The causal mechanism

Analytically, the theoretical argument centers on civilians as relevant actors who do or do not initiate change in civil–military relations (see also Agüero, Reference Agüero1995; Trinkunas, Reference Trinkunas2005). The argument starts from the finding that civilian control of the military is a relevant political goal for all new democracies, including those that developed from civilian dictatorships. During and after the transition from authoritarian rule, civilians must eliminate the military’s remaining prerogatives and establish new or strengthen existing institutions of effective civilian control. Institutional change in civil–military relations, however, will only be successful if the military complies with civilian reform initiatives. The success or failure of institutionalizing civilian control can be explained by the political elites’ ability to contain the military’s political power through ‘control strategies’, which make the military comply with their political decisions and reduce the armed forces’ ‘disposition and opportunity’ (Finer, Reference Finer1962) to resist them.

Civilian strategies vary in the extent of coercion applied against the military and the degree of intrusion into military autonomy. A strategy is robust if it includes a coercive element and intrudes deeply into military autonomy. This includes strategies such as sanctioning, counterbalancing, and monitoring the military. Conversely, a strategy is weak if it neither uses coercion nor intrudes deeply into internal military issues. This includes, for instance attempts to appease the military by offering corporate or personal political, material and ideational benefits to the military. Strategies of intermediate robustness are intrusive but do not include a coercive element, such as selecting the military leadership based on political instead of professional considerations.

We expect that robust strategies have a more profound and lasting impact on the military’s opportunity and disposition to counter civilian control than weaker strategies (see also Trinkunas, Reference Trinkunas2005): monitoring military conduct closely and punishing transgressions raises the costs of opposing the civilians and thus reduces the military’s incentives as well as their capabilities for defending institutional prerogatives and contesting political leaders. Intermediate strategies, in contrast, might reduce the military’s disposition to contest new rules of civilian control but do not impose costs for subversive behavior. Finally, weak strategies neither punish military opposition nor do they mold the military’s preferences, but rather accept the military’s political demands and thus solidify the status quo and reward opportunistic behavior; hence, they do not significantly weaken the military’s disposition and capability to resist institutional change.

The conditions of civilian control strategies

Political leaders’ decisions to employ specific control strategies and their effects do not occur in an historical or political vacuum, but are affected by the resources available to civilians: robust strategies require a better endowment with material and institutional resources, such as political capital, institutional channels, time, or expert knowledge (Trinkunas, Reference Trinkunas2005). This means that even though robust strategies may grant civilians the best leverage against the military, they will not always succeed in enforcing civilian control. In fact, an all-too-robust approach can have unintended consequences and might lead the military to openly challenge civilian authority if civilians do not possess sufficient resources to back up their attempts to enforce military compliance. Hence, civilians might need to resort to weaker strategies.

Therefore, we expect that the strategic choices of civilians are mainly determined by the specific context they are faced with: civilians will choose robust strategies if resources are plenty and weaker strategies if resources are scarce. ‘Context’, of course, is a very encompassing category and the list of all possible substantive factors that have been considered to affect civil–military relations is very long (Feaver, Reference Feaver1999; Pion-Berlin, Reference Pion-Berlin2001; Belkin and Schofer, Reference Belkin and Schofer2003; Kuehn, Reference Kuehn2013). However, not all of these potentially relevant factors can or should be integrated into a single explanatory model. Consequently, we only include potentially relevant factors into the analysis that fulfill three requisites. First, they must refer to the ‘structure’ in which civil–military interactions take place; actors’ behavior, their preferences or the control strategies themselves cannot be a condition. Second, they must be able to affect civil–military relations as resources for rational, strategic choices for control strategies; this excludes all ideational factors such as normative convictions of the officer corps or the civilian decision-makers. Third, theoretical expectations concerning the effects of a given condition must be consistent.Footnote ³

