Political opportunity structures, democracy, and civil war-

Political opportunity structures, democracy, and civil war- Kristian Skrede Gleditsch University of Essex & Centre for the Study of Civil War, PRIO Andrea Ruggeri University of Essex – Paper prepared for presentation at the 6th SGIR Pan-European International Relations Conference, Turin 12-15 September 2007. A previous version was presented at the 48th annual meeting of International Studies Association, Chicago, IL, USA, 28 February to 3 March 2007. We thank Hein Goemans and HÃ¥vard Strand for helpful discussions and comments. This is a preliminary draft of work in progress; Please ensure that you have the current version before citing. Kristian Skrede Gleditsch (corresponding author) is Reader, Department of Government, University of Essex, Colchester CO4 3SQ, UK (email: ksg@essex. ac. uk). Andrea Ruggeri is a PhD Candidate, Department of Government, University of Essex (email: arugge@essex. ac. uk). Abstract Theories of mobilization suggest that groups are more likely to resort to violence in the presence of political opportunity structures that afford greater prospects for government accommodation or opportunities to topple ruling governments. In empirical studies, however, efforts to test the possible influences of political opportunity structures on incentives for violence have almost invariably relied upon measures of democracy to proxy for the hypothesized mechanisms (e. g., the inverted u-curve hypothesis). We detail a number of problems with measures of democracy as proxies for opportunity structures. We suggest alternative measures based on the likely risks that leaders will lose power in irregular challenges, and evaluate empirically how the security with which leaders hold office influence their prospects for accommodating dissent and the decision to launch insurgencies. We find irregular leader entry and transitions increase the risk of conflict onset, and that democracy has a negative effect on the risk of civil war, once we control for our new measures of state weakness. 1 Introduction The salience of civil war in the contemporary world has generated a surge of interest in its causes. The theoretical literature on civil war has postulated a variety of possible explanations for why governments and insurgents may resort to violence against one another. Some researchers highlight the role of grievances as motives for protest, where the specific grievances can arise from issues such as deprived economic status, lack of political rights and civil liberties, or frustrated expectations (see, e. g., Davies 1962; Gurr 1970). Others emphasize conditions that can facilitate mobilization among potential insurgents, including the role of private benefits or incentives from conflict and the role of state strength in increasing the costs of protests and deterring potential insurgents from launching violent attacks (see, e. g., Collier and Hoeffler 2004; Fearon and Laitin 2003; Tilly 1978). In turn, the empirical literature on civil war has considered a large variety of country characteristics that may reflect either grievances that could give rise to motives for protest or opportunities for resort to violence (see, e. g., Sambanis 2002; 2004 for useful reviews of this literature). In this paper, we revisit arguments about democracy and state strength, and the idea that that political opportunity structure can encourage insurgent violence. We argue that many of the indicators used in empirical studies of civil war are relatively crude indicators of the underlying concepts, and only loosely related to the theoretical rationale. Moreover, since the same operational measure is often used as indicator for a large number of quite different concepts, it is often very difficult to discriminate between different interpretation and theories from the empirical results shown. In this paper, we focus on efforts to use democracy as a combined proxy for both grievances and state strength or repressive capacity. We argue that measures of to what extent institutions are democratic provide a poor indicator of state strength, and that we should seek separate rather than combined measure of the ability of democratic institutions to 2 encourage substitution to non-violent political means and political weakness that may encourage violence. We suggest alternative measures of state weakness based on irregular leader changes and the likelihood that leaders will be vulnerable to challenges from contenders who can seek power. Our results suggest that state weakness, as measured by irregular leader changes, indeed does appear to be associated with civil war onset. Moreover, we find that the risk of civil war depends upon factors influencing the anticipated state weakness as assessed by the risk of irregular leader changes. Although leaders that have entered irregularly can encourage civil war onset as they more likely to be susceptible to challenges and therefore more likely to make concessions to insurgents, the risk of civil war declines with longer tenure. Once we control for these measures of political weakness, we find that democracy has a clear negative effect on the risk of civil conflict onset Mobilization and civil war: State strength, repressive capacity, and political democracy The current literature on civil war has to a large extent focused on identifying government or country characteristics that can make countries more or less prone to civil war. For example, the inverse relationship between a country’s per capita income and civil war and the positive association between country size and civil war are often cited as key empirical facts about what we know about civil war (see, e. g., Hegre and Sambanis 2006). Although research on such country level propositions has provided many useful insights about civil war, the near exclusive focus on countries at large or the government side has led to a relative neglect of the role of nonstate actors in civil wars. Theories of civil war remind us that conflict is at least a minimally dyadic phenomenon. As such, understanding civil war requires us to consider how interaction between governments and non-state actors rather than the attributes of one in isolation influences the risk of civil war (see Buhaug, Cederman, and Rød 2006; Cunningham, Gleditsch, and Salehyan 2006). 