Preprint. Final appears in Human Nature
24: 33-58. doi:10.1007/s12110-012-9157-5
Risk, Uncertainty, and Violence in Eastern Africa
A Regional Comparison
Carol R. Ember
Teferi Abate Adem
Carol R. Ember
Human Relations Area Files
755 Prospect Street
New Haven, CT 06511
Teferi Abate Adem
Human Relations Area Files
Human Relations Area Files
Previous research on warfare in a worldwide sample of societies by Ember and Ember (1992a) found a strong relationship between resource unpredictability (particularly food scarcity caused by natural disasters) in nonstate, nonpacified societies and overall warfare frequency. Focusing on eastern Africa
, a region frequently plagued with subsistence uncertainty as well as violence, this paper explores the relationships between resource problems, including resource unpredictability, chronic scarcity, and warfare frequencies. It also examines whether resource scarcity predicts more resource-taking in land, movable property, and people, as well as the commission of atrocities. Results support previous worldwide results regarding the relationship between resource unpredictability and warfare frequency. Results regarding resource-taking and atrocities are more nuanced and complex. In almost all findings, relationships are generally in opposite directions in nonstate and state societies. In post-hoc analyses, atrocities are significantly more likely to be committed in states than in nonstates.
Warfare, resource unpredictability, resource scarcity, eastern Africa
, resource-taking, atrocities
Famines with high mortalities have become increasingly rare in recent years except in parts of Africa, particularly in the Horn (Devereux et al. 2002). The news from Somalia, Ethiopia, and parts of Kenya portrays shocking pictures of people suffering “mass mortality famines” of the kind previously occurring in many historically-famine-prone countries (Devereux 2000). As in the past, many of the current famines in eastern Africa are partially caused by natural disasters, primarily periods of prolonged drought. We recognize that droughts alone do not cause famine (Sen 1981) and that famines are multi-phased processes (Rangasami 1985) that evolve from bad to worse when people’s ability to access food is deliberately jeopardized by political and social means (de Waal 1989). Nonetheless, we think it important to systematically evaluate the relationships between various types of food risks to see to what degree they predict warfare in the eastern African region. We employ cross-cultural synchronic methods to address the relationships between different types of resource scarcities and various aspects of warfare—frequency, taking of land and other resources, and the conduct of combatants during warfare, particularly whether atrocities are committed. The food risks we examine range from extremely serious unpredictable events (e.g., famines or natural disasters that destroy substantial amounts of food supplies)1
to more predictable, chronic shortages (seasonal or nonseasonal).
The research reported on here focuses mostly on anthropological descriptions of eastern African societies while they were still fighting in traditional ways prior to the imposition of colonial rule and later post-independence governments, because many of the societies in eastern Africa had peace deliberately imposed upon them. To include such pacified societies as “low or rare” warfare societies would weaken our chances of finding relationships between resource problems and warfare.
This research is part of a larger project designed to develop agent-based computer models to predict the conditions under which violence occurs in eastern Africa, an area still frequently plagued by risk and uncertainty mostly caused by famine and drought. Ecological conditions are one of the main sets of conditions being modeled, along with subsistence/economic strategies, social organization, and the effects of international and national actions and conditions. The importance of models is that they allow manipulation of parameters and assumptions, but to be useful the modelers need to know what key assumptions they need to make and what reasonable parameters to put into the models. And so we employ a regional, cross-cultural comparison to empirically evaluate theories about relationships between resource scarcities and warfare.