A critical survey of the literature leaves five factors that could plausibly affect the successful application of civilian control strategies (Croissant et al., 2011, Reference Croissant, Kuehn, Lorenz and Chambers2013): (1) the absence of domestic security threats; (2) strong international support for democratic reforms; (3) a strong civil society; (4) the pre-existence of some effective institutions of civilian control before the transition to democracy; (5) and the degree of consolidation of the overall democratic institutions:

1. 1. Internal security threats and domestic armed conflict make civilians dependent on the military’s coercive capacities and thus reduce the willingness and ability of civilians to employ robust control strategies (Alagappa, Reference Alagappa2001). Consequently, we expect that the absence of domestic security threats is a condition for the strengthening of civilian control and that internal security threats lead to weak civilian control.

2. 2. If external actors support the establishment of civilian control, this increases civilians’ ability to employ more robust control strategies and enact them successfully (Bruneau and Trinkunas, Reference Bruneau and Matei2008). Therefore, we expect that international support contributes to the strengthening of civilian control.

3. 3. A strong civil society allows civilians to pursue more robust strategies, as it ‘constitute[s] a powerful safeguard against military intervention’ (Belkin and Schofer, Reference Belkin and Schofer2003: 605; Caparini et al., Reference Caparini, Fluri and Molnar2006). Consequently, we expect that a well-developed civil society leads to the strengthening of civilian control institutions.

4. 4. Civilians will find it easier to employ robust control strategies if they can rely on existing institutions of civilian control that have been established during the authoritarian regime (Agüero, Reference Agüero1995; Cottey et al., Reference Cottey, Edmunds and Forster2002). Thus, we expect that the historical legacies of a former civilian dictatorship will lead to strong civilian control, while legacies of military rule will lead to weak civilian control.

5. 5. Finally, the literature suggests that civilians will be better able to employ robust control strategies if the broader process of democratic consolidation supports them, that is, when institutional checks-and-balances limit executive power, popular support for democracy is strong, and unified political elites agree not to pull the military into politics (Belkin and Schofer, Reference Belkin and Schofer2003; Serra, Reference Serra2010).

To reiterate, these factors do not affect the degree of civilian control directly, but they are mediated through civilian control strategies. Furthermore, the context in which civilians decide on their control strategies is complex: individual factors might interact in complex causal ways or even offset each other, and causal relations between individual context factors and the outcome of civil–military relations might be asymmetric, such that some factors are sufficient for civilians to successfully employ robust control strategies, whereas others are necessary. As of now, civil–military relations theory is too weak to solve these issues analytically and to generate clear and unambiguous causal hypotheses beyond the expectations formulated above. This has implications for the most appropriate methodological approach to test the theoretical argument.

Case selection

In order to maximize the robustness and generalizability of our findings, we ideally should evaluate the development of civil–military relations in all newly democratized nations. Since tracing civil–military relations along the five dimensions introduced above requires in-depth research, and fsQCA relies on profound familiarity with the cases, we decided to draw on a stratified subset of all ‘third wave’ democracies. Cases to be included were chosen based on a three-step selection approach: first, we identified as the universe of cases all countries with a population of at least 500,000 that have made the transition to democracy between 1974 and 2010 according to the Polity IV dataset (Marshall and Jaggers, Reference Marshall and Jaggers2013).Footnote ⁴ From these, we purged all externally induced and supervised transitions (e.g., Iraq, Afghanistan, Cambodia, East Timor), as well as countries with no permanent armed forces (Haiti, Panama). We further removed countries with 3 years or less of democratic rule during the research period. This yielded a total of 67 countries from all geographical regions except Northwestern Europe (see online Appendix 2). Second, we differentiated these 67 countries into four world regions: Asia, Africa, Latin America, and Europe. Third, we classified countries in each of these groups based on the type of the outgoing authoritarian regime into military-led and civilian dictatorships based on Geddes et al. (Reference Geddes, Wright and Frantz2014). From each sub-group, we selected four countries for empirical analysis, except for those regions where the diversity of authoritarian regime types was limited, where we selected all military/former civilian regimes (see Table 1).

