Sampling Bias Does Not Exaggerate Climate–Conflict Claims

correspondence Sampling bias does not exaggerate climate– confict claims

To the Editor — In a recent Letter, claim, and the notion that security and Knowing that case selection is biased is Adams and colleagues1 argue that claims development are best addressed in concert useful, but not a reason to lower our estimate regarding climate–conflict links are is consistent with much political theory of the climate’s impact on conflict. ❐ overstated because of sampling bias. and practice4–6. However, this conclusion rests on logical Conceptually, the authors apply a Marc A. Levy fallacies and conceptual misunderstanding. curious kind of ‘piling on’ perspective in Center for International Earth Science Information There is some sampling bias, but it does not which each new study somehow ratchets Network, Columbia University, Palisades, NY, USA. have the claimed effect. up the consensus view of a country’s e-mail: [email protected] Suggesting that a more representative climate–conflict links, without regard Published online: 30 May 2018 literature would generate a lower estimate to methods or findings. Consider the https://doi.org/10.1038/s41558-018-0170-5 of climate–conflict links is a case of begging papers cited as examples of how selecting the question. It only makes sense if one cases on the conflict variable exaggerates References already accepts the conclusion that the the link: each uses a case selection 1. Adams, C., Ide, T., Barnett, J. & Detges, A. Nat. Clim. Change 8, 200–203 (2018). links are overstated. Otherwise it is possible strategy rooted in the qualitative methods 2. King, G., Keohane, R. O. & Verba, S. Designing Social Inquiry: that more representative cases might lead literature7. One, using a form of ‘crucial’ Scientifc Inference in Qualitative Research (Princeton Univ. Press, to stronger estimates. In fact, correcting case study, finds no evidence of climate Princeton, NJ, 1994). 3. Collier, D. & Mahoney World Politics 49, 56–91 (1996). sampling bias generally does tend to increase impacts on land-use conflicts in Mali, a 4. Hobbes, T. Leviathan or Te Matter, Forme and Power 2,3 effect estimates . region where climate–conflict links were of a Common Wealth Ecclesiasticall and Civill (Penguin, The authors’ claim that the literature’s particularly likely to be found8. The other, Harmondsworth, 1982). 5. Charter of the United Nations (United Nations, 1945); http://www. disproportionate focus on Africa using a ‘structured, focused comparison’, un.org/en/sections/un-charter/un-charter-full-text undermines sustainable development investigates two regions in the Middle East 6. Transforming Our World: the 2030 Agenda for Sustainable and climate adaptation rests on the same with similar climate stress but different Development A/RES/70/1 (United Nations General Assembly, 2015); http://go.nature.com/28TEATX fallacy. What if the links between climate conflict outcomes and concludes that 7. Bennett, A. & Elman, C. Annu. Rev. Polit. Sci. 9, 455–476 (2006). and conflict are as strong as people think? climate’s role as a conflict driver has been 8. Feitelson, E. & Tubi, A. Glob. Environ. Change 44, 39–48 (2017). It is far from obvious that acting as if exaggerated9. It is hard to see how these 9. Benjaminsen, T., Alinon, K., Buhaug, H. & Buseth, J. T. J. Peace they were not would somehow enhance papers mislead people into thinking Res. 49, 97–111 (2012). development and adaptation. The authors climate–conflict links are stronger than Competing interests offer no reasoning to support such a they really are. The author declares no competing interests.

Reply to ‘Sampling bias does not exaggerate climate–confict claims’

Ide et al. reply — We clarify three dependent variable’ strategy is widely and climate adaptation. We find, for arguments regarding our study1. recognized in the social sciences in general example, that some highly vulnerable First, sampling bias is a serious issue and in the environmental security literature countries receive very little attention in climate–conflict research. Although in particular2,3. from climate–conflict research (such as there are several forms of sampling bias Second, our study is concerned with the Bangladesh and Haiti)5. But by the same that do not affect estimates, the specific field as a whole rather than a critique of token, if our objective is to understand kind of bias we criticize — sampling individual studies. It is hence not helpful to how societies peacefully manage climate on the dependent variable — is very likely refer to individual studies to criticize our change and how such processes intersect to lead to an overrepresentation and conceptual approach and methods. We agree with development and conflict prevention6, overestimation of climate–conflict links. that both studies cited by Levy4 are excellent, then we must build explanations from cases Because cases experiencing both climate but they are not representative of the field in which climate risk is high but violent extremes and conflict are much more as a whole. This is like citing the only article conflict is not the outcome. The present widespread in the sample of cases studied on Oceania in our sample to prove that paucity of such analyses is a gap that needs than in the general population of cases, climate–conflict research has been well to be addressed. the relationship between these variables studied in this region. We do not deny a link between climate will seem to be more prevalent than it Third, the sampling biases we uncover change and conflict in principal. Indeed, is. This problem of a ‘sampling on the pose a problem for sustainable development some of our own recent work indicates