Conditional Relationships Between Drought and Civil Conflict in Sub-Saharan Africa

Conditional Relationships Between Drought and Civil Conflict in Sub-Saharan Africa Abstract Much of the literature on climate change adaptation claims the destabilizing consequences of environmental crises are mitigated by sociopolitical conditions that influence a state's susceptibility to scarcity-induced violence. However, few cross-national studies provide evidence of conditional scarcity-conflict relationships. This analysis of drought severity and civil conflict onset in sub-Saharan Africa (1962–2006) uncovers three sociopolitical conditions that influence the link between environmental scarcity and civil conflict: social vulnerability, state capacity, and unequal distribution of resources. Surprisingly, we find drought does not exacerbate the high risk of conflict in the vulnerable, incapable, and unequal states thought to be especially susceptible to increased scarcity. Instead, drought negates the peace-favoring attributes of stable states with less vulnerable populations. During severe drought, states with sociopolitical conditions that would otherwise favor peace are no less likely to suffer conflict than states with sociopolitical conditions that would otherwise increase the risk of violence. These findings, which are robust across several measures of these sociopolitical concepts, suggest environmental scarcity is most likely to increase the risk of conflict where populations have more to lose relative to periods with more favorable weather. The 2010 Conference of Parties of the United Nations Framework Convention on Climate Change (UNFCCC) highlighted an important transition in international environmental policy. Diplomats made little progress toward the organization’s original goal of climate change prevention. Instead, the conference resulted in an ambitious agenda focused on climate change adaptation. Fearing that adverse environmental change may now be inevitable, members renewed earlier pledges to help developing states craft adaptation plans and created a new Green Climate Fund to finance these efforts (UNFCCC 2012a). The UNFCCC estimates fully funding these plans will require contributions in excess of $100 billion UNFCCC 2012b), though this infeasible sum is nearly 50 times greater than the 2013 budget of the entire United Nations (UN 2011). This costly reaction to climate change is motivated in part by the popular notion that environmental catalysts prompt violent conflict where vulnerable states lack the capacity to adequately adapt.1 Events such as the genocide in Darfur, food riots in East Africa, and the 2010–2011 Arab Spring are sometimes offered as anecdotal support for this claim.2 But despite the popularity of the scarcity-conflict hypothesis among journalists, policymakers, and some conflict scholars, research has yet to generate consensus about a generalizable relationship between environmental catalysts and conflict (Gleditsch 2012; Salehyan 2008).3 Scholarly efforts to examine this relationship are impeded by an important disjuncture in the literature. Theories of environmental conflict, which are often drawn from case comparisons and small-n studies, emphasize several sociopolitical conditions that influence the likelihood with which scarcity-inducing catalysts might result in social conflict (i.e., Homer-Dixon 1991; Kahl 2006). This work stresses the mitigating effects of state capacity, social vulnerability to decreased resources, and barriers to efficient and equitable distribution of emergency resources. But very few cross-national studies are sensitive to these intervening sociopolitical conditions. Failure to account for conditional relationships can bias analyses of direct scarcity-conflict effects toward insignificance, leading scholars to conclude that no relationship exists when, in fact, scarcity-conflict linkages are only obscured by the conflation of vulnerable and invulnerable observations in the sample. The search for a universal effect may overlook salient conditional mechanisms. This study aims to advance our understanding of environmental drivers of political violence by testing for conditional relationships that link civil conflict onset to drought severity in 39 sub-Saharan states (1962–2006). It contributes to a nascent sub-literature that examines conditional causal mechanisms in a cross-national sample (Buhaug 2010; Couttenier and Soubeyran 2013; Fjelde and von Uexkull 2012; O’Loughlin et al. 2012; Raleigh and Urdal 2007; Slettebak 2012; Theisen, Holtermann, and Buhaug 2011–2012). We limit the scope of our project to African civil conflict because other types of violence (cattle raiding, riots, etc.) may have fundamentally different causes (Fjelde and von Uexkull 2012; Meier, Bond, and Bond 2007; Theisen 2012) and civil conflict poses an especially salient and consequential challenge to African political development. Further, most extant work on scarcity-driven conflict examines sub-Saharan Africa (Buhaug 2010; Couttenier and Soubeyran 2013; Hendrix and Glaser 2007; Hendrix and Salehyan 2012; Theisen, Holtermann, and Buhaug 2011–2012), and adopting this scope allows for more direct comparison of our results to previous research. We also follow convention by using climatological drought data to measure variation in resource scarcity. Today, more than one in three sub-Saharan Africans inhabits an area that is prone to severe drought and the United Nations estimates an additional 250–300 million Africans will be adversely affected by drought by 2020 (UNFCCC 2007, 5). Most sub-Saharan states are too poor to adequately respond to the forecasted increase in severe drought, and African economies are particularly susceptible to water scarcity (Müller et al. 2011). Sub-Saharan economies are more dependent upon agricultural production than those of any other region, and African agriculture is uniquely vulnerable because only a very small fraction of the region’s arable land is irrigated (World Bank 2013). Most of sub-Saharan Africa is expected to suffer more drought as a result of global climate change (UNFCCC 2007), and sub-Saharan Africa will also be the largest recipient of international adaptation aid if the UNFCCC National Adaptation Programs of Action are funded completely (UNFCCC 2012b). Any relationship between scarcity-inducing environmental change and large-scale violent conflict could be most apparent when drought strikes sub-Saharan Africa. Our analysis finds no evidence of drought increasing the risk of conflict wherever it occurs, and this corroborates previous research that reports no direct and universal relationship between drought and political violence in sub-Saharan Africa (Buhaug 2010; Couttenier and Soubeyran 2013; O’Loughlin et al. 2012; Theisen 2008). However, we highlight several conditions under which drought becomes a significant predictor of African civil conflict. Surprisingly, severe drought has no significant effect on the risk of civil conflict onset in the states with sociopolitical conditions that favor conflict. These states face a relatively high risk of political violence regardless of recent environmental conditions, and any additional risk posed by severe drought is small and statistically insignificant. To the contrary, we find severe drought to be a risk equalizer. States with greater food security, less ethnopolitical exclusion, democratic governance, superior public health, and longer histories of political stability are undermined by severe drought, and the magnitude of the effect is great enough to offset the stabilizing influences of these sociopolitical conditions. During severe drought, states with sociopolitical attributes that would normally promote peace are no less likely to suffer conflict than states with the weakest governments and most vulnerable populations. This finding, which is robust across several measures of these sociopolitical conditions, contradicts the argument that fragile states with vulnerable populations are at the greatest risk of scarcity-induced political violence. Instead, these results raise important questions about why environmental scarcity might destabilize relatively secure governments and populations. We discuss one plausible explanation that should be explored in future research. Work in psychology and behavioral economics on the “endowment effect” finds people are averse to loss to the point that absolute conditions may be less important than how those conditions compare to the perceived status quo (Kahneman and Tversky 1979; Thaler 1980). This logic was originally used to explain why sellers and buyers sometimes fail to agree upon an appropriate price (Brenner et al. 2007; Carmon and Ariely 2000; Kahneman, Knetsch, and Thaler 1991). Tendencies toward loss aversion cause those who are endowed with something (sellers) to place much greater value on what they possess than those who do not (buyers). This argument has been extended to conflict research to explain how adverse changes from an ex ante reference point (the status quo) can encourage risk-taking and conflict initiation (Levy 1992; Mercer 2005). In this context, endowment theory would expect the hardships caused by scarcity-inducing weather to elicit stronger reactions where the status quo is more comfortable. Drought-induced grievances will be less severe where the population is accustomed to an ineffective state and adverse sociopolitical conditions. Instead, frustration will be higher where a more favorable status quo has conditioned people to expect a higher level of welfare and more effectual government. Drought threatens a more adverse deviation from the ex ante reference point in these contexts, causing the risk of a drought-motivated conflict to be higher than it would be were the status quo less appealing. Literature Review Conditional theories of scarcity and conflict are often rooted in the work of Thomas Homer-Dixon (1991, 1994, 1999, 2000). Using case studies of conflicts that are typically traced to environmental causes, Homer-Dixon concludes conflict is more likely to occur where states are unable to effectively adapt to resource shortages. This thesis is far from deterministic, and scarcity-induced conflict is by no means inevitable. Rather, the argument is that scarcity-inducing events have the greatest destabilizing effects where governments cannot effectively insulate their populations from adverse change. Since the early 1990s, this work has advanced a broader multidisciplinary research program on the sociopolitical consequences of resource scarcity (Adger 1999; Brooks, Adger, and Kelly 2005; O’Brien et al. 2004; Tol and Yohe 2007) and has supported the international community’s recent commitment to international adaptation aid programs for states that are perceived to be especially susceptible. This literature has attracted pointed criticism for advancing complex causal mechanisms, biased case selection, and “theoretical myopia” (Gleditsch 1998, 390–92; 2012, 6–7; Theisen 2008, 813). These concerns give rise to a growing research program that uses cross-national statistics to seek more generalizable relationships, but, with very few exceptions, this work fails to account for conditional mechanisms. When indicators of scarcity are added to conflict models, researchers are testing for universal relationships that are consistent across all observations in the sample. This approach assumes scarcity will have a uniform effect wherever it occurs, and this insensitivity to conditional effects is widely acknowledged in reviews of the quantitative literature (see Barnett and Adger 2007, 644; Gleditsch 1998, 389–90; Salehyan 2008, 321–22). In fact, Gleditsch identifies this shortcoming as the research program’s “greatest weakness” (1998, 389; 2012, 6). A small group of large-n studies links increased abundance—rather than scarcity—to some types of violent conflict (Binningsbø, de Soysa, and Gleditsch 2007; de Soysa 2002; Meier et al. 2007; Theisen 2012). Many papers fail to find any connection between climatic extremes and conflict (Bergholt and Lujala 2012; Buhaug 2010; Theisen, Holtermann, and Buhaug 2011–2012), and others report curvilinear and non-monotonic relationships (Hendrix and Salehyan 2012; O’Loughlin et al. 2012; Raleigh and Kniveton 2012). Some conclude scarcity-inducing weather increases the risk of conflict (Burke et al. 2009; Hauge and Ellingsen 1998; Hsiang, Meng, and Cane 2011; Miguel, Satyanath, and Sergenti 2004), but many of these studies have been scrutinized and dismissed by other researchers over substantial methodological disagreements.4 Drawing inferences from the literature is further complicated by fractionalization into sub-literatures exploring various measures of scarcity (rainfall, temperature, deviation from local historical norms), units of analysis (state, sub-state administrative divisions, or geographic grid cells), and types of civil violence (civil conflict, inter-ethnic violence, communal cattle raiding, nonviolent protests).5 Some recent studies test conditional scarcity-conflict relationships, with most attention being dedicated to sub-state variation in ethnic exclusion and population density. O’Loughlin et al. (2012), Buhaug (2010), and Theisen, Holtermann, and Buhaug (2011–2012) use a sub-state grid cell unit of analysis and report no relationships between local scarcity and local conflict, regardless of the presence of marginalized ethnic groups, poverty, or population pressure within a specific grid cell. Fjelde and von Uexkull (2012), studying a different type of conflict (communal conflict) and unit of analysis (sub-national administrative district), find drought conditions very modestly increase the risk of small-scale communal conflict in districts containing at least one politically marginalized ethnic group. However, this significant result is a notable aberration in a group of studies that reports many non-findings. The absence of significant effects in even the most conflict-prone areas has led several scholars to forcefully reject the possibility that political violence can be “blamed” on drought (Buhaug 2010; Slettebak 2012). Local conditions certainly affect adaptive capacity and social vulnerability, but the literatures on these concepts also stress important state-level determinants (Adger 1999; Brooks, Adger, and Kelly 2005). The ethnic inclusivity of a national government is both empirically and theoretically distinct from the demographics of a sub-state district. Ethnic political dominance can create doubts about the equitable distribution of emergency resources, even in sub-state regions that are relatively homogeneous. The resilience of a state’s institutions is better captured by the statewide presence or absence of destabilizing violence or political change, rather than an indicator of recent conflict within a small, specific grid cell. A state’s efforts to aid villages in peaceful regions can be hindered by insurgencies that demand resources and attention far away, on a state’s loosely controlled frontiers. Further, the movement of effected populations could bring conflict to grid cells where scarcity is less severe. This is especially true in sub-Saharan Africa, where a significant share of the population practices pastoralism and many scarcity-inducing catalysts result in voluntary migration and social conflict in migrant-receiving areas (Barrios, Bertinelli, and Strobi 2006; Reuveny 2007; Salehyan and Gleditsch 2006). For these reasons, state-level determinants, such as state stability, statewide scarcity, and the characteristics of state political institutions, could affect the drought-conflict relationship as much as any local conditions. Our review of the literature uncovered only two published papers that focus on conditional scarcity-civil conflict relationships at the state level of analysis: Slettebak’s (2012) global study of climate-related disasters (including drought) and Couttenier and Soubeyran’s (2013) examination of drought in sub-Saharan Africa. Both papers report statistically significant conditional effects. Slettebak (2012) interacts the state population with a crisis indicator drawn from the Center for Research on the Epidemiology of Disasters (CRED 2013). He concludes, surprisingly, that population increases the pacifying effect of a disaster. States suffering disasters are less likely to suffer civil conflict onset, and the magnitude of this effect grows with state population. Couttenier and Soubeyran (2013) examine ethnic fractionalization, terrain roughness, and democratic institutions, finding the relationship between drought and civil conflict to be negative in homogeneous and non-mountainous states, but positive in diverse states with rugged terrain. Their tests for a conditional relationship between democracy, drought, and civil conflict offer mixed results. Both papers have important methodological shortcomings that drive these reported findings. Slettebak (2012) operationalizes disasters with a measure based on media reports of populations effected by weather events, but unequal media coverage causes disasters to be disproportionately overlooked in lesser-developed regions (Gall, Borden, and Cutter 2009). The United States suffers a qualifying disaster in every year in Slettebak’s sample (1950–2008), while less newsworthy but no less severe events in Africa are not represented in the data. There is also substantial temporal bias in the data; more recent years see more than three times as many disaster events as years before 1980. The oversampling of recent “disasters” in large Western states could explain the finding that disasters decrease the risk of conflict.6 The analysis offered by Couttenier and Soubeyran (2013) utilizes superior drought data, but the results are produced by some unconventional modeling decisions. First, the paper estimates civil conflict onset, which is dichotomous, with linear ordinary least squares regression (Couttenier and Soubeyran 2013, 7). This method is not typically used to estimate models with dichotomous dependent variables because it can generate predicted probabilities that lie outside the unit interval [0,1]. This practice is more common in economics than in political science (i.e., Miguel, Satyanath, and Sergenti 2004), but Beck (2011) has shown ordinary least squares regression to be unsatisfactory relative to logistic regression when independent variables are not distributed normally. Key independent variables such as the measures for drought (leptokurtic), ethnic fractionalization (negatively skewed), and terrain roughness (positively skewed) violate this important condition.7 Additionally, the strongest reported conditional effects—those related to ethnic fractionalization and terrain roughness—are obtained from models that do not include both of the independent variables that comprise the interaction term (see Couttenier and Soubeyran 2013, Table 7, Models 1–4). Because the interaction models omit one of the interacted variables, they may produce biased results (Brambor, Clark, and Golder 2006; Plümper and Troeger 2007). The models also include fixed effects for state and year, but no other control variables. When standard controls for economic performance and political stability are added to the models, the results diverge from those reported in the paper. Theory and Hypotheses Conditional scarcity-conflict arguments emerging from the environmental security literature generally link scarcity to civil conflict via one or more of three mechanisms. As these arguments have yet to be explicitly tested on a large cross-national sample, we proceed by delineating these mechanisms, identifying suitable measures, and proposing testable hypotheses that reflect these arguments. First, scarcity-inducing catalysts such as drought are expected to be more likely to motivate conflict where the population is more vulnerable to adverse change. Conditions that increase social vulnerability, such as poverty, poor public goods provision, and inadequate access to food and clean water, should reduce a population’s ability to tolerate further resource scarcity and therefore increase the risk of instability and conflict (Brooks, Adger, and Kelly 2005; Homer-Dixon 1994; Tol and Yohe 2007). With inadequate means, individuals have less capacity to generate and save surpluses that can reduce vulnerability during times of scarcity. As a result, they make greater demands of their governments during crises and are more likely to migrate to refugee camps and become temporarily dependent upon food aid (Gleditsch, Nordas, and Salehyan 2007; Reuveny 2007). The desperation that results from high social vulnerability can facilitate rebel recruitment by reducing the opportunity costs forgone by those who join anti-government movements. To test this social vulnerability mechanism, we leverage data on two indicators of well-being. Environmental scarcity is closely related to food insecurity, so our first measure, Food Supply, is the Food and Agriculture Organization’s best estimate of per capita daily caloric consumption (FAO 2012). Food Supply is recognized as an ideal indicator of social vulnerability because it is less prone to within-state inequality than measures of income, education, and other public goods (Conrad 2011). Wealthier people tend to not eat many, many times more calories than the poor. Instead, they replace cheaper calories (grains) with high-cost options such as meat and imported delicacies. In this sample, Food Supply averages 2,132 calories per person per day, with extremes at 1,439 (Burkina Faso 1962) and 2,978 (South Africa 2006). We lag this variable by one year to address the bias that could be caused by endogeneity between recent drought conditions and Food Supply. We also test this mechanism with Infant Mortality Rate, a widely adopted and comprehensive measure of well-being that is influenced by nutrition, access to health care, female education, and the availability of sanitary food and water. We use data from the World Bank (2013) and Abouharb and Kimball (2007) to create a time-variant measure that equals the number of children per 1,000 live births who perish