Pdf climate and social conflict

Pdf File 2,233.63 KByte, 54 Pages

1%(5:25.,1*3$3(56(5,(6

&/,0$7($1'&21)/,&7 0DUVKDOO%XUNH

6RORPRQ0+VLDQJ (GZDUG0LJXHO

:RUNLQJ3DSHU KWWSZZZQEHURUJSDSHUVZ

1$7,21$/%85($82)(&2120,&5(6($5&+ 0DVVDFKXVHWWV$YHQXH &DPEULGJH0$ 2FWREHU

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

&OLPDWHDQG&RQIOLFW 0DUVKDOO%XUNH6RORPRQ0+VLDQJDQG(GZDUG0LJXHO 1%(5:RUNLQJ3DSHU1R 2FWREHU -(/1R,2344

ABSTRACT

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effect by far, with each 1iLQFUHDVHWRZDUGZDUPHUWHPSHUDWXUHVLQFUHDVLQJ WKHIUHTXHQF\RIFRQWHPSRUDQHRXVLQWHUSHUVRQDOFRQIOLFWE\DQGRILQWHUJURXSFRQIOLFWE\ EXWWKDWWKHSHULRGFXPXODWLYHeffect of rainfall on intergroup conflict is also substantial (3.5%/i :HDOVRTXDQWLI\KHWHURJHQHLW\LQWKHVHHIIHFWHVWLPDWHVDFURVVVHWWLQJVWKDWLVOLNHO\LPSRUWDQW:H FRQFOXGHE\KLJKOLJKWLQJUHPDLQLQJFKDOOHQJHVLQWKLVILHOGDQGWKHDSSURDFKHVZHH[SHFWZLOOEHPRVW HIIHFWLYHDWVROYLQJWKHPLQFOXGLQJLGHQWLI\LQJPHFKDQLVPVWKDWOLQNFOLPDWHWRFRQIOLFWPHDVXULQJ WKHDELOLW\RIVRFLHWLHVWRDGDSWWRFOLPDWHFKDQJHVDQGXQGHUVWDQGLQJWKHOLNHO\LPSDFWVRIIXWXUH JOREDOZDUPLQJ

0DUVKDOO%XUNH &HQWHURQ)RRG6HFXULW\DQG(QYLURQPHQW 6WDQIRUG8QLYHUVLW\ 6WDQIRUG&$ PEXUNH#VWDQIRUGHGX

6RORPRQ0+VLDQJ *ROGPDQ6FKRRORI3XEOLF3ROLF\ 8QLYHUVLW\RI&DOLIRUQLD%HUNHOH\ +HDUVW$YHQXH %HUNHOH\&$ DQG1%(5 VKVLDQJ#EHUNHOH\HGX

(GZDUG0LJXHO 'HSDUWPHQWRI(FRQRPLFV 8QLYHUVLW\RI&DOLIRUQLD%HUNHOH\ (YDQV+DOO %HUNHOH\&$ DQG1%(5 HPLJXHO#HFRQEHUNHOH\HGX

1 Introduction

The goal of this article is to survey and synthesize the rapidly expanding econometric literature that studies the links between climate and conflict. Until the past decade, neither climate nor conflict have been core areas of inquiry within economics as a whole, and the same holds even for development economics, where their study is arguably most natural. As recently as 2007, when one of the authors carried out a survey of sixty-three development economics course syllabi (at both the undergraduate and graduate levels) in leading U.S. universities, only a handful of courses mentioned either conflict or climate, and leading development economics textbooks did not contain these words in their subject index.1 However, both of these topics have moved center stage over the past decade and are now widely taught and researched within economics and throughout the broader social science research community. This shift is at least partly a result of greater popular awareness of the critical role that climate might play in driving economic outcomes, and in particular rising public concern about climate change. Similarly, the violent aftermath of the Arab Spring revolutions, and the broader fact that conflict remains widespread in most low and middle income regions, have made it clear to many observers that economic development, political change and violent conflict are inextricably linked, and that armed conflict is not going away any time soon.

