Movatterモバイル変換

[0]ホーム
URL:

Skip to main content
NCBI home page
Search in PMCSearch
As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health.
Learn more:PMC Disclaimer | PMC Copyright Notice
Proceedings of the National Academy of Sciences of the United States of America logo

Civil conflict sensitivity to growing-season drought

Nina von Uexkulla,b,1,Mihai Croicua,Hanne Fjeldea,b,Halvard Buhaugb,c
aDepartment of Peace and Conflict Research, Uppsala University, SE-75120 Uppsala, Sweden;
bPeace Research Institute Oslo, NO-0134 Oslo, Norway;
cDepartment of Sociology and Political Science, Norwegian University of Science and Technology, NO-7491 Trondheim, Norway
1

To whom correspondence should be addressed. Email:nina.von_uexkull@pcr.uu.se.

Edited by B. L. Turner, Arizona State University, Tempe, AZ, and approved August 31, 2016 (received for review May 11, 2016)

Author contributions: N.v.U., M.C., H.F., and H.B. designed research; N.v.U. analyzed data; N.v.U., H.F., and H.B. wrote the paper; and M.C. created the dataset.

Issue date 2016 Nov 1.

Freely available online through the PNAS open access option.

PMCID: PMC5098672  PMID:27791091

Significance

Understanding the conflict potential of drought is critical for dealing effectively with the societal implications of climate change. Using new georeferenced ethnicity and conflict data for Asia and Africa since 1989, we present an actor-oriented analysis of growing-season drought and conflict involvement among ethnic groups. Results from naive models common in previous research suggest that drought generally has little impact. However, context-sensitive models accounting for the groups’ level of vulnerability reveal that drought can contribute to sustaining conflict, especially for agriculturally dependent groups and politically excluded groups in very poor countries. These results suggest a reciprocal nature–society interaction in which violent conflict and environmental shock constitute a vicious circle, each phenomenon increasing the group’s vulnerability to the other.

Keywords: armed conflict, climate variability, drought, ethnicity, georeferenced event data

Abstract

To date, the research community has failed to reach a consensus on the nature and significance of the relationship between climate variability and armed conflict. We argue that progress has been hampered by insufficient attention paid to the context in which droughts and other climatic extremes may increase the risk of violent mobilization. Addressing this shortcoming, this study presents an actor-oriented analysis of the drought–conflict relationship, focusing specifically on politically relevant ethnic groups and their sensitivity to growing-season drought under various political and socioeconomic contexts. To this end, we draw on new conflict event data that cover Asia and Africa, 1989–2014, updated spatial ethnic settlement data, and remote sensing data on agricultural land use. Our procedure allows quantifying, for each ethnic group, drought conditions during the growing season of the locally dominant crop. A comprehensive set of multilevel mixed effects models that account for the groups’ livelihood, economic, and political vulnerabilities reveals that a drought under most conditions has little effect on the short-term risk that a group challenges the state by military means. However, for agriculturally dependent groups as well as politically excluded groups in very poor countries, a local drought is found to increase the likelihood of sustained violence. We interpret this as evidence of the reciprocal relationship between drought and conflict, whereby each phenomenon makes a group more vulnerable to the other.

There is increasing acceptance within policy and national security circles that climate change and extreme weather events constitute a significant threat to societal stability and peace (1,2). Despite evidence provided by a few idiographic studies (3,4), the conflict research community has yet to agree on a statistical pattern consistent with a general causal climate–conflict connection (5–7). One reason for the scientific conundrum may be the failure to properly specify the socioeconomic and political context within which climatic extremes can undermine social stability and increase conflict risk. Drawing on insights from theoretical and single-case empirical research, we propose a conditional model of climate–security connections that explicitly considers the affected population’s socioeconomic context. Specifically, we examine how growing-season drought across the agricultural lands of spatially defined ethnic groups affects the risk that the groups engage in conflict against the state and the extent to which this effect is conditioned on the groups' livelihood vulnerability, political status, and economic development. In doing so, our study provides the most appropriate large-scale test to date of dominant environmental security thinking (8).

In brief, a group’s vulnerability to climatic extremes can be considered a function of its dependence on renewable resources, the sensitivity of that ecosystem to environmental changes, and the group’s coping capacity (9). The livelihoods of farming communities living off nonirrigated lands are often identified as particularly vulnerable (10,11). Central factors restricting coping capacity include a low level of socioeconomic development, a history of conflict, and limited access to economic and social capital that could facilitate alternate modes of livelihood (12,13). In addition, societal groups that are excluded from political processes are much less likely to be on the receiving end of government-sponsored relief aid and compensation programs in the wake of disaster (14,15).

Climate-induced crop failure or loss of pasture may imply a dramatic income loss, and limited material and human capital will aggravate the situation by narrowing the range of outside options. However, this process by itself does not explain how organized violent conflict might erupt or be sustained. Organizing unrest requires agency, a perception of common identity, and in the case of civil conflict, a belief that the government is to blame for the misery (16). Preexisting social structures, oftentimes in the form of ethnonational identities, constitute a key element necessary to solve the collective action problem for mobilization (17). In large parts of the developing world, particularly in Africa and Asia, ethnicity constitutes the predominant societal cleavage around which social identity and political preferences are formed and play out (18–20). Indeed, most contemporary civil conflicts are fought along ethnic lines, and ethnic conflicts have increased markedly since the end of the Cold War (21). For these reasons, the conflict potential of economic hardships is considered especially high where these coincide with distinct ethnic identities (22,23).

Analytical Approach

Earlier quantitative comparative assessments of the climate variability–armed conflict link typically rely on country-averaged data or use arbitrarily defined grid cells as units of analysis (24–27). Despite their merit, both approaches have notable limitations; country-level data mask considerable within-country variation in environmental and political conditions and may miss localized phenomena, whereas disaggregated grid analyses typically require spatial overlap between the treatment (climatic anomaly) and the outcome (conflict) for an effect to be detected. Neither approach is suited to capture and evaluate the group-level dynamic outlined above.

Remedying this shortcoming, we take advantage of a new generation of georeferenced data on ethnic groups (28), which we link to specific armed conflict events (29) by considering the ethnic claims and recruitment strategy of each nonstate conflict actor in each armed civil conflict (30). This linking procedure is an important innovation, because it permits considering how local drought affects ethnic groups’ conflict behavior, irrespective of the actual location of fighting relative to the drought or the group’s homeland. Moreover, by overlaying the spatial ethnicity layers with high-resolution land use rasters (31) and monthly remote sensing-based drought statistics (32), we are able to calculate, for each ethnic group and each calendar year, cropland-specific drought during the growing season of the dominant local crop (33). We use a group-specific Standardized Precipitation Evapotranspiration Index (SPEI) as our primary indicator of drought, because it captures both precipitation anomalies and variations in determinants of evaporation (e.g., temperature and wind speed). We focus on drought, because it is the environmental condition widely assumed to carry the largest conflict potential.

