Abstract
The large-n literature on political violence has paid little attention to the distinction between insurgencies that control territory and those that do not. Territorial control has consequences for the lethality of the group, its pattern of recruitment and bargaining power. The main determinant of territorial control, we argue, is state capacity: while territorial insurgencies are more frequent in poor countries, nonterritorial ones tend to occur in countries with intermediate levels of development (rich countries are free of internal violence). The authors show that the relationship between development and nonterritorial violence is a concave one, using a panel for the period 1970–1997 that combines existing data sets on civil wars and the Global Terrorism Database 1. The authors also find that nonterritorial violence is more likely in democratic, old states. Population, rough terrain, and inequality have a similar impact on both types of conflict. The authors discuss to what extent territorial conflicts correspond to civil wars and nonterritorial ones to terrorism.
In the 1960s, a wave of left-wing insurgencies swept Latin America. Inspired by the successful Cuban example, revolutionaries tried to create revolutionary focos in the countryside. After gaining control over the local population in the foco, they sought to expand their territory, conquering new space, the ultimate goal being the seizure of the capital and the collapse of the regime.
While in countries such as Colombia, Guatemala, Peru, and El Salvador, the insurgents were able to gain large territories during the conflict, the rural experiment was a nonstarter in Argentina, Uruguay, Venezuela, and Brazil, where rebels mostly fought in the cities, under the constraints of clandestinity.
Despite Fidel Castro’s dictum that “the city is the grave of the guerrilla,” insurgents in these four countries soon realized that they could not rely on territorial control in the countryside as their main strategy to defeat the government. Carlos Marighella (1985) theorized that in highly urbanized countries such as Brazil urban violence should activate the revolt and mobilize support for the cause. Abraham Guillén (1973, 217) foresaw that the theater of operations in flat, urbanized countries such as Uruguay and Argentina would be “the forests of cement” of the big cities.
For these two advocates of revolution, urban guerrilla was the most appropriate strategy for rebels that were unable to liberate territory from the state’s control. The Tupamaros in Uruguay were the purest manifestation of the urban strategy, their experience being emulated by many other groups around the world (Sánchez-Cuenca 2009). Given the relative affluence of the countries experiencing nonterritorial violence, urban violence was dubbed the guerrilla of the rich countries, as opposed to the traditional rural guerrillas of the poorer ones.
This article elaborates on the distinction between insurgencies with territorial control and insurgencies without it. Territorial control is an important issue, which affects the lethal capacity and offensive power of insurgencies, as well as their patterns of recruitment. Territorial control is mainly driven, we argue, by state capacity. When the state is weak, insurgents can aspire to gain territorial control. Confronted with stronger states, insurgents cannot but go underground. The strongest states, however, do not suffer violence at all, since would-be rebels are so powerless that they opt for peaceful ways of protest, if any. In a sense, nonterritorial violence is a residual one: it is the kind of challenge that states experience before reaching a threshold of development beyond which no incentives for rebellion exist.
Despite being a crucial issue in warfare, territorial control has been neglected in the quantitative literature on political violence, since the territorial factor has usually been omitted from any of the two sides of the regression equation. Here, we focus on territorial control as a dependent variable. We construct a three-valued variable with peace, territorial conflict onset, and the nonterritorial one. Information about insurgencies comes from data sets on civil war and on terrorism (basically, Fearon and Laitin 2003; Sambanis 2004a; and the Global Terrorism Database [LaFree and Dugan 2007]). We have checked for each conflict in these data sets whether violence was territorial or not, based on a wide number of sources.
The main result of the analysis concerns economic development (state capacity). While territorial insurgencies are concentrated in poor countries, nonterritorial ones are more likely in countries with intermediate levels of economic development. In fact, the relationship between nonterritorial insurgencies and gross domestic product (GDP) per capita is concave, while linear for territorial ones.
We also take into account the standard predictors of conflict, such as political regime, state age, rough terrain, population, and inequality. In general, nonterritorial conflict is more likely in democracies and in older countries. Inequality, population, and rough terrain have similar effects regarding both kinds of conflict: they predict conflict, but they do not help to account for their territorial nature.
The article is structured as follows. First, we discuss the connection between territorial control and insurgency type. Second, we provide several tests of the observable implications that follow from the distinction between territorial and nonterritorial insurgencies, proving that the rationale for analyzing them separately is not mere speculation. Third, we carry out the statistical analysis of conflict onset by type of conflict. We then discuss the main findings in the context of Latin America, with a brief qualitative analysis of deviant cases. The article ends with a discussion about how our results make sense of some findings in the largely disjointed literatures on civil wars and terrorism.
Insurgency and Territory
Insurgencies may be differentiated depending on whether they have territorial control or not (Sánchez-Cuenca and De la Calle 2009). Some insurgencies are able to seize territory from the state’s control, acting in these areas as local rulers and replacing therefore the authority of the state. The Liberation Tigers of Tamil Eelam (LTTE) in Sri Lanka fully controlled a large territorial area in the north and east of the country and it was to all intents and purposes the local ruler: it created a shadow state, including justice courts, postal service, police, and broadcasting (Hussain 2010, 382). Some other insurgencies, however, do not seize territory and act under the constraints imposed by clandestinity. The Red Army Faction in Germany was a fully clandestine organization that acted exclusively in cities. It spent most of its resources looking for safe places to hide its members (Aust 2009). This difference between insurgencies, incidentally, is independent of whether insurgents seek regime change or territorial concessions. At its peak, the Shining Path controlled large areas of Peru, but its aim was not territorial (they sought to overthrow the government and to create a Communist state). Euskadi Ta Askatasuna (ETA; Basque Homeland and Freedom) has always been an underground organization, but its aims are territorial: it seeks the secession of the Basque Country from Spain.
