# In a nutshell

This page discusses conditional subsidies (cash transfers or limited-liability interest-free loans) to support seasonal labor migration in northern Bangladesh. Low-income rural households primarily engaged in agricultural wage-labor in northern Bangladesh often experience a seasonal decrease in consumption in the months between planting and harvest. One strategy for smoothing consumption in this "lean season" is for a working member of the household to temporarily migrate elsewhere in Bangladesh, engage in wage labor, and remit money back home. Small conditional subsidies are intended to induce additional migration and lead to increased household income and consumption.

Several randomized controlled trials of conditional subsidies to increase migration have been conducted in northern Bangladesh. They have found mixed results on whether small conditional subsidies lead to increased migration. Among trials that found an impact on migration rates, there is moderately strong evidence that migration increases household income and consumption during the lean season.

Research on this program is ongoing. A randomized controlled trial is planned for the 2018 lean season. Given the mixed evidence to date, we do not currently consider conditional subsidies for seasonal labor migration to be a priority program.

Published: November 2017; Last updated: November 2018

• November 2017 version

## What is the lean season?

In some contexts, low-income people experience seasonal variation in consumption resulting in a time of especially low consumption (lower income and/or higher prices) during a predictable portion of the year. This "lean season" may also be referred to as a "hungry season" or "seasonal famine." In Bangladesh, the lean season is referred to as monga. Lean season cycles have been documented in contexts in South Asia and sub-Saharan Africa.1

In rural northern Bangladesh, the lean season is related to scarcity of agrarian work opportunities and to higher grain prices between planting and harvest of the major rice harvest.2 The lean season in Bangladesh traditionally refers to the period preceding the aman rice harvest (approximately September to November), although the timing of the lean season may vary by local rain and soil characteristics.3

## What is seasonal labor migration?

In rural northern Bangladesh, seasonal labor migration is a fairly commonly practiced strategy for coping with decreased local labor opportunities during the lean season. In villages of rural northern Bangladesh participating in the studies that inform our view of this intervention, approximately 37% of poor households send a working member to earn wages away from their home village during the lean season each year, without the intervention of any program supporting seasonal labor migration.4

Seasonal labor migrants may vary in terms of the duration of migration episodes, number of migration episodes, migration destination, and labor sector at the migration destination.5 Examples of migration employment include rickshaw pulling, construction, and agricultural labor.6 A seasonal labor migrant is almost always male, the household head, and the only migrant from the household.7 Approximately 60% of migration episodes involve group travel rather than traveling alone.8 Our impression is that some migrants arrange employment in advance of migration, while others search for employment once they have arrived at their destination.9 Households engage in seasonal labor migration at times other than the monga lean season, such as in the "lesser lean season" in approximately February to early March after the planting of the boro rice crop.10

## What are conditional subsidies for seasonal labor migration?

This report discusses the effect of two types of conditional subsidies for seasonal labor migration on household income and consumption: conditional cash and conditional limited-liability interest-free loans.

Conditional cash transfers for seasonal labor migration are small cash transfers (about the cost of round-trip bus travel for one migrant)11 provided conditional on a household member agreeing to migrate for at least some fixed amount of time (such as at least 7 days) in a given lean season. Conditional interest-free loans for seasonal labor migration are small, limited-liability,12 interest-free loans (about the cost of round-trip bus travel for one migrant) provided conditional on a household member agreeing to migrate for at least some fixed amount of time (such as at least 7 days) in a given lean season, and repaid after the migrant returns.

## What is the evidence that conditional subsidies for seasonal labor migration lead to increased migration?

Four studies have investigated the effect of conditional subsidies for labor migration during the lean season on rates of migration. Two of the studies found significant effects on migration rates and two did not.

• A 2008 RCT involving 1,900 households in 100 villages in northern Bangladesh included four arms: a conditional cash subsidy, a conditional interest-free loan subsidy, an information-only intervention, and a control group.13 In each village, a small proportion (approximately 14%) of the eligible population was targeted with the intervention.14 Migration rate in 2008 was 36.0% in the control group, 35.9% in the information-only group, 56.8% in the credit group, and 59.0% in the cash group. Pooling the credit and cash arms into an "incentivized" group and comparing this to the pooled information-only and pure control arms pool (the "non-incentivized" group), the migration rate in the incentivized group was 22 percentage points higher than in the non-incentivized group (58.0% vs 36.0%, p<0.01).15
• We are aware of a 2013 study involving 7,638 households in 133 villages in northern Bangladesh which included five arms: either job leads or credit (a no-interest loan covering approximately the cost of round-trip travel) offered to either 10% or 35% of the eligible population in villages, and a control arm.16 Results from this study are not published. Our understanding is that the researchers consider the study in that year to have failed, possibly due to political factors in Bangladesh in 2013 which may have interfered with migration in that year.17 We find this explanation for the 2013 study's apparent failure to induce migration plausible; however, we have not thoroughly investigated it, and our understanding is limited by the limited degree of analysis available on outcomes of the 2013 study.18
• A 2014 study involving 5,792 households in 133 villages in northern Bangladesh included three arms: conditional cash transfers offered to approximately 14% of the eligible population, to approximately 70% of the eligible population, and a control arm.19 Grants were disbursed in the later part of the lean season, and endline data collection took place in April 2015.20 In the control group, 34.2% of households sent a migrant. In villages where about 14% of the eligible population received a conditional cash transfer offer (comparable to the 2008 study above), 59% of households who received an offer participated in migration (24.8 percentage point difference, p<0.01). In villages where about 70% of the eligible population received a conditional cash transfer offer, 74% of households who received an offer participated in migration (39.8 percentage point difference, p<0.01). The difference in uptake in the two treatment arms is statistically significant (p<0.001)21
• A 2017 RCT, which evaluated the program when it was implemented at a larger scale than in the previous RCTs (40,557 households received loans) and in a manner that was intended to resemble how the program will be implemented in the future (see our review of No Lean Season), did not find an impact of conditional subsidies on rates of migration. We have reviewed preliminary, unpublished results. Further analysis, which the study authors expect to complete and share at a later date, may help explain what led to this result.22

There will also be an RCT conducted of the program in the 2018 lean season.23

### A note about the 2017 RCT results

Evidence Action, one of the partners implementing this program (more below), is exploring, with further analysis of the 2017 results and through further research, a range of hypotheses that might explain the lack of impact on migration rates in 2017, including whether it was due to ways in which the program was implemented that could be altered in subsequent seasons.

In particular, Evidence Action believes it is possible that adapting the program design to increase loan disbursement capacity early in the lean season, removing the use of delivery targets, and other changes to the implementation protocols could lead to greater impact on migration rates by ensuring that a greater proportion of eligible households have access to subsidies (particularly households from which someone would not have migrated without a subsidy).24 The program as designed includes up to three visits by program staff to eligible households to provide information about the program and to ask them if they would like to participate.25 If second and third visits are skipped, it is possible that the program will disproportionately enroll households who would have sent a migrant with or without the subsidy. Households that are induced to send a migrant by the program may be those that are less likely to take the first offer but choose to take the subsidy at the second or third offer.

While it is unclear if this is an important mechanism for inducing higher migration rates, there is suggestive evidence that program acceptance would have been higher in 2017 if program staff had followed up with a higher proportion of households that did not accept the first loan offer and/or followed up with these households earlier in the season. In particular:

• 31% of households accepted the first offer to participate in the program, 29% indicated that they were interested, and 41% indicated that they were not interested in the program.26 Initial offers were made between August 11, 2017 and October 29, 2017. Households made an offer earlier in the season were slightly more likely to accept.27
• Of households that indicated that they were interested in the offer but not immediately ready to accept, 60% were visited a second time in 2017.28 Of these households, 34% accepted the offer to participate in the program when asked a second time, while 3% were still interested but undecided, and 64% declined the offer to participate in the program.29 Of households that indicated during their first offer that they were not interested in the program, 20% were reached with a second offer.30 Among these, 60% accepted this second offer to participate in the program, 6% indicated interest in the program, and 35% declined the offer.31 This large proportion of initially uninterested households who, on a second visit, decide that they are interested in the program suggests that re-visiting a larger proportion of households who initially decline may increase participation in the program. Households made a second offer earlier in the season were much more likely to accept the offer, and it appears that in 2017 most (64%) of No Lean Season's second offers were made in November and December, when the acceptance rate had declined considerably.32
• Of households which received a second offer and indicated that they were interested in the program but not ready to accept at the time, 45% were visited a third time.33 Of households which received a second offer and indicated that they were not interested in the program, 6% received a third visit.34 Upon receiving a third visit, 24% of households accepted the offer to participate in the program, 8% said that they were interested, and 68% declined to participate in the program.35 2% of third visits were made prior to November 2017, 28% in November 2017, 64% in December 2017, and 5% after December 2017.36

We have not yet compared rates and timing of follow up with eligible households in 2017 with those in 2008 or 2014 (in which studies found that the program induced higher rates of migration).

