From ballot-boxes to barracks: Votes,institutions,and post-election coups

Abstract

The military often intervenes in politics shortly after elections. This might be because election results reveal information about the ease with which a coup can succeed. Would-be coup perpetrators use this information to infer whether the incumbent can be removed from office without provoking popular unrest. We argue that the informational content of elections depends on the electoral rules that translate votes into outcomes. In electoral systems that incentivize strategic voting, election returns are less informative about the distribution of political support than in electoral systems that incentivize sincere voting. An extensive battery of statistical tests shows that vote-shares of election winners do not predict coup attempts in plurality systems, which encourage strategic voting, but they do predict coup attempts in non-plurality electoral systems, which do not encourage strategic voting. Thus, incumbents who have performed well in elections face a lower risk of coup attempts, but only in institutional environments where voting results are highly informative about the distribution of political support. We apply this logic to illuminate the decisions of the military to intervene into politics during the famous failed 1936 coup in Spain and the successful 1973 coup in Chile.

Keywords

civil-military relations coup elections institutions military

If war is too important to be left to the generals, politics may be too contentious to be left to the civilians. At least this is what frequent military forays into the political arena suggest. But when and why does the military actively intervene in politics? In his landmark work, Finer (1975) argued that the military must have both opportunity and motive to intervene. There are multiple reasons why a clique of regime insiders or military officers may want to depose the government, but it is less clear what precisely constitutes an opportune moment for them to do so. A common refrain is that the military finds it easiest to challenge a government when the latter lacks ‘legitimacy’ (Nordlinger, 1977).¹ While this answer is plausible, it is fundamentally vague because it is ‘difficult to prove empirically whether deposed governments were legitimate or not’ (Wiking, 1983: 31).

Perhaps the most direct way to empirically operationalize the notion of legitimacy is by assessing a government’s endorsement by citizens in elections. Figure 1 (left panel) displays the frequency of coups from 1946 to 2009 broken down by the number of months since the closest preceding election. Strikingly, the largest number of coups occur in the immediate aftermath of elections. One could object that we observe this relationship only because there are but a few countries with long time intervals between elections. The right panel of Figure 1 rules out this interpretation. It depicts the average

Figure 1.

Elections and the timing of coups from 1946 to 2009

number of coups per month across five time intervals, broken down by quintiles. Coup risk shrinks as time passes since the last election. Thus, elections alone do not automatically prevent or precipitate coups. Instead, the data invite a more critical inspection of why some elections produce incumbents that are more vulnerable to coups than others.²

Recent scholarship has made substantial progress in this area. We now know that elections may elicit coups when the incumbent’s electoral support shrinks relative to the opposition or when elections prompt protests (Wig & Rød, 2014). This work is in line with more general theoretical literature arguing that elections offer an opportunity to gauge a regime’s degree of popular opposition, thereby inferring how easily it might be deposed, even if elections are non-democratic and feature manipulation and fraud (Little, 2012; Miller, 2015; Rozenas, 2016). We also know that elections can solidify autocratic regimes in the long run while increasing political instability in the short run (Knutsen, Nygård & Wig, 2017). Intensification of political conflicts (including coups) around the time of elections could be due to elections serving as ‘focal points’ in the struggle for political power (Fearon, 2011).

Our article builds upon this line of work by focusing on how elections can inform would-be coup plotters about the ease with which incumbents can be removed from office. The key point we make is that information contained in election results depends on the manner in which votes translate into political power. As in Wig & Rød (2014), we purport that election returns offer cues to coup plotters, but we argue that the information that coup plotters can infer from election results depends on electoral institutions as well as the informational environment. When electoral rules encourage strategic or ‘insincere’ voting, winning elections provides at best a weak signal that the office holder enjoys an extensive popular mandate to rule. Conversely, when electoral institutions do not encourage strategic voting, the winner’s vote-share serves as a more accurate signal of his true support. Therefore, the risk of a post-election coup should decrease as the incumbent’s electoral support increases under such electoral systems (e.g. proportional representation, majoritarian, or mixed). But under plurality rule – where strategic voting obfuscates the informational content of election returns – the relationship between vote-share and coup likelihood should be less pronounced.

This ‘institutionalist’ explanation of post-election coups makes three contributions. First, we contribute to the literature on post-election coups (Wig & Rød, 2014) by showing how election outcomes interact with institutions in impacting the calculus of coup-perpetrators. The literature on electoral institutions is well established (Duverger, 1954; Cox, 1997), but the insights of this literature have yet to be leveraged for understanding coups. We do so here. Second, we contribute to the literature on the short-term impact of elections on political volatility (Collier, 2009; Knutsen, Nygård & Wig, 2017). Whereas this literature has drawn

Figure 2.

Pinochet’s gamble: Electoral support and approximate ideological positions of the candidates in Chile’s 1970 presidential elections

connections between elections and short-term political instability, we identify some political and institutional conditions under which elections are likely to exert such negative effects (or alternatively, conditions under which elections are more likely to produce stability). Finally, the existing literature has shown how various signals of regime illegitimacy – poor economic performance (Kim, 2016), public protests (Casper & Tyson, 2014; Wig & Rød, 2014), or lack of access to political office through democratic procedures (Lehoucq & Pérez-Liñán, 2014) – increase the risk of coups. Our results suggest that electoral victory is insufficient to secure legitimacy and safeguard against coups when electoral institutions can allocate political power on the basis of a weak popular mandate.

We start with a motivating case of the 1973 coup in Chile. We then present the theoretical framework, predictions, and the empirical evidence. Finally, we discuss the historical case of pre-civil war Spain, which illustrates the mechanisms behind the argument, but also reveals its limitations.

Motivating example: The Chilean coup of 1973

Socialist party leader Salvador Allende lost the 1964 Chilean presidential election to Eduardo Frei. Although Allende earned 38.9% of the popular vote, it was not enough to best Frei’s 56.1%. Six years later Allende would become Chile’s next president by winning only 36.6% of the vote. Ironically, Allende earned more of the popular vote in defeat than in victory – an outcome enabled by a plurality voting system. In a field of three, Allende surpassed the next man – right-wing Nationalist Party leader Jorge Alessandri Rodríguez – by a mere 39,000 votes. Finishing third was center-right candidate Radomiro Tomic with 28.1% of the vote.³ As Figure 2 shows, the ideological alignment of the three candidates was such that Rodríguez and Tomic likely split the right-wing vote, thereby opening a path to office for Allende (Valenzuela, 1978).

