Impacts of transient and permanent environmental shocks on internal migration

ABSTRACT

We examined whether floods and cyclones, the shocks that are transient in nature, affect interregional migration differently compared to riverbank erosion that causes loss of lands and thus generates permanent shocks. We tracked Household Income and Expenditure Survey 2000 participants in nine coastal districts of Bangladesh and collected further information in 2015. Our analyses suggest that both transient and permanent shocks induce households to migrate, but the effect is higher for the latter category. Using a difference-in-differences setting, we find that migrants’ income and expenditure increase relative to their counterparts, indicating that facilitating migration may improve welfare in disaster-prone countries.

KEYWORDS:

• Natural disaster
• permanent shock
• transient shocks
• internal migration

JEL CLASSIFICATION:

• I38
• Q54
• Q56
• R23

Acknowledgements

We acknowledge support from the National Science Foundation (Award #1204762, #1832693). We thank the respondents who spent their valuable time participating in the ‘Coastal Vulnerability and Livelihood Security (CVLS)’ survey, the staff members at the Evaluation & Consulting Services (ECONS) who implemented the survey and Shyamal Chowdhury who provided valuable inputs in the survey design. We also thank anonymous reviewers, Tobias Pfutze, Mihaela Pintea, Mahadev Bhat, Abu Shonchoy, Shahe Imran, seminar participants at the Florida International University, Boston University, Massey University and International Food Policy Research Institute and conference participants at the Southern Economic Association, Western Economic Association International, Eastern Economic Association, BDI at Yale University and Canadian Economic Association for helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Mejia et al. (2018) found that global temperatures had uneven macroeconomic effects, with adverse consequences concentrated in most low-income countries with a warmer climate. Coronese et al. (2018) further indicated that available studies on the damages caused by natural disasters systematically underestimate the real losses in low-income countries.

² Disasters may also affect life in many other ways, like lowering job prospects, reducing life satisfaction, lowering schooling and deteriorating mental health of the victims (Kellenberg and Mobarak 2011; Karbownik and Wray 2019).

³ The survey question specifically asked whether a household head migrated permanently to a new location. Previous studies found that often only single household members migrate (e.g. Gröger and Zylberberg 2016). Our empirical setting did not focus on the issue.

⁴ The CVLS survey collected data from 2096 households, of which 1835 observations had relevant information. We dropped 1257 households as they were not included in HIES 2000 and thus could affect the representativeness of our sample. Figure 222 shows the locations of the analysis households in 2000 and 2015.

⁵ Our findings are robust to the exclusion of the split households.

⁶ This seems a bit high but consistent with some recent studies like Marshall and Rahman (2013).

⁷ Objective measures of disasters are better than their subjective counterpart. For instance, Guiteras, Jina, and Mobarak (2015) find self-reported exposure to flooding to be weekly correlated to true exposure. While it is possible to employ objective measures of floods and cyclones, it is difficult to get an objective measure of river erosion as it is concentrated in a narrow geographical location and often not tracked in a systematic way. As combining subjective and objective measures can be problematic, we relied on the former in our analyses. Please note that if the estimates are statistically significant, the issue is less of a concern for the validity of the results.

⁸ The change in price levels (using CPI) between 2000 and 2015 was around 300% (BBS 2011, 2018).

⁹ Our data includes information on exposure to salinity, drought and some other type of natural disasters. Since a very small group of households reported suffering from these disasters, we did not separately control for them in the reported analyses. Our conclusions remain unchanged when they are controlled for. Also, we do not have data on the number of occurrences of transient shocks that people have experienced, which restricts us from controlling for the factor in the model.

¹⁰ Unfortunately, our analysis sample is not large enough to control for the district fixed effects.

¹¹ The multinomial probit model for migration choices is motivated by the framework of the random utility model, discussed in Davies, Greenwood, and Li (2001).

¹² Equation (3)(3) $Y_{it}={\beta }_1+{\beta }_{2}Pos{t}_t+{\beta }_3{M}_i\times Pos{t}_t+{\lambda }_i+{\epsilon }_i$(3) is, in fact, a conventional difference-in-differences model with household fixed effects and so it drops the migration variable from the model.

¹³ Since the individual regression coefficients of probit models are difficult to interpret, we reported marginal effects. Full regression outputs, including all other robustness check results, are available from the authors upon request.

¹⁴ Credit can be determined simultaneously with migration, making it endogenous in our model. Dropping it from the control variable as a robustness check does not affect our results qualitatively.

¹⁵ Surprisingly, the regional public spending for disaster risk reduction in Bangladesh does not seem to be determined by risk and exposure but only weakly by vulnerability (Karim and Noy 2015).

¹⁶ To calculate the ATEs, we used the default set up in the Stata program psmatch2 that employs the single nearest neighbour (without caliper) to calculate the matched outcome. When we changed the matching method, the results indicated that our findings are largely immune to such changes.

¹⁷ The positive and significant estimates of the coefficient Post indicate that incomes and expenditures of stayers also increased. However, our empirical strategy and available data cannot identify whether it is due to migration or the overall improvements in the macroeconomic factors of the economy.

¹⁸ Private market may also react less to a slow onset process. For example, the disclosure of the future risk of sea-level rise to the properties of Kapiti Coast in New Zealand did not affect their prices (Filippova et al. 2020).

Additional information

Funding

The work was supported by the NSF [Award #1204762, #1832693].