As a programme scales, it is important to check that RCT results are replicable and that general equilibrium effects are considered
Northern Bangladesh sees high seasonal fluctuations in income, consumption, and hunger in rural areas. No Lean Season, an Evidence Action project offering small subsidies for seasonal migration, may come to reach millions of households in that region. The project originated from a randomised control trial (RCT) carried out in 2008-2011, and has expanded through a very close partnership between academics and practitioners.
The initial RCT on which No Lean Season is based found that providing travel subsidies to the rural poor to enable seasonal labour migration had positive effects on the household level. This includes raising expenditures, consumption, and caloric intake during the lean season (Bryan et al., 2014). In the first installment of this two-part series, we discussed the unique considerations that come into play when moving from an RCT to a larger intervention, particularly looking at potential negative collateral effects on targeted and non-targeted households.
In transforming RCT results into a scaled programme, two more elements are key:
- ascertaining that the results from a single RCT are replicable and not a one-off anomaly; and
- considering the general equilibrium effects that may arise at scale.
When exploring whether to transform an RCT into a programme, the first question is whether the results are replicable in the desired context. In 2014, we set out to answer this question for No Lean Season with a slow ramp-up involving migration subsidy offers to 5800 households (from 1292 in the original study). We also sought to identify any spillover effects on other members in the villages.
Specifically, we randomly assigned 133 villages to:
- a ‘low-intensity’ treatment group: in these, 14% of the eligible landless population was offered a loan conditional on migration (48 villages);
- a ‘high-intensity’ treatment: in these, 70% of eligible beneficiaries were offered the conditional loan (47 villages); or
- the control group (38 villages), where no one was offered a loan.
This design enabled us to explore whether the saturation of loans within a village matters for impact.
Findings
Seasonal migration
Reassuringly, we confirmed the positive impact of the loan on seasonal migration itself (Akram et al. 2017). Furthermore, we found that households offered the loan in the ‘high-intensity’ villages were 40 percentage points more likely to send a migrant, and those in ‘low-intensity’ villages were 25 percentage points more likely, relative to the control group.
The higher take-up in the high-intensity villages points to coordination in travel, as migrants often travel in groups and are more likely to go if a neighbour or friend also migrates. In fact, in ‘high-intensity’ villages, even those who were not offered the loan were 10 percentage points more likely to migrate than those in control villages. The combination of these effects led to 65% of the share of poor, landless households to emigrate from the high-intensity treatment villages, relative to a 35% out-migration rate from the control villages.
Local labour market
This 30 percentage point increase in the emigration rate caused the village agricultural wage rate to increase by 4.5-6.6%, benefiting both those who stay behind and migrants during their spells at home (as they usually do not migrate for the entire three to four-month period, but in blocks of a few weeks at a time, returning home in the interim).
Replicability
Through this iteration, we found evidence that the impact of the original RCT was not a fluke, particularly as the numbers we got were not only large and positive but also very close to those from the first study. We also learned that, due to travel coordination, it is more cost-effective to offer the loan to more households in fewer villages than fewer households in more villages. The intervention also potentially improves the labour market for rural labourers who stay behind, as migration decreases competition for scarce jobs during the season.
Further thoughts on general equilibrium: Permanent migration
Other big questions regarding No Lean Season relate to general equilibrium effects over time. A recurring question for this intervention has been its possible role in accelerating urbanisation and imposing further pressure on weak urban infrastructure systems in Bangladesh.
To answer this, in 2016, we carried out a follow-up survey of the households in the original RCT to explore whether there are long-term effects of the initial transfers on both recurring seasonal migration and permanent migration into urban destinations. Our results thus far do not point to any long-term acceleration of urbanisation, as the vast majority of rural migrants do not pick up and move either to an urban destination or another rural area during the eight-year interval studied. That is, No Lean Season could make rural living more viable and sustainable by providing households with a better tool to cope with the lean season, rather than pushing households into urban areas more quickly.
An RCT on the path to scale: Revisiting expenditures and consumption, and considering impacts on popular destination areas
Separately, we have a much larger RCT iteration underway, expected to disburse migration loans to 40,000 households in the current lean season. This multi-year study will again measure impacts on expenditures and consumption for direct beneficiaries, allowing for general equilibrium effects that may arise at this scale.
New to this iteration, we will also investigate how seasonal migration affects destination labour markets, as seasonal migrants might compete with or complement both workers already living in urban destinations and seasonal migrants from other rural areas (e.g. Ottaviano and Peri 2011). We will also explore health impacts that might surface as more and more migrants move between their villages of origin and seasonal destinations.
From RCT to scale: Testing replicability in new settings
A final question worth considering in the path to scale is whether an intervention can and should be replicated in other settings. Expanding a programme that is successful in one context – however impactful – to another location, brings new complexities.
No Lean Season will probably not produce similar success in areas where urban labour markets already have excess labour supply and high levels of unemployment, or in locations where transportation networks are weaker and physical mobility is restricted. In fact, through some research, we have determined that No Lean Season is unlikely to work in Malawi or eastern Zambia at this point, as urban labour markets in those countries are probably not able to absorb additional seasonal migration influxes.
On the other hand, a small pilot of No Lean Season is currently underway in Indonesia as well – with important modifications for the context. The policy landscape means we are working with the government through the Ministry of National Development and Planning (BAPENAS), focusing on intra-island migration. Land ownership changes the timing of the intervention as well, as the rural poor in Indonesia are generally subsistence farmers, who do not want to migrate during the lean season itself (leaving their planted crops behind) but prefer to travel post-harvest.
Lessons learned
We have learned a lot from our work on seasonal migration in Northern Bangladesh over the last nine years, not only in terms of the positive impact of No Lean Season on the lives of poor rural households, but also on how to move from a small RCT to an intervention directly impacting tens of thousands of people. We hope that the lessons we have gathered, and those we continue to encounter – along with our willingness to end or substantially modify the programme if we find negative effects at any dimension – contribute to the discussion and serve as a model for scale-up activities in development.
There is a lot more to scaling-up than finding initial positive effects and then hypothesising on general equilibrium. Considering these effects that may only arise or be relevant at scale requires a lot of work and a lot more data. However, it also affords more confidence that the intervention at scale will come to improve the lives of hundreds of thousands of poor households in Bangladesh – and possibly beyond.
Editor’s Note: This article is based on an IGC project.
References
Akram, A and M Mobarak (2016), "Effects of Emigration on Rural Labor Markets", NBER Working Paper No. 23929.
Bryan, G, S Chowdhury and A M Mobarak (2014), "Underinvestment in a profitable technology: The case of seasonal migration in Bangladesh", Econometrica 82(5).
Ottaviano, G and G Peri (2011,. “Rethinking the effect of immigration wages”, Journal of the European Economic Association 10(1): 152-197.