before reaching their second birthday. Southern countries like Botswana, Zimbabwe, and South Africa have experienced periods with relatively low rates of less than 50 per 1,000, while Infant Mortality Rate peaked just below 300 per 1,000 in Mozambique during its long civil war. Across roughly 1,600 country-year observations, the mean Infant Mortality Rate was 118 deaths per 1,000 live births. Social Vulnerability Hypotheses H1a: Drought most increases the risk of conflict where caloric consumption is low. H1b: Drought most increases the risk of conflict where infant mortality is high. The second conditional mechanism emphasizes the politics of resource distribution during times of crisis. Scarcity can be especially dangerous when access to remaining resources is perceived to be unequal (Homer-Dixon 1994), and many case studies find these perceptions to be more prevalent when diverse populations do not have comparable representation in the government. Kahl (1998, 2006) argues that preexisting ethnopolitical tensions were exacerbated during environmental crises in Kenya. Similarly, Raleigh (2010) studies the Sahel and stresses that diversity and drought are most likely to result in conflict where ethnic groups are also politically marginalized. Diversity increased perceptions of government favoritism and may have fueled conflict during Ethiopian resource shortages Martin (2005) and the Tuareg rebellion in Mali (Benjaminsen 2008). When scarcity causes migration, this migration is most likely to result in conflict between the migrant and host communities when these groups lack ethnic similarities (Gleditsch 2007; Reuveny 2007; Salehyan and Gleditsch 2006). Conversely, homogeneity can facilitate peace during periods of increased scarcity. Bogale and Korf (2007) find that regional homogeneity in some parts of Ethiopia helped communities arrange local resource-sharing agreements to reduce the effects of acute shortages. Shared identity may also decrease perceptions of relative deprivation (Bohle, Downing, and Watts 1994; Reardon and Taylor 1996) and help afflicted populations foster unity and collective identity during a community crisis (see Slettebak 2012). We hypothesize that scarcity will be most likely to result in conflict when it exacerbates pre-existing feelings of relative deprivation among salient ethnic groups with unequal access to political power. We test this argument with a measure of ethnic diversity that accounts for whether a group also suffers political marginalization. Ethnic Exclusion represents the share of a state’s population that belongs to ethnic groups that are excluded from political power. This information is compiled in the Ethnic Power Relations data set (Cederman, Wimmer, and Min 2010). The average of this time-variant variable is 23 percent, but approximately 43 percent of state-years have no exclusion whatsoever. Some of the highest observed values belong to 1970s Liberia (98 percent), 1960s–1970s Zimbabwe/Rhodesia (97 percent), and post-1994 Rwanda (84 percent). We expect the drought-conflict relationship to be weakly positive or insignificant when Ethnic Exclusion is low, but the magnitude of the correlation should increase with this variable. Exclusive forms of government promote favoritism and patronage and may increase perceptions of unequal resource allocation even where exclusion is not exacerbated by ethnic diversity. The leaders of democracies must attract a broader coalition to remain in power (Bueno de Mesquita et al. 2003), so populations living under democratic governments should expect state resources to be allocated more equitably to a greater share of the population. Because non-democratic leaders do not face electoral constraints and stay in power by pleasing a narrower coalition, leaders have substantially less incentive to allocate resources to a large share of the population during periods of extreme resource scarcity. We test this mechanism with Polity Score, which is the literature’s most frequently used measure of democratization (Marshall and Jaggers 2002). This 21-value ordinal scale (–10 to 10) increases with a state’s level of democratization. We use the Vreeland (2008) “X Polity” version of the data, which is optimized for use in time-series analysis. Sub-Saharan Africa trended slightly non-democratic in the overall sample (mean of –1.02), though Africa is much more democratic since 2000 (mean of +1.33) and includes several states (e.g., South Africa, Lesotho) that experienced periods of full-fledged democracy (Polity Score > 6). Equitable Distribution Hypotheses H2a: Drought most increases the risk of conflict where ethnic exclusion in high. H2b: Drought most increases the risk of conflict where democratization is low. A third mechanism emphasizes state capacity, defined as a government’s ability to mitigate the consequences of adverse environmental change. Whereas social vulnerability describes a population’s exposure to scarcity, capacity refers to the resources, expertise, and institutional stability that allow states to alleviate the impact of increased scarcity (Bohle, Downing, and Watts 1994; Busby et al. 2013; Reardon and Taylor 1996). Governments lacking adaptive capacity are less likely to possess the emergency resources they need to provide a sufficient response to increased scarcity, and citizens are less likely to believe the government will meet their needs (Homer-Dixon 1994; Reuveny 2007). Local norms of resource-sharing and social support networks can compensate for government shortcomings to some extent, but state support is often needed to provide security and supplement any local efforts with imported food and resources.8 Perceptions of government incompetence or indifference during crisis can then motivate rebel recruitment. There is a large literature on measuring adaptive capacity, and the most comprehensive juxtaposition of proposed measures finds political stability to be among the most important determinants of a government’s ability to act during environmental crises (Brooks, Adger, and Kelly 2005). During crises, stable governments are better able to provide security, manage emergency resource distribution, and adapt to internal migration (Reuveny 2007). Severe drought should be more likely to motivate civil conflict where states have been stable for a shorter period of time. Stability decreases the likelihood of civil conflict (Fearon and Laitin 2003; Hegre et al. 2001), increases the opportunity costs of rebellion (Collier and Hoeffler 2004), and allows for institutional consolidation (Huntington 1968). We test this mechanism with two measures of stability. Peace Years counts the number of years since either independence or the last year of civil conflict incidence (≥ 25 deaths) as recorded by the Uppsala Conflict Data Program Armed Conflict Dataset (Themnér and Wallensteen 2012). The average observation had 12 years of peace, while Tanzania reached nearly 50 years without suffering a qualifying incident of civil conflict. We also test for a conditional relationship with Political Stability, which is the duration of the state’s political system in years according to the Polity IV data set (Marshall and Jaggers 2002). Constitutional alternations of power between political parties or regime insiders do not constitute regime failures, but this measure will reset following unconstitutional power transitions and abrupt changes that result in an immediate three-point shift in Polity Score. It takes a value of zero during prolonged periods of instability or anarchy (e.g., Liberia 1990–1997). Beyond stability, state capacity can be influenced by the location of the population in need. Dispersed, rural populations are more difficult to reach, so states with lower levels of urbanization face additional challenges during scarcity-induced crises. To capture these increased demands, we test the conditional role of Urbanization, measured as the share of the population that lives in urban centers with at least 100,000 residents. Urbanization alone would make a poor proxy of state adaptive capacity, but among states with similar resources and capabilities, those with more urban populations should have greater capacity to adapt to increased scarcity than those with rural and dispersed populations. Therefore, we can shed light on the state capacity mechanism by testing for the conditional effect of urbanization while controlling for other determinants of state strength. State Capacity Hypotheses H3a: Drought most increases the risk of conflict where past civil conflict is more recent. H3b: Drought most increases the risk of conflict where political instability is more recent. H3c: Drought most increases the risk of conflict where urbanization is low. Drought Data and Research Design We test the hypotheses on a state-year sample of all sub-Saharan states for which data are available, 1960–2009. Due to some temporal and spatial constraints on the control variables, models estimated with every variable described below cover 39 states for the years 1962–2006 (1,579 observations).9 The dependent variable is dichotomous, so we utilize logistic regressions with robust standard errors clustered by state. We also describe several alternate specifications and estimators and provide the results in a supplemental appendix.10 Like much of the literature on environmental causes of civil conflict, our dependent variable is Conflict Onset, as reported by the Uppsala Conflict Data Program Armed Conflict Dataset, version 4-2012 (Gleditsch et al. 2002; Themnér and Wallensteen 2012). This dichotomous conflict indicator is equal to 1 in any year in which a new and sufficiently deadly (25 battle fatalities per year) civil conflict is initiated. States can combat multiple groups at one time, so states can suffer new onsets even while fighting ongoing conflicts with other domestic actors. Following the literature, new onsets are recorded only if relations between the state and the group in question are peaceful for at least two years preceding the onset. Otherwise, fighting is coded as an ongoing conflict and does not constitute a new onset. The frequency of Onset in the non-missing data is 66 civil conflict onsets in 32 states. We measure drought with PDSI, which is an original adaptation of the Palmer Drought Severity Index. This index was developed by climate scientists affiliated with the United States National Center for Atmospheric Research (Dai, Trenberth, and Qian 2004; Dai 2011; Palmer 1965, 1968) and is a leading measure of soil moisture in climatological research. PDSI accounts for several relevant determinants of drought conditions, including not only precipitation but also surface net radiation, humidity, wind speed, and air pressure. This is a strength of PDSI, as drought indices that rely only on precipitation are only able to capture surpluses or deficits of rainfall, but miss additional critical features that are relevant to agriculture, including the accumulation of water in the soil column and evaporative demand of the atmosphere (Guttman 1998; McKee, Doeskin, and Kleist 1993; Trenberth et al. 2014).11 The PDSI is calibrated for local relative sensitivity, allowing PDSI scores to reflect to local climate variability. Additionally, local sensitivity is used in a climatological sense, rather than a societal sense. In other words, sensitivity is driven by climatological variables (i.e., temperature, precipitation, evapotranspiration, humidity, etc.) and not human-influenced variables (e.g., irrigation water use, pollution, agricultural harvest, etc.) that would make it endogenous to human activity. A further advantage of PDSI is that drought conditions are triangulated via many different data sources. This allows researchers to estimate drought conditions with greater accuracy over a greater temporal period (Guttman 1998; McKee, Doeskin, and Kleist 1993). The result is an accurate and comprehensive measure of dryness that ranges from –10 (extremely dry) to +10 (extremely wet). Monthly PDSI scores are normalized to long-term local averages at a resolution of 2.5 degrees latitude by 2.5 degrees longitude. Data cover the entire post-independence period and extend to the early twentieth century for many African states. To transform PDSI for the state-year level of analysis, we first calculate an area-weighted value for each state-month. The PDSI score of each 2.5-by-2.5-degree cell is multiplied by the percentage of a state’s territory falling within that cell, and these products are then summed by state to create area-weighted averages at the state-month level of analysis. We aggregate from month to year by averaging the 12 state-month values that correspond to each year. Finally, to generate a measure that is better suited for comparison across states, we normalize these annual PDSI scores to long-term state averages (see Figure 1). The result is a continuous annual measure of statewide drought conditions where 0 represents average conditions, 1 represents soil moisture that is exactly one standard deviation above that state’s average conditions, and –1 represents soil moisture that is one standard deviation less than average conditions.12 Figure 1. View largeDownload slide Mean PDSI scores by state, 1960–2010 Note: The bars show the minimum and maximum PDSI observed, 1960–2010. Long-term averages are calculated with data reaching back to the early twentieth century, so the mean of the normalized is not always equal to zero. Each grid line marks one standard deviation. Darker states have lower (drier) PDSI averages. Figure 1. View largeDownload slide Mean PDSI scores by state, 1960–2010 Note: The bars show the minimum and maximum PDSI observed, 1960–2010. Long-term averages are calculated with data reaching back to the early twentieth century, so the mean of the normalized is not always equal to zero. Each grid line marks one standard deviation. Darker states have lower (drier) PDSI averages. Next, we compare our state-level PDSI index to several popular alternatives: GPCP precipitation anomaly data (e.g. Hendrix and Salehyan 2012), CRED data on media reports of drought-afflicted populations (e.g. Slettebak 2012), and the SPI precipitation index that is often collapsed into a sub-state-level ordinal variable (e.g., Theisen, Holtermann, and Buhaug 2011–2012). Table 1 is a correlation matrix that provides evidence of the considerable differences between these measures. The GPCP data are collected by NASA’s Global Precipitation Climatology Project and measure precipitation anomalies as deviations from historical norms. The result is a continuous but highly variable measure of drought that can oscillate between extremely wet and extremely dry conditions relatively quickly. The index does not account for other meteorological phenomena that determine drought conditions, and it is available over a much shorter temporal period relative to PDSI. The Standard Precipitation Index (SPI), presented here in a continuous and state-year format, similarly tracks rainfall anomalies and correlates highly with the GPCP (see Guttman 1998). Unfortunately, most conflict studies adopting the SPI transform it into a dichotomous or four-value ordinal variable that can be used at the sub-state level. Geographic precision is traded for a substantial loss of variation. Finally, the CRED data on media reports of drought-afflicted populations are not correlated with the SPI, GPCP, or PDSI data. This provides evidence of the reporting bias that is likely to be especially pervasive in underdeveloped regions of sub-Saharan Africa. Table 1. Correlation Matrix of Drought Data Used in African Conflict Research   PDSI  GPCP  CRED  GPCP  0.607      CRED  –0.053  –0.112    SPI  0.344  0.566  –0.061    PDSI  GPCP  CRED  GPCP  0.607      CRED  –0.053  –0.112    SPI  0.344  0.566  –0.061  Note: GPCP data are produced by NASA’s Global Precipitation Climatology Project and are derived from the replication data of Hendrix and Salehyan (2012). Slettebak’s (2012)CRED variable is collected by the Center for Research on the Epidemiology of Disasters and found in his replication files. SPI is continuous and was downloaded from the website of the International Research Institute for Climate and Society at Columbia University. View Large Each model includes a number of control variables that have been linked to African civil conflict in previously published research (see Table 2). Prior work emphasizes conditions that facilitate the organization and continuation of an insurgency. Following Fearon and Laitin (2003) and Couttenier and Soubeyran (2013), we consider Mountains, which is the logged percentage of a state’s land occupied by rough, mountainous terrain. States with rougher terrain have been shown to be more conflict-prone because inaccessible areas offer safe havens to insurgents and inhibit a state’s efforts to reach isolated areas. Fearon and Laitin (2003) use a similar logic to link civil conflict to low levels of Urbanization, which we operationalize as the share of the population living in towns of more than 100,000 residents.13 Opportunities for rebel recruitment are proxied by Population, which should have a positive correlation with conflict (Bruckner 2010), and per capita GDP, which is negatively associated with conflict (Collier and Hoeffler 2004). Both measures are logged and drawn from the Penn World Table, version 7.1 (Heston, Summers, and Aten 2012). There are fewer opportunities for new, distinct conflicts where states are fighting ongoing conflicts already, so we control for Ongoing Conflict as indicated by the UCDP Armed Conflict Dataset. To separate the immediate effects of drought from lagged effects, we also include PDSI at t – 1. Table 2. Descriptive Statistics   Min  Max  Mean  St. Dev.  Source  Onset  0  1  0.04  0.20  Themner & Wallensteen (2012)  PDSI  –4.65  3.21  –0.35  1.11  Dai (2011)  Food Supply  1,439  2,978  2,139  295  FAO (2012)  Inf. Mortality  39  237  111  38  Abouharb and Kimball (2007)  Eth. Exclusion  0  0.98  0.23  0.30  Cederman, Wimmer, and Min (2010)  Polity Score  –6  7  –1.25  3.71  Vreeland (2008)  Peace Years  0  46  12  11  Themner and Wallensteen (2012)  Pol. Stability (ln)  0  4.66  1.98  1.12  Marshall and Jaggers (2002)  Urbanization  0  0.58  0.11  0.10  Singer, Bremer, and Stuckey (1972)  Pop (ln)  6.02  11.83  8.64  1.17  Heston, Summers, and Aten (2012)  pc GDP (ln)  3.49  9.24  6.26  0.93  Heston, Summers, and Aten (2012)  Ong. Conflict  0  1  0.19  0.39  Themner and Wallensteen (2012)  Mountains (ln)  0  4.42  1.56  1.48  Fearon and Laitin (2003)    Min  Max  Mean  St. Dev.  Source  Onset  0  1  0.04  0.20  Themner & Wallensteen (2012)  PDSI  –4.65  3.21  –0.35  1.11  Dai (2011)  Food Supply  1,439  2,978  2,139  295  FAO (2012)  Inf. Mortality  39  237  111  38  Abouharb and Kimball (2007)  Eth. Exclusion  0  0.98  0.23  0.30  Cederman, Wimmer, and Min (2010)  Polity Score  –6  7  –1.25  3.71  Vreeland (2008)  Peace Years  0  46  12  11  Themner and Wallensteen (2012)  Pol. Stability (ln)  0  4.66  1.98  1.12  Marshall and Jaggers (2002)  Urbanization  0  0.58  0.11  0.10  Singer, Bremer, and Stuckey (1972)  Pop (ln)  6.02  11.83  8.64  1.17  Heston, Summers, and Aten (2012)  pc GDP (ln)  3.49  9.24  6.26  0.93  Heston, Summers, and Aten (2012)  Ong. Conflict  0  1  0.19  0.39  Themner and Wallensteen (2012)  Mountains (ln)  0  4.42  1.56  1.48  Fearon and Laitin (2003)  View Large To address the temporal dependence that often occurs in time-series analysis, we add cubic polynomial terms (t, t2, and t3 to each model where t = Year – 1960). Carter and Signorino (2010) demonstrate the efficiency of this strategy relative to time fixed effects, though we use this alternate approach in robustness testing. We also lag all time-variant variables by one year and test the robustness of each model using general estimating equation (GEE) estimation, which is more efficient than fixed effects regression when the number of panels (39 states) approaches or exceeds the number of observations in each panel (45 years) (Zorn 2001). Our models do not include state fixed effects because this would exclude from the sample any countries that do not experience an onset during the time period analyzed. This reduces variation and creates significant sample selection bias; countries without a history of political violence are eliminated from the sample, thereby concentrating the analysis only on within-panel variation in the most conflict-prone states. Results The following results tables provide the odds ratios and robust panel-clustered standard errors estimated with logistic regression. Odds ratios greater than 1 indicate that an increase in the independent variable raises the risk of conflict, while odds ratios less than 1 occur when higher values of the independent variable reduce the risk of conflict. There are few statistically significant odds ratios for the constituent terms of each interaction model, but it is well established that we can infer very little about interaction effects on limited dependent variables from these values (e.g., Brambor, Clark, and Golder 2006). Even where constituent terms of the interaction do not have an independent and significant effect, significant results may occur. We therefore evaluate our hypotheses by calculating marginal effects with 90 percent confidence intervals. For each model, we hold control variables at their means14 and generate predicted probabilities across the ranges of PDSI and the variable of interest. The marginal effects illustrated in the figures are equal to the change in the predicted probability of Onset that results from a one-standard-deviation decrease in PDSI. At values of the conditional variable at which the entire confidence interval lies above or below zero, there is a statistically significant relationship between drought severity and civil conflict onset. Model 1 (Table 3) tests for an unconditional relationship between PDSI and Onset. We find no evidence of an unconditional relationship, and this is congruent with several other recent studies (Buhaug 2010; Theisen, Holtermann, and Buhaug 2011–2012). Several control variables predict Onset, and there