The narrowness of an election bringing to power a socialist regime in Latin America prompted the following dispatch to Washington DC from the US ambassador to Chile, quoted in Kissinger (1970: 653):

Chile voted calmly to have a Marxist-Leninist state, the first nation in the world to make this choice freely and knowingly […] His margin is only about one percent but it is large enough in the Chilean constitutional framework to nail down his triumph as final. There is no reason to believe that the Chilean armed forces will unleash a civil war or that any other intervening miracle will undo his victory. It is a sad fact that Chile has taken the path to communism with only a little more than a third (36 percent) of the nation approving this choice, but it is an immutable fact.

The ambassador’s ‘immutable fact’ endured for a full two years – ultimately undone by General Pinochet’s coup d’état in the fall of 1973. Even though the presidential elections delivered a polarizing candidate who was not endorsed by the military, they alone did not precipitate a coup. Interestingly, the coup transpired only six months after another round of elections – Chile’s parliamentary election of March 1973. The parliamentary elections employed the proportional representation rule and they clearly showed that the two right-wing parties, which opposed Allende, were substantially more popular than the left-wing parties, which supported Allende (see Table I).

The outcome of the parliamentary elections indicated very starkly that by the spring of 1973 the incumbent president was likely opposed by the majority of the population. If the generals suspected Allende’s political capital to be weak in 1970, the elections of 1973 clearly showed that it was all but expended. If the military’s move was a gamble that the right-wing would back Allende’s removal, this bet certainly paid off: the military junta not only deposed Allende, but also received public support from both the Christian Democratic Party and

Table I.

Parliamentary elections in Chile in March 1973

	Party	Support
Right-wing (49%)	Christian Democrat Party of Chile	28%
	National Party	21%
Left-wing (38%)	Socialist Party of Chile	18%
	Communist Party of Chile	16%
	Radical Party	4%
Others		12%

Data from Political Database of the Americas (http://pdba.georgetown.edu).

the Nationalist Party. The results of Chile’s 1973 election may have sealed Allende’s fate.⁴

The Chilean episode illustrates how election outcomes and the institutions under which those outcomes were generated interact in shaping the calculus of coup-making. But is this episode a fluke or does it represent an example of a more systematic pattern? To answer this question, we begin by developing a theoretically plausible link between election results, electoral institutions, and coup onset.

Elections, institutions, and coups

The military attempts a coup only when ‘the expected rewards of the maneuver and its probability of victory are high enough to offset the dire consequences of a failed putsch’ (Powell, 2012: 1019). If coup perpetrators anticipate their actions will provoke a backlash from the population, they would also estimate a coup to be less likely to succeed and more likely to escalate into an outright civil conflict.⁵ Thus, coup plotters require an accurate assessment of the genuine support of the governments they seek to depose. If elections indicate that incumbents enjoy broad support, the expected value of attempting a coup should be low, and vice versa.

The general argument that perceptions of a government’s popularity matter in coup-calculus is not new, and can be traced back to Juan Linz who put it succinctly: ‘it seems unlikely that military leaders would turn their arms against the government unless they felt that a significant segment of society shared their lack of belief’ (Linz, 1978: 17). Similar conceptions of regime legitimacy have been invoked by other scholars (Finer, 1975; Nordlinger, 1977; Wiking, 1983). If coup perpetrators are reluctant to risk removing an unpopular government, upon what form of social analysis might they draw? Recently, Wig & Rød (2014) provided one interesting answer to this question by showing how election results (measured in terms of an increase in the opposition support relative to a previous election) are predictive of post-election coup attempts. We propose that, in addition to election results, we should consider the institutions under which those results are generated.

Electoral institutions matter for coup-calculus because election returns carry different informational content depending on the rules that translate votes into seats. It is a widely accepted finding that in elections under the plurality rule – where a party (or candidate) only needs a plurality of votes to win office – voters have strong incentives to vote strategically rather than sincerely (Duverger, 1954; Cox, 1997). That is, in such elections, voters tend not to vote for their most-preferred candidate if that candidate is not a viable competitor for office. In other electoral systems, like proportional representation, majoritarian run-off systems or mixed electoral systems, it is less likely that a vote for an unviable candidate will be wasted; thus the incentives to vote strategically are weaker than in the plurality system.⁶

Consider how the plurality system differs from its key counterpart – a majoritarian run-off system. In run-off systems, winning office requires securing a majority of votes; if no candidate receives a majority, the top candidates proceed to a second round for a run-off, thereby ensuring the winner has a majority. Both of these systems are non-proportional (because the ultimate victor earns all the spoils of office), but they have one profound difference: because of greater strategic voting, plurality elections can be won by a candidate (or a party) who would actually lose against any other candidate in a one-on-one race; under a majoritarian run-off system such an outcome is impossible (Niou, 2001). The technical term for a candidate who would lose in a pairwise election against every other candidate is a ‘Condorcet loser’. A

Table II.

Election returns, institutions, and information

		Electoral rule
		Plurality	Non-plurality
Winner’s vote-share	Low	Ambiguously weak mandate	Weak mandate
	High	Ambiguously strong mandate	Strong mandate

Condorcet loser can prevail in plurality elections, but he cannot prevail in majoritarian elections.

Why does this matter? Consider the calculus of coup perpetrators weighing a putsch against a government approved in elections under plurality rule. Because such a government can be a Condorcet loser, its victory under the plurality rule may signal that in a two-way contest between the incumbent and the military, a majority of the population is likely to stand behind the coup perpetrators (or at least not to object to the coup). Moreover, even if the government is not a Condorcet loser, prevailing in elections under plurality rule may not send a strong signal that the government enjoys a broad mandate. This is because strategic voting obfuscates the mapping of actual voter support to election outcomes. By contrast, in elections under majoritarian rules, the winner cannot be a Condorcet loser and is by definition assured to have a majority of support in the population. A large vote-share under majoritarian rules can send a more accurate signal that the winner of elections is widely supported in the population than the same vote-share in plurality elections.

The above argument can be extended to accommodate the proportional representation (PR) system. Since proportional representation allocates seats in direct proportion to votes, it does not encourage as much strategic voting as the plurality rule. Hence, the results of elections under proportional representation more or less reflect the population’s actual distribution of support for parties or candidates. Victory under a PR system may deliver a strong mandate; however, a narrow win in such a system may reveal a divided populace.⁷

Table II summarizes our argument. In plurality systems, due to strategic voting, the winner’s vote-share serves only as an ambiguous signal of his popular strength. In other electoral systems that do not encourage as much strategic voting, the winner’s vote-share serves as a highly informative signal of his support. Accordingly, when election returns are informative, good electoral performance can deter coups, whereas bad electoral performance can encourage coups. Thus, our first empirical prediction is the following:

Prediction 1: The propensity of a coup attempt is negatively associated with the winner’s vote-share after elections held under non-plurality rules, but not (or less so) after elections held under the plurality rule.