3 Although the current literature on civil war emphasizes country level or government characteristics, the earlier literature on violent protest and revolutions preceding the current wave of studies on civil war tended to emphasize the role of participants and mobilization, and the situations in which individuals decide whether to participate in protest or not. Many underlying grievances are either ubiquitous or constant features that change only very slowly over time. Socalled political opportunity structure (POS) theories tend to highlight the role of specific changes or events that provide windows of opportunity for protesters in achieving collective action (see, e. g., McAdam 1982; Tarrow 1994). More specifically, events or changes that decrease the deterrent capacity of the states or make it easier for individuals to achieve collective action should help in providing a more dynamic element to understanding the timing of protest. Most of literature on mobilization has focused on the role of a decline in state strength as an indicator of political opportunity structures (see, e. g., Skocpol 1979; Tilly 1978), although one can also imagine windows of opportunity that make it easier for insurgents to achieve mobilize, such as instances where dissatisfaction becomes revealed to be greater than commonly though and protests can rely on existing networks (see, e. g., Chwe 1999; Opp, Voss, and Gern 1995), or the potential role of entrepreneurs or organizers (e. g., Lichbach 1995). The idea that state weakness or political opportunities can encourage aggression has clear relevance for the risk of civil war as well. Civil war does not originate out of vacuum; Rather, potential insurgents are likely to weight their anticipated prospects of achieving something by resort to violence. Everything else being equal, one would expect weaker states to be more likely to become targeted, either because the insurgents have an opportunity to seize political power directly through toppling the government, or because weaker governments that are vulnerable to other challenges from other competitors — whether from opposing factors or individuals within a ruling coalition — will be more likely to offer some form of accommodation to insurgents (see, e. g., Rasler 1996 on the case of Iran). 4 However, although such “ state strength” or “ political opportunities” plays an important role in many accounts of protest and violent movements, it is much less clear how one would operationalize these concepts, in particular in a cross national setting. Meyer and Minkoff (2004) note that many conceptual discussions of political opportunities tend to leave issues of operationalization unspecified, and that many of the measures that have been suggested in empirical studies tend to be highly idiosyncratic and context specific. For example, McAdam’s use of the decline in lynching in the American South as a measure of political opportunities for African American cannot be generalized to other settings. In perhaps the best-known example from civil war studies, Fearon and Laitin (2003), for example, use GDP per capita as an indicator of state strength. However, this is at best a very indirect measure. Although it may generally be the case that wealthier states are likely to be generally stronger, the example of North Korea attests to how it is certainly not the case that all poor states can be characterized as weak states. Moreover, theories of the resource curse point out that an increase in income actually can decrease state strength and the governability of countries due the effects of wealth on rent-seeking and other disruptive activities (see, e. g., Auty 1993; Ron 2005; Ross 2004). Finally, GDP per capita has been used as an indicator of a number of quite different concepts. Collier and Hoeffler (2004), for example, see higher GDP per capita as an indicator of the opportunity costs of conflict, and GDP per capita could also be seen as an indicator of economic grievances among insurgents rather than a measure of state strength. Although many studies claim to find a relationship between GDP per capita and conflict, it is difficult to evaluate the relative merits of different arguments based on the same proxies. Much of the literature on political violence and civil has approached the issue of state strength through measures of political institutions. Lack of freedom, political rights, and opportunities for political participation can one the one hand be seen as an obvious cause of grievances, which may motivate resort to violence against a government (e. g., Gurr 1970; 5 Schnytzer 1994). This suggests that we should generally observe greater potential for insurgencies under autocratic regimes. By contrast, many have argued that democracies that afford greater opportunities for groups to pursue their aims by non-violent political means, and hence provide plausible substitutes for violence (see Eisinger 1973; Sandler, Tschirhart, and Cauley 1983). However, other researchers have pointed out many autocracies are likely to respond to dissent with harsh repression (e. g., Davenport 1995), and that countries with greater political openness may find it difficult to respond forcefully to violent conflict. Moreover, since regimes with high repressive capacity should be better able to deter conflict (e. g., Lichbach 1995; Tullock 1971), some researchers have argued the relationship between degree