is little variation in these predictions across the models. Mountains, Urbanization, Ethnic Exclusion, and Population (ln) increase the likelihood of civil conflict. Model 1 does not find a significant link between pc GDP (ln) and conflict, but this may be due to collinearity with Food Supply (Pearson’s r = .402) and Peace Years (Pearson’s r = .334). When these variables are excluded from the model, pc GDP (ln) has the expected negative effect on conflict onset. Onsets are also less likely where Food Supply is higher and states are already fighting ongoing civil conflicts. Table 3. Unconditional Link and Social Vulnerability Hypotheses   Model 1  Model 2 (H1a)  Model 3 (H1b)    No Conditions  Food Supply  Infant Mortality  PDSI  0.869    (0.103)  1.815    (1.462)  0.694*    (0.149)  Interaction Term        1.000    (0.000)  1.002    (0.002)  Food Supply  0.998***    (0.001)  0.998***    (0.001)        Infant Mortality              1.008*    (0.004)  PDSI (t – 1)  1.091    (0.138)  1.086    (0.139)  1.062*    (0.130)  Ethnic Exclusion  3.591**    (1.970)  3.575**    (1.971)  2.774*    (1.465)  Peace Years  0.996    (0.016)  0.996    (0.016)  0.991    (0.015)  Urbanization  39.741***    (39.325)  40.865***    (40.204)  49.699***    (65.697)  Population (ln)  1.252*    (0.145)  1.256**    (0.143)  1.204    (0.137)  pc GDP (ln)  0.974    (0.150)  0.971    (0.150)  0.790*    (0.110)  Mountains (ln)  1.202**    (0.106)  1.203**    (0.139)  1.294**    (0.131)  Ongoing Conflict  0.390**    (0.186)  0.386**    (0.184)  0.446*    (0.187)  Constant  0.243    (0.278)  0.329    (0.359)  0.003***    (0.004)  N      1,579      1,579      1,657  LL      –253.93      –253.52      –277.18    Model 1  Model 2 (H1a)  Model 3 (H1b)    No Conditions  Food Supply  Infant Mortality  PDSI  0.869    (0.103)  1.815    (1.462)  0.694*    (0.149)  Interaction Term        1.000    (0.000)  1.002    (0.002)  Food Supply  0.998***    (0.001)  0.998***    (0.001)        Infant Mortality              1.008*    (0.004)  PDSI (t – 1)  1.091    (0.138)  1.086    (0.139)  1.062*    (0.130)  Ethnic Exclusion  3.591**    (1.970)  3.575**    (1.971)  2.774*    (1.465)  Peace Years  0.996    (0.016)  0.996    (0.016)  0.991    (0.015)  Urbanization  39.741***    (39.325)  40.865***    (40.204)  49.699***    (65.697)  Population (ln)  1.252*    (0.145)  1.256**    (0.143)  1.204    (0.137)  pc GDP (ln)  0.974    (0.150)  0.971    (0.150)  0.790*    (0.110)  Mountains (ln)  1.202**    (0.106)  1.203**    (0.139)  1.294**    (0.131)  Ongoing Conflict  0.390**    (0.186)  0.386**    (0.184)  0.446*    (0.187)  Constant  0.243    (0.278)  0.329    (0.359)  0.003***    (0.004)  N      1,579      1,579      1,657  LL      –253.93      –253.52      –277.18  Note: Table shows odds ratios and robust standard errors. Models include t, t2, and t3. Control variables are lagged by one period. PDSI decreases with dry conditions. *p < .1; **p < .05; ***p < .01. View Large Models 2 and 3, which examine the social vulnerability hypotheses, find significant and surprising conditional effects (see Figure 2). In stark contrast to Hypothesis 1a, these results suggest drought increases the risk of civil conflict only where Food Supply is high. When control variables are held at their means, drought does not raise the risk of conflict in countries where people consumed less than approximately 2,050 calories per day in the previous year. Instead, drought is most likely to bring conflict to states with greater food security. The negligible slope that occurs above 2,050 calories per day indicates states above this threshold are similarly threatened by severe drought, with a shift from normal soil moisture (PDSI = 0) to one standard deviation below normal soil moisture (PDSI = –1), resulting in a .006-point increase in the risk of conflict. This is a small shift, but because Onset is relatively infrequent in the sample this means conflict is, all else being equal, roughly 50 percent more likely in a food-secure state that is one standard deviation drier than normal relative to a food-secure state that is one standard deviation wetter than normal. Figure 2. View largeDownload slide Effect of drought on conflict by social vulnerability Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Figure 2. View largeDownload slide Effect of drought on conflict by social vulnerability Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. This counterintuitive conditional relationship is supported by Model 3, which reveals a similar link between drought, conflict, and infant mortality. The social vulnerability literature would expect scarcity-conflict linkages to be strongest in societies with high infant mortality rates, but this test reveals the opposite relationship (see Figure 2). Above approximately 90 infant deaths per 1,000 live births, there is no relationship between drought and civil conflict onset. But in states with lower infant mortality rates, drought has a significant destabilizing effect. With controls at their means and Infant Mortality Rate at 60—a rate attained by states such as South Africa, Namibia, Senegal, and Gabon—a one-standard-deviation worsening in drought conditions increases the chance of conflict by 20 percent. The results of Models 2 and 3 hold when Food Supply and Infant Mortality Rate are replaced by logged versions of these variables. Next, we turn to our tests of the hypotheses related to equality of access to emergency resources (Table 4). The adaptive capacity literature expects conflict to be more likely where access to resources is perceived to be unequal or discriminatory, but the results do not support this expectation. Rather, the results echo the surprising findings presented above. Model 4 tests Hypothesis 2a and finds that the drought-conflict relationship is significant only where Ethnic Exclusion is low. The .776 odds ratio indicates that where none of the population faces ethnic exclusion, a one-standard-deviation improvement in PDSI decreases the likelihood of conflict by nearly 25 percent, ceteris paribus. The effect is illustrated in Figure 3. As Ethnic Exclusion increases, PDSI has a weaker effect on conflict. Above approximately 12 percent, this drought-conflict relationship is no longer statistically significant. This contradicts Hypothesis 2a but is similar to the finding that drought poses the greatest risk of increased conflict in states with lower levels of social vulnerability. Figure 3. View largeDownload slide Effect of drought on conflict by equality of access Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Figure 3. View largeDownload slide Effect of drought on conflict by equality of access Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Table 4. Equal Access Hypotheses   Model 4 (H2a)  Model 5 (H2b)  Model 6 (H2b)    Ethnic Exclusion  Democracy (All)  Democracy (Stable)  PDSI  0.776**  (0.097)  0.866  (0.101)  0.751**  (0.099)  Interaction Term  1.413  (0.401)  0.985  (0.021)  0.957  (0.026)  Ethnic Exclusion  4.118**  (2.432)          Polity Score      1.055  (0.039)  0.977  (0.057)  PDSI (t – 1)  1.085  (0.140)  1.091  (0.139)  0.953  (0.159)  Food Supply  0.998***  (0.001)  0.999**  (0.001)  0.999*  (0.001)  Peace Years  0.996  (0.016)  0.992  (0.017)  1.001  (0.022)  Urbanization  38.143***  (37.139)  10.363**  (10.425)  3.862  (3.643)  Population (ln)  1.265**  (0.144)  1.233  (0.116)  1.308**  (0.173)  pc GDP (ln)  0.995  (0.149)  0.949  (0.225)  1.394  (0.349)  Mountains (ln)  1.196**  (0.102)  1.141  (0.099)  1.159  (0.106)  Ongoing Conflict  0.395**  (0.186)  0.477  (0.258)  0.308*  (0.209)  Constant  0.203  (0.235)  0.408  (0.454)  0.034  (0.070)  N    1,606    1,606    1,171  LL    –253.47    –253.95    –155.99    Model 4 (H2a)  Model 5 (H2b)  Model 6 (H2b)    Ethnic Exclusion  Democracy (All)  Democracy (Stable)  PDSI  0.776**  (0.097)  0.866  (0.101)  0.751**  (0.099)  Interaction Term  1.413  (0.401)  0.985  (0.021)  0.957  (0.026)  Ethnic Exclusion  4.118**  (2.432)          Polity Score      1.055  (0.039)  0.977  (0.057)  PDSI (t – 1)  1.085  (0.140)  1.091  (0.139)  0.953  (0.159)  Food Supply  0.998***  (0.001)  0.999**  (0.001)  0.999*  (0.001)  Peace Years  0.996  (0.016)  0.992  (0.017)  1.001  (0.022)  Urbanization  38.143***  (37.139)  10.363**  (10.425)  3.862  (3.643)  Population (ln)  1.265**  (0.144)  1.233  (0.116)  1.308**  (0.173)  pc GDP (ln)  0.995  (0.149)  0.949  (0.225)  1.394  (0.349)  Mountains (ln)  1.196**  (0.102)  1.141  (0.099)  1.159  (0.106)  Ongoing Conflict  0.395**  (0.186)  0.477  (0.258)  0.308*  (0.209)  Constant  0.203  (0.235)  0.408  (0.454)  0.034  (0.070)  N    1,606    1,606    1,171  LL    –253.47    –253.95    –155.99  Note: Table shows odds ratios and robust standard errors. Models include t, t2, and t3. Control variables are lagged by one period. PDSI decreases with dry conditions. *p < .1; **p < .05; ***p < .01. View Large Models 5–6 replace Ethnic Exclusion with the Polity Score measure of democratization. Hypothesis 2b expects a drought-conflict connection only in exclusive non-democratic states, but again, results strongly suggest that drought increases the risk of conflict only where Exclusion is relatively low. The results produced by Model 5 (center plot of Figure 3) show that drought has no effect on conflict onset in non-democratic regimes, but this effect increases with democratization. The effect just misses the conventional threshold for statistical significance, but this may be because many of the democracies in the sample are transitional and fears of exclusion may persist even after very recent constitutional changes. Model 6 replicates Model 5 on a sub-sample of states that have had stable political systems for three years. The predicted magnitude of the drought-conflict relationship is very similar to the predictions of Model 5, though the confidence intervals tighten and the counterintuitive conditional relationship becomes statistically significant. In Africa’s most democratic states, a one-standard-deviation worsening of drought conditions increases the risk of war by nearly 1.5 percentage points, or roughly 45 percent. Table 5 reports results for the tests of the three hypotheses pertaining to state adaptive capacity. None of the hypothesized effects was supported. Once again, each of the three models produced statistically significant effects in the opposite direction. The literature on adaptive capacity links state capacity to a state’s history of conflict, claiming that more stable states should be more able to mitigate any adverse effects of scarcity-inducing environmental events. But Model 7, as illustrated in the leftmost plot of Figure 4, shows no drought-conflict relationship in states with very recent histories of violence. With controls at their means, PDSI has a significant effect on the risk of Onset only among states that have experienced more than a decade of peace. Stable states are undermined by drought; the risk of conflict increases 20 percent when PDSI drops by one standard deviation in states that have been peaceful for 15 years. The same change in PDSI raises the risk of conflict by 25 percent in states that have been peaceful for 30 years. Contrary to our expectations, the states that should be most capable are those that experience the greatest change in the risk of war during drought. Figure 4. View largeDownload slide Effect of drought on conflict by state capacity Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Figure 4. View largeDownload slide Effect of drought on conflict by state capacity Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Table 5. State Capacity Hypotheses   Model 7 (H3a)  Model 8 (H3b)  Model 9 (H3c)    Peace Years  Political Stability  Urbanization  PDSI  1.043  (0.191)  0.970  (0.157)  0.939  (0.169)  Interaction Term  0.984  (0.012)  0.927  (0.069)  0.572  (0.342)  Urbanization  40.93***  (40.85)  36.17***  (39.56)  29.86***  (29.93)  Peace Years  0.985  (0.019)          Political Stability (l      0.822  (0.119)      PDSI (t-1)  1.096  (0.138)  1.091  (0.141)  1.090  (0.136)  Food Supply  0.998***  (0.001)  0.998***  (0.001)  0.998***  (0.001)  Ethnic Exclusion  3.561**  (1.970)  3.886**  (2.138)  3.618**  (2.002)  Population (ln)  1.254**  (0.145)  1.198  (0.118)  1.257*  (0.150)  pc GDP (ln)  0.981  (0.151)  1.114  (0.180)  0.974  (0.150)  Mountains (ln)  1.203**  (0.108)  1.209*  (0.120)  1.194**  (0.104)  Ongoing Conflict  0.382**  (0.181)  0.379  (0.159)  0.418**  (0.163)  Constant  0.198  (0.238)  0.125  (0.154)  0.237  (0.277)  N    1,579    1,562    1,579  LL    –252.91    –    –253.82    Model 7 (H3a)  Model 8 (H3b)  Model 9 (H3c)    Peace Years  Political Stability  Urbanization  PDSI  1.043  (0.191)  0.970  (0.157)  0.939  (0.169)  Interaction Term  0.984  (0.012)  0.927  (0.069)  0.572  (0.342)  Urbanization  40.93***  (40.85)  36.17***  (39.56)  29.86***  (29.93)  Peace Years  0.985  (0.019)          Political Stability (l      0.822  (0.119)      PDSI (t-1)  1.096  (0.138)  1.091  (0.141)  1.090  (0.136)  Food Supply  0.998***  (0.001)  0.998***  (0.001)  0.998***  (0.001)  Ethnic Exclusion  3.561**  (1.970)  3.886**  (2.138)  3.618**  (2.002)  Population (ln)  1.254**  (0.145)  1.198  (0.118)  1.257*  (0.150)  pc GDP (ln)  0.981  (0.151)  1.114  (0.180)  0.974  (0.150)  Mountains (ln)  1.203**  (0.108)  1.209*  (0.120)  1.194**  (0.104)  Ongoing Conflict  0.382**  (0.181)  0.379  (0.159)  0.418**  (0.163)  Constant  0.198  (0.238)  0.125  (0.154)  0.237  (0.277)  N    1,579    1,562    1,579  LL    –252.91    –    –253.82  Note: Table shows odds ratios and robust standard errors. Models include t, t2, and t3. Control variables are lagged by one period. PDSI decreases with dry conditions. *p < .1; **p < .05; ***p < .01. View Large We would also expect unstable, transitional political systems (H3b) and rural, dispersed populations (H3c) to thwart states’ efforts to effectively adapt to environmental scarcity, yet the results suggest the opposite is true. The effect produced by Model 8 is weaker than that produced by Model 7, though this may be unsurprising because civil conflict can be more disruptive than other forms of political change. There is no link between PDSI and Onset where states have had less than five years of Political Stability, but drought increases the risk of war where political systems are more enduring. Most sub-Saharan states have very large rural populations, but drought has the greatest marginal effect on the risk of conflict in the more urbanized states. States that are 25 percent urban see a 17 percent increase in the risk of war when drought worsens by one standard deviation, yet drought has no discernible effect in very rural states. Discussion The results and robustness tests15 do not support the idea that drought amplifies the risk of conflict in states that are made susceptible by weak state capacity, inequality, or poor human welfare. In fact, the preponderance of the models evince the opposite: drought most destabilizes states with superior living conditions, greater food security, longer-enduring governments, more inclusive political systems, small rural populations, and longer histories of peace. This effect is strong enough to completely offset the pacifying effects that these conditions provide during normal weather conditions. In other words, drought is a conflict risk equalizer. During drought, states that would otherwise be the most prone to civil conflict are no more likely to see violence than states that are otherwise exceptionally unlikely to experience conflict. These results leave us with an important puzzle with pressing policy implications. Why would drought have a greater effect on the risk of conflict where political violence is less likely during normal weather conditions, and what might this mean for international efforts to insulate developing states from the adverse effects of climate change? Answering these questions will require further research, but we believe the answer may be found in the behavioral economics literature on the endowment effect (Kahneman and Tversky 1979; Thaler 1980). According to endowment theory, the value people place on objects and experiences is subjective and is biased by perceptions of the status quo, which serve as psychological reference points during opportunities for loss or gain. People tend to place great value on what they already possess, even to the point that they will accept irrational risks to prevent adverse change. A classic and oft-replicated experiment provides a simple illustration of this effect. Test subjects are divided into two groups. One group is given mugs, and the other is given candy. When the subjects are encouraged to trade these goods, researchers find that possession causes both mug-possessors and candy-possessors to demand much higher prices than their potential buyers are willing to accept. Even when the period of possession is very brief and subjects spend nothing to acquire their goods, endowment with these objects causes measurable loss aversion (Brenner et al. 2007; Carmon and Ariely 2000; Kahneman, Knetsch, and Thaler 1991). Exchanges of mugs and candy are somewhat trivial, but the same psychological bias can influence conflict behavior. In this context, reactions to an absolute condition like drought-induced scarcity should depend upon how one’s welfare during drought compares to one’s welfare during an ex ante reference point. Where welfare is already very poor, increased scarcity does not generate the same sense of loss that it does where welfare was once much more comfortable. We see this endowment effect in the results. All of the interaction models show that drought has no effect on the risk of conflict when sociopolitical conditions are already very unfavorable (high infant mortality, food insecurity, political instability, etc.). Rather, reactions to increased scarcity are strongest among those who perceive themselves to have more to lose. Endowment with a comfortable level of welfare in the ex ante status quo increases loss aversion, and this loss aversion can motivate political violence. We are not the first to apply work in this area to studies of political conflict. This endowment mechanism shares a common foundation with the International Relations literature on prospect theory and risk acceptance in the “domain of losses” (Levy 1992; Mercer 2005). Levy (1992) notes, for example, that wars sometimes occur because states that possess a contested territory are willing to fight to keep it even where they are militarily outmatched and a negotiated concession might be more advantageous. A weaker state facing territorial loss will fight a much stronger state because concession would represent an adverse change from the ex ante reference point. The resulting loss aversion drives the weak state to overvalue the territory and compels it to engage in an unwise armed conflict. This can also explain why declining world powers, such as Russia, may be more belligerent in their efforts to prolong their sphere of influence during a period of contraction. This logic is counterintuitive, but it is supported by the history of drought and civil conflict in sub-Saharan Africa. Populations that are accustomed to scarcity and ineffective government are no more likely to engage in civil conflict when drought threatens prolonged scarcity and poor welfare. But the threat of great loss relative to the status quo causes drought to significantly raise the risk of conflict in states that are normally less conflict-prone. Conclusion This paper was motivated by a significant gap in the literature on environmental drivers of violent civil conflict. Work based on case studies and small-n comparisons commonly argues that certain sociopolitical conditions can mitigate the destabilizing effects of drought, while statistical analyses of this relationship are generally insensitive to conditional mechanisms. We hoped to reconcile these branches of the literature with an array of statistical tests on the three intervening conditions that are proposed in the case literature: high social vulnerability, low state capacity, and unequal access to resources. To our surprise, none of the eight interaction models we used supported the hypotheses derived from this literature, though they did generate significant findings in the opposite direction. Drought does not have a greater destabilizing effect in insecure states with vulnerable, divided populations. It increases the risk of civil conflict only in states that would otherwise be relatively secure—states with greater food supply, lower infant mortality rates, more democratic governments, longer-enduring political systems, longer histories of peace, more urbanized populations, and larger populations suffering ethnic-based political exclusion. These results, which are robust to alternate measures of the intervening sociopolitical concepts and most alternate model specifications, give rise to a fascinating enigma: Why would states that are stable under normal conditions suffer the greatest destabilizing effects during drought? We drew on research by psychologists and behavioral economists to propose that the endowment effect, an established psychological bias that exaggerates feelings of loss and encourages risky efforts to avoid adverse change, could cause the threat of scarcity to most increase the risk of conflict where scarcity promises the largest deviation from the ex ante status quo. This argument has strong empirical support and has been used in the International Relations literature on interstate conflict, but we know of no other research to apply this logic to the study of environmental causes of civil conflict. We hope this analysis will motivate further work in several areas. It is important for future research to explore whether these counterintuitive conditional relationships can be generalized to other geographic regions or other types of political violence. We have no strong theoretical reason to believe these mechanisms should apply only to deadly conflicts in sub-Saharan Africa, and it would be particularly important for future work to examine similar drought-conflict linkages in non-African middle-income countries. This research program would also benefit from further exploration of scale effects. Work on sub-state variation in these conditions could shed light on how regional inequalities might influence the emergence of large-scale civil conflict. These advances would benefit the international policy community, which must respond to the divergence between the funds it requests and those it successfully collects with informed prioritization. Maintaining the stability provided by better social welfare and more effective government is paramount to the security of states that are projected to suffer from more extreme and more frequent severe weather events in the future. Work in this area is desperately needed, as demands on international donors exceed the resources available for adaptation efforts many times over. Footnotes 1 Barack Obama’s 2009 speech to the United Nations typifies this argument. He warns, “No nation, however large or small, wealthy or poor, can escape the impact of climate change… The security and stability of each nation and all peoples—our prosperity, our health, our safety—are in jeopardy” (Obama 2009). 