to hire workers from Mexico, Jamaica, and Guatemala.
Haitian workers and o cials were typically unfamiliar with the program and its associated requirements of bureaucratic procedure and documentation.
Many policy advocates expressed skepticism in personal communications about the potential for the visa to bene t Haiti or the United States. The US seasonal work visa in general has been heavily criticized as exposing workers to abusive recruitment practices (e.g. Owens et al. 2014) and requiring burdensome and expensive procedures for employers (e.g. U.S. Chamber of Commerce 2015). A common and fundamental objection cited the general goal of economic development policy as that of creating job opportunities within low-income countries-rendering overseas work at best a substitute for more traditional development policy.  These di culties in project implementation greatly restricted the size of the sample available for an impact evaluation of overseas temporary work. But they also created an opportunity to carefully assess that impact. Many Haitian workers eligible and willing to work overseas via this program, and selected by a US employer, were not allowed to travel and take up employment.
Almost all of these (44 out of 48) were barred from traveling for reasons outside of their control and unrelated to any personal characteristic of theirs. f Workers who were allowed to travel will therefore tend to be much more similar to eligible workers who were not allowed to travel than to typical Haitians, in all aspects other than the fact that they did or did not travel. By comparing outcomes for those who traveled to eligible, willing participants who did not travel, we can therefore assess more accurately the e ects of overseas work than by comparing participants who worked abroad to other, non-participant Haitians-even if those other Haitians were oth-erwise similar on easily-observable characteristics. But the small sample size renders any result necessarily preliminary.

Survey
A small evaluation survey was carried out in May 2016 to assess the program's e ects. The study was designed to measure comparative living conditions for individuals who worked in the US in 2015 and those who had not yet had a chance to travel. 32 households, each with a di erent program candidate, were originally contacted for survey participation, based on geographic proximity and availability. Six households were unable to participate at the last moment (including two households with a member that traveled to the U.S.); we successfully contacted an additional four households without a member that traveled, bringing the sample size to the nal 30. Twelve workers in the nal sample had traveled to work in the US in 2015. Five of these workers spent 13 weeks at an ornamental plant nursery in Alabama, earning US$10.59/hour; the other seven spent four weeks at apple farms in Oregon and Washington state, earning US$12.69/hour. All 12 returned to Haiti by the end of 2015. At the time of our survey in May 2016, eight of these had returned to the US for another season. Five workers surveyed had been originally selected by a US employer, but were unable to travel due to the visa and work certi cation denials mentioned above. The remaining 13 individuals in our sample had been selected for the candidate shortlist, but had not yet traveled.
Interviews were conducted at the central o ces of each agricultural cooperative, with PTP sta providing translation to and from Kreyòl. When the candidate himself was available, we spoke with him directly. If he had returned to the US on another work contract, the nearest relative was contacted. The survey questions were designed to mirror the 2012 Post-Earthquake Living Conditions Survey, or ECVMAS (Enquête sur les Conditions de Vie des Ménages Après le Séisme, described in World Bank 2014). g The major survey sections focused on basic household demographics, education and employment information, dwelling and asset details, estimated expenditures, and international transfers. An additional module was added to assess the impact of overseas work via the H-2A program (in the case of those who traveled) or expectations for future overseas work (in the case of those who had not yet traveled). The basic characteristics of the survey sample are summarized in Table 1. The average household is a family of six, with two school-age children, living on a one-hectare farm that is their primary livelihood. They are quite poor, living in basic structures generally without any services such as sanitation, and spend more than half their income on food.

Selection into the sample
Haitian farmers became eligible for the survey sample because 1) they were members of two established agricultural cooperatives that were known to eld workers at the Port-au-Prince o ce of the International Organization for Migration, 2) they were selected by the leadership of those cooperatives to be among the initial participants in the program, and 3) their households were available to be interviewed May 23-27, 2016. This triple selection creates the potential for surveyed workers to be substantially di erent from typical Haitian farmers, which could a ect the external validity of the evaluation results. We have less concern about selection a ecting the internal validity of the impact evaluation, given that the survey sample includes 12 of the 14 households that had a member travel in 2015, and the fact that many of the eligible households 6 whose members did not travel acquired this status due to a force majeure outside their control.

