Elections and civil violence in new multiparty regimes

Abstract

The introduction of multiparty competition around the world following the Cold War raises the specter of rising civil violence during election periods in emerging democracies and hybrid regimes. Yet there are also plausible theoretical reasons to expect dropping civil violence around elections in these states or, alternatively, no significant change in the level of such conflict. This article tests these hypotheses in Africa with the aid of event data on the daily rate of civil violence incidents (1997–2013). It asks if civil violence in that region is more frequent, less frequent or constant during election cycles compared to other times. To guard against definitional and data selection problems encountered in some prior cross-national studies of elections and use of force, the research design emphasizes the relative risk of social conflict at the national level. The analysis suggests three distinct patterns in Africa. Given countrywide norms, election periods in most countries run the same relative risk of a day with violent events as do non-election periods. A subset of African countries exists, however, with more civil violence during elections when judged against the national reference line for conflict. There is a smaller group of countries where the risk of electoral civil violence is comparatively low. While caution should be exercised in interpreting the findings, the policy implication is that no general reason exists to preclude or defer elections in Africa as a way to minimize social conflict associated with political campaigns, although there may be specific reasons to do so in particular countries.

Keywords

civil violence elections emerging democracies hybrid regimes

The end of the Cold War set off a global ‘electoral tsunami’ (Morse, 2012), with most military dictatorships and one-party states swept away by popular demands for contested elections. According to Freedom House, there were just 69 ‘electoral democracies’ worldwide in 1989–90 compared to 120 ten years later. Electoral democracies are defined as states with multiparty political systems and universal adult suffrage, which hold regular credible elections with generally open political campaigning.¹ During this same ten-year interlude, a majority of the world’s other political systems became hybrid regimes (Diamond, 2002: 26), where the opposition forces may legally contest for office but the ruling party retains major de facto advantages.

This political sea-change was propelled by the prospect that officially sanctioned rivalry for votes would provide countries with orderly means to recruit leadership and settle internal differences. Skeptical observers at the time cautioned against the idea that simply holding elections would guide political action into peaceful competition among elites (Schmitter & Karl, 1991: 78). They may have misrepresented the argument for elections to make it easier to critique and, predictably, researchers have found a range of ways that elections actually work in emerging democracies and hybrid regimes. A question mark in this field of inquiry, and the focal point here, is how likely these political contests are to whip up social conflict in the short run rather than to defuse it.

The electoral tsunami is particularly notable in Africa. Botswana, Gambia, and Mauritius were the only African electoral democracies in 1989–90, according to Freedom House’s tally. As of 1999–2000, the number had surged to 17 electoral democracies. Most remaining African states had adopted multiparty constitutions and term limits – at least on paper. By 2014, Freedom House categorized 21 African countries as electoral democracies, and classified 12 of the non-electoral, undemocratic countries as ‘partially free’. All but Eritrea and Somalia of the remaining ‘unfree’ states had held at least one election over the previous 25 years in which multiple political parties had nominal opportunities to gain control of government offices.² Africa thus presents a rich sample in which to observe perverse versus advantageous consequences of holding elections in countries that are inexperienced with democratic rule.

This article concentrates on the unwanted outcome of civil violence during and after political campaigns. By civil violence I follow Huntington’s (1972: 1) characterization: it is the use of force in which either a perpetrator or victim is not a government actor and which differs from criminal violence in that it aims to affect the way government is composed or performs. This definition excludes interstate conflict, while encompassing (a) state-incited repression of dissidents, (b) protest and rebellion against the state or regime, and (c) ethnic or partisan fighting among rival groups in society. The elections under consideration are normal planned popular votes for the national executive, a national legislative body, or both when there is a combined legislative and presidential election. To what extent is the process of selecting representatives in Africa via elections associated with contemporaneous civil violence? Even after 25 years of multipartyism, the correlation between vote-casting and violent conduct in the region is not exactly known.

Taking advantage of recently updated datasets on African armed conflict, this article provides a fresh descriptive perspective on elections by highlighting the relative national risk of civil violence. Existing empirical work has been apt to focus on the absolute amounts of electoral strife. Because the meaning of outright variations in the use of force is difficult to interpret across countries, a within-country research strategy is adopted here. I compare the national rate of daily violent disturbance when an election campaign is underway and the rate when an election is over or yet to start. From that basis, all countries may be classified according to their electoral violence risk ratios.

The article proceeds as follows. The first section outlines the debate over the effects of multiparty elections on political order in former autocracies. Two hypotheses are considered: national election periods suffer atypical conflict or they enjoy atypical amity. Section two reviews the political event data that provide the article’s empirical foundation and the methodological challenge these data pose. The discussion goes over the utility of relative risk ratio analysis as a way to measure national proneness to civil violence, and shows that under several different scenarios most countries in Africa (with a few consistent exceptions) run no greater risk of violence in election cycles compared to other times. There follows a brief theoretical and practical interpretation of this observation. The evidence comes from only one continent and covers less than two decades of elections, so it must be treated cautiously. However, the pattern described has possible consequences for emerging democratic states and competitive authoritarian regimes in other regions.

Voting and coterminous civil violence

Elections are meant as pressure relief valves that facilitate the expression of societal interests and reconcile the differences through mutual adjustment and accommodation. Even an entrenched dictatorship may concede to electoral competition in a bid to forestall violent revolt. But it may be idealistic to expect the political rivalry to be peaceful where mediating government organizations are weak and officials are unaccustomed to compromise and power-sharing. In reality, would not multipartyism be more likely to drive a wedge between different constituencies and create political instability?

The answer, according to a line of analysis that can be traced to Huntington (1968), is that pluralistic competition would tend to divide and destabilize. Looking at the independent states that emerged following World War II, Huntington postulated a threat to public order due to the imbalance between state institutional capability and rising claims for public goods and services. Elections under an unresponsive, low-capacity regime would be apt to revive disputes and sweep away the legally established political structures and procedures. Mansfield & Snyder (2002) updated this argument after the Cold War, noting how the end of autocracy typically released pent-up political demands. These demands could overload the new parties, legislature, and electoral system, setting the stage for insecurity and conflict.