Evidence Action provided its views on the above analysis as part of its comments on this page; see the following footnote for details.37

## What is the evidence that conditional subsidies for seasonal labor migration lead to increased consumption, given that they induce additional migration?

This section discusses evidence for the impact of conditional subsidies for seasonal labor migration on consumption when those subsidies successfully induce seasonal migration. We believe there is strong evidence from two randomized controlled trials (RCTs) that, when conditional subsidies for seasonal labor migration lead to increased migration, this results in greater income and consumption in the season of intervention and in the lean season one year after intervention. We believe there is weaker evidence from one RCT that such benefits persist for additional years and in lesser lean seasons.

### What is the evidence that conditional subsidies for seasonal labor migration lead to increased income or consumption in the year of intervention?

The two studies discussed above, from 2018 and 2014, that found increased rates of migration due to conditional subsidies, found that migration led to significant effects on consumption or income.

• In the 2008 RCT, the cash and credit arms experienced increased household consumption (food and non-food) and calories consumed in the second month of the lean season;38 total consumption in the pooled incentivized group was 7% higher than in the pooled non-incentivized group (p<0.05).39
• In the 2014 RCT, households in the control group earned about 23,903 taka in total income over a five-month recall period, while households offered a subsidy in villages where about 14% of eligible households were offered the subsidy earned on average an additional 2,589 taka (11% increase, p<0.01), and households offered a subsidy in villages where about 70% of eligible households were offered the subsidy earned on average an additional 2,105 taka (9% increase, p<0.05).40 We note that self-reported wages over a five-month recall period may be prone to bias. However, we do not have reason to believe this outcome may be biased more in one direction than the other.

### What is the evidence that conditional subsidies for seasonal labor migration lead to increased income or consumption in future lean seasons?

Follow-up surveys to the 2008 RCT of conditional subsidies (both cash transfers and no-interest loans) for seasonal labor migration conducted in 2009, in the lesser lean season of 2011, and in 2013 found some evidence of a persistent effect of the intervention on migration and consumption, without further intervention. These effects were statistically significant in 2009 (the year after intervention), but consumption effects were no longer statistically significant in 2011 and 2013. A follow-up survey to the 2014 RCT of conditional cash transfers for seasonal labor migration observed a large and statistically significant persistent effect on migration rates and migration income in 2015.

• A 2009 follow-up survey to the 2008 intervention assessed to what extent migration rates and consumption remained higher a year after treatment, without any further subsidy.41 This follow-up survey found that while the 2009 migration rate in control villages was 40.5%, the migration rate among households offered cash subsidies in 2008 was 44.6% (4.1 percentage point difference), and among households offered credit subsidies in 2008 was 49.1% (8.6 percentage point difference).42 2009 total consumption in the non-incentivized group (pooled control group and information-only group) was 1196.01 takas/person/month, and 64.992 takas/person/month higher (5% higher, p<0.01) among households offered either a cash or credit subsidy in the previous year.43
• We view the 2011 follow-up in the lesser lean season to the 2008 intervention as contributing evidence that there is some persistent re-migration effect without further intervention. In a July 2011 follow-up with households offered subsidies in 2008 and never since, 32% of households in the 2008 non-incentivized group (pooled control group and information-only group) migrated in the 2011 lesser lean season, while 39% of households in the 2008 incentivized group (pooled cash group and credit group) migrated in the 2011 lesser lean season, a 7 percentage point difference significant at p<0.05. Consumption results of this 2011 follow-up are not published; however, we have seen analysis showing that the pooled non-incentivized group had an average consumption of 1780.2 takas/person/month in the 2011 follow-up survey, and the incentivized group consumed on average 85.0 takas/person/month more (5% higher, not significant at p<0.10).44
• We have seen unpublished analysis of a 2013 follow-up survey to the 2008 intervention, showing 6% higher consumption (not statistically significant) in households offered subsidies in 2008 (five years prior, and never since) compared to households not offered subsidies in 2008 (and never since).45 We have not seen analysis of migration rates from this survey.
• 2015 follow-up to 2014 intervention:46 In control villages, the 2015 migration rate was 37.8%. In villages where in 2014 about 14% of the eligible population was offered a conditional cash transfer, the 2015 migration rate among those offered the 2014 subsidy was 56.6% (18.8 percentage points higher, p<0.01). In villages where in 2014 about 70% of the eligible population was offered a conditional cash transfer, the 2015 migration rate among those offered the 2014 subsidy was 67.1% (29.3 percentage points higher, p<0.01).47 In control villages, average migration-only income in 2015 was 9,204.65 taka, while average migration-only income for households offered cash subsidies in 2014 low-intensity treatment villages was 5,392 taka higher (59% higher, p<0.01) and average migration-only income for households offered cash subsidies in 2014 high-intensity treatment villages was 7,500 taka higher (81% higher, p<0.01).48 Note that these income figures include only income earned while migrating; they do not include income earned at home, and hence do not represent a total income effect. Because we expect that migration displaces some income that otherwise would have been earned at home, we expect that migration-only income results are larger than net income results or total consumption results. We note that it appears that migration income increased in 2015 compared to 2014, including in control villages.49

### What is the evidence that conditional subsidies for seasonal labor migration lead to increased income or consumption in future lesser lean seasons?

One follow-up survey to the 2008 RCT contributes evidence that subsidies may lead to increased migration and increased consumption in future lesser lean seasons, without further intervention. While the effect on consumption in the lesser lean season three years after intervention is not statistically significant, we expect that there may be a larger effect in the first and second lesser lean seasons after the intervention.

A July 2011 follow-up survey of households participating in the 2008 RCT and not offered subsidies since found a persistent, statistically significant seven percentage point difference in migration rates between the 2008 non-incentivized group and the 2008 incentivized group two and a half years later and during a lesser lean season.50 We have seen unpublished analysis of household consumption data collected in this follow-up survey showing 5% higher consumption (not statistically significant) two and a half years later in households offered subsidies in 2008, compared to households not offered subsidies in 2008.51

This effect in the third lesser lean season after intervention leads us to expect that the program led to at least a 5% increase in consumption in the first and second lesser lean seasons after intervention, since we guess that the effects of the intervention decrease with time.

## Are there negative impacts of the intervention?

In this section, we discuss possible negative effects of inducing migration in this context. Our best guess, based on limited evidence, is that the potential negative impacts of the intervention do not meaningfully offset the benefits of the intervention.

### Migrants' quality of life

No study of this intervention has attempted to measure the quality of life of migrants while they are migrating compared to while they are at home during the lean season. We speculate that there may be discomforts associated with the migration experience, such as lower-quality housing, a less familiar environment, or worry about how the household is faring in the migrant's absence. Our concerns about discomforts associated with migration are somewhat mitigated by observed rates of remigration without further incentive (more).

### Quality of life for nonmigrating household members

No study of this intervention has attempted to measure the quality of life of non-migrating household members while a member of the household is migrating, compared to quality of life when all household members are at home during the lean season. We are uncertain about the speculative ways in which quality of life for non-migrating household members may be greater or lesser when the migrant (usually the male head of household) is away.

We have a small amount of information about the impact of migration on intrahousehold relationships. In the 2014 RCT, women were asked whether they feel that their partner's migration impacted their relationship negatively, positively, or had no impact. Among women whose partners had migrated during the lean season (50%), 82% responded that they felt the migration had no impact on their relationship. 18% responded that they felt the migration had a positive impact on their relationship, and less than 1% responded that they felt the migration had a negative impact on their relationship.52

### Increased price of fish protein

The aforementioned 2014 RCT included an eight-week high-frequency survey53 of food prices which found a small but statistically significant increase in the price of fish.54 The survey also found small but statistically significant effects on the prices of prepared beverages (decrease in price), an index of 12 goods, and edible oil.55 It seems plausible that these small changes in prices are due to increased incomes leading to increased demand for protein, and increased male migrant departure leading to decreased demand for prepared beverages (tea and snacks in village tea shops).56 Given the magnitude of the observed effects of the intervention on income, we do not consider small increases in food prices to meaningfully offset the benefits of the intervention.