Results
Here we compare di erences in economic outcomes in the survey sample according to whether or not a member of the household was allowed to work in the US in 2015. The US decisions that disallowed some workers from entering the United States were largely independent of the workers' individual characteristics. We therefore interpret economic di erences between the allowed and disallowed groups as estimates of the e ect of overseas work.

Effects of overseas work on household income
Current income. Figure 3 suggests that working in the US had very large e ects on household income. Figure 3a considers household income during any given month that workers were present in the U.S. For the households with a worker in the U.S., these estimates conservatively assume that the household received income from Haitian sources pro-rated downward according to the fraction of the year that the US worker was absent. This is because farms of all workers who traveled continued to operate while they were gone, though often at levels of produc- project income in the 'traveled' sample and the 'did not travel' sample fails to reject equality with p = 0.081. But we do not necessarily expect equality of non-project income between the households of those who traveled and those who did not travel, because it is possible that income from temporary work overseas could have a ected Haiti-based sources of income. Several of the survey respondents with a household member who traveled reported spending some of the proceeds from overseas work on things that could have a ected Haiti-based household income by the time of the survey. These include agriculture inputs such as seeds and fertilizer, and inventory for home-based shops.
These estimates indicate that working temporarily in the United States raised the value of projecteligible Haitian farmers' labor while abroad by approximately 1,400 percent (a factor of 15), relative to its current value in Haiti. Even a short span of several weeks' work in the United States was su cient to more than double annual household income received by other household members. These e ects are so large that they are statistically signi cant even in this small sample. Respondents were asked about current consumption expenditures in the month prior to the interview. To capture any large expenditures that may have occurred immediately after some workers' return to Haiti-typically 5-6 months before the interview-respondents were separately asked to enumerate any major purchases they had made over the previous eight months.  We observe no e ect on overseas work on school attendance by children living in the workers' households; nearly all children age 5-18 in the households surveyed are attending school. The average fraction of children age 5-18 who are attending school is 1.0 in the households whose member worked in the US through this program, and 0.987 in the group whose member did not work in the U.S. These are statistically indistinguishable in a two-sample test for equality of proportions (p = 0.74).

13
Because prior research has shown that the objective economic impacts of migration on migrants can vary greatly from the impact on subjective well-being (Stillman et al. 2015), we asked all respondents questions related to subjective perceptions of temporary work in the U.S. Every individual surveyed from a household whose member had traveled-either the worker who traveled or a close family members-expressed a desire to repeat US employment. Every household whose member had not yet traveled expressed a desire for the member to travel in the near future. Those who had already traveled to the US universally rated their satisfaction with the program as a 10 out of 10. Beyond the sizable economic impacts of the program described above, workers who had traveled also noted improved agricultural skills and a sense of greater freedom and empowerment. "Now I can do anything I want, " a farmer's wife from the Nord department explained. The head of the agricultural cooperative Sohaderk expressed his satisfaction with the program, noting that those who had worked overseas passed on their new knowledge and skills to their neighbors upon return.
Households with workers who traveled reported substituting for the absent worker's labor either with other family members or with assistance from members of their cooperative association.
None reported hiring additional help for monetary compensation due to the absence. And none of these mentioned a substantial decline in home-plot production.
Eligible and willing workers who were not allowed to work in the US did not have unrealistically high impressions of their earning potential there. Each worker who had been unable to travel was asked what they believed they could have earned in the US Almost all reported a gure that was approximately correct (US$80-100 per day), though three respondents reported a gure of US$10 per day, a substantial underestimate of their true earning potential.
Beyond the household interviews in Haiti, the seven men who worked in Alabama were personally interviewed at the Alabama farm, each in isolation from the others and with a Kreyòl translator. Each was asked what concerns he had about the program and what he would change about it. None of the seven reported anything that he would change about the program, and the only concern-reported by several of the respondents-was their concern that the project might 14 not continue and they would be unable to work in the US again.
These perceptions di er markedly from descriptions of US work under the same visa as comparable to "slavery" (e.g. SPLC 2013). This suggests that well-known abuses of other (especially Mexican) seasonal workers on the same visa, such as the charging of illegal and exorbitant recruitment fees (e.g. Owens et al. 2014) is not an inherent trait of the visa or of seasonal work itself but of contextual factors, especially unregulated recruitment. This project and its e ects on participants suggest that recruitment can be properly designed to deliver very large bene ts to users of the visa.