In the empirical analysis presented below, we consider both naive models, in which local drought is assumed to have a direct and general effect on conflict risk, and context-sensitive models, where the effect of drought is conditioned on the groups’ livelihood vulnerability (agricultural dependence, expressed as share of the group’s settlement area covered by cropland), economic vulnerability (local economic development, expressed as area average night light emission per group settlement polygon), and political vulnerability (ethnopolitical exclusion). In all, we investigate subnational climate–conflict dynamics across Africa and Asia, 1989–2014, covering a larger spatiotemporal domain than any previous study of this kind. The geographical scope not only accounts for the large majority of the world’s armed conflicts, but it also encompasses the agricultural areas that are most vulnerable to drought and other extreme weather events (34–36). Additional details and alternative operationalizations are described inSI Text.

Figs. 1 and2 give a visual representation of key vulnerability dimensions captured in the analysis. The extent of agricultural dependence, used as proxy for livelihood vulnerability, varies substantially between countries as well as between ethnic groups within countries (Fig. 1). The largest shares of cropland are found in South and Southeast Asia. Most groups in Africa have modest proportions of agricultural lands, where pastoral and agropastoral systems constitute a more prominent part of the rural economy. The conflict data also reveal a distinctly nonrandom spatial pattern with notable clusters in West Africa, the African Great Lakes region, and the Horn, as well as in South Asia.Fig. 2 shows country-level infant mortality rates (IMRs) assigned to the respective groups. IMR arguably constitutes the best aggregate proxy for country-level socioeconomic development, being strongly and inversely related to human welfare and positively associated with state fragility (37,38). In contrast to gross domestic product (GDP) and other indicators of macroeconomic performance, IMR is less immediately affected by commodity price fluctuations and global market forces and also less endogenous to armed conflict. We use IMR statistics to identify a subset of “most likely” cases (i.e., ethnic groups in countries marked by chronic poverty and weak political institutions, which are conditions that can critically stunt communities’ capacity to cope with agroeconomic shocks) (9,39). Again, the data reveal large cross-sectional variation, with sub-Saharan Africa being the region of particular concern.

Fig. 1.

Fig. 1.

Agricultural dependence by ethnic group settlement area and location of armed conflict events according to the Uppsala Conflict Data Program (UCDP) GED dataset, 1989–2014. Gray denotes areas and groups with insufficient data.

Fig. 2.

Fig. 2.

Country-level IMRs assigned to respective ethnic groups in 2000. Gray denotes areas and groups with insufficient data.

Results

Fig. 3 presents the main results from the statistical analysis. The marginal effects displayed are derived from a comprehensive set of multilevel random effects models of group-specific armed conflict onset and incidence as a function of observed growing-season drought in each group’s agricultural areas. We estimated both direct drought effects and effects conditional on the groups’ local socioeconomic and political context. Furthermore, we alternated between models containing the complete Africa–Asia sample and models covering a subset of ethnic groups in countries where socioeconomic development is especially low (IMR > sample median).

Fig. 3.

Fig. 3.

Marginal effects of (A) growing-season drought and (B) cumulative years of growing-season drought on civil conflict onset and incidence for full sample (black) and high-IMR subsample (pink). Direct effects of drought are estimated in column 1; columns 2–4 show effects conditional on high agricultural dependence (85th percentile), low development (15th percentile), and ethnopolitical exclusion, respectively. Whiskers represent 95% confidence intervals.Tables S3–S6 have details on the estimated models.

Overall, the plots reveal a mixed pattern; the point estimates for most effects are positive, indicating that the occurrence of drought is more likely to be associated with an increase than a decrease in local civil conflict risk. At the same time, most coefficients fail to obtain statistical significance at conventional confidence levels. However, some models are better approximations of the proposed conditional relationships than others, and therefore, it would be a mistake to draw conclusions merely on the basis of the median or mean marginal effect among all models shown inFig. 3. In particular, we note that drought has a stronger association with conflict when it is modeled as a function of the group’s agricultural dependence. In other words, the occurrence of drought during the growing season of the main local crop increases the risk of conflict involvement for groups with large shares of cropland but not necessarily for predominantly nonagrarian groups. Moreover, across nearly all specifications, the estimated conflict-inducing effect of drought is larger for the subsample of groups located in the least developed countries—precisely where we would expect a causal link to be most likely to materialize. The analysis further suggests that drought has a larger and more systematic effect on civil conflict incidence than on onset. We return to the implication of this insight inDiscussion.

In terms of marginal impact,Fig. 4A shows how the effect of drought on conflict incidence increases with the group’s share of agricultural land. The positive effect is especially pronounced for the subset of groups in low-development countries, although the shape of the slope is determined by a modest fraction of the sample (Fig. 4B). For most groups, the estimated effect of drought is not significantly different from zero.Fig. 4C shows that conflict risk increases for each additional year of growing-season drought. For the average politically excluded group, an increase from no drought to 5 consecutive years of drought during the local growing season increases the estimated likelihood of conflict incidence from 12 to 15% ceteris paribus.

Fig. 4.

Fig. 4.

Marginal effects as a function of agricultural dependence for (A) the full sample (n = 5,381) and (B) the high-IMR subsample (n = 2,733).C shows the predicted risk of conflict incidence for each additional year of growing-season drought for groups in the high-IMR subsample (n = 2,733). Superimposed bars represent the distributions of observations. Details are inTables S3 andS4.

The main analysis, depicted inFig. 3, captures a variety of plausible combinations of local environmental conditions and conflict outcomes. Next, we carried out a set of sensitivity tests to explore the robustness of these patterns. First, we replaced the SPEI indicators with simpler group-specific measures of negative precipitation deviation. The results for agricultural-dependent groups remain substantively unchanged. Second, we relaxed the ethnic link criterion and consider all conflict events occurring within each ethnic group’s settlement area by means of a simpler spatial overlay procedure, analogous to the approach of earlier research. This test yielded more mixed results, which may be partly because the alternative conflict variables ignore relevant events occurring outside each groups’ homeland and partly because of noise introduced by additionally accounting for conflict events not involving the local population. Moreover, we found that the significant relationship between drought and conflict incidence is robust to replacing the two-level mixed effects logit estimator with ordinary least squares regression with group-level fixed effects. Details are inTables S1 andS7–S9.

Table S1.

Cross-tabulation of ethnicity- and spatial overlay-based conflict variables

Ethnic conflict onset/incidenceSpatial onsetSpatial incidence
0101
06,3532385,149845
15459130583

The table shows the number of observations (group-years) by outcome on the main ethnic conflict variables and the alternative spatial overlay variables. For the purpose of this cross-tabulation, ongoing conflict is coded as zero for the onset variables in contrast to the operationalization chosen in the main models, where it is coded missing.

Table S7.

Robustness drought agricultural dependence interaction

graphic file with name pnas.1607542113st07.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). In this table, we evaluate the robustness of the results originally reported in model 10 and replicated in model 33 against a similarly specified model 35 that uses the alternative spatial overlay-based conflict variable (Alt. DV). Model 34 presents results from a group fixed effects OLS regression (OLS fe), whereas model 36 uses standardized negative growing-season precipitation anomaly as an alternative indicator of drought (Alt. IV). All models in white are estimated on the full Africa–Asia sample. The same is done for model 12 replicated as model 37 on the sample of low-development countries (pink) in models 38–40. Models 34 and 38 are based on a split sample of observations with agricultural dependence larger than the sample median. Models 35 and 39 use alternative controls for temporal (decay) and spatial dependence (other groups in conflict) based on spatial overlay of conflict events and ethnic group polygons. All dependent variables (DV) are conflict incidence.