A contrasting perspective sees election-related violence as a diminishing threat. Focusing on sub-Saharan Africa, Lindberg (2006) argues that once introduced, elections have tended to lead states in that region incrementally toward greater respect of civil liberties (as measured by Freedom House), even though four of five African elections from 1990 to 2003 were tainted by some violence. Importantly, most of the conflict did not rise to major, debilitating levels. Based on a sample of post-Cold War elections in Asia, Latin America, Eastern Europe, and Africa, Howard & Roessler (2006) likewise show that elections accompanied by violence may still result in more democratic political systems as measured with a combination of the Freedom House and Polity indices.³

Another group of scholars takes an intermediate position about transitions to democracy in the 21st century, and sees many countries settling into a condition of competitive authoritarianism (Schedler, 2006; Levitsky & Way, 2010). While formal opposition and some open political debate occurs, the partisan playing field is tilted steeply in favor of the governing elites. Regardless of flawed elections, however, competitive autocracies may be quite stable. Because of the attendant cost and risk, high-intensity coercion is the exception not the rule in these states (Levitsky & Way, 2010: 58ff).

A pivotal point in this academic debate is exactly how pervasive electoral civil violence is in mixed and transitional regimes. Each perspective implies a different prediction. Following institutionalist logic, the hypothesis (H₁) is that most democratizing and semi-authoritarian states will experience an uptick in civil violence during political campaigns. This is mainly due to a problem of credible commitment – specifically, a lack of confidence in the longevity and effectiveness of representative institutions. If political actors feel they cannot count on established political structures and procedures to protect them in the future, they would understandably see physical assaults on property and individuals as an acceptable form of political participation to use now (Dunning, 2011: 7–8).

Civil violence could happen inadvertently, when efforts by political godfathers to mobilize followers spin out of control, or civil violence could be a conscious tactic of leaders expecting or reacting to an unfavorable election. Whoever holds elected office often has the upper hand in these disputes. Incidents might be started pre-election, as contending factions attempt to intimidate one another; on election day, as they endeavor to keep rivals away from the polling stations; and post-election as the losing side tries to overturn the results. The logic of retaliation could lead to a self-perpetuating cycle of coercion during election campaigns.

The democratic learning-process view of elections suggests the reverse (H₂): even where representative institutions are just emerging, elections should prove comparatively violence-free. Peaceful behavior does not depend on persuasion and compromise being prevailing political values, although if democratic norms are widely diffused in a society, so much the better. Tested and legitimate political institutions are helpful but not indispensable. Lack of civil violence theoretically owes mainly to partisan self-interest – provided all sides understand the cost of starting an arms race for political office.

During an election campaign, the rivals have incentives to act with restraint to avoid alienating swing voters and to maintain a dialogue with potential coalition partners (Saideman & Lanoue, 2005; Harish & Little, 2013). In the immediate aftermath of an election both winners and losers will want to keep their options open. There is also a likelihood of oversight by international organizations, which is a further inducement for everyone to behave well while under observation. These incentives appear to be particularly strong after a civil war, as long as the peace agreement allows former militant groups to run for office (Matanock, 2013). To follow this line of thought to its conclusion, a positive loop of violence avoidance would show up even in partially competitive elections.

The null hypothesis (H₀) is that such elections are times of neither abnormally high nor low internal strife. A lack of relationship is consistent with the way stable competitive autocracies are generally seen to run. If elites hold power via persistent, often low-level harassment, there is little need for further attacks on people and their possessions when a vote is looming. As Hafner-Burton, Hyde & Jablonski (2014) point out, piling on with additional force is a gamble. The governing party could face a popular backlash, which raises the likelihood it will be forced to step aside or hold new elections. The opposition may similarly want to avoid escalating confrontation so as to preserve its resources for a time when the regime has relaxed its attention or is more vulnerable for other reasons. If most political players see their interests served by bearing with the status quo during and after political campaigns, civil violence should keep at routine levels.

To be clear, H_1, H₂, and H₀ are not about whether competitive or partially competitive elections raise or lower the overall quantity of internal civil violence in post-Cold War ‘third wave’ democracies. The hypotheses only ask about the level of civil violence that occurs during elections, not if these contests disturb the national reference line for conflict on the whole. A number of studies have considered the wider effect of elections on civil violence, showing for example that hasty elections after an internal war may lead to a recurrence of conflict (Brancati & Snyder, 2012; Flores & Nooruddin, 2012). Cederman, Gleditsch & Hug (2013) find that competitive elections are a particular risk factor in ethnic wars. Based on African data, Linebarger & Salehyan (2015) estimate a large and statistically significant relationship between months in which an election is held and the frequency of social conflict events. On the other hand, Carey’s (2007) empirical work suggests elections decrease the probability of insurgencies. Her axiom: ‘any election is better than no election’ (Carey, 2007: 59).

Carey’s statement is a reminder not to rush to the conclusion that a violent transitional or hybrid regime election’s net impact is more intranational violence in total. Harish & Little (2013) develop a formal model that takes into account the likelihood that voting has both an intertemporal effect and a substitution effect on social conflict. The first displaces civil violence to election times while the second tempers civil violence by offering a peaceful means to contest for power. The final result after adding and subtracting these effects may be less overall violence with elections than without. Whether or not that is true in Africa, this article takes the background level of social conflict as given and only examines whether the occurrence of an election disturbs the standard pattern.

Assessing electoral civil violence

To test H₁ and H₂ in Africa’s third wave regimes, the first fact that has to be established is how violent their elections are. This is a larger methodological challenge than it might seem. One approach is to use binomial or ordinal values to characterize the degree of electoral peacefulness, based on preselected criteria. For example, Straus & Taylor (2012) divide African elections between those in which violence plays a minimal role and those in which violence is a central feature. Hyde & Marinov (2012) ask simply whether or not ‘significant violence’ relating to an election resulted in civilian deaths. The validity of the cutoff points on these scales is not clear.