2 Environmental roots of the conflict in Darfur are debated by Moon (2007), Faris (2007), Tubiana (2007), and Kevane and Gray (2008). Werrell and Femia (2013) and Ahmed (2013) discuss environmental causes of the Arab Spring. The links between food riots, climate change, and instability are weighed by Ahmed (2013), Barrett and Bellemare (2011), and Berazneva and Lee (2013). 3 See special issues of Political Geography (2007) and the Journal of Peace Research (2012), and cross-disciplinary exchanges in Science (Scheffran et al. 2012), Nature (Hsiang, Meng, and Cane 2011), and PNAS (Buhaug 2010; Burke et al. 2009; O’Loughlin et al. 2012). 4 See, among others, Buhaug’s (2010) response to Burke et al. (2009), Theisen’s (2008) rebuttal of Hauge and Ellingsen (1998), and Ciccone’s (2011) comment on Miguel, Satyanath, and Sergenti (2004). 5 Gleditsch (2012) discusses these recent sub-literatures as they appear in the largest and most recent special issue on environmental security. 6 For a more extensive critique of the CRED data, see Gall, Borden, and Cutter (2009). 7 We thank Couttenier and Soubeyran for sharing their replication data. 8 For work on sub-national variation in adaptive capacity and social vulnerability, see Busby et al. (2013). Brooks, Adger, and Kelly (2005) argue the adaptive capacity of state-level government is also important in responses to environmental crises. 9 The FAO does not record Food Supply for the Democratic Republic of the Congo, Equatorial Guinea, or Somalia. Cederman, Wimmer, and Min (2010) do not include Djibouti in their data set on Ethnic Power Relations. 10 Models are estimated with Stata 12, and all replication materials will be made available on the authors’ websites. 11 Specifically, we use scPDSIpm (self-calibrating Palmer Drought Severity Index with the Penman-Monteith method of calculating evapotranspiration), which includes these multiple-dimensions to fully characterize agriculturally relevant drought (Dai 2011; Dai, Trenberth, and Qian 2004; Trenberth et al. 2014). 12 Results are nearly identical when PDSI is not normalized. 13 We generate this variable using data from the National Capabilities dataset hosted by the Correlates of War Project (CINC). 14 t is set at the mean year (25). t2 and t3 are set at 252 and 253 instead of their means. 15 The results of more than 25 robustness tests are included in the online supplementary appendix. References Abouharb Rodwan M., Kimball Anessa L.. 2007. “ A New Dataset on Infant Mortality Rates, 1816–2002.” Journal of Peace Research  44: 743– 54. Google Scholar CrossRef Search ADS   Adger W. Neil. 1999. “ Social Vulnerability to Climate Change and Extremes in Coastal Vietnam.” World Development  27: 249– 69. Google Scholar CrossRef Search ADS   Ahmed Nafeez Mossaddeq. 2013. “Why Food Riots Are Likely to Become the New Normal.” Guardian, March 6. Accessed March 2013, www.guardian.co.uk. Barnett Jon, Adger W. Neil. 2007. “ Climate Change, Human Security and Violent Conflict.” Political Geography  26: 639– 55. Google Scholar CrossRef Search ADS   Barrett Christopher B., Bellemare Marc F.. 2011. “Why Food Price Volatility Doesn’t Matter.” Foreign Affairs, July 12. Accessed March 2013, www.foreignaffairs.com. Barrios Salvador, Bertinelli Luisito, Strobi Eric. 2006. “ Climatic Change and Rural-Urban Migration: The Case of Sub-Saharan Africa.” Journal of Urban Economics  60: 357– 71. Google Scholar CrossRef Search ADS   Beck Nathaniel. 2011. “Is OLS with a Binary Dependent Variable Really OK? Estimating (Mostly) TSCS Models with Binary Dependent Variables and Fixed Effects.” Accessed January 2014, http://politics.as.nyu.edu/docs/IO/2576/pgm2011.pdf. Benjaminsen Tor A. 2008. “ Does Supply-Induced Scarcity Drive Violent Conflicts in the African Sahel? The Case of the Tuareg Rebellion in Northern Mali.” Journal of Peace Research  45: 819– 36. Google Scholar CrossRef Search ADS   Berazneva Julia, Lee David R.. 2013. “ Explaining the African Food Riots of 2007–2008: An Empirical Analysis.” Food Policy  39: 28– 39. Google Scholar CrossRef Search ADS   Bergholt Drago, Lujala Paivi. 2012. “ Climate-Related Natural Disasters, Economic Growth, and Armed Civil Conflict.” Journal of Peace Research  49: 147– 62. Google Scholar CrossRef Search ADS   BinningsbØ Helga Malmin, de Soysa Indra, Gleditsch Nils Petter. 2007. “ Green Giant or Straw Man? Environmental Pressure and Civil Conflict, 1961–99.” Population and Environment  28: 337– 53. Google Scholar CrossRef Search ADS   Bogale Ayalneh, Korf Benedikt. 2007. “ To Share or Not to Share? (Non-) Violence, Scarcity, and Resource Access in Somali Region, Ethiopia.” Journal of Development Studies  43: 743– 65. Google Scholar CrossRef Search ADS   Bohle Hans G., Downing Thomas E., Watts Michael J.. 1994. “ Climate Change and Social Vulnerability: Toward a Sociology and Geography of Food Insecurity.” Global Environmental Change  4: 37– 48. Google Scholar CrossRef Search ADS   Brambor Thomas, Clark William Roberts, Golder Matt. 2006. “ Understanding Interaction Models: Improving Empirical Analyses.” Political Analysis  14: 63– 82. Google Scholar CrossRef Search ADS   Brenner Lyle, Rottenstreich Yuval, Sood Sanjay, Bilgin Baler. 2007. “ On the Psychology of Loss Aversion: Possession, Valence, and Reversals of the Endowment Effect.” Journal of Consumer Research  34: 369– 76. Google Scholar CrossRef Search ADS   Brooks Nick, Neil Adger W., Mick Kelly P.. 2005. “ The Determinants of Vulnerability and Adaptive Capacity at the National Level and the Implications for Adaptation.” Global Environmental Change  15: 151– 63. Google Scholar CrossRef Search ADS   Bruckner Markus. 2010. “ Population Size and Civil Conflict Risk: Is There a Causal Link?” Economic Journal  120: 535– 50. Google Scholar CrossRef Search ADS   Bueno de Mesquita Bruce, Smith Alastair, Siverson Randolph M., Morrow James D.. 2003. The Logic of Political Survival . Cambridge, MA: MIT Press. Buhaug Halvard. 2010. “ Climate Not to Blame for African Civil Wars.” Proceedings of the National Academy of Sciences of the United States of America  107: 16477– 82. Google Scholar CrossRef Search ADS PubMed  Burke Marshall B., Miguel Edward, Satyanath Shanker, Dykema John A., Lobell David B.. 2009. “ Warming Increases the Risk of Civil War in Africa.” Proceedings of the National Academy of Sciences of the United States of America  106: 20670–7 4. Google Scholar CrossRef Search ADS PubMed  Busby Joshua W., Smith Todd G., White Kaiba L., Strange Shawn M.. 2013. “ Climate Change and Insecurity: Mapping Vulnerability in Africa.” International Security  37: 132– 72. Google Scholar CrossRef Search ADS   Carmon Ziv, Ariely Dan. 2000. “ Focusing on the Forgone: How Value Can Appear So Different to Buyers and Sellers.” Journal of Consumer Research  27: 360– 70. Google Scholar CrossRef Search ADS   Carter David B., Signorino Curtis S.. 2010. “ Back to the Future: Modeling Time Dependence in Binary Data.” Political Analysis  18: 271– 92. Google Scholar CrossRef Search ADS   Cederman Lars-Erik, Wimmer Andreas, Min Brian. 2010. “ Why Do Ethnic Groups Rebel? New Data and Analysis.” World Politics  62: 87– 119. Google Scholar CrossRef Search ADS   Ciccone Antonio. 2011. “ Economic Shocks and Civil Conflict: A Comment.” American Economic Journal: Applied Economics  3 ( 1): 215– 27. Google Scholar CrossRef Search ADS   Collier Paul, Hoeffler Anke. 2004. “ Greed and Grievance in Civil War.” Oxford Economic Papers  56: 563– 96. Google Scholar CrossRef Search ADS   Conrad Courtenay R. 2011. “ Constrained Concessions: Beneficent Dictatorial Responses to the Domestic Political Opposition.” International Studies Quarterly  55: 1167– 87. Google Scholar CrossRef Search ADS   Couttenier Mathieu, Soubeyran Raphael. 2013. “ Drought and Civil War in Sub-Saharan Africa.” Economic Journal  124: 201– 44. Google Scholar CrossRef Search ADS   CRED. 2013. EM-DAT: The OFDA/CRED International Disasters Database . Brussels, Belgium: Centre for Research on the Epidemiology of Disasters. Dai Aiguo. 2011. “ Characteristics and Trends in Various Forms of the Palmer Drought Severity Index during 1900–2008.” Journal of Geophysical Research  116: 1– 26. Dai Aiguo, Trenberth Kevin E., Qian Taotao. 2004. “ A Global Dataset of Palmer Drought Severity Index for 1870–2002: Relationship with Soil Moisture and Effects of Surface Warming.” Hydrometeorology  5: 1117– 30. Google Scholar CrossRef Search ADS   de Soysa Indra. 2002. “ Ecoviolence: Shrinking Pie, or Honey Pot?” Global Environmental Politics  2: 1– 34. Google Scholar CrossRef Search ADS   FAO. 2012. FAOSTAT 2012 . New York: Food and Agriculture Organization of the United Nations. Faris Stephan. 2007. “The Real Roots of Darfur.” Atlantic Monthly. Accessed March 2013, www.theatlantic.com. Fearon James D., Laitin David D.. 2003. “ Ethnicity, Insurgency, and Civil War.” American Journal of Political Science  97: 75– 90. Google Scholar CrossRef Search ADS   Fjelde Hanne, von Uexkull Nina. 2012. “ Climate Triggers: Rainfall Anomalies, Vulnerability, and Communal Conflict in Sub-Saharan Africa.” Political Geography  31: 444– 53. Google Scholar CrossRef Search ADS   Gall Melanie, Borden Kevin A., Cutter Susan L.. 2009. “ When Do Losses Count? Six Fallacies of Natural Hazards Loss Data.” Bulletin of the American Meteorological Society  90: 799– 809. Google Scholar CrossRef Search ADS   Gleditsch Kristian Skrede. 2007. “ Transnational Dimensions of Civil War.” Journal of Peace Research  44: 293– 309. Google Scholar CrossRef Search ADS   Gleditsch Nils Petter. 1998. “ Armed Conflict and the Environment: A Critique of the Literature.” Journal of Peace Research  35: 381– 400. Google Scholar CrossRef Search ADS   Gleditsch Nils Petter. 2012. “ Whither the Weather? Climate Change and Conflict.” Journal of Peace Research  49: 3– 9. Google Scholar CrossRef Search ADS   Gleditsch Nils Petter, Nordas Ragnhild, Idean Salehyan. 2007. “Climate Change and Conflict: The Migration Link.” International Peace Academy. Gleditsch Nils Petter, Wallensteen Peter, Eriksson Mikael, Sollenberg Margareta, Strand Halvard. 2002. “ Armed Conflict 1946–2001: A New Dataset.” Journal of Peace Research  39: 615– 37. Google Scholar CrossRef Search ADS   Guttman N. B. 1998. “ Comparing the Palmer Drought Index and the Standardized Precipitation Index.” JAWRA.” Journal of the American Water Resources Association  34: 113– 21. Google Scholar CrossRef Search ADS   Hauge Wenche, Ellingsen Tanja. 1998. “ Beyond Environmental Scarcity: Causal Pathways to Conflict.” Journal of Peace Research  35: 299– 317. Google Scholar CrossRef Search ADS   Hegre Havard, Tanja Ellingsen, Gates Scott, Gleditsch Nils Petter. 2001. “ Toward a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816–1992.” American Political Science Review  95: 33– 48. Hendrix Cullen S., Glaser Sarah M.. 2007. “ Trends and Triggers: Climate, Climate Change and Civil Conflict in Sub-Saharan Africa.” Political Geography  26: 695– 715. Google Scholar CrossRef Search ADS   Hendrix Cullen S., n Salehyan Idea. 2012. “ Climate Change, Rainfall, and Social Conflict in Africa.” Journal of Peace Research  49: 35– 50. Google Scholar CrossRef Search ADS   Heston Alan, Summers Robert, Aten Bettina. 2012. Penn World Table Version 7.1. Philadelphia: Center for Production, Income, and Prices at the University of Pennsylvania. Homer-Dixon Thomas. 1991. “ On the Threshold: Environmental Changes as Causes of Acute Conflict.” International Security  16: 76– 116. Google Scholar CrossRef Search ADS   Homer-Dixon Thomas. 1994. “ Environmental Scarcities and Violent Conflict: Evidence from Cases.” International Security  19: 5– 40. Google Scholar CrossRef Search ADS   Homer-Dixon Thomas. 1999. Environment, Scarcity, and Violence . Princeton, NJ: Princeton University Press. Homer-Dixon Thomas. 2000. The Ingenuity Gap: Facing the Economic, Environmental, and Other Challenges of an Increasingly Complex and Unpredictable Future . New York: Vintage. Hsiang Solomon M., Meng Kyle C., Cane Mark A.. 2011. “ Civil Conflicts Are Associated with the Global Climate.” Nature  476: 438– 41. Google Scholar CrossRef Search ADS PubMed  Huntington Samuel P. 1968. Political Order in Changing Societies . Cambridge, MA: Harvard University Press. Kahl Colin. 1998. “ Population Growth, Environmental Degradation, and State-Sponsored Violence: The Case of Kenya, 1991–93.” International Security  23: 80– 119. Google Scholar CrossRef Search ADS   Kahl Colin. 2006. States, Scarcity, and Civil Strife in the Developing World . Princeton, NJ: Princeton University Press. Kahneman Daniel, Knetsch Jack L., Thaler Richard H.. 1991. “ Anomalies: The Endowment Effect, Loss Aversion, and Status Quo Bias.” Journal of Economic Perspectives  5: 193– 206. Google Scholar CrossRef Search ADS   Kahneman Daniel, Tversky Amos. 1979. “ Prospect Theory: An Analysis of Decision under Risk.” Econometrica  47: 263– 92. Google Scholar CrossRef Search ADS   Kevane Michael, Gray Leslie. 2008. “ Darfur: Rainfall and Conflict.” Environmental Research Letters  3: 034006. Google Scholar CrossRef Search ADS   Levy Jack S. 1992. “ Prospect Theory and International Relations: Theoretical Applications and Analytical Problems.” Political Psychology  13: 283– 310. Google Scholar CrossRef Search ADS   Marshall Monty G., Jaggers Keith. 2002. Polity IV Data Set [Computer File: version p4v2002]. Center for International Development and Conflict Management , University of Maryland. Martin Adrian. 2005. “ Environmental Conflict Between Refugee and Host Communities.” Journal of Peace Research  42: 329. Google Scholar CrossRef Search ADS   McKee T. B., Doeskin N. M., Kleist J.. 1993. “ The Relationship of Drought Frequency and Duration to Time Scales.” Proceedings of the 8th Conference on Applied Climatology, American Meteorological Society  1993: 179– 84. Meier Patrick, Bond Doug, Bond Joe. 2007. “ Environmental Influences on Pastoral Conflict in the Horn of Africa.” Political Geography  26: 716– 35. Google Scholar CrossRef Search ADS   Mercer Jonathan. 2005. “ Prospect Theory and Political Science.” Annual Review of Political Science  8: 1– 21. Google Scholar CrossRef Search ADS   Miguel Edward, Satyanath Shanker, Sergenti Ernest. 2004. “ Economic Shocks and Civil Conflict: An Instrumental Variable Approach.” Journal of Political Economy  112: 725– 53. Google Scholar CrossRef Search ADS   Moon Ban Ki. 2007. “A Climate Culprit in Darfur.” Washington Post. Accessed March 2013, June 16. www.washingtonpost.com. Müller Christoph, Cramer Wolfgang, Hare William L., Lotze-Campen Hermann. 2011. “ Climate Change Risks for African Agriculture.” Proceedings of the National Academy of Sciences of the United States of America  108: 4313– 15. Google Scholar CrossRef Search ADS PubMed  Obama Barack. 2009. Remarks by the President at the Morning Plenary Session of the United Nations Climate Change Conference, December 18. Washington, DC: White House. Accessed March 2013, www.whitehouse.gov. O’Brien Karen, Leichenko Robin, Kelkar Ulka, Venema Henry, Aandahl Guro, Tompkins Heather, Javed Akrem, et al.  . 2004. “ Mapping Vulnerability to Multiple Stressors: Climate Change and Globalization in India.” Global Environmental Change  14: 303– 13. Google Scholar CrossRef Search ADS   O’Loughlin John, Witmer Frank D. W., Linke Andrew M., Laing Arlene, Gettelman Andrew, Dudhia Jimy. 2012. “ Climate Variability and Conflict Risk in East Africa, 1990–2009.” Proceedings of the National Academy of Sciences of the United States of America  109: 18344– 49. Google Scholar CrossRef Search ADS PubMed  Palmer W. C. 1965. “Research Paper No. 45, US Weather Bureau.” NOAA Library and Information Services Division, Washington, DC. Palmer W. C. 1968. “ Keeping Track of Crop Moisture Conditions, Nationwide: The New Moisture Index.” Weatherwise  21: 156– 61. Google Scholar CrossRef Search ADS   Plümper Thomas, Troeger Vera E.. 2007. “ Efficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects.” Political Analysis  15: 124– 39. Google Scholar CrossRef Search ADS   Raleigh Clionadh. 2010. “ Political Marginalization, Climate Change, and Conflict in African Sahel States.” International Studies Review  12: 69– 86. Google Scholar CrossRef Search ADS   Raleigh Clionadh, Kniveton Dominic. 2012. “ Come Rain or Shine: An Analysis of Conflict and Climate Variability in East Africa.” Journal of Peace Research  49: 51– 64. Google Scholar CrossRef Search ADS   Raleigh Clionadh, Urdal Henrik. 2007. “ Climate Change, Environmental Degradation and Armed Conflict.” Political Geography  26: 674– 94. Google Scholar CrossRef Search ADS   Reardon Thomas, Taylor J. Edward. 1996. “ Agroclimatic Shock, Income Inequality, and Poverty: Evidence from Burkina Faso.” World Development  24: 901– 14. Google Scholar CrossRef Search ADS   Reuveny Rafael. 2007. “ Climate Change-Induced Migration and Violent Conflict.” Political Geography  26: 656– 73. Google Scholar CrossRef Search ADS   Salehyan Idean. 2008. “ From Climate Change to Conflict? No Consensus Yet.” Journal of Peace Research  45: 315– 26. Google Scholar CrossRef Search ADS   Salehyan Idean, Gleditsch Kristian Skrede. 2006. “ Refugees and the Spread of Civil War.” International Organization  60: 335– 66. Google Scholar CrossRef Search ADS   Scheffran Jurgen, Brzoska Michael, Kominek Jasmin, Michael Link P., Schilling Janpeter. 2012. “ Climate Change and Violent Conflict.” Science  336: 869– 71. Google Scholar CrossRef Search ADS PubMed  Singer J. David, Bremer Stuart, Stuckey John. 1972. Peace, War, and Numbers . Beverly Hills, CA: Sage. Slettebak Rune T. 2012. “ Don’t Blame the Weather! Climate-Related Natural Disasters and Civil Conflict.” Journal of Peace Research  49: 163– 76. Google Scholar CrossRef Search ADS   Thaler Richard H. 1980. “ Toward a Positive Theory of Consumer Choice.” Journal of Economic Behavior and Organization  1: 39– 60. Google Scholar CrossRef Search ADS   Theisen Ole Magnus. 2008. “ Blood and Soil? Resource Scarcity and Internal Armed Conflict Revisited.” Journal of Peace Research  45: 801– 18. Google Scholar CrossRef Search ADS   Theisen Ole Magnus. 2012. “ Climate Clashes? Weather Variability, Land Pressure, and Organized Violence in Kenya, 1989–2004.” Journal of Peace Research  49: 81– 96. Google Scholar CrossRef Search ADS   Theisen Ole Magnus, Holtermann Helge, Buhaug Halvard. 2011–2012. “ Climate Wars? Assessing the Claim That Drought Breeds Conflict.” International Security  36: 79– 106. Google Scholar CrossRef Search ADS   Themnér Lotta, Wallensteen Peter. 2012. “ Armed Conflict, 1946–2011.” Journal of Peace Research  49: 565– 75. Google Scholar CrossRef Search ADS   Tol Richard S. J., Yohe Gary W.. 2007. “ The Weakest Link Hypothesis for Adaptive Capacity: An Empirical Test.” Global Environmental Change  17: 218– 27. Google Scholar CrossRef Search ADS   Trenberth Kevin E., Dai Aiguo, van der Schrier Gerard, Jones Philip D., Barichivich Jonathan, Briffa Keith R., Sheffield Justin. 2014. “ Global Warming and Changes in Drought.” Nature Climate Change  4: 17– 22. Google Scholar CrossRef Search ADS   Tubiana Jerome. 2007. “ Darfur: A Conflict for Land?” In War in Darfur and the Search for Peace , edited by de Waal Alex, 68– 91. Cambridge, MA: Global Equity Initiative, Harvard University. UN. 2011. Secretary-General Presents $5.2 Billion 2012–2013 Budget to Fifth Committee, Says Figure Acknowledges Ongoing Financial Uncertainty, Austerity . New York: United Nations Department of Public Information. UNFCCC. 2007. Climate Change: Impacts, Vulnerabilities, and Adaptation in Developing Countries . New York: United Nations. UNFCCC. 2012a. National Adaptation Plans (Document FCCC/SBI/2012/L.41) . New York: United Nations. UNFCCC. 2012b. National Adaptation Programmes of Action: Index of NAPA Projects by Country . New York: United Nations. Vreeland James Raymond. 2008. “ The Effect of Political Regime on Civil War: Unpacking Anocracy.” Journal of Conflict Resolution  52: 401– 25. Google Scholar CrossRef Search ADS   Werrell Caitlin E., Femia Francesco, eds. 2013. The Arab Spring and Climate Change . Washington, DC: Center for American Progress. World Bank. 2013. World Bank Development Indicators . Washington, DC: World Bank. Zorn Christopher J. W. 2001. “ Generalized Estimating Equation Models for Correlated Data: A Review with Applications.” American Journal of Political Science  45: 470– 90. Google Scholar CrossRef Search ADS   © The Author (2016). Published by Oxford University Press on behalf of the International Studies Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Foreign Policy Analysis Oxford University Press

Conditional Relationships Between Drought and Civil Conflict in Sub-Saharan Africa

Loading next page...
 
/lp/ou_press/conditional-relationships-between-drought-and-civil-conflict-in-sub-yiOcEO5MwS
Publisher
Oxford University Press
Copyright
© The Author (2016). Published by Oxford University Press on behalf of the International Studies Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
1743-8586
eISSN
1743-8594
D.O.I.