Table S9.

Robustness cumulative drought exclusion interaction—high IMR subsample

graphic file with name pnas.1607542113st09.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). The models represent an extension ofTable S8 by introducing an interaction effect between cumulative drought and ethnopolitical exclusion. Model 45 is identical to model 32 (Table S6). Model 46 is estimated on a subsample of excluded groups and including group fixed effects. All DVs are conflict incidence. All models are estimated on the subsample of low-development countries (IMR > median; pink).

Table S8.

Robustness cumulative drought direct relationship–high IMR subsample

graphic file with name pnas.1607542113st08.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). In this table, we evaluate the robustness of the results originally reported in model 8 and replicated in model 41 against a similarly specified model 43 that uses the alternative spatial overlay-based conflict variable. Model 42 presents results from a group fixed effects OLS regression, whereas model 44 uses standardized negative growing-season precipitation anomaly as an alternative indicator of drought. All DVs are conflict incidence. The models are estimated on the subsample of low-development countries (IMR > median; pink).

Discussion

This study has investigated the extent to which the conflict behavior of spatially defined ethnic groups is sensitive to agroeconomic shocks. Although ethnicity is not the only marker of identity conducive to violent mobilization and collective action, it is the dominant societal cleavage in most countries in the developing world today (21). However, not all politically relevant ethnic groups are equally vulnerable to an environmental shock. Some groups constitute a crucial part of a regime’s “winning coalition” (40) and thereby enjoy economic benefits and political privileges. Even more important, groups vary in their dependence on agricultural production for income and livelihood. Accordingly, it should not come as a surprise that the majority of groups in this analysis seem relatively resilient to drought. However, for politically marginalized and agriculturally dependent groups—and especially those residing in countries characterized by very low socioeconomic development—the analysis has detected a consistent statistical pattern, whereby the occurrence and duration of drought increase the likelihood of sustained conflict involvement.

Overall, the models that estimated conflict incidence are more supportive of the drought hypothesis than the onset models. Although this result may partly be a consequence of the rare nature of conflict outbreak and consequently, few non-zero observations, we interpret this as evidence of a powerful reciprocal relationship between armed conflict and local drought, whereby each phenomenon makes a group more vulnerable to the other. Sustained fighting and insecurity deter investments, trigger capital flight, undermine public goods delivery, and have negative health implications (41,42), all of which may decrease the local population’s ability to cope with increased environmental hardship and increase their incentives to sustain ongoing resistance (43,44). The Tigrayan rebellion in Ethiopia and the Maoist insurgency in India are two examples where such a dynamic reportedly played out (45,46). We find less evidence that the occurrence or duration of drought significantly affects the risk of new conflict breaking out, even in the most likely subset of agriculturally dependent or politically marginalized ethnic groups in countries with very low economic development. Accordingly, although some research has suggested that a severe drought contributed to the onset of the Syrian civil war (3), our results do not indicate that this case is representative of a large number of contemporary conflicts.

Although this study provides insight regarding a more nuanced and conditional climate–conflict dynamic, it is clear that drought explains a small share of the observed variation in conflict involvement, implying that the substantive effect is modest compared with central drivers of conflict, such as ethnopolitical exclusion, temporal and spatial proximity to violence, and various country-specific risk factors. That said, this study has focused on only one theorized causal pathway. More research is needed to properly evaluate alternative mechanisms through which drought may translate into societal instability (e.g., forced displacement and consumer price shocks) as well as the impact on other manifestations of collective action (e.g., communal conflict and urban rioting but also, increased cooperation and conflict resolution).

Based on the most comprehensive and theoretically consistent assessment of its kind to date, we conclude that the impact of drought on conflict under most circumstances is limited. However, for segments of the population that are particularly vulnerable to natural forces because of their dependence on agriculture, drought does significantly increase the likelihood of sustained conflict, especially among groups in the least developed states. A severe drought threatens local food security, aggravates humanitarian conditions, often triggers large-scale human displacement, and as our results indicate, may also provide the breeding ground for sustained fighting. These results speak to the importance of strengthening adaptive capacities of agriculturally dependent communities, in particular in areas already affected by conflict. Conflict is development in reverse (47,48), and the adverse impact of intense violence on environmental vulnerability is probably many times greater than the effect of environmental shock on conflict risk. The international community should, thus, be especially attentive to climatic disasters striking zones of chronic conflict.

Materials and Methods

We estimate a series of two-level mixed effects logistic regression models, which allows examining both spatial and temporal variations in conflict risk across our observations, while taking into account dependence between observations within countries and groups over time. We specify random intercepts to account for systematic differences in baseline conflict risk between groups belonging to different countries, whereas the slopes or variable effect sizes are assumed constant across all groups.

The units of analysis are yearly observations of politically relevant ethnic groups with a geographical base drawn from the Ethnic Power Relations dataset (17,49). A group is politically relevant if at least one political organization claims to represent it in national politics or its members are subjected to state-led political discrimination. We exclude from our sample ethnic groups that dominate or have monopoly over governmental power, because these cannot simultaneously constitute a nonstate opposition actor.

Data on civil conflict events are taken from the Uppsala Conflict Data Program Georeferenced Event Dataset (UCDP GED), version 4.0., which records all lethal events in armed conflicts between state and nonstate actors that claim at least 25 battle-related deaths during at least one calendar year (29). We combine the conflict data with information on the ethnic claims and recruitment profiles of the nonstate actors from the ACD2EPR dataset (30). For each group-year, we code two alternative binary dependent variables. Onset is coded one in the first year that a lethal event can be linked to the ethnic group (recording a new onset after two calendar years of inactivity). Subsequent years of fighting are coded as missing, because groups involved in ongoing conflict are not at risk for experiencing a new onset. In addition, we consider conflict incidence, which is assigned the value of one in all years with at least one lethal event involving the group. Both variables are otherwise coded zero.

The independent variable is the severity of growing-season drought for crops cultivated within the group settlement areas. We first overlay ethnic settlement polygons (Geo-EPR-ETH v.2.0) with 0.5 × 0.5-decimal degrees grid cells using the PRIO-GRID structure (50). For each grid cell, we identify the primary crop [i.e., the crop covering the largest physical surface based on data from the Spatial Production Allocation Model (SPAM) 2005 (31)]. Growing seasons for each primary crop are identified using the Monthly Irrigated and Rainfed Crop Areas (MIRCA) cropping calendars (33) complemented by the SAGE Crop Calendar Dataset (51). This information is combined with monthly drought data by means of the SPEI, which considers the joint effect of precipitation and potential evapotranspiration compared with historical normal conditions (52). The severity of the growing-season drought is measured as the share of the growing season where the SPEI index is at least 1 SD below normal levels. The cell-specific drought index is then aggregated to the ethnic group level by taking the mean of all cells falling within each settlement polygon. We also construct a cumulative drought measure, which counts the number of consecutive years (up to 5 y) with growing-season drought values above the sample median. Additional details on the drought indices are inSI Text andFig. S1.