Clearly, the overall informational environment in which elections take place varies from country to country. When information regarding regime popularity is readily accessible, the military may not need to rely on electoral returns to infer the ripeness of their opportunity. However, when elections constitute only one of a few sources of information regarding regime legitimacy, coup perpetrators should rely more heavily on election results. Therefore, we expect the vote-share to affect coup propensity heterogeneously depending on the pre-existing informational environment.

Prediction 2: The winner’s vote-share is associated with coup propensity only in environments that are information-poor.

It is important to note certain limitations of the above argument. First, for theoretical parsimony and empirical tractability, we do not consider other potential sources of ambiguity that arise in elections, like coalition formation after multiparty elections or ideological differences between incumbents and the military. Second, our theoretical argument and its empirical treatment do not take into account how parties that compete in elections may represent various societal coalitions that can make it very difficult for the military to extract relevant information from election results. As our analysis of the Spanish case below shows, these considerations are important, but remain beyond the scope of our treatment.

Empirical evidence

In this section, we subject our predictions to a series of empirical tests. The data on coups come from Marshall & Marshall (2010) and cover the time period between 1946 and 2009. As our theory concerns the opportunities for the military to initiate coups, we do not distinguish between successful coups and coup attempts that failed (we refer to both as ‘coups’). Global election dates are taken from the NELDA dataset (Hyde & Marinov, 2012), which covers most national elections from 1945 to 2010.

The two key independent variables in our analyses are the vote-share of the incumbent in the preceding elections and the electoral system under which the incumbent was selected. To measure the former, we use the largest vote-share of the candidate or the party.⁸ This variable was created using multiple printed sources and internet resources.⁹ We analyze all elections for which we could gather data and where the largest party received at least 25% of the votes. For election rules, we used the Democratic Electoral Systems dataset (Bormann & Golder, 2013) and the IAEP dataset (Wig, Hegre & Regan, 2015), which we also supplemented with information from Nohlen, Grotz & Hartmann (2001), Nohlen, Krennerich & Thibaut (1999), and Nohlen (2005, 2010). We classify an electoral system as ‘plurality’ if it employs a first-past-the-post rule, block vote, party block vote, limited vote or single non-transferable vote – all versions of a plurality rule where a winner does not have to receive majority support and where strategic voting is encouraged (Farrell, 2011).

We adjust for a number of covariates that are standard in the literature on coups. To capture wealth effects we include the log of GDP per capita from Gleditsch (2002) and Teorell et al. (2016). We also include Polity IV scores from Marshall, Gurr & Jaggers (2013) ranging from –10 (autocracy) to +10 (full democracy). And because economic underperformance is often credited as a predictor of coups (Londregan & Poole, 1990; Powell, 2012), we construct a variable measuring yearly annual GDP growth. We lag all these covariates by one year to mitigate endogeneity concerns. Because a country with deeper democratic traditions might be less susceptible to coups but also less likely to have elections where incumbents win excessive majorities, we control for a country’s consecutive years of democracy (lagged). We do so using the binary measure of political regimes in Cheibub, Gandhi & Vreeland (2010), and set the variable to zero if the country is not a democracy. Since the informational content of elections may also be affected by electoral manipulation, we further control for election fraud using the variable from the NELDA dataset indicating whether there were pre-election concerns of electoral fraud (Hyde & Marinov, 2012). Additionally, since coups tend to cluster in time (Londregan & Poole, 1990), we add a variable measuring the number of days since the last coup. Finally, to account for secular time trends in coup propensity, we add cubic regression splines for time.

Probit regressions

We employ two modeling approaches: a flexible probit regression and survival analysis. In the probit analyses, the outcome variable is an indicator, Post-election coup_i , equal to 1 if there is a coup attempt within two years after election i. We use this two-year cutoff motivated by our theory: election results should only affect coup onset in close vicinity to elections. Our results remain qualitatively very similar if we define post-election coups when they occur within one, three, or four years after elections (see Online appendix). Note that if election i is followed by another election i + 1 within two years, then election i + 1 initiates a new spell. Again, this is grounded in our theoretical expectation that a new round of elections generates new information.

Testing Prediction 1 requires a specification that allows the association between vote-share and coup onset to vary by electoral system. To avoid strong functional assumptions, we fit the following semi-parametric probit regression:

Pr {Post - {election coup}_{i} = 1} = Φ (α \cdot z_{i} + f_{z_{i}} (v_{i}) + {x^{'}}_{i} β),

where Φ denotes the standard normal distribution function. The indicator variable z_i represents the type of electoral rule used in election i: if the electoral rule is non-plurality, then z_i = 0, otherwise, z_i = 1. Therefore, the coefficient α stands for the predictive effect of the electoral rule (irrespective of the election outcome) on coup onset.¹⁰ The variable v_i refers to the vote-share of the largest party or candidate. The functions f₀ and f₁ are two unknown smooth functions determining how vote-shares map into the probability of a post-election coup. Here, f₀ represents the function in non-plurality systems

Table III.

Probit regressions for post-election coup attempts

	Model 1	Model 2	Model 3
Vote-share AME (non-plurality)	−0.10** (−0.16, −0.05)	−0.13** (−0.19, −0.07)	−0.10** (−0.17, −0.05)
Vote-share AME (plurality)	0.00 (−0.07, 0.09)	−0.02 (−0.11, 0.07)	0.00 (−0.07, 0.10)
Plurality rule	−0.08 (−0.32, 0.17)	−0.04 (−0.30, 0.22)	−0.03 (−0.29, 0.23)
Polity score (lagged)	−0.03** (−0.05, −0.01)	−0.01 (−0.03, 0.02)	−0.01 (−0.03, 0.02)
GDP per capita (lagged)	−0.36** (−0.47, −0.25)	−0.34** (−0.47, −0.22)	−0.30** (−0.43, −0.17)
Age of democracy (lagged)		−0.02* (−0.03, 0.00)	−0.01 (−0.03, 0.01)
Election fraud		0.36** (0.09, 0.63)	0.31** (0.04, 0.58)
Growth (lagged)		−0.11** (−0.20, −0.02)	−0.10** (−0.20, −0.01)
Time since last coup			−0.14** (−0.22, −0.06)
Time trends (cubic splines)	Yes	Yes	Yes
Deviance explained	12.5%	15.9%	17.5%
Observations (elections)	1,458	1,383	1,383
Countries	144	143	143

AMEs represent averaged marginal effects. Other estimates represent regression coefficients. 95% confidence intervals in parentheses, clustered by countries. Significance levels: **p < .01; *p < .05; ^† p < .1 (two-tailed).

and f₁ represents the function in plurality systems. We expect f₀ to be a decreasing function and f₁ to be either a ‘flat’ function or ‘less decreasing’ than f₀. (The two functions need not be linear.) The model is estimated in a generalized additive model framework using thin plate regression splines (Wood, 2006). Finally, x_i is a vector of control variables, which also includes time trends approximated by natural cubic regression splines.