Table 1 Case selection

While not a selection criterion, it is worth mentioning that this sample also maximizes heterogeneity with respect to the outcome of institutionalizing civilian control, and in terms of the outcome of the democratization process at large. It includes stable democracies, where civilians were able to gain full control over the military (e.g., Greece, Taiwan, Portugal, and Poland), and countries where the failure to establish civilian control contributed to the sudden breakdown of democracy (e.g., Mali, Pakistan, Honduras, and Thailand).

Operationalizing civilian control of the military

To evaluate the degree of civilian control, we undertook extensive and in-depth surveys of the scholarly literature on civil–military relations for each of the countries. While the quantity and quality of the existing scholarship is uneven, at least 15 articles, book chapters, and monographs were consulted for each country. These data were complemented by additional information gathered from legal texts, interviews, and newspaper articles. Based on these data, we inductively identified ‘phases’ of civil–military relations in which significant changes in the degree of civilian control took place. Each of these phases constitutes a single ‘case’ for the analysis. All in all, this yields a total of 57 cases (see online Appendix 2). For countries that experienced a breakdown of democracy in the period under review (e.g., Russia, which Polity codes democratic from 2000 to 2007), or with multiple breakdowns of and transitions to democracy (e.g., Sri Lanka, from 2001 to 2003, and again from 2006 to 2010), we only included the democratic spells as ‘phases’ in the analysis.

The transformation of this qualitative information into coherent fuzzy set membership scores of the degree of civilian control in a given ‘phase’ followed a three-step approach. First, each ordinal value received a suitable fuzzy set score: low civilian control was coded 0, medium as 0.4, and high civilian control as 1. Second, the individual indicator values were aggregated into five sub-values of civilian control, one for each decision-making area [ER; PP; IS; ND; MO] by averaging their scores. Finally, we calculated the mean of these five values to receive a single measure of civilian control for each case. While all five areas matter for civilian control, we consent with recent scholarship on civil–military relations (Pion-Berlin, 1992; Trinkunas, Reference Trinkunas2005) that certain decision-making arenas are more important than others, depending on their proximity to the core of a civilian government’s function. To capture this, we combine the individual scores as a weighted average of the fuzzy set values of the five decision-making areas. Given that, on the one hand, civilian dominance over the rules and processes of elite recruitment (ER) is central for democracy, but on the other hand, even the complete absence of military influence over these matters is insufficient for fully fledged civilian control, we weigh it such that full civilian control over ER in combination with at least some degree of civilian control over the other areas is sufficient for the case to be a member of the set of countries with ‘civilian control’ (i.e., an aggregate fuzzy score of 0.5 or more). Similarly, civilian control over PP and IS is more relevant to the overall degree of civilian control than civilian control over ND and MO.Footnote ⁵ This yields the following formula for the overall membership value in the set ‘civilian control’:Footnote ⁶

$${{ER{\times}5{\plus}PP{\times}2{\plus}IS{\times}2{\plus}ND{\plus}MO} \over {11}}$$

Operationalizing the conditions of civilian control

Our model includes five conditions that affect the ability of elected civilians to employ control strategies and thus strengthen civilian control: internal security threats; international support; civil society; authoritarian civilian control institutions; and democratic consolidation:Footnote ⁷