It is no surprise that resulting estimates about the extent of recent electoral violence vary, sometimes widely. Straus & Taylor (2012) calculate the share of violent elections at 40% of the contests held in sub-Saharan Africa from 1990 through 2008. Hyde & Marinov (2012) put the figure at 28% for the same sample.⁴ Such uncertainty inhibits informed debate about the effects of competitive authoritarian and transitional democratic elections.

One way to sharpen the classification is to turn to event tallies – the number of reported violent incidents occurring in a country at a particular time (Linebarge & Salehyan, 2015). Do election periods have a high or low count, with non-election periods as the reference category? Or are the rates equivalent throughout both periods? While a seemingly more objective basis for comparison, event data are subject to selection bias. Moreover, elections are endogenous to conflict (Cheibub, Hays & Savun, 2012). These events do not happen by chance in a typical competitive autocracy, and parallel influences may affect the phasing of both elections and civil violence incidents. The relative approach taken here circumvents the counterfactual question of how much internal conflict would likely have occurred in the absence of elections, and looks at a simpler empirical issue at the national level. Does the proportion of country days with violent events go up, go down, or stay constant in election versus non-election periods?

The Armed Conflict Location and Event Dataset (ACLED) (Raleigh et al., 2010) is the principal source used in this article to operationalize Huntington’s three-way conception of civil violence. To reiterate, civil violence refers to acts that destroy or injure persons or property for political ends, in which popular uprising, ruling party repression, and intergroup fighting are the constituent elements. The state is the target or the instigator of the first two variants of civil violence, but not necessarily of the third type. I focus on incidents whose timing is associated with national elections or, conversely, appear to be independent of any particular electoral contest.

Not every act of civil violence during an election cycle may meet the strict criteria of electoral violence, in the sense of being consciously motivated by a desire to shape an election’s outcome. It is my premise, however, that abnormal clusters of events around an approaching or just-finished election are likely related to that election. Any single violent incident that takes place in proximity to voting could be a coincidence, but a spike in the number of incidents has high odds of reflecting efforts to affect the election results. Conversely, an unusual absence of events near an election is a probable indicator some political actors have openly or tacitly agreed to suspend acts of civil violence while the race is on.

ACLED collects comprehensive information regarding individual episodes of civil conflict in Africa. Version 4 of the dataset runs from the start of 1997 through the end of 2013. ACLED’s event data are gathered from many sources, particularly official reports and research publications. Specific types of events identified include riots, civilian killings, and battles, which are described by date and other distinguishing features. Multiday incidents are recorded on a daily basis. Thus, for example, strife that went on for three days would be reported as three separate incidents in the dataset.

To be more precise with regard to the coding of these events, ACLED defines a riot as a spontaneous act of organization that turns violent against people or property (Raleigh & Dowd, 2015: 6). Violence against civilians covers ‘deliberate violent acts perpetrated by an organized political group such as a rebel, militia or government force against unarmed non-combatants’ (Raleigh & Dowd, 2015: 13). A battle is defined as the use of armed force between two politically organized groups, with the intent of inflicting harm on the opposing side. Battles can include a range of participants such as rebels, militias, government forces, and communal groups (Raleigh & Dowd, 2015: 10). Africa suffered nearly 60,000 civil violence incidents from 1997 to 2013 – about 17,000 riots, 26,000 attacks on civilians, and 30,000 battles.⁵ These events occurred on close to 40,000 different country-days.

As a robustness check, I also employ the Uppsala Conflict Data Program’s Geo-referenced Event Dataset (UCDP GED) (Sundberg & Melander, 2013). While it covers the same geography, UCDP GED’s unit of analysis is narrower than ACLED’s. It only reports events in which lethal violence is employed. These fatal incidents (one death or more) entail the use of armed force by a government or by an organized group (formal or informal) against another organization or against civilians. The majority of events lasted one day or less, and for multiday events I only count the date on which the event is reported to have started. The most recent version of UCDP GED (1.5) available at the time of writing runs through the last day of 2010 and thus corresponds to 14 of the 17 years covered by ACLED. During the period in question, the dataset lists the inception of over 9,000 state-based conflicts, 5,500 non-state conflicts, and 2,100 one-sided violent incidents in Africa.⁶ There were approximately 10,000 country-days in Africa (1997–2010) with the onset of one or more of these fatal episodes.

Commentators have advised caution about these time series (Eck, 2012; Schrodt, 2012). Any cross-national comparison by means of event data may be skewed due to the media sources used to identify the violent incidents for each country. Urban bias is hard to avoid, and violent incidents in countries with large, remote land areas are probably inadequately represented in ACLED and UCDP GED. Similarly, countries that happen to have a greater media presence would likely have a deceptively large event count. Steps should be taken to control for such misreporting.

Yet, even allowing for population density, mountainous or desert territory, intensity of news coverage, and other relevant variables, it may be misleading to directly compare the number of incidents taking place in different countries. What thresholds should be used to conclude that the vote has been corrupted by violence? Taylor, Pevehouse & Straus (2013) code an election ‘violent’ if 20 or more deaths are associated with it. Bekoe (2010) makes the cutoff one death. These criteria seem arbitrary. Regardless of the number of casualties, the relationship between violent events and their national political impact is not constant. The political implications of 20 partisan killings must be vastly different in Gambia (estimated 2013 population 1.8 million) than they would in Nigeria (177 million). Is the difference simply proportional to population? Does it matter if lives are taken primarily in one district or region? Should the relevant number be adjusted some other way?

The research strategy I follow here to get around the problem of intercountry comparison of event data is to work forward from the intracountry time series. Looking at individual countries reduces the need to normalize the significant rate of conflict events. Longitudinal reports on civil violence at the national level are also less subject to selection effects than are variations in the rate of conflict incidents among nations. The unit of analysis I choose is a country-day with (or without) at least one violent event. I make no attempt to account for the magnitude of the underlying events, for example the numbers of participants or casualties. There is spotty coverage of such information in the databases, and the reported figures may be unreliable. Still, it seems likely the within-country variation in event magnitude is lower than the cross-country variation.

