94647 AUTHOR ACCEPTED MANUSCRIPT FINAL PUBLICATION INFORMATION The Development Impact of a Best Practice Seasonal Worker Policy The definitive version of the text was subsequently published in Review of Economics and Statistics, 96(2), 2014-05 Published by MIT Press and found at http://dx.doi.org/10.1162/REST_a_00383 THE FINAL PUBLISHED VERSION OF THIS ARTICLE IS AVAILABLE ON THE PUBLISHER’S PLATFORM This Author Accepted Manuscript is copyrighted by the World Bank and published by MIT Press. It is posted here by agreement between them. Changes resulting from the publishing process—such as editing, corrections, structural formatting, and other quality control mechanisms—may not be reflected in this version of the text. You may download, copy, and distribute this Author Accepted Manuscript for noncommercial purposes. Your license is limited by the following restrictions: (1) You may use this Author Accepted Manuscript for noncommercial purposes only under a CC BY-NC-ND 3.0 IGO license http://creativecommons.org/licenses/by-nc-nd/3.0/igo. (2) The integrity of the work and identification of the author, copyright owner, and publisher must be preserved in any copy. (3) You must attribute this Author Accepted Manuscript in the following format: This is an Author Accepted Manuscript of an Article by Gibson, John; McKenzie, David The Development Impact of a Best Practice Seasonal Worker Policy © World Bank, published in the Review of Economics and Statistics96(2) 2014-05 CC BY-NC-ND 3.0 IGO http://creativecommons.org/licenses/by-nc-nd/3.0/igo http://dx.doi.org/10.1162/ REST_a_00383 © 2015 The World Bank THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY John Gibson and David McKenzie* Abstract—Seasonal migration programs are widely used around the families. Temporary or circular migration programs are seen world, yet there is little evidence as to their development impacts. A multiyear prospective evaluation of New Zealand’s Recognised Seasonal as a way of overcoming such concerns and enabling poorer, Employer (RSE) seasonal worker program allows us to measure the less skilled workers to benefit from the higher incomes to be impact of participating in this program on households in Tonga and earned abroad as part of a ‘‘triple-win,’’ whereby migrants, the Vanuatu. Using a propensity-score prescreened difference-in-differences analysis based on surveys fielded before, during, and after participation, sending country, and the receiving country all benefit. Such we find that the RSE has indeed had positive development impacts that programs have been recommended as one of the most promis- dwarf those of other popular development interventions. It has increased ing ways to enhance the development benefits of migration by income, consumption, and savings of households; durable goods owner- ship; and subjective standard of living. The results also suggest that child a wide range of international organizations (United Nations, schooling improved in Tonga. 2004; Global Commission on International Migration, 2005; World Bank 2006), national governments (House of Commons First and foremost it will help alleviate poverty directly by pro- International Development Committee, 2004), and academics viding jobs for rural and outer island workers who often lack (Winters et al., 2003; Pritchett, 2006; Rodrik, 2007). income-generating work. The earnings they send home will sup- Almost all OECD countries have temporary worker port families, help pay for education and health, and sometimes migration programs, with seasonal workers the largest cate- provide capital for those wanting to start a small business. gory, totaling 576,000 workers in 2006 (OECD, 2008). Such Winston Peters, New Zealand minister of foreign affairs, programs remain controversial, especially when geared to on the approval of the RSE program, October 20061 low-skilled migrants. Some critics of such programs raise concerns that workers will overstay and they will compete A guest worker program is the most effective contribution we down the wages of native poorer workers (e.g. Borjas, can make to improving the lives of the world’s working poor. 2007), while others raise concerns about the possible exploi- Dani Rodrik, New York Times, June 1, 2007 tation of migrants and whether they can earn enough to make it worthwhile if the duration of work is short.2 I. Introduction Lacking in this debate is credible evidence on the devel- opment impact of international seasonal worker programs. I NTERNATIONAL migration is probably the most effec- tive mechanism we know to rapidly increase the incomes of poor people (Clemens, Montenegro, & Pritchett, 2008), and to The few existing studies are based on ex post surveys of migrants and lack credible counterfactuals for what would have happened to households in the absence of migration.3 help narrow global income gaps (Hanson, 2009). It is also one This paper seeks to provide credible evidence on the devel- of the most controversial, with migrant-receiving countries opment impacts of seasonal migration by means of a pro- worried about the costs of assimilating workers and their spective multiyear evaluation of New Zealand’s Recog- nised Seasonal Employer (RSE) program. The RSE began Received for publication September 27, 2011. Revision accepted for in 2007 and aims to ease labor shortages in New Zealand’s publication January 31, 2013. * Gibson: University of Waikato; McKenzie: World Bank, BREAD, horticulture and viticulture industries, while also promoting CEPR, CReAM, and IZA. economic development in the Pacific Islands. The policy We thank the editor and three anonymous referees for useful comments was developed taking account of lessons from previous sea- that helped sharpen the initial version of this paper; AusAID and the World Bank for funding for this project; Manjula Luthria for the catalyz- sonal worker programs elsewhere and is viewed as a possi- ing role she has played in the development of this policy and in support- ble model for other countries. For example, the ILO good ´a-Martinez, Halahingano Rohorua, and Alan ing its evaluation; Pilar Garcı Winters for their collaboration in earlier phases of this project; the New practices database states, ‘‘The comprehensive approach of Zealand Department of Labour, Tonga Department of Labour, Vanuatu the RSE scheme towards filling labour shortages in the hor- Department of Labour, MFAT, and other members of the RSE Intera- ticulture and viticulture industries in New Zealand and the gency Governance Committee for their collaboration in this research; Kim Robertson and Simil Johnson for work as field supervisors in system of checks to ensure that the migration process is Vanuatu; Emily Beam and Melanie Morten for research assistance; semi- orderly, fair, and circular could service as a model for other nar audiences and discussants at the Center for Global Development, destination countries.’’4 Dublin, Georgetown, Paris, Venice, and Waikato for helpful comments; and most of all, the interviewers and survey respondents in Tonga and 2 Vanuatu. All views expressed are our own, and do not necessarily repre- These concerns are discussed in Ruhs (2006), Pritchett (2006), and sent those of our employers. OECD (2008), among others. 3 A supplemental appendix is available online at http://www.mitpress For example, Basok (2000) conducted a snowball sample of Mexican journals.org/doi/suppl/10.1162/REST_a_00383. workers in Canada’s seasonal worker program in one area of Canada and 1 Quoted in ‘‘Seasonal Work Policy Benefits Pacific Says Peters,’’ in one village in Mexico. Macours and Vakis (2010) use a cross-sectional Islands Business, October 26, 2006, http://www.islandsbusiness.com survey of Nicaraguan households near the border with Honduras, where /news/index_dynamic/containerNameToReplace=MiddleMiddle/focus workers migrate seasonally. 4 ModuleID=130/focusContentID=6691/tableName=mediaRelease/overide http://www.ilo.org/dyn/migpractice/migmain.showPractice?p_lang=en SkinName=newsArticle-full.tpl. &p_practice_id=48. The Review of Economics and Statistics, May 2014, 96(2): 229–243 Ó 2014 The World Bank 230 THE REVIEW OF ECONOMICS AND STATISTICS Our evaluation was designed prospectively, alongside the opment projects less attractive. Our results are most directly program launch. We conducted baseline surveys of house- applicable to small countries, such as the 45 developing holds and communities in Tonga and Vanuatu before work- countries with populations below 1.5 million. But the find- ers left to work in New Zealand and then reinterviewed ings may have even broader relevance since New Zealand’s these same households 6, 12, and 24 months later. These program is sizable by international standards (greatly rich baseline data and institutional knowledge of how exceeding Australia’s fledging seasonal worker scheme and recruitment occurred let us use propensity score matching about one-third the size of Canada’s SWAP program, for to identify an appropriate set of households to act as a com- example) and is touted as a model for many destination parison group for the households in the RSE. Following countries. Crump et al. (2009) we use this propensity score to pre- The remainder of the paper is structured as follows. Sec- screen the sample used for panel difference-in-differences tion II describes the RSE policy and worker recruitment. and fixed effects estimation. With these methods, we assess Section III defines how we see development impact, and the impacts of the RSE on household incomes, consump- section IV describes our surveys and estimation methodol- tion, savings, durable assets, and subjective well-being and ogy. Household-level impacts are estimated in section V, also measure broader community-level impacts. and impacts at the community and macro level are dis- The results show large positive effects on sending house- cussed in section VI. Section VII concludes. holds in Tonga and Vanuatu; per capita income of partici- pating households rises by over 30% relative to the compari- son groups in both countries, while per capita expenditure II. The RSE Program and savings also rise. Subjective economic welfare increases The RSE was launched on April 30, 2007, and initially by almost half a standard deviation for participants in both let up to 5,000 seasonal workers come to New Zealand for a countries, who are also seen to have purchased more durable maximum of seven months per eleven-month period to assets. In Tonga, RSE households also doubled the rate of work in horticulture and viticulture. Preference is given to home improvement, and in both countries, households workers from Pacific countries (except Fiji), with Kiribati, became more likely to have a bank account, likely reflecting Samoa, Tonga, Tuvalu, and Vanuatu given special ‘‘kick- more formal savings. In addition, there is some evidence start’’ status that entailed deliberate and expedited efforts to that school attendance rates for 16- to 18-year-olds launch the program and recruit in these countries. Vanuatu increased in Tonga. An additional innovative feature of our and Tonga, the focus