1. i. We operationalize internal security threats (inthreat) based on the CONIAS dataset (Schwank, Reference Schwank2013). This dataset includes country-year data on domestic conflicts for all cases and differentiates five intensities of conflict, from 1 (non-violent latent conflict) to 5 (civil war). We calibrate ‘inthreat’ according to the conceptual thresholds of the CONIAS dataset: we code the absence of internal armed conflicts (CONIAS value 1) as absence of a substantial internal threat (fuzzy score 0). We consider limited or all-out internal wars (CONIAS values 4 or 5) as the presence of a meaningful internal threat (fuzzy value 1). Following CONIAS, we define the degree of violence as the central criterion for set membership of meaningful vs. not meaningful internal threat (fuzzy score 0.5). Consequently, we code non-violent crises (CONIAS intensity of 2) as more in the set of ‘no meaningful threats’ (fuzzy score 0.3), and low intensity conflict (CONIAS value of 3) as more in the set of ‘meaningful threats’ (fuzzy score 0.7). As we expect civilians to be risk averse, we draw on the maximum intensity a conflict reaches in a given phase to code the complete phase.

2. ii. The literature on civil–military relations suggests different forms of support civilians can receive from external actors, ranging from aid, international military co-operation, to membership in supranational organizations (Bruneau and Trinkunas, Reference Bruneau and Matei2008). In order to capture this broad range of international influence, we operationalize the condition ‘intif’ as an index of three individual conditions: (1) membership or credible expectation of membership, based on the beginning of membership application procedures, in NATO or the EU as organizations that demand strict civilian control of the military from member states (Barany, Reference Barany1997).Footnote ⁸ We code the condition as a crisp set (i.e., score of either 1 or 0), with 1 denoting (credible expectation of) membership, and 0 denoting non-membership in either organization. (2) The presence of US military forces in the country as a proxy indicator of direct military co-operation, based on the Global US Troop Deployment dataset (Kane, Reference Kane2004). We code this partial condition as a crisp set, with presence of US military troops in at least half the years of the phase sufficient to count the case within the set of cases with presence of US troops (membership score 1), otherwise, we code the case as outside that set (score 0). (3) The literature also suggests that a country’s trade or aid dependence can be used by external donors or trade partners to push for stronger civilian control. We measure international leverage as the sum of trade and aid ratios, drawing on World Bank data (Wahman et al., Reference Wahman, Teorell and Hadenius2013). We code this partial condition as a crisp set: a case belongs to the set of high dependency countries (membership score 1) if the average sum of trade and aid ratios within a phase is 0.5 or higher, otherwise, we code the case as outside that set (score 0). Because the literature is too weak to generate robust expectations on the interrelationship between these forms of international support, we assume first, that international influence will have a stronger impact on the ability of civilians to employ robust strategies the more forms of international influence combine; and second, that each of these partial conditions contributes equally to the overall degree of international influence. Consequently, we code the index condition ‘intif’ as fully out of the set of countries with high international influence (fuzzy score 0) if the sum of the fuzzy scores of the three partial conditions is 0, we code it 0.3 (more out of the set) if the sum is 1, we code it 0.7 (more in the set) if the sum is 2, and we code it 1 (fully in the set 1), if the sum is 3.

3. iii. The operationalization of the strength of civil society (civsoc) follows Belkin and Schofer (Reference Belkin and Schofer2003: 606). They suggest that in the absence of direct and reliable data on the strength of domestic civil society organizations, the ‘number of associational memberships that individuals and groups maintain in international nongovernmental organizations (INGOs)’ is a reliable proxy indicator. To measure this indicator, we draw on the annual data handbooks of the Union of International Associations (various issues). Missing data was interpolated linearly. Because there are no substantively or conceptually sound justifications for qualitative set membership thresholds, we follow Schneider and Wagemann’s (Reference Schneider and Rohlfing2012: 36) suggestion to identify notable ‘gaps’ in the empirical distribution and use them as ‘qualitative anchors’. Plotting the number of INGO memberships (online Appendix 4), we found three such ‘steps’ in the distribution, and defined the center of these ‘gaps’ as the thresholds. Thus, we defined 756 INGOs as the threshold for being out of the set of strong civil society (fuzzy score 0), while the fuzzy score 0.5 is located at 2370, and 4339 marking full membership in the set of countries with strong civil societies (fuzzy score 1). Based on these qualitative anchors, we then coded the condition according to Ragin’s ‘direct’ logistic transformation method of calibration (Ragin, Reference Ragin2008).