That said, the longitudinal data are not immune to selection error. Given that voting periods often come under extra scrutiny by journalists, aid workers, and diplomats, there could be an inflated number of reported conflicts while politicians run for office. Reporters and editors have an interest in stories with novelty and drama – a bias that might magnify reported incidents involving repression of sympathetic civilian underdog groups. Alternatively, a saturation effect is possible whereby an overabundance of internal conflicts makes them less newsworthy and leads to underreporting. Or the national authorities might be successful in limiting unfavorable reporting during a tense political season. The net effect of these intracountry phenomena is difficult to say, but their likely scale is smaller than intercountry distortions. The ACLED and UCDP GED intracountry time series thus seem generally more consistent and less debatable than the intercountry series.

To examine and interpret these panel data I estimate relative risk (RR) – the ratio of the probability of an event occurring (a country-day with civil violence in this article) in an ‘exposed’ group (country-days during election cycles) to the probability of the event occurring in a ‘non-exposed’ group (country-days outside of election cycles). Relative risk is useful in statistical analysis of binary outcomes where the outcome of interest is rare, as is the case with violent incidents in some countries. It also has the advantage of facilitating country-by-country comparison of the prevalence of these adverse events, which regression analysis may misrepresent if there are heterogeneous national-level correlations between elections and civil conflict.

Identifying election cycles

The next step to test H₁ and H₂ is to identify election periods. For election data, I draw mainly on the National Elections across Democracy and Autocracy (NELDA) dataset (Hyde & Marinov, 2012). NELDA is derived from a comprehensive list of secondary sources. Version 3 covers national elections in independent nations with populations over half a million starting in 1945.⁷ Not included are local elections or indirect elections and special by-elections. There were 273 relevant contests in Africa over the 1997–2013 period.⁸ Almost half were legislative elections; the rest were presidential or general elections for both the chief executive and the legislature. Of the elections, 50 were second- or third-round runoffs where those were required by law.

Most of these elections were ‘competitive’, using NELDA’s minimalist definition of electoral competition. Opposition parties are legal, they are allowed to compete, and they place candidates on the ballot for the office in question. How many of these contests were ‘high quality’ is another question. Irregularities in the registration of voters are common in Africa. Results from each polling station may not be reported accurately or added up correctly. Overall, the integrity of African elections remains suspect, according to the United Nations Economic Commission for Africa (2013: 21) and many other sources.

Some countries stick to a fairly regular schedule of elections. Kenya, for example, held successive general elections five years apart on 27 December 1997, 27 December 2002, and 29 December 2007, although the next one did not take place for five years and two months, on 4 March 2013. But predictability is the exception. Elections are frequently postponed due to disagreements between the government and opposition. Togo was supposed to hold parliamentary elections in October 2012, but rescheduled them for March 2013, then to 21 July 2013. The contest was finally held on 25 July. When an election is put off, there could be an anticipatory campaign that does not end with the casting of any ballots – implying electoral civil violence is possible even in the absence of an election.

Election cycles are discrete phases when special rules may apply to parties, candidates, and the media. There is no standard length to these cycles in Africa. They often officially begin (as announced by the national electoral management body) one or two months ahead of the day legally established for the voters to pick representatives, but those official time frames may not mean much in practice. Civil violence may start to heat up far ahead of any formal campaign proclamation. In Nigeria, for example, ethnic militias and private security units are said to create a nationwide climate of insecurity many months prior to election day (Anifowose & Odukoya, 2013: 294).

Table I.

Conditional logistic regression results, controlled for country effects

Predictor => election day in Africa	Response => violent day in Africa	Coefficient	Robust standard error	p-value
121-day e-cycle	}ACLED events (1997–2013)}	–0.025	0.021	0.221
61-day e-cycle		0.080	0.026	0.002
181-day e-cycle		–0.112	0.018	0.000
121-day e-cycle	}UCDP GED events (1997–2010)}	–0.152	0.039	0.000
61-day e-cycle		–0.130	0.052	0.013
181-day e-cycle		–0.256	0.034	0.000

121-day e-cycles are the 90 days before an election, the day of voting, and the following 30 days (overlapping days in the case of runoff elections are counted only once). 61-day (45 days before + 15 days after an election) and 181-day cycles (120 days before + 60 days after) are calculated in a similar manner. A violent day has one or more events as reported by ACLED or UCDP GED.

Sources: Author’s estimates based on Hyde & Marinov (2012), updated by the author to 2013; and ACLED V.3, see Raleigh et al. (2010).

Human Rights Watch (2011) reports that Zimbabwe security forces and their informal allies use the threat of political killings, torture, and abductions to foster a permanent state of fear among voters.

An election technically ends when the winners are announced or certified, but the counting and certification of ballots can be delayed in Africa, sometimes for weeks. Any uncertainty or perceived lack of transparency during the tabulation creates additional opportunities for civil disorder, in particular if the loser rejects the results. Take the case of Côte d’Ivoire’s long overdue presidential election in 2010. The runoff round took place on 28 November. The Independent Electoral Commission (CEI) declared a winner four days later, on 2 December. However, the Constitutional Council pre-emptively voided the announcement on grounds the CEI had failed to meet its mandated three-day deadline for announcing election results. On 3 December the Constitutional Council declared a different winner. Not until 6 May 2011 was the impasse resolved with the original winner finally sworn in as president – and that happened only after United Nations military intervention and capture of his opponent, who was soon extradited to the International Criminal Court.

It is thus far from self-evident how long before or after the polls close to look for unusual levels of possibly election-related civil unrest. Given the national inconsistencies, I settle on three arbitrary but reasonable general definitions of an election cycle. All are asymmetrical on the assumption that, voter fear or apathy aside, attentiveness to political issues is apt to build gradually in the lead up to voting and fade fairly quickly once the ballots are cast. Civil violence outside the high-interest intervals has political overtones but is unlikely to aim at affecting a specific election. To capture a range of possibilities regarding the length of these pre- and post-election phases in different countries at different times, I define a cycle as lasting 61 days (45 days prior to and 15 days after voting, plus election day itself), 121 days (90 days before and 30 days after election day), or 181 days (135 days before and 45 days after election day).