### Effects on destination labor markets

No study of this intervention has investigated the effect on destination labor markets.57 It is possible that increasing low-skill labor migration could depress wages in destination labor markets. It is also possible that destination labor markets are large enough to absorb this level of increased labor supply without wage changes or displacement of wages earned by local low-skill laborers.

## Are there additional positive impacts of the intervention?

There is some evidence for positive impacts of the intervention on wages at origin villages. We are uncertain about the degree to which the intervention may have additional positive impacts.

### Increased wages in origin villages

The aforementioned 2014 RCT included an eight-week high-frequency survey58 of earnings, time worked, and prices which found suggestive evidence that increasing seasonal labor migration rates increased the male agricultural wage rate in villages, as well as available work hours in the village, increasing income earned in the village.59 We have not closely examined these results. We aim to include such effects in our cost-effectiveness analysis of conditional subsidies for seasonal labor migration by using total income earned (at home and while migrating) as the metric of the intervention's benefit.

### Reduced exposure to domestic violence

The 2014 RCT, and a follow-up survey conducted in November of 2016, included survey questions related to gender and other social norms.60 This study did not find evidence of lasting positive or negative impacts on gender and social norms.61

The study also asked women about their experiences with threats, verbal abuse, and physical abuse by their husband or another family member. In the six months preceding the survey, 4% of women reported that their husbands threatened them with divorce or taking another wife, or acted on that threat. 11% of women reported physical abuse. 57% of women reported verbal abuse.62 There was no difference in the prevalence of reported abuse by women whose households received the intervention compared to women in control villages.63 This study did not explore whether the intervention leads to a difference in frequency of incidents of domestic abuse. We find it intuitively plausible that increased seasonal labor migration could decrease the frequency of women's (and potentially children's) exposure to domestic abuse during the lean season.

The 2016 follow-up survey asked male heads of household about decision-making in the previous (2015-2016) lean season. While previous survey results suggested that household assignment to treatment or control did not affect men's and women's overall decision-making spheres, this survey observed that men reported that women's decision spheres were temporarily expanded while the male migrant was temporarily away from home.64

### Developmental benefits to young children

We are aware of, but have not reviewed, literature on the effects of seasonal variability in consumption, or seasonal migration, on child development in contexts outside of Bangladesh.65 We do not currently include in our cost-effectiveness analysis any benefits to early childhood development due to increased lean season consumption.

### Consumption effects throughout the year

The effect of conditional subsidies for seasonal labor migration on consumption or income has been measured during the lean season, but not throughout the year. It is possible that the intervention has an effect on consumption outside of the lean season. For example, it may lead to decreased reliance on high-interest loans during the lean season, reducing expenditure on interest payments after the lean season.

## Who is working on the program?

No Lean Season, a program of RDRS Bangladesh (http://www.rdrsbangla.net/) and Evidence Action Beta (https://www.evidenceaction.org/evidence-action-beta/), provides conditional interest-free loans to low-income rural agricultural workers during the lean season in rural northern Bangladesh. See GiveWell's review of No Lean Season, and all GiveWell content on No Lean Season. We believe that No Lean Season is the only program providing conditional subsidies for seasonal labor migration in northern Bangladesh.

## How cost-effective is the program?

In 2017, we created a model of the cost-effectiveness of No Lean Season. With the information available at the time, our model estimated that No Lean Season was in the range of cost-effectiveness of our other top charities.66 A previous version of this report discussed major judgement calls and uncertainties in our model.

We have not completed an update of this model in 2018 to incorporate the 2017 RCT results and updated cost estimates. Depending on the outcome of research in 2018-2019 on the program's impact on migration rates and household income and consumption, we may fully update our cost-effectiveness analysis in 2019.

## Is there room for more funding?

See the relevant section of our review of No Lean Season for the most recent information on room for more funding for conditional subsidies for seasonal labor migration in northern Bangladesh.