We are not concerned here with the goals of the insurgents but rather with the way in which they fight for their cause. There is a strong association between territorial control and rural guerrillas, on one hand, and a lack of territory and urban insurgency, on the other. Most insurgencies that control territory are rural-based, having a base in the jungle or in the mountains, where they become local rulers. But the association is far from perfect (Staniland 2010). First, there are rural insurgencies of a nomadic nature that hide from security forces but fall short of controlling any territory. An interesting example is the Maquis in Spain under Franco: this was a rural guerrilla with no permanent territorial base, its members being always roving rebels trying to escape from security forces (Serrano 2006). Second, there may be territorial control of urban areas, as was the case for example in Beirut during the Lebanese civil war, where the contending parties had control of various areas of the city (Maksidi and Sadaka 2005, 66–67).
Our point is not exactly about rural versus urban guerrilla, but rather on territorial control, be it rural or urban. As we explain in greater detail in the Appendix, territorial control requires at least having camps or bases within the country borders, where the insurgents store weapons and train recruits. At most, territorial control means that the insurgents replace the authority of the state and create a parallel state that imposes order, administers justice, and extracts rents from the population. By contrast, underground or nonterritorial insurgencies are forced to hide all the time.
Both territorial and nonterritorial armed groups challenge the monopoly of violence the state is supposed to hold. The difference between the two is rather one related to sovereignty. Territorial control means that the insurgents break the state’s sovereignty over its own territory. In civil wars, sovereignty is segmented, or at least fragmented (Kalyvas 2006). In nonterritorial conflicts, the state retains sovereignty even if an armed group commits violent attacks.
The seizure of territory within the state’s borders has far-reaching implications for the production of insurgent violence. The dynamics of bargaining, recruitment, and lethality rely on the capacity of insurgents to seize territory and keep it.
First, in terms of bargaining power, territorial groups will have greater capacity to expand in size and also greater military resources that allow them to hold small-scale battles and employ hit-and-run tactics against the enemy. Rebels without territorial control, in contrast, must act clandestinely, which imposes severe limits on their capacity to grow. Nonterritorial insurgents cannot but restrict themselves to tactics such as improvised explosive devices, shootings, assassinations, hostage taking, and the like. As a consequence, the bargaining power of the two types of insurgencies differs: whereas many territorial insurgents claim and obtain a seat at the negotiation table, those without territorial control rarely get this privilege.
Second, in terms of recruitment, holding territorial control implies some measure of power over the local dwellers. The insurgents often become local rulers and develop institutions and norms resembling those of the state. This requires organizational resources and contact with the civilians living in the territory. Regardless of the initial strength of the insurgency, once rebels have territorial control, local dwellers may be forced to join their ranks, given the risk of repression or displacement if the offer is turned down (Kalyvas and Kocher 2007).
Insurgencies without territory have more superficial contact with civilians. The constraints of clandestinity make the process of recruitment harder. 1 Following the terminology of Weinstein (2007), underground groups can only rely on “social endowments” to build the organization. On one hand, the capacity to extract rents from the population is severely impaired for rebels who remain underground. On the other, governments could easily buy off looting-driven recruits, with devastating effects for the survival of the group. Thus, whereas territorial insurgencies, similarly to states, can attract recruits based on a combination of material incentives and ideology, clandestine groups depend more on ideology. This is why these groups are often formed by the most radical fringe of larger social movements (Della Porta and Tarrow 1986).
Finally, holding territorial control also has observable effects on lethality. As territorial groups have greater offensive capacity, we should expect that they produce a substantially larger number of fatalities. The 1,000 fatality threshold used to define civil wars is not perhaps entirely arbitrary. Many conflicts with less than 1,000 fatalities correspond to armed groups without territorial control, and almost every group with territorial control causes more than 1,000 fatalities, albeit with important exceptions (the Provisional Irish Republican Army [PIRA], for instance, killed over 1,600 people).
The literature on civil wars has shown that the main determinant of territorial control by insurgents is state capacity (Collier and Hoeffler 2004; Fearon and Laitin 2003; Hegre and Sambanis 2006). When the state administration works all over the national territory and the army and the police have an effective presence across the country, insurgents are not able to liberate territory from the state’s control. This is most likely to occur in rich countries, which are the ones with the capacity to maintain sovereignty intact. Rebels, therefore, gain territorial control in poor, defective states.
However, it might be argued that territorial control is, after all, nothing but a manifestation of a difference in recruitment capacity. According to this point of view, groups that are able to liberate territory are those that are successful in recruiting people to their side; groups who fail in attracting recruits are those that become underground and do not try to control territory. There is some evidence, nonetheless, that goes against this claim. Very large groups, such as the Montoneros in Argentina 2 and the PIRA in Northern Ireland, 3 were unable to liberate territory, and initially weak groups, such as the Maoists in Nepal 4 and the Castroists in Cuba, 5 managed to build strong organizations after seizing territory.