10.1093/fpa/orw002
Publisher site
See Article on Publisher Site

Abstract

Abstract Much of the literature on climate change adaptation claims the destabilizing consequences of environmental crises are mitigated by sociopolitical conditions that influence a state's susceptibility to scarcity-induced violence. However, few cross-national studies provide evidence of conditional scarcity-conflict relationships. This analysis of drought severity and civil conflict onset in sub-Saharan Africa (1962–2006) uncovers three sociopolitical conditions that influence the link between environmental scarcity and civil conflict: social vulnerability, state capacity, and unequal distribution of resources. Surprisingly, we find drought does not exacerbate the high risk of conflict in the vulnerable, incapable, and unequal states thought to be especially susceptible to increased scarcity. Instead, drought negates the peace-favoring attributes of stable states with less vulnerable populations. During severe drought, states with sociopolitical conditions that would otherwise favor peace are no less likely to suffer conflict than states with sociopolitical conditions that would otherwise increase the risk of violence. These findings, which are robust across several measures of these sociopolitical concepts, suggest environmental scarcity is most likely to increase the risk of conflict where populations have more to lose relative to periods with more favorable weather. The 2010 Conference of Parties of the United Nations Framework Convention on Climate Change (UNFCCC) highlighted an important transition in international environmental policy. Diplomats made little progress toward the organization’s original goal of climate change prevention. Instead, the conference resulted in an ambitious agenda focused on climate change adaptation. Fearing that adverse environmental change may now be inevitable, members renewed earlier pledges to help developing states craft adaptation plans and created a new Green Climate Fund to finance these efforts (UNFCCC 2012a). The UNFCCC estimates fully funding these plans will require contributions in excess of $100 billion UNFCCC 2012b), though this infeasible sum is nearly 50 times greater than the 2013 budget of the entire United Nations (UN 2011). This costly reaction to climate change is motivated in part by the popular notion that environmental catalysts prompt violent conflict where vulnerable states lack the capacity to adequately adapt.1 Events such as the genocide in Darfur, food riots in East Africa, and the 2010–2011 Arab Spring are sometimes offered as anecdotal support for this claim.2 But despite the popularity of the scarcity-conflict hypothesis among journalists, policymakers, and some conflict scholars, research has yet to generate consensus about a generalizable relationship between environmental catalysts and conflict (Gleditsch 2012; Salehyan 2008).3 Scholarly efforts to examine this relationship are impeded by an important disjuncture in the literature. Theories of environmental conflict, which are often drawn from case comparisons and small-n studies, emphasize several sociopolitical conditions that influence the likelihood with which scarcity-inducing catalysts might result in social conflict (i.e., Homer-Dixon 1991; Kahl 2006). This work stresses the mitigating effects of state capacity, social vulnerability to decreased resources, and barriers to efficient and equitable distribution of emergency resources. But very few cross-national studies are sensitive to these intervening sociopolitical conditions. Failure to account for conditional relationships can bias analyses of direct scarcity-conflict effects toward insignificance, leading scholars to conclude that no relationship exists when, in fact, scarcity-conflict linkages are only obscured by the conflation of vulnerable and invulnerable observations in the sample. The search for a universal effect may overlook salient conditional mechanisms. This study aims to advance our understanding of environmental drivers of political violence by testing for conditional relationships that link civil conflict onset to drought severity in 39 sub-Saharan states (1962–2006). It contributes to a nascent sub-literature that examines conditional causal mechanisms in a cross-national sample (Buhaug 2010; Couttenier and Soubeyran 2013; Fjelde and von Uexkull 2012; O’Loughlin et al. 2012; Raleigh and Urdal 2007; Slettebak 2012; Theisen, Holtermann, and Buhaug 2011–2012). We limit the scope of our project to African civil conflict because other types of violence (cattle raiding, riots, etc.) may have fundamentally different causes (Fjelde and von Uexkull 2012; Meier, Bond, and Bond 2007; Theisen 2012) and civil conflict poses an especially salient and consequential challenge to African political development. Further, most extant work on scarcity-driven conflict examines sub-Saharan Africa (Buhaug 2010; Couttenier and Soubeyran 2013; Hendrix and Glaser 2007; Hendrix and Salehyan 2012; Theisen, Holtermann, and Buhaug 2011–2012), and adopting this scope allows for more direct comparison of our results to previous research. We also follow convention by using climatological drought data to measure variation in resource scarcity. Today, more than one in three sub-Saharan Africans inhabits an area that is prone to severe drought and the United Nations estimates an additional 250–300 million Africans will be adversely affected by drought by 2020 (UNFCCC 2007, 5). Most sub-Saharan states are too poor to adequately respond to the forecasted increase in severe drought, and African economies are particularly susceptible to water scarcity (Müller et al. 2011). Sub-Saharan economies are more dependent upon agricultural production than those of any other region, and African agriculture is uniquely vulnerable because only a very small fraction of the region’s arable land is irrigated (World Bank 2013). Most of sub-Saharan Africa is expected to suffer more drought as a result of global climate change (UNFCCC 2007), and sub-Saharan Africa will also be the largest recipient of international adaptation aid if the UNFCCC National Adaptation Programs of Action are funded completely (UNFCCC 2012b). Any relationship between scarcity-inducing environmental change and large-scale violent conflict could be most apparent when drought strikes sub-Saharan Africa. Our analysis finds no evidence of drought increasing the risk of conflict wherever it occurs, and this corroborates previous research that reports no direct and universal relationship between drought and political violence in sub-Saharan Africa (Buhaug 2010; Couttenier and Soubeyran 2013; O’Loughlin et al. 2012; Theisen 2008). However, we highlight several conditions under which drought becomes a significant predictor of African civil conflict. Surprisingly, severe drought has no significant effect on the risk of civil conflict onset in the states with sociopolitical conditions that favor conflict. These states face a relatively high risk of political violence regardless of recent environmental conditions, and any additional risk posed by severe drought is small and statistically insignificant. To the contrary, we find severe drought to be a risk equalizer. States with greater food security, less ethnopolitical exclusion, democratic governance, superior public health, and longer histories of political stability are undermined by severe drought, and the magnitude of the effect is great enough to offset the stabilizing influences of these sociopolitical conditions. During severe drought, states with sociopolitical attributes that would normally promote peace are no less likely to suffer conflict than states with the weakest governments and most vulnerable populations. This finding, which is robust across several measures of these sociopolitical conditions, contradicts the argument that fragile states with vulnerable populations are at the greatest risk of scarcity-induced political violence. Instead, these results raise important questions about why environmental scarcity might destabilize relatively secure governments and populations. We discuss one plausible explanation that should be explored in future research. Work in psychology and behavioral economics on the “endowment effect” finds people are averse to loss to the point that absolute conditions may be less important than how those conditions compare to the perceived status quo (Kahneman and Tversky 1979; Thaler 1980). This logic was originally used to explain why sellers and buyers sometimes fail to agree upon an appropriate price (Brenner et al. 2007; Carmon and Ariely 2000; Kahneman, Knetsch, and Thaler 1991). Tendencies toward loss aversion cause those who are endowed with something (sellers) to place much greater value on what they possess than those who do not (buyers). This argument has been extended to conflict research to explain how adverse changes from an ex ante reference point (the status quo) can encourage risk-taking and conflict initiation (Levy 1992; Mercer 2005). In this context, endowment theory would expect the hardships caused by scarcity-inducing weather to elicit stronger reactions where the status quo is more comfortable. Drought-induced grievances will be less severe where the population is accustomed to an ineffective state and adverse sociopolitical conditions. Instead, frustration will be higher where a more favorable status quo has conditioned people to expect a higher level of welfare and more effectual government. Drought threatens a more adverse deviation from the ex ante reference point in these contexts, causing the risk of a drought-motivated conflict to be higher than it would be were the status quo less appealing. Literature Review Conditional theories of scarcity and conflict are often rooted in the work of Thomas Homer-Dixon (1991, 1994, 1999, 2000). Using case studies of conflicts that are typically traced to environmental causes, Homer-Dixon concludes conflict is more likely to occur where states are unable to effectively adapt to resource shortages. This thesis is far from deterministic, and scarcity-induced conflict is by no means inevitable. Rather, the argument is that scarcity-inducing events have the greatest destabilizing effects where governments cannot effectively insulate their populations from adverse change. Since the early 1990s, this work has advanced a broader multidisciplinary research program on the sociopolitical consequences of resource scarcity (Adger 1999; Brooks, Adger, and Kelly 2005; O’Brien et al. 2004; Tol and Yohe 2007) and has supported the international community’s recent commitment to international adaptation aid programs for states that are perceived to be especially susceptible. This literature has attracted pointed criticism for advancing complex causal mechanisms, biased case selection, and “theoretical myopia” (Gleditsch 1998, 390–92; 2012, 6–7; Theisen 2008, 813). These concerns give rise to a growing research program that uses cross-national statistics to seek more generalizable relationships, but, with very few exceptions, this work fails to account for conditional mechanisms. When indicators of scarcity are added to conflict models, researchers are testing for universal relationships that are consistent across all observations in the sample. This approach assumes scarcity will have a uniform effect wherever it occurs, and this insensitivity to conditional effects is widely acknowledged in reviews of the quantitative literature (see Barnett and Adger 2007, 644; Gleditsch 1998, 389–90; Salehyan 2008, 321–22). In fact, Gleditsch identifies this shortcoming as the research program’s “greatest weakness” (1998, 389; 2012, 6). A small group of large-n studies links increased abundance—rather than scarcity—to some types of violent conflict (Binningsbø, de Soysa, and Gleditsch 2007; de Soysa 2002; Meier et al. 2007; Theisen 2012). Many papers fail to find any connection between climatic extremes and conflict (Bergholt and Lujala 2012; Buhaug 2010; Theisen, Holtermann, and Buhaug 2011–2012), and others report curvilinear and non-monotonic relationships (Hendrix and Salehyan 2012; O’Loughlin et al. 2012; Raleigh and Kniveton 2012). Some conclude scarcity-inducing weather increases the risk of conflict (Burke et al. 2009; Hauge and Ellingsen 1998; Hsiang, Meng, and Cane 2011; Miguel, Satyanath, and Sergenti 2004), but many of these studies have been scrutinized and dismissed by other researchers over substantial methodological disagreements.4 Drawing inferences from the literature is further complicated by fractionalization into sub-literatures exploring various measures of scarcity (rainfall, temperature, deviation from local historical norms), units of analysis (state, sub-state administrative divisions, or geographic grid cells), and types of civil violence (civil conflict, inter-ethnic violence, communal cattle raiding, nonviolent protests).5 Some recent studies test conditional scarcity-conflict relationships, with most attention being dedicated to sub-state variation in ethnic exclusion and population density. O’Loughlin et al. (2012), Buhaug (2010), and Theisen, Holtermann, and Buhaug (2011–2012) use a sub-state grid cell unit of analysis and report no relationships between local scarcity and local conflict, regardless of the presence of marginalized ethnic groups, poverty, or population pressure within a specific grid cell. Fjelde and von Uexkull (2012), studying a different type of conflict (communal conflict) and unit of analysis (sub-national administrative district), find drought conditions very modestly increase the risk of small-scale communal conflict in districts containing at least one politically marginalized ethnic group. However, this significant result is a notable aberration in a group of studies that reports many non-findings. The absence of significant effects in even the most conflict-prone areas has led several scholars to forcefully reject the possibility that political violence can be “blamed” on drought (Buhaug 2010; Slettebak 2012). Local conditions certainly affect adaptive capacity and social vulnerability, but the literatures on these concepts also stress important state-level determinants (Adger 1999; Brooks, Adger, and Kelly 2005). The ethnic inclusivity of a national government is both empirically and theoretically distinct from the demographics of a sub-state district. Ethnic political dominance can create doubts about the equitable distribution of emergency resources, even in sub-state regions that are relatively homogeneous. The resilience of a state’s institutions is better captured by the statewide presence or absence of destabilizing violence or political change, rather than an indicator of recent conflict within a small, specific grid cell. A state’s efforts to aid villages in peaceful regions can be hindered by insurgencies that demand resources and attention far away, on a state’s loosely controlled frontiers. Further, the movement of effected populations could bring conflict to grid cells where scarcity is less severe. This is especially true in sub-Saharan Africa, where a significant share of the population practices pastoralism and many scarcity-inducing catalysts result in voluntary migration and social conflict in migrant-receiving areas (Barrios, Bertinelli, and Strobi 2006; Reuveny 2007; Salehyan and Gleditsch 2006). For these reasons, state-level determinants, such as state stability, statewide scarcity, and the characteristics of state political institutions, could affect the drought-conflict relationship as much as any local conditions. Our review of the literature uncovered only two published papers that focus on conditional scarcity-civil conflict relationships at the state level of analysis: Slettebak’s (2012) global study of climate-related disasters (including drought) and Couttenier and Soubeyran’s (2013) examination of drought in sub-Saharan Africa. Both papers report statistically significant conditional effects. Slettebak (2012) interacts the state population with a crisis indicator drawn from the Center for Research on the Epidemiology of Disasters (CRED 2013). He concludes, surprisingly, that population increases the pacifying effect of a disaster. States suffering disasters are less likely to suffer civil conflict onset, and the magnitude of this effect grows with state population. Couttenier and Soubeyran (2013) examine ethnic fractionalization, terrain roughness, and democratic institutions, finding the relationship between drought and civil conflict to be negative in homogeneous and non-mountainous states, but positive in diverse states with rugged terrain. Their tests for a conditional relationship between democracy, drought, and civil conflict offer mixed results. Both papers have important methodological shortcomings that drive these reported findings. Slettebak (2012) operationalizes disasters with a measure based on media reports of populations effected by weather events, but unequal media coverage causes disasters to be disproportionately overlooked in lesser-developed regions (Gall, Borden, and Cutter 2009). The United States suffers a qualifying disaster in every year in Slettebak’s sample (1950–2008), while less newsworthy but no less severe events in Africa are not represented in the data. There is also substantial temporal bias in the data; more recent years see more than three times as many disaster events as years before 1980. The oversampling of recent “disasters” in large Western states could explain the finding that disasters decrease the risk of conflict.6 The analysis offered by Couttenier and Soubeyran (2013) utilizes superior drought data, but the results are produced by some unconventional modeling decisions. First, the paper estimates civil conflict onset, which is dichotomous, with linear ordinary least squares regression (Couttenier and Soubeyran 2013, 7). This method is not typically used to estimate models with dichotomous dependent variables because it can generate predicted probabilities that lie outside the unit interval [0,1]. This practice is more common in economics than in political science (i.e., Miguel, Satyanath, and Sergenti 2004), but Beck (2011) has shown ordinary least squares regression to be unsatisfactory relative to logistic regression when independent variables are not distributed normally. Key independent variables such as the measures for drought (leptokurtic), ethnic fractionalization (negatively skewed), and terrain roughness (positively skewed) violate this important condition.7 Additionally, the strongest reported conditional effects—those related to ethnic fractionalization and terrain roughness—are obtained from models that do not include both of the independent variables that comprise the interaction term (see Couttenier and Soubeyran 2013, Table 7, Models 1–4). Because the interaction models omit one of the interacted variables, they may produce biased results (Brambor, Clark, and Golder 2006; Plümper and Troeger 2007). The models also include fixed effects for state and year, but no other control variables. When standard controls for economic performance and political stability are added to the models, the results diverge from those reported in the paper. Theory and Hypotheses Conditional scarcity-conflict arguments emerging from the environmental security literature generally link scarcity to civil conflict via one or more of three mechanisms. As these arguments have yet to be explicitly tested on a large cross-national sample, we proceed by delineating these mechanisms, identifying suitable measures, and proposing testable hypotheses that reflect these arguments. First, scarcity-inducing catalysts such as drought are expected to be more likely to motivate conflict where the population is more vulnerable to adverse change. Conditions that increase social vulnerability, such as poverty, poor public goods provision, and inadequate access to food and clean water, should reduce a population’s ability to tolerate further resource scarcity and therefore increase the risk of instability and conflict (Brooks, Adger, and Kelly 2005; Homer-Dixon 1994; Tol and Yohe 2007). With inadequate means, individuals have less capacity to generate and save surpluses that can reduce vulnerability during times of scarcity. As a result, they make greater demands of their governments during crises and are more likely to migrate to refugee camps and become temporarily dependent upon food aid (Gleditsch, Nordas, and Salehyan 2007; Reuveny 2007). The desperation that results from high social vulnerability can facilitate rebel recruitment by reducing the opportunity costs forgone by those who join anti-government movements. To test this social vulnerability mechanism, we leverage data on two indicators of well-being. Environmental scarcity is closely related to food insecurity, so our first measure, Food Supply, is the Food and Agriculture Organization’s best estimate of per capita daily caloric consumption (FAO 2012). Food Supply is recognized as an ideal indicator of social vulnerability because it is less prone to within-state inequality than measures of income, education, and other public goods (Conrad 2011). Wealthier people tend to not eat many, many times more calories than the poor. Instead, they replace cheaper calories (grains) with high-cost options such as meat and imported delicacies. In this sample, Food Supply averages 2,132 calories per person per day, with extremes at 1,439 (Burkina Faso 1962) and 2,978 (South Africa 2006). We lag this variable by one year to address the bias that could be caused by endogeneity between recent drought conditions and Food Supply. We also test this mechanism with Infant Mortality Rate, a widely adopted and comprehensive measure of well-being that is influenced by nutrition, access to health care, female education, and the availability of sanitary food and water. We use data from the World Bank (2013) and Abouharb and Kimball (2007) to create a time-variant measure that equals the number of children per 1,000 live births who perish before reaching their second birthday. Southern countries like Botswana, Zimbabwe, and South Africa