The idea that resource scarcity would increase the likelihood of war is not new, but it was brought into prominence by ecological anthropologists in the 1960s and 1970s, particularly by Vayda (1967, 1976, but see 1989), Rappaport (1967:114–117, 224–229), Harris (1974, 1984), and Gross (1975). For example, as Vayda and Rappaport discussed, a population may at times approach or exceed its carrying capacity, and the aggression and warfare that may result at those times could reduce the population pressure because of casualties and/or shifts in who possesses resources. In this scenario, warfare should cease when the population is more in line with its resources. More recently, Gat (2006:662-672), Kelly (2000:132-133), and LeBlanc (2003:69) have pointed to resource scarcity as an important cause of warfare, particularly in simpler societies. Given the high mortality estimated for males in warfare (simpler societies appear to have higher rates than complex societies—see Gat 2006:664 and Pinker 2011:53), how is willingness to risk your life for others individually adaptive? Durham (1976) proposed a complementary relationship between cultural and biological evolution, arguing that ultimately individual fitness must increase as a result. If resources are in short supply or threatened by others, group action in warfare, if successful, should enhance individuals’ fitness benefits.2 Bowles (2009) suggested that the emergence of willingness to sacrifice in warfare for others—parochial altruism—will be selected for as long as groups with a larger number of altruists are more likely to gain resources from losers. Although not said directly, resource scarcity would presumably increase the fitness gains to winners compared with losers.3
A more proximate mechanism that might lead to the increase of conflict is the increase in frustration or irritability that individuals would experience with chronic scarcity, which would, in turn, increase the likelihood of violence (LeVine 1980; Bolton 1973; Bolton and Vadheim 1973). LeBlanc (2003:69–70) postulates a different proximate mechanism, namely, people’s perception that they are falling below their minimal standard of living—LeBlanc suggests that hunger will motivate taking food or food-producing land from others if no higher political authority is there to stop them. States, on the other hand, he argues, can let some people starve. Humans work cooperatively, so they will likely try to take resources by group action, or warfare.
The “population pressure” theory was tested on a worldwide sample by the Embers (Ember and Ember 1992a). Although their results were consistent with the “population pressure” argument, in that some measures of resource scarcity strongly predicted more warfare, particularly in nonstate societies, a couple of puzzling results prompted them to suggest an alternative explanation. The first puzzling result was that in a multiple regression analysis, only unpredictable scarcity was significant—chronic scarcity was not. If anything, chronic shortages should be a better measure of population pressure than unpredictable scarcity. Second, the Embers’ data suggested that people were not just taking resources when they are in trouble. Not many societies had more than two unpredictable disasters in the 25-year time period measured. And yet, most of those societies had warfare at least once every year. That fact prompted the Embers (1992a:256) to suggest that it seemed to be the fear of economic loss, rather than actual loss, which seemed to motivate people in nonstate societies to fight. Using the word “fear” may suggest that the impetus for war is more psychological than economic. However we would argue that if you don’t know when disaster will hit, having somewhat more to start with would usually be better than starting with less. If the land will only yield one-third of a crop in a drought, then starting with more cropland would be better. If a larger herd can be supported in good times, a larger herd even if decimated 75% will still leave you with more livestock. Given that you don’t know when that disaster will hit—it could be any year—taking resources in advance may make some economic sense. It could also be argued that taking resources from the losers gives the winners even greater economic and political advantages over the losers in future confrontations.4 Obviously there are very high costs to lives and resources as well. It is difficult for researchers to evaluate whether the payoffs outweigh the costs.5
The effect of resource unpredictability is also supported by temporal studies in coastal southern California (Lambert 1997), the U.S. Southwest (Lekson 2002), Korea (Kang 2000), and China (Zhang et al. 2007). And a recent study of livestock raids in northwestern Kenya found the most intense violence occurred in drought years (Ember et al. 2012).6
Burtsev and Korotayev (2004:35; see also Korotayev 2008), reanalyzing the Embers’ (1992a, 1992b) data, reported that state societies showed strong opposite effects—in state societies, unpredictable disasters significantly predicted less overall warfare. These findings prompted us to analyze results separately for nonstate and state societies, and based on that reanalysis we now predict negative effects in state societies.7
If resource scarcity is a driver of warfare, we expect that the victors will take resources when they win to compensate for their resource deficits. Hence, we expect both that taking resources is the norm and that the more serious the resource problems, the more resources would be taken. However, we hypothesize that taking people would work oppositely in the face of resource scarcity, since generally more people require more food. The relationships between resource problems and outcomes of warfare or behaviors exhibited during warfare, which we will examine here, have not to our knowledge been systematically examined.