In the end, state capacity seems to overrun the initial levels of rebel support. If the state is strong, rebels will not be able to control territory, regardless of the initial size of their movement. If the state is weak, however, rebels could still thrive to build a strong organization thanks to territorial control, even if low support existed before the onset of conflict.
We hypothesize in this article, then, that the type of conflict is related to state capacity. Three scenarios can be envisioned. First, when the state is weak, rebels will likely be able to liberate territory and to fight a more lethal war. Second, when the state is very strong, no insurgency will emerge, either territorial or not. A sufficiently strong state keeps both sovereignty and the monopoly of violence. Finally, if the state is strong enough as to prevent rebels from seizing territory, but not as to deter challengers, nonterritorial rebellions should be the most common type of insurgency.
Nonterritorial insurgencies, therefore, emerge in rich, but not the richest, countries. Internal peace is the privilege of the strong states. From this perspective, it makes sense to consider that nonterritorial conflict is a residual form of violence. It is the violence that states experience before reaching the region of peace.
Data and Preliminary Evidence
In order to explore the differences between territorial and nonterritorial conflicts, we have compared all the insurgencies included in the Global Terrorism Database 1 (GTD1), compiled by Gary LaFree and Laura Dugan based on the files of the Pinkerton Global Intelligence Service (LaFree and Dugan 2007). It is, by far, the largest available data set on terrorism activity, covering both international and domestic events.
Typical data sets on large violent conflicts have a country/year panel structure in which the unit of observation is the conflict (Fearon and Laitin 2003; Gleditsch et al. 2002; Sambanis 2004a). The vast majority of conflicts in these data sets correspond to territorial ones. On the other hand, data sets on terrorism are much more disaggregated, the unit of observation being the attack and tend to cover international terrorist incidents only.
GTD1 contains 61,637 violent events of all kinds for the period 1970–1997. Observations for the year 1993 are, however, missing. Due to the vagueness of what counts as a terrorist attack in the data set (the threatened or actual use of illegal force and violence to attain a political, economic, religious, or social goal through fear, coercion, or intimidation), many coded events were carried out by territorial groups and correspond to conflicts that are routinely classified as civil wars. This comprehensiveness makes GTD1 rather unique. 6
We establish a minimum requirement for considering that an armed group poses a challenge. Thus, we only consider groups that killed at least ten people during more than one year of activity. 7 This excludes, for instance, a number of brutal attacks that have no continuity in time (such as the Oklahoma bombing, which was a unique event, not being part of a sustained campaign). The challenge to the state becomes serious when there is a real possibility that further attacks will be carried out.
We have also excluded groups that commit more than 95 percent of their attacks outside of their national borders, such as the Japanese Red Army or the Cuban Coordination of the United Revolutionary Organization. This last group, despite aspiring to end communism in Cuba, only operated abroad, attracting few Cubans from the exile communities. The fate of fully transnational organizations may depend more on the characteristics of the states wherein they operate than on the characteristics of the state they are fighting (Rosendorff and Sandler 2005).
There are 150 groups that fit these conditions. GTD1 does not include information on territorial control, but, based on external information (profiles of armed groups, cases studies, monographs, Web pages), we have checked in every case whether the armed group had territorial control or not. Appendix A includes information about the coding rules we have used. After careful examination of the cases, we have found seventy-one organizations without territorial control and seventy-nine with it.
Most of the territorial control groups intervene in well-known cases of civil wars (such as those that took place in Mozambique, Angola, El Salvador, Peru, or Myanmar). Nonterritorial groups, by contrast, are concentrated in conflicts which are usually not included in civil war data sets.
There may be several armed groups acting simultaneously in the same country and in the same conflict. We have then aggregated organizations by conflict, resulting in seventy-six conflicts during the period 1970–1997. For instance, in Northern Ireland, there were many nonterritorial armed groups, both Republican and Loyalist ones: they are all grouped together in a single conflict, the Troubles, in which the status of Northern Ireland was at stake. In Spain, by contrast, we find two nonterritorial conflicts going on simultaneously: first, the one featured by ETA, in which the goal is independence of the Basque country; second, the one featured by leftist and fascist armed groups during the transition to democracy. These were largely independent of each other. Of all the conflicts derived from GTD1, forty-seven are featured by insurgencies with territorial control and twenty-eight by insurgencies without it (see Table A1 in the online Appendix A for a full list of the nonterritorial conflicts). 8
Some preliminary evidence drawn from GTD1 confirms our theoretical expectations about the relevance of territorial control. Table 1 shows large differences in lethality as well as in the economic and political conditions under which each type of insurgency emerges. 9 Regarding lethality, the mean of killings in territorial conflicts is 1,417 (above the 1,000 fatality threshold of civil wars), whereas the mean in nonterritorial ones is only 289 (noticeably under it), the difference being statistically significant at 5 percent. 10
Some Covariates of Territorial Control
Note: Lethality: number of people killed by the nonstate armed groups (Global Terrorism Database 1 [GTD1]). Gross domestic product (GDP) per capita: data from Fearon and Laitin (2003). Democracy: 0 democracy, 1 dictatorship; data from ACLP, Przeworski et al. (2000).
*p < .05. **p < .01.