Fig. S1.

Fig. S1.

Stepwise construction of group-specific growing-season drought indicator. (A) Identification of main crop in 0.5 × 0.5-decimal degrees grid cell. (B) Identification of growing-season drought conditions (SPEI ≤ −1). (C) Annual growing-season drought measure calculated as share of the growing season of the main crop with drought conditions. (D) Repetition of stepsA–C for all grid cells with cropland. (E) Spatial overlay of ethnic group settlement polygons and grid cells. (F) Final drought indicator calculated as yearly mean of growing-season drought for all cells assigned to the group.

We construct three different group-level measures of vulnerability. First, we calculate agricultural dependence defined as the proportion of each ethnic group’s settlement area that overlaps with cropland according to the SPAM 2005 dataset (31). Second, we include a group-based measure of economic development, taking the mean yearly pixel value of night light emissions for each group’s settlement area from the Defense Meteorological Satellite Program—Operational Linescan System (53). The variable is log-transformed before use. Third, we measure ethnopolitical exclusion as a binary indicator based on the Ethnic Power Relations dataset (17).

Because we anticipate systematic differences in groups’ vulnerability to environmental shocks between countries, we account for the host countries’ level of socioeconomic development represented by IMRs from the World Development Indicators (48). IMR measures the number of infants dying before reaching 1 y of age per 1,000 live births in a given year. Some models are estimated on a subset of the full sample, where we only include groups in the least developed countries [i.e., where the IMR score is above the sample median score (IMR > 60)].

As control variables, we include the size of the population living in the ethnic group settlement areas from the Center for International Earth Science Information Network (CIESIN) obtained via the Geographical Research on War, United Platform (GROWup) database (53,54). We also include an indicator of country-level gross domestic product per capita (GDPpc) from the World Development Indicators (55). Both variables are log-transformed before use. We account for temporal dependence in conflict risk through a decay function of time since the ethnic group last was involved in an armed conflict with the state. The risk of spillover effects from nearby conflict is accounted for by including a dummy variable indicating whether at least one other ethnic group in the same country was engaged in armed conflict against the state during the previous year. Finally, we include a group-specific time trend to account for possible trending patterns in, for example, climate variability and agricultural productivity (56). All independent variables are entered into the models att − 1 to ensure the correct sequence of events. Replication data are available from PRIO’s data repository athttps://www.prio.org/Data/Replication-Data as well as fromwww.pcr.uu.se/data.

SI Text

Introduction.

This document provides details on sources and measurements of the empirical data, tables with complete model output of all results reported in the text, and documentation of additional sensitivity tests. Replication data are available from PRIO’s data repository athttps://www.prio.org/Data/Replication-Data as well as fromwww.pcr.uu.se/data.

Data Structure.

The dataset used in the analysis consists of yearly observations of all politically relevant ethnic groups for all countries in Africa and Asia (excluding the Middle East) for the period 1989–2014 (Fig. 1). The spatiotemporal domain was determined by the coverage of the conflict data. The definition of ethnicity follows the Ethnic Power Relations (EPR-ETH) project (17) and captures ethnolinguistic, ethnosomatic, and ethnoreligious differences but not clans and tribes. An ethnic group is considered politically relevant if at least one significant political actor claims to represent the interests of that group in the national political arena or if group members are systematically and intentionally discriminated against in the domain of public politics. The EPR-ETH project classifies all groups according to their status and influence on national politics ranging from monopoly on state power to active and targeted discrimination. Given our focus on the behavior of nonstate actors in response to local drought, we exclude groups that enjoy “monopoly” or “dominant” power; by definition, these groups cannot constitute the nonstate challenger in an ethnic conflict against the state.

The spatial delineation of ethnic groups is taken from the GeoEPR-ETH v.2.0 dataset (49). This data source contains geographic information on the settlement areas of all ethnic groups included in the EPR-ETH dataset in the form of polygon shapefiles stored in a Geographic Information System. Transnational ethnic groups with settlement areas that span international borders are treated as separate observations in each country. For instance, the Uzbeks have separate settlement polygons for Afghanistan, Uzbekistan, and Tajikstan. The GeoEPR data are time-variant, reflecting significant changes in the ethnic geography and national boundaries over time, although for most groups in our sample, the polygons do not change. In total, our dataset contains time-varying statistics for 316 unique politically relevant ethnic groups (excluding monopoly and dominant groups) in 63 countries in Asia and Africa between 1989 and 2014.

Note that we aggregate the group-level data to the calendar year, although we make use of monthly climate statistics and conflict data that register each battle event by date. Many ethnic groups involved in active conflict do not engage in battles in every month. This pattern is especially true for groups residing in regions with significant seasonal variation, where environmental conditions hamper military activities during some months (57), and groups involved in low-intensity conflicts. By relying on yearly data, we minimize the problem of conflating nonconflict observations that are truly peaceful with nonconflict observations that reflect a random or tactical lull in fighting.

Dependent Variable: Ethnic Civil Conflict.

The primary conflict variables are group-specific binary indicators of civil conflict outbreak and incidence based on the new UCDP Georeferenced Event Dataset (GED), version 4.0 (29), which contains detailed information on the location and timing of all civil conflict battles in Africa and Asia 1989–2014. The definition of civil conflict follows that of the parent UCDP/PRIO Armed Conflict Dataset (58,59) [i.e., armed contests between a government and a nonstate actor over governance and/or a specific territory that result in at least 25 battle-related deaths (BRDs) in a calendar year]. However, the UCDP GED dataset stores information on relevant battle events even for years when the conflict failed to cross the 25 BRDs, which also are taken into account here. Because our units of observations are ethnic groups, we only consider conflict events involving rebel organizations that recruit from or claim to fight on behalf of a politically relevant ethnic group. The linking of the UCDP GED conflict data to the EPR groups was made possible by drawing on the ACD2EPR dataset (30,60) updated through 2014 by the authors. By linking events to the identity of the actors, we capture all relevant events for each ethnic group, regardless of whether the fighting takes place within or outside the group’s settlement area. This design is an important refinement over earlier geographically disaggregated research, which typically requires spatial overlap between the unit of analysis (grid cell or administrative region) and the phenomenon of interest (conflict) for an event to be counted.

We aggregate the events to the group-year level and construct two binary dependent variables.

  • Onset: coded one in the first year of reported fighting for the group as well as in the first year of fighting after at least 2 calendar years without reported battle events. Subsequent years of fighting are coded as missing (i.e., the group is not at risk for initiating another conflict with the state), whereas years without reported fighting involving the group are coded zero.

  • Incidence: coded one for all years of reported fighting for the group and zero otherwise.

Some civil conflicts involve several ethnic groups fighting jointly against government forces. In these cases, each group is coded as taking part in the conflict in accordance with the battle event information provided in the UCDP GED dataset. Accordingly, groups that are part of the same conflict may be coded with onset and incidence in different years depending on when the rebel organizations that they support were engaged in military battles against the state.