4. iv. In the absence of reliable data on the degree of civilian control under authoritarian regimes we operationalize the legacies of the authoritarian institutions of civilian control as a proxy condition (autinst) developed from the type and the degree of institutionalization of the last regime prior to the transition to democracy: (1) in military regimes the institutions for controlling the military-as-institution are less developed and effective than in civilian dictatorships (Agüero, Reference Agüero1995). The measurement of regime type draws on the classification of military or civilian dictatorship by Cheibub et al. (Reference Cheibub, Gandhi and Vreeland2010). (2) Personal rulers assure control over the military less by institutional means but informally and through patron–client relations. Consequently, democracies that follow a personalist regime cannot make use of institutional resources, but have to build them up from scratch, which poses greater challenges to the employment of robust control strategies. We measure personalization of authoritarian rule based on the Geddes et al. (Reference Geddes, Wright and Frantz2014) dataset, and consider all regimes that include the ‘personal’ characteristic as personalized, all others as institutionalized regimes. Combining these two dimensions, we expect that personalized military regimes will provide the least institutional resources for post-authoritarian governments and code them as fully out of the set of cases with authoritarian control institutions (membership score 0). Institutionalized, non-personal civilian dictatorships are coded as fully in that set (score 1); institutionalized military regimes are more in the set (score 0.7) and personalized civilian regimes are more out of the set (score 0.3).

5. v. Democratic consolidation denotes a complex and multi-dimensional process that involves multiple analytical and substantive levels, including the consolidation of democratic institutions, the interactions and patterns of conflict of political elites, and the expectations and trust in the democratic institutions of the mass public (Linz and Stepan, Reference Linz and Stepan1996). Moreover, because consolidation is generally considered to be a process, it is hard to specify exactly when it has occurred (Svolik, Reference Svolik2008). In order to operationalize this concept, we draw on the age of the democratic regime as a necessarily crude proxy indicator for the degree of consolidation.Footnote ⁹ This is substantively based on Rustow’s (Reference Rustow1970) argument, that democratic institutions and actors’ expectations and behavior become stable over time. In defining the relevant qualitative thresholds, we draw on recent research on military coups, which finds that in new democracies coups are particularly likely to occur in the first 5 years after the transition to democracy, while ‘the risk of a coup almost disappears once a democracy survives for two decades’ (Svolik, Reference Svolik2014: 5). Consequently, we code the age of the democratic regime at the end of each phase: all democracies 5 years or younger are considered to be fully out of the set of consolidated democracies (fuzzy score 0), while democracies aged 20 years or older are coded as fully in that set (fuzzy score 1). The qualitative cut-off (fuzzy score 0.5) was defined as a democratic age of 10 years, which is based substantively on Valenzuela’s (Reference Valenzuela1992) argument that the minimum threshold for democratic consolidation is that at least one democratic election has taken place after the founding elections. Based on these qualitative anchors, we then coded the condition according to the ‘direct method of calibration’. In the following, we denote this condition as ‘age’.Footnote ¹⁰

Conditions for civilian control

The first step in any QCA analysis is to test if any of the conditions might be necessary. In fsQCA, a condition is necessary if the outcome can be described as a consistent subset of a condition. Following Ragin (Reference Ragin2000), we set the minimal consistency benchmark for necessity at a level of 0.9. We find that none among the five conditions or their negations is a necessary condition for the presence of civilian control. This is an important empirical finding, as it not only corroborates our expectations, but also goes against the tendency in some of the literature to consider certain variables as preconditions for civilian control (Barany, Reference Barany2012).