The advantage of the semi-parametric approach is that we do not need to rely on a strong functional assumption of linearity. The disadvantage is that we cannot present the main results in terms of regression coefficients. We circumvent this problem in the following way: for each electoral system, we report the sample average marginal effect (AME) of vote-share. The AME under electoral system z (with z = 0 and z = 1 standing for non-plurality and plurality system, respectively) is defined as:

AME(z) = \frac{1}{n_{z}} \sum_{i : z_{i} = z} \frac{\partial}{\partial v} Φ (\hat{α} \cdot z + {\hat{f}}_{z} (v_{i}) + {x^{'}}_{i} β),

where n_z is the number of elections under the electoral rule z in our sample. Thus, the marginal effects are estimated for each observation in the dataset and then averaged across each electoral system. Intuitively, AME(z) represents the expected marginal effect of vote-share on the probability of a coup for a randomly drawn election under electoral system z. We compute confidence intervals and p-values for the AMEs using Monte Carlo simulations.

Table III shows results from three models. Model 1, the baseline specification, includes lagged Polity IV scores and lagged measures of GDP per capita (logged). Model 2 adds the age of democracy, annual growth in GDP per capita in the year prior to the election, and the indicator for election fraud. Finally, Model 3 includes a measure of the time since the last coup, measured in days (on the logarithmic scale).

The main quantities of interest are the average marginal effects (AMEs) of vote-shares in plurality and non-plurality elections. The AME of the vote-share in non-plurality elections is negative and statistically significant. In contrast, the AME for the vote-share in plurality elections is effectively zero across all three specifications. In substantive terms, we can think about the magnitude of these effects as follows. If we randomly draw an election from our dataset and calculate the marginal effect of increasing vote-share for that election (conditional on the covariates and the observed vote-share) assuming that the election is not held under a plurality system, this effect would be equal to −0.10 (based on Model 1); but if the same election were held under a plurality system, this effect would be essentially equal to zero. Adding additional covariates in Models 2 and 3 does not alter the main conclusions, as the estimated AMEs change very little across specifications.

The estimated coefficients for the covariates are in line with prior studies. Coups are more likely to occur in poorer and in less democratic countries, but the effect of democracy (as captured by Polity scores) attenuates substantially once we control for election fraud. Coups are also more likely after fraudulent elections. Furthermore, coups are less likely to occur after elections preceded by a growing economy and after elections that were not preceded by other coups.

To explore more closely the relationship between vote-share and coup onset, Figure 3 depicts the predicted probabilities of a coup attempt conditional on the vote-

Figure 3.

Predicted probabilities of coup attempts

share under the two electoral systems. The left panel of the figure shows how the sample-averaged probability of a coup attempt changes as the vote-share of the largest party increases in plurality systems. The figure clearly reveals that coup propensity does not vary with vote-share when elections are held under the plurality rule. The right panel of the figure depicts the equivalent graph for non-plurality systems with remarkably different patterns. The probability of a coup attempt is strongly and monotonically decreasing in vote-share. In substantive terms, increasing the vote-share of the largest party from 40% to 60% reduces the probability of a coup attempt from about 0.10 to 0.08. Although the effect might appear minor, one should bear in mind that the overall risk of a post-election coup (within two years after an election) is very small – equal to about 0.07. In that context, the 0.02 point reduction is quite substantial.

As a robustness check, we conducted other estimations and tests (see Online appendix). We employed a one-year, three-year, and four-year cutoff to define post-election coups (instead of two years), but these changes do not meaningfully alter the results. We added region-specific controls and controls for a country’s colonial past, to alleviate concerns that colonial legacies impacted electoral institutions and coup propensity. We included country random effects¹¹ and fixed year effects instead of cubic splines for time. We also estimated the above models by excluding one by one individual countries or regions, and excluding elections that were fully non-competitive. In all these cases, the main results remained very similar. Despite these additional measures, however, one should be cautious not to overinterpret our results as definitive evidence of a causal effect of electoral institutions and outcomes on coup onset.

Survival analysis

The preceding probit analysis faces two shortcomings. First, it ignores the actual timing of coups – as long as a coup happens in a fixed time window, it is treated in the same way irrespective of whether it occurred five days or five months after an election. Second, it does not address the censoring problem: since each new election starts a new spell, the data are right-censored. We address these shortcomings with survival analysis.

We first consider the proportional hazards model (Cox model). Here, the outcome of interest is the hazard rate of a coup at time t, denoted by $h (t | θ_{i})$ , where i stands for an election. This hazard rate is modeled as the product of $h_{0} (t)$ , the baseline hazard rate, and the mean θ_i such that $h (t | θ_{i}) = h_{0} (t) θ_{i}$ and θ_i depend on the interaction of the vote-share and the electoral system (as well as the covariates):

ln θ_{i} = α_{0} (1 - z_{i}) v_{i} + α_{1} z_{i} v_{i} + α_{2} z_{i} + {x^{'}}_{i} β

Here, z_i is an indicator equal to 1 if election i is run under plurality rule and 0 otherwise; v_i is the vote-share; and x_i is the vector of covariates. This is simply an

Table IV.

