Climate and Civil War: Is the Relationship Robust?

Marshall Burke, J. A. Dykema, David B. Lobell, Edward Miguel, Shanker Satyanath2010 · DOI

Abstract

A recent paper by Burke et al. (2009) found a strong historical relationship between warmer-than-average temperatures and the incidence of civil war in Africa. These findings were subsequently challenged by Buhaug (2010), who questioned the methodology, variable controls, and data choices. This paper demonstrates that the original findings are robust to different climate datasets and war codings. Furthermore, we show that using preferred climate data under sound econometric assumptions suggests an even stronger relationship for the 1981–2002 period than originally reported. However, we find that this relationship has weakened over the last decade, a period marked by significant African economic growth and a decline in large-scale conflict.

Introduction

The link between environmental stress and political instability is a cornerstone of the "climate security" discourse. In 2009, Burke et al. provided empirical weight to this discussion, identifying a significant correlation between rising temperatures and civil war in sub-Saharan Africa. This finding was met with skepticism, most notably by Halvard Buhaug (2010), who argued that the relationship was a byproduct of specific methodological choices. This paper serves as a formal rebuttal and a deeper exploration of the climate-conflict nexus, seeking to determine if heat is a genuine "threat multiplier" or merely a statistical artifact.

Methodological Rigor and the "Bad Control" Problem

A central point of contention in the climate-conflict debate involves how to properly control for confounding variables. Buhaug (2010) argued that the climate effect disappears when controlling for factors like GDP growth. However, we argue that this approach suffers from serious econometric shortcomings, specifically simultaneity bias.

Because climate shocks (like extreme heat) likely influence conflict through their impact on the economy (e.g., crop failure leading to lower GDP), including GDP as a control variable "soaks up" the very causal pathway being studied. By using country fixed effects and time trends, we isolate the exogenous variation in weather—essentially treating year-to-year temperature changes as a natural experiment—without neutralizing the economic mechanisms that link heat to violence.

Robustness Across Datasets

To address concerns regarding data sensitivity, we tested our models against multiple sources:

  • Climate Data: We compared the Climate Research Unit (CRU) dataset with the University of Delaware (UDel) and Global Precipitation Climatology Project (GPCP) datasets.
  • Conflict Data: We utilized the Uppsala/PRIO database (v4, 2009), distinguishing between "large wars" (over 1,000 deaths) and "small wars" (over 25 deaths).

The results were striking. When applying sound econometric models to the climate data preferred by our critics (UDel), the estimated impact of temperature on conflict for the 1981–2002 period was twice as strong as our original estimate. Specifically, a 1°C increase in temperature was associated with a 10% or greater increase in the risk of civil war. This relationship proved most robust for high-intensity "large wars," suggesting that while climate may not trigger every minor skirmish, it is a significant driver of major societal breakdowns.

The Recent Decoupling: Development as a Buffer

Perhaps the most significant finding in this re-evaluation is the breakdown of the climate-conflict link after 2002. Between 2003 and 2008, the statistical significance of temperature as a predictor of war vanished.

This period coincided with a transformative era for the African continent:

  1. Economic Growth: Average GDP growth rose from near-zero to 2.8%.
  2. Democratization: Average Polity IV scores improved from -2.5 (autocratic) to +1.8 (leaning democratic).

This "decoupling" suggests that the relationship between climate and war is not deterministic. Instead, economic resilience and improved governance appear to act as a buffer, effectively insulating societies from the destabilizing effects of environmental shocks.

Conclusion

The historical link between heat and African civil war is robust and survives rigorous methodological scrutiny. However, the recent weakening of this link provides a roadmap for future policy. If the "heat-war" cycle has been broken by economic and political progress, then investing in development and democratic institutions is not just a humanitarian goal—it is a critical strategy for climate adaptation. The challenge remains to ensure these gains are permanent, as a reversal in growth or governance could once again leave the continent vulnerable to the volatile effects of a warming planet.

Summary updated

Abstract

A recent paper by Burke et al. (henceforth "we") finds a strong historical relationship between warmerthan-average temperatures and the incidence of civil war in Africa (Burke et al. 2009).These findings have recently been challenged by Buhaug (2010) who finds fault with how we controlled for other potential explanatory variables, how we coded civil wars, and with our choice of historical time period and climate dataset.We demonstrate that Buhaug's proposed method of controlling for confounding variables has serious econometric shortcomings and show that our original findings are robust to the use of different climate data and to alternate codings of major war.Using Buhaug's preferred climate data under sound econometric assumptions yields results that suggest an even stronger relationship between temperature and conflict for the 1981-2002 period than we originally reported.We do find that our historical relationship between temperature and conflict weakens over the last decade, a period of unprecedented African economic growth and very few large wars.
BibTeX
@misc{burke2010climate,
  author = {Marshall Burke and J. A. Dykema and David B. Lobell and Edward Miguel and Shanker Satyanath},
  title = {{Climate and Civil War: Is the Relationship Robust?}},
  year = {2010},
  doi = {10.3386/w16440},
}
Figure