To identify sufficient conditions, that is, those causal factors or combinations of factors that bring the outcome about, we create a truth table, which assigns each case to exactly one of the 2^k logically possible combinations of conditions (so called ‘types’, which define the rows in the truth table), with k being the number of conditions. For our set of cases, the truth table has 2⁵=32 rows. Table 2 reproduces an abbreviated truth table for the outcome civilian control, purged from all logical types for which there are no empirical cases. Table 2 shows that the 57 cases are distributed over 20 out of the 32 logically possible combinations of conditions. The remaining 12 types are ‘logical remainders’, that is configurations of conditions for which there are no empirical cases. To identify the conditions or combination of conditions that are sufficient for the outcome civilian control, we need to define a minimal consistency value, which should not be smaller than 0.75 (Ragin, Reference Ragin2000). Inspecting the truth table, we notice a substantial drop in consistency between types 26 and 18. Including cases below that threshold would significantly reduce overall consistency of the solution. Consequently, we decided to set the consistency cut-off at 0.8 and only include the rows above that threshold in the Boolean minimization to identify the sufficient (combinations of) conditions for the outcome civilian control. The minimization produces a ‘complex solution’ of eight individual combinations of conditions, which is highly consistent (0.867) and covers a relatively broad range of the cases (coverage 0.755; see Table A in Appendix 7).

Table 2 Summary of conditions for civilian control

Table 3 Truth table for the outcome civilian controlFootnote ^a

inthreat=internal threat; autinst=authoritarian institutions; civsoc=strength of civil society; age=democracy’s age; intif=international influence.

^a Type identifies a given combination of conditions; n specifies the number of cases that fall into a given type; outcome states if the cases that fall under a given type exhibit the outcome civilian control (1) or not (0); consistency gives the consistency value; cases shows the id number of the cases that fall under a given type as shown in Appendix 3.

To reduce the complexity of these results, we follow Ragin (Reference Ragin2008: 160–175) and derive the ‘intermediate solution’ by specifying assumptions about the expected effect of the individual conditions on the outcome. Based on our theoretical discussion above, we assume that the presence of each individual condition will contribute to civilian control, except the presence of an internal threat, which we expect will lead to the absence of civilian control. This yields a solution term that identifies one individual condition (the presence of a strong civil society, CIVSOC) and three combinations (democratic consolidation together with international influence, AGE*INTIF; existing authoritarian control institutions and democratic consolidation, AUTINST*AGE; and the combination of high internal threat, the presence of existing control institutions of the authoritarian regime and international support, INTHREAT*AUTINST*INTIF) as individually sufficient for civilian control. The overall solution consistency of 0.855 and overall coverage of 0.803 indicate a good model fit. We summarize the full solution term along with consistency and coverage values and the cases that are members of each partial solution in Table 4.Footnote ¹²

Table 4 Intermediate solution of sufficient conditions for civilian controlFootnote ^a

Intermediate solution consistency: 0.855.

Intermediate solution coverage: 0.803.

inthreat=internal threat; autinst=authoritarian institutions; civsoc=strength of civil society; age=democracy’s age; intif=international influence.

^a Capital letters in the column ‘Sufficient combination’ indicates that the presence of the combination is sufficient for the outcome; * denotes logical AND; + denotes logical OR.

The findings suggest five conclusions concerning the theoretical expectations formulated above. First, as expected, there is no single condition to explain the establishment of civilian control in new democracies. Instead, different institutional and structural variables can serve civilians as resources to successfully enact control strategies. This is also underscored by the fact that the four different partial solutions each have a relatively small unique coverage term, which indicates a high degree of overlap between the different (combinations of) conditions. Second, all conditions discussed above are part of the solution term. This means that all of them can provide civilians with the necessary institutional, structural, and political resources to enforce their authority over a hesitant military.