In turn, territorial groups tend to emerge in poorer countries than underground groups. Armed groups without territorial control are more likely to fight democracies, whereas groups with territorial control are observed in dictatorships.
In light of these differences, it seems necessary to separate the two kinds of insurgencies for an analysis of the conditions under which they emerge. Before going into the empirical analysis, we explore first the consequences of not doing so by focusing on the effect that economic development (GDP per capita) has on the onset of violent conflict. As data sets of domestic violent conflicts become more inclusive, conflating territorial and nonterritorial insurgencies, the effect of economic development becomes less important: when all kinds of insurgencies are merged into a single category, they are observed all along the range of development. The data sets we have analyzed are those of Fearon and Laitin (2003), Sambanis (2004a), and the PRIO Armed Conflict Data Set (Gleditsch et al. 2002). Regarding GTD1, we use two versions: the original one, which includes all conflicts, territorial or not, and then the version only with nonterritorial conflicts.
To make the lists of conflicts strictly comparable, we restrict all data sets to the period covered by GTD1, 1970–1997. There is very little overlap between cases of civil wars and our operationalization of cases of nonterritorial violence. Regarding Fearon and Laitin’s data set, there are only two cases of their list of civil wars that we think correspond to nonterritorial violence: Argentina in the early 1970s and Turkey in the late 1970s. The Montoneros in Argentina tried to emulate guerrillas, but the state prevented this and their violence was mostly urban and clandestine in nature (Gillespie 1982, 203). In Turkey, there was a bloody conflict between the left and the right in the 1970s, but it was mainly urban violence and the parties did not control territory (Sayari 2010). 11 We have excluded these cases from the civil war data sets. Though this does not alter the results, it is important for consistency with our coding rules.
Table 2 contains various models in which the independent variable is GDP per capita (taken from Fearon and Laitin 2003). The dependent variable is conflict onset: it has value 1 in the first year of the conflict and 0 otherwise. We exclude those observations with ongoing conflicts, except new onsets. GDP per capita is lagged one period. 12 Countries with fewer than one million inhabitants have been excluded from the analysis.
The Vanishing Negative Effect of Economic Development on Conflict Onset
Note: GDP = gross domestic product; GTD1 = Global Terrorism Database 1. Country clustered standard errors in parentheses.
*p < .05. **p < .01. ***p < .001.
Model 1 is based on Fearon and Laitin’s data set (corrected for the two cases mentioned earlier). It contains sixty-eight civil wars. Model 2 is based on Sambanis’s list, with ninety-one civil wars. Model 3 is based on PRIO’s Armed Conflict Data Set, with 104 conflicts (the threshold is much lower now, at least twenty-five “battle-related deaths” in one year). Model 4 contains the seventy-five violent conflicts of GTD1—as defined above. Model 5 contains only our own list of twenty-eight nonterritorial insurgencies. As the number of positives in each data set attests, Fearon and Laitin’s is the most demanding data set, followed by Sambanis’s and PRIO’s.
The coefficient of economic development becomes smaller from model 1 to model 5. Whereas in Fearon and Laitin’s data set, the coefficient is −.32, it goes down to −.23 in Sambanis’s data set, to −.16 in the Armed Conflict data set, and to a mere −.07 in the version of GTD1 in which all conflicts are included. If only nonterritorial conflicts are analyzed, the coefficient becomes positive, but is no longer significant. The effect of economic development, therefore, varies dramatically with the inclusiveness of data sets.
This section has shown that, apart from theoretical reasons, preliminary data analysis confirms the necessity of differentiating types of insurgencies in terms of territorial control. Their conflation in a single category would conceal interesting contrasts with regard to the determinants of onset. We move now to a more detailed analysis of the conditions associated with the two types of insurgencies.
The Determinants of Conflict Onset
We consider three different scenarios: no conflict, conflict onset featured by territorial insurgencies, and conflict onset by nonterritorial underground groups. The dependent variable, therefore, has three values. Several comparisons are possible. On one hand, we can compare the occurrence of the two kinds of conflicts, following the preliminary results shown in Table 1. The question here is under what circumstances conflicts are territorial and under which ones they are nonterritorial. On the other hand, we can analyze the determinants of onset for each type of conflict as compared to peace.
Territorial versus Nonterritorial Conflicts
According to our previous argument on state capacity, we should find that nonterritorial conflicts tend to happen in countries that are richer than those that suffer territorial warfare. Civil wars tend to be observed in poor countries, where the rulers lack the capacity to keep internal order due to the weakness of the state. Thus, in a comparison between the two types of conflict, the effect of state capacity should be positive: the richer the country, the more likely that the conflict observed is nonterritorial.
The literature on political violence has discussed other factors that may play a role in conflict onset (see the reviews of Hegre and Sambanis 2006; Mason 2009). Population size and the presence of ragged terrain help promote civil wars: larger, more mountainous countries offer additional possibilities for insurgents to fight against the state. In turn, recently created states have more prospects of a civil war: institutions in new states may be more fragile and ill-tuned to cope with internal conflicts. In general, the level of inequality is not a good predictor of the onset of civil war. Finally, there is no consensus on the effect of regime type. Some authors find no relationship, whereas others find an inverted-U relationship, with more conflicts in mixed regimes (see Gleditsch, Hegre, and Strand 2009). We control for all these variables, and we test whether they have a differential effect on each type of conflict.
