Quantifying the Influence of Climate on Human Conflict ...users.clas.ufl.edu/prwaylen/Geo3280articles/Climate... · East Africa Civil conflict onset Global tropics Civil war incidence

DOI 101126science1235367 (2013)341 Science

et alSolomon M HsiangQuantifying the Influence of Climate on Human Conflict

This copy is for your personal non-commercial use only

clicking herecolleagues clients or customers by you can order high-quality copies for yourIf you wish to distribute this article to others

herefollowing the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

) September 12 2013 wwwsciencemagorg (this information is current as of

The following resources related to this article are available online at

httpwwwsciencemagorgcontent34161511235367fullhtmlversion of this article at

including high-resolution figures can be found in the onlineUpdated information and services

httpwwwsciencemagorgcontentsuppl20130731science1235367DC1html can be found at Supporting Online Material

httpwwwsciencemagorgcontent34161511235367fullhtmlref-list-1 40 of which can be accessed freecites 98 articlesThis article

CopyrightAmerican Association for the Advancement of Science 1200 New York Avenue NW Washington DC 20005 (print ISSN 0036-8075 online ISSN 1095-9203) is published weekly except the last week in December by theScience

13 SEPTEMBER 2013 VOL 341 SCIENCE wwwsciencemagorg

Quantifying the Infl uence of Climate on Human Confl ictSolomon M Hsiang Marshall Burke Edward Miguel

Introduction Despite the existence of institutions designed to promote peace interactions between individuals and groups sometimes lead to confl ict Understanding the causes of such confl ict is a major project in the social sciences and researchers in anthropology economics geography history political science psychology and sociology have long debated the extent to which climatic changes are responsible Recent advances and interest have prompted an explosion of quantitative studies on this question

Methods We carried out a comprehensive synthesis of the rapidly growing literature on climate and human confl ict We examined many types of human confl ict ranging from interpersonal violence and crime to intergroup violence and political instability and further to institutional breakdown and the collapse of civilizations We focused on quantitative studies that can reliably infer causal associations between climate variables and confl ict outcomes The studies we examined are experi-ments or ldquonatural experimentsrdquo the latter exploit variations in climate over time that are plausibly independent of other variables that also affect confl ict In many cases we obtained original data from studies that did not meet this criterion and used a common statistical method to reanalyze these data In total we evaluated 60 primary studies that have examined 45 different confl ict data sets We collected fi ndings across time periods spanning 10000 BCE to the present and across all major world regions

Results Deviations from normal precipitation and mild temperatures systematically increase the risk of confl ict often substantially This relationship is apparent across spatial scales ranging from a single building to the globe and at temporal scales ranging from an anomalous hour to an anoma-lous millennium Our meta-analysis of studies that examine populations in the post-1950 era sug-gests that the magnitude of climatersquos infl uence on modern confl ict is both substantial and highly statistically signifi cant (P lt 0001) Each 1-SD change in climate toward warmer temperatures or more extreme rainfall increases the frequency of interpersonal violence by 4 and intergroup confl ict by 14 (median estimates)

Discussion We conclude that there is more agreement across studies regarding the infl uence of cli-mate on human confl ict than has been recognized previously Given the large potential changes in precipitation and temperature regimes projected for the coming decadesmdashwith locations through-out the inhabited world expected to warm by 2 to 4 SDs by 2050mdashamplifi ed rates of human confl ict could represent a large and critical social impact of anthropogenic climate change in both low- and high-income countries

FIGURES AND TABLE IN THE FULL ARTICLE

Fig 1 Samples and spatiotemporal resolu-tions of 60 studies examining intertemporal associations between climatic variables and human confl ict

Fig 2 Empirical studies indicate that clima-tological variables have a large effect on the risk of violence or instability in the modern world

Fig 3 Examples of paleoclimate reconstruc-tions that fi nd associations between climatic changes and human confl ict

Fig 4 Modern empirical estimates for the effect of climatic events on the risk of inter-personal violence

Fig 5 Modern empirical estimates for the effect of climatic events on the risk of inter-group confl ict

Fig 6 Projected temperature change by 2050 as a multiple of the local historical SD (σ) of temperature

Table 1 Primary quantitative studies testing for a relationship between climate and con-fl ict violence or political instability

SUPPLEMENTARY MATERIALS

Supplementary TextFigs S1 to S4Tables S1 to S4References (140 141)

minus2 minus1 0 1 2

minus40

∆ Pixel temp (degC)

t risk (

minus1 0 1 2∆ Pacific Ocean temp (degC)

minus05 0 05 1∆ Country temp (degC)

Local violence

East Africa

Civil conflict onset

Global tropics

Civil war incidence

Sub-Saharan Africa

Climate and confl ict across spatial scales Evidence that temperature infl uences the risk of modern human confl ict (A) local violence in 1deg grid cells (B) civil war in countries and (C) civil confl ict risk in the tropics The map depicts regions of analysis corresponding to nonparametric watercolor regressions in (A) to (C) The color intensity in (A) to (C) indicates the level of certainty in the regression line

READ THE FULL ARTICLE ONLINEhttpdxdoiorg101126science1235367

Cite this article as S Hsiang et al Science 341 1235367 (2013) DOI 101126science1235367

The list of author affi liations is available in the full article onlineCorresponding author E-mail shsiangberkeleyedu

RESEARCH ARTICLE SUMMARY

Published by AAAS

Quantifying the Influence of Climateon Human ConflictSolomon M Hsiang12daggerDagger Marshall Burke3dagger Edward Miguel24

A rapidly growing body of research examines whether human conflict can be affected byclimatic changes Drawing from archaeology criminology economics geography history politicalscience and psychology we assemble and analyze the 60 most rigorous quantitative studiesand document for the first time a striking convergence of results We find strong causal evidencelinking climatic events to human conflict across a range of spatial and temporal scales andacross all major regions of the world The magnitude of climatersquos influence is substantial foreach one standard deviation (1s) change in climate toward warmer temperatures or more extremerainfall median estimates indicate that the frequency of interpersonal violence rises 4 andthe frequency of intergroup conflict rises 14 Because locations throughout the inhabitedworld are expected to warm 2s to 4s by 2050 amplified rates of human conflict could represent alarge and critical impact of anthropogenic climate change

Human behavior is complex and despitethe existence of institutions designed topromote peace interactions between in-

dividuals and groups sometimes lead to conflictWhen such conflict becomes violent it can havedramatic consequences on humanwell-beingMor-tality fromwar and interpersonal violence amountsto 05 to 1 million deaths annually (1 2) withnonlethal impacts including injury and lost eco-nomic opportunities affecting millions more Be-cause the stakes are so high understanding thecauses of human conflict has been a major projectin the social sciences

Researchers working across multiple dis-ciplines including archaeology criminology eco-nomics geography history political science andpsychology have long debated the extent to whichclimatic changes are responsible for causing con-flict violence or political instability Numerouspathways linking the climate to these outcomeshave been proposed For example climatic changesmay alter the supply of a resource and causedisagreement over its allocation or climatic con-ditions may shape the relative appeal of usingviolence or cooperation to achieve some precon-ceived objective Qualitative researchers have awell-developed history of studying these issues(3ndash7) dating back at least to the start of the 20thcentury (8) Yet in recent years growing recog-nition that the climate is changing coupled with

improvements in data quality and computing hasprompted an explosion of quantitative analysesseeking to test these theories and quantify thestrength of these previously proposed linkagesThus far this work has remained scattered acrossmultiple disciplines and has been difficult to syn-thesize given the disparate methodologies dataand interests of the various research teams

Here we assemble the first comprehensivesynthesis of this rapidly growing quantitative lit-eratureWe adopt a broad definition of ldquoconflictrdquousing the term to encompass a range of outcomesfrom individual-level violence and aggression tocountry-level political instability and civil warWe then collect all available candidate studies andguided by previous criticisms that not all corre-lations imply causation (9ndash11) focus on onlythose quantitative studies that can reliably infercausal associations (9 12) between climate var-iables and conflict outcomes The studies weexamine exploit either experimental or natural-experimental variation in climate the latter termrefers to variation in climate over time that isplausibly independent of other variables that alsoaffect conflict To meet this standard studies mustaccount for unobservable confounding factorsacross populations as well as for unobservabletime-trending factors that could be correlated withboth climate and conflict (13) In many cases weobtained data from studies that did not meet thiscriterion and reanalyzed it with a common sta-tistical model that did meet the criterion (see sup-plementary materials) The importance of thisrigorous approach is highlighted by an exam-ple in which our standardized analysis generatedfindings consistent with other studies but at oddswith the original conclusions of the study in ques-tion (14)

In total we obtained 60 primary studies thateither met this criterion or were reanalyzed with amethod that met this criterion (Table 1) Collect-ively these studies analyze 45 different conflict

data sets published across 26 different journalsand represent the work of more than 190 re-searchers from around the world Our evaluationsummarizes the recent explosion of research onthis topic with 78 of studies released since2009 and the median study released in 2011 Wecollected findings across a wide range of conflictoutcomes time periods spanning 10000 BCE tothe present day and all major regions of theworld (Fig 1)

Although various conflict outcomes differ inimportant ways we find that the behavior ofthese outcomes relative to the climate system ismarkedly similar Put most simply we find thatlarge deviations from normal precipitation andmild temperatures systematically increase the riskof many types of conflict often substantially andthat this relationship appears to hold over a varie-ty of temporal and spatial scales Ourmeta-analysisof studies that examine populations in the post-1950 era suggests that these relationships contin-ue to be highly important in the modern worldalthough there are notable differences in the mag-nitude of the relationshipwhen different variablesare considered The standardized effect of tem-perature is generally larger than the standardizedeffect of rainfall and the effect on intergroupviolence (eg civil war) is larger than the effecton interpersonal violence (eg assault) We con-clude that there is substantially more agreementand generality in the findings of this burgeoningliterature than has been recognized previouslyGiven the large potential changes in precipitationand temperature regimes projected for the comingdecades our findings have important implicationsfor the social impact of anthropogenic climatechange in both low- and high-income countries

Estimation of Climate-Conflict LinkagesReliably measuring an effect of climatic condi-tions on human conflict is complicated by the in-herent complexity of social systems In particulara central concern is whether statistical relation-ships can be interpreted causally or if they areconfounded by omitted variables To address thisconcern we restrict our attention to studies withresearch designs that are scientific experiments orthat approximate one (ie ldquonatural experimentsrdquo)After describing how studies meet this criterionwe discuss how we interpret the precision of re-sults assess the importance of climatic factorsand address choices over functional form

Research DesignIn an ideal experiment we would observe twoidentical populations change the climate of oneand observe whether this treatment leads to moreor less conflict relative to the control conditionsBecause the climate cannot be experimentally ma-nipulated researchers primarily rely on naturalexperiments in which a given population is com-pared to itself at different moments in time whenit is exposed to different climatic conditionsmdashconditions that are exogenously determined by

RESEARCHARTICLE

1Program in Science Technology and Environmental PolicyWoodrow Wilson School of Public and International AffairsPrinceton University Princeton NJ 08544 USA 2National Bu-reauof Economic Research CambridgeMA02138USA 3Depart-ment of Agricultural and Resource Economics University ofCalifornia Berkeley Berkeley CA 94720 USA 4Department ofEconomics University of California Berkeley Berkeley CA94720 USA

Present address Goldman School of Public Policy Universityof California Berkeley Berkeley CA 94720 USAdaggerThese authors contributed equally to this workDaggerCorresponding author E-mail shsiangberkeleyedu

wwwsciencemagorg SCIENCE VOL 341 13 SEPTEMBER 2013 1235367-1

Table 1 Primary quantitative studies testing for a relationship betweenclimate and conflict violence or political instability ldquoStat testrdquo is Y if theanalysis uses formal statistical methods to quantify the influence of climatevariables and uses hypothesis testing procedures (Y yes N no) ldquoLarge effectrdquois Y if the point estimate for the effect size is considered substantial by theauthors or is greater in magnitude than 10 of the mean risk level for a 1s

change in climate variables ldquoReject b =0rdquo is Y if the study rejects an effect sizeof zero at the 95 confidence level ldquoReject b = 10rdquo is Y if the study is ableto reject the hypothesis that the effect size is larger than 10 of the mean risklevel for a 1s change in climate variables ndash not applicable SSA sub-SaharanAfrica PDSI Palmer Drought Severity Index ENSO El NintildeondashSouthern Os-cillation NAO North Atlantic Oscillation N Hem Northern Hemisphere

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10

Interpersonal conflict (15)Anderson et al 2000 1950ndash1997 USA Annual Country Temp Violent crime Y Y Y ndash (34)Auliciems et al 1995dagger 1992 Australia Week Municipality Temp Domestic violence Y Y Y ndash (29)Blakeslee et al 2013 1971ndash2000 India Annual Municipality Rain Violent and

property crimeY Y Y ndash (42)

Card et al 2011daggerDagger 1995ndash2006 USA Day Municipality Temp Domestic violence Y Y Y ndash (37)Cohn et al 1997sect 1987ndash1988 USA Hours Municipality Temp Violent crime Y Y Y ndash (30)Jacob et al 2007dagger 1995ndash2001 USA Week Municipality Temp Violent and