of our impact analysis, supplied the analysis was to survey community leaders, who reported most workers under the RSE in the first two seasons: 3,590 positive broader impacts on the community. workers in the case of Vanuatu and 1,971 from Tonga Overall these results show that the seasonal worker pro- (including return workers).5 gram has been a powerful development intervention for the Ramasamy et al. (2008) detail the origins of the RSE as a participating households, with aggregate effects that are solution to the long-standing problems the horticulture and important relative to aid flows and export earnings. Thus, the viticulture industries had in meeting seasonal labor needs, RSE policy appears to have succeeded in its development while contributing to New Zealand’s broad development objectives in the short run. There are very few rigorous impact goals in the Pacific region. Design of the RSE paid careful evaluations that show large gains in income from develop- attention to previous experience with seasonal worker pro- ment interventions, and the development impacts of this sea- grams around the world to reduce risks of overstaying, dis- sonal worker program on participating households dwarf placement of New Zealand workers, and worker exploitation. those found in recent evaluations of other popular develop- The risk of overstaying is mitigated in a number of ways: ment interventions like conditional cash transfers, microfi- workers may be reemployed in subsequent years, with the nance, business training, and grants to microenterprises. same or a new employer, which can be contrasted with A common issue for any in-depth evaluation of a single single-entry programs that provide high incentives for work- intervention is the extent to which the results may general- ers to overstay; employers are required to pay the costs asso- ize to other settings. We note first that seasonal worker pro- ciated with worker removal from New Zealand if workers grams are an important policy lacking rigorous evidence become illegal, giving employers incentives not to be compli- from any setting. Second, we examine impacts in two coun- cit in their overstaying; and competition for places among tries that differ in many dimensions: Tonga and Vanuatu communities and countries leads to social pressures not to have different ethnic populations (one Polynesian, one Mel- jeopardize future possibilities for others by overstaying and anesian), different government structures, and different thereby creating a negative reputation for one’s community. prior histories of migration. Finding similar results in both The risk of displacement of New Zealand workers is countries therefore suggests the results are not particular to mitigated through a New Zealanders first principle that a single context. Third, small countries are indeed one of the most relevant contexts to look at such programs: emi- 5 A return worker is an individual who participates in the RSE in one gration rates are highest for small island countries, where season and then returns again in a second season; he or she is counted as the lack of market size and remoteness makes other devel- two workers in the administrative data. THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 231 requires employers to first lodge their vacancies with the The RSE has been viewed as a success from New Zeal- Ministry of Social Development (which provides welfare and’s point of view. An evaluation of the first two years benefits and job search services) before attempting to conducted by the New Zealand Department of Labour recruit offshore. The risk of exploitation is mitigated (2010, p. xvii) concluded, ‘‘Overall, the RSE Policy has through regulations stating that workers must not be achieved what it set out to do.’’ The policy is found to have charged recruitment fees and that employers must pay mar- provided employers in the horticulture and viticulture ket wages and offer workers at least a minimum remunera- industries with access to a reliable and stable workforce, tion, which depends on the length of the contract. Employ- with productivity gains starting to emerge as workers return ers also must arrange suitable accommodation, internal for repeat seasons. The main concerns raised about tempor- transportation, access to personal banking services, provi- ary labor programs have been mitigated: the evaluation sion of protective equipment, and opportunities for recrea- finds little displacement of New Zealand workers, overstay tion and religious observance. rates were only 1% in the first season and less than 1% in Recruitment options were fit to the needs of each coun- the second, and concerns about worker exploitation have at try. In Tonga employers could recruit workers directly or most arisen in a couple of isolated cases. The question this from a work-ready pool of Tongans prescreened and paper addresses is then whether the RSE has also lived up selected by the Labour Ministry. In the first year, recruit- to the policy goal of improving development in the Pacific. ment was mainly from the work-ready pool. Preselection and screening was by district and town officers and church and community leaders. There was tremendous interest in III. What Do We Mean by Development Impact? the program, evident from the fact that more than 5,000 In order to measure whether the RSE has improved devel- Tongans registered for the work-ready pool within three opment outcomes in the Pacific Islands, we must first define months of the launch of the program. We describe this what we mean by development impact for such a program. selection process (Gibson, McKenzie, & Rohorua, 2008) We see three related definitions of development impact and and show the main attributes used by village committees in attempt to provide evidence on each. The first, and narrow- preselection, which favored low-income applicants. The ests, views development as anything that raises the income Tongan Labour Ministry ensured that all villages in the of people from poor countries. The large wage differences country had workers in the scheme. between a developed and a developing country would In Vanuatu employers could hire directly or through an strongly suggest we should expect to see an impact on agent. Direct recruitment is facilitated by the Vanuatu household income from participating, and so the main ques- Department of Labour, which in the first year also used a tion is then measuring how large this impact is. However, work-ready pool of workers from walk-ins who registered there is a view that guest-worker programs in agriculture are directly with the department. These workers were typically ‘‘close to slavery,’’ with workers being exploited and not from the more urban areas. In rural areas, direct recruitment paid the promised wages, so that after meeting expenses, and agents relied heavily on community contacts through they are no better off (Southern Poverty Law Center, 2007). village councils, again using villages to prescreen workers. Our analysis will help rule out this possibility. In McKenzie, Martinez, and Winters (2008), we study this A second, broader view sees development as increasing process, and find that agents and villages looked for similar the utility of households in developing countries. The key characteristics to those selected on in Tonga except that it issue in this view is whether the gain in income is more than was not the poorest households that applied and had work- offset by any negative issues to the household that arise from ers selected. This perhaps reflected the newness of interna- physical separation of household members. A related litera- tional migration in Vanuatu, with communities more con- ture on the impacts of separation in military families has cerned with sending workers who would represent the applied two approaches to addressing this issue. The first is a village; in addition, the poorest households lacked informa- revealed preference approach, which examines the trade-offs tion about the program in the first year and lacked the soldiers make between time and money (Dunn, 2003), noting resources to finance the costs of the travel process. that to the extent the decision is a choice, households must Typical work under the RSE includes working in vine- have decided that expected disutility is more than offset by yards to prune vines and pick grapes, harvesting apples and the additional income. This approach may be less useful for a kiwifruit and other fruit picking, and working in the pack- new program in which information may be incomplete. A house to sort, grade, and pack the fruit. The work was typi- second approach attempts to measure impacts on specific out- cally physically demanding and included work in both cold comes without trying to aggregate up to an overall impact.7 and hot conditions. In part due to the nature of the work, the majority of RSE workers recruited were male: in the first 7 For example, Angrist and Johnson (2000) find that deployment in the year in our sample, 82% of the ni-Vanuatu RSE workers First Gulf War led to no changes in divorce rates or child disability when and 87% of the Tongan RSE workers were male.6 men were deployed, but increases in divorce when women were deployed. Lyle (2006) finds that military deployments to Iraq and Afghanistan have a negative impact on children’s test scores, with the impact greater when 6 Ni-Vanuatu is the term used to refer to citizens of Vanuatu. women leave. 