Differences with traditional aid policy
These preliminary results from a small survey are su cient to suggest three important di erences between this project, designed to assist poor Haitians, and more traditional aid policy designed with the same goal. The available data suggest that temporary overseas work by Haitians directly bene ted the economies of both the United States and Haiti, whereas traditional aid would bene t the economy of Haiti at best; that overseas work resulted in much larger nancial bene ts per participant than traditional aid; and that overseas work targeted and bene ted poor households much more directly than traditional aid. Here we review the evidence for each of these assessments.

Economic effects in both Haiti and the United States
Labor mobility by these Haitian workers is likely to have e ects beyond the household level.
First, it is likely to have economic multiplier e ects in Haiti. The survey evidence suggests that, for every worker-month of overseas work, approximately US$1,700 will eventually be spent in Haiti. Very little of the actual or planned expenditure described by survey respondents could include imported goods, and essentially none of it will be invested outside Haiti. This means that these expenditures ripple through the local and national economy, raising GDP by more than one dollar for each dollar spent (Djajić 2014). The economics literature contains no estimate of the size of this multiplier e ect in Haiti, but the few available estimates for other developing 15 countries range from 1.9 for poor villages in Mexico (Taylor 2004, p. 164) to 2.8 for Morocco (Glytsos 2005). A rough, reasonable estimate would therefore be that each worker-month of Haitian seasonal agricultural work in the US contributes at least US$3,000 to Haiti's current GDP.
Second, when US farmers expand their seasonal workforce, they add value to the US economy.
In this way, labor mobility di ers sharply from a transfer of traditional foreign aid. The degree to which it adds value in the US depends on the degree to which the US farmers who hire the workers could achieve the same productivity by substituting other production inputs, such as Haitian wages in the U.S., between $3.65 (short run) and $1.11 (long run) are added to the US farms' productivity, which would then ripple through other sectors. Such multiplier e ects of added farm productivity through the rest of the regional US economy are well-studied, and are approximately 1.5 for Oregon and 2.0 for Alabama. o These regional multipliers imply, conservatively using the 1.11 long-run multiplier on farm productivity, that a worker-month of Haitian labor expands the US economy by more than $3,720 (Oregon)-$4,140 (Alabama). p The short-run e ects would be substantially larger. in a year, during less than seven months of work. This implies that a temporary work visa to

Magnitude of financial flows
New Zealand increases the current value of household labor in Tonga by a factor of more than 14, and in Vanuatu by a factor of more than eight. q

Targeting to beneficiaries
A third way that overseas work visas di er from traditional programs to assist the poor lies in their unusually direct targeting of bene ciaries. Easterly and Williamson (2011,   18 These evaluation results contain only preliminary evidence from a small sample in a tightly constrained pilot project. They do suggest unexplored potential in policy interventions to shape labor mobility as a form of development and antipoverty policy. The policy intervention described here acted to overcome informational and bureaucratic barriers to Haiti-US labor mobility-but did not require any changes to existing legislation in either country, and was able to act in a post-disaster setting. The results imply that this project di ered sharply from post-earthquake aid to Haiti in three ways. First, these estimates suggest substantial economic bene ts, in both countries, from seasonal The analysis above implies that such a potential future ow, if workers stayed only three months a year, would add on the order of US$100 million annually to the Haitian economy-and more than this to the US economy. Facilitating access to lawful channels of Haiti-US labor mobility is also likely to partially substitute for unlawful migration between the two countries (Hanson 2009). The principal constraint to such expansion appears to be demonstrating the e cacy of to experiment with unknown workers during critical planting and harvesting periods-and the bureaucratic burdens of using the seasonal work visa program, which make it broadly unpopular with employers. But the possibility of large and shared bene ts presented by temporary labor mobility suggest that innovation in this area deserves development policymakers' scarce resources.  e As is standard practice around the world, no reasons were given for the group-based denial of visas at the embassy in Port-au-Prince. The most common reasons for denial of a nonimmigrant visa include the o cers' assessment that home-country ties such as property ownership are insu cient to motivate return. In this program, applicants were vetted and trained by Sohaderk and Feccano for several months in a process designed to ensure return. As they vetted potential participants, association leaders were aware that continued participation in the program would be jeopardized if a substantial number of workers overstayed their visas. In the event, all of the workers who traveled returned as scheduled.