Kenrick et al 1986para 1985 USA Day Site Temp Hostility Y Y Y ndash (27)Larrick et al 2011daggerDagger 1952ndash2009 USA Day Site Temp Violent retaliation Y Y Y ndash (36)Mares 2013 1990ndash2009 USA Month Municipality Temp Violent crime Y Y Y ndash (39)Miguel 2005daggerDagger 1992ndash2002 Tanzania Annual Municipality Rain Murder Y Y N N (40)Mehlum et al 2006 1835ndash1861 Germany Annual Province Rain Violent and

Ranson 2012dagger 1960ndash2009 USA Month County Temp Personal violence Y Y Y ndash (38)Rotton et al 2000sect 1994ndash1995 USA Hours Municipality Temp Violent crime Y Y Y ndash (31)Sekhri et al 2013dagger 2002ndash2007 India Annual Municipality Rain Murder and

domestic violenceY Y Y ndash (41)

Vrij et al 1994para 1993 Netherlands Hours Site Temp Police use of force Y Y Y ndash (28)Intergroup conflict (30)

Almer et al 2012 1985ndash2008 SSA Annual Country Raintemp Civil conflict Y Y N N (65)Anderson et al 2013 1100ndash1800 Europe Decade Municipality Temp Minority expulsion Y Y Y ndash (63)Bai et al 2010 220ndash1839 China Decade Country Rain Transboundary Y Y Y ndash (50)Bergholt et al 2012Dagger 1980ndash2007 Global Annual Country Floodstorm Civil conflict Y N N Y (75)Bohlken et al 2011 1982ndash1995 India Annual Province Rain Intergroup Y Y N N (44)Buhaug 2010 1979ndash2002 SSA Annual Country Temp Civil conflict Y N N N (22)Burke 2012Dagger 1963ndash2001 Global Annual Country Raintemp Political instability Y Y N N (71)Burke et al 2009Daggerdaggerdagger 1981ndash2002 SSA Annual Country Temp Civil conflict Y Y Y ndash (64)Cervellati et al 2011 1960ndash2005 Global Annual Country Drought Civil conflict Y Y Y ndash (54)Chaney 2011 641ndash1438 Egypt Annual Country Nile floods Political Instability Y Y Y ndash (70)Couttenier et al 2011 1957ndash2005 SSA Annual Country PDSI Civil conflict Y Y Y ndash (53)Dell et al 2012 1950ndash2003 Global Annual Country Temp Political instability

and civil conflictY Y Y ndash (21)

Fjelde et al 2012Dagger 1990ndash2008 SSA Annual Province Rain Intergroup Y Y N N (55)Harari et al 2013 1960ndash2010 SSA Annual Pixel (1deg) Drought Civil conflict Y Y Y ndash (52)Hendrix et al 2012Dagger 1991ndash2007 SSA Annual Country Rain Intergroup Y Y Y ndash (46)Hidalgo et al 2010Dagger 1988ndash2004 Brazil Annual Municipality Rain Intergroup Y Y Y ndash (25)Hsiang et al 2011 1950ndash2004 Global Annual World ENSO Civil conflict Y Y Y ndash (51)Jia 2012 1470ndash1900 China Annual Province Droughtflood Peasant rebellion Y Y Y ndash (56)Kung et al 2012 1651ndash1910 China Annual County Rain Peasant rebellion Y Y Y ndash (47)Lee et al 2013 1400ndash1999 Europe Decade Region NAO Violent conflict Y Y Y ndash (57)Levy et al 2005Dagger 1975ndash2002 Global Annual Pixel (25deg) Rain Civil conflict Y Y N N (49)Maystadt et al 2013 1997ndash2009 Somalia Month Province Temp Civil conflict Y Y Y ndash (66)Miguel et al 2004DaggerDagger 1979ndash1999 SSA Annual Country Rain Civil war Y Y Y ndash (48)OrsquoLaughlin et al 2012Dagger 1990ndash2009 E Africa Month Pixel (1deg) Raintemp Civilintergroup Y Y Y ndash (23)Salehyan et al 2012 1979ndash2006 Global Annual Country PDSI Civilintergroup Y Y Y ndash (76)Sarsons 2011 1970ndash1995 India Annual Municipality Rain Intergroup Y Y Y ndash (45)Theisen et al 2011Dagger 1960ndash2004 Africa Annual Pixel (05deg) Rain Civil conflict Y N N N (24)Theisen 2012Dagger 1989ndash2004 Kenya Annual Pixel (025deg) Raintemp Civilintergroup Y Y N N (14)Tol et al 2009 1500ndash1900 Europe Decade Region Raintemp Transboundary Y Y Y ndash (60)Zhang et al 2007sectsect 1400ndash1900 N Hem Century Region Temp Instability Y Y Y ndash (59)

Institutional breakdown and population collapse (15)Bruumlckner et al 2011 1980ndash2004 SSA Annual Country Rain Inst change Y Y Y ndash (78)

Continued on next page

13 SEPTEMBER 2013 VOL 341 SCIENCE wwwsciencemagorg1235367-2

RESEARCH ARTICLE

the climate system (9 15) In this researchdesign a single population serves as both thecontrol population (eg just before a change inclimatic conditions) and the treatment population(eg just after a change in climatic conditions)Thus inferences are based only on how a fixedpopulation responds to different climatic condi-tions that vary over time and time-series or lon-gitudinal analysis is used to construct a credibleestimate for the causal effect of climate on con-flict (12 15 16)

To minimize statistical bias and improve thecomparability of studies we focus on studies thatuse versions of the general model

conflict variableit frac14 b climate variableit thornmi thorn qt thorn isinit eth1THORN

where locations are indexed by i observationalperiods are indexed by t b is the parameter ofinterest and isin is the error If different locationsin a sample exhibit different average levels of

conflictmdashperhaps because of cultural historicalpolitical economic geographic or institutionaldifferences between the locationsmdashthis will beaccounted for by the vector of location-specificconstants m (commonly known as ldquofixed effectsrdquo)The vector of time-specific constants q (a dum-my for each time period) flexibly accounts forother time-trending variables such as economicgrowth or gradual demographic changes that couldbe correlated with both climate and conflict Insome cases such as in time series the qt parameters

municipal

province

country

region

globalS

tial s

Duration of climatic event (log scale)

Hsiang et al(2011)

8000 BCE 0 1000 1800 1950 2000 2010

Years in study (log scale)

Kuper ampKroumlepelin

(2006)

Vrij et al (1994)

Americas

AustraliaEurasia amp

DrsquoAngou et al(2012)

Africa

Global

Fig 1 Samples and spatiotemporal resolutions of 60 studies examiningintertemporal associations between climatic variables and human con-flict (A) The location of each study region (y axis) plotted against the period oftime included in the study (x axis) The x axis is scaled according to log years beforethe present but is labeled according to the year of the common era (B) The level

of aggregation in social outcomes (y axis) plotted against the time scale of climaticevents (x axis) The envelope of spatial and temporal scales where associations aredocumented is shaded with studies at extreme vertices labeled for referenceMarker size indicates the number of studies at each location with the smallestbubbles marking individual studies and the largest bubble denoting 10 studies

Sampleperiod

Sampleregion

Timeunit

Spatialunit

Indepen-dent

variable

Dependentvariable

Stattest

Larg-eef-fect

Re-jectb = 0

Rejectb = 10 Re-

Buckley et al 2010 1030ndash2008 Cambodia Decade Country Drought Collapse N ndash ndash ndash (85)Buumlntgen et al 2011 400 BCEndash2000 Europe Decade Region Raintemp Instability N ndash ndash ndash (62)Burke et al 2010Dagger 1963ndash2007 Global Annual Country Raintemp Inst change Y Y Y ndash (77)Cullen et al 2000 4000 BCEndash0 Syria Century Country Drought Collapse N ndash ndash ndash (83)DrsquoAnjou et al 2012 550 BCEndash1950 Norway Century Municipality Temp Collapse Y Y Y ndash (89)Ortloff et al 2003 500ndash2000 Peru Century Country Drought Collapse N ndash ndash ndash (80)Haug et al 2003 0ndash1900 Mexico Century Country Drought Collapse N ndash ndash ndash (84)Kelly et al 2013 10050 BCEndash1950 USA Century State Temprain Collapse Y Y Y ndash (88)Kennett et al 2012 40 BCEndash2006 Belize Decade Country Rain Collapse N ndash ndash ndash (87)Kuper et al 2006 8000ndash2000 BCE N Africa Millennia Region Rain Collapse N ndash ndash ndash (81)Patterson et al 2010 200 BCEndash1700 Iceland Decade Country Temp Collapse N ndash ndash ndash (86)Stahle et al 1998 1200ndash2000 USA Multiyear Municipality PDSI Collapse N ndash ndash ndash (82)Yancheva et al 2007 2100 BCEndash1700 China Century Country Raintemp Collapse N ndash ndash ndash (79)Zhang et al 2006 1000ndash1911 China Decade Country Temp Civil conflict

and collapseY Y Y ndash (58)

Number of studies (60 total) 50 47 37 1Fraction of those using statistical tests 100 94 74 2

Also see (33) daggerShown in Fig 4 DaggerReanalyzed using the common statistical model containing location fixed effects and trends (see supplementary materials) sectAlso see discussionin (32) Shown in Fig 2 paraActual experiment Shown in Fig 5 Effect size in the study is statistically significant at the 10 level but not at the 5 level daggerdaggerAlso seediscussion in (22 132ndash137) DaggerDaggerAlso see discussion in (138 139) sectsectAlso see (61) Shown in Fig 3

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10 Ref

RESEARCH ARTICLE

may be replaced by a generic trend (eg q t)that is possibly nonlinear and is either commonto all locations or may be location-specific (egqi t) Our conclusions from the literature arebased only on those studies that implement Eq 1or one of the mentioned alternatives In selectcases when studies did not meet this criterionbut the data from these analyses were publiclyavailable or supplied by the authors we usedthis common method to reanalyze the data (seesupplementary materials) Many estimates ofEq 1 in the literature and in our reanalysis ac-count for temporal andor spatial autocorrelationin the error term isin although this adjustment was

not considered a requirement for inclusion hereIn the case of some paleoclimatological and ar-chaeological studies formal statistical analysis isnot implemented because the outcome variablesof interest are essentially singular cataclysmicevents However we include these studies becausethey follow populations over time at a fixed lo-cation and are thus implicitly using the model inEq 1 (these cases are noted in Table 1)

We do not consider studies that are purelycross-sectional that is studies that only comparerates of conflict across different locations andattribute differences in average levels of conflictto average climatic conditions Populations differ

from one another in numerous ways (culture his-tory etc) many of them unobserved and theseldquoomitted variablesrdquo are likely to confound theseanalyses In the language of the natural experi-ment the treatment and control populations inthese analyses are not comparable units so wecannot infer whether a climatic treatment has acausal effect or not (12 13 15ndash17) For examplea cross-sectional study might compare averagerates of civil conflict in Norway and Nigeriaattributing observed differences to the differentclimates of these countries despite the fact thatthere are clearlymany other relevant ways inwhichthese countries differ Nonetheless some studies

minus30 minus20 minus10 0 10 20 30minus60

minus40

minus20

Civil conflict onset (Global)Pixelminusbyminusyear N = 3809432

Levy et al (2005)

Pixel Standardized Precipitation Loss(Weighted Anomaly Index)

minus10 0 10

minus20

minus10

Rape (USA)Countyminusbyminusmonth N = 1434832

Ranson (JEEM in-press)

Mean Daily Max County Temperature(Anomaly in ˚C)

minus20 minus10 0 10 20

minus20

minus10

Violent interminusgroup retalitation (USA)Playminusbyminusday N = 595500

Larrick et al (PS 2011)

Stadium temperature(Anomaly in ˚C)

rminuste

minus2 minus1 0 1 2

minus50

Political amp interminusgroup violence (East Africa)Pixelminusbyminusmonth N = 91656

OrsquoLaughlin et al (PNAS 2012)

Pixel Temperature(Anomaly in ˚C)

minus100

minus50

Political amp interminusgroup violence (Kenya)Pixelminusbyminusyear N = 13520

Theisen (JPR 2012)

minus05 0 05 1

minus20

Civil war incidence (Africa)Countryminusbyminusyear N = 1049

Burke et al (PNAS 2009)

Country Temperature(Anomaly in ˚C)

minus1 0 1

minus20

minus10

Political leader exit (Global)Countryminusbyminusyear N = 5491

Burke (BEJM 2012)

minus1 0 1 2

minus20

Redistributive interminusgroup conflict (Brazil)Municipalityminusbyminusyear N = 50521

Hidalgo et al (REStat 2010)

Municipality Rainfall Deviation(Absolute value in σ)

minus1 0 1 2 3minus20

Riot political amp interminusgroup violence (Africa)Countryminusbyminusyear N = 1344

Hendrix amp Salehyan (JPR 2012)

Country Rainfall Deviation(Absolute value in σ)

( change from prior year)

minus1 0 1 2

minus20

Civil conflict onset (Global tropics)Annual observations N = 54Hsiang et al (Nature 2011)

Pacific Ocean Temperature(NINO3 index MayminusDec in ˚C)

minus40 minus20 0 20

minus80

minus40

Interminusgroup riots (India)Stateminusbyminusyear N = 206

Bohlken amp Sergenti (JPR 2010)