have experienced periods with relatively low rates of less than 50 per 1,000, while Infant Mortality Rate peaked just below 300 per 1,000 in Mozambique during its long civil war. Across roughly 1,600 country-year observations, the mean Infant Mortality Rate was 118 deaths per 1,000 live births. Social Vulnerability Hypotheses H1a: Drought most increases the risk of conflict where caloric consumption is low. H1b: Drought most increases the risk of conflict where infant mortality is high. The second conditional mechanism emphasizes the politics of resource distribution during times of crisis. Scarcity can be especially dangerous when access to remaining resources is perceived to be unequal (Homer-Dixon 1994), and many case studies find these perceptions to be more prevalent when diverse populations do not have comparable representation in the government. Kahl (1998, 2006) argues that preexisting ethnopolitical tensions were exacerbated during environmental crises in Kenya. Similarly, Raleigh (2010) studies the Sahel and stresses that diversity and drought are most likely to result in conflict where ethnic groups are also politically marginalized. Diversity increased perceptions of government favoritism and may have fueled conflict during Ethiopian resource shortages Martin (2005) and the Tuareg rebellion in Mali (Benjaminsen 2008). When scarcity causes migration, this migration is most likely to result in conflict between the migrant and host communities when these groups lack ethnic similarities (Gleditsch 2007; Reuveny 2007; Salehyan and Gleditsch 2006). Conversely, homogeneity can facilitate peace during periods of increased scarcity. Bogale and Korf (2007) find that regional homogeneity in some parts of Ethiopia helped communities arrange local resource-sharing agreements to reduce the effects of acute shortages. Shared identity may also decrease perceptions of relative deprivation (Bohle, Downing, and Watts 1994; Reardon and Taylor 1996) and help afflicted populations foster unity and collective identity during a community crisis (see Slettebak 2012). We hypothesize that scarcity will be most likely to result in conflict when it exacerbates pre-existing feelings of relative deprivation among salient ethnic groups with unequal access to political power. We test this argument with a measure of ethnic diversity that accounts for whether a group also suffers political marginalization. Ethnic Exclusion represents the share of a state’s population that belongs to ethnic groups that are excluded from political power. This information is compiled in the Ethnic Power Relations data set (Cederman, Wimmer, and Min 2010). The average of this time-variant variable is 23 percent, but approximately 43 percent of state-years have no exclusion whatsoever. Some of the highest observed values belong to 1970s Liberia (98 percent), 1960s–1970s Zimbabwe/Rhodesia (97 percent), and post-1994 Rwanda (84 percent). We expect the drought-conflict relationship to be weakly positive or insignificant when Ethnic Exclusion is low, but the magnitude of the correlation should increase with this variable. Exclusive forms of government promote favoritism and patronage and may increase perceptions of unequal resource allocation even where exclusion is not exacerbated by ethnic diversity. The leaders of democracies must attract a broader coalition to remain in power (Bueno de Mesquita et al. 2003), so populations living under democratic governments should expect state resources to be allocated more equitably to a greater share of the population. Because non-democratic leaders do not face electoral constraints and stay in power by pleasing a narrower coalition, leaders have substantially less incentive to allocate resources to a large share of the population during periods of extreme resource scarcity. We test this mechanism with Polity Score, which is the literature’s most frequently used measure of democratization (Marshall and Jaggers 2002). This 21-value ordinal scale (–10 to 10) increases with a state’s level of democratization. We use the Vreeland (2008) “X Polity” version of the data, which is optimized for use in time-series analysis. Sub-Saharan Africa trended slightly non-democratic in the overall sample (mean of –1.02), though Africa is much more democratic since 2000 (mean of +1.33) and includes several states (e.g., South Africa, Lesotho) that experienced periods of full-fledged democracy (Polity Score > 6). Equitable Distribution Hypotheses H2a: Drought most increases the risk of conflict where ethnic exclusion in high. H2b: Drought most increases the risk of conflict where democratization is low. A third mechanism emphasizes state capacity, defined as a government’s ability to mitigate the consequences of adverse environmental change. Whereas social vulnerability describes a population’s exposure to scarcity, capacity refers to the resources, expertise, and institutional stability that allow states to alleviate the impact of increased scarcity (Bohle, Downing, and Watts 1994; Busby et al. 2013; Reardon and Taylor 1996). Governments lacking adaptive capacity are less likely to possess the emergency resources they need to provide a sufficient response to increased scarcity, and citizens are less likely to believe the government will meet their needs (Homer-Dixon 1994; Reuveny 2007). Local norms of resource-sharing and social support networks can compensate for government shortcomings to some extent, but state support is often needed to provide security and supplement any local efforts with imported food and resources.8 Perceptions of government incompetence or indifference during crisis can then motivate rebel recruitment. There is a large literature on measuring adaptive capacity, and the most comprehensive juxtaposition of proposed measures finds political stability to be among the most important determinants of a government’s ability to act during environmental crises (Brooks, Adger, and Kelly 2005). During crises, stable governments are better able to provide security, manage emergency resource distribution, and adapt to internal migration (Reuveny 2007). Severe drought should be more likely to motivate civil conflict where states have been stable for a shorter period of time. Stability decreases the likelihood of civil conflict (Fearon and Laitin 2003; Hegre et al. 2001), increases the opportunity costs of rebellion (Collier and Hoeffler 2004), and allows for institutional consolidation (Huntington 1968). We test this mechanism with two measures of stability. Peace Years counts the number of years since either independence or the last year of civil conflict incidence (≥ 25 deaths) as recorded by the Uppsala Conflict Data Program Armed Conflict Dataset (Themnér and Wallensteen 2012). The average observation had 12 years of peace, while Tanzania reached nearly 50 years without suffering a qualifying incident of civil conflict. We also test for a conditional relationship with Political Stability, which is the duration of the state’s political system in years according to the Polity IV data set (Marshall and Jaggers 2002). Constitutional alternations of power between political parties or regime insiders do not constitute regime failures, but this measure will reset following unconstitutional power transitions and abrupt changes that result in an immediate three-point shift in Polity Score. It takes a value of zero during prolonged periods of instability or anarchy (e.g., Liberia 1990–1997). Beyond stability, state capacity can be influenced by the location of the population in need. Dispersed, rural populations are more difficult to reach, so states with lower levels of urbanization face additional challenges during scarcity-induced crises. To capture these increased demands, we test the conditional role of Urbanization, measured as the share of the population that lives in urban centers with at least 100,000 residents. Urbanization alone would make a poor proxy of state adaptive capacity, but among states with similar resources and capabilities, those with more urban populations should have greater capacity to adapt to increased scarcity than those with rural and dispersed populations. Therefore, we can shed light on the state capacity mechanism by testing for the conditional effect of urbanization while controlling for other determinants of state strength. State Capacity Hypotheses H3a: Drought most increases the risk of conflict where past civil conflict is more recent. H3b: Drought most increases the risk of conflict where political instability is more recent. H3c: Drought most increases the risk of conflict where urbanization is low. Drought Data and Research Design We test the hypotheses on a state-year sample of all sub-Saharan states for which data are available, 1960–2009. Due to some temporal and spatial constraints on the control variables, models estimated with every variable described below cover 39 states for the years 1962–2006 (1,579 observations).9 The dependent variable is dichotomous, so we utilize logistic regressions with robust standard errors clustered by state. We also describe several alternate specifications and estimators and provide the results in a supplemental appendix.10 Like much of the literature on environmental causes of civil conflict, our dependent variable is Conflict Onset, as reported by the Uppsala Conflict Data Program Armed Conflict Dataset, version 4-2012 (Gleditsch et al. 2002; Themnér and Wallensteen 2012). This dichotomous conflict indicator is equal to 1 in any year in which a new and sufficiently deadly (25 battle fatalities per year) civil conflict is initiated. States can combat multiple groups at one time, so states can suffer new onsets even while fighting ongoing conflicts with other domestic actors. Following the literature, new onsets are recorded only if relations between the state and the group in question are peaceful for at least two years preceding the onset. Otherwise, fighting is coded as an ongoing conflict and does not constitute a new onset. The frequency of Onset in the non-missing data is 66 civil conflict onsets in 32 states. We measure drought with PDSI, which is an original adaptation of the Palmer Drought Severity Index. This index was developed by climate scientists affiliated with the United States National Center for Atmospheric Research (Dai, Trenberth, and Qian 2004; Dai 2011; Palmer 1965, 1968) and is a leading measure of soil moisture in climatological research. PDSI accounts for several relevant determinants of drought conditions, including not only precipitation but also surface net radiation, humidity, wind speed, and air pressure. This is a strength of PDSI, as drought indices that rely only on precipitation are only able to capture surpluses or deficits of rainfall, but miss additional critical features that are relevant to agriculture, including the accumulation of water in the soil column and evaporative demand of the atmosphere (Guttman 1998; McKee, Doeskin, and Kleist 1993; Trenberth et al. 2014).11 The PDSI is calibrated for local relative sensitivity, allowing PDSI scores to reflect to local climate variability. Additionally, local sensitivity is used in a climatological sense, rather than a societal sense. In other words, sensitivity is driven by climatological variables (i.e., temperature, precipitation, evapotranspiration, humidity, etc.) and not human-influenced variables (e.g., irrigation water use, pollution, agricultural harvest, etc.) that would make it endogenous to human activity. A further advantage of PDSI is that drought conditions are triangulated via many different data sources. This allows researchers to estimate drought conditions with greater accuracy over a greater temporal period (Guttman 1998; McKee, Doeskin, and Kleist 1993). The result is an accurate and comprehensive measure of dryness that ranges from –10 (extremely dry) to +10 (extremely wet). Monthly PDSI scores are normalized to long-term local averages at a resolution of 2.5 degrees latitude by 2.5 degrees longitude. Data cover the entire post-independence period and extend to the early twentieth century for many African states. To transform PDSI for the state-year level of analysis, we first calculate an area-weighted value for each state-month. The PDSI score of each 2.5-by-2.5-degree cell is multiplied by the percentage of a state’s territory falling within that cell, and these products are then summed by state to create area-weighted averages at the state-month level of analysis. We aggregate from month to year by averaging the 12 state-month values that correspond to each year. Finally, to generate a measure that is better suited for comparison across states, we normalize these annual PDSI scores to long-term state averages (see Figure 1). The result is a continuous annual measure of statewide drought conditions where 0 represents average conditions, 1 represents soil moisture that is exactly one standard deviation above that state’s average conditions, and –1 represents soil moisture that is one standard deviation less than average conditions.12 Figure 1. View largeDownload slide Mean PDSI scores by state, 1960–2010 Note: The bars show the minimum and maximum PDSI observed, 1960–2010. Long-term averages are calculated with data reaching back to the early twentieth century, so the mean of the normalized is not always equal to zero. Each grid line marks one standard deviation. Darker states have lower (drier) PDSI averages. Figure 1. View largeDownload slide Mean PDSI scores by state, 1960–2010 Note: The bars show the minimum and maximum PDSI observed, 1960–2010. Long-term averages are calculated with data reaching back to the early twentieth century, so the mean of the normalized is not always equal to zero. Each grid line marks one standard deviation. Darker states have lower (drier) PDSI averages. Next, we compare our state-level PDSI index to several popular alternatives: GPCP precipitation anomaly data (e.g. Hendrix and Salehyan 2012), CRED data on media reports of drought-afflicted populations (e.g. Slettebak 2012), and the SPI precipitation index that is often collapsed into a sub-state-level ordinal variable (e.g., Theisen, Holtermann, and Buhaug 2011–2012). Table 1 is a correlation matrix that provides evidence of the considerable differences between these measures. The GPCP data are collected by NASA’s Global Precipitation Climatology Project and measure precipitation anomalies as deviations from historical norms. The result is a continuous but highly variable measure of drought that can oscillate between extremely wet and extremely dry conditions relatively quickly. The index does not account for other meteorological phenomena that determine drought conditions, and it is available over a much shorter temporal period relative to PDSI. The Standard Precipitation Index (SPI), presented here in a continuous and state-year format, similarly tracks rainfall anomalies and correlates highly with the GPCP (see Guttman 1998). Unfortunately, most conflict studies adopting the SPI transform it into a dichotomous or four-value ordinal variable that can be used at the sub-state level. Geographic precision is traded for a substantial loss of variation. Finally, the CRED data on media reports of drought-afflicted populations are not correlated with the SPI, GPCP, or PDSI data. This provides evidence of the reporting bias that is likely to be especially pervasive in underdeveloped regions of sub-Saharan Africa. Table 1. Correlation Matrix of Drought Data Used in African Conflict Research   PDSI  GPCP  CRED  GPCP  0.607      CRED  –0.053  –0.112    SPI  0.344  0.566  –0.061    PDSI  GPCP  CRED  GPCP  0.607      CRED  –0.053  –0.112    SPI  0.344  0.566  –0.061  Note: GPCP data are produced by NASA’s Global Precipitation Climatology Project and are derived from the replication data of Hendrix and Salehyan (2012). Slettebak’s (2012)CRED variable is collected by the Center for Research on the Epidemiology of Disasters and found in his replication files. SPI is continuous and was downloaded from the website of the International Research Institute for Climate and Society at Columbia University. View Large Each model includes a number of control variables that have been linked to African civil conflict in previously published research (see Table 2). Prior work emphasizes conditions that facilitate the organization and continuation of an insurgency. Following Fearon and Laitin (2003) and Couttenier and Soubeyran (2013), we consider Mountains, which is the logged percentage of a state’s land occupied by rough, mountainous terrain. States with rougher terrain have been shown to be more conflict-prone because inaccessible areas offer safe havens to insurgents and inhibit a state’s efforts to reach isolated areas. Fearon and Laitin (2003) use a similar logic to link civil conflict to low levels of Urbanization, which we operationalize as the share of the population living in towns of more than 100,000 residents.13 Opportunities for rebel recruitment are proxied by Population, which should have a positive correlation with conflict (Bruckner 2010), and per capita GDP, which is negatively associated with conflict (Collier and Hoeffler 2004). Both measures are logged and drawn from the Penn World Table, version 7.1 (Heston, Summers, and Aten 2012). There are fewer opportunities for new, distinct conflicts where states are fighting ongoing conflicts already, so we control for Ongoing Conflict as indicated by the UCDP Armed Conflict Dataset. To separate the immediate effects of drought from lagged effects, we also include PDSI at t – 1. Table 2. Descriptive Statistics   Min  Max  Mean  St. Dev.  Source  Onset  0  1  0.04  0.20  Themner & Wallensteen (2012)  PDSI  –4.65  3.21  –0.35  1.11  Dai (2011)  Food Supply  1,439  2,978  2,139  295  FAO (2012)  Inf. Mortality  39  237  111  38  Abouharb and Kimball (2007)  Eth. Exclusion  0  0.98  0.23  0.30  Cederman, Wimmer, and Min (2010)  Polity Score  –6  7  –1.25  3.71  Vreeland (2008)  Peace Years  0  46  12  11  Themner and Wallensteen (2012)  Pol. Stability (ln)  0  4.66  1.98  1.12  Marshall and Jaggers (2002)  Urbanization  0  0.58  0.11  0.10  Singer, Bremer, and Stuckey (1972)  Pop (ln)  6.02  11.83  8.64  1.17  Heston, Summers, and Aten (2012)  pc GDP (ln)  3.49  9.24  6.26  0.93  Heston, Summers, and Aten (2012)  Ong. Conflict  0  1  0.19  0.39  Themner and Wallensteen (2012)  Mountains (ln)  0  4.42  1.56  1.48  Fearon and Laitin (2003)    Min  Max  Mean  St. Dev.  Source  Onset  0  1  0.04  0.20  Themner & Wallensteen (2012)  PDSI  –4.65  3.21  –0.35  1.11  Dai (2011)  Food Supply  1,439  2,978  2,139  295  FAO (2012)  Inf. Mortality  39  237  111  38  Abouharb and Kimball (2007)  Eth. Exclusion  0  0.98  0.23  0.30  Cederman, Wimmer, and Min (2010)  Polity Score  –6  7  –1.25  3.71  Vreeland (2008)  Peace Years  0  46  12  11  Themner and Wallensteen (2012)  Pol. Stability (ln)  0  4.66  1.98  1.12  Marshall and Jaggers (2002)  Urbanization  0  0.58  0.11  0.10  Singer, Bremer, and Stuckey (1972)  Pop (ln)  6.02  11.83  8.64  1.17  Heston, Summers, and Aten (2012)  pc GDP (ln)  3.49  9.24  6.26  0.93  Heston, Summers, and Aten (2012)  Ong. Conflict  0  1  0.19  0.39  Themner and Wallensteen (2012)  Mountains (ln)  0  4.42  1.56  1.48  Fearon and Laitin (2003)  View Large To address the temporal dependence that often occurs in time-series analysis, we add cubic polynomial terms (t, t2, and t3 to each model where t = Year – 1960). Carter and Signorino (2010) demonstrate the efficiency of this strategy relative to time fixed effects, though we use this alternate approach in robustness testing. We also lag all time-variant variables by one year and test the robustness of each model using general estimating equation (GEE) estimation, which is more efficient than fixed effects regression when the number of panels (39 states) approaches or exceeds the number of observations in each panel (45 years) (Zorn 2001). Our models do not include state fixed effects because this would exclude from the sample any countries that do not experience an onset during the time period analyzed. This reduces variation and creates significant sample selection bias; countries without a history of political violence are eliminated from the sample, thereby concentrating the analysis only on within-panel variation in the most conflict-prone states. Results The following results tables provide the odds ratios and robust panel-clustered standard errors estimated with logistic regression. Odds ratios greater than 1 indicate that an increase in the independent variable raises the risk of conflict, while odds ratios less than 1 occur when higher values of the independent variable reduce the risk of conflict. There are few statistically significant odds ratios for the constituent terms of each interaction model, but it is well established that we can infer very little about interaction effects on limited dependent variables from these values (e.g., Brambor, Clark, and Golder 2006). Even where constituent terms of the interaction do not have an independent and significant effect, significant results may occur. We therefore evaluate our hypotheses by calculating marginal effects with 90 percent confidence intervals. For each model, we hold control variables at their means14 and generate predicted probabilities across the ranges of PDSI and the variable of interest. The marginal effects illustrated in the figures are equal to the change in the predicted probability of Onset that results from a one-standard-deviation decrease in PDSI. At values of the conditional variable at which the entire confidence interval lies above or below zero, there is a statistically significant relationship between drought severity and civil conflict onset. Model 1 (Table 3) tests for an unconditional relationship between PDSI and Onset. We find no evidence of an unconditional relationship, and this is congruent with several other recent studies (Buhaug 2010; Theisen, Holtermann, and Buhaug 2011–2012). Several control variables predict Onset, and there is little variation in these predictions across the models. Mountains, Urbanization, Ethnic Exclusion, and