In sensitivity tests, we relax the identity criterion (i.e., the explicit link between ethnic groups and rebel organizations’ claims/recruitment strategy) and consider all battle events that occur within each ethnic group’s settlement area, regardless of the actor and nature of the civil conflict event. The “spatial overlay” variables (onset and incidence) were generated by intersecting georeferenced UCDP GED point data with yearly group polygons; whenever a battle event occurred within the perimeters of a group’s settlement area, that group was coded as being part of a civil conflict without considering the true link between the rebel organization and the local population. Naturally, the alternative onset and incidence variables contain a larger number of positive observations, because they also capture conflicts that are nonethnic in nature (Table S1).

Independent Variable: Growing-Season Drought.

Distinctly different environmental shocks (e.g., floods, droughts, storms, and heatwaves) are unlikely to generate the same social response within the same spatiotemporal domain (7,61). To increase analytical consistency and maximize the probability of detecting a general effect, we, therefore, limit focus to the environmental condition widely assumed to carry the largest conflict potential: drought. The main independent variables are two group-specific annualized measures of growing-season drought in the group’s crop-producing areas. The construction of these indicators is visualized inFig. S1 and explained in further detail below.

  • a)

    We first divide the world into 0.5 × 0.5-decimal degrees grid cells using the PRIO-GRID structure (50), into which we import crop production data from the SPAM 2005 (62). For each grid cell, we identify the main crop defined as the crop type covering the largest physical surface. Cells without crop production are coded as missing.

  • b)

    Next, we import gridded monthly SPEI-1 data from SPEIbase v.2.3 (32,52). The SPEI-1 index gives the difference between observed climatic water balance [precipitation minus potential evapotranspiration (PET) for a given month] and the long-term (1901–2013) and historical median water balance for the same month. PET is calculated in accordance with the Food and Agriculture Organization of the United Nations (FAO)-56 Penman–Monteith estimation of PET, taking into account the influence of daily wind speed, relative humidity, and solar radiation as well as temperature levels. For any location, SPEI = 0 represents the median observed value for that month during the complete time series (1901–2013); SPEI = −1 is equal to 1 SD below the local median water balance, which serves as our threshold for short-term drought.

  • c)

    In a third step, we identify the growing-season months for the main crop in the grid cell based on the MIRCA cropping calendar (33) complemented by the SAGE Crop Calendar Dataset (51). We then construct a simple yearly grid-level drought variable operationalized as the share of growing-season months with SPEI ≤ −1. InFig. S1C, the observed cell is assigned a value of 0.6 (60%), consistent with 3 of 5 growing-season months (April to August) being unusually dry.

  • d)

    This procedure is repeated for all cells in the grid for all years in the spatiotemporal sample: Asia and Africa, 1989–2014. The result of this stage is a dataset of PRIO-GRID cells containing information on the main crop in each cell as well as a yearly measure of the fraction of each main crop’s growing season affected by drought.

  • e)

    The next task is to aggregate the yearly grid cell drought measures to the level of ethnic group. This task is accomplished by importing ethnic group settlement polygons from GeoEPR (49) into the grid and then assigning each cell to the group that covers the largest share of the cell’s terrestrial area. Each cell is allocated to one and only one group.

  • f)

    In a final step, we construct the main independent variable, ethnic group-level SPEI drought, by taking the yearly mean of the growing-season drought values for all cells assigned to the group. The drought indicator has a theoretical range from zero [implying that no part of the group’s crop-producing areas experienced anomalously dry conditions (SPEI-1 ≤ –1) during any month of the growing season] to one (all grid cells covering agricultural areas experienced anomalously dry conditions during all months of the growing season).

Because successive years of drought erode coping capacity and have especially adverse impacts on agricultural income and food production (63), we also create a group-specific cumulative drought variable that counts the number of consecutive years up tot with drought values, as defined above, larger than the sample median (>0.152) up to a maximum of 5 y. We use the complete time series of SPEI data (1901–2013), so that early years in our dataset have the same chance to observe successive droughts as later years.

Conditioning Factors.

As outlined in the text, we find it unreasonable to assume that a growing-season drought of some magnitude has the same knock-on effect on societal stability across contexts. For this reason, we consider three group-level factors that influence a group's vulnerability to environmental shocks as well as a country-level factor that captures socioeconomic development.

Agricultural dependence.

A group’s vulnerability to growing-season drought is directly related to its extent of agricultural production; where large segments of group members are reliant on agriculture for livelihood and food supplies, crop failures will have a more devastating impact. To capture this variation, we create a variable agricultural dependence, which gives the share of each group’s settlement area (49) covered by cropland (62). Obviously, this variable is a relatively crude measure; although it is likely to correlate with the share of the population employed in the agricultural sector, it says little about vulnerability to weather anomalies as such and also does not capture variations in productivity and technological sophistication. We considered constructing separate drought and agricultural dependence indices for irrigated vs. nonirrigated areas but ultimately, decided that the coarse, time-invariant irrigation data were unsuitable. However, a country-averaged agricultural dependence indicator (cropland area as share of total land area) compares reasonably well (r = 0.18) with statistics from the World Development Indicators (55) on the share of the country’s population employed in agriculture (Fig. S2).

Fig. S2.

Fig. S2.

Country-level agricultural dependence vs. share of population used in agriculture.

Economic development.

Poverty and low economic development are generally associated with low coping capacity and limited outside options in the wake of a drought (11), increasing the incentives for protest and rebellion. Although conventional statistics of income and welfare are unavailable for most groups in our sample, we proxy local economic development using remote sensing data on night light emission. These data originate from the Defense Meteorological Satellite Program—Operational Linescan System and were obtained through the GROWup data portal (53). We take the natural log of the average pixel emission values by group polygon as an indicator of local economic development. Earlier research has shown that night light emission data are useful to detect spatial variation in wealth within countries (64).

Exclusion.

Ethnic groups that are politically marginalized or subject to outright discrimination are overrepresented among actors in armed conflict. Aside from its direct effect on opportunities and motivation for protest (65), the configuration of political power may also exert an indirect influence on conflict propensity by denying or restricting assistance and compensation to excluded groups in the wake of climatic disasters, thereby increasing frustration and animosity toward the central government (15). To capture the political dimension of vulnerability, we include a binary exclusion variable that identifies whether the group is excluded from national political processes in the given year according to the EPR-ETH dataset (17).

IMR.

In addition to group-level vulnerability, the society’s overall capacity to mitigate negative effects of climate shocks is an important determinant of conflict risk. To account for the groups’ larger socioeconomic context, we use time-varying statistics on countries’ IMRs. Unlike GDP per capita and other common measures of macroeconomic development, IMR directly captures human wellbeing—arguably the best indication of socioeconomic development—and it is less immediately affected by fluctuations in international commodity prices and global financial trends. IMR is defined as the number of infants dying before reaching 1 year of age per 1,000 live births in a given year. Data are taken from the World Development Indicators (55). We use the sample median IMR score to identify the subset of least developed countries (IMR > 60), for which we run separate regression models.

Control Variables.