State Rainfall Losses

minusM

minus5

Violent personal crime (USA)Jurisdictionminusbyminusweek N = 26567

Jacob et al (JHR 2007)

Jurisdiction Temperature(Anomaly in ˚C)

Fig 2 Empirical studies indicate that climatological variables have alarge effect on the risk of violence or instability in the modern world(A to L) Examples from studies of modern data that identify the causal effect ofclimate variables on human conflict Both dependent and independent variableshave had location effects and trends removed so all samples have a mean of zeroRelationships between climate and conflict outcomes are shown with nonpara-metric watercolor regressions where the color intensity of 95 CIs depicts the like-lihood that the true regression line passes through agiven value (darker ismore likely)(128) The white line in each panel denotes the conditional mean (129 130)

Climate variables are indicated by color red temperature green rainfall deviationsfrom normal blue precipitation loss black ENSO Panel titles describe theoutcome variable location unit of analysis sample size and study Because thesamples examined in each study differ the units and scales change across eachpanel (see Figs 4 and 5 for standardized effect sizes) ldquoRainfall deviationrdquorepresents the absolute value of location-specific rainfall anomalies with bothabnormally high and abnormally low rainfall events described as having a largerainfall deviation ldquoPrecipitation lossrdquo is an index describing how much lowerprecipitation is relative to the prior yearrsquos amount or the long-term mean

RESEARCH ARTICLE

use cross-sectional analyses and attempt to con-trol for confounding variables in regression analy-ses typically using a handful of covariates suchas average income or political indices Howeverbecause the full suite of determinants of conflictis unknown and unmeasured it is probably im-possible that any cross-sectional study can explic-itly account for all important differences between

populations Rather than presuming that all con-founders are accounted for the studies we eval-uate compareNorway orNigeria only to themselvesat different moments in time thereby ensuring thatthe structure history and geography of compar-ison populations are nearly identical

Some studies implement versions of Eq 1 thatare expanded to explicitly control for potential

confounding factors such as average income Inmany cases this approach is more harmful thanhelpful because it introduces bias in the coef-ficients describing the effect of climate on con-flict This problem occurswhen researchers controlfor variables that are themselves affected by cli-mate variation causing either (i) the signal in theclimate variable of interest to be inappropriately

Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Empire of Angkor city-state collapses

Mexico

Major phases of collapse forthe ldquoClassicrdquo Mayan empire

(Wet) 40

(Dry) 20

Major dynasties collapse and transition

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

minus3000 minus2500 minus2000 minus1500 minus1000 minus500 0 500 1000 1500 2000Years BCECE

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

Akkadian empirecollapses

FRoman

conquestGreat Famineamp Black Death

Fig 3 Examples of paleoclimate reconstructions that find associationsbetween climatic changes and human conflict Lines are climate recon-structions (red temperature blue precipitation orange drought smoothedmoving averages when light gray lines are shown) and dark gray bars indicateperiods of substantial social instability violent conflict or the breakdown ofpolitical institutions (A) Alluvial sediments from the Cariaco Basin indicate sub-stantial multiyear droughts coinciding with the collapse of the Maya civilization(84) (B) Reconstruction of a drought index from tree rings in Vietnam thePalmer drought severity index (PDSI) shows sustainedmegadroughts prior to thecollapse of the Angkor kingdom (85) (C) Sediments from Lake Huguang Maar inChina indicate abrupt and sustained periods of reduced summertime

precipitation that coincided with most major dynastic transitions (79) Thecollapse of the Tang Dynasty (907) coincided with the terminal collapse of theMaya (A) both of which occurred when the Pacific Ocean altered rainfall patternsin both hemispheres (79) Similarly the collapse of the Yuan Dynasty (1368)coincided with collapse of Angkor (B) which shares the same regional climate(D) Tiwanaku cultivation of the Lake Titicaca region ended abruptly after a dryingof the region as measured by ice accumulation in the Quelccaya Ice Cap Peru(80) (E) Continental dust blown from Mesopotamia into the Gulf of Omanindicates terrestrial drying that is coincident with the collapse of the Akkadianempire (83) (F) European tree rings indicate that anomalously cold periods wereassociated with major periods of instability on the European continent (62)

RESEARCH ARTICLE

absorbed by the control variable or (ii) the esti-mate to be biased because populations differ inunobserved ways that become artificially cor-related with climate when the control variable isincluded Thismethodological error is commonlytermed ldquobad controlrdquo (12) andwe exclude resultsobtained using this approach The difficulty inthis setting is that climatic variables affect manyof the socioeconomic factors commonly includedas control variables things like crop productioninfant mortality population (via migration or mor-tality) and even political regime type To theextent that these outcome variables are used ascontrols in Eq 1 studies might draw mistakenconclusions about the relationship between cli-mate and conflict Because this error is so salientin the literature we provide examples below Afull treatment can be found in (12 18)

For an example of (i) consider whether var-iation in temperature increases conflict In manystudies of conflict researchers often employ astandard set of controls that are correlates of con-flict such as per capita income However evi-dence suggests that income is itself affected bytemperature (19ndash21) so if part of the effect oftemperature on conflict is through income thencontrolling for income in Eq 1 will lead the

researcher to underestimate the role of temper-ature in conflict This occurs because much ofthe effect of temperature will be absorbed by theincome variable biasing the temperature coeffi-cient toward zero At the extreme if temperatureinfluences conflict only through income then con-trolling for income would lead the researcher inthis example to draw exactly the wrong conclusionabout the relationship between temperature andconflict that there is no effect of temperature onconflict

For an example of (ii) imagine that a measureof politics (ie democracy) and temperature bothhave a causal effect on conflict and both poli-tics and temperature have an effect on incomebut that income has no effect on conflict If poli-tics and temperature are uncorrelated estimatesof Eq 1 that do not control for politics will stillrecover the unbiased effect of temperature How-ever if income is introduced to Eq 1 as a controlbut politics is left out of the model perhaps be-cause it is more difficult to measure then therewill appear to be an association between incomeand conflict because income will be serving asa proxy measure for politics In addition this ad-justment to Eq 1 also biases the estimated effectof temperature This bias occurs because the types

of countries that have high income when tem-perature is high are different in terms of their av-erage politics from those countries that have highincomewhen temperature is low Thus if incomeis held fixed as a control variable in a regressionmodel the comparison of conflict across temper-atures is not an ldquoapples-to-applesrdquo comparison be-cause politics will be systematically different acrosscountries at different temperatures generating abias that can have either sign In this examplethe inclusion of income in the model leads to twoincorrect conclusions It biases the estimated rela-tionship between climate and conflict and impli-cates income as playing a role in conflict when itdoes not

Statistical PrecisionWe consider each studyrsquos estimated relationshipbetween climate and conflict as well as the esti-matersquos precision Because sampling variability andsample sizes differ across studies some analysespresent results that are more precise than otherstudies Recognizing this fact is important whensynthesizing a diverse literature as some appar-ent differences between studies can be reconciledby evaluating the uncertainty in their findingsFor example some studies report associations that

minus8

minus4

minus8

minus4

liatio

minusminusminusminusminusminusminusminusminus

minusminusminusminusminusminusminus

Allstudies

Temperaturestudies

Median = 39

Mean = 23

Fig 4 Modern empirical estimates for the effect of climatic events onthe risk of interpersonal violence Each marker represents the estimatedeffect of a 1s increase in a climate variable expressed as a percentage changein the outcome variable relative to its mean Whiskers represent the 95 CIon this point estimate Colors indicate the forcing climate variable Acoefficient is positive if conflict increases with higher temperature (red)greater rainfall loss (blue) or greater rainfall deviation from normal (green)The dashed line indicates the median estimate the top solid black line denotes

the precision-weighted mean with its 95 CI shown in gray The panels onthe right show the precision-weighted mean effect (circles) and thedistribution of study results for all 11 results looking at individual conflict orfor the subset of 8 results focusing on temperature effects Distributions ofeffect sizes are either precision-weighted (solid lines) or derived from aBayesian hierarchical model (dashed lines) See the supplementary materialsfor details on the individual studies and the calculation of mean effects andtheir distribution

RESEARCH ARTICLE

are very large or very small but with uncertaintiesthat are also very large leading us to place lessconfidence in these extreme findings This intui-tion is formalized in our meta-analysis which ag-gregates results across studies by down-weightingresults that are less precisely estimated

The strength of a finding is sometimes sum-marized in a statement regarding its statisticalsignificance which describes the signal-to-noiseratio in an individual study However in prin-ciple the signal is a relationship that exists in thereal world and cannot be affected by the researcherwhereas the level of noise in a given studyrsquos

finding (ie its uncertainty) is a feature specificto that studymdasha feature that can be affected by aresearchers decisions such as the size of the sam-ple they choose to analyze Thus although it isuseful to evaluate whether individual findings arestatistically significant and it is important to down-weight highly imprecise findings individual studiesprovide useful information even when their find-ings are not statistically significant

To summarize the evidence that each statisti-cal study provides while also taking into accountits precision we separately consider three ques-tions for each study in Table 1 (i) Is the estimated

average effect of climate on conflict quantitative-ly ldquolargerdquo in magnitude (discussed below) regard-less of its uncertainty (ii) Is the reported effectlarge enough and estimated with sufficient pre-cision that the study can reject the null hypothesisof ldquono relationshiprdquo at the 5 level (iii) If thestudy cannot reject the hypothesis of ldquono rela-tionshiprdquo can it reject the hypothesis that therelationship is quantitatively large In the litera-ture often only the second question is evaluatedin any single analysis Yet it is important toconsider the magnitude of climate influence (firstquestion) separately from its statistical precision

1 2 3 40

Standard deviations

Fig 6 Projected temperature change by 2050 as a multiple of the localhistorical SD (s) of temperature Temperature projections are for the A1Bscenario and are averaged across 21 global climate models reporting in theCoupled Model Intercomparison Project (CMIP3) (96) Changes are the difference

between projected annual average temperatures in 2050 and average temper-atures in 2000 The historical SD of temperature is calculated fromannual averagetemperatures at each grid cell over the period 1950ndash2008 using data from theUniversity of Delaware (131) The map is an equal-area projection

minus20

minus10

minus20

minus10

t al 2

itutio

minusminusminusminusminusminusminusminusminusminusminusminus

minusminusminusminusminus

minusminusminusminusminusminus

Allstudies

Temperaturestudies

Median = 136Mean = 111

Fig 5 Modern empirical estimates for the effect of climatic events onthe risk of intergroup conflict Eachmarker represents the estimated effectof a 1s increase in a climate variable expressed as a percentage change in theoutcome variable relative to its mean Whiskers represent the 95 CI on thispoint estimate Colors indicate the forcing climate variable A coefficient ispositive if conflict increases with higher temperature (red) greater rainfall loss(blue) greater rainfall deviation from normal (green) more floods and storms(gray) more El Nintildeondashlike conditions (brown) or more drought (orange) ascaptured by different drought indices The dashed line indicates the median

estimate the top solid black line denotes the precision-weighted mean withits 95 CI shown in gray The panels at right show the precision-weightedmean effect (circles) and the distribution of study results for all 21 resultslooking at intergroup conflict or for the subset of 12 results focusing ontemperature effects (which includes the ENSO and drought studies)Distributions of effect sizes are either precision-weighted (solid lines) orderived from a Bayesian hierarchical model (dashed lines) See thesupplementary materials for details on the individual studies and on thecalculation of mean effects and their distribution

RESEARCH ARTICLE

because the magnitude of these effects tells ussomething about the potential importance of cli-mate as a factor that may influence conflict solong as we are mindful that evidence is weaker ifa studyrsquos results are less certain In cases in whichthe estimated effect is smaller in magnitude andnot statistically different from zero it is importantto consider whether a study provides strong evi-dence of zero associationmdashthat is whether thestudy rejects the hypothesis that an effect is largein magnitude (third question)mdashor relatively weakevidence because the estimated confidence inter-val (CI) spans large effects as well as zero effect

Evaluating Whether an Effect Is ImportantEvaluating whether an observed causal relation-ship is ldquoimportantrdquo is a subjective judgment thatis not essential to our scientific understanding ofwhether there is a causal relationship Nonethelessbecause importance in this literature has some-times been incorrectly conflated with statisticalprecision or inferred from incorrect interpreta-tions of Eq 1 and its variants we explain our ap-proach to evaluating importance

Our preferred measure of importance is to aska straightforward question Do changes in climatecause changes in conflict risk that an expertpolicy-maker or citizenwould consider large Toaid comparisons we operationalize this questionby considering an effect important if authors of aparticular study state that the size of the effect issubstantive or if the effect is greater than a 10change in conflict risk for each one SD (1s)change in climate variables This second criterionuses an admittedly arbitrary threshold and otherthreshold selections would be justifiable How-ever we contend that this threshold is relativelyconservative as most policy-makers or citizenswould be concerned by effects well below 10per 1s For instance because random variation ina normally distributed climate variable lies in a4s range for 95 of its realizations even a 3per 1s effect size would generate variation in con-flict of 12 of its mean which is probably im-portant to those individuals experiencing these shifts