In this paper, we focus on over 50 quantitative papers that examine the link between climate and conflict using modern econometric methods that make credible attempts to draw causal inferences from data. Illustrating just how new research interest on these topics is within economics, the median year of publication among the studies we consider is 2012. While the quantitative literature on this topic is very recent, researchers working in other disciplines -- including archaeology, criminology, geography, history, political science, and psychology -- have long debated the extent to which climatic changes are responsible for conflict, violence or political instability (see for example, Huntington (1917); Levy (1995); Homer-Dixon (1999); Anderson et al. (2000); Davis (2001); DeMenocal (2001); Kuper and Kr?opelin (2006); Grove (2007); Scheffran et al. (2012); Gleditsch (2012)), and historians have connected prolonged periods of extreme climate with the collapse of major human civilizations (see Buckley et al. (2010); Cullen et al. (2000); Diamond (2005); Fagan (2000); Haug et al. (2003); Yancheva et al. (2007)). Numerous pathways linking the climate to these outcomes have been proposed. For example, climatic changes may alter the supply of natural resources and lead to disagreement over their allocation, or climatic conditions may shape the relative appeal of using violence versus cooperation to achieve an objective. Now, improvements in data availability, computing, and statistical methods have prompted an explosion of quantitative analyses seeking to test these theories and quantify the strength of these previously proposed linkages. A central goal of this piece is to make sense of this diverse and growing body of literature, and to chart a productive path forward for future research.

In this article, we use the terms climate and conflict and to describe broad classes of variables, and it is worth clarifying our use of these key terms up front.

1See Blattman and Miguel (2010) for details on the survey.

2

Climate We use climate to refer to observations of climatic variables: temperature, rainfall, and water availability, as well as climate indices that proxy for these measures, such the El Nin~o-Southern Oscillation index or the Palmer Drought Severity Index. These variables may be averaged over longer or shorter observational periods. Some authors argue that short averaging periods (e.g., annual) only describe the "weather" or "climate variability" and thus have little to say about the impact of "climate." We do not agree with this view. Societies experience climatic variables in continuous time and respond to both short-lived and long-term changes, making the frequency of short-lived events an economically-relevant feature of the climate. For example, if hot temperatures increase the likelihood of riots in a city ? even if hot temperatures are only experienced for a few hours ? then this is important for understanding climate impacts because the frequency of these momentary events may change if the distribution of daily temperatures changes.

We use the term conflict to describe events where regular patterns of dispute resoConflict lution fail. These events are usually violent in nature (although they need not be in all cases), they may involve individuals or groups, they may be organized or disorganized, and they may be personally, politically or otherwise motivated. While most existing empirical studies examine only one type of conflict at a time, in this review we examine this comprehensive set of outcomes because dierent types of conflict are potentially related and that their responses to climate might exhibit some commonalities. Our hope is that evaluating these phenomena together might help us better understand each individually.

The rest of this article is organized as follows. Secton 2 presents the existing evidence linking climate to conflict, and is the core of the paper. We begin the section by discussing the key methodological issues in estimating causal relationships in this area, and then survey the existing evidence across dierent types of conflict with particular attention to those studies capable of making credible causal claims. After collecting and standardizing estimated eect sizes across papers, we carry out a hierarchical Bayesian meta-analysis that both allows us to estimate the mean eect of climate variation on conflict outcomes and also quantifies the degree of variability in this eect size across studies. Section 3 lays out the leading theoretical mechanisms linking extreme conflict to conflict -- including both economic theories and non-economic explanations, such as those from psychology -- evaluates the limited body of empirical evidence regarding the importance of these channels and recommends methods for identifying causal pathways in future work. Section 4 discusses remaining challenges in this field, including data limitations, the challenge of understanding whether societies eectively adapt to climatic conditions, and the need to better understand the likely impacts of future global climate change. Section 5 concludes.

3

2 Evidence linking climate to conflict

Climatic conditions never cause conflict alone, but changes in climate can alter the conditions under which certain social interactions occur and thus have the potential to change the likelihood that conflict results. The situation is similar to the rise in car accident rates during rainy days. Car accidents themselves are almost always due to some form of driver or mechanical error; however, heavy rainfall may increase the probability of a critical error or the risk that a small error has cascading eects that in turn generate a crash (perhaps the car begins to fishtail, setting o a multi-car accident). Without the possibility of driver or mechanical errors, rainfall would have no eect on car accident rates, but without rainfall, there would still be some accidents. Similarly, climatic conditions are neither necessary nor su cient for conflicts to occur, but changes in climatic conditions could have measurable impact on the probability and intensity of conflict, holding other conflict-related factors fixed. The central empirical challenge addressed by the literature to date has been to quantify this eect.

