A growing number of policymakers, journalists and scholars are linking climate change to violent conflict9. Nevertheless, scientific evidence of this relationship remains elusive due to heterogeneous research designs, variables, data sets and scales of analysis10,11. Amid the array of disparate findings is a core of meta-analyses that are based on statistical methods12,13 as well as several in-depth studies linking climate change to highly prominent conflicts such as those in Darfur or Syria14,15.
Critics of this research point to an array of methodological problems, and to a lesser extent a deeper underlying problem with a study design that selects only cases where conflict is present or where data are readily available1,2,3,4,10. Researchers have, for instance, intensively studied the impact of a multi-year drought on the onset of the Syrian civil war in 2011, while there is little analysis of responses to the same drought in Jordan or Lebanon, where no large-scale violence erupted16. So, if the evidence of a causal association between climate and violent conflict is informed only by exceptional instances where violent conflict arises and climate also varies in some way, it is unable to explain the vastly more ubiquitous and continuing condition of peace under a changing climate.
Other critics of the research claiming a link between climate change and violent conflict have pointed to the way it stigmatizes some places—most often ‘Africa’ or a few African countries—as being more naturally violent than others. It does this ignoring the many similar and/or proximate places where peaceful responses are the norm, and the complex political, economic and institutional factors that cause violence and peace4,6,8,17. Such ‘mappings of danger’ can undermine the confidence of investors, local people and international donors and hence undermine sustainable development. They change the climate policy challenge from being one of adaptation with and in the interests of local people, to one of interventions to secure peace in the interests of those who fear the risk of contagious conflict and instability6,18.
So, it is important to understand whether the research claiming a link between climate change and violent conflict is based on a biased sampling strategy. Yet the extent to which this is the case remains untested. We therefore survey the relevant academic literature for the period 1990–2017 using the Scopus database and a systematic review—a method often used to analyse large bodies of literature with a high degree of rigour and replicability, and which is described in the Methods section with data provided in Supplementary Datasets 1 and 219,20.
The analysis of the relevant literature shows that Africa is by far the most frequently mentioned continent (77 mentions), followed by Asia (45) (see Table 1). The dominant focus on Africa in the literature is largely stable over time (see Fig. 1). This is surprising given that Asia is also home to places that are politically fragile and highly vulnerable to climate change21,22, but much more populous. Other continents with significant vulnerabilities to climate change (and that are at least in some places also prone to violent conflict), such as South America or Oceania, are hardly considered at all21.
Table 1 Most frequently mentioned continents and world regions in climate–conflict publications
Full size table
Fig. 1: Frequency of mentions of continents in the climate–conflict literature per year.
Fig. 1
The bars illustrate how frequently a continent was mentioned in the climate–conflict literature per year (2007–2017). No bar indicates that the continent was not mentioned in this year.
Full size image
With respect to world regions, Sub-Saharan Africa was by far most frequently mentioned in the literature analysed (44 times), although the Middle East (22) and the Sahel (22) were also discussed often (see Table 1). At the country level, Kenya and Sudan were most frequently analysed by climate–conflict researchers (11 mentions), followed by Egypt (8) as well as India, Nigeria and Syria (7). Complete lists of the continents, world regions and countries discussed in climate–conflict research can be found in Supplementary Dataset 1.
To check whether the selection of cases is biased towards the dependent variable, we run a number of Poisson regressions (see Supplementary Tables 1–3 for the full results) using data on, among others, the number of times a country is mentioned in the literature and on battle-related deaths between 1989 and 201522. Although the battle-related deaths data set is far from perfect and tends to underestimate small-scale violence (which many scholars believe is likely to be the most affected by climate change), it is currently the best global data set on violent conflict prevalence available.
The correlation between the number of mentions and a high death toll is positive and significant in all models (Fig. 2). This suggests that studies on climate–conflict links that research one or a few individual countries are disproportionally focusing on cases that are already experiencing violent conflict. Holding other factors constant, we estimate that countries with more than 1,000 battle-related deaths are mentioned almost three times as often as countries with a lower death toll. This is further supported by a comparison of the top ten countries of each list (Table 2). Six of the ten most-often-mentioned countries are also among the ten countries with the most battle-related deaths. The four remaining countries are also characterized by significant numbers of battle-related deaths, ranging from 2,775 (Egypt) to 8,644 (South Sudan).
Fig. 2: Changes in the frequency of mentions in the climate–conflict literature depending on country characteristics.
Fig. 2
Relative changes in the frequency with which countries are mentioned in the climate–conflict literature depending on climatic and other characteristics (estimated incidence rate ratios are shown, with 95% confidence intervals in grey). Estimated changes are not significant at the 5% level where confidence intervals cross the dashed line. Model 1 analyses the full sample. Model 2 includes English-speaking country instead of former British colony. Model 3 replaces Agriculture>25% of GDP with Agriculture>25% of employment. Model 4 uses high vulnerability rather than high exposure to climate change. Model 5 drops Kenya and Sudan from the analysis. Model 6 includes only African countries.
Full size image
Table 2 Countries most often mentioned in climate–conflict literature and countries with most battle-related deaths
Full size table
In contrast, the sampling of countries to be studied seems to be barely informed by the independent variable. A high exposure and a high vulnerability to climate change according to the ND-GAIN index23 are negatively, but not significantly, correlated with the number of times a country is mentioned (Fig. 2). The same holds true for the correlation with our climate risk measure based on the Global Climate Risk Index (CRI)24, although correlations are mostly significant here (Fig. 2), indicating that countries less at risk from climate change are more often discussed in the climate–conflict literature.