232 THE REVIEW OF ECONOMICS AND STATISTICS In our context, we look at a number of specific outcomes, as Ex ante it was not known how many individuals from each well as subjective well-being as a measure of overall utility. country would participate in the scheme, but the numbers A final view of development is more concerned about likely to be involved were certainly too small for a simple whether interventions have long-term impacts on house- random sample of households to pick up enough RSE house- holds. The concern here is that either income earned abroad holds in a cost-effective way. Hence, we needed to know does not make it back to the household in the first place RSE status before surveying. Survey design was then com- because it is consumed by the migrant, or that the money plicated by the fact that approvals to recruit workers and that is received is spent on things that increase short-term recruitment took place on a rolling basis. Once workers were household utility but leave the household no better off in selected for recruitment, they often had only two or three the long term. The question is, then, whether money re- weeks before they departed for New Zealand, leaving a very ceived by the family is spent on durable asset accumula- short window of time to interview them and their household. tion, saved, or invested in children’s education and in busi- Therefore, we used a rolling sampling methodology, add- nesses, as well as whether there are spillover benefits for the ing sample as we received updates of when, where, and broader community. We examine these outcomes, although who employers were recruiting, with the baseline survey we also note that this view is based on the presumption that conducted between October 2007 and April 2008. In Tonga, participating in seasonal migration for many years is not a our survey has nearly national coverage, covering the possibility, whereas in practice, households could increase islands of Tongatapu, Vava’u, and ‘Eua (containing 90% of long-term well-being by participating in this program for the population and 92% of year 1 RSE workers). Officials many years. helped us identify households with RSE workers and house- holds with members of the RSE work-ready pool who had IV. Our Surveys and Estimation Methodology not been selected yet. We also surveyed randomly selected households in the same villages who had no one yet apply A. The Surveys for the program. In each village, we aimed for approxi- mately five households with an RSE worker, three house- There was keen interest from national governments on holds with a member of the work-ready pool who was not both sides of the migration relationship and from the World selected, and four households with nonapplicants. The Bank in learning whether the new RSE program would have choice of a control sample from the same villages as the the development impacts envisioned as one of its core ratio- treated was informed by our previous experience matching nales. It was therefore decided ex ante to have a rigorous for a different migration program in Tonga (McKenzie, evaluation of development impacts, focusing on Tonga and Gibson, & Stillman, 2010) and work on evaluation of labor Vanuatu. These were expected to be the two countries parti- training programs, which stresses the importance of match- cipating most, and they offer an interesting contrast in prior ing on the same or similar labor markets (Dehejia, 2005). migration history. Tonga (population 100,000) has tradi- Our resulting baseline survey covered 448 households con- tionally had high emigration rates to New Zealand, Australia, taining 2,335 individuals in 46 villages.8 and the United States, with most recent migration through Vanuatu’s rugged geography and high transportation costs family-sponsored categories and a special annual permanent made it infeasible to survey in all islands, so we limited the migration quota to New Zealand, the Pacific Access Category. evaluation to three islands from which we believed there was The 2006 New Zealand Census enumerated 20,520 Tongan- a high chance of workers coming: Efate (population 50,000), born in New Zealand. In contrast, Vanuatu (population containing the capital city, Port Vila; Ambrym (population 215,000) has had relatively little international emigration, 10,000); and Tanna (population 20,000). In contrast to Tonga, with only 1.5% of its population abroad prior to the RSE not all villages in Vanuatu initially participated in the RSE, so (World Bank, 2008) and fewer than 1,000 Vanuatu-born in as well as sampling nonapplicant households in villages with the 2006 New Zealand Census. participating RSE workers, we also sampled households from Given that recruiting occurred at the employer level, the nearby villages that had not participated in the RSE. Ulti- interests of some employers in screening workers them- mately our baseline survey covered 456 households contain- selves, and the large number of employers involved, it was ing 2,173 individuals in 48 villages or communities. never going to be feasible to attempt to get employers to Three rounds of follow-up surveys were then conducted. randomly select workers. Therefore, we decided the most The first took place between April and July 2008, approxi- credible impact evaluation strategy would be a matched mately six months after the baseline survey. This was difference-in-differences approach. This would entail a intended to be a time when RSE workers were still in the baseline survey of households that would participate in the midst of their seven-month stint abroad. However, as in RSE before the workers left, along with surveys of nonpar- practice many contracts were for shorter than seven months, ticipating households, and then following these households over time. Nonparticipating households would be separated 8 Further details of the baseline sampling methodology for Tonga are into whether they had a member of the work-ready pool contained in Gibson et al. (2008); McKenzie et al. (2008) provide more who had applied for the program but not been selected. details on the Vanuatu sampling methodology. THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 233 TABLE 1.—MEANS OF BASELINE CHARACTERISTICS OF HOUSEHOLDS FOR TONGAN SAMPLE Full Sample PS-1 in [0.1, 0.9] PS-2 in [0.1, 0.9] RSE Non-RSE RSE Non-RSE RSE Non-RSE Households Households Households Households Households Households Household size 5.70 4.82*** 5.61 5.08** 5.37 5.05 Number of males 18 to 50 1.50 1.25*** 1.50 1.34* 1.48 1.37 Share of males 18 to 50s who: Are literate in English 0.92 0.85** 0.91 0.90 0.92 0.89 Have more than 10 years schooling 0.46 0.49 0.47 0.52 0.48 0.52 Have very good self-reported health 0.68 0.60* 0.68 0.63 0.65 0.59 Drank alcohol in the last month 0.42 0.39 0.41 0.42 0.44 0.46 Mean days hard labor in past week 4.56 3.97*** 4.53 4.19* 4.46 4.04 Share of adults who previously have 0.38 0.20*** 0.39 0.23*** 0.36 0.21*** worked or studied in New Zealand Number of relatives in New Zealand 5.41 4.80* 5.33 4.87 4.93 4.64 Household durable assets index 0.07 À0.06 0.05 0.01 0.02 À0.15 Number of pigs 5.57 5.49 5.52 5.40 5.42 5.12 Number of chickens 5.11 5.12 5.05 5.12 4.82 4.69 Number of cattle 0.45 0.47 0.48 0.42 0.52 0.44 Have a traditional-style dwelling 0.15 0.13 0.13 0.10 0.14 0.14 Located on Tongatapu or Efate 0.81 0.80 0.82 0.83 0.83 0.83 Semiannual per capita income (pa’anga) 979 1,342*** 991 1,142 1,064 1,103 Semiannual per capita consumption (pa’anga) 829 1,184*** 831 948* 874 978 Proportion with income per capita below $US1 per day 0.19 0.12** 0.19 0.14 0.18 0.17 Proportion with income per capita below $US2 per day 0.49 0.36*** 0.49 0.41 0.44 0.44 Had a male aged 18 to 50 work for pay in early 2007 0.21 0.27 0.21 0.26 0.23 0.25 Mean change in weekly wage income 8.91 14.67 13.83 15.06 16.60 13.12 2006 to 2007 (pa’anga) Median change in weekly wage income 0.00 0.00 0.00 0.00 0.00 0.00 2006 to 2007 (pa’anga) Sample size 197 251 183 196 153 121 *, **, and *** indicate that differs in mean from the RSE households at the 10%, 5%, and 1% levels, respectively. PS-1 and PS-2 are the two propensity-score matched groups. Both match on household demo- graphics, characteristics of the 18- to 50-year-old males in the household, the household’s previous experience and network in New Zealand, household baseline assets and housing, geography, and past household wage and salary history in the first half of 2006 and 2007. PS-2 additionally restricts the sample to households that had a member apply to participate in the RSE. approximately two-thirds of Tongan RSE workers and one- households (58% Tonga and 54% Vanuatu) had only one fifth of ni-Vanuatu RSE workers in our sample had returned seasonal worker spell during the two years of our study, by the time of this survey. The second follow-up survey while the rest had multiple spells. Since the number of took place between October 2008 and February 2009, times a household participates in the RSE conditional on approximately one year after the baseline, while the third participating is potentially the result of household choices, and final follow-up survey took place between October we focus on a binary measure of RSE participation: RSEi,t, 2009 and March 2010, two years after baseline. which takes a value of 1 if household i has at least one Attrition was remarkably low in the Tongan sample. Of member who has worked in the RSE by time t, where t ¼ the 448 households in the baseline, we were able to reinter- 1,2,3 and 4 corresponds to our four survey waves. Estimat- view 442 households in the second round survey, 444 in the ing the impact of RSEi,t then involves estimating the aver- third round, and 440 in the fourth round. Attrition was age impact of ever participating in the RSE over the first higher in Vanuatu. Of the 456 households in the baseline two years of the program. survey, 382, 388, and 348 households were reinterviewed in We then begin with panel data regressions of the impact rounds 2, 3, and 4, respectively, while 33 households were of the RSE, using the full sample of households separately interviewed only in round 1. In an online appendix, we for each country. Letting Yi,t be an outcome of interest for show our main results are robust to this attrition. household i in survey round t, we begin with the following difference-in-differences specification, X4 B. Estimation Methodology Yi;t ¼ a þ bEverRSEi þ d þ cRSEi;t þ ei;t ; ð1Þ t¼2 t Table 1 in the online appendix has the number of house- where EverRSEi indicates whether household i ever partici- holds participating in the RSE by survey round. Most new pates in the RSE over the four waves of our sample and dt entries into the RSE in our sample came between the first are survey round dummies. The coefficient of interest is and second rounds, with few additional households from then g, which gives the average treatment effect of partici- our sample joining the RSE in the later rounds.9 Most RSE pating in the RSE. We do not include additional time-vary- 9 ing controls in this regression, since we have few time- Of course, other households from the general population also joined the RSE during the time of subsequent survey rounds, but these house- varying variables that are not potentially themselves holds were outside our initial sample. affected by the RSE. Standard errors are clustered at the 234 THE REVIEW OF ECONOMICS AND STATISTICS household level to account for autocorrelation in the error Our surveys of RSE and non-RSE households were fielded term ei,t across survey waves. at the same time in the same villages (and hence local labor Difference-in-differences controls for any baseline-level markets) using the same questionnaire. We know the char- differences in the outcome Yi,t at the group level. An alter- acteristics that villages and employers used in selecting native approach is to control for baseline differences at the workers and can include these in the matching specification. household level with household-level fixed effects. We esti- Because this was a new program, employers relied largely mate this using on the prescreening and observable characteristics like Eng- X4 lish literacy to choose workers. Also, we have more than Yi;t ¼ li þ d þ cRSEi;t þ ei;t ; t¼2 t ð2Þ one period of pre-RSE wage earning data (though not many earned wage income). Furthermore, we know whether where mi is the fixed effect for household i. households tried to participate in the RSE (by having a In both specifications, g measures the average impact of member register for the work-ready pool or apply directly participating in the RSE over the two-year period of our to an employer). Finally, we have a plausible reason that study. Pooling multiple rounds of posttreatment