Third, only the existence of a strong civil society suffices on its own for the strengthening of civilian control in new democracies. All other conditions are only efficacious in combination with each other: solid democratic consolidation can unfold its positive influence on civilians’ ability to employ robust control only if combined with strong international support or if civilians can also make use of the existing institutions of civilian control that are the historical legacies of the preceding authoritarian regime. Similarly, the existence of historical legacies of civilian control can only provide sufficient resources if combined with international support or an overall smooth process of democratic consolidation. We cannot answer why exactly these combinations and not others are sufficient for civilian control in new democracies because the cross-case fsQCA does not shed light on within-case causal processes. Similarly, the current state of civil–military relations theory does not provide sufficient explanatory leverage to answer this question. Consequently, we consider our study as an explorative step in better understanding the conditions of civilian control and their interaction, which requires further and more qualitative process-oriented research.

Fourth, four of the five conditions work in the theoretically expected directions and there are no contradictions in the causal efficacy of a strong civil society, existing authoritarian control institutions, democratic consolidation, and international influence.

Fifth, it is only in the fourth combination that an unexpected relationship is observed, with the presence of an internal threat being part of one partial solution term: the combination of internal security threats, existing authoritarian institutions, and international support is sufficient for the existence of civilian control. 37 out of 57 cases in our analysis (65%) are confronted with an internal security threat, of which 23 (62%) do belong to the set of cases with civilian control. At this point there is no plausible explanation for this finding, as both the theoretical literature as well as our previous empirical research (Croissant et al., Reference Croissant, Kuehn, Lorenz and Chambers2013) strongly indicates a detrimental effect of internal threats on civilians’ ability to establish civilian control. This finding, therefore, sheds some doubt on the conventional wisdom and strongly demands further qualitative research.

To test the robustness and validity of these findings, it is useful to visually inspect the distribution of cases in an XY-graph that plots the membership of each case in the solution term against its membership in the set of civilian control. Figure 1 presents such an XY-plot for the intermediate solution term. The diagonal line signifies the 0.5 threshold for membership in both sets, and allows for identifying necessary and sufficient solutions: for the solution term to be a consistently sufficient solution, all cases must be situated above the diagonal; for a consistently necessary condition, they need to be below the bisecting line. The horizontal and vertical lines denote membership in the outcome set of civilian control (all cases above the horizontal line are members of that set), and in the intermediate solution set (all cases right to the vertical line are members of that set). The plot visualizes, first, that the intermediate solution is actually consistently sufficient for the outcome: the majority of cases (34 out of 57, about 60%) are located under the diagonal, and the 23 non-consistent cases are not too far from that dividing line. Second, all cases located in the upper right quadrant of the plot and above the diagonal can be considered ‘typical’ cases for the solution (Schneider and Rohlfing, Reference Schneider and Wagemann2013).Footnote ¹³ Third, of all deviant cases, that is cases that are situated in the right half of the XY-plot (X>0.5) and below the diagonal, only seven cases are ‘deviant in kind’, that is cases whose degree of civilian control is considerably lower than their membership in the solution term would suggest.Footnote ¹⁴

Figure 1 XY-plot of intermediate solution term: civilian control.

Conditions for the absence of civilian control

To identify the conditions for the absence of civilian control in new democracies, we follow the same steps as above. The analysis of possible necessary conditions demonstrates that none of the five conditions qualifies as a strong necessary condition for the absence of civilian control. A consistency score of 0.935 indicates that the absence of a strong civil society is a necessary condition for weak civilian control; however, with only 0.529, its coverage is rather low. In other words, only a very small proportion of all cases that exhibit weak civilian control also has a weak civil society (see online Appendix 6). Consequently, we do not consider weak civil society as a necessary condition for the failure of civilians to strengthen civilian control. Turning to the identification of sufficient conditions for the failure to strengthen civilian control in new democracies, Table 5 presents the truth table for the five conditions and the outcome ‘no civilian control’.

Table 5 Truth table for the outcome no civilian control

inthreat=internal threat; autinst=authoritarian institutions; civsoc=strength of civil society; age=democracy’s age; intif=international influence.