In April 2013, we published a shallow investigation (a brief look at an area that we use to decide how to prioritize further research) of seasonal migration. In March 2014, GiveWell recommended that Good Ventures grant $250,000 to Evidence Action to support the development of additional programs. Evidence Action used these funds to support the No Lean Season program in Bangladesh. We recommended additional grants to support the growth of No Lean Season in March 2015, March 2016, and December 2016. This evaluation of and recommended support to No Lean Season was part of GiveWell Incubation Grants, with the aim of supporting the development of a potential future GiveWell top charity. We added No Lean Season to our list of top charities in November 2017. In mid-2018, Evidence Action shared preliminary results from the 2017 RCT with us. We and Evidence Action agreed that No Lean Season should not be considered a GiveWell top charity in 2018, and Evidence Action told us that it was not seeking additional funding for No Lean Season at the time and would spend available funding implementing the program and conducting further research in 2019-2020. We updated this report and our review of No Lean Season to reflect this information. See also all content on No Lean Season. ## Questions for further investigation • Will the program have an impact on migration rates when operated at a large scale? How consistent will these results be? How cost-effective will the program be when operated at a large scale? • Are there contexts outside of northern Bangladesh where conditional subsidies for seasonal labor migration would be cost-effective? • Can we achieve a better understanding of the mechanisms by which this intervention works? What are the pre-existing barriers to seasonal labor migration and how does a small conditional subsidy overcome these barriers? • What effect on income and consumption do households experience in years after intervention, and what effect do they experience in lesser lean seasons? • Do intervention households experience higher income or consumption throughout the year? (For example, does the program reduce their loan interest burden at other times of the year?) • Does the intervention's effect on seasonal income and consumption affect children's development? • In terms of quality of life, how does the experience of migrating compare to the experience of lean season at home, for the migrant? For household members not migrating, how does the experience of having a household member migrate compare to not having a household member migrate? • What, if any, are the economic effects of induced migration on destination economies? ## Sources Document Source Ahsan and Iqbal 2016 Source (archive) Akram, Chowdhury and Mobarak 2017 Source (archive) Bryan, Chowdhury and Mobarak 2014 Source (archive) Chowdhury and Mobarak 2016 note regarding 2013 hartals, email correspondence 19 October 2016 Source Khandker 2012 Source Khandker and Mahmud 2012 Source (archive) Macours and Vakis 2010 Source Mobarak, Reimão, and Shenoy 2017 Source No Lean Season Preliminary 2016 Data Unpublished No Lean Season Research Timeline Source • 1. "Seasonal poverty has been documented in Ethiopia (Dercon and Krishnan (2000)), where poverty and malnourishment increase 27% during the lean season, Mozambique and Malawi (Brune, Gine, Goldberg, and Yang (2011)), where people refer to a “hungry season,” Madagascar, where Dostie, Haggblade, and Randriamamonjy (2002) estimated that 1 million people fall into poverty before the rice harvest, Kenya, where Swift (1989) distinguished between years that people died and years of less severe shortage, Francophone Africa (the soudure phenomenon), Indonesia (Basu and Wong (2012)) (‘musim paceklik’ or ‘famine season’ and ‘lapar biasa’ or ‘ordinary hunger period’), Thailand (Paxson (1993)), India (Chaudhuri and Paxson (2002)), and inland China (Jalan and Ravallion (2001))." Bryan, Chowdhury and Mobarak 2014, pp. 1-2 (footnote). • 2. • For background on seasonal variation in consumption in rural northern Bangladesh, see Khandker 2012 and Khandker and Mahmud 2012. • In particular, Khandker and Mahmud 2012 discuss seasonal patterns of income, expenditures, and poverty on pp. 43-61, based on the authors' analysis of data from the Household Income and Expenditure Survey of the Bangladesh Bureau of Statistics, showing marked dips in income and consumption in the lean season. See especially data on seasonal employment patterns on p. 60, showing that household employment in the farm sector varies seasonally from a peak of approximately 17 person-days per month to a low of approximately 8 person-days per month in the lean season. • Bryan, Chowdhury and Mobarak 2014 p. 1676 shows seasonal variation in total expenditures per capita, food expenditure per capita, price of rice, and quantity of rice per household based on the Household Income and Expenditure Survey of the Bangladesh Bureau of Statistics. • 3. "Although the change in the rice crop cycle has helped reduce the seasonal spread in rice prices, the traditional lean season preceding the aman harvest and spreading from September to November has changed little in its characteristics. It is the season of the least crop-related activity with no major crops planted or harvested, as can be seen from the crop calendar shown in figure 3.4. Although the wheat-growing season starts in November, the crop is grown in only about 5 percent of the cultivated land area. Moreover, the aftermaths of natural disasters like floods, drought, or excessive rains are usually felt most severely in this season. The cropping patterns in Bangladesh are delicately balanced within the natural cycles of rains and annual floods. Thus, farmers’ production options and perception of risk are often determined by the physical environment: the degree of seasonal flooding, the timing and quantity of rainfall, and the soil characteristics (Mahmud, Rahman, and Zohir 1994, 2000)." Khandker and Mahmud 2012, pp. 40-41. See especially figure 3.4, "Crop Calendar of Major Crops", which illustrates that no other major crop activities occur while the aman rice crop is in its growing period from September to November. • 4. A series of studies in rural northern Bangladesh has measured seasonal migration rates: • In a 2008 study, 36.0% of households in the control group participated in seasonal migration. Bryan, Chowdhury and Mobarak 2014, Table II p. 1683, column "Control", row "Migration rate in 2008". Eligible households for this study were selected on the basis of poverty, as measured by self-reported land ownership and self-reported skipped meals in the prior lean season: "Seventy-one percent of the census households owned less than 50 decimals of land, and 63% responded affirmatively to the question about missing meals. Overall, 56% satisfied both criteria, and our sample is therefore representative of the poorer 56% of the rural population in the two districts." Bryan, Chowdhury and Mobarak 2014, footnote 9, pp. 1677-1678. • In a 2009 follow-up to the 2008 study, 40.5% of households in the control group participated in seasonal migration. This migration rate was measured as part of a follow-up to the 2008 study above: "To study the longer-run effects of migration, and re-migration behavior during the next monga season, we conducted another follow-up survey in December 2009." Bryan, Chowdhury and Mobarak 2014, p. 1679. Therefore, the characteristics of participating households were as above. • In a 2014 study, 34.2% of households in the control group participated in seasonal migration. Akram, Chowdhury and Mobarak 2017, Table 1 p. 36, column "At least one migrant (2014-2015)", row "Control Mean". The eligibility criteria for this study and its follow-up were similar to the eligibility for the 2008 study and its follow-up: "The experiment was conducted in 133 randomly selected villages in Kurigram and Lalmonirhat districts of Rangpur. We first conducted village censuses to identify all households that would be “eligible” to receive this intervention in each of these villages. A household was deemed eligible if (1) it owned less than 0.5 acres of land, and (2) it reported back in 2008 that a member had experienced hunger (i.e., skipped meals) during the 2007 monga season." Akram, Chowdhury and Mobarak 2017, p. 7. • In a 2015 follow-up to the 2014 study, 37.8% of households in the control group participated in seasonal migration. Akram, Chowdhury and Mobarak 2017, Table 1 p. 36, column "Re-migration in 2016, at least one migrant", row "Control Mean". • Overall, across these four nonconsecutive years for which we have data on migration rates in populations not exposed to migration interventions, on average (36.0%+40.5%+34.2%+37.8%)/4 = 37.1% of poor households participated in seasonal migration. • We have not yet included data from the 2017 RCT in this calculation. • 5. • "Appendix Table A.VI provides further descriptive statistics on the number of migration episodes and average earnings by sector and by destination. Dhaka (the largest urban area) is the most popular migration destination, and a large fraction of migrants to Dhaka work in the transport sector (i.e., rickshaw-pulling). Many others work for a daily wage, often as unskilled labor at construction sites. At or around other smaller towns that are nearer to Rangpur, many migrants work in agriculture, especially in potato-growing areas that follow a different seasonal crop cycle than in rice-growing Rangpur." Bryan, Chowdhury and Mobarak 2014, p. 1691. Appendix Table A.VI is on p. 1732. • "Appendix Table A.II shows that the treatment does not significantly alter whether the household sends a male or female migrant, or the number of trips per migrant, or the number of migrants or trips per household (on the intensive margin, conditional on someone in the household migrating once)." ... footnote 17: "Migrants make 1.73 trips on average during the season, which implies that migrants often travel multiple times within the season. The first trip lasts 42 (56) days for treatment (control) group migrants. They return home with remittance and to rest, and travel again for 40 (40) days or less on any subsequent trips." Bryan, Chowdhury and Mobarak 2014, p. 1688. • 6. "Appendix Table A.VI provides further descriptive statistics on the number of migration episodes and average earnings by sector and by destination. Dhaka (the largest urban area) is the most popular migration destination, and a large fraction of migrants to Dhaka work in the transport sector (i.e., rickshaw-pulling). Many others work for a daily wage, often as unskilled labor at construction sites. At or around other smaller towns that are nearer to Rangpur, many migrants work in agriculture, especially in potato-growing areas that follow a different seasonal crop cycle than in rice-growing Rangpur." Bryan, Chowdhury and Mobarak 2014, p. 1691 • 7. "Appendix Table A.II shows that the treatment does not significantly alter whether the household sends a male or female migrant, or the number of trips per migrant, or the number of migrants or trips per household (on the intensive margin, conditional on someone in the household migrating once). The effects are concentrated on the extensive margin, inducing migration among households who were previously not migrating at all. However, the treatment does make it more likely that older, heads of households become more likely to migrate." ... Footnote 17: "The migrant is almost always male (97%), and often the household head (84% in treatment villages and 76% in control), who is often the only migrant from that household (93%)." Bryan, Chowdhury and Mobarak 2014, p. 1688. • 8. Bryan, Chowdhury and Mobarak 2014, Table VII, Panel C: Percentage of Migrants Traveling Alone, p. 1705: In the column "Non-Incentive", 32% of migrants in the control group traveled alone for their first migration episode, while 39% traveled alone for during any migration episode. • 9. In GiveWell's rough analysis of No Lean Season Preliminary 2016 Data (unpublished), program data from a conditional subsidy program for seasonal labor migration, we estimate that of households reporting time spent finding work, 44% report spending 0 days finding work at the migration destination. The median number of days reported spent searching for work is 1, and the maximum is 17. (Note that 0.4% of migrating households in this dataset reported not finding work.) • 10. • "In 2011, we conducted one more round of randomized interventions in the same sample of 1900 households (in 100 villages), plus 247 new households in 13 new randomly selected villages from the same two districts (Kurigram and Lalmonirhat). The treatments (most of which encouraged migration, like the 2008 experiments) were randomized at the village level. They were offered in February 2011, just before the onset of the 2011 “mini-monga season,” which is the pre-harvest lean season associated with the lesser of the two annual rice harvests. The treatments were therefore designed to encourage migration during this lean season." Bryan, Chowdhury and Mobarak 2014, p. 1722. • 2011 follow-up with the households in the 2008 control group, found that 32% of these control households participated in seasonal labor migration in the 2011 lesser lean season. Bryan, Chowdhury and Mobarak 2014, Table II, p. 1683, column "Not Incentivized", row "Migration rate in 2011". Text on p. 1684 clarifies that "The second and third rows of Table II compare re-migration rates in subsequent years across the incentive and non-incentive groups. We conducted follow-up surveys in December 2009 and in July 2011 and asked about migration behavior in the preceding lean seasons, but we did not repeat any of the treatments in the villages used for the comparisons in 2008." • 11. • "We randomly assign a cash or credit incentive (of$8.50, which covers the round-trip travel cost) conditional on a household member migrating during the 2008 monga season." Bryan, Chowdhury and Mobarak 2014, p. 1672, describing the subsidy amount in the 2008 randomized controlled trial (RCT).