In the first analysis, we investigate if these factors can predict whether the conflict is territorial or not. The predictors are the following. First, economic development, as measured by GDP per capita, taken from Fearon and Laitin (2003). Economic development is usually taken as a proxy of state capacity. Second, political regime: we use the ACLP dummy measurement of regime type (0 = democracy; 1 = dictatorship; Przeworski et al. 2000). Third, state age: based on the State System Membership List, coded by the Correlates of War project (2008), we take the number of years the state has been functioning, starting in 1816. Fourth, economic inequality: given the problematic nature of the data, we decided to use the mean value of income inequality for each country in our study period, 1970–1997, based on Deininger and Squire (1996). Admittedly, it is a very imperfect measurement; however, it is the only way to avoid a massive loss of observations. Fifth, we also include population (in log) and rough terrain—both variables are taken from Fearon and Laitin (2003).
The data set is a panel for all countries with a population over one million people during the period 1970–1997. The dependent variable has value 0 for all country years with no conflict onset, value 1 for territorial conflict onset, and value 2 for nonterritorial conflicts. Given that GTD1 is intended to capture terrorist incidents rather than civil wars, we use GTD1 only for value 2 (nonterritorial conflicts). The list of civil wars included in GTD1 is not particularly reliable. Thus, for value 1 (territorial conflicts), we have used the list of Fearon and Laitin (2003), corrected as explained earlier. This is the most demanding data set in terms of the inclusiveness of conflicts. We have replicated all the analyses of this section using Sambanis (2004a) for value 1 of the dependent variable, finding no relevant differences (see the online Appendix B). There are sixty-eight episodes of territorial conflict onset and twenty-eight of nonterritorial ones.
Given the structure of the data, we estimate multinomial logit models with country-clustered standard errors. All independent variables are lagged one period to avoid potential issues of endogeneity. Observations corresponding to spells of violence after the conflict onset year are removed, except for new onsets. One problem is that of temporal dependency. In panels with a dichotomous dependent variable, we can use cubic splines (Beck, Katz, and Tucker 1998). With a three-valued variable, there is not a standard solution. We have adapted the usual procedure to the multinomial setting as follows: we define the “peace years” variable for the emergence of any violent conflict, regardless of its territorial nature. Cubic splines are then defined in terms of this variable. This makes sense in so far as the effect of years of peace is the same for both types of conflict. We think that this is a reasonable assumption for our estimation problem. We do not see any reason to consider that, keeping everything else constant, the eruption of some type of conflict is more likely than the other one for any given period of peace.
Table 3 shows the results for the comparison between the two types of conflict onset. Territorial conflict is the base category. In model 1, inequality has been excluded as this variable reduces the sample considerably (we lose nine cases of territorial conflict onset of the sixty-eight). The first column of model 1 contains the analysis of territorial conflict onset. Basically, it produces the same results as a standard analysis of civil war onset, though coefficients have the opposite sign to the ones that are usually reported due to the fact that territorial conflict is here the base category. We find, in accord with previous studies, that these conflicts occur in poorer countries, with a large population and rough terrain. We do not find an effect for political regime.
Nonterritorial versus Territorial Violence
Note: GDP = gross domestic product. Country clustered standard errors in parentheses. Spline coefficients not shown.
*p < .05. **p < .01. ***p < .001.
The most interesting results, however, have to do with the comparison between the two types of conflict. In the second column of model 1 we observe, in line with the results of Table 1, a positive coefficient for GDP per capita: the richer the country, the more likely that the conflict is a nonterritorial one. As suggested previously, this means that stronger states are better able to prevent any loss of territorial control if challenged by domestic insurgents. In weaker states, insurgents may gain control over a certain area that becomes their territorial base; when this is so, the conflict can then easily escalate into a civil war.
There is a strong effect of new versus old states; the older the state, the more likely that the violent conflict is a nonterritorial one. Territorial conflicts are more frequent in newer, younger states. This is consistent with our previous finding on state capacity. It seems reasonable to suppose that younger states will have lesser state capacity than old, well-established ones, everything else constant.
According to model 1, the coefficients of population and rough terrain are not significant for the comparison between the two types of conflict: both types tend to happen in highly populated countries and with similar mountainous areas. The lack of effect of rough terrain goes against the expectation that nonterritorial violence should thrive more in flatter countries, where rebels have fewer chances to seize territory. Regarding political regime, it is not significant. Whereas in the simple mean comparison in Table 2, there seemed to be a difference (nonterritorial conflicts being more likely in democracies), the multivariate analysis does not reveal a significant regime effect.
In model 2, inequality is added. It does not have any noticeable effect. It cannot be said that the more brutal conflicts (the territorial ones) are triggered by deep inequalities, whereas milder conflicts (the nonterritorial ones) occur in less inegalitarian societies.
In general, Table 3 reveals that the emergence of nonterritorial violence as opposed to the territorial one is mainly driven by the capacity of the state to prevent territorial loss. The states that are challenged by nonterritorial insurgencies are more developed and older than the states in which insurgents are able to capture part of the territory. Moreover, the nature of the conflict does not depend on the geographical shape of the country, as rough terrain does not differentiate between the two types of insurgencies.