2.1 The Empirical Problem

In an ideal experiment, we would observe two identical populations or societies, change the climate of one, and observe whether this "treatment" leads to more or less conflict relative to the "control" conditions. Because the climate cannot (yet) be experimentally manipulated, research has relied on natural experiments where plausibly exogenous variation in climatic variables generates changes in conflict risk that can be measured by an econometrician. The central challenge in this context is to identify plausibly homogenous populations, only some of which are naturally treated with a climatic event, that one can reasonably believe would behave similarly had neither been subject to a climate treatment (Freedman (1991); Holland (1986)).

2.1.1 Cross-sectional approaches

One approach to the above problem would be to assume that populations or societies inhabiting dierent locations are identical to one another in all respects except their climate, usually after regression adjustment for observable economic, social and political correlates of conflict. For example, Buhaug (2010a) compares the rate of civil war across dierent countries in Africa. It seems implausible that the conditions needed for causal inference are met in this setting: there are many ways in which populations and societies dier from one another (i.e., culture, history, etc.), many of them unobserved or hard to measure, so we cannot infer whether a climatic "treatment" has a causal eect or not (Angrist and Pischke (2008); Wooldridge (2002)). In the above example, the cross-sectional analysis by Buhaug (2010a) compares average rates of civil conflict in South Africa and Nigeria (among many comparisons), attributing observed dierences to the dierent climates of these countries ? despite the fact that there are many other relevant ways in which these countries dier. Hsiang and Meng (2014) revisit this example and explicitly test the assumption that no important omitted variables are missing from the analysis. Perhaps unsurprisingly, they strongly reject the assumption that baseline conflict rates in these countries are comparable, suggesting that they are unlikely to be valid counterfactuals for one another.

4

We take the critique by Hsiang and Meng (2014) seriously and argue that in general, the handful of covariates such as national per capita income or political indices that are commonly used in cross-sectional regression analyses are insu cient to credibly account for the numerous ways in which populations and societies dier from one another. Because the full suite of determinants of conflict are unknown and unmeasured, it is likely impossible that any cross-sectional study can explicitly account for all important dierences. For this reason, we do not draw causal inferences on the relationship between climate and conflict from cross-sectional analyses in this article, and instead rely on panel data approaches.

2.1.2 Identification in time series

Rather than presuming that all confounders are accounted for in a cross-sectional regression,

the bulk of recent studies estimate the eect of climate on conflict by using time-series variation

for identification, usually in a panel data context. In this research design, a single population

serves as both the "control" population ? e.g. just before a change in climatic conditions? and

the "treatment" population ? e.g. just after a change in climatic conditions. Inferences are

thus based on how a fixed population responds to dierent climatic conditions which vary over

time. Here the assumptions necessary for causal inference are more likely to be met, since the

structure, history and geography of comparison populations are nearly identical. Therefore, we

follow Hsiang, Burke, and Miguel (2013a) and restrict our attention in this review on studies

that use variation over time in a given location to study the climate/conflict relationship.

As pointed out by Hsiang and Burke (2013), the central shortcoming of this approach is

the frequency-identification tradeo that emerges because populations and societies evolve at a

much faster rate than many low-frequency climatic changes of interest. For example, if we are

interested in the eect of a climate change that takes one hundred years to manifest, then the

"control" and "treatment" populations in our sample must necessarily be roughly one hundred

years apart on average. However, human populations may change dramatically over one hundred

years, violating the assumption that the "control" and "treatment" populations are largely

comparable. This generates a direct tension between our ability to credibly identify causal

eects of climate and our ability to examine slow-moving climatic changes. Stated generally, for

an outcome Y observed at time t, conditional on contemporaneous climatic conditions C , the

t

t

estimate for the eect of a change after a time interval t is

^ = E[Y |C ] E[Y |C ].

(1)

t+ t t+ t

tt

This estimate approaches the true parameter of interest

= E[Y |C ] E[Y |C ]

(2)

t t+ t

tt

so long as Y is comparable to Y conditional on C (and possibly other covariates). This is

t

t+ t

the identifying assumption of this research design. However, as the frequency 1 of the climatic

t

variation of interest becomes lower (climate changes become more gradual) t becomes larger

and the assumption that Y and Y are comparable becomes increasingly di cult to justify.

t

t+ t

5