Table 3 adds further evidence to this claim. None of the ten most climate change-affected countries according to the ND-GAIN exposure score or the CRI are among the top ten countries considered in the climate–conflict literature. Further, the literature on climate change and conflict does not discuss 11 of these 20 high-climate risk-countries at all (Guatemala, Haiti, Honduras, Kiribati, Marshall Islands, Micronesia, Nicaragua, Philippines, Seychelles, Tuvalu and Yemen), despite many of them being characterized by significant political instability. There may be several reasons for these disparities, which include a greater interest in conflict-prone countries, issues of accessibility (discussed in the next paragraph) and a preference for studying countries with a higher global political relevance.
Table 3 Countries most often mentioned in the climate–conflict literature compared with the countries most exposed to and at risk from climate change
Full size table
The literature largely agrees that climate change is a ‘threat multiplier’ that aggravates existing tensions. It would hence make little sense to focus predominantly on countries that are politically very stable. Also, several analyses explicitly select their cases based on a number of scope conditions that are hypothesized to make climate–conflict links more likely16,25. But if studies (especially when analysing a small number of cases) focus on places that are already suffering from intense violent conflict, while highly vulnerable countries receive little attention, results may be distorted and significant knowledge gaps left unaddressed. In line with this, we find that further climate sensitivity measures such as the contribution of the agricultural sector to employment (negative, insignificant effect) and to gross domestic product (GDP; slightly positive and significant, but not robust effect) are weak predictors for the number of mentions (Fig. 2).
Our results further indicate a streetlight effect in climate–conflict research, that is, researchers tend to focus on particular places for reasons of convenience5. On the continent level, the availability of conflict data might have played an important role, especially as statistical analyses are very widespread in climate–conflict research10. Large geo-referenced conflict data sets spanning several countries and longer time periods were until very recently only available for Africa26. Indeed, when just considering statistical studies (n = 35 in our sample), the focus on Africa as a continent (65%) and Sub-Saharan Africa as a region (57%) is even stronger than in the full sample.
On the country level, all models reveal a positive and significant correlation between the numbers of mentions in the literature and countries that are former British colonies (Fig. 2). A likely explanation for this finding is that countries formerly colonized by Great Britain have better data (for example, historic weather records), which makes research more convenient5. Further, in four of the six most-mentioned countries (Sudan, Kenya, India and Nigeria). English is an official language (which makes research more practicable for many Western scholars). However, the positive correlation between these two factors indicated by model 2 (Fig. 2) is not significant. The presence of a streetlight effect in climate–conflict research is a reason for concern as it suggests that case selection (and hence knowledge production) is driven by accessibility rather than concerns for the explanation or practical relevance27.
One should note that the database we used for the literature search (Scopus) mainly captures journal articles that are written in English. Including French and Spanish language journals would probably yield a different picture of countries and regions most frequently mentioned.
The statistical findings provided by this study are robust to the use of different model specifications, the inclusion of further control variables, and the removal of the two most frequently mentioned countries (Kenya and Sudan) from the analysis (see Fig. 2 and the Supplementary Information for further information). Results also hold when analysing Africa only, hence suggesting that the detected sampling biases occur not only on a global scale, but are also valid for the continent most intensively discussed in climate–conflict research.
To conclude, critics have warned for some time that environmental security and climate–conflict research tend to choose cases on the dependent variable2,3,28. Our study provides the first systematic, empirical evidence that such claims are warranted. Studies focusing on one or a few cases tend to study places where the dependent variable (violent conflict) is present and hardly relate to the independent variable (vulnerability to climate change). In addition, climate–conflict research strongly focuses on cases that are most convenient in terms of field access or data availability.
To be clear, we do not intent to criticize individual studies, which often have good reasons to focus on specific regions, countries and phenomena. However, the sampling biases of the climate–conflict research field as a whole are deeply problematic for at least four reasons.
First, they convey the impression that climate–conflict links are stronger or more prevalent than they actually are3. This is especially the case for studies using few cases. Large-N studies usually contain a large number of non-conflict cases in their sample, although they draw all of these cases from a few regions or countries (see below).
Second, focusing strongly on cases of violent conflict limits the ability of (qualitative) researchers to study how people adapt peacefully to the impacts of climate change or carry out the associated conflicts non-violently4,29. Such knowledge, however, would be particularly valuable from a policy-making perspective.
Third, evidence of climate–conflict links comes primarily from few regions and countries that are convenient to access, such as (Sub-Saharan) Africa. This is even more of an issue in large-N, statistical analyses. While such a bias is not problematic per se as considerable parts of (Sub-Saharan) Africa are vulnerable to both climate change and conflict, this also implies that other very vulnerable regions, for instance in Asia and especially in South America and Oceania, receive little scholarly attention.
Finally, over-representing certain places leads to them being stigmatized as inherently violent and unable to cope with climate change peacefully4,6. This is particularly the case for Africa as a continent, the world regions Sub-Saharan Africa and the Middle East, and countries such as Kenya, Sudan or Egypt. Such stigmatization might contribute to the re-production of colonial stereotypes, especially as 81% of the first authors in our sample were affiliated with institutions in countries that are members of the Organisation for Economic Co-operation and Development (OECD). And it can also provide legitimation for the imposed security responses in certain places at the expense of co-produced adaptation responses in all places at risk from climate change17,18,30.