The Onset of Nonterritorial Conflicts
According to our argument, nonterritorial conflicts are a form of residual violence that developed countries experience before entering into the region of internal peace in which no armed group challenges the state. This region of internal peace is reserved for the most developed states. Nonterritorial violence will be most likely observed in countries with intermediate levels of development, richer than those that suffer territorial violence.
The base category in the dependent variable is now peace. In order to test the concave relationship between development and nonterritorial violence, we introduce a quadratic term of GDP per capita. If the quadratic term has a negative sign, the form of the relationship is that of an inverted U, whereby the poorest and the richest countries experience little nonterritorial violence.
The rest of the independent variables are the same as in the previous analysis. Model 1 of Table 4 shows that the quadratic specification of GDP per capita does not work for territorial conflicts. However, it works nicely for nonterritorial conflict onset. The quadratic term of GDP per capita is negative and strongly significant, confirming the existence of a concave relationship between the variables. The probability of having a nonterritorial conflict only starts to decline at the far right of the distribution of country development. In fact, it is beyond the 75th percentile of the income variable where the probability of nonterritorial onset starts to go downward. The richest countries (the top 25 percent) tend to be free of any kind of violence. No country with a per capita income higher than that of the United States in 1971 has suffered the emergence of new domestic nonterritorial violence as defined in this article.
Absence of Conflict versus Nonterritorial Conflicts
Note: GDP = gross domestic product. Country clustered standard errors in parentheses. Spline coefficients not shown.
a p < .1. *p < .05. **p < .01. ***p < .001.
In general, the statistical analysis shows that nonterritorial insurgencies emerge in countries well above the world average GDP per capita. Confronted with strong states, insurgents go underground. A graphical representation of the concave relationship can be found in Figure A1 of the online Appendix A.
Regarding regime, democracies are more likely to experience nonterritorial conflicts (regime is irrelevant for territorial ones). The effect, however, is not a robust one: the coefficient is only significant at 15 percent, and it varies considerably with different specifications of the model. Having said this, the effect of regime is a substantial one: in general, democracies are twice as likely to experience some episode of nonterritorial onset as dictatorships (see Table A2 in the online Appendix A).
State age has opposite signs for territorial and nonterritorial conflicts, though the effect is only statistically significant for nonterritorial ones, which tend to happen in old states. Population is strongly significant for both types of conflict. As for rough terrain, the coefficients are almost identical, but due to different standard errors, the impact of mountains is more significant for territorial violence than for the nonterritorial one (as expected).
When income inequality is added (model 2 of Table 4), it has a positive effect on both types of conflict. More unequal countries have a greater risk of experiencing internal violence. The coefficient is somewhat stronger for territorial violence.
To recap, when we compare the likelihood of territorial and nonterritorial violence onset, we find that the latter is more likely in richer, older countries. Population, rough terrain, and inequality do not discriminate between the two types of conflict. When we analyze the onset of nonterritorial conflicts the picture changes slightly. The relationship with economic development is concave, this violence tends to emerge in richer countries, but not in the richest ones, which are free of any kind of violence. It tends to emerge in old countries, and in democracies rather than in dictatorships, as well as in highly populated and unequal countries.
Robustness Checks
In addition to the analyses reported in this article, we have tried a number of alternative analyses to evaluate the strength of our findings. First, we checked whether the results hold if the dependent variable for the multinomial analysis uses the list of civil wars compiled by Sambanis (2004a) instead of the one by Fearon and Laitin (2003). Second, we checked whether the results are robust to the use of different measures of GDP per capita and regime type. Regarding GDP per capita, we used the Penn World Tables as adjusted for purchasing power parities (Heston, Summers, and Aten 2002). Regarding regime type, we used the yearly country scores of the Scalar Index of Polities (SIP2), compiled by Gates et al. (2006). 13 Third, we explored whether different ways of modeling time dependence alter the results. We followed Urdal (2006) in modeling the probability of conflict as an exponential decaying function, instead of a linear one. Fourth, we tested whether there is some influential case within the list of countries with nonterritorial conflicts. Finally, we also tried alternative estimation techniques, such as rare event logit and random-effects probit models. 14
In general, the concave relationship between GDP per capita and the onset of a nonterritorial conflict is very robust to alternative measures of civil war and GDP per capita, as well as to different estimation techniques. The effect of democracy on nonterritorial violence is less reliable. The ACLP indicator does not reach standard significance levels, but the expected effect shows up when a continuous indicator is used (SIP2). Regarding other independent variables, we obtain very similar estimates across analyses.
An Application: Latin America
We opened this article with a brief reference to the transition from rural insurgencies to urban ones in Latin America. In the period under study, 1970–1997, we observe episodes of both types of insurgency. According to the data sets, we have used (GTD1; Fearon and Laitin 2003), Argentina, Chile, Colombia, Ecuador, Honduras, and Mexico faced nonterritorial challenges, whereas El Salvador, Haiti, Nicaragua, and Peru experienced territorial ones. 15 Can our statistical results account for this variation? And how can we explain deviant cases? By addressing these questions, we try to make sense, in the limited context of Latin America, of some of the previous findings.