In cases where a country holds closely spaced elections, the election period needs to be adjusted. Throughout this article, the count is restarted from the earliest election in the sequence. To illustrate with the mid-range 121-day period, if there are two rounds of voting that take place 14 days apart, the combined election cycle for the overlapping events would be 135 days (14 days plus 121 days).⁹

Relative risk of electoral civil violence

Conditional logistic regression is a basic model conventionally used to relate a response variable with two states (in this article a country-day with either civil violence or the absence of civil violence) to a binary explanatory predictor (an election cycle day or a non-election day). The method is designed to control for omitted variables with cross-sectional models, here by using only within-country differences and disregarding information about differences between countries.

Taking the mid-range (121-day) cycles and violent days as determined with ACLED, the regression equation indicates there is no significant relationship between election cycles and violent country-days across Africa. In other words, the model (the top row in Table I) appears to support H_0.

This finding in Table I could be spurious because the estimates are based on an arbitrarily sized election cycle. For comparison, I turn to a 61-day cycle. As with the 121-day cycle, where follow-up rounds of multistage national elections occur, those rounds are counted as part of one election sequence depending on the proximity of each stage. In contrast to the first model, this one (the second row in Table I) predicts a small but statistically significant positive correlation with violent country days, thus lending support to H₁. Strangely, however, a 181-day election cycle produces the opposite result. The correlation is significant and negative, which appears to confirm H₂.

These contradictory estimates could be due to errors in the dependent variable, violent days, taken from ACLED. The higher threshold UCDP GED time series (deaths must occur, not just the use of force) generally support H₂, and the idea that elections tend to be unusually quiet times. The coefficients are all negative and statistically significant, as reported in the bottom three rows of Table I. However, the time frame for the UCDP GED is shorter and the results could be sensitive to the period under investigation as well as to the type of violence being assessed in the different databases. It is difficult to have much conviction about H₂ under the circumstances.

An alternative explanation for Table I’s ambiguous continent-wide regression predictions is that substantial national variation might be the cause. When predictor variables vary greatly across individuals but do not change much over time for each individual, then fixed effects estimates will be imprecise. As I noted earlier, a simple and transparent method to focus on the country level is to calculate the risk ratio – the fractional increase or decrease in the incidence of civil violence during any specific nation’s election cycles, which is given by:

R R = \frac{violent election days/(violent + nonviolent election days)}{violent nonelection days/(violent + nonviolent nonelection days)}

Bear in mind that the metric is relative risk. A conflict-prone country could have routine day-to-day civil violence, yet experience somewhat fewer of these violent days in concurrence with its elections. Its election periods would thus be comparatively free of civic conflict even if the cumulative incidence of conflict is high. Conversely, a generally tranquil country could see relatively large jumps in civil violence around elections. Because the change would be off a low base, the absolute difference in violent days might be minor.

Table II shows the relevant information for each African country using 121-day election cycles. The risk ratios (reported in the fifth column) indicate the chance of a violent day in each reported African country over the duration of these cycles compared to the chance when no vote is approaching or recently concluded. A risk ratio significantly greater than 1 corresponds to an elevated risk, a ratio less than 1 indicates a reduced risk, while a ratio equal to or close to 1 signals no appreciable change in the probability of a violent day occurring during an election cycle.

The country-by-country results reported in Table II reveal a more complex pattern than is shown by the aggregate regression results for all of Africa in Table I. The sample countries are sorted into three groups, according to their individual RRs. For the 14 countries arranged at the top of Table II, the coefficient was significantly greater than 1 – that is, they are consistent with H₁. Not all the countries in this category are stereotyped for holding troubled elections or having high levels of social unrest. Botswana, for example, had only 39 violent country-days, but six of them (15%) fell during its 121-day election-periods. Ghana had 28 violent election-cycle days, but that was a disproportionately high number compared to other days in this generally peaceful country. Several other countries in the high relative risk category seem to belong here more intuitively – Kenya and Zimbabwe, for instance, which attracted worldwide attention for election turmoil in 2007 and 2008.

There are ten countries (put together at the bottom of the table) that appear to satisfy H₂. For this group, the probability of a day with civil violence was significantly less likely in a 121-day election period compared to other times. Again, there are some surprising countries in the category. Angola had more than 1,000 violent country-days, but just eight during an election cycle; Sierra Leone experienced over 600 violent country-days but only four fell during an election. Thus their relative exposure to electoral violence was very low. Cases such as these suggest how calm elections can be in post-conflict countries where there is a clear military resolution or an effective ceasefire agreement in place. The Angolan civil war lasted, with some interludes, from 1975 to 2002; the Sierra Leone civil war went from 1991 to 2002.

Table II.