data pro- some households participated in the RSE and other house- vides more power to identify the effect of interest, espe- holds with similar characteristics did not: there was excess cially for outcomes like income and consumption that are demand for RSE employment, so not all households that not highly autocorrelated (McKenzie, 2012).10 wanted to participate were able to. The difference-in-differences and fixed-effects specifica- We estimate two versions of the propensity score PS-1 tions let us estimate differential effects on RSE households and PS-2 which differ only in that PS-2 restricts the sample but cannot capture macro effects benefiting the non-RSE to RSE-applicant households, removing nonapplicant households. The RSE has broader effects at the community households. This lets us explicitly screen on demand for the level, as shown in section VI, but mainly through remit- RSE, although given that the reason many nonapplicant tances earmarked for community projects rather than households said they did not apply was lack of information through the absence of migrants, leading to more job oppor- about the program (Gibson et al., 2008; McKenzie et al., tunities or higher wages for those remaining. Similarly, 2008), failure to apply need not imply lack of demand, and there appears to be little effect on prices; many of these vil- the exclusion of nonapplicants reduces our sample size, lages lack markets, with imported durable goods and even which has costs in terms of power. food markets often available only in the main large towns. We use six main categories of variables we believe may A second underlying assumption of difference-in-differ- influence participation in the RSE to estimate the propensity ences and fixed effects is that after controlling for level dif- score: demographic variables (household size, numbers of ferences, households would have exhibited similar trends in adults, school-aged children, and males aged 18 to 50); the outcome variables in the absence of the RSE. This is less characteristics of the 18- to 50-year-old males in the house- credible if the households we are comparing have very dif- hold, who are the individuals most likely to participate ferent characteristics. We therefore follow the recommenda- (share literate in English, share with schooling beyond tions of Crump et al. (2009) of estimating a propensity score grade 10, share with self-reported health rated as very good, and dropping observations with estimated propensity scores share who drank alcohol in the past month, and the mean outside the range [0.1,0.9].11 This systematic approach to number of days of hard labor carried out in the past month); prescreening the sample ensures the regression is estimated the household’s previous experience and network in New only for the sample where the covariate distribution overlaps Zealand (share of adults who had previously been to New for the RSE and non-RSE households. Angrist and Pischke Zealand, number of relatives in New Zealand); household (2009) show that this approach works well in approximating baseline assets and housing infrastructure (an asset index the experimental results obtained in a U.S. work experience from the first principal component of durable goods, pigs, program. An alternative approach would be to use the pro- cattle, and chickens and whether the dwelling was tradi- pensity scores directly in estimation. The online appendix tional style); geography (on Tongatapu or Efate as opposed shows our results are robust to this alternative, although we to one of the other islands); and past household wage and prefer the prescreened regression approach given that we salary history (household wage income for the first half of have multiple rounds of posttreatment data to use. 2006 and 2007, and whether the household had any male Our study includes many of the features identified as aged 18 to 50 who worked for pay in 2006 and 2007). For desirable for propensity score matching (Dehejia, 2005). each variable, we include both the variable and its square in estimating the propensity score. 10 An alternative approach would be to use each of the subsequent sur- For Tonga, estimating the propensity score and restrict- vey rounds separately with the baseline and attempt to estimate the trajec- tory of treatment effects. But our power to do this is very low, and hence ing to the range [0.1, 0.9] reduces our sample of 448 house- we focus on the average impact over the two years. holds (197 RSE, 251 non-RSE) to 379 households using 11 In both Tonga and Vanuatu, the common support of our propensity PS-1 (183 RSE and 196 non-RSE) and 283 households score distributions is wider than the [0.1, 0.9] range. Crump et al. (2009) show that the [0.1, 0.9] cut-offs have good optimality properties and using PS-2 (153 RSE, 121 non-RSE). In Vanuatu the sam- closely approach data-dependent optimal thresholds in most contexts. ple of 456 households (147 RSE, 309 non-RSE) reduces to THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 235 TABLE 2.—MEANS OF BASELINE CHARACTERISTICS OF HOUSEHOLDS IN VANUATU Full Sample PS-1 in [0.1, 0.9] PS-2 in [0.1, 0.9] RSE Non-RSE RSE Non-RSE RSE Non-RSE Households Households Households Households Households Households Household Size 4.72 4.83 4.71 4.68 4.68 4.71 Number of males 18 to 50 1.25 1.21 1.26 1.23 1.22 1.17 Share of male 18 to 50 who: Are literate in English 0.85 0.70*** 0.85 0.79 0.85 0.75** Have more than 10 years schooling 0.06 0.07 0.07 0.06 0.06 0.05 Have very good self-reported health 0.83 0.69*** 0.84 0.77 0.83 0.70*** Drank alcohol in last month 0.52 0.55 0.51 0.56 0.52 0.53 Mean days of hard labor in past week males 3.05 3.38 3.02 3.37 3.15 3.35 Share of adults who previously have worked 0.08 0.01*** 0.05 0.01*** 0.04 0.00*** or studied in New Zealand Number of relatives in New Zealand 0.10 0.06 0.09 0.03* 0.08 0.05 Household durable assets index 0.60 À0.29*** 0.60 À0.12*** 0.14 À0.48** Number of pigs 3.82 3.42 3.67 3.62 3.30 3.37 Number of chickens 9.99 12.75* 10.16 11.72 10.20 11.73 Number of cattle 1.39 1.73 1.38 1.72 1.38 1.32 Have a traditional-style dwelling 0.70 0.75 0.70 0.75 0.72 0.78 Located on Tongatapu or Efate 0.46 0.33 0.46 0.38 0.42 0.33 Semiannual per capita income (vatu) 85,282 71,961 79,188 69,805 73,384 72,442 Semiannual per capita consumption (vatu) 65,872 55,462* 68,795 58,953 65,366 60,909 Proportion with income per capita below $US1 per day 0.19 0.21 0.19 0.20 0.20 0.20 Proportion with income per capita below $US2 per day 0.37 0.44 0.36 0.43 0.41 0.45 Had a male aged 18 to 50 work for pay in early 2007 0.41 0.35 0.42 0.34 0.40 0.28** Mean change in weekly wage income 2006 to 2007 (vatu) 753 À1,030 À95 À136 À146 À322 Median change in weekly wage income 2006 to 2007 (vatu) 0.00 0.00 0.00 0.00 0.00 0.00 Sample size 147 309 129 231 123 146 *, **, and *** indicate that differs in mean from the RSE households at the 10%, 5%, and 1% levels, respectively. PS-1 and PS-2 are the two propensity-score matched groups. Both match on household demo- graphics, characteristics of the 18- 50-year-old males in the household, the household’s previous experience and network in New Zealand, household baseline assets and housing, geography, and past household wage and salary history in the first half of 2006 and 2007. PS-2 additionally restricts the sample to households that had a member apply to participate in the RSE. 360 households using PS-1 (129 RSE, 231 non-RSE) and to of our study we estimate equation (3) without including the 269 households using PS-2 (123 RSE, 146 non-RSE). Trim- baseline lag since these variables were not asked at baseline. ming mainly removes non-RSE households too dissimilar to RSE households to be appropriate comparators, plus a few RSE households that differ too much from any non- C. Summary Statistics RSE household. We reestimate equations (1) and (2) for The main outcomes of interest are household income and households with propensity scores in the range [0.1, 0.9]. expenditure, savings, asset ownership, and schooling. The Again the differencing or fixed effects will eliminate both online appendix details how we measure income and expen- observed and unobserved time-invariant differences among diture. Following Paxson (1992), we measure savings as a households, and the assumption of a common underlying flow measure, defined as the difference between income trend in the absence of the RSE is likely to be more credible and expenditure in a period. for households with propensity scores within this range. Table 1 reports baseline means of household characteris- We use equations (1) and (2) to look at the impact of the tics in Tonga for RSE and non-RSE households for the full RSE on flow variables of interest like income and consump- sample and for the PS-1 and PS-2 screened subsamples. tion. For impacts on stock variables like assets owned, we Table 2 presents the same comparisons for Vanuatu. Aster- instead estimate, for households within the propensity score isks show the results of tests for difference in means. Con- range [0.1, 0.9], the following equation: sider first the Tongan sample. The average RSE household has 5.7 members. The rural subsistence farming nature Yi;4 ¼ a þ bYi;1 þ cRSEi;4 þ ei;4 : ð3Þ of these households is seen in only 21% of these house- holds having any male wage or salary worker in the six For example, estimating equation (3) for whether the months prior to the launch of the RSE, as well as in their household owns a TV is equivalent to asking whether, con- ownership of pigs and chickens. Semiannual per capita ditional on their TV ownership status in the baseline, house- income and consumption, including the value of goods holds in the RSE are more likely to own a TV two years produced for own consumption, averaged 830 pa’anga later than non-RSE households with similar covariates. (around US$430).12 This is less than an RSE worker could Finally, for the variables subjective well-being, making a earn in a good week in New Zealand. dwelling improvement over the two years of our study and making a major asset purchase (200 pa’anga or more in 12 In April 2008, NZ$1 ¼ 1.52 pa’anga and US$1 ¼ 1.92 pa’anga; Tonga, 10,000 vatu or more in Vanuatu) over the two years NZ$1 ¼ 73.08 vatu, and US$1 ¼ 92.50 vatu. 236 THE REVIEW OF ECONOMICS AND STATISTICS Tongan RSE households tend to be larger and poorer relatives in New Zealand, compared to 0.1 relatives for the than the average non-RSE household in our sample, and average ni-Vanuatu RSE worker. With higher levels of school- their males worked more days of hard labor on average than ing in Tonga, a larger share of adult males in Tonga are literate in non-RSE households, reflecting selection of workers bet- in English, and 46% of 18- to 50-year-old males in RSE house- ter able or more inclined to do physical work (table 1). The holds exceed ten years of schooling versus only 6% in RSE households are also more connected to New Zealand, Vanuatu. But the Vanuatu sample is more likely to have pre- with adults in the household more likely to have previously viously worked for pay, and in the end, the