Population (ln) increase the likelihood of civil conflict. Model 1 does not find a significant link between pc GDP (ln) and conflict, but this may be due to collinearity with Food Supply (Pearson’s r = .402) and Peace Years (Pearson’s r = .334). When these variables are excluded from the model, pc GDP (ln) has the expected negative effect on conflict onset. Onsets are also less likely where Food Supply is higher and states are already fighting ongoing civil conflicts. Table 3. Unconditional Link and Social Vulnerability Hypotheses   Model 1  Model 2 (H1a)  Model 3 (H1b)    No Conditions  Food Supply  Infant Mortality  PDSI  0.869    (0.103)  1.815    (1.462)  0.694*    (0.149)  Interaction Term        1.000    (0.000)  1.002    (0.002)  Food Supply  0.998***    (0.001)  0.998***    (0.001)        Infant Mortality              1.008*    (0.004)  PDSI (t – 1)  1.091    (0.138)  1.086    (0.139)  1.062*    (0.130)  Ethnic Exclusion  3.591**    (1.970)  3.575**    (1.971)  2.774*    (1.465)  Peace Years  0.996    (0.016)  0.996    (0.016)  0.991    (0.015)  Urbanization  39.741***    (39.325)  40.865***    (40.204)  49.699***    (65.697)  Population (ln)  1.252*    (0.145)  1.256**    (0.143)  1.204    (0.137)  pc GDP (ln)  0.974    (0.150)  0.971    (0.150)  0.790*    (0.110)  Mountains (ln)  1.202**    (0.106)  1.203**    (0.139)  1.294**    (0.131)  Ongoing Conflict  0.390**    (0.186)  0.386**    (0.184)  0.446*    (0.187)  Constant  0.243    (0.278)  0.329    (0.359)  0.003***    (0.004)  N      1,579      1,579      1,657  LL      –253.93      –253.52      –277.18    Model 1  Model 2 (H1a)  Model 3 (H1b)    No Conditions  Food Supply  Infant Mortality  PDSI  0.869    (0.103)  1.815    (1.462)  0.694*    (0.149)  Interaction Term        1.000    (0.000)  1.002    (0.002)  Food Supply  0.998***    (0.001)  0.998***    (0.001)        Infant Mortality              1.008*    (0.004)  PDSI (t – 1)  1.091    (0.138)  1.086    (0.139)  1.062*    (0.130)  Ethnic Exclusion  3.591**    (1.970)  3.575**    (1.971)  2.774*    (1.465)  Peace Years  0.996    (0.016)  0.996    (0.016)  0.991    (0.015)  Urbanization  39.741***    (39.325)  40.865***    (40.204)  49.699***    (65.697)  Population (ln)  1.252*    (0.145)  1.256**    (0.143)  1.204    (0.137)  pc GDP (ln)  0.974    (0.150)  0.971    (0.150)  0.790*    (0.110)  Mountains (ln)  1.202**    (0.106)  1.203**    (0.139)  1.294**    (0.131)  Ongoing Conflict  0.390**    (0.186)  0.386**    (0.184)  0.446*    (0.187)  Constant  0.243    (0.278)  0.329    (0.359)  0.003***    (0.004)  N      1,579      1,579      1,657  LL      –253.93      –253.52      –277.18  Note: Table shows odds ratios and robust standard errors. Models include t, t2, and t3. Control variables are lagged by one period. PDSI decreases with dry conditions. *p < .1; **p < .05; ***p < .01. View Large Models 2 and 3, which examine the social vulnerability hypotheses, find significant and surprising conditional effects (see Figure 2). In stark contrast to Hypothesis 1a, these results suggest drought increases the risk of civil conflict only where Food Supply is high. When control variables are held at their means, drought does not raise the risk of conflict in countries where people consumed less than approximately 2,050 calories per day in the previous year. Instead, drought is most likely to bring conflict to states with greater food security. The negligible slope that occurs above 2,050 calories per day indicates states above this threshold are similarly threatened by severe drought, with a shift from normal soil moisture (PDSI = 0) to one standard deviation below normal soil moisture (PDSI = –1), resulting in a .006-point increase in the risk of conflict. This is a small shift, but because Onset is relatively infrequent in the sample this means conflict is, all else being equal, roughly 50 percent more likely in a food-secure state that is one standard deviation drier than normal relative to a food-secure state that is one standard deviation wetter than normal. Figure 2. View largeDownload slide Effect of drought on conflict by social vulnerability Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Figure 2. View largeDownload slide Effect of drought on conflict by social vulnerability Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. This counterintuitive conditional relationship is supported by Model 3, which reveals a similar link between drought, conflict, and infant mortality. The social vulnerability literature would expect scarcity-conflict linkages to be strongest in societies with high infant mortality rates, but this test reveals the opposite relationship (see Figure 2). Above approximately 90 infant deaths per 1,000 live births, there is no relationship between drought and civil conflict onset. But in states with lower infant mortality rates, drought has a significant destabilizing effect. With controls at their means and Infant Mortality Rate at 60—a rate attained by states such as South Africa, Namibia, Senegal, and Gabon—a one-standard-deviation worsening in drought conditions increases the chance of conflict by 20 percent. The results of Models 2 and 3 hold when Food Supply and Infant Mortality Rate are replaced by logged versions of these variables. Next, we turn to our tests of the hypotheses related to equality of access to emergency resources (Table 4). The adaptive capacity literature expects conflict to be more likely where access to resources is perceived to be unequal or discriminatory, but the results do not support this expectation. Rather, the results echo the surprising findings presented above. Model 4 tests Hypothesis 2a and finds that the drought-conflict relationship is significant only where Ethnic Exclusion is low. The .776 odds ratio indicates that where none of the population faces ethnic exclusion, a one-standard-deviation improvement in PDSI decreases the likelihood of conflict by nearly 25 percent, ceteris paribus. The effect is illustrated in Figure 3. As Ethnic Exclusion increases, PDSI has a weaker effect on conflict. Above approximately 12 percent, this drought-conflict relationship is no longer statistically significant. This contradicts Hypothesis 2a but is similar to the finding that drought poses the greatest risk of increased conflict in states with lower levels of social vulnerability. Figure 3. View largeDownload slide Effect of drought on conflict by equality of access Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Figure 3. View largeDownload slide Effect of drought on conflict by equality of access Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Table 4. Equal Access Hypotheses   Model 4 (H2a)  Model 5 (H2b)  Model 6 (H2b)    Ethnic Exclusion  Democracy (All)  Democracy (Stable)  PDSI  0.776**  (0.097)  0.866  (0.101)  0.751**  (0.099)  Interaction Term  1.413  (0.401)  0.985  (0.021)  0.957  (0.026)  Ethnic Exclusion  4.118**  (2.432)          Polity Score      1.055  (0.039)  0.977  (0.057)  PDSI (t – 1)  1.085  (0.140)  1.091  (0.139)  0.953  (0.159)  Food Supply  0.998***  (0.001)  0.999**  (0.001)  0.999*  (0.001)  Peace Years  0.996  (0.016)  0.992  (0.017)  1.001  (0.022)  Urbanization  38.143***  (37.139)  10.363**  (10.425)  3.862  (3.643)  Population (ln)  1.265**  (0.144)  1.233  (0.116)  1.308**  (0.173)  pc GDP (ln)  0.995  (0.149)  0.949  (0.225)  1.394  (0.349)  Mountains (ln)  1.196**  (0.102)  1.141  (0.099)  1.159  (0.106)  Ongoing Conflict  0.395**  (0.186)  0.477  (0.258)  0.308*  (0.209)  Constant  0.203  (0.235)  0.408  (0.454)  0.034  (0.070)  N    1,606    1,606    1,171  LL    –253.47    –253.95    –155.99    Model 4 (H2a)  Model 5 (H2b)  Model 6 (H2b)    Ethnic Exclusion  Democracy (All)  Democracy (Stable)  PDSI  0.776**  (0.097)  0.866  (0.101)  0.751**  (0.099)  Interaction Term  1.413  (0.401)  0.985  (0.021)  0.957  (0.026)  Ethnic Exclusion  4.118**  (2.432)          Polity Score      1.055  (0.039)  0.977  (0.057)  PDSI (t – 1)  1.085  (0.140)  1.091  (0.139)  0.953  (0.159)  Food Supply  0.998***  (0.001)  0.999**  (0.001)  0.999*  (0.001)  Peace Years  0.996  (0.016)  0.992  (0.017)  1.001  (0.022)  Urbanization  38.143***  (37.139)  10.363**  (10.425)  3.862  (3.643)  Population (ln)  1.265**  (0.144)  1.233  (0.116)  1.308**  (0.173)  pc GDP (ln)  0.995  (0.149)  0.949  (0.225)  1.394  (0.349)  Mountains (ln)  1.196**  (0.102)  1.141  (0.099)  1.159  (0.106)  Ongoing Conflict  0.395**  (0.186)  0.477  (0.258)  0.308*  (0.209)  Constant  0.203  (0.235)  0.408  (0.454)  0.034  (0.070)  N    1,606    1,606    1,171  LL    –253.47    –253.95    –155.99  Note: Table shows odds ratios and robust standard errors. Models include t, t2, and t3. Control variables are lagged by one period. PDSI decreases with dry conditions. *p < .1; **p < .05; ***p < .01. View Large Models 5–6 replace Ethnic Exclusion with the Polity Score measure of democratization. Hypothesis 2b expects a drought-conflict connection only in exclusive non-democratic states, but again, results strongly suggest that drought increases the risk of conflict only where Exclusion is relatively low. The results produced by Model 5 (center plot of Figure 3) show that drought has no effect on conflict onset in non-democratic regimes, but this effect increases with democratization. The effect just misses the conventional threshold for statistical significance, but this may be because many of the democracies in the sample are transitional and fears of exclusion may persist even after very recent constitutional changes. Model 6 replicates Model 5 on a sub-sample of states that have had stable political systems for three years. The predicted magnitude of the drought-conflict relationship is very similar to the predictions of Model 5, though the confidence intervals tighten and the counterintuitive conditional relationship becomes statistically significant. In Africa’s most democratic states, a one-standard-deviation worsening of drought conditions increases the risk of war by nearly 1.5 percentage points, or roughly 45 percent. Table 5 reports results for the tests of the three hypotheses pertaining to state adaptive capacity. None of the hypothesized effects was supported. Once again, each of the three models produced statistically significant effects in the opposite direction. The literature on adaptive capacity links state capacity to a state’s history of conflict, claiming that more stable states should be more able to mitigate any adverse effects of scarcity-inducing environmental events. But Model 7, as illustrated in the leftmost plot of Figure 4, shows no drought-conflict relationship in states with very recent histories of violence. With controls at their means, PDSI has a significant effect on the risk of Onset only among states that have experienced more than a decade of peace. Stable states are undermined by drought; the risk of conflict increases 20 percent when PDSI drops by one standard deviation in states that have been peaceful for 15 years. The same change in PDSI raises the risk of conflict by 25 percent in states that have been peaceful for 30 years. Contrary to our expectations, the states that should be most capable are those that experience the greatest change in the risk of war during drought. Figure 4. View largeDownload slide Effect of drought on conflict by state capacity Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Figure 4. View largeDownload slide Effect of drought on conflict by state capacity Note: Each figure includes 90 percent confidence intervals around the first difference estimates. Dashed line segments indicate p < .10, meaning the drought/conflict relationship is statistically insignificant at that value of the conditional variable. Thin solid lines appear when .05 < p < .10 (90–95 percent confidence in a conditional relationship), and thick solid lines with endpoints appear when p < .05 (95 percent confidence). The 10th and 90th percentiles bound the illustrated ranges of the conditional variables. Table 5. State Capacity Hypotheses   Model 7 (H3a)  Model 8 (H3b)  Model 9 (H3c)    Peace Years  Political Stability  Urbanization  PDSI  1.043  (0.191)  0.970  (0.157)  0.939  (0.169)  Interaction Term  0.984  (0.012)  0.927  (0.069)  0.572  (0.342)  Urbanization  40.93***  (40.85)  36.17***  (39.56)  29.86***  (29.93)  Peace Years  0.985  (0.019)          Political Stability (l      0.822  (0.119)      PDSI (t-1)  1.096  (0.138)  1.091  (0.141)  1.090  (0.136)  Food Supply  0.998***  (0.001)  0.998***  (0.001)  0.998***  (0.001)  Ethnic Exclusion  3.561**  (1.970)  3.886**  (2.138)  3.618**  (2.002)  Population (ln)  1.254**  (0.145)  1.198  (0.118)  1.257*  (0.150)  pc GDP (ln)  0.981  (0.151)  1.114  (0.180)  0.974  (0.150)  Mountains (ln)  1.203**  (0.108)  1.209*  (0.120)  1.194**  (0.104)  Ongoing Conflict  0.382**  (0.181)  0.379  (0.159)  0.418**  (0.163)  Constant  0.198  (0.238)  0.125  (0.154)  0.237  (0.277)  N    1,579    1,562    1,579  LL    –252.91    –    –253.82    Model 7 (H3a)  Model 8 (H3b)  Model 9 (H3c)    Peace Years  Political Stability  Urbanization  PDSI  1.043  (0.191)  0.970  (0.157)  0.939  (0.169)  Interaction Term  0.984  (0.012)  0.927  (0.069)  0.572  (0.342)  Urbanization  40.93***  (40.85)  36.17***  (39.56)  29.86***  (29.93)  Peace Years  0.985  (0.019)          Political Stability (l      0.822  (0.119)      PDSI (t-1)  1.096  (0.138)  1.091  (0.141)  1.090  (0.136)  Food Supply  0.998***  (0.001)  0.998***  (0.001)  0.998***  (0.001)  Ethnic Exclusion  3.561**  (1.970)  3.886**  (2.138)  3.618**  (2.002)  Population (ln)  1.254**  (0.145)  1.198  (0.118)  1.257*  (0.150)  pc GDP (ln)  0.981  (0.151)  1.114  (0.180)  0.974  (0.150)  Mountains (ln)  1.203**  (0.108)  1.209*  (0.120)  1.194**  (0.104)  Ongoing Conflict  0.382**  (0.181)  0.379  (0.159)  0.418**  (0.163)  Constant  0.198  (0.238)  0.125  (0.154)  0.237  (0.277)  N    1,579    1,562    1,579  LL    –252.91    –    –253.82  Note: Table shows odds ratios and robust standard errors. Models include t, t2, and t3. Control variables are lagged by one period. PDSI decreases with dry conditions. *p < .1; **p < .05; ***p < .01. View Large We would also expect unstable, transitional political systems (H3b) and rural, dispersed populations (H3c) to thwart states’ efforts to effectively adapt to environmental scarcity, yet the results suggest the opposite is true. The effect produced by Model 8 is weaker than that produced by Model 7, though this may be unsurprising because civil conflict can be more disruptive than other forms of political change. There is no link between PDSI and Onset where states have had less than five years of Political Stability, but drought increases the risk of war where political systems are more enduring. Most sub-Saharan states have very large rural populations, but drought has the greatest marginal effect on the risk of conflict in the more urbanized states. States that are 25 percent urban see a 17 percent increase in the risk of war when drought worsens by one standard deviation, yet drought has no discernible effect in very rural states. Discussion The results and robustness tests15 do not support the idea that drought amplifies the risk of conflict in states that are made susceptible by weak state capacity, inequality, or poor human welfare. In fact, the preponderance of the models evince the opposite: drought most destabilizes states with superior living conditions, greater food security, longer-enduring governments, more inclusive political systems, small rural populations, and longer histories of peace. This effect is strong enough to completely offset the pacifying effects that these conditions provide during normal weather conditions. In other words, drought is a conflict risk equalizer. During drought, states that would otherwise be the most prone to civil conflict are no more likely to see violence than states that are otherwise exceptionally unlikely to experience conflict. These results leave us with an important puzzle with pressing policy implications. Why would drought have a greater effect on the risk of conflict where political violence is less likely during normal weather conditions, and what might this mean for international efforts to insulate developing states from the adverse effects of climate change? Answering these questions will require further research, but we believe the answer may be found in the behavioral economics literature on the endowment effect (Kahneman and Tversky 1979; Thaler 1980). According to endowment theory, the value people place on objects and experiences is subjective and is biased by perceptions of the status quo, which serve as psychological reference points during opportunities for loss or gain. People tend to place great value on what they already possess, even to the point that they will accept irrational risks to prevent adverse change. A classic and oft-replicated experiment provides a simple illustration of this effect. Test subjects are divided into two groups. One group is given mugs, and the other is given candy. When the subjects are encouraged to trade these goods, researchers find that possession causes both mug-possessors and candy-possessors to demand much higher prices than their potential buyers are willing to accept. Even when the period of possession is very brief and subjects spend nothing to acquire their goods, endowment with these objects causes measurable loss aversion (Brenner et al. 2007; Carmon and Ariely 2000; Kahneman, Knetsch, and Thaler 1991). Exchanges of mugs and candy are somewhat trivial, but the same psychological bias can influence conflict behavior. In this context, reactions to an absolute condition like drought-induced scarcity should depend upon how one’s welfare during drought compares to one’s welfare during an ex ante reference point. Where welfare is already very poor, increased scarcity does not generate the same sense of loss that it does where welfare was once much more comfortable. We see this endowment effect in the results. All of the interaction models show that drought has no effect on the risk of conflict when sociopolitical conditions are already very unfavorable (high infant mortality, food insecurity, political instability, etc.). Rather, reactions to increased scarcity are strongest among those who perceive themselves to have more to lose. Endowment with a comfortable level of welfare in the ex ante status quo increases loss aversion, and this loss aversion can motivate political violence. We are not the first to apply work in this area to studies of political conflict. This endowment mechanism shares a common foundation with the International Relations literature on prospect theory and risk acceptance in the “domain of losses” (Levy 1992; Mercer 2005). Levy (1992) notes, for example, that wars sometimes occur because states that possess a contested territory are willing to fight to keep it even where they are militarily outmatched and a negotiated concession might be more advantageous. A weaker state facing territorial loss will fight a much stronger state because concession would represent an adverse change from the ex ante reference point. The resulting loss aversion drives the weak state to overvalue the territory and compels it to engage in an unwise armed conflict. This can also explain why declining world powers, such as Russia, may be more belligerent in their efforts to prolong their sphere of influence during a period of contraction. This logic is counterintuitive, but it is supported by the history of drought and civil conflict in sub-Saharan Africa. Populations that are accustomed to scarcity and ineffective government are no more likely to engage in civil conflict when drought threatens prolonged scarcity and poor welfare. But the threat of great loss relative to the status quo causes drought to significantly raise the risk of conflict in states that are normally less conflict-prone. Conclusion This paper was motivated by a significant gap in the literature on environmental drivers of violent civil conflict. Work based on case studies and small-n comparisons commonly argues that certain sociopolitical conditions can mitigate the destabilizing effects of drought, while statistical analyses of this relationship are generally insensitive to conditional mechanisms. We hoped to reconcile these branches of the literature with an array of statistical tests on the three intervening conditions that are proposed in the case literature: high social vulnerability, low state capacity, and unequal access to resources. To our surprise, none of the eight interaction models we used supported the hypotheses derived from this literature, though they did generate significant findings in the opposite direction. Drought does not have a greater destabilizing effect in insecure states with vulnerable, divided populations. It increases the risk of civil conflict only in states that would otherwise be relatively secure—states with greater food supply, lower infant mortality rates, more democratic governments, longer-enduring political systems, longer histories of peace, more urbanized populations, and larger populations suffering ethnic-based political exclusion. These results, which are robust to alternate measures of the intervening sociopolitical concepts and most alternate model specifications, give rise to a fascinating enigma: Why would states that are stable under normal conditions suffer the greatest destabilizing effects during drought? We drew on research by psychologists and behavioral economists to propose that the endowment effect, an established psychological bias that exaggerates feelings of loss and encourages risky efforts to avoid adverse change, could cause the threat of scarcity to most increase the risk of conflict where scarcity promises the largest deviation from the ex ante status quo. This argument has strong empirical support and has been used in the International Relations literature on interstate conflict, but we know of no other research to apply this logic to the study of environmental causes of civil conflict. We hope this analysis will motivate further work in several areas. It is important for future research to explore whether these counterintuitive conditional relationships can be generalized to other geographic regions or other types of political violence. We have no strong theoretical reason to believe these mechanisms should apply only to deadly conflicts in sub-Saharan Africa, and it would be particularly important for future work to examine similar drought-conflict linkages in non-African middle-income countries. This research program would also benefit from further exploration of scale effects. Work on sub-state variation in these conditions could shed light on how regional inequalities might influence the emergence of large-scale civil conflict. These advances would benefit the international policy community, which must respond to the divergence between the funds it requests and those it successfully collects with informed prioritization. Maintaining the stability provided by better social welfare and more effective government is paramount to the security of states that are projected to suffer from more extreme and more frequent severe weather events in the future. Work in this area is desperately needed, as demands on international donors exceed the resources available for adaptation efforts many times over. Footnotes 1 Barack Obama’s 2009 speech to the United Nations typifies this argument. He warns, “No nation, however large or small, wealthy or poor, can escape the impact of climate change… The security and stability of each nation and all peoples—our prosperity, our health, our safety—are in jeopardy” (Obama 2009). 