Survival regressions for coup attempts

	Cox model	Weibull	Log-logistic
Vote-share (non-plurality)	−1.11* (−2.09, −0.13)	1.82* (0.21, 3.44)	1.88* (0.22, 3.55)
Vote-share (plurality)	−0.02 (−1.06, 1.01)	0.01 (−1.70, 1.72)	0.07 (−1.73, 1.87)
Plurality rule	−0.51 (−1.45, 0.43)	0.94 (−0.58, 2.46)	0.92 (−0.64, 2.48)
Polity score (lagged)	−0.01 (−0.05, 0.03)	0.01 (−0.05, 0.07)	0.01 (−0.06, 0.07)
GDP per capita (lagged)	−0.52** (−0.71, −0.34)	0.82** (0.53, 1.11)	0.86** (0.55, 1.17)
Age of democracy (lagged)	−0.03^† (−0.06, 0.00)	0.05^† (−0.01, 0.10)	0.04 (−0.01, 0.09)
Election fraud	0.23 (−0.17, 0.62)	−0.34 (−0.99, 0.31)	−0.47 (−1.13, 0.19)
Growth (lagged)	−0.09 (−0.21, 0.03)	0.13 (−0.06, 0.32)	0.17 (−0.03, 0.38)
Time since last coup	−0.26** (−0.37, −0.15)	0.39** (0.19, 0.58)	0.42** (0.21, 0.63)
Time trends (cubic splines)	Yes	Yes	Yes
Log likelihood	−918.95	−1447.21	−1441.64
Observations (elections)	1,389	1,389	1,389
Countries	143	143	143

95% confidence intervals in parentheses, clustered by countries. Significance levels: **p < .01; *p < .05; ^† p < .1 (two-tailed).

interactive model with linear terms for vote-shares: if election i uses the plurality rule, z_i = 1 and the effect of vote-share is captured by the coefficient α₁; if election i uses a non-plurality rule, z_i = 0 and the effect of vote-share is captured by the coefficient α₀.

The Cox model assumes that the baseline hazard rates are proportional to the time of survival. To ensure that this assumption is not driving our results, we also estimate two accelerated failure time models (AFP). In these models, the dependent variable is the time (measured in days) to a coup following election i, denoted by T_i. The regression model is given by $ln (T_{i}) = ln (θ_{i}) + ∊_{i}$ , where θ_i takes the same interactive form as in Equation 3 and the $∊_{i s}$ are the residuals. We consider two common distributions for these residuals – the Weibull and log-logistic distributions. We estimate the survival models using methodology and software by Therneau & Grambsch (2013).

The results are shown in Table IV. Column 1 shows the estimated coefficients for the Cox proportional hazards model. The two quantities of interest are the coefficient estimates for the vote-share in non-plurality and plurality elections. The vote-share has a negative and statistically significant effect on the hazard rate of a coup in non-plurality elections. By comparison, the coefficient for vote-share in plurality elections is an order of magnitude smaller and statistically insignificant. Holding all else constant, raising the vote-share by 0.25 points in a non-plurality system reduces the predicted relative hazard of a coup by a factor of exp(−1.11 × 0.25) = 0.76, or 24%. By comparison, the effect on the relative hazard in plurality systems is equal to exp(−0.02 × 0.25) = 0.99, or 1%.

In columns 2 and 3 we show estimated coefficients from two AFP models. The signs of the estimated coefficients in column 1 are the opposite of those in columns 2 and 3 because the outcomes of the two types of regressions are inversely related (hazard rates in column 1 and survival time in columns 2 and 3). The coefficient estimates in both AFP models tell a similar story to the Cox model. In the Weibull regression, the coefficient for the vote-share in non-plurality elections is equal to 1.82, which implies that a 0.25 increase in the vote-share increases the predicted time without a coup by a factor of exp(1.82 × 0.25) = 1.6, or 60%. The respective effect in plurality elections is drastically smaller, and statistically not significant. The same results are obtained with the log-logistic specification (column 3).

In summary, the evidence of the probit and survival analysis points in the same direction. First, there is no consistent statistical relationship between post-election coup attempts and the vote-share of election winners under plurality rule. Second, vote-share is strongly and negatively associated with post-election coup attempts in non-plurality elections. Consistent with Prediction 1, coup plotters are more sensitive to information in election returns produced under non-plurality rules than under the plurality rule.

Mechanisms

According to our argument, vote-shares can help predict coup onset because of their informational content. We now investigate the plausibility of this informational mechanism. Our first probe into the mechanism is a direct test of Prediction 2 given in the theoretical section. The prediction is that vote-shares should not be

Table V.

Average marginal effects (AMEs) of the vote-share in subsamples

	Plurality		Non-plurality
	AME	95% CI	AME	95% CI	N
Democracy	0.01	(–0.20, 0.19)	–0.01	(–0.14, 0.11)	813
Non-democracy	–0.02	(–0.28, 0.21)	–0.16^*	(–0.30, –0.01)	569
Polls available	–0.42	(–3.72, 0.81)	0.01	(–0.12, 0.13)	638
Polls unavailable	–0.10	(–0.28, 0.09)	–0.15^**	(–0.27, –0.05)	738

The estimates control for baseline covariates. Significance levels: **p < .01; *p < .05; ^† p < .1 (two-tailed).

associated with coup onset in information-rich environments (where actors can gauge political support for parties even in the absence of elections), but they should be associated with coup onset in information-poor environments (where elections constitute one of very few sources of information).¹²

We discriminate information-rich from information-poor environments in two ways. First, we posit democracies are more likely to be information-rich than autocracies. This is on account of freedom of the media and the ability of citizens to openly express their preferences in democratic contexts. Such outlets serve as extensive sources of political information.¹³ Second, and more directly, information provided by elections is of less value when pre-election public opinion surveys exist. Thus, if such an informational mechanism is at work, we propose the following. Vote-share (under non-plurality rule) should be at best weakly negatively associated with coup onset in democracies and elections with polls. However, we anticipate a strong negative association between coups and vote-share in non-democracies and elections not preceded by polling. In sum, an information-rich setting should obviate a dependency on elections to discern a government’s popular appeal.

To test these hypotheses, we re-estimate the probit specifications by slicing the dataset in two different ways: elections held under democracies vs. non-democracies (using the binary democracy measure (Cheibub, Gandhi & Vreeland, 2010)) and elections with and without public opinion polls (based on the nelda25 variable in the NELDA dataset (Hyde & Marinov, 2012)). In Table V, we show the average marginal effects of the vote-share in plurality and non-plurality systems in the four subsets of the data that result from this slicing.¹⁴ Consistent with prior results, vote-share does not have a statistically significant effect on coup onset in plurality elections across all four subsets. However, the vote-share does have a negative and statistically significant association with coup onset in non-plurality systems, but not in democracies and not when polling information was available. These two pieces of evidence speak to the plausibility of the informational mechanism.

Our second test is premised on the idea that whatever informational content elections carry should dissipate with time. The incumbent party might have garnered a large margin in an election that could have deterred coup plotters in a short period after the election, but it is unlikely that these effects would persist for long into the future. Thus, if the informational mechanism is driving our results, vote-share should be a strong predictor of coups in a short period after elections, but a weaker predictor of coups in time periods further away from elections.