Warfare is not pretty; whenever there is armed combat, people are killed. But societies vary in their customary behavior regarding combatants and noncombatants. We are dealing with societies that for the most part in the time frame described had no equivalent of international conventions and agreements regarding appropriate conduct in warfare; therefore we are not using the term “war crimes.” Nevertheless, some societies had customs that limited the harm directed toward noncombatants or the number of combatants killed, whereas others exhibited little restraint. We focus our research on behaviors that the international community agrees are war crimes (ICRC 2012): killing noncombatants purposely
, torture, rape, and the needless destruction of property. We also looked at killing combatants to see if it appeared excessive (e.g., slaughtering the entire group). We use the word “atrocities” as a shorthand term. Why terrorize others? It can be argued that terrorizing will make it more likely that the victims and their families will flee or give up access to their current resources,8
so we expected that resource scarcity should increase such behaviors in all types of societies (states and nonstates).
With regard to behavior toward combatants, a few cross-cultural studies and reviews have looked at killing captured combatants and noncombatants (Hobhouse, Wheeler, and Ginsberg 1915; Keeley 1996:87; Otterbein 2000:439). Although previous results are contradictory regarding the type of society likely to kill captured warriors (Otterbein found less-complex societies less likely to do so; the other studies mentioned above found such societies more likely), such behaviors appear to be partially related to political organization, which we plan to examine here. In addition, Otterbein (2000) found that societies that kill enemy captives are likely to also kill women and children, suggesting a pattern of terrorizing behavior in different domains.
Because previous research found that the state-nonstate contrast makes a difference in both the strength and direction of correlations, we distinguish the states from nonstates throughout the analyses.
To summarize, our hypotheses regarding resource scarcity and warfare frequencies, resource-taking, and atrocities are as follows:
Resource scarcity will be associated with higher warfare frequencies in nonpacified societies.
1a. More specifically, in nonstate societies, resource scarcity will be positively associated with warfare (and, by extension, with internal, external attacking, and external attacked frequencies).
1b. In state societies, resource scarcity will be negatively associated with warfare (and, by extension, with internal, external attacking, and external attacked frequencies).
1c. Unpredictable scarcity will be more predictive of higher warfare frequencies than predictable (chronic) scarcity.
Taking of resources occurs in most societies during the course of warfare.
Resource scarcity will be associated with frequency with which resources are taken in the course of warfare.
3a. More specifically, the more resource scarcity, the more likely land and movable property will be taken in warfare.
3b. The more resource scarcity, the less likely people will be taken in warfare.
Hypothesis 4. The more resource scarcity, the more likely atrocities will be committed in warfare.
Most systematic cross-cultural studies employ synchronic analyses, looking to see if variables are related in ways predicted by the theories and hypotheses. In a synchronic comparison it is vital to measure all the variables for each case in more or less the same time period (Ember et al. 1991; Ember and Ember 2009) If there is a causal relationship, the presumed cause and effect should occur in close time proximity. If different variables are measured at different time periods, there will be considerable measurement error (Divale 1975). By “same time period” we do not mean that each society in the sample has the same time focus. We mean that all the variables for each society are measured for that society’s focal time and place. For the purpose of studying warfare, the earliest time frame is generally preferable. Regardless of the time frame chosen for each society, the assumption is made that if there are lawlike patterns, the patterns should hold for any time and place.
Every case rated in our study was pinpointed to a 25-year time period, usually −15 and +10 years around a selected “ethnographic present.” The initial starting point for a focus came from the sample we used (see below). If the society was already pacified (that is, all warfare ceased because of colonial authorities), we tried to go back earlier in time to find traditional warfare patterns. However, in a few cases we moved the time period forward to a time that was no longer pacified. By definition, pacification means that fighting was effectively halted.
Warfare and resource problems were rated by the second and third authors, who were not aware of the hypotheses when the variables were coded. Although we initially decided that each would read the ethnographic material independently, we did not have enough time to continue this practice, so in most cases one of the authors read the ethnographic literature and took notes, usually with verbatim quotes; the other coder independently rated from the notes.