of democracy and the risk of conflict will be non-linear and non-monotonic, due to countervailing influences of declining repressiveness and greater opportunities for non-violent political avenues. More specifically, the relationship between degree of democracy should take the shape of an inverted-u, with the greatest risk of violence among semi-democratic countries that combine insufficient repressiveness to deter and insufficient political openness to induce substitution to non-violent activities (see, e. g., Hegre et al. 2001; Muller and Weede 1990). Many studies have indeed found evidence that seem consistent with the inverted u-curve argument. However, this line of research seems problematic for a number of reasons. First, degree of democracy is here used to proxy for two opposing trends. Instead of using the hypothesized implications of democracy as a proxy for repressive capacity or state strength based on assumptions about democracy and substitution, it would be desirable to have direct measures of repressive capacity or state strength separate from the extent to which democracy facilitates substitution to non-violent political means. We will detail one alternative based on information on political leaders and how they enter office below. 6 Second, tests of the inverted u-curve arguments have typically relied on the Polity data, which provide a 21 point scale indicating a country’s degree of democracy based on its institutional characteristics. However, although many studies appear to find evidence that the risk of civil war seems highest for countries with values in the middle of this scale, this finding may follow partly from construction. In particular, Polity contains a large number of observations where regular values on the subcomponents that make up the Polity scale, and the overall Polity scale is given a special code reflecting how institutional characteristics could not be classified due to either “ foreign interruption” (-66), cases of “ interregnum” (-77), defined by Gurr et al. (1990: 6) as periods “ in which there is a complete collapse of central political authority”, or periods of “ transition” (-88). The Polity project has recently released versions of the data where they implement a set of imputations to replace these special codes. More specifically, Marshall and Jaggers (2005) suggest replacing cases of interregnum with a Polity score of 0, linearly interpolating transition periods, and setting periods of foreign interruption to system missing. These imputations are highly problematic, since countries may be coded as being in an “ interregnum” precisely due to conflict and violence (as Gurr et al 1990: 6 note), rather than their institutions being somehow “ in-between” democracy and autocracy. Moreover, transitions rarely follow the smooth pattern assumed by linear interpolation (see Lichbach 1984). Third, Vreeland (2006) suggests that Polity data are problematic in testing arguments about institutions and civil war since two of the subcomponents in the Polity scale — namely, the Competitiveness of Participation (PARCOMP) and the Regulation of Political Participation (PARREG) — can acquire particular values based on whether a country experiences civil war. In particular, PARREG can take on a value of 1 in the event of “ unregulated participation”, or a situation where there is no systematic control of political activity, or a value of 2 for “ factional” polities, with restricted patterns of competition between competing factions. However, Gurr, Jaggers and Moore (1989) explicitly note that “ unregulated participation” could entail “ violent 7 conflict among partisan groups”, and that “ factionalism” is characterized by “ intense, hostile, and frequently violent” competition, which in extreme cases “ may be manifested in the establishment of rival governments and in civil war” (p. 12). Likewise, PARCOMP is coded as 0 in cases deemed “ not applicable” or “ unregulated competition” or 1 for “ factional competition”, which, as noted above, may reflect situations where a country experiences civil war. 1 Vreeland (2006) argues that this risk introducing an inverted u-shape by construction, and that none of the other subcomponents of the Polity index appear to display the hypothesized inverted u-shaped pattern or to be associated with conflict. Finally, anocratic polities with scores in the middle range of the Polity scale are often countries in transition, either on the way to democracy or autocracy, and tend to be less persistent than clear autocracies or democracies (see Gleditsch and Ward 1997; Gurr 1974). Many have argued that political transitions may influence the risk of conflict, and it can be difficult to separate the potential impact of transitions from the impact of anocratic polities per se (Gleditsch 2002; Hegre et al. 2001). Moreover, looking at data on degree of democracy alones provides at best a partial indication of political change and its potential impact on conflict. Any measure that identifies whether countries are democracies treats non-democracies as a residual category, where regimes essentially are defined by what they are not. This risks lumping together a great deal of quite disparate regimes as non-democracies, including socialist states such as the Soviet Union and kleptocracies such as Zaire under Mobutu. Gleditsch and Choung (2006) note how political transitions such as the Iranian and Cuban revolutions would not show up as large 1 These definitions might seem to imply that PARREG and PARCOMP in these cases must have the same values, but this is not the case in the observed data. Although PARREG= 1 implies that PARCOMP= 0,. the reverse is not the case. Moreover, PARCOMP= 1 actually implies that PARREG= 4 (“ restricted”) and only 6% of observations with PARREG= 2 has PARCOMP= 1. 