To investigate this mechanism, we start with the semi-parametric probit regression where the dependent variable is equal to 1 if a coup occurs within 12 months after an election. We then shift this time window forward by a month, and estimate the AMEs of the vote-share for non-plurality systems; we repeat this procedure 12 times. Figure 4 shows the results: the average marginal effects are negative, substantial in magnitude, and statistically significant when we predict coups occurring within 18 months of an election. However, the vote-share becomes a less reliable predictor of those coups transpiring further away from elections.

Figure 4.

Average marginal effects of vote-share in non-plurality systems for different time intervals of post-election coups (with 95% confidence intervals)

In sum, when other sources of information are missing, an incumbent’s vote-share serves as a valuable signal of how easy he can be removed from office. Collectively, these findings indicate strong support for the informational mechanism relating election outcomes and the risk of a coup.

A historical application: Pre-civil war Spain

We now examine the case of pre-civil war Spain to investigate how the electoral rules influenced the calculus and timing of the military’s decision to leave the barracks in 1936, but to stay put in 1931. This analysis is best qualified not as a rigorous empirical test, but as an idiographic theory-guided case study (Levy, 2008), which allows us to assess how well the above theory can explain this very important and puzzling historical event.

In April 1931 the Spanish military, which had ruled the country with the backing of the monarchy since 1923, called municipal elections to restore constitutional rule. These elections would turn into a referendum on the monarchy itself (Borkenau, 1963). With few exceptions, the socialists and liberal republicans captured every provincial capital in Spain. Electoral results in hand and throngs of Spaniards filling the streets of Madrid, the last hope for the Spanish throne was its longstanding bastion of support: the military. However, the generals stood down when, on 14 April 1931, the Republic was proclaimed. The denouement of the Spanish monarchy in the 1930s is puzzling in terms of its timing. Certainly the military was no friend of the republican alliance into whose hands the regime had fallen. It was inconceivable that the army should long remain aloof from the highly contentious political realm. The motive to strike was certainly present. However, the coup d’état did not transpire in 1931, when the outgoing monarchists in fact invited it, but rather in 1936. The puzzle, then, is explaining why the generals chose to lie low in 1931 when Spanish rule shifted inexorably to the liberals and socialists.

We argue that an inspection of the three Spanish Cortes elections between 1931 and 1936, and the rules governing them, suggests the generals moved against the regime when their opportunity for success was greatest. According to Thomas (2001), the constituent Cortes held on 28 June 1931 was the fairest election that had been held in Spain. It also suggested the majority of the people were behind the new regime. The top of Table VI shows the results of these elections. In all, the left and center-left won 400 of the 470 seats in the Cortes (Beevor, 2006). The non-republican right fared terribly, gaining only 57 seats. The Catholic party, also inclined to the right, won a mere six seats. The new republican government – formed by Manuel Azana – received a clear mandate for its rule.

With roughly 84% of the total votes cast, the socialists, republicans, and liberals won 89% of the seats. The largest party owed its majority representation to changes in Spanish electoral law favoring coalitions. The Spanish electoral system improvised in May 1931 was in response to the old plurality system which favored local bosses (‘caciques’). The new hybrid system, however, preserved some undesirable criteria of the system it was intended to replace. All the single member districts were replaced

Table VI.

Votes and seats in the Spanish elections

Political bloc	Votes	Seats
1931 provisional Cortes elections
Left (socialists, republicans, and liberals)	84%	89%
Communists	1%	0%
Right (Catholics and monarchists)	15%	11%
1933 Cortes elections
Socialists	20%	13%
Republicans	14%	8%
Communists	2%	0%
Center-right (liberals)	30%	37%
Right (Catholics and monarchists)	34%	42%
1936 Cortes elections
Popular front (communists, socialists, republicans)	47%	60%
Center-right (liberals)	22%	15%
Right (Catholics and monarchists)	31%	25%

Data from Colomer (2004).

with small and medium multi-member districts with limited ballot – or what is called an open list system with semi-proportional (limited) rule. The new rules and the multi-member districts were even more restrictive than those they replaced; they were more likely to create artificial majorities with a minority of voting support (Colomer, 2004). They were also likely to produce overrepresentation to the candidates with most votes.

The incentives for strategic voting were also present. Voters could cast votes for between 67% and 80% of the candidates in each district, depending on the number of seats up for grabs in the district. However, the seats in each district were divided between the two most voted lists: the ‘majority’ and the ‘minority’. A qualified plurality of at least 20% (later changed to 40%) was necessary to earn the number of seats for which the voter could vote (between 67% and 80%) in that district. This was the majority. The minority received the remaining number of seats (between 33% and 20%).

Colomer (2004) examines the Spanish rules in great detail and offers an insightful hypothetical example of a typical district with eight seats in which each voter could vote for up to six candidates. If three candidates obtained 41%, 30%, and 29% of the votes, respectively, they were allocated six (the majority), two (the minority), and zero seats, respectively. This translates to a seat-share ratio of 75% (six out of eight) awarded to the candidacy with only 41% of the popular vote, while the candidacy earning nearly 30% of the votes would go unrepresented. Moreover, under these rules, it was theoretically possible for a candidate to win as much as 42% of the overall votes and garner zero seats – the threshold of maximum exclusion. No candidate was guaranteed Cortes representation in this eight-seat district unless they won roughly 43% of the votes. Furthermore, the rules encouraged large parties to attempt to win both the majority and minority candidates in an eight-seat district by running two separate lists and winning at least 57% of the vote (100 – 43) between them.

And so it was that in 1931 the winning party was overrepresented. Its victory encouraged the defeated side to mount more effective opposition from the right. The official monarchist party was formed later the same year. By 1932, anti-republican schemes were prospering. The Carlist movement was revived in the wake of the republican victory. It found common cause with the army and also allied itself with the Catholics. General Sanjurjo elected to attempt an ill-fated pronunciamiento against the regime in the summer of 1932. Finer (1975: 75) would characterize the endeavor as one with ‘disposition, but no opportunity’, adding that ‘Sanjurjo himself knew that his rising was bound to fail, but felt honor bound to try’. But, aware of what was afoot, the government put down the insurrection with ease.

At the time of Sanjurjo’s plot, the possibility of legitimizing a military takeover in Spain was weak. Two primary factors worked against Sanjurjo. Not only did he lack adequate support, but the regime was in no way an illegitimate representation of the collective Spanish will. Sanjurjo was without even modest backing from the larger Spanish aristocracy and large landowners. Only two out of 262 Spanish nobles supported Sanjurjo. Notwithstanding the outcome, there was some clarity in failure. No longer would the age-old military pronunciamiento suffice to supplant an unpalatable regime: active civilian support would be essential.