We use multivariate regression to evaluate the independent effects of unpredictable resource scarcity versus chronic scarcity on warfare frequencies in nonstate and state societies separately. Because the earlier study (Ember and Ember 1992a:250) noted but did not publish the multivariate analyses comparing unpredictable with predictable scarcity, we have reanalyzed the data for the worldwide sample. The other relational hypotheses are evaluated bivariately.
To maximize variation in resource problems, we needed to include a broad range of econiches and so our regional comparison encompasses a broad region from the Horn of Africa in the east to the eastern Congo in the west, the southern Sudan in the north, and the northern parts of Zambia, Malawi, and Mozambique in the south. This region does not technically have a name, but we will refer to it as “eastern Africa.”
We resorted to three sources of data to get a sample size of about 40 cases. We first included all the societies in this area in the HRAF Collection of Ethnography (paper and eHRAF World Cultures) as of 2009. We next included all the eastern African cases in the Standard Cross-Cultural Sample (SCCS) (Murdock and White 1969) that were not in eHRAF, and then we randomly sampled (using a table of random numbers) additional societies from this region from the Ethnographic Atlas (EA) (Murdock 1962 and subsequent issues). Because eHRAF includes many time and place foci, we used the time and place foci in the SCCS or EA as a starting point. In all cases, if there wasn’t sufficient information to code warfare frequency for the ethnographic present listed in the SCCS or EA, or if the society was already pacified, we looked for a nonpacified time period that could be rated with sufficient ethnographic information. If we could not find warfare descriptions for a nonpacified time period, we went on to the next randomly sampled case. About 60 societies were examined in this study, but only 38 societies had sufficient information on at least one nonpacified time period (see the sample cases and the foci in the Appendix). In general, the societies in the HRAF Collection of Ethnography had more information than those randomly sampled from the EA.
This sampling procedure resulted in overlap between the worldwide Ember and Ember (1992a) study based on the SCCS and the eastern African sample. To ensure that the worldwide results were not conflated by the inclusion of eastern African societies, we reanalyzed the data from the worldwide sample (Ember and Ember 1992b),9 omitting all the overlapping eastern African societies. To alleviate concern about the fact that the eastern African sample contains neighboring societies, possibly compromising the independence of cases (if one society attacks the other, the ratings of overall warfare may be higher in each), we reanalyze our eastern African results by omitting one case from a set of neighbors.10
Definitions and Coding Procedures
This section describes the variables we considered in broad terms—the coding scales are described in the Appendix. The coded data can be found at www.yale.edu/hraf/Ember2013.
Warfare. As in the Ember and Ember (1992a:248) study, warfare is defined as “socially organized armed combat between members of different territorial units (communities or aggregates of communities).” Note that the scale and organization of warfare in our sample generally differs considerably from warfare in modern nation-states. Any socially organized armed combat between communities or larger units was considered warfare regardless of the stated intent. By this definition, we focus on armed combat of socially organized groups, not on motives for fighting. Accordingly, some feuding will be considered warfare, if the episode is between communities or larger units and if at least one socially organized group is present on at least one side.11
For this study, we rated a number of dimensions of warfare during the focal period for each society—frequency (overall frequency, internal frequency, external frequency, external attacking frequency, and external attacked frequency), outcomes of warfare (taking of land, taking of people, taking of nonland, nonpeople resources—“movable property” for short), and harm inflicted on combatants and noncombatants.
With regard to internal versus external warfare frequency, we follow Ember and Ember (1992b:173) in distinguishing internal from external warfare as follows: “internal warfare is defined as socially organized armed combat between territorial units (communities or larger aggregates) within the same society. . . . External warfare refers to war between the focal society and other societies.” External attacking refers to the frequency with which the focal society attacks other societies, and external attacked refers to the frequency with which the focal society is attacked by others. Because external attacking and external attacked are very highly correlated with overall external warfare, we omit analyses with overall external warfare frequency here and focus on the components.