8 changes in the Polity data, despite large institutional changes, since the countries remain “ nondemocracies”. Leader entry and stability as measures of political opportunities In this section we describe how information on leaders can help provide independent measures to assess state strength and political opportunities for potential protesters to mount violent challenges, separate from measures of the degree to which political institutions are democratic. We have previously argued that rebels are strategic, and more likely to resort to violence when they stand a higher chance of achieving some success. As such, we would expect that weaker leaders, vulnerable to challenges from competitors, would be more likely to encourage insurgent violence. Whether a government is “ weak” is sometimes classified based on whether regimes in the end turn out to fall or not. However, this is a post hoc classification that can only be made after the fact. Moreover, it is not necessary for a leader to actually ultimately fall to encourage aggression, and concessions to insurgents may under some circumstances suffice to prevent violent conflict. Although whether leaders fail cannot be known with certainty ex ante, it is not farfetched to argue that insurgents can recognize instances when leaders are likely to be weak and more likely to offer concessions. Moreover, there are clear empirically observable relationships between certain ex ante observable leader characteristics and resulting political stability that we can use to identify such opportunity structures. More specifically, using a new data set on political leaders entitled Archigos, Goemans, Gleditsch, and Chiozza (2006) show that the manner in which a leader enter office is strongly related with how leaders lose office. Leaders may enter power in a regular fashion, or in accordance with prevailing rules and practices through for example election or designation, or in an irregular fashion such as seizing power through a coup. In particular, Goemans, Gleditsch, 9 and Chiozza (2006) show that leaders who have entered power irregularly are more than three times more likely to leave office in an irregular manner. Indeed, irregular exit is the modal form of exit for leaders who have entered power irregularly. Goemans, Gleditsch, and Chiozza (2006) and Lacina (2006) furthermore show that irregular leader entry is positively associated with civil war. However, although irregular leader entry is associated with a higher observed risk, it would be unreasonable to expect that mode of entry should have a constant effect on conflict irrespective of the duration of a leader’s tenure. Rather, we would expect the impact of irregular entry to decay with time. For example, although Castro entered power irregularly, the length of his rule beyond the initial years should suffice to demonstrate the strength of his position. Goemans and Bas (2006) show that the negative impact of irregular leader entry on economic growth declines with time, making leaders who have entered regularly and irregular indistinguishable after an initial window. In this paper we examine the positive effect of revealed political opportunity structures by examining whether civil conflict is more likely in the wake of irregular regime changes where leaders first exit by irregular means and a new leader then assumes power by irregular mean. We expect that irregular transitions would signal windows of opportunity and state weakness that may encourage onset. We then examine the impact of irregular entry and how this varies with time. We expect that leaders who have entered power by irregular means will have a high likelihood of conflict, but that the positive effect will dissipate with longer tenure, indicating more secure leaders. We also consider separately the components of the Polity scale that may underlie the inverted u-shape, more specifically the irregular Polity values (-66,-77,-88) and the values of PARCOMP and PARREG that may reflect conflict. We expect that when we control for state weakness and partition out these components of the Polity scale, democracy should have a negative monotonic relationship with civil war. 10 Empirical analysis Our unit of analysis is the state year, for all independent countries since 1945. Most of our data are available at a yearly level only, although we have more fine grained data for some characteristics such as leaders, political institutional changes, and conflict. We generally use the features in place at the end of the calendar year, when converting these to annual data. Data and measures Our measure of conflict is based Uppsala armed conflict data (see Gleditsch et al. 2002). More specifically, we consider whether countries experience the onset of armed intrastate or conflict claiming more than 25+ deaths in a calendar year. Since we are only interested in new conflicts, subsequent years of the same conflict as coded as system missing. Our measure of irregular transitions and leader changes are taken from the Archigos data. We use Archigos to identify whether the leader in power entered irregularly and the tenure of a leader (measured in days at the end of the year). Obviously not all leader changes imply political transitions. We define as irregular transitions or alterations between distinct coalitions in autocratic regimes all instances where a leader leaves power irregularly and a new leader enters irregularly within a 12 month period. We focus on both exit and entry to avoid situations where a leader is forced to give up power but a constitutionally or designated successor then assumes power in a regular manner. There can be cases where an outgoing leader may exit in a regular manner, but a new leader enters irregularly. Many monarchies, for example, have unclear lines of succession, with the consequences that the exit of one leader may give way to instability as contenders vie for power (hence the phrase “ long live the