In some prior studies authors have arguedthat a particular estimated effect is unimportantbased onwhether a climatic variable substantiallychanges goodness-of-fit measures (eg R2) for aparticular statistical model sometimes in com-parison to other predictor variables (14 22ndash24)We do not use this criterion here for two reasonsFirst goodness-of-fit measures are sensitive tothe quantity of noise in a conflict variable Morenoise reduces goodness of fit thus under thismetric irrelevant measurement errors that intro-duce noise into conflict data will reduce the ap-parent importance of climate as a cause of conflicteven if the effect of climate on conflict is quan-titatively large Second comparing the goodnessof fit acrossmultiple predictor variables oftenmakeslittle sense in many contexts because (i) longi-tudinal models typically compare variables thatpredict both where a conflict will occur and whena conflict will occur and (ii) thesemodels typically

compare the causal effect of climatic variableswith the noncausal effects of confounding var-iables such as endogenous covariates These areldquoapples-to-orangesrdquo comparisons and the faultylogic of both types of comparison is made clearwith examples

For an example of (i) consider an analystcomparing violent crime over time in New YorkCity andNorth Dakota who finds that the numberof police on the street each day is important forpredicting how much crime occurs on that daybut that a population variable describes more ofthe variation in crime because crime and pop-ulation in North Dakota are both low Clearly thiscomparison is not informative because the rea-son that there is little crime in North Dakota hasnothing to do with the reason why crime is lowerin New York City on days when there are manypolice on the street The argument that variationsin climate are not important to predicting whenconflict occurs because other variables are goodpredictors of where conflict occurs is analogousto the strange statement that the number of policein New York City is not important for predictingcrime rates because North Dakota has lowercrime that is attributable to its lower population

For an example of (ii) suppose that both higherrainfall and higher household income lower thelikelihood of civil conflict but household incomeis not observed and instead a variable describingthe average observable number of cars each house-hold owns is included in the regression Becausewealthier households are better able to affordcars the analyst finds that populations with morecars have a lower risk of conflict This relation-ship clearly does not have a causal interpretationand comparing the effect of car ownership onconflict with the effect of rainfall on conflict doesnot help us better understand the importance ofthe rainfall variable Published studies that makesimilar comparisons do so with variables that theauthors suggest are more relevant than cars butthe uninformative nature of comparisons be-tween causal effects and noncausal correlationsis the same

Functional Form and Evidence of NonlinearitySome studies assume a linear relationship be-tween climatic factors and conflict risk whereasothers assume a nonlinear relationship Taken asa whole the evidence suggests that over a suf-ficiently large range of temperatures and rainfalllevels both temperature and precipitation appearto have a nonlinear relationship with conflict atleast in some contexts However this curvature isnot apparent in every study probably because therange of temperatures or rainfall levels containedwithin a sample may be relatively limited Thusmost studies report only linear relationships thatshould be interpreted as local linearizations of amore complex and possibly curved responsefunction

As we will show all modern analyses thataddress temperature impacts find that higher tem-peratures lead to more conflict However a few

historical studies that examine temperate loca-tions during cold epochs do find that abrupt cool-ing from an already cold baseline temperaturemay lead to conflict Taken together this collectionof locally linear relationships indicates a globalrelationship with temperature that is nonlinear

In studies of rainfall impacts the distinctionbetween linearity and curvature is made fuzzy bythe multiple ways that rainfall changes have beenparameterized in existing studies Not all studiesuse the same independent variable and because asimple transformation of an independent variablecan change the response function from curved tolinear and visa versa it is difficult to determinewhether results agree In an attempt to make find-ings comparable when replicating the studiesthat originally examine a nonlinear relationshipbetween rainfall and conflict we follow the ap-proach of Hidalgo et al (25) and use the absolutevalue of rainfall deviations from the mean as theindependent variable In studies that originallyexamined linear relationships we leave the inde-pendent variable unaltered Because these twoapproaches in the literature (and our reanalysis)differ wemake the distinction clear in our figuresthrough the use of two different colors

Results from the Quantitative LiteratureWe divide this section topically examining inturn the evidence on how climatic changes shapepersonal violence group-level violence and thebreakdown of social order and political institu-tions Results from 12 example studies of recentdata (post-1950) are displayed in Fig 2 Thesefindings were chosen to represent a broad crosssection of outcomes geographies and time pe-riods and we used the common statistical frame-work described above to replicate these results(see supplementary materials) Findings fromseveral studies of historical data are collected inFig 3 where the different time scales of climaticevents can be easily compared Table 1 lists anddescribes all primary studies For a detailed de-scription and evaluation of each individual studysee (26)

Personal Violence and CrimeStudies in psychology and economics have re-peatedly found that individuals are more likely toexhibit aggressive or violent behavior towardothers if ambient temperatures at the time of ob-servation are higher (Fig 2 A to C) a result thathas been obtained in both experimental (27 28)and natural-experimental (29ndash39) settings Docu-mented aggressive behaviors that respond to tem-perature range from somewhat less consequential[eg horn-honking while driving (27) and inter-player violence during sporting events (36)] tomuch more serious [eg the use of force duringpolice training (28) domestic violencewithinhouse-holds (29 37) and violent crimes such as assaultor rape (30ndash35 38)] Although the physiologicalmechanism linking temperature to aggressionremains unknown the causal association appearsrobust across a variety of contexts Importantly

RESEARCH ARTICLE

because aggression at high temperature increasesthe likelihood that intergroup conflicts escalate insome contexts (36) and the likelihood that policeofficers use force (28) it is possible that thismechanism could affect the prevalence of group-level conflicts on a larger scale

In low-income settings extreme rainfall eventsthat adversely affect agricultural income are alsoassociated with higher rates of personal violence(40ndash42) and property crime (43) High temper-atures are also associated with increased propertycrime (34 35 38) but violent crimes appear torise with temperature more quickly than propertycrimes (38)

Group-Level Violence and Political InstabilitySome forms of intergroup violence such asHindu-Muslim riots (Fig 2D) tend to be more likelyafter extreme rainfall conditions (44ndash47) This rela-tionship between intergroup violence and rainfallis primarily documented in low-income settingssuggesting that reduced agricultural productionmaybe an important mediating mechanism althoughalternative explanations cannot be excluded

Low water availability (23 46 48ndash57) verylow temperatures (58ndash63) and very high temper-atures (14 21 23 51 64ndash66) have been as-sociated with organized political conflicts in avariety of low-income contexts (Fig 2 E F H IK and L) The structure of this relationship againseems to implicate a pathway through climate-induced changes in income either agricultural(48 67ndash69) or nonagricultural (20 21) althoughthis hypothesis remains speculative Large devia-tions from normal precipitation have also beenshown to lead to the forceful reallocation ofwealth (25) (Fig 2G) or the nonviolent replace-ment of incumbent leaders (70 71) (Fig 2J)

Some authors recently suggested that contra-dictory evidence is widespread among quantita-tive studies of climate and human conflict (72ndash74)but the level of disagreement appears overstatedTwo studies (22 24) estimate that temperatureand rainfall events have a limited impact on civilwar in Africa but the CIs around these estimatesare sufficiently wide that they do not reject arelatively large effect of climate on conflict that isconsistent with 35 other studies of modern dataand 28 other studies of intergroup conflict Withinthe broader literature of primary statistical studiesthese results represent 4 of all reported findings(Table 1) Isolated studies also suggest that wind-storms and floods have limited observable effecton civil conflicts (75) and that anomalously highrainfall is associated with higher incidence of ter-rorist attacks (76)

Institutional BreakdownUnder sufficiently high levels of climatologicalstress preexisting social institutions may strainbeyond recovery and lead to major changes ingoverning institutions (77ndash79) (Fig 3C) a pro-cess that often involves the forcible removal ofrulers High levels of climatological stress havealso led to major changes in settlement patterns

and social organization (80 81) (Fig 3D) Final-ly in extreme cases entire communities civili-zations and empires collapse entirely after largechanges in climatic conditions (62 79 80 82ndash89)(Fig 3 A to C E and F) These documentedcatastrophic failures all precede the 20th centuryyet the level of economic development in thesecommunities at the time of their collapse wassimilar to the level of development in many poorcountries of the modern world [see (26) for acomparison] an indicator that these historicalcases may continue to have modern relevance

Synthesis of FindingsOnce attention is restricted to those studies ableto make rigorous causal claims about the relation-ship between climate and conflict some generalpatterns become clear Here we identify for thefirst time commonalities across results that spandiverse social systems climatological stimuli andresearch disciplines

Generality Samples Spatial Scales and Ratesof Climate ChangeSocial conflicts at all scales and levels of organi-zation appear susceptible to climatic influenceand multiple dimensions of the climate systemare capable of influencing these various outcomesStudies documenting this relationship can be foundin data samples covering 10000 BCE to thepresent and this relationship has been identifiedmultiple times in each major region as well as inmultiple samples with global coverage (Fig 1A)

Climatic influence on human conflict appearsin both high- and low-income societies althoughsome types of conflict such as civil war are rarein high-income populations and do not exhibit astrong dependence on climate in those regions(51) Nonetheless many other forms of conflictin high-income countries such as violent crime(35 38) police violence (28) or leadership changes(71) do respond to climatic changes These formsof conflict are individually less extreme but theirtotal social cost may be large because they arewidespread For example during 1979ndash2009 therewere more than 2 million violent crimes (assaultmurder and rape) per year on average in theUnited States alone (38) so small percentagechanges can lead to substantial increases in theabsolute number of these types of events

Climatic perturbations at spatial scales rang-ing from a building (27 28 36) to the globe (51)have been found to influence human conflict orsocial stability (Fig 1B) The finding that climateinfluences conflict across multiple scales sug-gests that coping or adaptation mechanisms areoften limited Interestingly as shown in Fig 1Bthere is a positive association between the tem-poral and spatial scales of observational units instudies documenting a climate-conflict link Thismight indicate that larger social systems are lessvulnerable to high-frequency climate events or itmay be that higher-frequency climate events aremore difficult to detect in studies examining out-comes over wide spatial scales

Finally it is sometimes argued that societiesare particularly resilient to climate perturbationsof a specific temporal scale Perhaps these so-cieties are capable of buffering themselves againstshort-lived climate events or alternatively theyare able to adapt to conditions that are persistentWith respect to human conflict the availableevidence does not support either of these claimsClimatic anomalies of all temporal durationsfrom the anomalous hour (28) to the anomalousmillennium (81) have been implicated in someform of human conflict (Fig 1B)

The association between climatic events andhuman conflict is general in the sense that it hasbeen observed almost everywhere across typesof conflict human history regions of the worldincome groups the various durations of climaticchanges and all spatial scales However it is nottrue that all types of climatic events influence allforms of human conflict or that climatic condi-tions are the sole determinant of human conflictThe influence of climate is detectable across con-texts but we strongly emphasize that it is onlyone of many factors that contribute to conflict[see (90) for a review of these other factors]

The Direction and Magnitude of ClimaticInfluence on Human ConflictWe must consider the magnitude of the climatersquosinfluence to evaluate whether climatic events playan important role in the occurrence of conflictand whether anthropogenic climate change hasthe potential to substantially alter future conflictoutcomes Quantifying the magnitude of climaticimpact in archaeological and paleoclimatologicalstudies is difficult because outcomes of interestare often one-off cataclysmic events (eg so-cietal collapse) and we typically do not observehow the universe of societies would have re-sponded to similar-sized shocks Modern datasamples however generally contain a large num-ber of comparable social units (eg countries)that are repeatedly exposed to climatic variationand this setting is more amenable to statisticalanalyses that quantify how changes in climateaffect the risk of conflict within an individualsocial unit

To compare quantitative results across studiesof modern data we computed standardized effectsizes for those studies where it was possible to doso evaluating the effect of a 1s change in theexplanatory climate variable and expressing theresult as a percentage change in the outcomevariable Because we restrict our attention tostudies that examine changes in climate varia-bles over time the relevant SD is based only onintertemporal changes at each specific locationinstead of comparing variation in climate acrossdifferent geographic locations

Our results are displayed in Figs 4 and 5(colors match those in Figs 2 and 3) Nearly allstudies suggest that warmer temperatures loweror more extreme rainfall or warmer El NintildeondashSouthern Oscillation (ENSO) conditions lead to a2 to 40 increase in the conflict outcome per

RESEARCH ARTICLE

1s in the observed climate variable The consist-ent direction of temperaturersquos influence is partic-ularly notable because all 27 modern estimates(including ENSO and temperature-based droughtindices 20 estimates are shown in Figs 4 and 5)indicate that warmer conditions generate moreconflict a result that would be extremely unlikelyto occur by chance alone if temperature had noeffect on conflict It is more difficult to interpretwhether the signs of rainfall-related variablesagree because these variables are parameterizedseveral different ways so Figs 4 and 5 presentlikelihoods for different parameterizations sepa-rately However if all modern rainfall estimatesare pooled (including ENSO and rainfall-baseddrought indices 13 estimates are shown in Figs4 and 5) using signs shown in Figs 4 and 5 thenthe signs of the effects in 16 out of 18 esti-mates agree