• "The basic form of our intervention was the offer of a cash grant worth 1,000 Taka ($13.00 USD) to rural households in northern Bangladesh to cover the round-trip cost of travel to nearby cities where there are job opportunities during the lean season." Akram, Chowdhury and Mobarak 2017, p. 6, describing the subsidy amount in the 2014 RCT. • 12. "The loan was offered by our partner micro-credit NGOs that have a history of lending money in these villages. There is an implicit understanding of limited liability on these loans since we are lending to the extremely poor during a period of financial hardship." Bryan, Chowdhury and Mobarak 2014, p. 1678. • 13. • "The two districts where the project was conducted (Lalmonirhat and Kurigram) represent the agro-ecological zones that regularly witness the monga famine. We randomly selected 100 villages in these two districts and first conducted a village census in each location in June 2008. Next, we randomly selected 19 households in each village from the set of households that reported (a) that they owned less than 50 decimals of land, and (b) that a household member was forced to miss meals during the prior (2007) monga season." [footnote 9] "Seventy-one percent of the census households owned less than 50 decimals of land, and 63% responded affirmatively to the question about missing meals. Overall, 56% satisfied both criteria, and our sample is therefore representative of the poorer 56% of the rural population in the two districts." [main text] "In August 2008, we randomly allocated the 100 villages into four groups: Cash, Credit, Information, and Control. These treatments were subsequently implemented on the 19 households in each village in collaboration with PKSF through their partner NGOs with substantial field presence in the two districts. The partner NGOs were already implementing micro-credit programs in each of the 100 sample villages. The NGOs implemented the interventions in late August 2008 for the monga season starting in September. Sixteen of the 100 study villages (consisting of 304 sample households) were randomly assigned to form a control group. A further 16 villages (consisting of another 304 sample households) were placed in a job-information-only treatment. These households were given information on types of jobs available in four preselected destinations, the likelihood of getting such a job, and approximate wages associated with each type of job and destination (see Appendix A for details). Seven hundred three households in 37 randomly selected villages were offered cash of 600 Taka (~US$8.50) at the origin conditional on migration, and an additional bonus of 200 Taka (~US$3) if the migrant reported to us at the destination during a specified time period. We also provided exactly the same information about jobs and wages to this group as in the information-only treatment. Six hundred Taka covers a little more than the average round-trip cost of safe travel from the two origin districts to the four nearby towns for which we provided job information. We monitored migration behavior carefully and strictly imposed the migration conditionality, so that the 600 Taka intervention was practically equivalent to providing a bus ticket. The 589 households in the final set of 31 villages were offered the same information and the same Tk. 600 + Tk. 200 incentive to migrate, but in the form of a zero-interest loan to be paid back at the end of the monga season. The loan was offered by our partner micro-credit NGOs that have a history of lending money in these villages. There is an implicit understanding of limited liability on these loans since we are lending to the extremely poor during a period of financial hardship. As discussed below, ultimately 80% of households were able to repay the loan. In the 68 villages where we provided monetary incentives for people to seasonally out-migrate (37 cash + 31 credit villages), we sometimes randomly assigned additional conditionalities to subsets of households within the village. A trial profile in Figure 2 provides details. Some households were required to migrate in groups, and some were required to migrate to a specific destination. These conditionalities created random within-village variation, which we use as instrumental variables to study spillover effects from one person to another." Bryan, Chowdhury and Mobarak 2014, pp. 1677-79. • "We conducted a baseline survey of the 1900 sample households in July 2008, just before the onset of the 2008 monga. We collected follow-up data in December 2008, at the end of the 2008 monga season. These two rounds involved detailed consumption modules in addition to data on income, assets, credit, and savings. The follow-up also asked detailed questions about migration experiences over the previous four months. We learned that many migrants had not returned by December 2008, and therefore conducted a short follow-up survey in May 2009 to get more complete information about households’ migration experiences." Bryan, Chowdhury and Mobarak 2014, p. 1679. • 14. A subsequent study (the 2014 RCT) described its treatment arms as follows, and notes that the "low-intensity" treatment of approximately 14% of the eligible population was comparable to the previous study (the 2008 RCT): "We randomly assigned the 133 villages into three groups: (a) Low Intensity – 48 villages where we targeted migration subsidies to roughly 14% of the eligible (landless, poor) population. This is comparable to the Bryan et al. (2014) treatment. (b) High Intensity – 47 villages where we targeted roughly 70% of the eligible population with migration subsidy offers. (c) Control – 38 randomly selected villages where nobody was offered a migration subsidy." Akram, Chowdhury and Mobarak 2017, p. 7. • 15. Table II, Bryan, Chowdhury and Mobarak 2014, p. 1683. • 16. • 17. • "In 2013, we implemented interventions to encourage seasonal migration similar to those described in Bryan, Chowdhury and Mobarak, Econometrica 2014. However, beneficiaries did not react much to the migration subsidy offers in 2013, which is very unlike the responses we observed in 2008, 2011 and 2014, or even 2009 and 2015 (which were longer run responses to the subsidies offered in 2008 and 2014 respectively). The lack of a "first stage" effect on migration means that the research project failed, in that we couldn’t study the effects of migration on other outcomes, as we were not successful in generating any random variation in migration that year. Given these unexpected and counter-intuitive results, we conducted several rounds of discussions in 2016 with the field staff of IPA (our research management partner), and of RDRS (the implementation partner) to explore both implementation problems and external events that may have contributed to this failure. We found that the most plausible explanation of low migration rate in 2013 is the unprecedented political strikes (called hartals in Bangladesh) and strike-induced violence in late 2013, which deterred economic activity, and likely deterred migration due to the safety risks associated with movement during that period." Chowdhury and Mobarak 2016 note regarding 2013 hartals, email correspondence 19 October 2016. This note goes on to discusses the relevance of Ahsan and Iqbal 2016 to the 2013 study. • Ahsan and Iqbal 2016, figure 1, pdf p. 36 shows annual frequency of hartals, or labor strikes, in Bangladesh from 2005-2013, showing approximately 75 hartals in 2013, a far outlier compared to fewer than 15 hartals per year in 2007-2012. Other than 2013, the highest number of hartals per year was approximately 30 hartals in 2006. According to Ahsan and Iqbal 2016, p. 2, "These hartals disrupt the country’s transportation network". • Chowdhury and Mobarak 2016 note regarding 2013 hartals, email correspondence 19 October 2016 also discuss violence and deaths associated with hartals in 2013: "While hartals have been part of political protests and people’s lives in Bangladesh for a long time, the violence accompanying hartals in 2013 was extraordinary. Figure 2 reports the number of deaths due to hartal-induced violence across years. Deaths increased many-fold in 2013, and probably exceeded the combined death toll from all previous hartals called during the entire history of Bangladesh. Hartal combined with extreme violence was most likely was a strong deterrent to travel by road in 2013. From newspaper reports, it is clear that such violence not only takes place on hartal days, but also the days before and after a hartal is called." Figure 2 shows between 175-200 deaths due to hartal-induced violence in 2013, compared to fewer than 15 per year in 2007-2012. • Chowdhury and Mobarak 2016 note regarding 2013 hartals, email correspondence 19 October 2016 specifically find that hartals were prevalent during the 2013 monga: "Ahsan and Iqbal (2015) constructed the hartal data from newspaper reports, and are therefore also able to report the incidence of strikes by week or month of occurrence. We reproduce a figure from the 2015 version of their paper using monthly data to see whether the 2013 strikes and violence corresponded to the monga and migration periods that year. Unfortunately for us, hartals and hartal-induced violence were most prevalent in late 2013, which corresponds to the period right after our migration subsidies were disbursed. Our interventions in other years suggest that November and December are among the two most popular months for people to seasonally migrate from Rangpur." Figure 3 shows over 10 hartals and over 20 deaths in November 2013, and approximately 20 hartals and 40 deaths in December 2013. • 18. If we are able to see more detailed results from this study, it may help us to better understand the plausibility of different theories for why the program failed in 2013. For example, if political unrest led to decreased migration, we would expect to see lower rates of migration in the control group than occurred in the 2008 and 2014 RCTs. We have not yet seen such analysis. • 19. • "We offer to subsidize transport costs for 5792 potential seasonal migrants in Bangladesh, randomly varying saturation of offers across 133 villages." Akram, Chowdhury and Mobarak 2017, p. 1. • "The basic form of our intervention was the offer of a cash grant worth 1,000 Taka ($13.00 USD) to rural households in northern Bangladesh to cover the round-trip cost of travel to nearby cities where there are job opportunities during the lean season. This was a conditional transfer, where the subsidy is conditional on one person from the household agreeing to out-migrate during the lean season. As offers were made, we let households know that they may have a better chance of finding work outside of their village, but we did not offer to make any connections to employers. No requirement was imposed on who within the household had to migrate, or what city they had to go to. As in Bryan et al. (2014), migration was carefully and strictly monitored by project staff to ensure adherence to the conditionality." Akram, Chowdhury and Mobarak 2017, pp. 6-7.
• "The experiment was conducted in 133 randomly selected villages in Kurigram and Lalmonirhat districts of Rangpur. We first conducted village censuses to identify all households that would be “eligible” to receive this intervention in each of these villages. A household was deemed eligible if (1) it owned less than 0.5 acres of land, and (2) it reported back in 2008 that a member had experienced hunger (i.e., skipped meals) during the 2007 monga season. We focused on landownership because land is the most important component of wealth in rural Bangladesh, and it is easily measurable and verifiable. We used the second question on skipping meals to avoid professional, non-agricultural households (who may not own much land, but who are comparatively well off). Our census data suggest that about 57% of households in these villages were eligible to receive the intervention after applying these two criteria." Akram, Chowdhury and Mobarak 2017, p. 7.
• "We randomly assigned the 133 villages into three groups: (a) Low Intensity – 48 villages where we targeted migration subsidies to roughly 14% of the eligible (landless, poor) population. This is comparable to the Bryan et al. (2014) treatment. (b) High Intensity – 47 villages where we targeted roughly 70% of the eligible population with migration subsidy offers. (c) Control – 38 randomly selected villages where nobody was offered a migration subsidy." Akram, Chowdhury and Mobarak 2017, p. 7.
• "The sample of 133 villages included the 100 villages that were part of the earlier Bryan et al. (2014) experiment, but the majority of the households in our sample are new, and were not included in the earlier experiment. We show in Appendix Tables A2-A5 that participation in the earlier rounds of the experiment has no significant effect on migration decisions this year, and therefore does not materially affect the main results reported in this paper on the downstream effects of migration on income, labor supply and other outcomes." Akram, Chowdhury and Mobarak 2017, pp. 7-8.
• 20.
• "We disbursed grants during the latter part of the monga season, in early November, 2014. Figure 2 provides a timeline of project activities. Ideally, seasonal migration subsidy offers should be made in September after the rice planting work is done, but our disbursement was delayed due to political disturbance in Bangladesh at that time. Despite this delay, we observe high overall take-up and migration during the late Monga, as well as some post-harvest migration after January." Akram, Chowdhury and Mobarak 2017, p. 8.
• "Next, we conducted a detailed endline survey of 3,602 households during April 2015, before the start of the next rice-planting season. Figure 1 displays the sample breakdown across treatment arms and across types of households (those who were offered grants and those who were not). This endline survey collected information on a broader set of questions on migration and other socioeconomic outcomes that were not sensible or possible to ask repeatedly on a weekly basis, as in the high frequency survey. Core modules focused on collecting detailed information on the migration experience, including number of members who migrated, timing of migration events and destinations. The survey also delved into income generated by households (especially from migration), behavior and attitude changes, risk coping, credit and savings." Akram, Chowdhury and Mobarak 2017, p. 11.
• We are uncertain as to whether surveyors were blinded to the treatment status of those they surveyed.
• 21.
• Table 1, Akram, Chowdhury and Mobarak 2017, pdf p. 43.
• "About one-third of the households in control villages sent a seasonal migrant (34.2%), which is the same as what Bryan et al. (2014) and Khandker and Mahmud (2012) find in multiple years using other datasets. Households offered a grant in the low-intensity group were 24.8 percentage points more likely to migrate than a household in the control group, where no grant offers were made.11 In contrast, households offered a grant in the high-intensity group had a 39.8 percentage point higher propensity to migrate compared to the control group. This large difference in their reactions to the exact same offer is statistically significant (p-valueAkram, Chowdhury and Mobarak 2017, p. 16.
• 22.