By 1933 the Azana government was beset with crisis. In the face of continued revolts and unrest, it was forced to form a new cabinet and call for new elections in the fall of 1933. Sensing opportunity, the right and center-right parties united into a single coalition called the ‘Unión de Derecha y Agrarios’. The move proved successful. The right and center-right parties won 374 of a possible 467 seats in the November Cortes elections. The left fared badly, earning only 93 seats.

The distribution of votes in 1933 was more balanced and there was semblance of a center (see Table VI). The newly formed government, with the support of the right, began reversing many of the policies and reforms of the previous administration. Their side comfortably in power, the generals were now without even the motive to strike. However, developments would ensure that the military remained quite active in the coming months.

In the wake of electoral defeat, the extreme left agitated for socialist reform. In Asturias the tension would eventually culminate in what is often referred to as the Asturias revolution of 1934. It resulted in the government declaring martial law and cost over 1,000 lives. The military suppressed the rebellion without much difficulty. While the episode was a disaster for the militants, it was a clear and frightening indication to the nation precisely how precarious was its state of democracy. The event also reinforced the relationship between the generals and the coalitions on the right. After the Asturias revolution, the army became the clear and understood safeguard of the right and ‘old Spain’.

By 1935, the political and social fabric of Spain had almost completely unraveled. A premature dissolution of the Cortes in December 1935 made inevitable new legislative elections set for February 1936. The right consisted principally of an alliance between the monarchists, Carlists, and the CEDA (Spanish Confederation of the Autonomous Right). The Catholic church was also decidedly right-leaning and helped to finance the right’s campaign. The parties of the left and center-left – including the Socialist Party (PSOE), the Spanish Communist Party (PCE), and the Workers Party of Marxists Unification (POUM) – united into a single group, presenting a Popular Front program. In an election anticipated to be close, the left badly needed the anarchist vote. The anarchists, hoping to free many of their comrades from prison, overcame their anti-election penchant, turning out in large numbers (Beevor, 2006).

The results barely favored the left: the Popular Front won the election by a margin of less than 2%. However, the electoral law encouraging coalitions permitted a kind of first-past-the-post result for the Popular Front in the Cortes. The Popular Front attained an absolute majority in the Cortes. With only 47% of the popular vote, the left earned 60% of the seats. Moreover, the right and center-right clearly won more than 50% of the popular vote. The parties of the left, ignoring the narrowness of their victory, proceeded to behave as if they had received an overwhelming mandate for revolutionary change (Beevor, 2006). Within six months, the generals revolted; Spain was at war.

Reflecting on the Spanish case of 1931–36, the observations of Linz (1967: 237) himself are telling: ‘Perhaps a different electoral system would have crystalized these social tensions in a formal split of the PSOE or a strengthening of the Communist Party, and who knows if that rather than the polarization and radicalization of the major democratic parties would not have served the country better.’ Rather than ameliorating social tensions in Spain, the characteristics of the electoral system only served to reinforce and deepen social polarization (Balcells, 2017). Moreover, the coalitions that formed during the period only reinforced social positions. There was no left–right or cross-party coalition formation that could have served to ameliorate or mitigate the military’s fears. Instead, the military gleaned from the electoral outcomes and the incumbent regime’s behavior that its chance to strike would not improve in the near future.

In light of the theory and statistical models presented above, what does the Spanish episode suggest? Several issues merit a mention. As always, reality is more complex than theory and statistical averages. The hybrid electoral rules system at play within the years discussed here does not fit neatly within the coding schema utilized above. While it clearly promoted strategic voting, it reflects characterizations of both plurality and non-plurality systems. Moreover, it promoted the formation of polarizing coalitions, which our theory does little to account for. But the three Cortes elections clearly indicate that the Spanish generals remained sensitive to the ruling regime’s popular support.

Conclusion

Coups and elections are by no means independent events, as the former often follow the latter. This article has been an attempt to impose a theoretical construct sufficient to explain this phenomenon and to better understand what constitutes a coup opportunity. We have leveraged the rich literature on electoral institutions to argue that the rules governing elections should have a bearing on the military’s judgment to intervene in politics. The empirics we have examined are consistent with the idea that electoral institutions matter for explaining post-election coups.

As our historical discussion of the Spanish case shows, our approach has important limitations. The electoral system at play in Spain between 1931 and 1936 certainly fostered strategic voting, but it also encouraged parties to game the system as well. Moreover, its ‘hybrid’ system did not easily fit into the discrete categories employed in our statistical analysis – and the same is likely to be the case in multiple other systems. Furthermore, neither our theory nor the empirics properly contextualize the types of pre-election and post-election coalitions that can emerge from different constellations of societal alignments. These are obvious areas for future research.

Footnotes

Replication data

The datasets and replication code for the empirical analysis in this article, along with the Online appendix, can be found at http://www.prio.org/jpr/datasets, as well as the authors’ website at .

Acknowledgements

For comments and suggestions, we thank Laia Balcells, Emerson Niou, and D. John Robinson. Ashley Weiler provided valuable research assistance.

Notes

References

Balcells

Laia

(2017) Rivalry and Revenge: The Politics of Violence during Civil War. Cambridge: Cambridge University Press.

Beevor

Anthony

(2006) The Battle for Spain: The Spanish Civil War 1936–1939. New York: Penguin.

Borkenau

Franz

(1963) The Spanish Cockpit. Ann Arbor, MI: Faber & Faber.

Bormann

Nils-Christian

Golder

Matt

(2013) Democratic electoral systems around the world, 1946–2011. Electoral Studies 32(2): 360–369.

Casper

Brett A

Tyson

Scott A

(2014) Popular protest and elite coordination in a coup d’état. Journal of Politics 76(2): 548–564.

Cheibub

Jose A

Gandhi

Jennifer

Vreeland

James R

(2010) Democracy and dictatorship revisited. Public Choice 143(1): 67–101.

Collier

Paul

(2009) Wars, Guns, and Votes: Democracy in Dangerous Places. New York: Harper.

Colomer

Josep M

(2004) Handbook of Electoral System Choice. London: Palgrave Macmillan.

Cox

Gary W

(1997) Making Votes Count. Cambridge: Cambridge University Press.

10.

Cox

Gary W

Shugart

Matthew Soberg

(1996) Strategic voting under proportional representation. Journal of Law, Economics, and Organization 12(2): 299–324.

11.