Under the assumption that there is some un-derlying similarity across studies we computethe average effect of climate variables acrossstudies by weighting each estimate according toits precision (the inverse of the estimated var-iance) a common approach that penalizes uncer-tain estimates (91) We also calculate the CI onthismean by assuming independence across studiesalthough this assumption is not critical to ourcentral findings (in the supplementary materialswe present results wherewe relax this assumptionand show that it is not essential) The precision-weighted average effect on interpersonal conflictis a 23 increase for each 1s change in climaticvariables (SE = 012 P lt 0001 Fig 4 and tableS1) and the analogous estimate for intergroupconflict is 111 (SE = 13 P lt 0001 Fig 5and table S1) These precision-weighted averagesare relatively uninfluenced by outliers becauseoutlier estimates in our sample tend to have lowprecision and thus low weight in the meta-analysis The corresponding medians which arealso insensitive to outliers are comparable 39for personal conflict and 136 for group con-flict If we restrict our attention to only the effectsof temperature the precision-weighted averageeffect is similar for interpersonal conflict (23)however for intergroup conflict the effect risesto 132 per 1s in temperature (SE = 20 P lt0001 Fig 5) Regarding the interpretation ofthese effect sizes we note that whereas the aver-age effect for interpersonal violence is smallerthan the average effect for intergroup conflict inpercentage terms the baseline number of inci-dents of interpersonal violence is dramaticallyhigher meaning a small percentage increase canrepresent a substantial increase in total incidents

We estimate the precision-weighted proba-bility distribution of study-level effect sizes inFigs 4 and 5 and in table S1 These distributionsare centered at the precision-weighted averagesdescribed above and can be interpreted as thedistribution of results from which studiesrsquo find-ings are drawn The distribution for interpersonalconflict is narrow around its mean probably be-causemost interpersonal conflict studies focus on

one country (the United States) and use very largesamples and derive very precise estimates Thedistribution for intergroup conflict is broader andcovers values that are larger inmagnitude with aninterquartile range of 6 to 14 per 1s and the5th to 95th percentiles spanning ndash5 to 32 per1s (table S1) We estimate that for the intergroupand interpersonal conflict studies respectively10 and 0 of the probability mass of the distri-butions of effect sizes lies below zero

Figures 4 and 5make it clear that even thoughthere is substantial agreement across results someheterogeneity across estimates remains It is pos-sible that some of this variation is meaningfulperhaps because different types of climate var-iables have different impacts or because the so-cial economic political or geographic conditionsof a societymediate its response to climatic eventsFor instance poorer populations appear to havelarger responses consistent with prior findingsthat such populations are more vulnerable to cli-matic shifts (51) However it is also possible thatsome of this variation is due to differences in howconflict outcomes are definedmeasurement errorin climate variables or remaining differences inmodel specifications that we could not correct inour reanalysis

To formally characterize the variation in esti-mated responses across studies we use a Bayesianhierarchical model that does not require knowl-edge of the source of between-study variation (92)(see supplementarymaterials)Under this approachestimates of the precision-weighted mean are es-sentially unchanged and we recover estimatesfor the between-study SD (a measure of the un-derlying dispersion of true effect sizes acrossstudies) that are half of the precision-weightedmean for interpersonal conflict and two-thirds ofthe precision-weighted mean for intergroup con-flict (median estimates see supplementary mate-rials fig S3 and tables S2 and S3) By comparisonif variation in effect sizes across studies wasdriven by sampling variation alone then this SDin the underlying distribution of effect sizeswould be zero This finding suggests that trueeffects probably differ across settings and under-standing this heterogeneity should be a primarygoal of future research

Publication BiasPublication bias is a long-standing concern acrossthe sciences with a common form of bias arisingfrom the research communityrsquos perceived prefer-ence for positive rather than null results Al-though it is always possible that publication biasplayed a role in the publication of a specificanalysis there are multiple reasons why publica-tion bias is unlikely to be driving our findingsabout the literature on climate and conflict Firstwe include working papers in our analysis (as iscommon practice in the social sciences) therebyeliminating editorial selection Second the cen-tral results presented here are replicated in mul-tiple disciplines and across diverse samples Thirdthe large number of positive findings present in

the literature since 2009 could provide limitedprofessional incentive for researchers to publishyet another positive finding and benefits mightbe higher to those who publish results with al-ternative findings Fourth many analyses are notexplicitly focused on the direct effect of climateon conflict but instead use climatic variationsinstrumentally (25 35 48 71 77) or account forit as an ancillary covariate in their analysis [eg(37)] while trying to study a different researchquestion indicating that these authors have littleprofessional stake in the sign magnitude or sta-tistical significance of the climatic effects they arepresenting Fifth we reanalyze the raw data frommany studies using a common statistical frame-work possibly undoing adjustments that authorsmight be making to their analysis (consciously orunconsciously) that make their findings appearstronger Partial support for this idea is providedby individual studies that present significant re-sults but whose results are only marginally signif-icant or no longer significant after our reanalysis(see supplementary materials for details) Finallywe look for evidence of publication bias by ex-amining whether the statistical strength of indi-vidual studies reflects their sample size (93) anddo not find systematic evidence of strong bias inabsolute terms or in comparison to other socialscience literature (see fig S4 table S4 and sup-plementary materials)

Implications for Future Climatic ChangesThe above evidence taken at face value makesthe case that future anthropogenic climate changecould worsen conflict outcomes across the globein comparison to a future with no climatic changesgiven the large expected increase in global surfacetemperatures and the likely increase in variabilityof precipitation across many regions over comingdecades (94 95) Recalling our finding that a1s change in a locationrsquos temperature is associatedwith an average 23 increase in the rate of in-terpersonal conflict and a 132 increase in therate of intergroup conflict and assuming thatfuture populations will respond to climatic shiftssimilarly to how current populations respond onecan consider the potential effect of anthropogenicwarmingby rescaling expected temperature changesaccording to each locationrsquos historical variabil-ity Although not all conflict outcomes have beenshown to be responsive to changes in temper-ature many have and the results uniformly indi-cate that increasing temperatures are harmful inregions that are temperate or warm initially InFig 6 we plot expected warming by 2050 com-puted as the ensemble mean for 21 climate mod-els running the A1B emissions scenario in termsof location-specific SDs (96) Almost all inhab-ited locations warm by gt2s with the largest in-creases exceeding 4s in tropical regions thatare already warm and currently experience rel-atively low interannual temperature variabilityThese large climatological changes combinedwith the quantitatively large effect of climate onconflictmdashparticularly intergroup conflictmdashsuggest

RESEARCH ARTICLE

that amplified rates of human conflict could rep-resent a large and critical impact of anthropogenicclimate change

Two reasons are often given as to why climatechange might not have a substantive impact onhuman conflict Future climate change will occurgradually and will thus allow societies to adaptand the modern world today is less susceptible toclimate variation than it has been in the pastHowever if slower-moving climate shocks havesmaller effects or if the world has become lessclimate-sensitive it is unfortunately not obviousin the data Gradual climatic changes appear toadversely affect conflict outcomes and the ma-jority of the studies we review use a sample periodthat extends into the 21st century (recall Fig 1)Furthermore some studies explicitly examinewhether populations inhabiting hotter climatesexhibit less conflict when hot events occur butfind little evidence that these areas are moreadapted (31 38) We also note that many of themodern linkages between high-temperatureanomalies and intergroup conflict have been char-acterized in Africa (14 23 52 64 66) or theglobal tropics and subtropics (21 51) regionswith hot climates where we would expect pop-ulations to be best adapted to high temperaturesNevertheless it is always possible that futurepopulations will adapt in previously unobservedways but it is impossible to know if and to whatextent these adaptations will make conflict moreor less likely

Studies of nonconflict outcomes do indicatethat in some situations historical adaptation toclimate is observable albeit costly (97ndash100)whereas in other cases there is limited evidencethat any adaptation is occurring (19 101) To ourknowledge no study has characterized the scaleor scope for adaptation to climate in terms ofconflict outcomes and we believe this is an im-portant area for future research Given the quan-titatively large effect of climate on conflict futureadaptations will need to be dramatic if they areto offset the potentially large amplification ofconflict

Future ResearchGiven the marked consistency of available quan-titative evidence linking climate and conflict inour view the top research priority in this fieldshould be to narrow the number of competingexplanatory hypotheses Beyond efforts to miti-gate future warming limiting climatersquos future in-fluence on conflict requires that we understandthe causal pathways that generate the observedassociation This task is made difficult by thelikely situation that multiple mechanisms con-tribute to the observed relationships and thatdifferent mechanisms dominate in different con-texts The rich qualitative literature (3ndash7) sug-gests that a multiplicity of mechanisms may beat work

To date no study has been able to conclu-sively pin down the full set of causal mechanismsalthough some studies find suggestive evidence

that a particular pathway contributes to the ob-served association in a particular context In mostcases this is accomplished by ldquofingerprintingrdquothe effect of climate on an intermediary variablesuch as income and showing that the same sta-tistical fingerprint is visible in the climatersquos effecton conflict This approach typically called ldquoinstru-mental variablesrdquo (12) in the social sciences iden-tifies a mechanism linking climate and conflictunder the assumption that climatersquos only influenceon conflict is through the particular intermediatevariable in question Because this assumption isoften difficult or impossible to test evidence fromthis approach is more suggestive than conclusivein uncovering mechanisms (51)

An alternate and promising research designthat can help rule out certain hypotheses is tostudy situations in which plausibly exogenousevents block a proposed pathway in a treatedsubpopulation and then to compare whether theclimate-conflict association persists or disappearsin both the treatment and control subpopulationsIn an example of this approach Sarsons exam-ines whether rainfall shortages in India lead toriots because they depress local agricultural in-come (45) By showing that rainfall shortagesand riots continue to occur together in districtswith dams that supply irrigation investments thatpartially decouple local agricultural income fromtemporary rain shortfalls Sarsons argues that therainfall effect on riots is unlikely to be operat-ing solely through changes in local agriculturalincome

Plausible MechanismsThe following hypotheses have in our judgmentreceived the strongest empirical support in exist-ing analyses although the evidence is still ofteninconclusive A common hypothesis focuses onlocal economic conditions and labor markets andargues that when climatic events cause economicproductivity to decline (19ndash21 68 69 102ndash104)the value of engaging in conflict is likely to riserelative to the value of participating in normaleconomic activities (48 52 105ndash110) A compet-ing hypothesis on state capacity argues that thesedeclines in economic productivity reduce thestrength of governmental institutions (eg if taxrevenues fall) curtailing their ability to suppresscrime and rebellion or encouraging competitorsto initiate conflict during these periods of relativestate weakness (61 70 71 77ndash79 84 85)

A second set of hypotheses focuses on whathave more generally been termed ldquogrievancesrdquoHypotheses about inequality contend that whenclimatic events increase actual (or perceived) so-cial and economic inequalities in a society (111 112)this could increase conflict by motivating at-tempts to redistribute assets (25 34 35 43)Evidence linking changes in food prices to con-flict (61 113ndash115) can be interpreted similarlymdashfor example food riots due to a governmentrsquosperceived inability to keep food affordablemdashparticularly when some members of society caninfluence food markets (111 116)

Climate-induced migration and urbanizationmight also be implicated in conflict If climaticevents cause large population displacements orrapid urbanization (97 117 118) this might leadto conflicts over geographically stationary resourcesthat are unrelated to the climate (119) but becomerelatively scarce where populations concentrateChanges in climate might also affect the logisticsof human conflict (76 120) for example by al-tering the physical environment (eg road quali-ty) in which disputes or violence might occur(52 120 121) Finally climate anomalies mightresult in conflict because they can make cogni-tion and attribution more difficult or error-proneor they may affect aggression through somephysiological mechanism For instance climaticevents may alter individualsrsquo ability to reason andcorrectly interpret events (27 28 30 31 34ndash36)possibly leading to conflicts triggered by mis-understandings Alternatively if climatic changesand their economic consequences are inaccu-rately attributed to the actions of an individualor group (63 122ndash125)mdashfor example an ineptpolitical leader (71)mdashthis may lead to violentactions that try to return economic conditions tonormal by removing the ldquooffendingrdquo population

Selecting Climate Variables andConflict OutcomesClimate variables that have been analyzed pre-viously such as seasonal temperatures precipi-tation water availability indices and climateindices may be correlated with one another andautocorrelated across both time and space Forinstance temperature and precipitation time se-ries tend to be negatively correlated in much ofthe tropics and drought indices tend to be spa-tially correlated (51 126) Unfortunately only afew of the existing studies account for the cor-relations between different variables so it may bethat some studies mistakenly measure the influ-ence of an omitted climate variable by proxy [see(126) for a complete discussion of this issue]Except for the experiments linking temperatureto aggression (27 28) only a few studies demon-strate that a specific climate variable is more im-portant for predicting conflict than other climatevariables or that climatic changes during a spe-cific season are more important than during otherseasons Furthermore no study isolates a partic-ular type of climatic change as the most influ-ential and no study has identifiedwhether temporalor spatial autocorrelations in climatic variablesare mechanistically important Identifying the cli-matic variables timing of events and forms ofautocorrelation that influence conflict will help usbetter understand the mechanisms linking cli-matic changes to conflict

A similar situation exists with the choice ofconflict outcomes Most analyses simply docu-ment changes in the rate at which conflicts arereported in aggregate but this approach providesonly limited insight into how the evolution ofconflict is affected by climatic variables A pathfor future investigation is to link climate data