We define 0.8 as the consistency cut-off point. Based on the cases that are part of the types above this cut-off point, the minimization produces a marginally consistent ‘complex solution’ of three individual combinations of conditions with relatively low coverage (see Table C in online Appendix 7). Grounded on our expectation that the absence of all conditions except internal threats will weaken civilians’ ability to employ robust control strategies and lead to weak civilian control, we derive an intermediate solution term that identifies three configurations as sufficient for weak civilian control: the lack of authoritarian control institutions combined with a weak civil society and unconsolidated democracy (autinst*civsoc*age); a weak civil society plus unconsolidated democracy and a lack of international support (civsoc*age*intif); and the combination of significant internal threat combined with a lack of pre-existing control institutions, an unconsolidated democratic regime, and the absence of international support (INTHREAT*autinst*age*intif). However, while the consistency values of the individual partial solutions are acceptably high, the overall consistency of the intermediate solution (0.753) indicates that the three sufficient configurations are not particularly strong predictors of weak civilian control. Similarly, the solution coverage is only 0.642, which indicates that the solution does not capture a large share of the variance of the cases with weak civilian control. We summarize the full solution term along with consistency and coverage values and the cases that are members of each partial solution in Table 6.Footnote ¹⁵

Table 6 Intermediate solution of sufficient conditions for NO civilian controlFootnote ^a

Intermediate solution consistency: 0.753.

Intermediate solution coverage: 0.642.

inthreat=internal threat; autinst=authoritarian institutions; civsoc=strength of civil society; age=democracy’s age; intif=international influence.

^a Small letters in the column ‘Sufficient combination’ indicates that the absence of the combination is sufficient for the outcome; * denotes logical AND; + denotes logical OR.

This analysis points to three insights. First, the significantly lower consistency and coverage scores show that the five structural and institutional factors perform worse in explaining the absence of civilian control in new democracies than in accounting for its presence. This suggests that another theoretical model might be necessary to explain the failure to strengthen civilian control in new democracies, and that future research should consider including additional conditions into the analysis. The insight that different causal arguments might be necessary to explain the negation of the outcome is in line with existing research on democratization: Svolik (Reference Svolik2008), for instance, argues that explaining the breakdown of democracy is explained by different factors than democratic stability and consolidation. Second, there is no single structural or causal factor whose absence (or presence) alone will totally inhibit the ability of civilians to strengthen civilian control of the military. The results, rather, suggest that the failure to strengthen civilian control is due to a ‘syndrome’ of pathologies and unfavorable conditions that interact in constraining civilians’ ability to employ robust control strategies. In fact, even domestic armed conflict, which many authors view as the most unfavorable restriction (e.g., Finer, Reference Finer1962; Alagappa, Reference Alagappa2001), will only lead to weak civilian control if a new democracy also lacks pre-existing institutions of civilian control, is unconsolidated, and does not receive meaningful support from the outside. Third, the analysis shows that our theoretical expectations hold: all five conditions affect the absence of civilian control as we expected them to do and there are no results that contradict our theoretical model.

Figure 2 demonstrates that the solution is much less consistent than the solution to the outcome ‘civilian control’ (see the Conditions for civilian control section): While there are relatively few cases below the diagonal (13 out of 57), these are much farther from the diagonal, and thus, pose a much starker contradiction to the theoretical expectations. Furthermore, there are only four cases that could be considered typical. This explains the relatively low coverage of the solution term, as the majority of cases (situated in the upper left quadrant) reduce the solution coverage. These cases have high outcome scores, but relatively low membership scores in the solution term. Moreover, 11 cases can be considered deviant from a consistently sufficient condition.Footnote ¹⁶ These cases, which are situated in the right half of the XY-plot (X>0.5) and below the diagonal, are particularly interesting to analyze in-depth for improving the theoretical model.

Figure 2 XY-plot of intermediate solution term: no civilian control.