poverty rates are visited New Zealand and the household having more rela- similar for our evaluation samples in both countries. These dif- tives in New Zealand. We believe that in both countries, ferences in contexts offer the possibility of examining how dif- RSE households in some cases reported the current RSE ferent the effects are in quite different starting circumstances, episode, so the actual difference in pre-RSE experience in offering some degree of external validity for the results. New Zealand is likely less than the gap shown here.13 The After prescreening with the propensity score, both the third through sixth columns of table 1 show that matching Tongan and ni-Vanuatu samples are balanced on initial and restricting to households with propensity scores incomes, consumption, and poverty, our key outcomes of between 0.1 and 0.9 makes the RSE and non-RSE house- interest. The income-generating processes in these countries holds more similar. The PS-2 subsample in particular does were fairly stable over the period examined; households not differ significantly in baseline demographics, income, mainly do semisubsistence farming as they had been doing or consumption from the subsample of RSE households for years. Hence, assuming parallel trends in the absence of with propensity scores in the [0.1, 0.9] range and differs the RSE seems reasonable. Ideally one would have several only in previous experience in New Zealand, which, as rounds of preintervention outcome data to check this, but noted, may be overstated for RSE households. the difficulty of recalling consumption and agricultural A natural concern is then whether our estimation is also income from previous years makes this infeasible in our perhaps picking up the value of being more connected to case, as it likely is in any similar evaluation. Wage income New Zealand, and not just the impact of the RSE. We note is more readily recalled, so as a further check, the bottoms that after matching, there is not a significant difference of tables 1 and 2 show no difference in the growth in wage between groups in whether they have relatives in New Zeal- income between RSE and non-RSE households over 2006– and or in whether other members of the household have been 2007, a full year before the program began. to New Zealand. Second, to the extent a difference remains and is not due to the measurement issue noted above, our dif- V. Household-Level Impacts ference-in-differences and fixed-effects estimation will still eliminate any time-invariant impact of this on household A. Impact on Incomes, Expenditure, and Savings outcomes. There then does not seem to be a strong reason to think there should be a large time-varying impact of being Table 3 presents the results of estimating equation (1) in more connected to New Zealand during our study period. columns 1 to 5 and equation (2) in columns 6 to 10 for In contrast to Tonga, table 2 shows that RSE households Tonga in panel A and for Vanuatu in panel B. Full sample in Vanuatu are richer than the average non-RSE household, results are shown first and then those for the propensity- with higher baseline asset ownership, income, and con- score screened samples. To check if results are driven by a sumption. But many participants are still poor by interna- few observations at the upper tail, columns 4 and 9 trim the tional standards; 37% of the RSE households have per top 1% of observations from the sample. Finally, columns 5 capita income below $US2 per day. Restricting to propen- and 10 use nearest-neighbor matching with replacement sity scores between 0.1 and 0.9 makes the RSE and non- based on PS-2 and include only the observations that are RSE households more similar, but in contrast to Tonga, used in at least one match. This further reduces the sample using just applicant households in PS-2 does not seem to size and power but serves as an additional robustness check. improve on PS-1. This may reflect the less widespread nat- Participating in the RSE has a large and statistically sig- ure of the work-ready pool in Vanuatu and the fact that nificant positive impact on household income per capita in many people who would like to have applied did not do so both countries. In Tonga, semiannual income is 233 to 249 because they lacked information, meaning that some nonap- pa’anga higher as a result of the RSE, relative to a baseline plicants may be better matches for RSE workers in Vanuatu income of 979 pa’anga. Trimming for potential outliers than we can find among our sample of applicants. increases the gain to 300 to 325 pa’anga. Log income is less Comparing the Tongan and ni-Vanuatu samples shows the sensitive to outliers and also shows large and statistically much greater prior exposure of Tongans to international significant increases. Using the estimates that screen on PS- migration; the average Tongan RSE worker surveyed has 5.4 2 and are thus restricted to RSE applicants, log income rises by 0.29 to 0.32, corresponding to a 34% to 38% increase in 13 per capita income as a result of the RSE. Semiannual Consistent with this view, the rate of previous experience in New Zealand is not significantly higher for other members of the RSE worker’s income in Vanuatu is approximately 44,000 vatu higher, household than it is for the average member in a non-RSE household. relative to a baseline of 85,000 vatu. In log terms, per capita TABLE 3.—AVERAGE IMPACT OF RSE MIGRATION ON HOUSEHOLD INCOME AND EXPENDITURE Difference-in-Differences Fixed Effects All PS-1 PS-2 PS-2-Trim Nearest N All PS-1 PS-2 PS-2-Trim Nearest N Baseline Mean for Outcome Variable RSE Households (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) A. Tonga Per capita income 979 331.0*** 278.4*** 233.1* 325.3*** 297.8** 347.4*** 271.0*** 248.7** 300.1*** 312.6*** (99.3) (105.3) (129.5) (90.9) (143.1) (87.5) (87.8) (106.2) (80.6) (108.3) Log per capita income 6.57 0.355*** 0.346*** 0.290*** 0.331*** 0.377*** 0.383*** 0.355*** 0.324*** 0.350*** 0.431*** (0.071) (0.077) (0.094) (0.087) (0.104) (0.066) (0.070) (0.084) (0.080) (0.0920) Per capita expenditure 829 224.1** 127.1 104.6 142.4** 106.7 249.8** 117.8* 133.0 145.0** 120.8 (111.6) (81.8) (104.1) (63.1) (116.4) (111.5) (65.6) (82.2) (61.7) (89.58) Log per capita expenditure 6.58 0.124** 0.117** 0.083 0.106* 0.124 0.128*** 0.093* 0.090 0.103* 0.124* (0.052) (0.054) (0.066) (0.058) (0.0755) (0.048) (0.051) (0.060) (0.056) (0.0686) Per capita savings 150 106.8 151.3** 128.5 183.0** 191.1** 97.59 153.2** 115.7 155.1** 191.8*** (109.2) (65.12) (78.10) (77.67) (78.65) (119.4) (66.44) (73.02) (68.27) (69.82) Household size 5.70 0.098 À0.029 0.137 0.086 0.300 À0.037 À0.026 0.015 0.010 0.0667 (0.159) (0.152) (0.224) (0.217) (0.266) (0.077) (0.084) (0.100) (0.097) (0.109) Number of observations 1,774 1,499 1,092 1,080 1,025 1,774 1,499 1,092 1,080 1,025 Number of households 448 379 274 274 257 448 379 274 274 257 B. Vanuatu Per capita income 85,282 42,861*** 44,441*** 48,241*** 24,491*** 55,111*** 29,522* 32,760** 37,717** 17,489* 39,026** (15,201) (15,659) (16,388) (8,291) (15,157) (15,585) (14,938) (15,389) (9,400) (16,267) Log per capita income 10.73 0.320*** 0.301*** 0.364*** 0.310*** 0.516*** 0.186* 0.167 0.267** 0.227* 0.373*** (0.104) (0.107) (0.116) (0.115) (0.126) (0.109) (0.109) (0.121) (0.120) (0.139) Per capita expenditure 65,872 8,495 12,353** 13,020** 9,289* 17,264* 1,093 2,228 2,978 1,213 9,510 (6,590) (6,131) (5,559) (5,195) (8,987) (5,761) (6,217) (5,912) (5,777) (7,744) Log per capita expenditure 10.63 0.240*** 0.261*** 0.254*** 0.227*** 0.413*** 0.134* 0.137* 0.132* 0.125 0.306*** (0.0745) (0.0778) (0.0761) (0.0766) (0.0916) (0.0746) (0.0806) (0.0799) (0.0805) (0.0935) Per capita savings 19,410 34,366** 32,088** 35,222** 15,202** 37847** 28,429* 30,531** 34,739** 16,275* 29,516* (14,460) (13,839) (14,478) (7,195) (15,369) (14,995) (13,809) (14,131) (8,672) (16,663) Household size 4.72 0.0163 À0.141 À0.0995 À0.0691 0.0816 À0.102 À0.265* À0.262* À0.250* À0.232 (0.158) (0.168) (0.179) (0.178) (0.206) (0.133) (0.137) (0.142) (0.141) (0.146) Number of observations 1,574 1,225 977 967 775 1,574 1,225 977 967 775 Number of households 456 360 269 269 215 456 360 269 269 215 All outcomes are converted to six-month values. The subsamples used are (a) all: the full sample; (b) PS-1, the propensity-score screened subsample; (c) PS-2, the propensity-score screened subsample restricted to RSE applicant households only; (d) PS-2 after trimming obser- vations above the 99th percentile for income in the full sample; and (e) nearest N: subsample restricted to RSE households in the common support of PS-2 and the RSE applicant households which are nearest neighbors. Robust standard errors in parentheses, clustered at the house- hold level: ***p < 0.01, **p < 0.05, *p < 0.1. THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 237 238 THE REVIEW OF ECONOMICS AND STATISTICS income is 0.30 to 0.36 log points higher, which is equiva- households sent a worker in one year only, the per capita lent to a 35 to 43 percentage increase. These gains are large per year effect for these households has to be divided by and significant regardless of the estimation method and two. Finally, households also lose both the wage income sample used. and contribution to agricultural production the household Household expenditure per capita is also found to member would have contributed while in New Zealand. increase with participation in the RSE in Tonga. The Working through these calculations, NZ$5,500 in remit- increase is less than the increase in per capita income and, tances and repatriated savings for 1.5 years of participation if we restrict ourselves to the PS-2 screened sample, is sig- equates to a 550 pa’anga semiannual per capita income nificant only after trimming outliers. The log per capita con- increase in Tonga, compared to the 250 to 350 pa’anga sumption results suggest increases of approximately 9% to increase in semiannual per capita income seen in table 3; 10%, just one-third the increase in per capita income. This is and to 32,000 vatu semiannual per capita increase in consistent with some of the additional income being saved, Vanuatu, compared to the 18,000 to 48,000 vatu increase and we see that the flow savings per capita has an increase measured in table 3. They are thus of the same order of approximately equal to the baseline mean; that is, house- magnitude, with differences between the estimated impact holds double their savings. In Vanuatu, semiannual per and the impact calculated from wages and remittances capita expenditure is approximately 12,000 to 13,000 vatu reflecting the opportunity cost of labor that the migrant higher, relative to a baseline of 65,000 vatu, and the effect would have provided (as well as approximation error). on log expenditure is equivalent to approximately a 28% Nevertheless, the gain in income (equivalent to US$260 increase. The increase in savings of 15,000 to 30,000 vatu is per capita per annum in Tonga and US$860 per capita per equal to a doubling or tripling of baseline savings rates. Of annum in Vanuatu) still dwarfs other popular