Notes
f That is, 44 of the workers were unable to travel because their potential employers did not receive certi cation to hire them, a bureaucratic process that does not involve consideration of the workers as individuals. The group that was denied visas at the US Embassy in Port-au-Prince, which as mentioned above largely overlapped with those whose 23 positions were not certi ed, were denied collectively-with no concerns noted by consular o cers about particular individuals in that group.
g The ECVMAS interviews were conducted between August and December 2012. In October 2012 one US dollar averaged 42.2 gourdes.
h In one household whose member was allowed to work in the U.S., the same worker while in Haiti did halftime hired construction work, and the survey respondent (his spouse) did not know how much he earned in that occupation. We impute the value of that income source using the average value of half-time construction wage earnings among ECVMAS households, in 2016 dollars: US$118 per month.
i This is close to a Purchasing Power Parity-adjusted estimate of the earnings di erences, because the vast majority of the earnings in both groups are spent in Haiti. The seven workers in Alabama, interviewed in person at their worksite, estimated that 10-15 percent of their US earnings were spent in the U.S.; the US seasonal work visa requires employers to pay for housing, transportation to and from the U.S., and weekday meals, minimizing workers' expenses while abroad. The median in the 'traveled' group is US$2,292, and in the 'did not travel' group US$108.
j Workers who traveled abroad and their families were asked how much money the worker brought/sent back to Haiti.
k A t-test for the equality of population means rejects the null of equality with p < 0.0001. The median in the 'traveled' group is US$354, and in the 'did not travel' group US$108. l A t-test fails to reject equality of means with p = 0.89. The median in the 'traveled' group is US$208, and in the 'did not travel' group US$186. m A t-test fails to reject equality of means with p = 0.95. n This method is discussed further in Clemens (2013a). In Cobb-Douglas production, the output elasticity of an input is well approximated by its cost share. The simplest version of the dual problem proceeds as min K, L wL + rK s.t.
AK 1−α L α = Q − −− → FOC α = w L w L+r K , where K, L are capital and labor, with prices r, w; A is total factor productivity; and Q is some output quantity. For this reason the industrial organization literature commonly approximates rm-level output elasticity with industry-level cost share of the input (e.g. Griliches 1963;Syverson 2004;Foster et al. 2008). o These are the 2010 US Bureau of Economic Analysis Regional Input-Output Modeling System (RIMS II) regional economic model agriculture-sector multipliers for Alabama and Oregon for output across all sectors of the state economies. These are the 'Type I' multipliers, which omit any local spending of wages by the Haitian workers. 24 p For Alabama, a worker-month of Haitian labor is paid US$10.59/hour × ∼5.5 days/week × 8 hours/day × 4 weeks/month = $1,864, thus $1,864 × (1.11 × 2.0) = $4,138. For Oregon, a worker-month of Haitian labor is paid US$12.69/hour × ∼5.5 days/week × 8 hours/day × 4 weeks/month = $2,233, thus $2,233 × (1.11 × 1.5) = $3,719.
These estimates are furthermore conservative because, by using the Cobb-Douglas estimate of the farm productivity multiplier (1.11), they assume that manual labor could be fully substituted with su cient investment in machinery.
But machinery to fully automate the tasks performed by these workers does not exist. A more reasonable, but still conservative estimate would take the ratio of marginal revenue product to wage as something greater than 2.0 rather than 1.11, resulting in an estimate of over $6,700-7,455 per Haitian worker-month of added value to all sectors of the US economy. The estimates are moreover conservative because they ignore multiplier e ects in the US from the 10-15 percent of Haitian workers' earnings that are spent in the U.S. q For example, for Tonga: NZ$12,000/7 months = NZ$1,714/month; NZ$1,400/12 = NZ$/117/month; and 1,714/117 = 14.7. This is a lower bound because not all Tongan workers stay the full duration of the maximum 7 months. r ∼US$1,700 per worker-month × 5 months × 4,000 workers = US$34 million. s The announcement (DHS 2010) was accompanied by a warning that "attempting to leave Haiti now will only bring more hardship to the Haitian people and nation. . . . Those who attempt to travel to the United States after January 12, 2010 will not be eligible for TPS and will be repatriated. " (Napolitano 2010). This policy was revised a year and a half later, after it was clear that Haitians were not departing in large numbers, to include people who had departed after the earthquake (DHS 2011).