Karen Levy, Director - Global Innovation at Evidence Action, conversations with GiveWell, July 13 and 20, 2018

• 23.

Karen Levy, Director - Global Innovation at Evidence Action, conversations with GiveWell, August 6, 2018

• 24.

Karen Levy, Director - Global Innovation at Evidence Action, conversations with GiveWell, July 13 and 20, 2018

• 25.

See discussion of the phases of Evidence Action's No Lean Season program here.

• 26.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 158,014 total eligible households, 48,960 (31%) are also coded as responding "Yes Accepted" (code 1) to their first offer to participate in the program (column AC), 64,018 (41%) are coded as "Interested in the Offer" (code 2), and 45,036 (29%) are coded as "No, has not accepted the offer" (code 3). The coding for this question can be seen in the No Lean Season 2017 CommCare Module Translation (unpublished), tab "module3_form1".

• 27.
Outcome of first offer, by month of offer, out of 158,014 total households made an offer.
August September October Overall
Accepted 10,344 (36%) 20,926 (35%) 17,690 (26%) 48,960 (31%)
Interested 12,050 (42%) 21,947 (36%) 30,021 (44%) 64,018 (41%)
Declined 6,397 (22%) 17,647 (29%) 20,992 (31%) 45,036 (29%)
Total offers 28,791 (18%) 60,520 (38%) 68,703 (43%) 158,014 (100%)

This table is based on the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished). Date of the first offer is the variable attempt1_date (column AD), and response to the first offer is the variable attempt1_outcome (column AC), coded as "Yes Accepted" (code 1), "Interested in the Offer" (code 2), and "No, has not accepted the offer" (code 3). The coding for this question can be seen in the No Lean Season 2017 CommCare Module Translation (unpublished), tab "module3_form1".

• 28.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 64,018 households coded as interested but not immediately accepting the initial offer to participate in the program (column AC), 38,191 (60%) have a recorded date of a second offer visit.

• 29.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 38,191 households recorded as interested in the program but not accepting the first offer (code 2, column AC)and visited a second time (column AF), 12,838 (34%) were recorded as accepting the second offer to participate in the program (code 1, column AE), 1,063 (3%) were recorded as interested in the program (code 2, column AE), and 24,290 (64%) were recorded as not accepting the second offer to participate in the program (code 3, column AE).

• 30.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 45,036 households that responded that they were not interested in the first offer to participate in the program (code 3, column AC), 8,841 (20%) were visited a second time (have a recorded date of a second visit in column AF).

• 31.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 8,841 households which, when first offered, responded that they were not interested in the program (column AC), and which were visited a second time (column AF), 5,273 (60%) responded to the second offer by accepting to participate in the program (code 1, column AE), 515 (6%) responded that they were interested (code 2, column AE) and 3,053 (35%) again did not accept the offer to participate in the program.

• 32.
Outcome of second offer, by month of offer, out of 47,032 total households made a second offer.
August September October November December January Overall
Accepted 206 (82%) 3,838 (98%) 9,628 (78%) 3,408 (35%) 1,031 (5%) 0 (0%) 18,111 (39%)
Interested 17 (7%) 23 (1%) 486 (4%) 668 (7%) 384 (2%) 0 (0%) 1,578 (3%)
Declined 27 (11%) 51 (1%) 2,282 (18%) 5,706 (58%) 18,985 (93%) 292 (100%) 27,343 (58%)
Total offers 250 (0.5%) 3,912 (8%) 12,396 (26%) 9,782 (21%) 20,400 (43%) 292 (0.6%) 47,032 (100%)

This table is based on the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished). Date of the second offer is the variable attempt2_date (column AF), and response to the second offer is the variable attempt2_outcome (column AE), coded as "Yes Accepted" (code 1), "Interested in the Offer" (code 2), and "No, has not accepted the offer" (code 3). The coding for this question can be seen in the No Lean Season 2017 CommCare Module Translation (unpublished), tab "module3_form1".

• 33.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 1,578 households still undecided after a second offer (response code 2 column AE), 703 (45%) were reached with a third visit (any response to attempt3_outcome, column AG).

• 34.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 27,343 households which declined a second offer (response code 3 column AE), 1585 (6%) were reached with a third visit (any response to attempt3_outcome, column AG).

• 35.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 2,290 households which received a third visit, 555 (24%) were recorded as accepting the offer to participate in the program (code 1, column AG), 187 (8%) were recorded as indicating interest in the program (code 2, column AG), and 1,548 (68%) declined to accept the offer to participate in the program (code 3, column AG).

• 36.

In the 2017 No Lean Season "2b. Household Offer Meeting" dataset (unpublished), out of 2,290 households which received a third visit (any date listed for attempt3_date, column AH), the date of that visit was prior to November 2017 for 52 households (2%), during November for 639 households (28%), during December for 1,477 households (64%) and after December 2017 for 122 households (5%).

• 37.

"The research team did a similar analysis of the number of visits and decided ultimately that it was not, on its own, a good indicator of access to loans, since MOs might have made judgments about which households to visit early and often, and also because “visits” as recorded in CommCare can be initiated by either the MO or a household member. The perspective shared here is, however, generally consistent with our current hypothesis that ‘mis-targeting’ – a combination of prioritizing households that were already most likely to send a migrant and working towards targets that were set too low – was the most likely explanation for why the program did not induce migration. Under this theory, MOs would have prioritized visits to ‘always migrants’, and ‘always migrants’ would have been the most likely to initiate contact with the MOs. This would also explain the pattern explained here." From Evidence Action's comment on a draft of this review, November 9, 2018.

• 38.

"Our survey only asked about expenditures during the second month of monga, and the modal migrant in our sample had not yet returned home (which includes cases where they may have returned once, but left again). We therefore expect the effects to persist for at least another month, and the total expenditure increase therefore easily exceeds the amount of the treatment ($8.50). Furthermore, if households engage in consumption smoothing, then some benefits may persist even further in the future. In any case, the$8.50 is spent on transportation costs two months prior to the consumption survey." Bryan, Chowdhury and Mobarak 2014, p. 1689.