development particular note is that the point estimates for expenditures interventions, which have struggled to generate large gains are smaller with fixed effects than with difference-in- in income. For example, Banerjee et al. (2010) find no differences in Vanuatu, which may reflect attenuation bias increase in average per capita income or expenditure from a due to measurement error.14 microfinance expansion; conditional cash transfer programs Finally, table 3 also shows little change in household size in Nicaragua and Mexico (involving the government giving as a result of participating in the RSE. The only marginally transfers rather than providing a means to generate income) significant changes come from the fixed-effects specifica- have increased per capita incomes by only US$20 to US$40 tion in Vanuatu. This is consistent with our direct questions, (Fiszbein & Schady, 2009); grants to microenterprise own- in which few households reported changing household com- ers in Sri Lanka (de Mel, McKenzie, & Woodruff, 2008) position as an adjustment mechanism to cope with absent and Ghana (Fafchamps et al., 2011) increased microenter- RSE workers. prise incomes for male owners at least, but per capita The median after-tax income earned in New Zealand income gains are in the order of approximately US$20 per reported by the seasonal migrants is approximately year; and recent business training programs for microenter- NZ$12,000.15 This is several multiples of mean annual prises have some effects on revenues in bad months but no household income per capita of RSE households at baseline significant impacts on average incomes (Karlan & Valdivia, of about NZ$1,400 in Tonga and NZ$2,500 in Vanuatu. 2011; Drexler, Fischer, & Schoar, 2011). Bryan, Chowdh- Despite the large increase in income from the RSE, one ury, and Mobarak (2011) examine the impact of grants and might then ask why the increase in per capita incomes is loans to induce seasonal internal migration during the hun- ‘‘only’’ 35%. First, workers face costs in New Zealand from gry season in Bangladesh and find large percentage living expenses (including rent and health insurance) and increases in per capita consumption from a small base, so repaying their share of the airfare. From the NZ$12,000 in monthly per capita consumption increases US$5 in the hun- income, the average worker remitted or brought back an gry season. Compared with all these other interventions, the average of NZ$5,500; it was half remitted and half repa- gains from international seasonal emigration are enormous, triated for Tongans, but just 10% in the form of remittances especially considering that this intervention involves and 90% as repatriated savings for Vanuatu. Second, when removing a policy distortion and so, apart from set-up costs, we consider per capita income, this amount is divided by is free, whereas the others require taxes or aid to fund. 5.7 in Tonga and 4.7 in Vanuatu. Third, we are looking at average impacts over two years, so since just over half the B. Impact on Subjective Well-Being 14 The Vanuatu data are considerably noisier than the Tongan data. The In addition to measuring household welfare through baseline coefficient of variation of per capita income for the RSE house- income and expenditure, our final-round survey measured holds is 0.90 in Tonga compared to 1.40 in Vanuatu, while the correlation in per capita income from one wave to the next for the non-RSE house- subjective well-being. This addresses the second view of holds varies from 0.43 to 0.77 in Tonga, compared to between 0.19 and development impact discussed above—that we should look 0.27 in Vanuatu. There is thus more signal relative to noise in the Tongan at broader measures than income. One adult per household data than in the Vanuatu data. 15 This number accords well with what migrants should have been earn- (the one with the next birthday) was asked to imagine a ten- ing given prevailing wage rates in the sector. step ladder, where on the bottom step were the poorest peo- THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 239 TABLE 4.—IMPACT OF RSE PARTICIPATION ON HOUSEHOLD ASSETS AND SUBJECTIVE STANDARD OF LIVING TWO YEARS LATER Tonga Ever in RSE Vanuatu Ever in RSE Asset PS-1 PS-2 PS-1 PS-2 Subjective standard of living 0.431*** 0.427*** 0.766*** 0.648*** (0.0940) (0.114) (0.202) (0.199) Made any dwelling improvement 0.106*** 0.108** 0.0642 0.0744 (0.0391) (0.0450) (0.0451) (0.0486) Household bank account 0.0956** 0.140*** 0.185*** 0.166*** (0.0373) (0.0462) (0.0562) (0.0607) Made any major asset purchase 0.163*** 0.113** 0.298*** 0.269*** (0.0426) (0.0521) (0.0592) (0.0628) Number of households 372 271 269 225 All asset regressions and bank account regression control for baseline asset levels. Robust standard errors shown in parentheses: ***p < 0.01, **p < 0.05, ***p < 0.1. ple and the top step the richest people, and to state which years earlier). As a result we do not believe our results are step this person thought his or her household was on today the result of negative spillover impacts on the control group. and on which step two years ago. Ravallion and Lokshin (2001) refer to this as an economic ladder question and note that it leaves it up to the individual to define what constitu- C. Impact on Dwelling Improvements and Durable Assets tes ‘‘poor’’ and captures subjective economic welfare. The second row of table 4 shows that Tongan households We estimate equation (3) without including baseline sub- participating in the RSE were 10 to 11 percentage points more jective well-being as a control since it is only measured ex likely to make a dwelling improvement over the period of our post. The results are shown in the first row of table 4.16 In surveys almost double the rate for non-RSE households. Tonga, participating in the RSE is estimated to increase Home improvements were the most commonly mentioned subjective welfare by 0.43 steps on the ladder, about 45% main use of money from the RSE in Vanuatu, but the point of a standard deviation. This effect is strongly significant. estimates suggest that RSE households were just 7 to 8 per- Adding the household’s recalled subjective well-being from centage points more likely to make dwelling improvements, two years earlier only slightly reduces this coefficient, to which is only marginally significant. Renovations are reported 0.36 for the PS-2 screened group, and is still strongly signif- to be much more commonplace in Vanuatu than in our Ton- icant (p < 0.001). Participating in the RSE is estimated to gan survey, with 79% of non-RSE households making a increase subjective welfare by 0.71 to 0.83 steps on the lad- dwelling improvement over the two years of our surveys.17 der in Vanuatu, which is 43% to 50% percent of a standard The impact of the RSE may also then be for households to deviation and strongly significant. Adding the household’s make more substantive changes, such as the transition from recalled subjective well-being from two years earlier does traditional to modern dwellings, which our surveys do not not change these results, with coefficients in the 0.74 to capture since dwelling type was recorded only at baseline. 0.85 range and again strong significance (p < 0.001). Sub- In the baseline survey, 65% of Tongan RSE households jective economic welfare has therefore risen in both coun- reported having a bank account. By the fourth round, this tries for households participating in the RSE. Moreover, the had increased to 83%. This represents a statistically signifi- increase in subjective welfare is of similar magnitude in cant 10 to 14 percentage point increase relative to non-RSE terms of standard deviations as the increases in income: the households over the two-year period (table 4, row 3). This estimated impacts on per capita income in table 3 translates increase in bank account use likely reflects bank accounts to a 0.24 to 0.43 standard deviation increase in per capita being set up for the purpose of household savings rather income in Tonga and 0.31 to 0.47 standard deviation than bank accounts directly being used to receive remit- increase in per capita income in Vanuatu. tances. For the Tongan sample, over 90% of remittances A potential concern here would be if the increased income were made via money transfer operators that did not require of migrant households lowered the subjective well-being of a bank account. In Vanuatu we find the share of RSE house- nonmigrant households due to relative consumption com- holds with a bank account increasing from 55% in the base- parisons. However, subjective well-being did not fall in line to 74% two years later. This is estimated to be a statisti- either country for the control group. In Tonga the PS-1 con- cally significant 17 to 18 percentage point increase relative trol group had essentially stable subjective well-being (4.30 to the non-RSE households over the same period. As in in wave 4, with a recall of 4.26 for two years earlier); in Tonga, it is likely that this increase in household bank Vanuatu subjective well-being rose slightly for the PS-1 account use reflects the use of banks for savings rather than control group (4.62 in wave 4, with a recall of 3.62 to two just to receive remittances. 16 17 Table 4 shows just the PS-1 and PS-2 screened results for reasons of This reflects the much higher proportion of households living in tradi- space. The results using the full sample are similar in terms of both mag- tional (bush material) dwellings in Vanuatu, which have short life expec- nitudes and statistical significance. tancy compared with modern dwellings. 240 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 5.—IMPACT OF THE RSE ON CHILDREN’S SCHOOL ATTENDANCE IN FINAL ROUND SURVEY Aged 6 to 14 at Baseline Aged 15 to 18 at Baseline Full Sample PS-1 PS-2 Full Sample PS-1 PS-2 A. Tongan children Household is ever in the RSE À0.004 À0.004 À0.005 0.129** 0.136** 0.094 (0.004) (0.004) (0.005) (0.059) (0.063) (0.083) Number of observations 478 414 286 167 144 97 Proportion of non-RSE students attending school 0.983 0.984 0.977 0.603 0.599 0.576 B. Ni-Vanuatu children Household is ever in the RSE À0.006 À0.048 À0.044 À0.027 0.027 0.018 (0.048) (0.055) (0.062) (0.107) (0.123) (0.127) Number of observations 337 256 214 101 71 60 Proportion of non-RSE students attending school 0.816 0.817 0.799 0.388 0.403 0.379 Results show regression coefficients after controlling for baseline school attendance and age. Robust standard errors in parentheses: ***p < 0.01, **p < 0.05, *p < 0.1. Each follow-up round asked if the household had made remittance income reported in our surveys, and when asked any major purchase (200 pa’anga or more in Tonga or in the final round survey what the most important use of the 10,000 vatu or more in Vanuatu) of assets since the last sur- money earned in the RSE has been, 40% of Tongans and vey. Row 4 of table 4 shows that Tongan RSE households 28% of ni-Vanuatu said school expenses. The question is were 12 to 15 percentage points more likely to have made whether this translates into higher schooling attainment for such a purchase; double the rate of non-RSE households. children. Schooling attainment is also useful to examine as Ni-Vanuatu RSE households were 27% to 30% more likely a measure of whether parental absence is causing negative to have made such a purchase, which more than doubles the consequences on children. 