To account for structural variations in baseline risk, we include a limited set of controls. First, we account for group size (i.e., the size of the population living in each ethnic group’s settlement area) based on high-resolution population raster data from Gridded Population of the World v.3 (54) obtained through the GROWup database (53). Data are available for 5-y intervals starting in 1990. We interpolate values between data points and take the natural logarithm to account for a nonlinear effect. Second, we control for the host countries’ economic performance by means of (log) GDP per capita in constant 2005 purchasing power parity-adjusted US dollars taken from the World Development Indicators (55). In addition, we include three group-specific controls for serial and spatial dependence: a decay function of the number of years since the previous conflict involving the group (half-life parameter α = 2), a dummy variable indicating whether at least one other ethnic group in the country was involved in civil conflict during the previous year, and a group-specific linear time trend to account for trends in climate variables as well as trends in agricultural productivity unrelated to drought.

Estimation Strategy and Sensitivity Tests.

The complete dataset consists of nested observations, where a group’s conflict risk, the binary dependent variable, is determined partly by observed group-specific characteristics and partly by factors that apply to all groups in the host country. Our key explanatory variable, drought, is a shock variable that varies across groups as well as over time, whereas the conditioning factors exhibit mostly cross-sectional variation. For this reason, a two-level mixed effects logistic regression is our preferred model, which allows examining both spatial and temporal variations in conflict risk while taking into account dependence between observations within countries and groups over time (66). We specify random intercepts to account for systematic differences in baseline conflict risk between groups belonging to different countries, whereas the slopes or variable effect sizes are assumed constant across all groups, ceteris paribus. To ensure a correct sequencing of events, all time-varying right-hand side regressors are applied a 1-y time lag.

Alternative model specification.

In sensitivity tests (Tables S7–S9), we use an alternative identification strategy, running ordinary least squares (OLS) regression with group fixed effects that are robust to unobserved heterogeneity between groups but limited to estimating strictly temporal effects. In these models, we exclude control variables that may be endogenous to temporal variations in drought. The fixed effects models corroborate the main results, where the association between drought exposure/duration and conflict incidence is most pronounced for agriculturally dependent groups (models 34 and 38/models 42 and 46).

Alternative dependent variable.

As referred to in the presentation of the conflict data above, we also explore an alternative conflict indicator in sensitivity tests, coding conflict involvement based on simple spatial overlay between the groups settlement areas and georeferenced conflict events. Although we consider this simplistic approach inferior—among other things, it ignores available information about conflict actors and their relation (or lack thereof) to the local ethnic group, and it also misses conflict events taking place outside a participating group’s main territory—it has the advantage of also capturing nonethnic conflict. In models using this alternative dependent variable, we include a spatiotemporal lag of conflict events as control for conflict diffusion. The spatiotemporal lag is coded one if there was at least one civil conflict event according to the UCDP GED (29) located within a 150-km buffer zone around each group’s polygon in the previous year and zero otherwise. These alternative models (models 35, 39, 43, and 47) provide mixed results for the drought–conflict relationship. However, we note that most other explanatory variables also present very different associations with the spatial overlay-based conflict variable. We believe the weaker results for the alternative conflict indicator may be ascribed at least partly to its ignorance of events that, for tactical reasons, take place outside the group’s homeland as well as data noise from including events that may be unrelated to the wellbeing of the local population.

Alternative independent variable.

The main models use a group-aggregated drought index (SPEI-1), which combines local rainfall and temperature statistics. Both temperature and precipitation matter for agricultural output, which makes this measure a more convincing drought measure than precipitation or temperature alone (52). In sensitivity tests, we consider negative precipitation anomaly as an alternative and simpler proxy for agricultural income shock. The variable is constructed in a similar manner as the main drought measure: intersecting high-resolution precipitation data with gridded data on agricultural areas and then, calculating group-averaged standardized deviations from historical mean rainfall levels for each year’s growing-season months. The values of the variable negative precipitation anomaly are reversed, so that the larger the value, the larger the negative rainfall anomaly. We also create an alternative group-specific cumulative drought variable based on the precipitation data that counts the number of consecutive years up tot with negative rainfall anomaly values, as defined above, larger than the sample median (>0.0256) up to a maximum of 5 y. The precipitation data originate from the Global Precipitation Climatology Centre (67) and were obtained via PRIO-GRID (50). Although results examining conditional effects based on agricultural dependence remain substantially unchanged (models 36 and 40), the results for cumulative drought measures are less significant with the alternative operationalization (models 44 and 48).

Tables.

Tables display summary statistics (Table S2) and the results summarized inFigs. 3 and4 (Tables S3–S6) followed by documentation of various sensitivity tests referred to above (Tables S7–S9).

Table S2.

Summary statistics

Variable namesNMeanSDMinimumMaximum
Ethnic civil conflict incidence6,7070.110.310.001.00
Ethnic civil conflict onset6,0740.020.140.001.00
Incidence (spatial overlay)6,7070.210.410.001.00
Onset (spatial overlay)5,4680.050.230.001.00
SPEI drought6,7070.170.130.000.80
SPEI cumulative drought6,7071.331.720.005.00
Negative rainfall anomaly6,7070.140.220.002.06
Cumulative drought rainfall6,7070.941.320.005.00
Agricultural dependence6,7070.150.160.000.80
Economic development5,4518.802.510.0015.85
Exclusion6,7070.500.500.001.00
Population (ln)6,61314.131.808.5819.87
Country GDPpc (ln)t−16,6637.490.934.6810.45
IMRt−16,62067.3331.896.90161.80
Decay conflict6,7070.130.320.001.00
Other groups in conflict6,7070.070.250.001.00
Decay conflict (spatial overlay)6,7070.280.410.001.00
Other conflicts/spatial lag (spatial overlay)6,7070.410.490.001.00
Observations6,707

Table S3.

Direct relationship

graphic file with name pnas.1607542113st03.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). The models test the direct association between our preferred drought measure (models 1–4) as well as the cumulative drought indicator (models 5–8) and conflict outbreak and conflict incidence for the complete Asia–Africa sample (white) as well as for a subsample of ethnic groups in low-development countries (IMR > median; pink).

Table S6.

Exclusion

graphic file with name pnas.1607542113st06.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). The table estimates the effect of drought (models 25–28) and cumulative drought (models 29–32) conditional on the third group-specific vulnerability factor, ethnopolitical exclusion, for the complete Asia–Africa sample (white) as well as for a limited sample of ethnic groups in low-development countries (IMR > median; pink).

Table S4.

Agricultural dependence

graphic file with name pnas.1607542113st04.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). The table estimates the effect of drought (models 9–12) and cumulative drought (models 13–16) conditional on the group’s extent of agricultural dependence for the complete Asia–Africa sample (white) as well as for a subsample of ethnic groups in low-development countries (IMR > median; pink).

Table S5.

Economic development

graphic file with name pnas.1607542113st05.jpg

Two-level mixed effects logistic regression estimates with SEs in parentheses (*P < 0.10; **P < 0.05; ***P < 0.01). The table estimates the effect of drought (models 17–20) and cumulative drought (models 21–24) conditional on the group’s level of economic development (measured using night light emissions) for the complete Asia–Africa sample (white) as well as for a subsample of ethnic groups in low-development countries (IMR > median; pink).