We have calculated the probability of each type of conflict in the onset year. For territorial insurgencies, we have used the linear specification of development (model 1 in Table 3) and for nonterritorial ones, the quadratic one (model 1 in Table 4). We compare, among the countries that suffered violence, the probabilities of having either type of conflict. Results in Table 5 show that the models we have estimated are quite helpful to reconstruct the historical path of countries that experimented violence. 16 In 1970, Argentina, given its conditions (GDP per capita, political regime, age, population, etc.) had a 12.3 percent probability of having a nonterritorial conflict onset, compared to a 4.1 percent of having a territorial one. This fits the observed outcome—nonterritorial violence by the Montoneros and other leftist groups. Haiti had a 1.9 percent chance of a territorial conflict starting and only a 0.1 percent chance of a nonterritorial one; the country experienced the former.
In general, Table 5 shows that the statistical analysis performs well in accounting for the different types of conflict in Latin America. In some countries, however, the probabilities for each conflict are very similar. Thus, in Nicaragua and El Salvador, two countries that went through territorial violence, the probabilities for territorial and nonterritorial conflicts are close, though the highest ones correspond to territorial civil wars, as was indeed the case. There are two cases where the model clearly fails: Peru and Honduras. In Peru nonterritorial violence was more likely, but there was a bloody territorial conflict. In Honduras, a territorial conflict was expected, but a nonterritorial one emerged. We shall now provide some potential explanations for Peru and Honduras.
Predicted and Actual Conflicts in Latin America (%)
According to McClintock (1994), Peru was not a likely scenario for a civil war. However, a bloody conflict took place, in which the Shining Path was able to control large areas of the country and a high number of municipalities. The Shining Path started in the remote region of Ayacucho. In-depth research has shown that the administration in this mountainous terrain had no capacity to deal with the insurgents’ challenge (Degregori 1998; De Witt and Gianotten 1994). Before the launching of violence, Shining Path had built a powerful network of support in the villages of Ayacucho (Weinstein 2007, 117). When violence erupted against local authorities and members of the bourgeoisie, the Shining Path was able to replace the state quickly, imposing order in its own way, which gained the support of large segments of the peasantry. When the army intervened, it was too late. Indiscriminate repression alienated the peasants even more and by then the Shining Path was already a formidable military organization. If Peru evolved into a territorial conflict, it was mainly due to successful recruitment and the weakness of the state, whose capacity was probably lower than expected given Peru’s GDP per capita. This is a case where GDP per capita may not work well as a proxy of state capacity.
Honduras has been a puzzle for those working on civil wars: despite being the poorest country in Central America, large-scale violence did not break out in the 1980s (Booth 1991; Brockett 2005, 127). Still, the process of left-wing radicalization in Honduras resembled the formation of guerrilla groups in neighboring countries such as El Salvador. Thus, extreme left-wing urban groups started to carry out attacks against politicians, kidnapped businessmen, and threatened US personnel after the restoration of democracy in 1980, all revolutionary groups coalescing into a single organization named the United National Directorate in 1983 (Schooley 1987, 165, 190). However, the insurgents were unable to go to the countryside and seize territory. And without controlling territory, they were very quickly defeated.
Scholars have underlined two factors to account for this failure. Given the absence of a strong landholding upper class, the military enjoyed relative autonomy within the political realm. This allowed military officers to play the “reformist” card during the 1970s with several land reforms and granted a peaceful transition of power to civilians in 1980. As a consequence of this relative autonomy, the military did not embark on massive programs of indiscriminate repression against potential internal enemies, and this soft strategy avoided the alienation of broad segments of society (Brockett 2005, 250–51; Goodwin 2001, 169–76).
Second, the US government watched with increasing concern the reining turmoil in Central America, and therefore opted to turn Honduras into a “buffer state” in the region. To that end, the United States invested hundreds of millions of dollars, mostly through military aid in the country, and contributed to transforming its soil into a safe haven for the Contra guerrillas fighting against the Sandinista regime in Nicaragua, and also giving military training to Salvadorian military personnel. The massive support of the US forces to the regime very much limited the capacity of the Honduran insurgents to seize territory and doomed them to failure (Brockett 2005, 255; Schooley 1987, 241). In brief, the nonterritorial conflict in Honduras was a truncated cycle of guerrilla war, where insurgents did not go beyond an underground organization.
Our statistical model helps to explain why some left-wing groups in Latin America seized territory and fostered highly lethal guerrilla wars, whereas other groups were obliged to fight clandestinely, using the tactics of underground violence. When the statistical models fail, it is basically because of measurement problems or because of factors we have not considered in the quantitative analysis, such as the implication of third countries in domestic conflicts.
Conclusion
Conflicts featured by nonterritorial insurgencies have been overlooked in the literature on political violence. On one hand, studies on civil wars deal with conflicts with more than 1,000 killings, most of which correspond to insurgencies with territorial control. On the other hand, studies on terrorism go too far in the disaggregation of violence and analyze attacks or incidents, without paying much attention to the actors that carry out the violence. Despite a rich literature on small-n studies, there has not been any quantitative effort to directly compare territorial and nonterritorial armed conflicts.
Our main finding is that nonterritorial conflicts thrive more in richer, but not the richest, countries. There is a concave relation between economic development (understood as a proxy for state capacity) and nonterritorial conflicts: at one extreme, we have poor countries that suffer territorial conflicts; at the other one, rich countries that are violence-proof; in the middle, countries with intermediate levels of development and nonterritorial conflicts. This contrasts with the linear, negative relationship between territorial conflicts and economic development. We also find that nonterritorial conflicts are more likely in old states and in democracies. Population, rough terrain, and inequality have similar effects on both types of conflict.