King”). However, irregular entry is here partly a function of lack of a clear manner of succession, and the new entrants tend to be closely associated with the previous leader. We thus prefer to err on the side of caution and not 11 count these as irregular transitions among autocracies, although they will of course be counted as irregular leader entries where conflict may follow in the wake on leader changes. We use the Polity data to measure the extent to which a country has political institutions that can be characterized as democratic. Despite the recommendation of the Polity data, we do not impute a value of 0 for “ interregnum” or linearly interpolate cases “ in transition”. The predicted value for these polities based on alternative data sources such as the Freedom House data suggest values very close to -10 or most autocratic end of the scale. Hence, we assign these observations a value of -10 instead. We also introduce separate dummy variables identifying observations with irregular Polity values to see if these differ in their expected influence on civil war. Finally, we introduce dummy variables for observations with the values of PARREG and PARCOMP seen as problematic by Vreeland as they may be influenced buy whether countries experience conflict. We also include a number of relevant control variables based on existing studies of civil war that plausibly could be associated with our values of interest. First, we include the log GDP per capita, which is a potentially rival measure of state strength. This is a central control variable, as political instability and state weakness potentially could reflect variation in wealth and income. Second, we include the log of population, as larger countries potentially could be systematically more stable (for example through a screening effect where weaker states disintegrate over time), or have more opportunities for instability, due to their larger or more diverse populations. Finally, we include a count of the success years of peace since conflict or independence, whichever is shorter, since time since the last conflict is known to influence the risk of civil war and could be associated with political instability or state weakness. We do not expect the influence of conflict history py to be linear, and we therefore use an exponential function exp − py / α . Trial and error suggested that α = 4 , provides a reasonable fit to the data. This implies that the original risk of conflict is roughly halved after 2 years. 12 Empirical results We first illustrate the relationship of civil war to irregular transitions through some simple descriptive statistics. As can be seen from Table 1, civil wars often tend to coincide with irregular regime transitions. More specifically, we find that whereas 31% of the observations with irregular leader transitions either in the current or previous, compared to only 14% of the observations without irregulars observations. 13 Table 1: Civil war by irregular transitions Civil war No No transition Irregular transition 6, 568 (86%) 223 (69%) Yes 1, 070 (14%) 98 (31%) Looking at conflict incidence is potentially problematic, as conflict could precede transitions in ways which would make transitions outcomes of civil war rather than “ causes”. However, Table 2 shows that the relationship holds if we restrict attention to civil war onset. We see civil war onset in about 12% of all the cases with irregular transitions, compared to less than 4% of the observations where we do not see irregular transitions. Moreover, the overrepresentation of onset in years with transitions holds even if we lag transitions by one year to ensure that transitions are not due to civil wars breaking out prior to transitions. This supports are arguments that irregular transitions signaling state weakness and political opportunities can encourage aggression. Table 2: Civil war onset by irregular transitions Civil war onset No No transition Irregular transition 6, 568 (96%) 223 (88%) Yes 250 (4%) 30 (12%) Table 3 examines the relationship of civil war onset to changes in the Polity scale that result in changes between democracies and non-democracies, using the threshold for democracy 14 suggested by Jaggers and Gurr (1995). 2 As can be seen, there is some evidence that relative share of conflict onset is higher following transitions. However, the evidence is very weak for transitions to democracy, which have only a marginally higher share of conflict onset than years without transitions. Moreover, the number of transitions on the Polity scale (135) is much smaller than the number of irregular transitions (253). This again supports our claim that although changes in transitions reflected in the Polity scale may tell us something about state weakness or political opportunity structures that could encourage aggression, they leave out a great of relevant political changes and instability, especially among autocracies. Table 3: Civil war onset by Polity transitions Civil war onset No Transition to autocracy No transition Transition to democracy 40 (87%) 6, 575 (96%) 82 (94%) Yes 6 (13%) 258 (4%) 5 (6%) Although suggestive, the bivariate relationships shown so far could be due to third factors that influence both the risk of civil war and political instability and leader changes. We now turn to our multivariate analysis, including factors that are commonly believed to be associated with civil war. Model 1 in Table 4 first presents the results for a null model, without taking into account the impact of irregular transitions or the potentially problematic components of the Polity scale. The results for Model 1 indicate that that the irregular Polity categories of 2 More specifically, countries with a Polity score of 6 and above are considered coherent democracies. Jaggers and Gurr also distinguish between “ coherent autocracies”, which have Polity scores of -6 or below, and “ anocracies”, but we do not consider this difference among non-democracies here. 15 “ interregnum” (-77) and “ in transition” (-88) are strongly positively associated with conflict. This in turn implies that imputation techniques assigning these values or 0 or values toward the middle of the Polity scale through interpolation may generate a seeming inverted u-shape by construction. The coefficient estimate for “ foreign interruption” (-66) is actually negative, perhaps reflecting the fact that states with foreign interruption may be more likely to have conflicts involving other states that are less likely to be coded as civil wars. However, the enormous size of the coefficient suggest a problem of separation (see Zorn 2005), 3 and we omit this category from the remainder of the analysis. Even after taking into account all of the transition categories, the results still suggest an inverted u-shaped relationship between the Polity scale and civil war onset, with a maximum effect when a state has a Polity score of 0. Replacing the quadratic specification with a linear term results in a positive (albeit not statistically significant) coefficient. Hence, the inverted ushaped relationship between degree of democracy and conflict onset thus seems supported by the data, even when we account for the potential problems due to imputation of values on Polity scale for the special categories. 3 The problem of separation refers to cases where some predictors have little or no variation in the response, which makes it difficult to estimate meaningful coefficients. In this case, we only have one case of conflict onset under “ foreign interruption”, namely Uganda in 1979. 16 Table 4: Logit estimates of conflict onset Model 1 Coef SE -4. 492 0. 792 -0. 289 0. 084 0. 0003 0. 012 -0. 0101 0. 003 -11. 848 299. 798 2. 874 0. 447 1. 378 0. 383 Model 2 Coef SE -4. 893 0. 819 -0. 265 0. 086 -0. 035 0. 019 -0. 007 0. 003 2. 226 0. 753 -0. 397 0. 446 0. 376 2. 044 0. 735 383. 6 11 6103 0. 517 0. 461 0. 227 0. 226 0. 042 0. 182 0. 245 Model 3 Coef SE -4. 760 0. 819 -0. 312 0. 084 -0. 048 0. 017 (Intercept) log of GDP pc Polity Polity^2 Polity = -66 Polity = -77 Polity = -88 PARCOMP = 0 | 1 PARREG = 1 | 2 log of population Conflict history Irregular transition Chi-square Df N 0. 378 2. 127 0. 042 0. 181 1. 614 0. 161 -0. 656 0. 499 0. 369 2. 052 0. 779 377. 8 10 6103 0. 444 0. 384 0. 197 0. 226 0. 042 0. 183 0. 245 370. 8 9 6103 In Model 2, we include a term for the presence of an irregular transition in the current or prior years, as well as separate dummy variables for the potentially problematic values of the PARCOMP and PARREG subcomponents. As can be seen, we find a clear positive effect of irregular transitions, suggesting that the log-odds of conflict more than doubles. This is strongly consistent with our claim that such instances of political instability can encourage violent conflict. We find support for Vreeland’s claim that certain values of PARREG that could reflect the presence of civil war are positively associated with conflict onset. However, in the case of PARCOMP, the values identified by Vreeland as problematic are actively negatively associated with civil conflict onset. Finally, we note that the sign for the linear part of the now switches sign and becomes negative. The net implied relationship between civil war is still non-monotonic, since negative values on the polity scale will be rendered positive by the quadratic term. However, the flex point or value where the net impact is the largest is now shifted much lower, to a Polity value of -3. Moreover, we show in Model 3 that if we replace the quadratic 17 specification with a linear term, we get a clear and significant negative coefficient for democracy. Although a model with more terms by construction must fit the data better, we do not normally let all our terms enter models non-linearly, and any gain in the statistical fit of a model must be tempered by the loss in degrees of freedom. As such, it seems reasonable to infer that there is a great deal more evidence of democracy having a negative effect on the risk of civil war onset once we have taken into account the positive effects of conflict from state or leader weaknesses, consistent with the idea that political democracy allows for substitution to nonviolent methods. We now turn to examine how the irregular entry affects the prospects for civil war onset, and whether this effect is stable over the tenure of a leader. We omit the term for the presence of irregular transition for Model 4 in Table 5; Although an irregular transition is not by definition the same as irregular leader entry, we already know that the likelihood of irregular transition will generally be much higher for leaders that have entered irregularly, and decline with longer tenure as leaders that originally entered power in an irregular manner become more secure in office. Including a term for irregular leader entry still suggests that the time around irregular transitions see an increased likelihood of conflict, but the estimated coefficient is obviously smaller when a term for irregular leader entry and time in office is included. We will return to the issue of political opportunity windows due to transitions later. Model 4 indicates a strong positive coefficient for leader entry, suggesting that leaders that have entered power irregularly are almost three times more likely to experience conflict than leaders who have assumed power in regular ways. However, since the model now has an interactive term between irregular entry and time in office, the actual impact of irregular entry will depend on time in office. Figure 1 illustrates how the risk of conflict onset varies by type of leader entry and the length of a leader’s time in office. As can be seen, the differences in risk of conflict onset are very dramatic at the outset of a