Relative risk: 121-day election cycles and civil violence in Africa, 1997–2013

Relative risk group	Country	E-cycle days with violent incident(s)	E-cycle days incident free	Non e-cycle days with a violent incident(s)	Non e-cycle days incident free	RR of a violent e-cycle day	p-value
Significantly greater risk of civil violence during election cycles	Libya	54	67	570	5,518	4.77	0.00
	Togo	57	911	96	5,145	3.21	0.00
	Botswana	6	357	33	5,813	2.93	0.01
	Lesotho	13	471	54	5,671	2.85	0.00
	Zimbabwe	424	271	1,448	4,066	2.32	0.00
	Djibouti	12	835	33	5,329	2.30	0.01
	Mozambique	38	325	267	5,579	2.29	0.00
	Senegal	115	703	379	5,012	2.00	0.00
	Kenya	249	235	1,529	4,196	1.93	0.00
	Ghana	28	504	168	5,509	1.78	0.00
	Côte d’Ivoire	83	392	627	5,107	1.60	0.00
	Madagascar	64	727	284	5,134	1.54	0.00
	Zambia	61	423	503	5,222	1.43	0.01
	Mali	75	814	339	4,981	1.32	0.02
No significant difference between election cycles and non-election cycles	Equatorial Guinea	5	705	20	5,479	1.94	0.18
	Guinea-Bissau	25	802	108	5,274	1.51	0.06
	Swaziland	11	473	92	5,633	1.41	0.27
	Tanzania	30	340	351	5,488	1.35	0.10
	Gambia	10	748	55	5,396	1.31	0.44
	Cameroon	31	816	154	5,208	1.27	0.28
	Gabon	10	771	57	5,371	1.22	0.56
	Ethiopia	85	288	1,098	4,738	1.21	0.07
	Guinea	45	700	293	5,171	1.13	0.44
	Congo Rep.	26	626	202	5,355	1.10	0.66
	Mauritania	34	825	192	5,158	1.10	0.59
	Dem. Rep. of Congo	175	158	2,873	3,003	1.07	0.19
	Egypt	198	956	829	4,226	1.05	0.53
	Namibia	21	342	329	5,517	1.03	0.90
	Malawi	7	356	118	5,728	0.96	0.91
	Uganda	139	330	1,771	3,969	0.96	0.58
	Nigeria	245	391	2,355	3,218	0.91	0.07
	Algeria	201	646	1,425	3,937	0.89	0.09
	Chad	30	538	338	5,303	0.88	0.49
	Benin	8	797	62	5,342	0.87	0.70
	Liberia	34	385	592	5,198	0.79	0.17
	Niger	15	565	231	5,398	0.63	0.07
Significantly lower risk of civil violence during election cycles	Burundi	54	213	1,588	4,354	0.76	0.02
	South Africa	64	299	1,360	4,486	0.76	0.01
	Rwanda	21	498	356	5,334	0.65	0.04
	Sudan	40	202	1,728	4,239	0.57	0.00
	Morocco	12	351	361	5,485	0.54	0.03
	Burkina Faso	10	876	142	5,181	0.42	0.01
	Central Afr. Rep.	23	601	698	4,887	0.29	0.00
	Angola	8	234	1,017	4,950	0.19	0.00
	Sierra Leone	4	387	625	5,193	0.10	0.00
	Tunisia	1	362	548	5,298	0.03	0.00
	Eritrea	0	0	171	6,038	NA
	Somalia	0	0	3,131	2,898	NA

121-day e-cycles are the 90 days before an election, the day of voting, and the following 30 days (overlapping days in the case of runoff elections are counted only once). A violent day has one or more riots, battles, or incidents of violence against citizens.

Sources: Author’s estimates based on Hyde & Marinov (2012), updated by the author to 2013; and ACLED V.3, see Raleigh et al. (2010).

The electoral lull may not last long, however. Take the Central African Republic, which is also in the low RR group in Table II. A 2007 peace agreement between the government and rebel forces in that country led to a unity government and, after several postponements, national elections in 2011. The sitting president was re-elected to a second term. Two years later, however, a rebel coalition emerged and attacked the capital city, and the president fled. The insurgent leader took over the office, but he resigned within months and was succeeded by an interim caretaker government. Factional fighting was ongoing in the Central African Republic at the start of 2015.

Sudan is a similar example. In 2005 the government and the southern secessionist movement agreed to a truce, which ended their long-running war and set the stage for national elections and an independence referendum in 2011. The elections were relatively uneventful, as Table II suggests, but tension between Sudan and South Sudan (now technically an interstate conflict rather than civil violence) still runs high. Internal conflicts are also widely reported in both countries as of 2015.

Data from the largest number of African countries accord with the null hypothesis: the relative risk of a violent day was statistically indistinguishable regardless of a proximate election. Included are countries celebrated for democratic consolidation, such as Benin and Namibia, and others notorious for flawed elections and insurgent activity, such as Algeria and Nigeria. Whatever the general incidence rate of violence, a forthcoming or recently completed vote was not associated with a synchronic escalation in civil conflict in these nations.

How much confidence can we place in Table II? The sequence of elections could introduce a degree of systematic error if the first or ‘founding’ elections in countries are outliers. A formative early political contest could be uncharacteristically peaceful with all eyes fixed on the conduct of candidates and voters. The institutionalist model of political violence implies that the calmness of election cycles might degenerate over time in African countries, where elites often have irreconcilable interests and governing structures are generally weak. The opposite might also be true, with less civil violence taking place over time should the democratic learning process model apply. In either case, subsequent political campaigns might be more representative of electoral civil conflict patterns within a country.

Thus, to check whether Table II’s results are unduly affected by founding elections, I dropped seven elections (DR Congo 2006, Egypt 2005, Guinea 2010, Libya 2012, Mauritania 2006, Mozambique 2009, and Uganda 2006) that were the first multiparty contests to be held following a significant period of non-democratic rule. Using the base 121-day election cycle, Mozambique and the DR Congo switched places between the high risk and the neutral category, and Libya was dropped because it did not hold any follow up elections during the period, as presented in the left-hand columns in Table III. It thus appears that founding elections sometimes are uncharacteristic in terms of civil violence, but not always in the same direction and not so far off the norm as to change the modal relative risk pattern in Africa.

Table III also reports the country-by-country RR estimates in abbreviated form of three additional models – all elections with 61-day election cycles and 181-election cycles (using the ACLED data), and 121-day cycles with UCDP GED violent events. Each African nation is listed in the tripartite relative risk categories (RR > 1, RR ≈ 1, RR < 1) as determined under the four sets of assumptions. Three points stand out from visual inspection of the matrix.

First, using four plausible means of measuring relative election cycle violence, a plurality of countries in the region hold elections that are uncorrelated with statistically significant changes in the daily frequency of civil violence events. The RR coefficients change, but the number of countries in this middle category is always substantial in Table III. This observation is grounds to reject both H₁ and H₂ as applying across the board: African countries that hold elections do not as a rule have exceptional amounts (compared to the national baseline) of domestic aggression and counter-aggression during and after the campaigns.