20% of non-RSE households making such a purchase. In Table 5 shows the impact of being in an RSE household both countries the impact is statistically significant. on school attendance in the final-round survey, conditional In addition to households purchasing assets, RSE workers on baseline school attendance status and age, that is, equa- sometimes returned with durable goods acquired abroad. tion (3). This is carried out for children who were in the DVD players were the most common such asset. Each household at both baseline and follow-up. As we have seen, round of the survey also directly asked about ownership of household size and composition did not significantly certain durable goods, enabling us to capture the net effect change with RSE participation. of purchases, sales, and durable goods that migrants bought In Tonga, schooling is compulsory from ages 6 to 15, back with them. The Tongan RSE households are signifi- and there is near universal school enrollment for children of cantly more likely to have acquired a cell phone, television, these ages. It is therefore no surprise that we see no impact DVD player, and bicycle over the two-year period than of the RSE on children aged 6 to 14 at baseline since over similar non-RSE households (online appendix table 3). The 97% of them are attending school whether or not their ni-Vanuatu RSE households are significantly more likely to household is in the RSE. In contrast, the last three columns have acquired a radio or stereo, a DVD player, a computer, of panel A show large positive effects of the RSE on school a gas or electric oven, and a boat over the two-year period attendance of youth 15 to 18 years old. These effects are than similar non-RSE households. We do not see any signif- statistically significant for the full sample and PS-1 samples icant impact on livestock ownership in either country. and of similar magnitude but not significant in the smaller If nonmigrant households are benefiting from having PS-2 sample (which has less power).18 The magnitude of other households in their village participate in the RSE, we the effect is sizable: a 10 to 14 percentage point increase in should expect them to accumulate more assets. The online the proportion attending school, relative to 60% of youth in appendix examines durable asset changes in the control non-RSE households attending school on average over this group and concludes that there do not appear to be large spill- two-year period. overs, with asset ownership largely stable over two years. In contrast, panel B shows no significant effect of the RSE on school attendance in Vanuatu. One reason may be D. Impact on Children’s Education and Business that starting in 2010, when the final-round survey was in the Ownership field, primary schooling became fully subsidized, whereas In addition to raising household incomes and assets, an previous fees were 7,000 vatu per year (10% of per capita important motive for many RSE participants was the chance average income). Moreover, many schools had allowed stu- to raise money to pay for school fees. In our baseline sur- dents to remain enrolled even with unpaid fees from pre- vey, 85% of Tongan RSE households and 98% of ni- vious years; the main incentive to clearing these debts was Vanuatu RSE households said that earning money to pay 18 for school fees was a very important or somewhat important Moreover, pooling attendance rates over all survey waves, the aver- age impact of the RSE on attendance over the two years is positive and motive for participating in the RSE. In addition, school fees significant even in the PS-2 sample in Tonga (coefficient of 0.19, standard are one of the most common earmarked purposes for RSE error of 0.10). THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 241 to allow students to sit the leaving examinations at the end of the monetary contribution that migrants make to their com- grades 6 and 8. Hence, those RSE workers who report the pay- munities, either through remitting to a community group ment of school fees as a motivation may have been repaying while abroad or contributing some of their repatriated earn- school fee debts, which would not show up in current enroll- ings to this group on return. We asked return migrants how ment except for a possibly higher transition rate to high much they had contributed in this way to the community. school, which also depends on examination performance. The mean response aggregated over the two years was 157 The apparently divergent impacts of RSE participation pa’anga in Tonga and 11,733 vatu in Vanuatu—or approxi- on school enrollment also may reflect the nature of the mately US$80 to $130 per migrant. Our expenditure mod- selection into the RSE in the two countries. In Vanuatu, the ule also collected expenditure on community obligations, households in the RSE are relatively better off, and their but only for a recall period of one month, thereby likely children have higher baseline school attendance rates than missing one-off contributions made by migrants on return. non-RSE households. It is therefore possible that credit con- The difference-in-differences regression then gives a posi- straints were not limiting schooling for this group. In con- tive but insignificant impact on this item. trast, Tongan RSE households were relatively poorer at To further gauge the impacts at the community level, we baseline than non-RSE households, with lower school atten- conducted surveys of community leaders. This was done at dance rates. The extra income earned through the RSE baseline and at the time of the second-round survey in allows them to then catch up to (and surpass) the school Vanuatu, and at baseline and at the times of the third- and attendance rates of the non-RSE group. fourth-round surveys in Tonga. These data are thus less use- Our data do not show any evidence that the RSE has fos- ful for Vanuatu, since they measure only immediate effects tered the development of nonagricultural businesses among while most workers were still away. The Tongan surveys the households in our sample in Tonga. None of the house- reveal the mean (median) community saying it received holds surveyed mentioned investment in a business as a 633 (500) pa’anga from RSE workers, which is consistent main use of the money earned in the RSE, and we do not with the household surveys, given a median of five workers observe any individuals in RSE households starting a new per village participating coupled with the amounts reported business over the two-year period of our surveys. In by workers. The main use of this fund in 83% of cases in Vanuatu we had only five households in the round 4 survey Tonga was funding the village water supply in the first year. say the most important use of the money earned through the In the second year, villages were also using this for street RSE was starting a business or supporting an existing busi- lighting, a school scholarship fund, community halls, and, ness. Given the low population densities and small local in one case, adding Internet to a community hall. markets, it is not clear what the scope for such business Village leaders were directly asked the main benefit and start-ups actually is, but over the first two years of the RSE main disadvantages of the RSE for their community. In policy, there does not appear to be much evidence that it Vanuatu this was only asked in wave 2, six months into the fostered self-employment. Indeed the recent Pacific Futures RSE. The main benefits reported at this stage were job report (World Bank, 2011) cites small market sizes and opportunities for people in the village, money to support long distances to other markets as reasons Pacific Island the village church, and improvements in housing. Disad- economies are unlikely to have much firm growth. vantages were fewer people to do the community work, Our surveys do not directly measure agricultural produc- cases where a worker was not contributing to church or tion techniques or other work skills. However, there are rea- family, and concerns about the potentially bad influence of sons to believe that to date, the RSE is unlikely to have had a alcohol abroad. In Tonga these questions were asked in major impact on the productivity of workers on their own waves 3 and 4, approximately one year and two years into farms. The crops tended to in New Zealand (apples, grapes, the RSE. The main benefit reported at one year is income kiwifruit) are different from those grown in the Pacific for families, along with some saying income for the com- Islands (bananas, yams, cassava, squash), and the climate and munity and church donations. After two years, there are soil conditions are also vastly different. Our surveys asked also a few mentions of improved skills and improved Eng- return workers directly if they had learned any new skills. lish of workers and positive impacts on school enrollments. The main one workers mentioned was pruning plants— When asked the main disadvantage, more than half at one important in the New Zealand fruit industries but not for the year, said none, the main other answer being family separa- crops they produce. Absent any short-run evidence on pur- tion. At two years, one-third said family separation; 30% chases of livestock or farm machinery, it therefore seems said less labor available for the village, church, and com- unlikely that home production skills or technology have munity projects; and about 15% said fewer members for improved dramatically from participation in this program. church activities. Online appendix table 5 summarizes the results of ques- VI. Community-Level and Country-Level Impacts tions in both the household and community leader surveys intended to measure qualitatively the impressions of the Finally we consider the broader impacts of the RSE on broader community-level impacts of the RSE. The RSE the sending communities. The most direct impact is through workers themselves believe that participation in the RSE 242 THE REVIEW OF ECONOMICS AND STATISTICS TABLE 6.