Duverger

Maurice

(1954) Political Parties. New York: Wiley.

12.

Farrell

David M

(2011) Electoral Systems: A Comparative Introduction. London: Palgrave Macmillan.

13.

Fearon

James D

(2011) Self-enforcing democracy. Quarterly Journal of Economics 126(4): 1661–1708.

14.

Finer

Samuel Edward

(1975) The Man on Horseback: The Role of the Military in Politics. Middlesex: Penguin.

15.

Gleditsch

Kristian S

(2002) Expanded trade and GDP data. Journal of Conflict Resolution 46(5): 712–724.

16.

Gschwend

Thomas

(2007) Ticket-splitting and strategic voting under mixed electoral rules: Evidence from Germany. European Journal of Political Research 46(1): 1–23.

17.

Hyde

Susan

Marinov

Nikolay

(2012) Which elections can be lost? Political Analysis 20(2): 191–210.

18.

Kim

Nam Kyu

(2016) Revisiting economic shocks and coups. Journal of Conflict Resolution 60(1): 3–31.

19.

Kissinger

Henry

(1970) White House Years. New York: Simon & Schuster.

20.

Knutsen

Carl Henrik

Nygård

Håvard Mokleiv

Wig

Tore

(2017) Autocratic elections: Stabilizing tool or force for change? World Politics 69(1): 98–143.

21.

Laver

Michael

Shepsle

Kenneth A

(1996) Making and Breaking Government. Cambridge: Cambridge University Press.

22.

Lehoucq

Fabrice

Pérez-Liñán

Aníbal

(2014) Breaking out of the coup trap: Political competition and military coups in Latin America. Comparative Political Studies 47(8): 1105–1129.

23.

Levy

Jack S

(2008) Case studies: Types, designs, and logics of inference. Conflict Management and Peace Science 25(1): 1–18.

24.

Lindberg

Staffan

(2006) Democracy and Elections in Africa. Baltimore, MD: Johns Hopkins University Press.

25.

Linz

Juan J

(1967) The party system of Spain: Past and future. In: Lipset

Seymour Martin

Rokkan

Stein

(eds) Party Systems and Voter Allignments: Cross-National Perspectives. New York: Free Press, 197–282.

26.

Linz

Juan

(1978) The Breakdown of Democratic Regimes: Crisis, Breakdown and Reequilibration. Baltimore, MD: Johns Hopkins University Press.

27.

Little

Andrew T

(2012) Elections, fraud, and election monitoring in the shadow of revolution. Quarterly Journal of Political Science 7(3): 249–283.

28.

Londregan

John B

Poole

Keith T

(1990) Poverty, the coup trap, and the seizure of executive power. World Politics 42(2): 151–183.

29.

Luo

Zhaotian

Rozenas

Arturas

(2018) Strategies of election rigging: Trade-offs, determinants, and consequences. Quarterly Journal of Political Science 13(1): 1–28.

30.

Marshall

Monty G

Marshall

Donna Ramsey

(2010) Coup d’état events, 1946–2009. Center for Systemic Peace (http://www.systemicpeace.org/inscrdata.html).

31.

Marshall

Monty G

Gurr

Ted Robert

Jaggers

Keith

(2013) Polity IV Project: Political Regime Characteristics and Transitions, 1800–2013. College Park, MD: University of Maryland.

32.

Miller

Michael K

(2015) Elections, information, and policy responsiveness in autocratic regimes. Comparative Political Studies 48(6): 691–727.

33.

Niou

Emerson MS

(2001) Strategic voting under plurality and runoff rules. Journal of Theoretical Politics 13(2): 209–227.

34.

Nohlen

Dieter

(2005) Elections in the Americas: A Data Handbook. Volume 2: South America. Oxford: Oxford University Press.

35.

Nohlen

Dieter

(2010) Elections in Europe: A Data Handbook. Baden-Baden: Nomos.

36.

Nohlen

Dieter

Grotz

Florian

Hartmann

Christof

(2001) Elections in Asia and the Pacific: A Data Handbook. Volume I: Middle East, Central Asia, and South Asia. Oxford: Oxford University Press.

37.

Nohlen

Dieter

Krennerich

Michael

Thibaut

Bernard

(1999) Elections in Africa: A Data Handbook. Oxford: Oxford University Press.

38.

Nordlinger

Eric A

(1977) Soldiers in Politics: Military Coups and Governments. Englewood Cliffs, NJ: Prentice Hall.

39.

Powell

Jonathan

(2012) Determinants of the attempting and outcome of coups d’état. Journal of Conflict Resolution 56(6): 1017–1040.

40.

Rozenas

Arturas

(2016) Office insecurity and electoral manipulation. Journal of Politics 78(1): 232–248.

41.

Shugart

Matthew Soberg

Wattenberg

Martin P

(2001) Mixed-member electoral systems: A definition and typology. In: Shugart

Matthew Soberg

Wattenberg

Martin P

(eds) Mixed-Member Electoral Systems: The Best of Both Worlds? Oxford: Oxford University Press, 9–24.

42.

Svolik

Milan

(2009) Power sharing and leadership dynamics in authoritarian regimes. American Journal of Political Science 53(2): 477–494.

43.

Teorell

Jan

Charron

Nicholas

Samanni

Marcus

Holmberg

Soeren

Rothstein

(2016) The Quality of Government dataset, Version 17. University of Gothenburg: Quality of Government Institute (http://www.qog.pol.gu.se).

44.

Therneau

Terry M

Grambsch

Patricia M

(2013) Modeling Survival Data: Extending the Cox Model. New York: Springer Science & Business Media.

45.

Thomas

Hugh

(2001) The Spanish Civil War. New York: Modern Library.

46.

Valenzuela

Arturo

(1978) The Breakdown of Democratic Regimes: Chile Volume 4. Baltimore, MD: Johns Hopkins University Press.

47.

Wig

Tore

Rød

Espen Geelmuyden

(2014) Cues to coup plotters? Elections as coup triggers in dictatorships. Journal of Conflict Resolution 60(5): 787–812.

48.

Wig

Tore

Hegre

Håvard

Regan

Patrick M

(2015) Updated data on institutions and elections 1960–2012: Presenting the IAEP dataset version 2.0. Research & Politics 2(2): 2053168015579120.

49.

Wiking

Stephan

(1983) Military Coups in Sub-Saharan Africa: How to Justify Illegal Assumptions of Power. Uppsala: Scandinavian Institute of African Studies.

50.

Wood

Simon

(2006) Generalized Additive Models: An Introduction with R. Boca Raton, FL: CRC.