https://orcid.org/0000-0003-1189-2019
![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
ABSTRACT
Recent research has found evidence for a linkage between conflict induced-displacement and violence. Yet, displacement is also caused by natural disasters, whose implications for security have until now not received much attention. Drawing on spatial data on flood-induced disasters and forced migration in Africa, we investigate the impact of migration caused by natural disasters on social conflict. We show that disaster-induced displacement differs from conflict-induced displacement and raises distinct security implications. We also consider if areas simultaneously affected by conflict and disaster-induced migration are particularly at risk of conflict. The results suggest that there is no such amplifying effect.
Acknowledgments
Early drafts of the article were presented at the ENCoRe meeting in Uppsala, October 16th-18th, 2014, the International Study Association (ISA) 55th Annual Convention March 26th–29th, 2014, Toronto, Canada and at the Midwest Political Science Association (MPSA) 72nd Annual Convention April 3rd–6th, 2014, Toronto. We thank Alex Braithwaite, Christian Davenport, Joseph Hongoh, Sandra Lavenex, Idean Salehyan, Burcu Savun and two anonymous reviewers for helpful comments on earlier drafts of this paper. This work was supported by the the AXA Research Fund under a grant for a project on ‘Forced migration, environmental risks, and conflict’. In addition, Fabien Cottier acknowledges support from the National Science Foundation through an OIA award (nº 1934798).
Disclosure statement
No potential conflict of interest was reported by the author(s).
Supplementary material
Supplemental data for this article can be accessed here.
Correction Statement
Replication files for this article can be found here: https://doi.org/10.7910/DVN/VKGLHAThis article has been republished with minor changes. These changes do not impact the academic content of the article.
Notes
1. For an overview of the link between conflict-induced displacement and violence, see Rüegger and Bohnet (Citation2020) or the 2019 Journal of Peace Research special issue on ‘Refugees, Forced Migration, and Conflict.’
2. In Pakistan, for example, Islamist militants have provided aid in return for membership, in addition to relying on forced recruitment
3. Adams et al. (Citation2018) warns that research on climate change generally focuses on areas already experiencing frequent armed conflicts, such as Africa, which may lead to biased inferences. We do not wish to downplay these concerns. Yet, we believe these concerns are limited here, since we primarily examine the plausibility of our hypotheses, but cannot directly test the postulated mechanisms.
4. We refer to the appendix for a discussion of alternative data sources on flood-induced displacement.
5. By ‘large’ flood events, the Brakenridge (Citation2014) refers to episodes, which have caused ‘significant damage, to structures or agriculture, long (decades) reported intervals since the last similar event, and/or fatalities.’
6. We chose this coding scheme to generate a more granular measure of the extent to which an area has been affected by flood-induced displacement. We do not have a priori expectations as to which of these three types of zones is more likely to be associated with social conflict (see also the discussion in the appendix).
7. For more information on the FAO GAUL dataset, we refer to footnote 3 in the appendix.
8. We chose 1ʹ000 persons as a threshold to exclude floods that might have led to population displacement but the scale of which is unlikely to have caused a significant burden for the state and local authorities.
9. These three displacement variables displacement are by definition a subset of the corresponding flood variables. In other words, if an administrative zone is completely affected by flood- induced displacement, the same zone is also coded as fully affected by floods. Nevertheless, because a few administrative zones may have been affected by more than one flood in a given year, which may not all have caused displacement, there is a risk of a discrepancy in the coding of the flood and displacement variables. To solve this issue, we force the coding of the flood variables to reflect the coding of the flood-induced displacement variables.
10. For more information about the content of the GIDP dataset, see Bohnet, Cottier and Hug (Citation2018).
11. Social conflict events that revolved exclusively around any of the five following issues were systematically excluded, as they are unlikely to be related to disaster-induced displacement: ‘Elections,’ ‘foreign affairs/relations,’ ‘domestic wars,’ ‘violence, terrorism,’ ‘pro-government’ (Salehyan and Hendrix Citation2012). Social conflict events, which involved ‘pro-government violence’ or ‘intra-governmental violence,’ are also exclused, because the theory does not address repression, nor conflict within the state. Finally, we removed events which could not be geo-located at a level of precision equal or higher to the first order administrative zone.,
12. We note that the SCAD is not exempt of limitations. In particular, its reliance on newswires from the Agence France Presse and Associated Press is susceptible to induce a selection bias with events occurring in major urban areas, as well as events of larger magnitude, more likely to be reported (Weidmann Citation2015). In spite of its limitations, we opted for the SCAD to measure social conflict events over existing alternatives, such as the ACLED (Raleigh et al. Citation2010, ACLED Citation2019), because its exclusive reliance on newswires make events reporting more consistent over time and space.
13. While it would have been interesting to examine how these control variables may moderate the effects of disaster-induced displacement on social conflict, we are here primarily interested in how displacement directly affects the odds of conflict. Moreover, extending the empirical analysis to the examination of interaction effects would have gone beyond the scope of the present article. We thus leave this aspect for future research.
14. We also replaced the flood count variable by a variable counting the number of flood events having caused a displacement of at least a 1ʹ000 persons, but the results are similar to those we report in .
15. The 1 degree cell resolution of the G-Econ dataset is problematic as the G-Econ cells frequently overlap administrative boundaries. Therefore, we generate a population-weighted dataset with a resolution equal to a 2.5 arc-minute.,
16. Both the GPW and the Nordhaus datasets are available at five year intervals (from 1990 to 2000 for the former, respectively 2005 for the latter). We extrapolate interval years for the entire period 1990–2011.
17. Variables at the country level only serve as controls and, given the hierarchical structure, will yield coefficients with standard errors that are underestimated.
18. Unfortunately, our data only allows us to examine the link between disaster-displacement and social conflict, but we cannot directly test the causal mechanisms posited.
19. In order to obtain conservative estimates, each set of simulations was conducted only on the sub-sample of administrative zones, which has been fully, respectively partially affected by floods, or was located in close proximity to one.
20. The estimated differences in the average predicted differences for fully affected administrative zones is distinct from zero at the 90 percentile for the immediate response.
21. We provide further illustration in Figure A.1 in the Appendix, which depicts average simulated probabilities at varying levels of past occurrence of floods.
22. The 95 per cent confidence interval extends over the interval bounded between 0.001 and
0.070.
23. Because we encounter a problem of separation due to the large number of interaction terms included in this specification and the restricted time range, we estimate this model using Bayesian generalised linear models (see arm package in R, Gelman and Hill Citation2006).
24. This result must be interpreted in light of the fact that administrative units hosting conflict IDPs already possess a higher than average risk of conflict.
25. The point estimate of the difference in the average predicted differences for administrative zones in proximity to a flood is distinct from zero at the 90 per cent confidence interval.
26. Average simulated probabilities for the effects of flood-induced displacement in the absence of conflict IDPs are depicted in Figure A2 of the Appendix.
27. Models 5–16 in Table A.2–A.3 of the Appendix present the results of the sensitivity analysis. With minor caveats, the results of the sensitivity analysis do not alter the main conclusions of the analysis.