• 39.
• "Our consumption data are detailed and comprehensive: we collect expenditures on 318 different food (255) and non-food (63) items (mostly over a week recall, and some less-frequently-purchased items over bi-weekly or monthly recall), and aggregate up to create measures of food and non-food consumption and caloric intake." Bryan, Chowdhury and Mobarak 2014, p. 1685.
• We are uncertain as to whether surveyors were blinded to the treatment status of those they surveyed.
• Table III, Bryan, Chowdhury and Mobarak 2014, p. 1686. Total consumption above control in the pooled incentivized group (the fourth data column of Table III) is 68.359 taka/person/month, 7% higher than the mean total consumption in the pooled non-incentivized group of 1000.87 taka/person/month (p<0.05).
• 40.
• "Our endline survey collected data on all sources of income (including migration income) over a 5-month recall period covering the entire main Aman rice growing season (including the preharvest lean period)." Akram, Chowdhury and Mobarak 2017, p. 19.
• Table 5, Akram, Chowdhury and Mobarak 2017, pdf p. 41, shows treatment effects on various types of income. We use column (4), "Migration and net non-migration income", which combines migration income with non-migration income from local wages and the net income from household enterprise (earnings minus costs). We confirmed our interpretation of this table with Mushfiq Mobarak, one of the lead researchers of this study. (Mushfiq Mobarak, email correspondence, October 17 2017)
• 41.

"To study the longer-run effects of migration, and re-migration behavior during the next monga season, we conducted another follow-up survey in December 2009. This survey only included the consumption module and a migration module." Bryan, Chowdhury and Mobarak 2014, p. 1679.

• 42.

Table II, Bryan, Chowdhury and Mobarak 2014, p. 1683. We are uncertain about the p-values for these comparisons. Table II includes p-values for the comparison of the pooled "incentivized" group (cash treatment pooled with credit treatment) compared to the pooled "not incentivized" group (information-only group pooled with pure control group) (rightmost column). For this comparison, the difference in re-migration rates in 2009 is 9.2 percentage points, significant at p<0.01. Other columns in this table do not include p-value asterisks and so are either not significant at p<0.1 or are not marked as such.

• 43.

Table III Panel B, Bryan, Chowdhury and Mobarak 2014, p. 1687.

• 44.

Table 9, email correspondence with Mushfiq Mobarak, Seungmin Lee, and Elizabeth Carls, October 13 2015 (unpublished).

• 45.

The 2008 "non-incentivized group" consists of the 2008 control group pooled with the households which received an information-only intervention in 2008. The 2008 "incentivized group" consists of the households which received conditional cash transfer offers in 2008, pooled with households which received conditional interest-free credit offers in 2008. Average consumption in the 2008 non-incentivized group in 2013 was 1698.3 takas/person/month, and was 98.6 takas higher (6% higher, not significant at p<0.10) in the incentivized group. Table 9, email correspondence with Mushfiq Mobarak, Seungmin Lee, and Elizabeth Carls, October 13 2015 (unpublished).

• 46.

"To study the longer-term behavior of households, we conducted a follow-up survey in August 2016 enquiring about a number of items over the time period beginning mid-August 2015 through mid-August 2016. This survey included questions on migration – specifically, timing and number of episodes, income from migration and questions about resource sharing by migrants – and the household’s experience of hunger over the previous year. This was administered to the original endline sample from the 2014-2015 round of study and we were able to effectively re-interview 94% of the sample (3,386 of the original 3,602 households). The migration subsidy program was not implemented again during the 2015 monga season, so this survey captures any longer-run changes from the intervention carried out during the prior lean season." Akram, Chowdhury and Mobarak 2017, p. 12.

• 47.

Table 1, Akram, Chowdhury and Mobarak 2017, pdf p. 43.

• 48.

Table 5, Akram, Chowdhury and Mobarak 2017, pdf p. 41.

• 49.
• 2014 migration rate among those made an offer in high-intensity villages: 39.8 percentage points higher than the migration rate in the control group (34.2%). Table 1, Akram, Chowdhury and Mobarak 2017, pdf p. 37.
• 2015 migration rate among those made an offer in high-intensity villages in 2014: 29.3 percentage points higher than the migration rate in the control group (37.8%). Table 1, Akram, Chowdhury and Mobarak 2017, pdf p. 37.
• We use migration-only income when comparing the program effects in 2014 and 2015 because total income figures are not available for 2015.
• 2014 migration-only income among those made an offer in high-intensity villages: 4,815 taka higher than migration-only income in the control group. The average household in the control group earned about 5,016 taka in migration-only income in 2014. Table 5, Akram, Chowdhury and Mobarak 2017, pdf p. 41.
• 2015 migration-only income among those made an offer in high-intensity villages: 7,500 taka higher than migration-only income in the control group. The average control group household earned about 9,205 taka in migration-only income in 2015. Akram, Chowdhury and Mobarak 2017, pdf p. 41.
• Assuming that the increase in migration rates in 2015 treatment is due to remigration of households induced to migrate in 2014 by the program, it appears that roughly 29.3/39.8 = 74% of those induced to migrate by the program in 2014 re-migrated in 2015.
• In 2014, 34.2% + 39.8% = 74% of the population that was eligible, in a high-intensity village, and made an offer, migrated. The eligible-and-offered population in high intensity villages as a whole earned an average of 5,016 + 4,815 = 9,831 takas in migration-only income, or 9,831/.74 = 13,285 takas per migrant.
• In 2015, 37.8% + 29.3% = 67.1% of the population that was eligible, in a high-intensity village, and made an offer in 2014, migrated. The eligible-and-offered population in high intensity villages as a whole earned an average of 9,205 + 7,500 = 16,705 takas in migration-only income, or 16,705/.671 = 24,896 takas per migrant.
• This suggests that the average migrant in 2015 earned (24,896-13,285)/13,285 = 87% more while migrating than the average migrant in 2014. The difference in migration income earned in 2015 compared to 2014 appears to be to a large degree secular, not an effect of the intervention: In the control group, the average migrant earned 5,016/.342 = 14,667 taka in 2014 and 9,205/.378 = 24,352 taka in 2015. That is an increase in migration-only income in the control group of about (24,352 - 14,667) / 14,667 = 66%.
• We do not know what conditions (economic, political, behavioral, etc) may have resulted in higher migration earnings in 2015 compared to 2014.
• 50.

The 2008 "non-incentivized group" consists of the 2008 control group pooled with the households which received an information-only intervention in 2008. The 2008 "incentivized group" consists of the households which received conditional cash transfer offers in 2008, pooled with households which received conditional interest-free credit offers in 2008. 32% of households in the 2008 non-incentivized group migrated in the 2011 lesser lean season, while 39% of households in the 2008 incentivized group migrated in the 2011 lesser lean season, a 7 percentage point difference significant at p<0.05. Table II, Bryan, Chowdhury and Mobarak 2014, p. 1683.

• 51.

The 2008 "non-incentivized group" consists of the 2008 control group pooled with the households which received an information-only intervention in 2008. The 2008 "incentivized group" consists of the households which received conditional cash transfer offers in 2008, pooled with households which received conditional interest-free credit offers in 2008. Average consumption in the 2008 non-incentivized group in 2011 was 1780.2 takas/person/month, and was 85.0 takas higher (5% higher, not significant at p<0.10) in the incentivized group. Table 9, email correspondence with Mushfiq Mobarak, Seungmin Lee, and Elizabeth Carls, October 13 2015 (unpublished).

• 52.

"Lastly, we asked women their perception of the impact of their partner’s migration on their relationship. Among women whose partners migrated during the lean season (50%), the vast majority reported that the migration itself had no impact on their relationship (consistent with the preliminary results we gather). That said, one in five women felt it had a positive impact on their relationship, and, reassuringly, less than 1% reported a perceived negative impact." Mobarak, Reimão, and Shenoy 2017, p. 4. See the figures in Table 2, p. 4.

• 53.
• "After the travel grants were disbursed in November 2014, we started surveying 2,294 households in both treatment and control villages about their wage and employment conditions. The survey was administered once every 10 days for six rounds starting on December 22, 2014. We refer to this as “High Frequency Origin Surveys”. The survey asked respondents about labor market outcomes (income, time spent working, location, industry) and a brief set of questions on consumption (essential food and non-food items) and migrant remittances." Akram, Chowdhury and Mobarak 2017 pp. 9-10.
• "We paired the brief consumption module in the high-frequency survey described above with repeated surveys of 399 shopkeepers (i.e. grocery store owners), or three in each of the 133 villages in our sample. These were administered simultaneously with the consumption module to collect prices for the same food items that the consumption module asked households about. We collected data on the prices of major food items, including rice, wheat, pulses, edible oil, meat, fish, eggs, milk, salt and sugar. These data allow us to explore whether encouraging migration at large scale in a village (and the extra income that generates) leads to price effects on food markets. It also allows us to convert the food consumption effects into monetary values." Akram, Chowdhury and Mobarak 2017, p. 11.
• 54.