In terms of outcomes, we distinguish the taking of three kinds of resources when the victors win—land, movable property, and people. We also coded harm inflicted on combatants and noncombatants (see Appendix for the scales). Both outcomes and inflicted harms were coded separately for internal and external warfare.
Resource Scarcity. Two kinds of resource scarcity are distinguished here—unpredictable scarcity and chronic scarcity. Two types of unpredictable resource problems are analyzed for the 25-year period—(1) frequency of famine and (2) frequency of natural disasters that seriously destroy food supplies. These two variables follow the coding rules of Ember and Ember (1992a, 1992b). Although we also initially followed the Embers’ earlier coding scheme for chronic scarcity, we subsequently decided that the scale really tapped two different kinds of chronic problems, and so for analytic purposes we separated the scale into two parts: (1) prevalence of chronic shortages for the population (“chronic nonseasonal scarcity”) and (2) chronic seasonal shortages (i.e., hungry months) and we recoded accordingly—see Appendix.
Resource Unpredictability, Chronic Shortages, and Frequencies of Warfare
The hypotheses regarding resource scarcity and warfare frequencies (1–1c) are largely supported in the eastern Africa sample (Table 1) and are generally consistent with reanalyzed data from the Ember and Ember (1992a) study shown in Table 2. Because our hypotheses predict an overall effect of one or more resource scarcities on warfare frequency and also postulate that resource unpredictability has more of an effect than chronic scarcity, we have put both types of scarcity into the multiple regression models. We only use the stronger measure of unpredictable scarcity—disasters—rather than famine in the multiple regression model because independent variables should not be highly correlated (the rho between famine and disasters is 0.85, p< 0.001, two-tailed, n=41). We leave both chronic scarcity measures in the model because they are not highly correlated.
[Tables 1 and 2 near here]
As noted earlier, since few of the cases in the sample are neighbors, we display our results for nonstate societies with all usable cases in the first of a pair of columns and with one of each pair of neighbors omitted in the second. The results are virtually the same, suggesting that the inclusion of some neighbors was not problematic.
Consistent with hypothesis 1, regarding the predicted relationships between resource scarcity and warfare, all of the nonstate regression results (Table 1, cols. 1–8) show one or more types of resource scarcity significantly or marginally significantly predicting warfare frequencies (cols. 1–8), judging by the standardized betas for the individual resource scarcity measures. Half of the overall models (cols. 1, 2, 5, 6) are also significant. In state societies (cols. 9–12), despite the small sample size, in three of the models (cols. 9, 11, and 12) one or more significant or marginally significant resource scarcity measure predicts warfare frequency.12 And two of the overall regressions are significant or marginally so (cols. 9 and 11). The strongest models for both nonstates and states involve overall warfare frequency and external attack frequency. Hypothesis 1 is generally supported.
However, while resource scarcity of some kind appears to predict warfare frequencies, they do not do so in the same directions. Hypotheses 1a and 1b postulate that nonstates and states will have relationships in the opposite direction—resource scarcity is predicted to be positively associated with higher warfare frequencies in nonstate societies and negatively in state societies. In Table 1, 11 of 12 of the standardized betas for the main nonstate models (cols. 1, 3, 5, 7) are positive. In the state societies (cols. 9–12), all 4 of the betas for disasters and all 4 of the betas for chronic nonseasonal scarcity are negative. The only exception is chronic seasonal scarcity, for which all the betas are positive. When we examine the significant results, all 7 of the significant betas in the nonstate columns are positive, and 4 of 5 of the significant betas in the state columns are negative. We conclude that the nonstate/state divide is critical for predicting the direction of the effect of resource scarcity on warfare frequencies.
Hypothesis 1c predicts that unpredictable scarcity will be more predictive of warfare frequency than chronic or predictable scarcity. Hypothesis 1c is supported in nonstate societies (cols. 1–8), but not in state societies (cols. 9–12). In nonstate societies our measure of unpredictable scarcity (“disasters”) has a higher beta in all models than either of the chronic scarcity measures. Moreover, in the main models (1,3,5,7), the betas for disasters are significant or marginally significant, whereas only one beta out of 8 for chronic scarcity is significant (see column 5 for external attacking frequency). However, in state societies (cols. 9–12), disasters is only higher in column 11 for external attacking frequency, but not in other models.