Second, a few countries tend to be listed in the high-risk cell regardless of the length of the election cycle or the source of the violent event data. Djibouti, Kenya, Senegal, Togo, Zambia, and Zimbabwe appear in this category in all four columns. The repeated appearance of these countries is fairly compelling evidence for the existence of a subset of countries satisfying H₁: elections are linked to statistically significant increases in the daily incidence of civil violence in recent years.

Third, countries with significantly lower relative risk of civil violence during election cycles are somewhat atypical in Africa. Only three countries appear in the low-risk cell in all four models, raising questions about how often H₂ applies. Voting seldom seems associated with a temporary explicit or tacit ceasefire among political rivals and a drop in violent events. The exceptions seem to be where civil wars have recently ended.

Table III.

Relative risk of days of civil violence during national election cycles in Africa, alternative definitions

RR Group	Country code	RR 121-day e-cycle/ACLED events w/o founding election (1997–2013)	Country code	RR 61-day e-cycle/ACLED events (1997–2013)	Country code	RR 181-day e-cycle/ACLED events (1997–2013)	Country code	RR 121-day e-cycle/ UCDP GED events (1997–2010)
High: RR >1	GUI	4.70**	LIB	4.73**	LIB	4.62**	TOG√	54.68**
	TOG√	3.21**	MZM	3.72**	LES	3.24**	ZIM√	26.58**
	BOT	2.93*	TOG√	3.35**	TOG√	2.72**	ZAM√	20.64*
	LES	2.85**	EQG	3.04*	GAM	2.10*	DJI√	4.91**
	ZIM√	2.32**	GHA	2.75**	ZIM√	1.99**	SAF	3.25**
	DJI√	2.30*	DJI√	2.67*	DJI√	1.96*	KEN√	2.17**
	SEN√	2.00**	GAB	2.50**	KEN√	1.90**	SEN√	2.17**
	KEN√	1.93**	LES	2.40*	SEN√	1.78**	ALG	1.19**
	GHA	1.78**	ZIM√	2.34**	GHA	1.67**	BOT	14.07
	CDI	1.60**	KEN√	2.17**	MZM	1.63**	LES	14.07
	MAG	1.54**	SEN√	2.06**	MLI	1.48**	MAW√	14.07
	MAA	1.46*	SWZ	2.06*	ZAM√	1.41**	SWZ	14.07
	ZAM√	1.43*	ZAM√	2.02**	CDI	1.35**	EQG	8.05
	SWZ	1.41**	MAG	1.89**	ETH	1.21*	GAM	7.57
	ETH	1.36*	CDI	1.85**	DRC	1.17*	TUN	7.04
	DRC	1.33**	EGY	1.24**	BOT	2.30	BFO	6.16
	EGY	1.20*	NAM√	1.47	EQG	1.50	GAB	6.16
Neutral: RR ≈1	EQG	1.94	GNB	1.46	SWZ	1.43	BEN√	5.80
	MZM	1.54	BEN√	1.44	MAG	1.26	MOR	3.52
	GNB	1.39	CON√	1.42	NAM√	1.20	MZM	3.52
	TAZ√	1.35	GUI	1.39	GNB	1.13	MAA	1.61
	GAM	1.31	MAA	1.34	CAO√	1.12	CHA√	1.52
	CAO√	1.27	GAM	1.27	TAZ√	1.08	TAZ√	1.38
	GAB	1.22	ETH	1.26	GUI	1.06	SUD	1.19
	MLI	1.18	MAW√	1.09	EGY	1.00	EGY	1.12
	CON√	1.10	UGA√	1.08	CON√	0.98	ETH	1.08
	NAM√	1.03	LBR	1.04	UGA√	0.94	NIG√	1.04
	MAW√	0.96	CAO√	1.03	MAA	0.92	GHA	1.03
	NIG√	0.91	DRC	0.94	NIG√	0.91	NAM√	1.01
	CHA√	0.88	TAZ√	0.94	CHA√	0.90	CDI	0.93
	UGA√	0.88	ALG	0.92	ALG	0.89	UGA√	0.88
	BEN√	0.87	NIG√	0.91	GAB	0.86	DRC	0.82
	LBR	0.79	BUI	0.90	MAW√	0.72	CON√	0.74
	BUI	0.76	CHA√	0.90	NGR	0.71	MLI	0.47
	RWA	0.65	MLI	0.88	BEN√	0.64	CAO√	0.44
	NGR	0.63	BOT	0.88	SAF	0.69**	BUI	0.60**
Low: RR <1	ALG	0.89**	RWA	0.83	BUI	0.63**	CEN√	0.23**
	SAF	0.76*	NGR	0.74	MOR	0.63*	GNB	0.19**
	SUD	0.57**	MOR	0.45	LBR	0.63**	ANG√	0.16**
	MOR	0.54*	SAF	0.54*	BFO	0.61*	MAG	0.12**
	BFO	0.42*	SUD	0.30**	RWA	0.58**	NGR	0.10**
	CEN√	0.29**	CEN√	0.25**	SUD	0.53**	GUI	0.09**
	ANG√	0.19**	BFO	0.24**	CEN√	0.34**	LBR	0.06**
	SIE√	0.10**	ANG√	0.16**	ANG√	0.24**	SIE√	0.06**
	TUN	0.03**	SIE√	0.16**	SIE√	0.14**	RWA	0.05**
	ERI	.	TUN	0.05**	TUN	0.06**	LIB	.
	LIB	.	ERI	.	ERI	.	ERI	.
	SOM	.	SOM	.	SOM	.	SOM	.

ACLED events are riots, incidents of violence against citizens and battles; UCDP GED events are state-based conflicts, non-state-based conflicts, and incidents of one-sided violence. A correction of 0.5 was added to each count in the cross tabulation for UCDP GED summary cells with zero violent days. See text for other definitions.

*Significantly different from 1 at the 0.05 level. **Significantly different from 1 at the 0.01 level. √ Countries appearing in the same RR category in each model.

Sources: Author’s estimates based on Hyde & Marinov (2012), updated by the author to 2013; ACLED V.3, see Raleigh et al. (2010); and UCDP GED 1.5, see Sundberg & Melander (2013).