Akram, Chowdhury and Mobarak 2017, Table 16, pdf p. 50. Column 5 shows a 0.147 (p<0.1) log increase in the price of fish (5.299 taka per kg of fish in the control group) per share of eligible households migrating. This implies that for every 10% of additional migration, the price of fish increases by 1.5%: "For every 10% increase in emigration, fish price increases by 1.5% (p<0.1)", Akram, Chowdhury and Mobarak 2017 p. 28.

• 55.

Akram, Chowdhury and Mobarak 2017, Table 16, pdf p. 50. Per share of the eligible population migrating, the price of prepared beverages decreases by 0.146 (p<0.1) in log units, the Laspeyres index of 12 goods increases by 0.0884 (p<0.05) in log units, and the price of edible oil increases by 0.0321 (p<0.1) in log units. This means that for every 10% increase in migration, the price of prepared beverages falls by 1.5%, the Laspeyres index of 12 goods increases by 0.9%, and the price of edible oil increases by 0.3%.

• 56.
• "Fifth, food prices in the village increase slightly on net, driven by an increase in the price of fish (the main source of protein), no change in the prices of main staples (like rice and wheat), and offset by a decrease in the price of prepared food and tea at the village tea shops." [footnote] "Bryan et al. (2014) documented an increase in protein consumption in migrant households. It appears that large-scale male migrant departures led to a negative demand shock for tea and snacks in village tea shops." Akram, Chowdhury and Mobarak 2017, p. 2.
• "We don’t detect any systematic increase in the prices of the most common staples. There is absolutely no effect of our intervention on the price of rice (or of flour, daal (lentils), sugar, salt or milk). Rice accounts for about 70% of the food budget in our sample. However, we see a statistically significant increase in the price of fish protein. For every 10% increase in emigration, fish price increases by 1.5% (pAkram, Chowdhury and Mobarak 2017, p. 28.
• 57.

The pre-analysis plan for the 2017 RCT (archived here) notes that one of the goals of the study is to "investigate the program’s spillover effects on workers at the migration destination who are not offered migration incentives." Given the lack of an effect found on migration rates due to the program, our understanding is that it will not be possible to use data from 2017 to investigate impact on workers at the migration destination.

• 58.

"After the travel grants were disbursed in November 2014, we started surveying 2,294 households in both treatment and control villages about their wage and employment conditions. The survey was administered once every 10 days for six rounds starting on December 22, 2014. We refer to this as “High Frequency Origin Surveys”. The survey asked respondents about labor market outcomes (income, time spent working, location, industry) and a brief set of questions on consumption (essential food and non-food items) and migrant remittances." Akram, Chowdhury and Mobarak 2017 pp. 9-10.

• 59.
• "The transport subsidies increase beneficiaries’ income due to better employment opportunities in the city, and also generate the following spillovers: (a) A higher density of offers increases the individual take-up rate, and induces those connected to offered recipients to also migrate. The village emigration rate increases from 35% to 65%. (b) This increases the male agricultural wage rate in the village by 4.5-6.6%, and the available work hours in the village by 11-14%, which combine to increase income earned in the village, (c) There is no intra-household substitution in labor supply, but primary workers within households earn more during weeks in which many of their village co-residents moved away. (d) The wage bill for agricultural employers increases, which reduces their profit, with no significant change in yield. (e) Food prices increase by 2.7% on net, driven by an increase in the price of (fish) protein, and offset by (f) a decrease in the price of non-tradables like prepared food and tea." Akram, Chowdhury and Mobarak 2017, abstract.
• See especially: discussion on Akram, Chowdhury and Mobarak 2017 pp. 19-26, and Tables 7-10 in the appendix.
• Households' primary wage earners (including households which participated in migration and those which did not) reported being without work 57% of the time when they were at home in control villages, and 49% of the time in villages where 70% of the eligible population was offered a conditional subsidy for seasonal labor migration. (p=0.352) Akram, Chowdhury and Mobarak 2017, Table 7, pdf p. 43.
• Households' primary wage earners' (including households which participated in migration and those which did not) average reported earnings in home villages during the 8-week survey period was 3,726 taka in the control group, and 4,003 taka for households in villages where 70% of the eligible population was offered a conditional subsidy for seasonal labor migration. This 7% difference was not statistically significant (p=.217). Akram, Chowdhury and Mobarak 2017, Table 7, pdf p. 43. Note that this survey includes both households that participated in seasonal migration and those that did not: "This implies that increases in income earned outside the village during the 8 week heavy migration period did not simply displace income that would have otherwise been earned in the village. Further, the direction of these effects is somewhat surprising, even though it is not statistically significant, because we induced the main income earner to leave home and go work outside the village. That the household continues to earn more at home even when the primary worker is not present (while also earning more at migration destinations) may indicate some changes in the village labor market, or intra-household changes in labor supply, and we need to understand this better." Akram, Chowdhury and Mobarak 2017, p. 22. Analysis in Table 13 (pdf p. 49) finds statistically significant differences in per-survey-period primary worker income earned at home (15% increase, p<0.01), total days worked at home (10% increase, p<0.01), and daily income at home (3% increase, p<0.05) for primary workers in villages where 70% of the eligible population was offered a conditional travel subsidy compared to control villages.
• "More out-migration causes the agricultural wage rate in the village to increase, but has no detectable effect on the non-agricultural wage rate. For every extra 10% of the landless population that emigrates, wages increase by 2.2% (pAkram, Chowdhury and Mobarak 2017, p. 24. Table 10, pdf p. 46, shows the intervention's effect on employer-reported wages.
• 60.

"To explore the impact of No Lean Season on social norms, we use data from the 2014 RCT and the follow-up survey conducted in 2016. The first dataset – and main one used here – is the endline survey to the 2014 RCT on spillovers, which was collected in April-May 2015, after the relevant lean season ended. In November 2016, a follow-up survey was conducted with the same households to assess the longer term effects of the intervention. We use some data from the follow-up to validate and assist in the interpretation of our results." Mobarak, Reimão, and Shenoy 2017, p. 2.

• 61.
• Mobarak, Reimão, and Shenoy 2017 Table 5, p. 7 presents responses to questions about spheres of decision-making. Only three of 66 coefficients in this table are statistically significant.
• Mobarak, Reimão, and Shenoy 2017 Table 7, p. 8 presents responses to questions about political beliefs and civic participation. The authors note that "In the regressions for political beliefs, a few of the coefficients on the treatment variable are significant at the 10% level, but, again, none are significant at the 5% level or higher. Also note that questions generating this marginal significance tend to be those with largely homogenous responses. For example, 98% of all respondents said they know a “woman who is capable of managing household affairs without help from her husband”. So, a difference of just a handful of respondents on either side tends to lead to these marginally significant coefficients." (p. 8)
• 62.

"In the survey, we also asked women about threats and violence they may have been subjected to by their husbands and/or other family members during the 6 months preceding the survey. A small share of women (4%) reported that their husbands either threatened them with divorce or taking another wife or acted on that threat during the previous 6 months. In contrast to this somewhat rare occurrence, more than half of women reported being verbally abused by their spouse or other family members, and 10% reported physical abuse within the last 6 months." [footnote 3:} "Note here that we are very aware of the literature discussing the challenge of measuring violence and abuse through self-reported data. Nonetheless, the tendency to underreport these cases is only an issue for our analysis if there is a systematic difference in this tendency between treated and untreated households. We have no reason to believe this is the case." Mobarak, Reimão, and Shenoy 2017, p. 3. See also Table 2, p. 4.

• 63.

Table 6, Mobarak, Reimão, and Shenoy 2017, p. 8 shows that the differences in response between treatment and control households are very small in magnitude and none is statistically significant.

• 64.

For example, 2% of male migrants reported that when they are at home, women alone make decisions about household expenditures. 17% reported that when they are at home, these decisions are made jointly, and 81% reported that they are made by the man alone. 43% of male migrants reported that when they are migrating, women alone make decisions about household expenditures. 16% reported that these decisions are made jointly, and 40% reported that they are made by the man alone. Table 9, Mobarak, Reimão, and Shenoy 2017, pp. 10-11. See also discussion on pp. 9-11.

• 65.

For example, we are aware of but have not reviewed Macours and Vakis 2010.

• 66.

See our November 2017 cost-effectiveness analysis here