Since we were looking to see if the eastern African data replicated the Embers’ (1992a) previous worldwide finding regarding resource unpredictability and overall warfare frequency, we have displayed the original worldwide published results (labeled Model 1) in column 1 of Table 2 along with reanalyses of the worldwide data in columns 2–4, 6, and 7. Comparable eastern African analyses from the current study appear in columns 5 and 8. The first set of reanalyses (cols. 2 and 4 in Table 2) of the worldwide data exclude any eastern African societies because we want to ensure that the overlap in samples did not account for similar results in the two studies. Although we were not able to rate socialization for mistrust for this study, we note that results regarding disasters in the worldwide sample are basically similar to those we have found for eastern Africa. First, the standardized betas for disasters are strong, positive, and significant in nonstate societies (cols. 1 and 3 in Table 2) even when the eastern African cases are omitted from the worldwide sample (see cols. 2 and 4 in Table 2). Second, in nonstate societies, when disasters and chronic scarcity are both in the model (labeled Model 2), only disasters predicts overall warfare frequency significantly (see cols. 3 and 4). Third, the sign of the beta coefficients for disasters reverses in state societies (cols. 6 and 7) as predicted. In other words, disasters in state societies appear to be associated with less, not more, overall warfare. Note that in Table 2 we use the overall chronic scarcity measure that the Embers (1992a) used, rather than the two new disaggregated measures, to enable a more direct comparison of our data with previous results (see col. 5 in Table 2). In column 5, the beta for natural disasters in nonstate societies is significant and the beta for overall chronic scarcity is not significant.
[Table 2 near here]
Outcomes of Warfare
We hypothesized that most societies will take resources during warfare if they can. Looking at the median score (see Appendix for scales), our data are consistent with hypothesis 2. In eastern Africa, the median scale score for taking movable property (almost always livestock) in both internal and external warfare is “always.” Taking people is the next most common category—the median is “usually” for internal warfare and “always” for external warfare. Land is somewhat less likely to be taken, although it is not uncommon. The median score for both internal and external warfare is “the defeated are sometimes driven from their territory and the victorious sometimes use the land of the defeated.” Our eastern African findings parallel the findings from the Embers’ (1992a) worldwide comparison that resource-taking is the norm.
Turning to our hypotheses about resource scarcity and taking of resources (3, 3a, and 3b), our results are equivocal. We hypothesized that resource scarcity would increase the taking of land or movable property during warfare (3a) but make it less likely to take people (3b). Our strongest results are consistent with the “people” hypothesis (3b) for nonstate societies (Tables 3 and 4). Of the 8 possible relationships in nonstate societies, 5 of the results are significantly negative, as expected, and another is negative and marginally significant. The two correlations that were not negative for nonstate societies were between “taking people” and chronic seasonal scarcity. Since chronic seasonal scarcity is probably the least troublesome type of resource problem, we think the hypothesis regarding the lower likelihood of taking people during warfare in nonstate societies is mostly supported. The number of state societies that could be rated on outcomes and resource problems is very small, so it is difficult to get significant results. However, we note that the three substantial correlations (0.66, 0.67, and 0.60) are in the opposite direction. These results tentatively suggest a different direction of relationship in state societies. We discuss why the result might be positive for state societies, rather than negative, in the discussion section. In sum, the hypothesis is partially supported and appears to apply to nonstate societies.
[Tables 3 and 4 near here]
Hypothesis 3a postulates that both movable property and land will be more likely to be taken the higher the resource scarcity. The results for movable property and land are not the same. Taking movable property first (middle rows of Tables 3 and 4), the results only narrowly support the hypothesis in two specific conditions—nonstate societies and external warfare. Table 4 (external warfare) shows three significant or marginally significant positive results. In external warfare, nonstate societies with more famine and chronic seasonal scarcity are significantly more likely to take movable property, and societies with more disasters are marginally significantly more likely to take movable property. Notice that for state societies, external warfare correlations are in the opposite direction, contrary to the hypothesis (albeit only one correlation is marginally significant). With internal warfare in nonstate societies (the top of the middle row in Table 3) there are no significant relationships between taking movable property and any of the resource scarcity measures; in state societies, only famine is marginally significantly related to more taking.
