With climate change, we are seeing more extreme weather events and increasingly frequent hazards, putting agriculture and rural livelihoods at ever greater risk. Temporary migration can help households withstand these types of short-term hardships and income deficits, and may be a viable mechanism for adapting “in place” to changing environmental conditions. However, adapting in place keeps people in vulnerable areas, potentially magnifying losses – human and financial – in the future. Understanding migration as adaptation is, therefore, essential to effective policy design for climate mitigation and disaster relief.

According to the United Nations High Commissioner for Refugees, by 2002 an estimated 20-25 million people had already been “internally displaced” due to environmental factors. In earlier work on Bangladesh, my co-authors and I find that weather and other environmental factors have significant effects on permanent migration though, in some cases, vulnerable groups end up being trapped by adverse shocks. Other studies also find mixed results (e.g. heat stress triggers out-migration but few households report environment as a reason for migration). The impact of environment on temporary migration has been even more difficult to characterise. Many surveys record only permanent changes in household membership and/or location. Even when temporary migration is enumerated. Such moves are often limited in duration and occur over short distances, raising concerns about recall bias.

Study design

To address these limitations, we implemented a unique high frequency survey with generous support from the International Growth Centre and the Institute for Population Research and Sustainable and Resilient Economy Discovery Theme at The Ohio State University. To identify our sample, we utilised a long-standing panel data collection, the Nutrition Survey of Rural Bangladesh, and interview households by phone every two months over a one-year period. This approach has the advantage of allowing us to link detailed data from the previous in-person survey (conducted in 2007-08), as well as including a broad range of households around the country without pre-selecting on mobile phone access.

Although mobile phone penetration is quite high in Bangladesh, we encountered several challenges in implementing this study design. A number of households, roughly 8.6%, had to be provided with phones, while another 8% were found to be entirely unreachable by phone. Other sub-groups required special outreach efforts to encourage participation and minimise attrition, including the use of translators for local dialects, transportation to nearby areas with better network coverage and phone charging stations, and local liaisons to update respondents’ contact information (i.e. changing SIM cards). Respondents were compensated for their participation with mobile minutes, to offset direct costs as well as foregone wages.

To estimate the effect of weather shocks on temporary migration, we match the bi-monthly data to satellite-based measures of flooding and precipitation at the upazila (sub-units of districts) level. To account for underlying differences in climate across regions, we also incorporate data on a broad range of historical weather conditions and environmental characteristics. The sub-district-specific long-run averages and standard deviations of precipitation, growing degree days and sun exposure from weather stations, and flooding are all included. Additionally, for each sub-district, we include a measure of river density based on the proportion of the total area covered by rivers, and the percentage of land with saline-contaminated soil as identified in the most recent soil survey.

Contemporaneous shocks reduce migration

Preliminary results indicate that contemporaneous shocks, rainfall, and flooding deviations from the upazila-specific mean in the one to two months prior to the interview, reduce the likelihood of the household having a temporary migrant. However, this effect displays strong seasonality and is significant only for shocks in the late summer and early fall. Higher than usual precipitation at this time of year is beneficial for the growth of aman rice, so favourable rainfall shocks in this period reduce the need for income diversification/supplementation through temporary migration.

Conversely, flooding has adverse effects on both aman and aus rice crops at this point in the season. And, during this particular period (August-September 2017), especially heavy rains in Bangladesh led to widespread flash flooding and destruction of crops. But, since crops had already been planted when this flooding occurred, temporary migration was less appealing as a diversification strategy. Instead, households could more effectively minimise income losses by maintaining labour power to complete the harvest and seeking supplemental income locally.

Adding controls for climate/environmental factors and household characteristics has little to no effect on the point estimates for weather shocks. This indicates that there is little correlation between contemporaneous seasonal shocks and underlying climate/environment characteristics and provides reassurance that our measures of weather shocks are not correlated with household socio-economic status and migration preferences. Self-reported shocks are also found to reduce out-migration suggesting that, in the immediate aftermath of an adverse shock, households are more likely to focus on local income smoothing opportunities, perhaps because migration itself may be risk-increasing. This also points to liquidity constraints as a potential factor limiting migration decisions and raises concerns about a trapped population dynamic.

But distant shocks increase migration

In contrast, higher than usual flooding in the year prior to the survey significantly increases the likelihood of temporary migration. This suggests that the effect of contemporaneous weather shocks on migration is more constrained by current labour needs and/or liquidity constraints, while adaptation to more distant shocks can be more forward-looking. The positive effect of distant shocks on migration is confirmed in our pre-survey data, which uses a traditional (one-time) in-person interview approach. However, without disaggregating the timing of migration episodes by season, we mask important and economically/socially significant causal relationships between weather shocks and temporary migration.

Policy implications

These findings suggest that, when high frequency interviews cannot be conducted, surveys should attempt to obtain information about the timing of migration episodes, where possible, and weather shocks should be disaggregated to allow for seasonal variation in the environment-migration relationship. Our study also reveals a much more nuanced relationship between weather shocks and temporary migration, with both contemporaneous and distant shocks being relevant. In Bangladesh, temporary migration appears to be part of a long-term ex ante income diversification strategy, rather than a short-term response to shocks. Thus, although this may alleviate concerns about environmental migration exacerbating aggregate shocks (e.g. rural migrants flowing into urban areas that have also been flooded), policymakers should be prepared for the effects of such shocks to ripple out over time.