RESEARCH ARTICLE

with richer conflict data that describes differentstages of the conflict ldquolife cyclerdquo For example fu-ture studies could examine how often nonviolentgroup disputes become violent Two studies citedin this paper (28 36) demonstrate the usefulnessof selecting conflict variables other than total con-flict rates By examining the probability that aninitial confrontation escalates rather than just count-ing the total number of conflicts these studiesdemonstrate that high temperatures lead to moreviolence by increasing the likelihood that a smallconflict escalates into a larger conflict

ConclusionFindings from a growing corpus of rigorous quan-titative research acrossmultiple disciplines suggestthat past climatic events have exerted consider-able influence on human conflict This influenceappears to extend across the world throughouthistory and at all scales of social organizationWe do not conclude that climate is the sole oreven primary driving force in conflict but we dofind that when large climate variations occurthey can have substantial effects on the incidenceof conflict across a variety of contexts The me-dian effect of a 1s change in climate variablesgenerates a 14 change in the risk of intergroupconflict and a 4 change in interpersonal vio-lence across the studies that we review where itis possible to calculate standardized effects Iffuture populations respond similarly to past pop-ulations then anthropogenic climate changehas the potential to substantially increase conflictaround the world relative to a world without cli-mate change

Although there is marked convergence ofquantitative findings across disciplines many openquestions remain Existing research has success-fully established a causal relationship betweenclimate and conflict but is unable to fully explainthe mechanisms This fact motivates our pro-posed research agenda and urges caution whenapplying statistical estimates to future warmingscenarios Importantly however it does not im-ply that we lack evidence of a causal associationThe studies in this analysis were selected for theirability to provide reliable causal inferences andthey consistently point toward the existence of atleast one causal pathway To place the state of thisresearch in perspective it is worth recalling thatstatistical analyses identified the smoking of tobac-co as a proximate cause of lung cancer by the 1930s(127) although the research community was un-able to provide a detailed account of the mech-anisms explaining the linkage until many decadeslater So although future research will be critical inpinpointing why climate affects human conflictdisregarding the potential effect of anthropogenicclimate change on human conflict in the interimis in our view a dangerously misguided interpre-tation of the available evidence

Numerous competing theories have been pro-posed to explain the linkages between the climateand human conflict but none have been con-vincingly rejected and all appear to be consistent

with at least some existing results It seems likelythat climatic changes influence conflict throughmultiple pathways that may differ between con-texts and innovative research to identify thesemechanisms is a top research priority Achievingthis research objective holds great promise as thepolicies and institutions necessary for conflictresolution can be built only if we understand whyconflicts arise The success of such institutionswill be increasingly important in the comingdecades as changes in climatic conditions am-plify the risk of human conflicts

References and Notes1 C Mathers T Boerma D Ma Fat The Global Burden

of Disease 2004 Update (World Health OrganizationGeneva 2008)

2 R Lozano et al Global and regional mortality from235 causes of death for 20 age groups in 1990 and2010 A systematic analysis for the Global Burden ofDisease Study 2010 Lancet 380 2095ndash2128 (2012)doi 101016S0140-6736(12)61728-0 pmid 23245604

3 M A Levy Is the environment a national security issueInt Secur 20 35ndash62 (1995) doi 1023072539228

4 T F Homer-Dixon Environment Scarcity and Violence(Princeton Univ Press Princeton NJ 1999)

5 J Scheffran M Brzoska H G Brauch P M LinkJ Schilling Eds Climate Change Human Security andViolent Conflict Challenges for Societal Stability vol 8(Springer Berlin 2012)

6 T Deligiannis The evolution of environment-conflictresearch Toward a livelihood framework Glob EnvironPolit 12 78ndash100 (2012) doi 101162GLEP_a_00098

7 K W Butzer Collapse environment and societyProc Natl Acad Sci USA 109 3632ndash3639 (2012)doi 101073pnas1114845109 pmid 22371579

8 E Huntington Climatic change and agriculturalexhaustion as elements in the fall of rome Q J Econ31 173ndash208 (1917) doi 1023071883908

9 P W Holland Statistics and causal inference J AmStat Assoc 81 945ndash960 (1986) doi 10108001621459198610478354

10 N P Gleditsch Armed conflict and the environment Acritique of the literature J Peace Res 35 381ndash400(1998) doi 1011770022343398035003007

11 J D Angrist J-S Pischke The credibility revolution inempirical economics How better research design istaking the con out of econometrics J Econ Perspect 243ndash30 (2010) doi 101257jep2423

12 J D Angrist J-S Pischke Mostly Harmless EconometricsAn Empiricistrsquos Companion (Princeton Univ PressPrinceton NJ 2008)

13 J Wooldridge Econometric Analysis of Cross Section andPanel Data (The MIT Press Cambridge MA 2002)

14 O M Theisen Climate clashes Weather variability landpressure and organized violence in Kenya 1989-2004J Peace Res 49 81ndash96 (2012) doi 1011770022343311425842

15 D Freedman Statistical models and shoe leather SociolMethodol 21 291ndash313 (1991) doi 102307270939

16 W H Greene Econometric Analysis (Prentice HallUpper Saddle River NJ ed 5 2003)

17 A Gelman J Hill Data Analysis Using Regression andMultilevelHierarchical Models (Cambridge Univ PressCambridge 2006)

18 J D Angrist A B Krueger in Empirical Strategies inLabor Economics vol 3 (Elsevier Science Amsterdam1999) chap 23 pp 1277ndash1366

19 W Schlenker M J Roberts Nonlinear temperatureeffects indicate severe damages to US crop yields underclimate change Proc Natl Acad Sci USA 10615594ndash15598 (2009) doi 101073pnas0906865106pmid 19717432

20 S M Hsiang Temperatures and cyclones stronglyassociated with economic production in the Caribbeanand Central America Proc Natl Acad Sci USA 107

15367ndash15372 (2010) doi 101073pnas1009510107pmid 20713696

21 M Dell B F Jones B A Olken Climate change andeconomic growth Evidence from the last half centuryAm Econ J Macroecon 4 66ndash95 (2012) doi 101257mac4366

22 H Buhaug Reply to Burke et al Bias and climate warresearch Proc Natl Acad Sci USA 107 E186ndashE187(2010) doi 101073pnas1015796108

23 J OrsquoLoughlin et al Climate variability and conflictrisk in East Africa 1990-2009 Proc Natl AcadSci USA 109 18344ndash18349 (2012) doi 101073pnas1205130109 pmid 23090992

24 O Theisen H Holtermann H Buhaug Climate warsAssessing the claim that drought breeds conflict IntSecur 36 79ndash106 (2011) doi 101162ISEC_a_00065

25 F Hidalgo S Naidu S Nichter N Richardson Economicdeterminants of land invasions Rev Econ Stat 92505ndash523 (2010) doi 101162REST_a_00007

26 S M Hsiang M Burke Climate conflict and social stabilityWhat does the evidence say Clim Change 101007s10584-013-0868-3 (2013) available at httppapersssrncomsol3paperscfmabstract_id=2302245

27 D T Kenrick S W Macfarlane Ambient temperatureand horn honking A field study of the heataggressionrelationship Environ Behav 18 179ndash191 (1986)doi 1011770013916586182002

28 A Vrij J Van Der Steen L Koppelaar Aggressionof police officers as a function of temperatureAn experiment with the fire arms training systemJ Community Appl Soc 4 365ndash370 (1994)doi 101002casp2450040505

29 A Auliciems L DiBartolo Domestic violence in asubtropical environment Police calls and weather inBrisbane Int J Biometeorol 39 34ndash39 (1995) doi101007BF01320891

30 E Cohn J Rotton Assault as a function of time andtemperature A moderator-variable time-series analysisJ Pers Soc Psychol 72 1322ndash1334 (1997)doi 1010370022-35147261322

31 J Rotton E G Cohn Violence is a curvilinear function oftemperature in Dallas A replication J Pers Soc Psychol78 1074ndash1081 (2000) doi 1010370022-35147861074 pmid 10870909

32 B J Bushman M C Wang C A Anderson Is the curverelating temperature to aggression linear or curvilinearA response to Bell (2005) and to Cohn and Rotton(2005) J Pers Soc Psychol 89 74ndash77 (2005)doi 1010370022-351489174 pmid 16060746

33 C A Anderson B J Bushman R W Groom Hot yearsand serious and deadly assault Empirical tests of theheat hypothesis J Pers Soc Psychol 73 1213ndash1223(1997) doi 1010370022-35147361213pmid 9418277

34 C Anderson K Anderson N Dorr K DeNeve M FlanaganTemperature and aggression Adv Exp Soc Psychol 3263ndash133 (2000) doi 101016S0065-2601(00)80004-0

35 B Jacob L Lefgren E Moretti The dynamics of criminalbehavior Evidence from weather shocks J Hum Resour42 489ndash527 (2007)

36 R P Larrick T A Timmerman A M Carton J AbrevayaTemper temperature and temptation Heat-relatedretaliation in baseball Psychol Sci 22 423ndash428 (2011)doi 1011770956797611399292 pmid 21350182

37 D Card G B Dahl Family violence and football Theeffect of unexpected emotional cues on violent behaviorQ J Econ 126 103ndash143 (2011) doi 101093qjeqjr001 pmid 21853617

38 M Ranson Crime weather and climate change J EnvironEcon Manage (in press) httppapersssrncomsol3paperscfmabstract_id=2111377

39 D Mares Climate change and levels of violence insocially disadvantaged neighborhood groups J UrbanHealth 90 768ndash783 (2013) doi 101007s11524-013-9791-1 pmid 23435543

40 E Miguel Poverty and witch killing Rev Econ Stud 721153ndash1172 (2005) doi 1011110034-652700365

41 S Sekhri A Storeygard Dowry deaths Consumptionsmoothing in response to climate variability in Indiaworking paper (2012) httpgoogl2pfZr

RESEARCH ARTICLE

42 D Blakeslee R Fishman Rainfall shocks and propertycrimes in agrarian societies Evidence from India SSRNworking paper (2013) httppapersssrncomsol3paperscfmabstract_id=2208292

43 H Mehlum E Miguel R Torvik Poverty and crime in19th century Germany J Urban Econ 59 370ndash388(2006) doi 101016jjue200509007

44 A T Bohlken E J Sergenti Economic growth and ethnicviolence An empirical investigation of Hindu-Muslimriots in India J Peace Res 47 589ndash600 (2010)doi 1011770022343310373032

45 H Sarsons Rainfall and conflict Harvard working paper(2011) wwweconyaleeduconferenceneudc11paperspaper_199pdf

46 C S Hendrix I Salehyan Climate change rainfall andsocial conflict in Africa J Peace Res 49 35ndash50 (2012)doi 1011770022343311426165

47 J K Kung C Ma Can cultural norms reduce conflictsConfucianism and peasant rebellions in Qing Chinaworking paper (2012) httpahec2012orgpapersS6B-2_Kai-singKung_Mapdf

48 E Miguel S Satyanath E Sergenti Economic shocks andcivil conflict An instrumental variables approach J PolitEcon 112 725ndash753 (2004) doi 101086421174

49 M Levy et al Freshwater availability anomalies andoutbreak of internal war Results from a global spatialtime series analysis International Workshop on HumanSecurity and Climate Change Holmen Norway 21 to 23June 2005 wwwciesincolumbiaedupdfwaterconflictpdf

50 Y Bai J Kung Climate shocks and Sino-nomadicconflict Rev Econ Stat 93 970ndash981 (2011)doi 101162REST_a_00106

51 S M Hsiang K C Meng M A Cane Civil conflicts areassociated with the global climate Nature 476 438ndash441(2011) doi 101038nature10311 pmid 21866157

52 M Harari E La Ferrara Conflict climate and cells Adisaggregated analysis working paper (2013) www-2iiessuseNobel2012PapersLaFerrara_Hararipdf

53 M Couttenier R Soubeyran Drought and civil war inSub-Saharan Africa Econ J 101111ecoj12042 (2013)doi 101111ecoj12042

54 M Cervellati U Sunde S Valmori Disease environmentand civil conflicts IZA discussion paper no 5614 (2011)httppapersssrncomabstract=1806415

55 H Fjelde N von Uexkull Climate triggers Rainfallanomalies vulnerability and communal conflict insub-Saharan Africa Polit Geogr 31 444ndash453 (2012)doi 101016jpolgeo201208004

56 R Jia Weather shocks sweet potatoes and peasantrevolts in historical China Econ J (2013) doi 101111ecoj12037

57 H F Lee D D Zhang P Brecke J Fei Positivecorrelation between the North Atlantic oscillation andviolent conflicts in Europe Clim Res 56 1ndash10 (2013)doi 103354cr01129

58 D Zhang et al Climatic change wars and dynastic cyclesin China over the last millennium Clim Change 76459ndash477 (2006) doi 101007s10584-005-9024-z

59 D D Zhang P Brecke H F Lee Y Q He J ZhangGlobal climate change war and population decline inrecent human history Proc Natl Acad Sci USA 10419214ndash19219 (2007) doi 101073pnas0703073104pmid 18048343

60 R Tol S Wagner Climate change and violent conflict inEurope over the last millennium Clim Change 9965ndash79 (2010) doi 101007s10584-009-9659-2

61 D D Zhang et al The causality analysis of climatechange and large-scale human crisis Proc Natl AcadSci USA 108 17296ndash17301 (2011) doi 101073pnas1104268108 pmid 21969578