new leader, where leaders that have entered irregularly have estimated odds of 18 conflict onset almost three times higher than that of a leader who has entered irregularly, everything else being held constant. The risk of conflict onset generally declines with tenure, but the net effect once we take into account the negative interaction between leader entry and time in office suggests that the declining risk of conflict onset with time in office is much greater for leaders who have entered irregularly. Keeping everything else constant, we find that leaders who have entered irregularly but manage to hold onto power for over 15 years have a lower estimated risk of leaders who have entered regularly. Although staying in power for periods as long as this is very uncommon among leaders who enter regularly, the tails of leaders with very long tenure is much higher among leaders who have entered irregularly but able to perpetuate their rule beyond the first couple of years. Table 5: Logit estimates of conflict onset Model 4 Coef SE -5. 017 0. 852 -0. 262 0. 088 -0. 031 0. 02 -0. 007 0. 003 2. 766 0. 781 -0. 456 0. 486 0. 384 2. 019 1. 055 -0. 036 -0. 372 399. 2 13 6093 0. 539 0. 467 0. 231 0. 227 0. 043 0. 186 0. 313 0. 109 0. 167 Model 5 Coef SE -4. 869 0. 851 -0. 307 0. 085 -0. 044 0. 018 Model 6 Coef SE -4. 226 0. 888 -0. 334 0. 088 -0. 048 0. 019 (Intercept) log of GDP pc Polity Polity^2 Polity = -66 Polity = -77 Polity = -88 PARCOMP = 0 | 1 PARREG = 1 | 2 log of population Conflict history Irregular or aut. trans. Irrgular entry log of tenure Entry * log tenure Chi-square Df N 2. 193 0. 216 -0. 702 0. 542 0. 378 2. 02 1. 087 -0. 05 -0. 386 394. 4 12 6093 0. 465 0. 382 0. 199 0. 227 0. 043 0. 186 0. 313 0. 108 0. 167 1. 776 0. 132 -0. 68 0. 577 0. 343 2. 181 0. 369 0. 686 -0. 128 -0. 21 397. 9 13 5915 0. 499 0. 391 0. 203 0. 234 0. 045 0. 193 0. 252 0. 343 0. 114 0. 177 19 1. 0 Irregular entry Regular entry Effect on log odds -0. 5 0 0. 0 0. 5 2000 4000 Tenure (in days) 6000 Figure 1: Effect of irregular entry and time in office on conflict onset Whereas Model 4 still suggests a somewhat curvilinear relationship between Polity and risk of conflict (where the risk of conflict is maximized for a Polity value of -3), Model 5 shows that excluding the square term yields a negative and statistically significant estimate for a linear Polity term. Although the quadratic specification is statistically significant and again fits the data marginally better than the linear term, the differences are sufficiently slight that the linear form can be said to provide a reasonable, yet more parsimonious representation of the data. Finally, Model 6 shows adding the presence of either an irregular transition or a transition to autocracy increases the risk of conflict. Taking a step back, the results of Models 5 and 6 (and the gist of the implications of Model 4) are consistent with our argument that greater democracy affords greater possibility for substitution to non-violent political means than an autocracy, once we have taken into account 20 the effect of political opportunities and state weakness that can be gleaned from information on political leaders. Although we may be able to find support for an inverted u-shape in data on democracy and conflict onset, there is nothing about the institutions of a partial democracy per se that increases the risk of conflict, but rather perhaps something about the ways in which a country comes to be a partial democracy that may be associated with a higher risk of conflict. More specifically, we know that countries that are classified as “ anocratic” in the Polity data often appear to be states where weak leaders who would prefer to rule in an autocratic manner try offer some half-hearted reform in order to strengthen their position, yet are unwilling to offer reforms that would suffice to bring about any meaningful form of democratic rule. Conclusion A large number of studies have examined non-linear specifications of democracy and civil war to proxy for a large number of disparate mechanisms. In this paper, we have shown that it is possible to devise separate measure of the “ opportunities” for violence that emanate from political instability and weaker leaders, as the “ willingness” for violence that stems from restricted opportunities for political participation and advancing political claims by non-violent means. We have argued that data on leader changes can be used to devise independent measure of state strength. Periods of leader changes indicate moments of political opportunities, where potential insurgents may mobilize and resort to violence. Furthermore, information on when leaders enter office as well as how long they have held office can tell us a great deal about how vulnerable leaders will be to challenges, either from rebels directly, or from other contenders who can challenge leaders weakened by domestic dissents. Our results lend strong support to the claim that political opportunities, as measured by irregular political leader changes, indeed does appear to be associated with civil war onset. Moreover, we find that the risk of civil war depends upon factors influencing the anticipated state weakness as assessed by the risk of irregular leader 21 changes. Although leaders that have entered irregularly can encourage civil war onset as they more likely to be susceptible to challenges and therefore more likely to make concessions to insurgents, the risk of civil war declines with longer tenure. Once we control for these measures of political weakness, we find that democracy has a clear negative effect on the risk of civil conflict onset. There is probably nothing about the institutions of partial democracies per se that make them more prone to conflict. Rather, we should shift our attention to how state weakness may compel autocracies both to introduce half-hearted democratic reforms and increase incentives for resort to violence. 