—THE BIG PICTURE: NET IMPACTS AT THE MACROLEVEL led to sizable increases in income in both Tonga and Tonga Vanuatu Vanuatu. Under a broader utility-based view of develop- Number of RSE workers in New Zealand, 1,971 3,590 ment, we find subjective standards of living have improved, 2007–2008 and 2008–2009 households are spending more, and community leaders Net income gain to country from first 5,333,526 9,714,540 view the policy as having an overall positive impact. two years of program (NZ$), Recognised Seasonal Employer program Finally, under a view that what matters is whether the New Zealand bilateral aid received 12,700,050 20,701,000 income gained from this program is spent or invested in in 2009–2010 ways that will continue to benefit the household in the Australian bilateral aid received 20,664,000 56,088,000 in 2009–2010 longer term, we find increases in household durable asset Total export earnings 2008 11,340,600 43,296,000 accumulation, increases in savings, and, in Tonga, some Sources: New Zealand bilateral aid from MFAT annual report 2009–2010, http://www.mfat.govt.nz evidence that it improved child school attendance for older /downloads/media-and-publications/annual-report/annualreport09-10.pdf; Australian bilateral aid from Ausaid annual report 2009–2010, http://www.ausaid.gov.au/anrep/rep10/pdf/anrep09-10entirereport.pdf; children. However, we do not find evidence of increased export statistics from Prism: http://www.spc.int/prism/trade-export-values-us-000. self-employment or substantial accumulation of skills that can increase home productivity. This reflects the limited either improved or left unchanged their family and commu- market sizes in the participating countries and is consistent nity life. Non-RSE households in Tonga also saw benefits with the view that migration is likely to be a long-term part in terms of community life, availability of paid jobs,19 and of the economic organization of Pacific Island economies schooling opportunities. To the extent that such benefits are (World Bank, 2011). really accruing to non-RSE households, our estimates com- These results make this seasonal migration program one paring RSE to non-RSE households will be a lower bound of the most effective development interventions for which on the positive development impact of the RSE program. rigorous evaluations are available. In addition, although Non-RSE households in Vanuatu were more likely than there has been nontrivial investment by both the New Zeal- those in Tonga to say there had been no change in commu- and and Pacific governments in the set-up phase, it does not nity life or in job or schooling opportunities. Finally, com- involve grants and appears to be benefiting both private munity leaders were asked their assessment of the overall employers and workers. The design features of the program impact of the RSE on their communities. In Tonga 92% of and the low rate of overstaying have already led to this pol- leaders said that it has had positive effects after two years, icy being heralded as international best practice. The large and in Vanuatu, even at six months, 72% of leaders said the development impacts seen here should lead other countries overall impact was positive. to consider similar policies. Finally, table 6 shows that the overall development Nevertheless, there are several caveats to these conclu- impact of the RSE program is important for Tonga and sions. The first is that development is a long-term process, Vanuatu. We take the per worker estimates of the average and some of the effects of the RSE may materialize only impact of the program on household income over the first over many years of community involvement. These could two years and scale up by the total number of workers hired include positive effects such as greater asset building, from each country to get total positive development impacts investments, and skill development if workers return for of NZ$5.3 million in Tonga and NZ$9.7 million in many seasons, and potential longer-term negative effects of Vanuatu. This amount is equivalent to 42% to 47% of total continual absence of family members on family and com- annual bilateral aid from New Zealand to these countries munity relations. Second, although the gains to households and is still nontrivial when compared to bilateral aid from from this seasonal migration are large, they still pale in Australia, the other main donor. Moreover, this aggregate comparison to the gains from permanent international development impact of the RSE program is equivalent to migration (Clemens et al., 2008; McKenzie et al., 2010). A almost 50% of annual export earnings for Tonga and 25% key policy issue is therefore the extent to which seasonal of annual export earnings for Vanuatu. migration can or cannot eventually open up avenues for per- manent migration. VII. Conclusion New Zealand’s RSE program was designed with promot- REFERENCES ing development in the Pacific Islands as an explicit goal. Angrist, Joshua, and John Johnson, ‘‘Effects of Work-Related Absences The results of our multiyear prospective impact evaluation on Families: Evidence from the Gulf War,’’ Industrial and Labor show that it has largely achieved this goal. We consider Relations Review 54(2000), 41–58. Angrist, Joshua, and Jo ¨ rn-Steffen Pischke, Mostly Harmless Econo- three views of development impact and find evidence to metrics: An Empiricist’s Companion (Princeton, NJ: Princeton support impacts under each view. Under a narrow view of University Press, 2009). impact as increasing household incomes, we find the RSE Banerjee, Abhijit, Esther Duflo, Rachel Glennerster, and Cynthia Kinnan, ‘‘The Miracle of Microfinance? Evidence from a Randomized Eva- luation,’’ mimeograph MIT (2010). 19 Our impression is that this response refers largely to the availability Basok, Tanya, ‘‘Migration of Mexican Seasonal Farm Workers to Canada of paid jobs in New Zealand for members of the community rather than and Development: Obstacles to Productive Investment,’’ Interna- more jobs in the village. tional Migration Review 34 (2000), 79–97. THE DEVELOPMENT IMPACT OF A BEST PRACTICE SEASONAL WORKER POLICY 243 Borjas, George, ‘‘A Lemon in the Senate: The Immigration Deal Is a Tra- on Children’s Academic Achievement,’’ Journal of Labor Eco- vesty of a Mockery of a Sham,’’ National Review Online, May 17, nomics 24 (2006), 319–350. 2007, http://www.hks.harvard.edu/fs/gborjas/Papers/A%20Lem- Macours, Karen, and Renos Vakis, ‘‘Seasonal Migration and Early Child- on%20in%20the%20Senate,%20NRO.pdf. hood Development,’’ World Development 38 (2010), 857–869. Bryan, Gharad, Shyamal Chowdhury, and Mushfiq Mobarak, ‘‘Seasonal McKenzie, David, ‘‘Beyond Baseline and Follow-Up: The Case for More Migration and Risk Aversion,’’ mimeograph, Yale School of Man- T in Experiments,’’ Journal of Development Economics 99 (2012), agement (2011). 210–221. Clemens, Michael, Claudio Montenegro, and Lant Pritchett, ‘‘The Place McKenzie, David, John Gibson, and Steven Stillman, ‘‘How Important Is Premium: Wage Differences for Identical Workers across the U.S. Selection? Experimental vs. Non-Experimental Measures of the Border,’’ Center for Global Development working paper 148 Income Gains from Migration,’’ Journal of the European Eco- (2008). nomic Association 8 (2010), 913–945. Crump, Richard, V. Joseph Hotz, Guido Imbens, and Oscar Mitnik, McKenzie, David, Pilar Garcı ´a Martinez, and L. Alan Winters, ‘‘Who Is ‘‘Dealing with Limited Overlap in Estimation of Average Treat- Coming from Vanuatu to New Zealand under the New Recognised ment Effects,’’ Biometrika 96 (2009), 187–199. Seasonal Employer (RSE) Program?’’ Pacific Economic Bulletin Dehejia, Rajeev, ‘‘Practical Propensity Score Matching: A Reply to Smith 23 (2008), 205–228. and Todd,’’ Journal of Econometrics 125 (2005), 355–364. New Zealand Department of Labour, Final Evaluation Report of the De Mel, Suresh, David McKenzie, and Christopher Woodruff, ‘‘Returns Recognised Seasonal Employer Policy (2007–2009) Wellington: to Capital in Microenterprises: Evidence from a Field Experi- Department of Labour (2010), http://dol.govt.nz/publications ment,’’ Quarterly Journal of Economics 123 (2008), 1329–1372. /research/rse-evaluation-final-report/rse-final-evaluation.pdf. Drexler, Alejandro, Greg Fischer, and Antoinette Schoar, ‘‘Keeping It OECD, International Migration Outlook—SOPEMI—2008. (Paris: Simple: Financial Literacy and Rule of Thumbs,’’ mimeograph, OECD, 2008.) London School of Economics (2011). Paxson, Christina, ‘‘Using Weather Variability to Estimate the Response Dunn, Lucia F., ‘‘Is Combat Pay Effective? Evidence from Operation of Savings to Transitory Income in Thailand,’’ American Eco- Desert Storm,’’ Social Science Quarterly 84 (2003), 344–358. nomic Review 82 (1992), 15–33. Fafchamps, Marcel, David McKenzie, Simon Quinn, and Christopher Pritchett, Lant, Let Their People Come: Breaking the Gridlock on Global Woodruff, ‘‘Female Microenterprises and the Fly-Paper Effect: Labor Mobility (Washington, DC: Center for Global Development, Evidence from a Randomized Experiment in Ghana,’’ mimeo- 2006). graph, World Bank (2011). Ramasamy, Sankar, Vasantha Krishnan, Richard Bedford, and Charlotte Fiszbein, Ariel, and Norbert Schady, ‘‘Conditional Cash Transfers: Redu- Bedford, ‘‘The Recognised Seasonal Employer Policy: Seeking the cing Present and Future Poverty’’ (Washington, DC: World Bank, Elusive Triple Wins for Development through International Migra- 2009). tion,’’ Pacific Economic Bulletin 23 (2008), 171–186. Gibson, John, David McKenzie, and Halahingano Rohorua, ‘‘How Pro- Ravallion, Martin, and Michael Lokshin, ‘‘Identifying Welfare Effects Poor Is the Selection of Seasonal Migrant Workers from Tonga from Subjective Questions,’’ Economica 68 (2001), 335–357. under New Zealand’s Recognised Seasonal Employer (RSE) Pro- Rodrik, Dani, ‘‘Be Our Guest,’’ New York Times, June 1, 2007, http:// gram?’’ Pacific Economic Bulletin 23 (2008), 187–204. www.nytimes.com/2007/06/01/opinion/01rodrik.html. Global Commission on International Migration, Migration in an Intercon- Ruhs, Martin, ‘‘The Potential of Temporary Migration Programmes in nected World: New Directions for Action (Geneva: Global Com- Future International Migration Policy,’’ International Labour mission on International Migration, 2005). Review, 145 (2006), 7–35. Hanson, Gordan, ‘‘The Economic Consequences of the International Southern Poverty Law Center, Close to Slavery: Guestworker Programs Migration of Labor,’’ Annual Review of Economics 1 (2009), 179– in the United States (Montgomery, AL: Southern Poverty Law 207. Center, 2007). House of Commons International Development Committee, Migration United Nations, World Economic and Social Survey (New York: United and Development: How to Make Migration Work for Poverty Nations, 2004). Reduction 2004, http://www.publications.parliament.uk/pa/cm200304 Winters, L. Alan, Terrie Walmsley, Wang Zhen Kun, and Roman Grynberg, /cmselect/cmintdev/79/79.pdf. ‘‘Liberalizing Temporary Movement of Natural Persons: An Agenda Karlan, Dean, and Martin Valdivia, ‘‘Teaching Entrepreneurship: Impact for the Development Round,’’ World Economy 26 (2003), 1137–1161. of Business Training on Microfinance Institutions and Clients,’’ World Bank, At Home and Away: Expanding Job Opportunities for Paci- this REVIEW, 93 (2011), 510–527. fic Islanders through Labour Mobility (Washington, DC: World Lee, David, ‘‘Training, Wages, and Sample Selection: Estimating Sharp Bank, 2006). Bounds on Treatment Effects,’’ Review of Economic Studies 76 ——— Migration and Remittances Factbook 2008 (Washington, DC: (2009), 1071–1102. World Bank, 2008). Lyle, David, ‘‘Using Military Deployments and Job Assignments to Esti- ——— ‘‘Pacific Futures’’ World Bank Pacific Department discussion mate the Effect of Parental Absences and Household Relocations note (2011).