Acknowledgments

We thank Henning Tamm, Idean Salehyan, and participants in the meetings of the European Network of Conflict Research (Barcelona, Spain, October of 2015), the Second PRIO Workshop on Climate Anomalies and Violent Environments (October of 2015), the Workshop on Urban Insecurity and Civil Conflict (Oxford, United Kingdom, November of 2015), and the Annual Meeting of the International Studies Association (Atlanta, GA, March of 2016) for helpful comments on earlier versions of the manuscript. This research was financially supported by Research Council of Norway Grants 240315/F10 and 217995/V10, the Swedish International Development Cooperation Agency, and European Research Council Grant 648291.

Footnotes

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The replication data reported in this paper are available from Peace Research Institute Oslo’s data repository,https://www.prio.org/Data/Replication-Data, as well as the Department of Peace and Conflict Research, Uppsala University,www.pcr.uu.se/data.

This article contains supporting information online atwww.pnas.org/lookup/suppl/doi:10.1073/pnas.1607542113/-/DCSupplemental.

References

  • 1.NATO Parliamentary Assembly 2015 Resolution 427 on Climate Change and International Security. Available atwww.nato-pa.int/default.asp?CAT2=3924&CAT1=16&CAT0=576&SHORTCUT=3742&SEARCHWORDS=Resolution,427. Accessed May 5, 2016.
  • 2.US Government 2015 US National Security Strategy. Available athttps://www.whitehouse.gov/sites/default/files/docs/2015_national_security_strategy.pdf. Accessed May 5, 2016.
  • 3.Kelley CP, Mohtadi S, Cane MA, Seager R, Kushnir Y. Climate change in the Fertile Crescent and implications of the recent Syrian drought. Proc Natl Acad Sci USA. 2015;112(11):3241–3246. doi: 10.1073/pnas.1421533112. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Maystadt J-F, Calderone M, You L. Local warming and violent conflict in North and South Sudan. J Econ Geogr. 2015;15(3):649–671. [Google Scholar]
  • 5.Adger WN, et al. Human security. In: Field CB, et al., editors. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel of Climate Change. Cambridge Univ Press; Cambridge, UK: 2014. pp. 755–791. [Google Scholar]
  • 6.Salehyan I. Climate change and conflict: Making sense of disparate findings. Polit Geogr. 2014;43:1–5. [Google Scholar]
  • 7.Buhaug H, et al. One effect to rule them all? A comment on climate and conflict. Clim Change. 2014;127(3-4):391–397. [Google Scholar]
  • 8.Homer-Dixon TF. Environmental scarcities and violent conflict: Evidence from cases. Int Secur. 1994;19(1):5–40. [Google Scholar]
  • 9.Barnett J, Adger WN. Climate change, human security and violent conflict. Polit Geogr. 2007;26(6):639–655. [Google Scholar]
  • 10.Morton JF. The impact of climate change on smallholder and subsistence agriculture. Proc Natl Acad Sci USA. 2007;104(50):19680–19685. doi: 10.1073/pnas.0701855104. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Olsson L, et al. Livelihoods and poverty. In: Field CB, et al., editors. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel of Climate Change. Cambridge Univ Press; Cambridge, UK: 2014. pp. 793–832. [Google Scholar]
  • 12.Busby JW, Smith TG, White KL, Strange SM. Climate change and insecurity: Mapping vulnerability in Africa. Int Secur. 2013;37(4):132–172. [Google Scholar]
  • 13.Field C, Barros V, Stocker T, Dahe Q, editors. Managing the Risks of Extreme Events and Disasters to Advance Climate Change Adaptation. Cambridge Univ Press; Cambridge, UK: 2012. [Google Scholar]
  • 14.Oliver-Smith A. Antropological research on hazards and disaster. Annu Rev Anthropol. 1996;25(1):303–328. [Google Scholar]
  • 15.Raleigh C. Political marginalization, climate change, and conflict in African Sahel states. Int Stud Rev. 2010;12(1):69–86. [Google Scholar]
  • 16.Gurr TR. Why Men Rebel. Princeton Univ Press; Princeton: 1970. [Google Scholar]
  • 17.Cederman L-E, Wimmer A, Min B. Why do ethnic groups rebel? New data and analysis. World Polit. 2010;62(1):87–119. [Google Scholar]
  • 18.Horowitz DL. Ethnic Groups in Conflict. Univ of California Press; Berkeley, CA: 1985. [Google Scholar]
  • 19.Fearon JD. Ethnic mobilization and ethnic violence. In: Weingast BR, Wittman DA, editors. The Oxford Handbook of Political Economy. Oxford Univ Press; New York: 2006. pp. 852–868. [Google Scholar]
  • 20.Sambanis N, Shayo M. Social identification and ethnic conflict. Am Polit Sci Rev. 2013;107(02):294–325. [Google Scholar]
  • 21.Denny EK, Walter BF. Ethnicity and civil war. J Peace Res. 2014;51(2):199–212. [Google Scholar]
  • 22.Schleussner C-F, Donges JF, Donner RV, Schellnhuber HJ. Armed-conflict risks enhanced by climate-related disasters in ethnically fractionalized countries. Proc Natl Acad Sci USA. 2016;113(33):9216–9221. doi: 10.1073/pnas.1601611113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Stewart F. Horizontal Inequalities and Conflict. Understanding Group Violence in Multi-Ethnic Societies. Palgrave Macmillan; New York: 2008. [Google Scholar]
  • 24.Burke MB, Miguel E, Satyanath S, Dykema JA, Lobell DB. Warming increases the risk of civil war in Africa. Proc Natl Acad Sci USA. 2009;106(49):20670–20674. doi: 10.1073/pnas.0907998106. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Buhaug H. Climate not to blame for African civil wars. Proc Natl Acad Sci USA. 2010;107(38):16477–16482. doi: 10.1073/pnas.1005739107. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Hsiang SM, Burke M, Miguel E. Quantifying the influence of climate on human conflict. Science. 2013;341(6151):1235367. doi: 10.1126/science.1235367. [DOI] [PubMed] [Google Scholar]
  • 27.O’Loughlin J, Linke AM, Witmer FDW. Effects of temperature and precipitation variability on the risk of violence in sub-Saharan Africa, 1980-2012. Proc Natl Acad Sci USA. 2014;111(47):16712–16717. doi: 10.1073/pnas.1411899111. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Vogt M, et al. Integrating data on ethnicity, geography, and conflict: The ethnic power relations data set family. J Conflict Resolut. 2015;59(7):1327–1342. [Google Scholar]
  • 29.Sundberg R, Melander E. Introducing the UCDP Georeferenced Event Dataset. J Peace Res. 2013;50(4):523–532. [Google Scholar]
  • 30.Wucherpfennig J, Metternich NW, Cederman L-E, Gleditsch KS. Ethnicity, the state, and the duration of civil war. World Polit. 2012;64(01):79–115. [Google Scholar]