22 References Auty, Richard M. 1993. Sustaining Development in Mineral Economies: The Resource Curse Thesis. London: Routledge. Buhaug, Halvard, Lars-Erik Cederman, and Jan Ketil Rød. 2006. Modeling Ethnic Conflict in Center-Periphery Dyads. Paper Presented at the Annual Meeting of the American Political Science Association, Philadelphia, PA, USA, 31 August – 3 September. Chwe, Michael. 1999. “ Structure and Strategy in Collective Action”. American Journal of Sociology 105: 128-156. Collier, Paul, and Anke Hoeffler. 2004. “ Greed and Grievance in Civil War”. Oxford Economic Papers 56 (4): 663-595. Cunningham, David, Kristian Skrede Gleditsch, and Idean Salehyan. 2006. Dyadic Analyses of Civil War Duration. Typescript, Department of Government, University of Essex. Davenport, Christian. 1995. “ Multi-Dimensional Threat Perception and State Repression: An Inquiry into Why States Apply Negative Sanctions”. American Journal of Political Science 39: 683 -713. Davies, John. 1962. “ Toward a Theory of Revolution”. American Sociological Review 27: 5-18. Eisinger, Peter. 1973. “ The Conditions of Protest Behavior in American Cities”. American Political Science Review 81 (1): 11-28. Fearon, James D., and David D. Laitin. 2003. “ Ethnicity, Insurgency, and Civil War”. American Political Science Review 97 (1): 75-90. Gleditsch, Kristian S., and Michael D. Ward. 1997. “ Double Take: Reexamining Democracy and Autocracy in Modern Polities”. Journal of Conflict Resolution 41 (3): 361-83. Gleditsch, Kristian Skrede. 2002. All International Politics is Local: The Diffusion of Conflict, Integration, and Democratization. Ann Arbor, MI: University of Michigan Press. Gleditsch, Kristian Skrede, and Jinhee Lee Choung. 2006. Autocratic Transitions and Democratization. Typescript, Department of Government, University of Essex. Gleditsch, Nils Petter, Peter Wallensteen, Mikael Eriksson, Margareta Sollenberg, and HÃ¥vard Strand. 2002. “ Armed Conflict 1946-2001: A New Dataset”. Journal of Peace Research 39 (5): 615—637. Goemans, Hein, Kristian Skrede Gleditsch, and Giacomo Chiozza. 2006. The Empirical Study of Political Leaders: Introducing Archigos. Typescript, University of Rochester, University of Essex, and University of California, Berkeley. Gurr, Ted R. 1970. Why Men Rebel. Princeton, NJ: Princeton University Press. Gurr, Ted R. 1974. “ Persistence and Change in Political Systems, 1800-1971″. American Political Science Review 68 (4): 1482-1504. Gurr, Ted R., Keith Jaggers, and Will H. Moore. 1989. Polity II Codebook. Center for Comparative Politics, Department of Political Science, University of Colorado, Modified Online Version Available at http://weber. ucsd. edu/~kgledits/Polity. html. Hegre, HÃ¥vard, Tanja Ellingsen, Scott Gates, and Nils Petter Gleditsch. 2001. “ Toward a Democratic Civil Peace? Democracy, Political Change, and Civil War, 1816—1992″. American Political Science Review 95: 33—48. Hegre, HÃ¥vard, and Nicholas Sambanis. 2006. “ Sensitivity Analysis of Empirical Results on the Causes of Civil War Onset”. Journal of Conflict Resolution 50 (4): 508-535. Jaggers, Keith, and Ted R. Gurr. 1995. “ Tracking Democracy’s Third Wave with the Polity III Data”. Journal of Peace Research 32 (4): 469-482. Lichbach, Mark I. 1984. Regime Change and the Coherence of European Governments. Denver, CO: University of Denver Graduate School of International Studies monograph series. 23 Lichbach, Mark I. 1995. The Rebel’s Dilemma. Ann Arbor, MI: Michigan University Press. Marshall, Monty G., and Keith Jaggers. 2005. Polity IV Project: Political Regime Characteristics and Transitions, 1800-2004, Dataset Users’ Manual. Center for Global Policy, George Mason University, Arlington, VA. McAdam, Doug. 1982. Political Process and the Development of Black Insurgency, 1930-1970. Chicago, IL: University of Chicago Press. Meyer, David S., and Debra C. Minkoff. 2004. “ Conceptualizing Political Opportunity”. Social Forces 82 (4): 1457-1492. Muller, Edward N., and Erich Weede. 1990. “ Cross-National Variation in Political Violence: A Rational Action Approach”. Journal of Conflict Resolution 34: 624-651. Opp, Karl-Dieter, Peter Voss, and Christian Gern. 1995. The Origins of a Spontaneous Revolution. Ann Arbor, MI: Michigan University Press. Rasler, Karen. 1996. “ Concessions, Repression and Political Protest in the Iranian Revolution”. American Sociological Review 61: 132-52. Ron, James. 2005. “ Paradigm in Distress? Primary Commodities and Civil War”. Journal of Conflict Resolution 49 (4): 443-450. Ross, Michael. 2004. “ What Do We Know About Natural Resources and Civil War? ” Journal of Peace Research. Sambanis, Nicholas. 2002. “ A Review of Recent Advances and Future Directions in the Literature on Civil War”. Defense and Peace Economics 13 (3): 215-243. Sambanis, Nicholas. 2004. “ Using Case Studies to Expand Economic Models of Civil War”. Perspectives on Politics 2 (2): 259-279. Sandler, Todd, John Tschirhart, and Jon Cauley. 1983. “ A Theoretical Analysis of Transnational Terrorism”. American Political Science Review 77 (1): 36-54. Schnytzer, Adi. 1994. “ An Economic Model of Regime Change: Freedom as a Public Good”. Public Choice 79: 325-39. Skocpol, Theda. 1979. States and Social Revolutions: A Comparative Analysis of France, Russia and China. New York: Cambridge. Tarrow, Sidney. 1994. Power in Movement: Social Movements, Collective Action, and Politics. New York: Cambridge University Press. Tilly, Charles. 1978. From Mobilization to Revolution. Reading, MA: Addison-Wesley. Tullock, Gordon. 1971. “ The Paradox of Revolution”. Public Choice 1: 88-99. Vreeland, James. 2006. Research Note: A Problem with Polity — Unpacking Anocracy. Manuscript, Yale University. Zorn, Christopher. 2005. “ A Solution to Separation in Binary Response Models”. Political Analysis 13 (2): 157-170. 24