With regard to taking land in nonstate societies (Tables 3 and 4), none of the relationships are significant. Only two results for state societies are significant or marginally significant, and they only relate to state societies during internal warfare; with chronic scarcity of either type, there is more taking of land in internal warfare (see Table 3).
Behavior toward Combatants and Noncombatants
Before we discuss the test of the hypothesis regarding atrocities, we note that atrocities are not infrequently committed in the course of warfare in this region of the world. The median rating for both killing noncombatants and the rape of women is “sometimes.” Both torture and destruction of resources occur more commonly—the median rating for both is “usually” in both internal and external warfare. Non-physical intimidation varies by type of warfare. For internal warfare, the median rating is “always”; for external warfare it is between “sometimes” and “usually.” We discuss nonstate/state differences in the following section.
Do resource problems increase the likelihood of atrocities as hypothesis 4 predicted? On the whole, the overall pattern of results does not support the hypothesis. At best, there is only narrow support in state societies, where chronic seasonal scarcity predicts more atrocities committed in the course of external warfare. However, in nonstate societies, most of the relationships are in the opposite direction.
Tables 5 and 6 show the relationships between unpredictable scarcities and atrocities in internal warfare and external warfare for both nonstates and states. The nonstate and state societies clearly show very different patterns. With regard to internal warfare (Table 5), 11 of the 12 correlations in nonstate societies are negative, 5 of them significant or marginally significant, suggesting, contrary to expectation, the more there are unpredictable scarcities in nonstate societies, the less likely atrocities (we show results of two-tailed tests when results are opposite our predictions). On the other hand, 11 of the 12 correlations for state societies are positive, suggesting that unpredictable scarcity has the opposite effect in state societies. Even though the number of cases on the state side is very small, two of the positive correlations are significant or marginally significant—with more famine and more natural disasters, killing of noncombatants in internal warfare is more likely. The picture with regard to external warfare (Table 6) is similar. Ten of the 12 nonstate correlations are negative; 2 are marginally significant. On the other hand, for state societies all 12 correlations are positive, and just as with internal warfare, both famine and disasters strongly and significantly predict more killing of noncombatants.
[Tables 5 and 6 near here]
Now let us turn to the relationships between chronic resource problems and atrocities in internal warfare (Table 7) and in external warfare (Table 8) in nonstates and states. As mentioned above, there is narrow support for the hypothesis regarding chronic scarcity in state societies. The most obvious pattern (Table 8) is that in state societies, both types of chronic resource problems have only positive relationships to atrocities when warfare is external (12 of 12). Five of the 6 correlations are significant or marginally significant with chronic seasonal scarcity—the more seasonal scarcity in state societies, the more atrocities are committed. Other patterns are far from clear. Turning to internal warfare, there are few significant correlations. In nonstate societies, the correlation between more chronic seasonal resource problems and higher killing of combatants is significant, but the other atrocities generally do not show the same pattern. In state societies there are only two significant correlations, between seasonal and nonseasonal scarcity and the greater likelihood of killing noncombatants.
[Tables 7 and 8 near here]
In sum, our expectation that resource problems would increase the likelihood of atrocities at best only applies to state societies with regard to chronic scarcity. Perhaps more interesting is the fact that the relationships in states and nonstates are generally in the opposite direction. Forty-five of the 48 correlations shown in Tables 5 through 8 are positive in state societies. More telling, in state societies all of the significant or marginally significant associations (12 of 12) are positive. Most of the relationships in nonstate societies are negative (33 of 48), and 8 of the 11 significant or marginally significant correlations are negative. Admittedly, our sample of state societies is quite small; nevertheless the difference in direction is striking, suggesting that resource problems have opposite effects in the two types of society.