The distinction between the two kinds of conflicts may shed light on some of the controversies in the literature on political violence. While most territorial conflicts can be assimilated to civil wars, many nonterritorial ones are usually described as terrorist in nature, in which the violence is carried out by underground groups. The association between nonterritorial violence and terrorism is not new in the literature (Cronin 2009, 147; Della Porta 1995, chap. 5; Hoffman 1998, 41; Merari 1993, 224; Sánchez-Cuenca and de la Calle 2009; Weinstein 2007, 17). In the case of civil wars, Sambanis, for instance, establishes some degree of territorial control as a prerequisite for a conflict to be considered a civil war (Sambanis 2004a, 829).
Of course, the correspondence is far from perfect. For instance, many authors consider that the Taliban insurgency today in Afghanistan is a terrorist one, but the Taliban control important areas on the border with Pakistan. Despite cases of mismatch, however, the fact is that when the connection between civil war and territorial conflict, on one hand, and nonterritorial conflict and terrorism, on the other, is made some apparently incongruent results start to make sense.
Regarding economic development, studies on civil war systematically find it has a strong, negative effect (see Mason [2009] for a recent review). This fits our finding for territorial conflicts. By contrast, in studies on international terrorism (Bloomberg, Hess, and Weerapana 2004; Burgoon 2006; Krueger 2007; Li and Schaub 2004; Piazza 2006; Sambanis 2008), GDP per capita usually has no effect, being a rather weak predictor. Most authors conclude that terrorism is independent of economic conditions, but it may well be because the concave relationship between the variables has not been tested.
Although our findings on regime are not robust, they point to some association between democracy and nonterritorial violence. This is consistent also with a wide literature that finds that democracy facilitates terrorism (Bloomberg and Hess 2008; Burgoon 2006; Chenoweth 2010; Eubank and Weinberg 1994; Li 2005; Piazza 2006; Savun and Phillips 2009).
We do not want to imply that types of insurgency (territorial vs. nonterritorial ones) are coterminous with the traditional categories of conflict (civil war vs. terrorism). But we think that they may be useful to establish links between the two and to shed some light on the different determinants of each (see De la Calle and Sánchez-Cuenca 2011). In an age where insurgents are taking advantage of all available warfare techniques to hit the state, it seems wise to better understand how the control of territory influences the dynamics of political violence.
Appendix
This Appendix explains how we have addressed the coding problems regarding the territorial control variable. According to our general rule, there is territorial control when the insurgents are able to do some or all of these things:
Set up camps or bases within the country’s borders in which they store weapons, train recruits, and so on.
Establish stable roadblocks, disrupting the flow of goods and people within the country.
Rule the civil population in the localities they seize (e.g., extracting rents or administering justice). To be recognized as the new authority, insurgents may wear uniforms and carry arms in the controlled areas.
If any (or all) of the preceding three conditions are met, we conclude that the group had territorial control. Otherwise, we classify it as underground.
Some contentious cases are the following:
Some insurgents may have safe havens beyond the country borders. This may range from some degree of territorial control in the foreign territory, such as the Palestinians in Jordan between 1968 and 1970, before Black September, to mere permissiveness regarding the presence of insurgents in the state’s territory, such as ETA’s recruits in the south of France or PIRA members in the Republic of Ireland. As long as these groups do not control any territory within their own country, we do not count them as guerrilla.
A problematical case is the situation described by Kalyvas (2006) as fragmented sovereignty (in which, typically, the state rules during the day and the rebels during the night). This situation, we contend, is only possible if the rebels already have some territorial base somewhere else. Fragmented sovereignty occurs in territory under dispute, but in order for a territory to be disputed, rebels must have some territorial base somewhere else.
Some armed groups have the capacity to coerce the population but without fully replacing the authority of the state. When ETA kills an informer or the PIRA engages in punishing petty criminals, they are competing with state institutions but this does not necessarily prevent citizens from resorting to security forces or to the judicial system. This kind of coercion, typical of some terrorist groups, does not require territorial control as defined earlier.
Finally, we adopt a sort of absorption rule for the case of groups that start as underground organizations and then become full guerrillas with territorial control. The leftist organizations in El Salvador that created in 1980 one of the classical guerrillas in Central America, the Frente Farabundo Martí para la Liberación Nacional (FMLN), started in the 1970s as small clandestine groups. In those cases, we only consider the final outcome: the FMLN is codified as a guerrilla group, regardless of its nonterritorial origins.
Footnotes
Acknowledgments
The authors are thankful for comments from audiences at the Program on Order, Conflict, and Violence (Yale), Midwest Political Science Association Annual Meetings (Chicago), Stanford University, Centro Superior de Investigaciones Científicas (Madrid) and Institut Barcelona d’Estudis Internacionals (Barcelona). They are especially grateful to Neal Beck for methodological advice. Online appendix and replication materials are available at http://jcr.sagepub.com and http://www.march.es/dtv.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This article is part of the research project “The Emergence of Terrorist Groups,” funded by the Spanish Ministry of Science (CSO2010-21704).
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