2 Environmental roots of the conflict in Darfur are debated by Moon (2007), Faris (2007), Tubiana (2007), and Kevane and Gray (2008). Werrell and Femia (2013) and Ahmed (2013) discuss environmental causes of the Arab Spring. The links between food riots, climate change, and instability are weighed by Ahmed (2013), Barrett and Bellemare (2011), and Berazneva and Lee (2013). 3 See special issues of Political Geography (2007) and the Journal of Peace Research (2012), and cross-disciplinary exchanges in Science (Scheffran et al. 2012), Nature (Hsiang, Meng, and Cane 2011), and PNAS (Buhaug 2010; Burke et al. 2009; O’Loughlin et al. 2012). 4 See, among others, Buhaug’s (2010) response to Burke et al. (2009), Theisen’s (2008) rebuttal of Hauge and Ellingsen (1998), and Ciccone’s (2011) comment on Miguel, Satyanath, and Sergenti (2004). 5 Gleditsch (2012) discusses these recent sub-literatures as they appear in the largest and most recent special issue on environmental security. 6 For a more extensive critique of the CRED data, see Gall, Borden, and Cutter (2009). 7 We thank Couttenier and Soubeyran for sharing their replication data. 8 For work on sub-national variation in adaptive capacity and social vulnerability, see Busby et al. (2013). Brooks, Adger, and Kelly (2005) argue the adaptive capacity of state-level government is also important in responses to environmental crises. 9 The FAO does not record Food Supply for the Democratic Republic of the Congo, Equatorial Guinea, or Somalia. Cederman, Wimmer, and Min (2010) do not include Djibouti in their data set on Ethnic Power Relations. 10 Models are estimated with Stata 12, and all replication materials will be made available on the authors’ websites. 11 Specifically, we use scPDSIpm (self-calibrating Palmer Drought Severity Index with the Penman-Monteith method of calculating evapotranspiration), which includes these multiple-dimensions to fully characterize agriculturally relevant drought (Dai 2011; Dai, Trenberth, and Qian 2004; Trenberth et al. 2014). 12 Results are nearly identical when PDSI is not normalized. 13 We generate this variable using data from the National Capabilities dataset hosted by the Correlates of War Project (CINC). 14 t is set at the mean year (25). t2 and t3 are set at 252 and 253 instead of their means. 15 The results of more than 25 robustness tests are included in the online supplementary appendix. References Abouharb Rodwan M., Kimball Anessa L.. 2007. “ A New Dataset on Infant Mortality Rates, 1816–2002.” Journal of Peace Research  44: 743– 54. Google Scholar CrossRef Search ADS   Adger W. Neil. 1999. “ Social Vulnerability to Climate Change and Extremes in Coastal Vietnam.” World Development  27: 249– 69. Google Scholar CrossRef Search ADS   Ahmed Nafeez Mossaddeq. 2013. “Why Food Riots Are Likely to Become the New Normal.” Guardian, March 6. Accessed March 2013, www.guardian.co.uk. Barnett Jon, Adger W. Neil. 2007. “ Climate Change, Human Security and Violent Conflict.” Political Geography  26: 639– 55. Google Scholar CrossRef Search ADS   Barrett Christopher B., Bellemare Marc F.. 2011. “Why Food Price Volatility Doesn’t Matter.” Foreign Affairs, July 12. Accessed March 2013, www.foreignaffairs.com. Barrios Salvador, Bertinelli Luisito, Strobi Eric. 2006. “ Climatic Change and Rural-Urban Migration: The Case of Sub-Saharan Africa.” Journal of Urban Economics  60: 357– 71. Google Scholar CrossRef Search ADS   Beck Nathaniel. 2011. “Is OLS with a Binary Dependent Variable Really OK? Estimating (Mostly) TSCS Models with Binary Dependent Variables and Fixed Effects.” Accessed January 2014, http://politics.as.nyu.edu/docs/IO/2576/pgm2011.pdf. Benjaminsen Tor A. 2008. “ Does Supply-Induced Scarcity Drive Violent Conflicts in the African Sahel? The Case of the Tuareg Rebellion in Northern Mali.” Journal of Peace Research  45: 819– 36. Google Scholar CrossRef Search ADS   Berazneva Julia, Lee David R.. 2013. “ Explaining the African Food Riots of 2007–2008: An Empirical Analysis.” Food Policy  39: 28– 39. Google Scholar CrossRef Search ADS   Bergholt Drago, Lujala Paivi. 2012. “ Climate-Related Natural Disasters, Economic Growth, and Armed Civil Conflict.” Journal of Peace Research  49: 147– 62. Google Scholar CrossRef Search ADS   BinningsbØ Helga Malmin, de Soysa Indra, Gleditsch Nils Petter. 2007. “ Green Giant or Straw Man? Environmental Pressure and Civil Conflict, 1961–99.” Population and Environment  28: 337– 53. Google Scholar CrossRef Search ADS   Bogale Ayalneh, Korf Benedikt. 2007. “ To Share or Not to Share? (Non-) Violence, Scarcity, and Resource Access in Somali Region, Ethiopia.” Journal of Development Studies  43: 743– 65. Google Scholar CrossRef Search ADS   Bohle Hans G., Downing Thomas E., Watts Michael J.. 1994. “ Climate Change and Social Vulnerability: Toward a Sociology and Geography of Food Insecurity.” Global Environmental Change  4: 37– 48. Google Scholar CrossRef Search ADS   Brambor Thomas, Clark William Roberts, Golder Matt. 2006. “ Understanding Interaction Models: Improving Empirical Analyses.” Political Analysis  14: 63– 82. Google Scholar CrossRef Search ADS   Brenner Lyle, Rottenstreich Yuval, Sood Sanjay, Bilgin Baler. 2007. “ On the Psychology of Loss Aversion: Possession, Valence, and Reversals of the Endowment Effect.” Journal of Consumer Research  34: 369– 76. Google Scholar CrossRef Search ADS   Brooks Nick, Neil Adger W., Mick Kelly P.. 2005. “ The Determinants of Vulnerability and Adaptive Capacity at the National Level and the Implications for Adaptation.” Global Environmental Change  15: 151– 63. Google Scholar CrossRef Search ADS   Bruckner Markus. 2010. “ Population Size and Civil Conflict Risk: Is There a Causal Link?” Economic Journal  120: 535– 50. Google Scholar CrossRef Search ADS   Bueno de Mesquita Bruce, Smith Alastair, Siverson Randolph M., Morrow James D.. 2003. The Logic of Political Survival . Cambridge, MA: MIT Press. Buhaug Halvard. 2010. “ Climate Not to Blame for African Civil Wars.” Proceedings of the National Academy of Sciences of the United States of America  107: 16477– 82. Google Scholar CrossRef Search ADS PubMed  Burke Marshall B., Miguel Edward, Satyanath Shanker, Dykema John A., Lobell David B.. 2009. “ Warming Increases the Risk of Civil War in Africa.” Proceedings of the National Academy of Sciences of the United States of America  106: 20670–7 4. Google Scholar CrossRef Search ADS PubMed  Busby Joshua W., Smith Todd G., White Kaiba L., Strange Shawn M.. 2013. “ Climate Change and Insecurity: Mapping Vulnerability in Africa.” International Security  37: 132– 72. Google Scholar CrossRef Search ADS   Carmon Ziv, Ariely Dan. 2000. “ Focusing on the Forgone: How Value Can Appear So Different to Buyers and Sellers.” Journal of Consumer Research  27: 360– 70. Google Scholar CrossRef Search ADS   Carter David B., Signorino Curtis S.. 2010. “ Back to the Future: Modeling Time Dependence in Binary Data.” Political Analysis  18: 271– 92. Google Scholar CrossRef Search ADS   Cederman Lars-Erik, Wimmer Andreas, Min Brian. 2010. “ Why Do Ethnic Groups Rebel? New Data and Analysis.” World Politics  62: 87– 119. Google Scholar CrossRef Search ADS   Ciccone Antonio. 2011. “ Economic Shocks and Civil Conflict: A Comment.” American Economic Journal: Applied Economics  3 ( 1): 215– 27. Google Scholar CrossRef Search ADS   Collier Paul, Hoeffler Anke. 2004. “ Greed and Grievance in Civil War.” Oxford Economic Papers  56: 563– 96. Google Scholar CrossRef Search ADS   Conrad Courtenay R. 2011. “ Constrained Concessions: Beneficent Dictatorial Responses to the Domestic Political Opposition.” International Studies Quarterly  55: 1167– 87. Google Scholar CrossRef Search ADS   Couttenier Mathieu, Soubeyran Raphael. 2013. “ Drought and Civil War in Sub-Saharan Africa.” Economic Journal  124: 201– 44. Google Scholar CrossRef Search ADS   CRED. 2013. EM-DAT: The OFDA/CRED International Disasters Database . Brussels, Belgium: Centre for Research on the Epidemiology of Disasters. Dai Aiguo. 2011. “ Characteristics and Trends in Various Forms of the Palmer Drought Severity Index during 1900–2008.” Journal of Geophysical Research  116: 1– 26. Dai Aiguo, Trenberth Kevin E., Qian Taotao. 2004. “ A Global Dataset of Palmer Drought Severity Index for 1870–2002: Relationship with Soil Moisture and Effects of Surface Warming.” Hydrometeorology  5: 1117– 30. Google Scholar CrossRef Search ADS   de Soysa Indra. 2002. “ Ecoviolence: Shrinking Pie, or Honey Pot?” Global Environmental Politics  2: 1– 34. Google Scholar CrossRef Search ADS   FAO. 2012. FAOSTAT 2012 . New York: Food and Agriculture Organization of the United Nations. Faris Stephan. 2007. “The Real Roots of Darfur.” Atlantic Monthly. Accessed March 2013, www.theatlantic.com. Fearon James D., Laitin David D.. 2003. “ Ethnicity, Insurgency, and Civil War.” American Journal of Political Science  97: 75– 90. Google Scholar CrossRef Search ADS   Fjelde Hanne, von Uexkull Nina. 2012. “ Climate Triggers: Rainfall Anomalies, Vulnerability, and Communal Conflict in Sub-Saharan Africa.” Political Geography  31: 444– 53. Google Scholar CrossRef Search ADS   Gall Melanie, Borden Kevin A., Cutter Susan L.. 2009. “ When Do Losses Count? Six Fallacies of Natural Hazards Loss Data.” Bulletin of the American Meteorological Society  90: 799– 809. Google Scholar CrossRef Search ADS   Gleditsch Kristian Skrede. 2007. “ Transnational Dimensions of Civil War.” Journal of Peace Research  44: 293– 309. Google Scholar CrossRef Search ADS   Gleditsch Nils Petter. 1998. “ Armed Conflict and the Environment: A Critique of the Literature.” Journal of Peace Research  35: 381– 400. Google Scholar CrossRef Search ADS   Gleditsch Nils Petter. 2012. “ Whither the Weather? Climate Change and Conflict.” Journal of Peace Research  49: 3– 9. Google Scholar CrossRef Search ADS   Gleditsch Nils Petter, Nordas Ragnhild, Idean Salehyan. 2007. “Climate Change and Conflict: The Migration Link.” International Peace Academy. Gleditsch Nils Petter, Wallensteen Peter, Eriksson Mikael, Sollenberg Margareta, Strand Halvard. 2002. “ Armed Conflict 1946–2001: A New Dataset.” Journal of Peace Research  39: 615– 37. Google Scholar CrossRef Search ADS   Guttman N. B. 1998. “ Comparing the Palmer Drought Index and the Standardized Precipitation Index.” JAWRA.” Journal of the American Water Resources Association  34: 113– 21. Google Scholar CrossRef Search ADS   Hauge Wenche, Ellingsen Tanja. 1998. “ Beyond Environmental Scarcity: Causal Pathways to Conflict.” Journal of Peace Research  35: 299– 317. Google Scholar CrossRef Search ADS   Hegre Havard, Tanja Ellingsen, Gates Scott, Gleditsch Nils Petter. 2001. “ Toward a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816–1992.” American Political Science Review  95: 33– 48. Hendrix Cullen S., Glaser Sarah M.. 2007. “ Trends and Triggers: Climate, Climate Change and Civil Conflict in Sub-Saharan Africa.” Political Geography  26: 695– 715. Google Scholar CrossRef Search ADS   Hendrix Cullen S., n Salehyan Idea. 2012. “ Climate Change, Rainfall, and Social Conflict in Africa.” Journal of Peace Research  49: 35– 50. Google Scholar CrossRef Search ADS   Heston Alan, Summers Robert, Aten Bettina. 2012. Penn World Table Version 7.1. Philadelphia: Center for Production, Income, and Prices at the University of Pennsylvania. Homer-Dixon Thomas. 1991. “ On the Threshold: Environmental Changes as Causes of Acute Conflict.” International Security  16: 76– 116. Google Scholar CrossRef Search ADS   Homer-Dixon Thomas. 1994. “ Environmental Scarcities and Violent Conflict: Evidence from Cases.” International Security  19: 5– 40. Google Scholar CrossRef Search ADS   Homer-Dixon Thomas. 1999. Environment, Scarcity, and Violence . Princeton, NJ: Princeton University Press. Homer-Dixon Thomas. 2000. The Ingenuity Gap: Facing the Economic, Environmental, and Other Challenges of an Increasingly Complex and Unpredictable Future . New York: Vintage. Hsiang Solomon M., Meng Kyle C., Cane Mark A.. 2011. “ Civil Conflicts Are Associated with the Global Climate.” Nature  476: 438– 41. Google Scholar CrossRef Search ADS PubMed  Huntington Samuel P. 1968. Political Order in Changing Societies . Cambridge, MA: Harvard University Press. Kahl Colin. 1998. “ Population Growth, Environmental Degradation, and State-Sponsored Violence: The Case of Kenya, 1991–93.” International Security  23: 80– 119. Google Scholar CrossRef Search ADS   Kahl Colin. 2006. States, Scarcity, and Civil Strife in the Developing World . Princeton, NJ: Princeton University Press. Kahneman Daniel, Knetsch Jack L., Thaler Richard H.. 1991. “ Anomalies: The Endowment Effect, Loss Aversion, and Status Quo Bias.” Journal of Economic Perspectives  5: 193– 206. Google Scholar CrossRef Search ADS   Kahneman Daniel, Tversky Amos. 1979. “ Prospect Theory: An Analysis of Decision under Risk.” Econometrica  47: 263– 92. Google Scholar CrossRef Search ADS   Kevane Michael, Gray Leslie. 2008. “ Darfur: Rainfall and Conflict.” Environmental Research Letters  3: 034006. Google Scholar CrossRef Search ADS   Levy Jack S. 1992. “ Prospect Theory and International Relations: Theoretical Applications and Analytical Problems.” Political Psychology  13: 283– 310. Google Scholar CrossRef Search ADS   Marshall Monty G., Jaggers Keith. 2002. Polity IV Data Set [Computer File: version p4v2002]. Center for International Development and Conflict Management , University of Maryland. Martin Adrian. 2005. “ Environmental Conflict Between Refugee and Host Communities.” Journal of Peace Research  42: 329. Google Scholar CrossRef Search ADS   McKee T. B., Doeskin N. M., Kleist J.. 1993. “ The Relationship of Drought Frequency and Duration to Time Scales.” Proceedings of the 8th Conference on Applied Climatology, American Meteorological Society  1993: 179– 84. Meier Patrick, Bond Doug, Bond Joe. 2007. “ Environmental Influences on Pastoral Conflict in the Horn of Africa.” Political Geography  26: 716– 35. Google Scholar CrossRef Search ADS   Mercer Jonathan. 2005. “ Prospect Theory and Political Science.” Annual Review of Political Science  8: 1– 21. Google Scholar CrossRef Search ADS   Miguel Edward, Satyanath Shanker, Sergenti Ernest. 2004. “ Economic Shocks and Civil Conflict: An Instrumental Variable Approach.” Journal of Political Economy  112: 725– 53. Google Scholar CrossRef Search ADS   Moon Ban Ki. 2007. “A Climate Culprit in Darfur.” Washington Post. Accessed March 2013, June 16. www.washingtonpost.com. Müller Christoph, Cramer Wolfgang, Hare William L., Lotze-Campen Hermann. 2011. “ Climate Change Risks for African Agriculture.” Proceedings of the National Academy of Sciences of the United States of America  108: 4313– 15. Google Scholar CrossRef Search ADS PubMed  Obama Barack. 2009. Remarks by the President at the Morning Plenary Session of the United Nations Climate Change Conference, December 18. Washington, DC: White House. Accessed March 2013, www.whitehouse.gov. O’Brien Karen, Leichenko Robin, Kelkar Ulka, Venema Henry, Aandahl Guro, Tompkins Heather, Javed Akrem, et al.  . 2004. “ Mapping Vulnerability to Multiple Stressors: Climate Change and Globalization in India.” Global Environmental Change  14: 303– 13. Google Scholar CrossRef Search ADS   O’Loughlin John, Witmer Frank D. W., Linke Andrew M., Laing Arlene, Gettelman Andrew, Dudhia Jimy. 2012. “ Climate Variability and Conflict Risk in East Africa, 1990–2009.” Proceedings of the National Academy of Sciences of the United States of America  109: 18344– 49. Google Scholar CrossRef Search ADS PubMed  Palmer W. C. 1965. “Research Paper No. 45, US Weather Bureau.” NOAA Library and Information Services Division, Washington, DC. Palmer W. C. 1968. “ Keeping Track of Crop Moisture Conditions, Nationwide: The New Moisture Index.” Weatherwise  21: 156– 61. Google Scholar CrossRef Search ADS   Plümper Thomas, Troeger Vera E.. 2007. “ Efficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects.” Political Analysis  15: 124– 39. Google Scholar CrossRef Search ADS   Raleigh Clionadh. 2010. “ Political Marginalization, Climate Change, and Conflict in African Sahel States.” International Studies Review  12: 69– 86. Google Scholar CrossRef Search ADS   Raleigh Clionadh, Kniveton Dominic. 2012. “ Come Rain or Shine: An Analysis of Conflict and Climate Variability in East Africa.” Journal of Peace Research  49: 51– 64. Google Scholar CrossRef Search ADS   Raleigh Clionadh, Urdal Henrik. 2007. “ Climate Change, Environmental Degradation and Armed Conflict.” Political Geography  26: 674– 94. Google Scholar CrossRef Search ADS   Reardon Thomas, Taylor J. Edward. 1996. “ Agroclimatic Shock, Income Inequality, and Poverty: Evidence from Burkina Faso.” World Development  24: 901– 14. Google Scholar CrossRef Search ADS   Reuveny Rafael. 2007. “ Climate Change-Induced Migration and Violent Conflict.” Political Geography  26: 656– 73. Google Scholar CrossRef Search ADS   Salehyan Idean. 2008. “ From Climate Change to Conflict? No Consensus Yet.” Journal of Peace Research  45: 315– 26. Google Scholar CrossRef Search ADS   Salehyan Idean, Gleditsch Kristian Skrede. 2006. “ Refugees and the Spread of Civil War.” International Organization  60: 335– 66. Google Scholar CrossRef Search ADS   Scheffran Jurgen, Brzoska Michael, Kominek Jasmin, Michael Link P., Schilling Janpeter. 2012. “ Climate Change and Violent Conflict.” Science  336: 869– 71. Google Scholar CrossRef Search ADS PubMed  Singer J. David, Bremer Stuart, Stuckey John. 1972. Peace, War, and Numbers . Beverly Hills, CA: Sage. Slettebak Rune T. 2012. “ Don’t Blame the Weather! Climate-Related Natural Disasters and Civil Conflict.” Journal of Peace Research  49: 163– 76. Google Scholar CrossRef Search ADS   Thaler Richard H. 1980. “ Toward a Positive Theory of Consumer Choice.” Journal of Economic Behavior and Organization  1: 39– 60. Google Scholar CrossRef Search ADS   Theisen Ole Magnus. 2008. “ Blood and Soil? Resource Scarcity and Internal Armed Conflict Revisited.” Journal of Peace Research  45: 801– 18. Google Scholar CrossRef Search ADS   Theisen Ole Magnus. 2012. “ Climate Clashes? Weather Variability, Land Pressure, and Organized Violence in Kenya, 1989–2004.” Journal of Peace Research  49: 81– 96. Google Scholar CrossRef Search ADS   Theisen Ole Magnus, Holtermann Helge, Buhaug Halvard. 2011–2012. “ Climate Wars? Assessing the Claim That Drought Breeds Conflict.” International Security  36: 79– 106. Google Scholar CrossRef Search ADS   Themnér Lotta, Wallensteen Peter. 2012. “ Armed Conflict, 1946–2011.” Journal of Peace Research  49: 565– 75. Google Scholar CrossRef Search ADS   Tol Richard S. J., Yohe Gary W.. 2007. “ The Weakest Link Hypothesis for Adaptive Capacity: An Empirical Test.” Global Environmental Change  17: 218– 27. Google Scholar CrossRef Search ADS   Trenberth Kevin E., Dai Aiguo, van der Schrier Gerard, Jones Philip D., Barichivich Jonathan, Briffa Keith R., Sheffield Justin. 2014. “ Global Warming and Changes in Drought.” Nature Climate Change  4: 17– 22. Google Scholar CrossRef Search ADS   Tubiana Jerome. 2007. “ Darfur: A Conflict for Land?” In War in Darfur and the Search for Peace , edited by de Waal Alex, 68– 91. Cambridge, MA: Global Equity Initiative, Harvard University. UN. 2011. Secretary-General Presents $5.2 Billion 2012–2013 Budget to Fifth Committee, Says Figure Acknowledges Ongoing Financial Uncertainty, Austerity . New York: United Nations Department of Public Information. UNFCCC. 2007. Climate Change: Impacts, Vulnerabilities, and Adaptation in Developing Countries . New York: United Nations. UNFCCC. 2012a. National Adaptation Plans (Document FCCC/SBI/2012/L.41) . New York: United Nations. UNFCCC. 2012b. National Adaptation Programmes of Action: Index of NAPA Projects by Country . New York: United Nations. Vreeland James Raymond. 2008. “ The Effect of Political Regime on Civil War: Unpacking Anocracy.” Journal of Conflict Resolution  52: 401– 25. Google Scholar CrossRef Search ADS   Werrell Caitlin E., Femia Francesco, eds. 2013. The Arab Spring and Climate Change . Washington, DC: Center for American Progress. World Bank. 2013. World Bank Development Indicators . Washington, DC: World Bank. Zorn Christopher J. W. 2001. “ Generalized Estimating Equation Models for Correlated Data: A Review with Applications.” American Journal of Political Science  45: 470– 90. Google Scholar CrossRef Search ADS   © The Author (2016). Published by Oxford University Press on behalf of the International Studies Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com

Journal

Foreign Policy AnalysisOxford University Press

Published: Jan 1, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off