62 U Buumlntgen et al 2500 years of European climatevariability and human susceptibility Science 331578ndash582 (2011) doi 101126science1197175pmid 21233349

63 R W Anderson N D Johnson M Koyama From thepersecuting to the protective state Jewish expulsionsand weather shocks from 1100 to 1800 SSRN workingpaper (2013) httpssrncomabstract=2212323

64 M B Burke E Miguel S Satyanath J A DykemaD B Lobell Warming increases the risk of civil war in

Africa Proc Natl Acad Sci USA 106 20670ndash20674(2009) doi 101073pnas0907998106 pmid 19934048

65 C Almer S Boes Climate (change) and conflictResolving a puzzle of association and causation workingpaper dp1203 (2012) httpideasrepecorgpubedpvwibdp1203html

66 J-F Maystadt O Ecker A Mabiso Extreme weather andcivil war in Somalia Does drought fuel conflict throughlivestock price shocks IFPRI working paper (2013)wwwifpriorgpublicationextreme-weather-and-civil-war-somalia

67 W Schlenker M J Roberts Nonlinear effects of weatheron corn yields Rev Agric Econ 28 391ndash398 (2006)doi 101111j1467-9353200600304x

68 W Schlenker D Lobell Robust negative impacts ofclimate change on African agriculture Environ Res Lett5 014010 (2010) doi 1010881748-932651014010

69 D Lobell M Burke Climate Change and Food SecurityAdapting Agriculture to a Warmer World (SpringerDordrecht Netherlands 2010)

70 E Chaney Revolt on the Nile Economic shocks religionand political influence Econometrica (in press) httpscholarharvardeduchaneypublicationsrevolt-nile-economic-shocks-religion-and-political-power-0

71 P J Burke Economic growth and political survivalB E J Macroecon 12 1ndash43 (2012)

72 J Scheffran M Brzoska J Kominek P M LinkJ Schilling Climate change and violent conflict Science336 869ndash871 (2012) doi 101126science1221339pmid 22605765

73 N Gleditsch Whither the weather Climate changeand conflict J Peace Res 49 3ndash9 (2012)doi 1011770022343311431288

74 T Bernauer T Boumlhmelt V Koubi Environmentalchanges and violent conflict Environ Res Lett 7015601 (2012) doi 1010881748-932671015601

75 D Bergholt P Lujala Climate-related natural disasterseconomic growth and armed civil conflict J Peace Res49 147ndash162 (2012) doi 1011770022343311426167

76 I Salehyan C Hendrix Climate shocks and politicalviolence Annual Convention of the InternationalStudies Association San Diego CA 1 April 2012httpgoogl7RGEZd

77 P J Burke A Leigh Do output contractions triggerdemocratic change Am Econ J Macroecon 2 124ndash157(2010) doi 101257mac24124

78 M Bruumlckner A Ciccone Rain and the democratic windowof opportunity Econometrica 79 923ndash947 (2011)doi 103982ECTA8183

79 G Yancheva et al Influence of the intertropical convergencezone on the East Asian monsoonNature 445 74ndash77 (2007)doi 101038nature05431 pmid 17203059

80 C R Ortloff A L Kolata Climate and collapseAgro-ecological perspectives on the decline of theTiwanaku State J Archaeol Sci 20 195ndash221 (1993)doi 101006jasc19931014 pmid 11303088

81 R Kuper S Kroumlpelin Climate-controlled Holoceneoccupation in the Sahara Motor of Africarsquos evolutionScience 313 803ndash807 (2006) doi 101126science1130989 pmid 16857900

82 D W Stahle M K Cleaveland D B Blanton M D TherrellD A Gay The lost colony and Jamestown droughts Science280 564ndash567 (1998) doi 101126science2805363564pmid 9554842

83 H Cullen et al Climate change and the collapse of theAkkadian empire Evidence from the deep sea Geology28 379ndash382 (2000) doi 1011300091-7613(2000)28lt379CCATCOgt20CO2

84 G H Haug et al Climate and the collapse of Mayacivilization Science 299 1731ndash1735 (2003)doi 101126science1080444 pmid 12637744

85 B M Buckley et al Climate as a contributing factor inthe demise of Angkor Cambodia Proc Natl Acad SciUSA 107 6748ndash6752 (2010) doi 101073pnas0910827107 pmid 20351244

86 W P Patterson K A Dietrich C Holmden J T AndrewsTwo millennia of North Atlantic seasonality andimplications for Norse colonies Proc Natl AcadSci USA 107 5306ndash5310 (2010) doi 101073pnas0902522107 pmid 20212157

87 D J Kennett et al Development and disintegration ofMaya political systems in response to climate changeScience 338 788ndash791 (2012) doi 101126science1226299 pmid 23139330

88 R L Kelly T A Surovell B N Shuman G M SmithA continuous climatic impact on Holocene humanpopulation in the Rocky Mountains Proc Natl Acad Sci USA 110 443ndash447 (2013) doi 101073pnas1201341110pmid 23267083

89 R M DrsquoAnjou R S Bradley N L Balascio D B FinkelsteinClimate impacts on human settlement and agriculturalactivities in northern Norway revealed through sedimentbiogeochemistry Proc Natl Acad Sci USA 10920332ndash20337 (2012) doi 101073pnas1212730109pmid 23185025

90 C Blattman E Miguel Civil war J Econ Lit 48 3ndash57(2010) doi 101257jel4813

91 L V Hedges I Olkin Statistical Method forMeta-Analysis (Academic Press London 1985)

92 A Gelman J B Carlin H S Stern D B RubinBayesian Data Analysis (Chapman amp HallCRC PressBoca Raton FL 2004)

93 D Card A B Krueger Time-series minimum-wage studiesA meta-analysis Am Econ Rev 85 238ndash243 (1995)

94 G Meehl et al Global climate projections in ClimateChange 2007 The Physical Science Basis Contributionof Working Group I to the Fourth Assessment Report ofthe Intergovernmental Panel on Climate Change(Cambridge Univ Press Cambridge 2007)pp 747ndash845

95 Intergovernmental Panel on Climate Change Managingthe Risks of Extreme Events and Disasters to AdvanceClimate Change Adaptation (Cambridge Univ PressCambridge 2012)

96 G A Meehl et al The WCRP CMIP3 Multimodel DatasetA new era in climate change research Bull AmMeteorol Soc 88 1383ndash1394 (2007) doi 101175BAMS-88-9-1383

97 R Hornbeck The enduring impact of the American DustBowl Short and long-run adjustments to environmentalcatastrophe Am Econ Rev 102 1477ndash1507 (2012)doi 101257aer10241477

98 G D Libecap R H Steckel Eds The Economicsof Climate Change Adaptations Past and Present(University of Chicago Press Chicago 2011)

99 O Deschecircnes M Greenstone Climate change mortalityand adaptation Evidence from annual fluctuations inweather in the US Am Econ J Appl Econ 3 152ndash185(2011) doi 101257app34152

100 S M Hsiang D Narita Adaptation to cyclone riskEvidence from the global cross-section Climate ChangeEcon 3 1250011ndash1250039 (2012)

101 M Burke K Emerick Adaptation to climate changeEvidence from US agriculture working paper (2013)httpssrncomabstract=2144928

102 S Barrios L Bertinelli E Strobl Trends in rainfall andeconomic growth in Africa A neglected cause of theAfrican growth tragedy Rev Econ Stat 92 350ndash366(2010) doi 101162rest201011212

103 J Graff Zivin M Neidell Temperature and the allocationof time Implications for climate change J Labor Econ(in press) wwwnberorgpapersw15717

104 B Jones B Olken Climate shocks and exports Am EconRev Pap Proc 100 454ndash459 (2010) doi 101257aer1002454

105 O Dube J Vargas Commodity price shocks and civilconflict Evidence from Colombia Rev Econ Stud(2013) httprestudoxfordjournalsorgcontentearly20130215restudrdt009abstract

106 J Angrist A Kugler Rural windfall or a new resourcecurse Coca income and civil conflict in Colombia RevEcon Stat 90 191ndash215 (2008) doi 101162rest902191

107 S Chassang G Padro-i-Miquel Economic shocks andcivil war Q J Polit Sci 4 211ndash228 (2009)doi 10156110000008072

108 E Dal Boacute P Dal Boacute Workers warriors and criminalsSocial conflict in general equilibrium J EurEcon Assoc 9 646ndash677 (2011) doi 101111j1542-4774201101025x

RESEARCH ARTICLE

109 E Berman J Shapiro J Felter Can hearts andminds be bought The economics of counterinsurgencyin Iraq J Polit Econ 119 766ndash819 (2011)doi 101086661983

110 Y Lei G Michaels Do giant oilfield discoveries fuelinternal armed conflicts CEP discussion paper (2011)httpceplseacukpubsdownloaddp1089pdf

111 R Grove The Great El Nintildeo of 1789-93 and its globalconsequences Reconstructing an an extreme climateevent in world environmental history Mediev Hist J 1075ndash98 (2007) doi 101177097194580701000203

112 J K Anttila-Hughes S M Hsiang Destructiondisinvestment and death Economic and human lossesfollowing environmental disaster SSRN working paper(2012) httppapersssrncomabstract_id=2220501

113 M Lagi K Bertrand Y Bar-Yam The food crises andpolitical instability in North Africa and the Middle EastarXiv working paper (2011) httparxivorgabs11082455

114 R Arezki M Bruumlckner Food prices and politicalinstability IMF working paper (2011) wwwimforgexternalpubsftwp2011wp1162pdf

115 C B Barrett Ed Food or Consequences Food Securityand Its Implications for Global Sociopolitical Stability(Oxford Univ Press New York 2013)

116 B Carter R Bates Public policy price shocks and civilwar in developing countries Harvard working paper(2012) wwwwcfiaharvardedusitesdefaultfilesbcarter_publicpolicycivilwarpdf

117 S Barrios L Bertinelli E Strobl Climatic change andrural-urban migration The case of Sub-Saharan AfricaJ Urban Econ 60 357ndash371 (2006) doi 101016jjue200604005

118 S Feng M Oppenheimer W Schlenker Climatechange crop yields and internal migration in theUnited States NBER working paper 17734 (2012)wwwnberorgpapersw17734

119 P S Jensen K S Gleditsch Rain growth and civil warThe importance of location Defence Peace Econ 20359ndash372 (2009) doi 10108010242690902868277

120 J Fearon D Laitin Ethnicity insurgency and civil warAm Polit Sci Rev 97 75ndash90 (2003) doi 101017S0003055403000534

121 C K Butler S Gates African range wars Climateconflict and property rights J Peace Res 49 23ndash34(2012) doi 1011770022343311426166

122 C Achen L Bartels Blind retrospection Electoralresponses to drought flu and shark attacks Centro deEstudios Avanzados en Ciencias Sociales working paper

(2004) wwwmarchesceacspublicacionesworkingarchivos2004_199pdf

123 D Hibbs Jr Voting and the macroeconomy in TheOxford Handbook of Political Economy B R WeingastD A Wittman Eds (Oxford Univ Press New York2006) pp 565ndash586

124 A J Healy N Malhotra C H Mo Irrelevant events affectvotersrsquo evaluations of government performance ProcNatl Acad Sci USA 107 12804ndash12809 (2010)doi 101073pnas1007420107 pmid 20615955

125 M Manacorda E Miguel A Vigorito Governmenttransfers and political support Am Econ J Appl Econ 31ndash28 (2011) doi 101257app331 pmid 22199993

126 M Auffhammer S Hsiang W Schlenker A Sobel Usingweather data and climate model output in economicanalyses of climate change Rev Environ Econ Policy 7181ndash198 (2013) doi 101093reepret016

127 H Witschi A short history of lung cancer ToxicolSci 64 4ndash6 (2001) doi 101093toxsci6414pmid 11606795

128 S M Hsiang Visually-weighted regression SSRNworking paper (2013) httppapersssrncomsol3paperscfmabstract_id=2265501

129 G S Watson Smooth regression analysis Sankhya 26359ndash372 (1964)

130 E A Nadaraya On estimating regression Theory ProbabAppl 9 141ndash142 (1964) doi 1011371109020

131 D R Legates C J Willmott Mean seasonal and spatialvariability global surface air temperature Theor ApplClimatol 41 11ndash21 (1990) doi 101007BF00866198

132 A E Sutton et al Does warming increase the risk of civilwar in Africa Proc Natl Acad Sci USA 107 E102(2010) doi 101073pnas1005278107 pmid 20538978

133 H Buhaug H Hegre H Strand Sensitivity analysis ofclimate variability and civil war PRIO working paper(2010) httpgooglAr3xox

134 H Buhaug Climate not to blame for African civil warsProc Natl Acad Sci USA 107 16477ndash16482 (2010)doi 101073pnas1005739107 pmid 20823241

135 M Burke J Dykema D Lobell E Miguel S SatyanathClimate and civil war Is the relationship robust NBERworking paper 16440 (2010) wwwnberorgpapersw16440

136 M B Burke E Miguel S Satyanath J A DykemaD B Lobell Climate robustly linked to African civil warProc Natl Acad Sci USA 107 E185 (2010)doi 101073pnas1014879107 pmid 21118990

137 M B Burke E Miguel S Satyanath J A DykemaD B Lobell Reply to Sutton et al Relationship betweentemperature and conflict is robust Proc Natl Acad