  • 31.You L, et al. 2014 Spatial Production Allocation Model (SPAM) 2005 v2.0. Available atmapspam.info. Accessed June 29, 2015.
  • 32.Beguería S, Vicente-Serrano SM, Reig F, Latorre B. Standardized precipitation evapotranspiration index (SPEI) revisited: Parameter fitting, evapotranspiration models, tools, datasets and drought monitoring. Int J Climatol. 2014;34(10):3001–3023. [Google Scholar]
  • 33.Portmann FT, Siebert S, Döll P. MIRCA2000—Global monthly irrigated and rainfed crop areas around the year 2000: A new high-resolution data set for agricultural and hydrological modeling. Global Biogeochem Cycles. 2010;24(1):1–24. [Google Scholar]
  • 34.Müller C, Cramer W, Hare WL, Lotze-Campen H. Climate change risks for African agriculture. Proc Natl Acad Sci USA. 2011;108(11):4313–4315. doi: 10.1073/pnas.1015078108. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Springmann M, et al. Global and regional health effects of future food production under climate change: A modelling study. Lancet. 2016;387(10031):1937–1946. doi: 10.1016/S0140-6736(15)01156-3. [DOI] [PubMed] [Google Scholar]
  • 36.Knox J, et al. Climate change impacts on crop productivity in Africa and South Asia. Environ Res Lett. 2012;7(3):034032. [Google Scholar]
  • 37.Abouharb MR, Kimball AL. A new dataset on infant mortality rates, 1816—2002. J Peace Res. 2007;44(6):743–754. [Google Scholar]
  • 38.OECD . States of Fragility 2015: Meeting Post-2015 Ambitions. OECD Publishing; Paris: 2015. [Google Scholar]
  • 39.Schmidhuber J, Tubiello FN. Global food security under climate change. Proc Natl Acad Sci USA. 2007;104(50):19703–19708. doi: 10.1073/pnas.0701976104. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Bueno de Mesquita B, Smith A, Siverson RM, Morrow JD. The Logic of Political Survival. MIT Press; Cambridge, MA: 2005. [Google Scholar]
  • 41.Justino P. Poverty and violent conflict: A micro-level perspective on the causes and duration of warfare. J Peace Res. 2009;46(3):315–333. [Google Scholar]
  • 42.Oppenheimer M, et al. Emergent risks and key vulnerabilities. In: Field CB, et al., editors. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part A: Global and Sectoral Aspects. Contribution of Working Group II to the Fifth Assessment Report of the Intergovernmental Panel of Climate Change. Cambridge Univ Press; Cambridge, UK: 2014. pp. 1039–1099. [Google Scholar]
  • 43.Wischnath G, Buhaug H. Rice or riots: On food production and conflict severity across India. Polit Geogr. 2014;43(0):6–15. [Google Scholar]
  • 44.Eastin J. Fuel to the fire: Natural disasters and the duration of civil conflict. Int Interact. 2016;42(2):322–349. [Google Scholar]
  • 45.De Waal A. 1991 Evil Days. 30 Years of War and Famine in Ethiopia. An Africa Watch Report. Available athttps://www.hrw.org/sites/default/files/reports/Ethiopia919.pdf. Accessed May 5, 2016.
  • 46.Parenti C. Tropic of Chaos: Climate Change and the New Geography of Violence. Nations Books; New York: 2011. [Google Scholar]
  • 47.Gates S, Hegre H, Nygård HM, Strand H. Development consequences of armed conflict. World Dev. 2012;40(9):1713–1722. [Google Scholar]
  • 48.World Bank 2011 World Development Report. Available atgo.worldbank.org/1BOIJMD8H0. Accessed May 5, 2016.
  • 49.Wucherpfennig J, Weidmann NB, Girardin L, Cederman L-E, Wimmer A. Politically relevant ethnic groups across space and time: Introducing the GeoEPR Dataset. Confl Manage Peace Sci. 2011;28(5):423–437. [Google Scholar]
  • 50.Tollefsen AF, Strand H, Buhaug H. PRIO-GRID: A unified spatial data structure. J Peace Res. 2012;49(2):363–374. [Google Scholar]
  • 51.Sacks WJ, Deryng D, Foley JA, Ramankutty N. Crop planting dates: An analysis of global patterns. Glob Ecol Biogeogr. 2010;19(5):607–620. [Google Scholar]
  • 52.Vicente-Serrano SM, Beguería S, López-Moreno JI. A multiscalar drought index sensitive to global warming: The standardized precipitation evapotranspiration index. J Clim. 2009;23(7):1696–1718. [Google Scholar]
  • 53.Girardin L, Hunziker P, Cederman L-E, Bormann N-C, Vogt M. 2015 Growup - Geographical Research on War, Unified Platform. Available athttps://growup.ethz.ch. Accessed May 5, 2016.
  • 54.CIESIN CIAT 2005 Gridded Population of the World (GPW), v4. Available atsedac.ciesin.columbia.edu/data/collection/gpw-v4. Accessed May 5, 2016.
  • 55.WDI . World Development Indicators 2014. World Bank; Washington, DC: 2014. [Google Scholar]
  • 56.Lobell DB, Gourdji SM. The influence of climate change on global crop productivity. Plant Physiol. 2012;160(4):1686–1697. doi: 10.1104/pp.112.208298. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 57.Landis ST. Temperature seasonality and violent conflict: The inconsistencies of a warming planet. J Peace Res. 2014;51(5):603–618. [Google Scholar]
  • 58.Pettersson T, Wallensteen P. Armed conflicts, 1946–2014. J Peace Res. 2015;52(4):536–550. [Google Scholar]
  • 59.Gleditsch NP, Wallensteen P, Eriksson M, Sollenberg M, Strand H. Armed conflict 1946-2001: A new dataset. J Peace Res. 2002;39(5):615–637. [Google Scholar]
  • 60.Bormann N-C, et al. 2014 ACD2EPR Version 2014 Codebook. Available atwww.icr.ethz.ch/data/acd2epr/ACD2EPR-2014_Codebook.pdf. Accessed May 5, 2016.
  • 61.Selby J. Positivist climate conflict research: A critique. Geopolitics. 2014;19(4):829–856. [Google Scholar]
  • 62.You L, Wood S, Wood-Sichra U, Wu W. Generating global crop distribution maps: From census to grid. Agric Syst. 2014;127(0):53–60. [Google Scholar]
  • 63.von Uexkull N. Sustained drought, vulnerability and civil conflict in Sub-Saharan Africa. Polit Geogr. 2014;43(0):16–26. [Google Scholar]
  • 64.Henderson JV, Storeygard A, Weil DN. A bright idea for measuring economic growth. Am Econ Rev. 2011;101(3):194–199. doi: 10.1257/aer.101.3.194. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65.Cederman L-E, Gleditsch KS, Buhaug H. Inequalities, Grievances, and Civil War. Cambridge Univ Press; Cambridge, UK: 2013. [Google Scholar]
  • 66.Gelman A, Hill J. Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge Univ Press; New York: 2007. [Google Scholar]
  • 67.Schneider U, et al. 2015. GPCC Full Data Reanalysis Version 7.0 at 0.5: Monthly Land- Surface Precipitation from Rain-Gauges Built on GTS-Based and Historic Data (Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory, Boulder, CO)

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy ofNational Academy of Sciences

ACTIONS

RESOURCES

[8]ページ先頭
©2009-2025 Movatter.jp