Sci USA 107 E103 (2010) doi 101073pnas1005748107

138 A Ciccone Economic shocks and civil conflict Acomment Am Econ J Appl Econ 3 215ndash227 (2011)doi 101257app34215 pmid 22199993

139 E Miguel S Satyanath Re-examining economic shocksand civil conflict Am Econ J Appl Econ 3 228ndash232(2011) doi 101257app34228 pmid 22199993

Acknowledgments We thank C Almer D BlakesleeD Card P Burke H Buhaug P deMenocal N P GleditschM Harari G Haug R Larrick H Lee L Lefgren M LevyS Naidu J OrsquoLoughlin M Ranson O M Theisen andN von Uexkull for providing results and data We also thankthree anonymous reviewers I Bannon D Card M CaneM Kremer D Lobell K Meng S Mullainathan M OppenheimerM Roberts I Salehyan W Schlenker J Shapiro R SinghD Stahle L Thompson D Zhang and seminar participants atColumbia University Harvard University the InternationalStudies Association Annual Meeting Oxford University thePacific Development Conference Princeton UniversityUniversity of California (UC) Berkeley UC San Diego andthe World Bank for comments suggestions and referencesWe thank C Baysan and F Gonzalez for excellent researchassistance SMH was funded by a Postdoctoral Fellowship inScience Technology and Environmental Policy at PrincetonUniversity MB was funded by a Graduate Research Fellowshipfrom the NSF EM was funded by the Oxfam Faculty Chairin Environmental and Resource Economics at UC BerkeleySMH served as consultant for Scitor a company that hasbeen analyzing security risks that emerge from climaticchanges Data and replication code for all results in this paperare available for download at httpcegaberkeleyeduassetsmiscellaneous_filesHsiangBurkeMiguel-2013-datazip Authorcontributions SMH and MB conceived of and designedthe study SMH and MB collected and analyzed dataSMH MB and EM interpreted the findings and wrotethe paper

Supplementary Materialswwwsciencemagorgcgicontentfullscience1235367DC1Supplementary TextFigs S1 to S4Tables S1 to S4References (140 141)

18 January 2013 accepted 23 July 2013Published online 1 August 2013101126science1235367

RESEARCH ARTICLE

contentscivol341issue6151pdf1235367pdf

Quantifying the Influence of Climate on Human Conflict

Estimation of Climate-Conflict Linkages

Research Design

Statistical Precision

Evaluating Whether an Effect Is Important

Functional Form and Evidence of Nonlinearity

Results from the Quantitative Literature

Personal Violence and Crime

Group-Level Violence and Political Instability

Institutional Breakdown

Synthesis of Findings

Generality Samples Spatial Scales and Rates of Climate Change

The Direction and Magnitude of Climatic Influence on Human Conflict

Publication Bias

Implications for Future Climatic Changes

Future Research

Plausible Mechanisms

Selecting Climate Variables and Conflict Outcomes

Conclusion

References and Notes

13 SEPTEMBER 2013 VOL 341 SCIENCE wwwsciencemagorg

Quantifying the Infl uence of Climate on Human Confl ictSolomon M Hsiang Marshall Burke Edward Miguel

Introduction Despite the existence of institutions designed to promote peace interactions between individuals and groups sometimes lead to confl ict Understanding the causes of such confl ict is a major project in the social sciences and researchers in anthropology economics geography history political science psychology and sociology have long debated the extent to which climatic changes are responsible Recent advances and interest have prompted an explosion of quantitative studies on this question

Methods We carried out a comprehensive synthesis of the rapidly growing literature on climate and human confl ict We examined many types of human confl ict ranging from interpersonal violence and crime to intergroup violence and political instability and further to institutional breakdown and the collapse of civilizations We focused on quantitative studies that can reliably infer causal associations between climate variables and confl ict outcomes The studies we examined are experi-ments or ldquonatural experimentsrdquo the latter exploit variations in climate over time that are plausibly independent of other variables that also affect confl ict In many cases we obtained original data from studies that did not meet this criterion and used a common statistical method to reanalyze these data In total we evaluated 60 primary studies that have examined 45 different confl ict data sets We collected fi ndings across time periods spanning 10000 BCE to the present and across all major world regions

Results Deviations from normal precipitation and mild temperatures systematically increase the risk of confl ict often substantially This relationship is apparent across spatial scales ranging from a single building to the globe and at temporal scales ranging from an anomalous hour to an anoma-lous millennium Our meta-analysis of studies that examine populations in the post-1950 era sug-gests that the magnitude of climatersquos infl uence on modern confl ict is both substantial and highly statistically signifi cant (P lt 0001) Each 1-SD change in climate toward warmer temperatures or more extreme rainfall increases the frequency of interpersonal violence by 4 and intergroup confl ict by 14 (median estimates)

Discussion We conclude that there is more agreement across studies regarding the infl uence of cli-mate on human confl ict than has been recognized previously Given the large potential changes in precipitation and temperature regimes projected for the coming decadesmdashwith locations through-out the inhabited world expected to warm by 2 to 4 SDs by 2050mdashamplifi ed rates of human confl ict could represent a large and critical social impact of anthropogenic climate change in both low- and high-income countries

FIGURES AND TABLE IN THE FULL ARTICLE

Fig 1 Samples and spatiotemporal resolu-tions of 60 studies examining intertemporal associations between climatic variables and human confl ict

Fig 2 Empirical studies indicate that clima-tological variables have a large effect on the risk of violence or instability in the modern world

Fig 3 Examples of paleoclimate reconstruc-tions that fi nd associations between climatic changes and human confl ict

Fig 4 Modern empirical estimates for the effect of climatic events on the risk of inter-personal violence

Fig 5 Modern empirical estimates for the effect of climatic events on the risk of inter-group confl ict

Fig 6 Projected temperature change by 2050 as a multiple of the local historical SD (σ) of temperature

Table 1 Primary quantitative studies testing for a relationship between climate and con-fl ict violence or political instability

SUPPLEMENTARY MATERIALS

Supplementary TextFigs S1 to S4Tables S1 to S4References (140 141)

minus2 minus1 0 1 2

minus40

∆ Pixel temp (degC)

t risk (

minus1 0 1 2∆ Pacific Ocean temp (degC)

minus05 0 05 1∆ Country temp (degC)

Local violence

East Africa

Civil conflict onset

Global tropics

Civil war incidence

Sub-Saharan Africa

Climate and confl ict across spatial scales Evidence that temperature infl uences the risk of modern human confl ict (A) local violence in 1deg grid cells (B) civil war in countries and (C) civil confl ict risk in the tropics The map depicts regions of analysis corresponding to nonparametric watercolor regressions in (A) to (C) The color intensity in (A) to (C) indicates the level of certainty in the regression line

READ THE FULL ARTICLE ONLINEhttpdxdoiorg101126science1235367

Cite this article as S Hsiang et al Science 341 1235367 (2013) DOI 101126science1235367

The list of author affi liations is available in the full article onlineCorresponding author E-mail shsiangberkeleyedu

RESEARCH ARTICLE SUMMARY

Published by AAAS

RESEARCHARTICLE

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10

RESEARCH ARTICLE

municipal

province

country

region

globalS

tial s

Hsiang et al(2011)

8000 BCE 0 1000 1800 1950 2000 2010

(2006)

Vrij et al (1994)

Americas

Africa

Global

Sampleperiod

Sampleregion

Timeunit

Spatialunit

Indepen-dent

variable

Dependentvariable

Stattest

Larg-eef-fect

Re-jectb = 0

Rejectb = 10 Re-

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10 Ref

RESEARCH ARTICLE

minus40

minus20

Levy et al (2005)

minus10 0 10

minus20

minus10

minus20

minus10

rminuste

minus2 minus1 0 1 2

minus50

minus100

minus50

Theisen (JPR 2012)

minus05 0 05 1

minus20

minus1 0 1

minus20

minus10

Burke (BEJM 2012)

minus1 0 1 2

minus20

minus1 0 1 2

minus20

minus80

minus40

minusM

minus5

RESEARCH ARTICLE

Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Mexico

(Wet) 40

(Dry) 20

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

FRoman

RESEARCH ARTICLE

minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCHARTICLE

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10

RESEARCH ARTICLE

municipal

province

country

region

globalS

tial s

Hsiang et al(2011)

8000 BCE 0 1000 1800 1950 2000 2010

(2006)

Vrij et al (1994)

Americas

Africa

Global

Sampleperiod

Sampleregion

Timeunit

Spatialunit

Indepen-dent

variable

Dependentvariable

Stattest

Larg-eef-fect

Re-jectb = 0

Rejectb = 10 Re-

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10 Ref

RESEARCH ARTICLE

minus40

minus20

Levy et al (2005)

minus10 0 10

minus20

minus10

minus20

minus10

rminuste

minus2 minus1 0 1 2

minus50

minus100

minus50

Theisen (JPR 2012)

minus05 0 05 1

minus20

minus1 0 1

minus20

minus10

Burke (BEJM 2012)

minus1 0 1 2

minus20

minus1 0 1 2

minus20

minus80

minus40

minusM

minus5

RESEARCH ARTICLE

Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Mexico

(Wet) 40

(Dry) 20

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

FRoman

RESEARCH ARTICLE

minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10

RESEARCH ARTICLE

municipal

province

country

region

globalS

tial s

Hsiang et al(2011)

8000 BCE 0 1000 1800 1950 2000 2010

(2006)

Vrij et al (1994)

Americas

Africa

Global

Sampleperiod

Sampleregion

Timeunit

Spatialunit

Indepen-dent

variable

Dependentvariable

Stattest

Larg-eef-fect

Re-jectb = 0

Rejectb = 10 Re-

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10 Ref

RESEARCH ARTICLE

minus40

minus20

Levy et al (2005)

minus10 0 10

minus20

minus10

minus20

minus10

rminuste

minus2 minus1 0 1 2

minus50

minus100

minus50

Theisen (JPR 2012)

minus05 0 05 1

minus20

minus1 0 1

minus20

minus10

Burke (BEJM 2012)

minus1 0 1 2

minus20

minus1 0 1 2

minus20

minus80

minus40

minusM

minus5

RESEARCH ARTICLE

Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Mexico

(Wet) 40

(Dry) 20

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

FRoman

RESEARCH ARTICLE

minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


municipal

province

country

region

globalS

tial s

Hsiang et al(2011)

8000 BCE 0 1000 1800 1950 2000 2010

(2006)

Vrij et al (1994)

Americas

Africa

Global

Sampleperiod

Sampleregion

Timeunit

Spatialunit

Indepen-dent

variable

Dependentvariable

Stattest

Larg-eef-fect

Re-jectb = 0

Rejectb = 10 Re-

StudySampleperiod

Sampleregion

Timeunit

Spatialunit

Independentvariable

Dependentvariable

Stattest

Largeeffect

Rejectb = 0

Rejectb = 10 Ref

RESEARCH ARTICLE

minus40

minus20

Levy et al (2005)

minus10 0 10

minus20

minus10

minus20

minus10

rminuste

minus2 minus1 0 1 2

minus50

minus100

minus50

Theisen (JPR 2012)

minus05 0 05 1

minus20

minus1 0 1

minus20

minus10

Burke (BEJM 2012)

minus1 0 1 2

minus20

minus1 0 1 2

minus20

minus80

minus40

minusM

minus5

RESEARCH ARTICLE

Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Mexico

(Wet) 40

(Dry) 20

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

FRoman

RESEARCH ARTICLE

minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


minus40

minus20

Levy et al (2005)

minus10 0 10

minus20

minus10

minus20

minus10

rminuste

minus2 minus1 0 1 2

minus50

minus100

minus50

Theisen (JPR 2012)

minus05 0 05 1

minus20

minus1 0 1

minus20

minus10

Burke (BEJM 2012)

minus1 0 1 2

minus20

minus1 0 1 2

minus20

minus80

minus40

minusM

minus5

RESEARCH ARTICLE

Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Mexico

(Wet) 40

(Dry) 20

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

FRoman

RESEARCH ARTICLE

minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


Europe

(Dry) 180

(Wet) 100

Cambodia

1250 1300 1400 1450 1500 1550 1600

(Dry) minus4

minus2

(Wet) 4

Mexico

(Wet) 40

(Dry) 20

(Wet) 20

(Dry) 08

Tiwanakuabandonment

1350900800

minus2

minus1

y (deg

Migrationperiod

Celticexpansion

30 YearsWar

Modernmigration

(Dry) 8

(Wet) 2

FRoman

RESEARCH ARTICLE

minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


minus8

minus4

minus8

minus4

liatio

Allstudies

Temperaturestudies

Median = 39

Mean = 23

RESEARCH ARTICLE

1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


1 2 3 40

Standard deviations

minus20

minus10

minus20

minus10

t al 2

itutio

Allstudies

Temperaturestudies

RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


RESEARCH ARTICLE

Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


Sci USA 107 E103 (2010) doi 101073pnas1005748107

RESEARCH ARTICLE



Research Design











Publication Bias


Future Research



Conclusion


Documents

Quantifying the Influence of Climate on Human Conflict ...users.clas.ufl.edu/prwaylen/Geo3280articles/Climate... · East Africa Civil conflict onset Global tropics Civil war incidence