Electoral civil violence further decomposed

Thus far I have treated all days of civil violence as if they were identical. But civil violence covers an array of activities and all the components need not line up the same way. ACLED reports three subcategories of conflict events: riots (spontaneous or unorganized demonstrations against government institutions), violence against civilians (group attacks on unarmed members of the public), and battles (violent interactions between two politically organized armed groups at a particular time and location). The relative risk ratios for these events using 121-day election cycles are isolated by country in Table IV.

Looking across the table, the three variants of social conflict appear as one in about one-third of the sample. Riots, attacks on civilians, and battles all happen more often during election than non-election periods in Kenya, Libya, Senegal, and Togo. By contrast, the relative risk of all three types of events is lower during an election compared to other times in Angola, Central African Republic, and Sierra Leone. There is no change in the incidence rates across the board for Benin, Cameroon, Chad, Congo, Malawi, Namibia, Nigeria, Tanzania, and Uganda.

The remaining countries overlap from one category to another, with the most common pattern a constant risk of violent events in elections and out. Table IV thus suggests that for a majority of African countries, election-related civil violence is quite distinctive. Some experience a significant rise in riots but not in attacks on civilians or battles, which may mean that electoral conflict is largely a grassroots phenomenon in these places. Other countries face an increase in attacks during elections, possibly indicating that election civil violence is predominantly directed by government authorities against the general public. In a few countries, the relative risk of battles climbs, pointing toward more organized activity on both sides of electoral civil conflict. These are issues worth exploring in greater depth in another article.

Discussion and conclusion

In this article I considered an important empirical question about elections and civil violence in emerging democracies and competitive autocracies: are these political contests associated with a rise or drop in social conflict incidents compared to the ‘standing rate’ of conflict, which varies depending on qualities of the country? Focusing on Africa because of data availability and, more importantly, because it is a region that has recently and quickly moved away from single-party and military rule, I reported the relative risk of a violent country-day under a variety of scenarios.

Similar to some prior research, I found that Africa is far from homogeneous with regard to election-related civil violence. But prior studies have sometimes also been casual about differentiating the level of social conflict among African countries. Their results have loosely supported popular conjecture that election times in that region are frequently disruptive and violent – without being specific or consistent about what ‘frequently’ means.

After estimating the incidence rate of violent days at the national level, I observed three patterns that correspond to three hypotheses about elections in hybrid states. Institutionalists expect perverse outcomes, and for several African states anomalous surges in violent incidents around an election are the norm – although these

Table IV.

Relative risk of types of electoral civil violence in Africa

	121-day e-cycle with the occurrence of an ACLED event (1997–2013)
RR Group	Country code	RR riot	Country code	RR violence against civilians	Country code	RR battle
	EQG	7.75**	LIB√	5.20**	NAM	9.91**
	DJI	6.33**	LES	4.73**	TOG√	4.51**
	LIB√	5.71**	MZM	3.81**	MAG	4.49**
	BOT	4.03**	TOG√	3.25**	LIB√	3.33**
	GNB	3.13**	ZIM	2.77**	SEN√	2.44**
	TOG√	3.08**	KEN√	2.47**	KEN√	2.01**
High: RR >1	KEN√	2.83**	ZAM	2.38**	ETH	1.27*
	ETH	2.54**	CDI	2.19**	BEN√	6.71
	CDI	2.08**	SEN√	1.94**	MOR	3.22
	GHA	2.02*	SWZ	2.86**	ZAM	2.96
	SEN√	2.01**	GAB	2.32	SAF	1.96
	MLI	1.89**	GNB√	2.17	TAZ	1.88
	TAZ	1.81*	GAM√	1.86	ZIM	1.52
	DRC	1.80**	GHA	1.83	GHA	1.33
	LES	2.57	BEN√	1.83	CDI	1.32
	BDI	2.53	MAA√	1.64	CAO√	1.18
	CON√	2.13	CAO√	1.61	CON√	1.13
	ANG	1.81	DJI	1.46	DRC	1.09
	MZM	1.62	TAZ	1.40	UGA	1.01
	GNB√	1.55	EQG	1.29	MLI	1.00
	RWA	1.37	CHA	1.26	DJI	0.97
	LBR	1.32	ETH	1.21	ALG	0.89
	MAG	1.30	LBR	1.21	GNB√	0.87
	MAA√	1.17	MOR	1.18	NGR	0.80
	ZAM	1.12	DRC	1.10	GAM√	0.80
	MWI	1.12	MAG	1.08	MZM	0.79
Neutral: RR ≈1	CAO√	1.10	EGY	1.02	CHA	0.72
	GAB	1.06	MLI	1.01	MAA√	0.69
	ZIM	0.98	BDI	0.96	GNB	0.65
	EGY	0.96	CON√	0.93	BFO	0.60
	SWZ	0.86	GNB	0.90	NIG	0.77**
	UGA	0.86	MWI	0.89	SUD	0.56**
	SUD	0.80	NIG	0.89	EGY	0.53**
	CEN	0.73	ALG	0.88	LBR	0.44**
	NGR	0.73	SAF	0.83	BDI	0.39**
	GAM√	0.63	RWA	0.81	CEN	0.26**
	BEN√	0.52	BFO	0.75	RWA	0.17**
	NIG	0.69**	NAM	0.63	SIE√	0.06**
	ALG	0.59**	UGA	0.78*	ANG	0.00†
	SAF	0.52**	SUD	0.22**	BOT	0.00†
	BFO	0.37**	CEN	0.20**	EQT	0.00†
Low: RR <1	MOR	0.27**	ANG	0.19**	GAB	0.00†
	NAM	0.27**	NGR	0.16*	LES	0.00†
	SIE√	0.23**	SIE√	0.07**	MWI	0.00†
	TUN√	0.04**	BOT	0.00†	SWZ	0.00†
	CHA	0.00†	TUN√	0.00†	TUN√	0.00†

*Significantly different from 1 at the 0.05 level. **Significantly different from 1 at the 0.01 level. † There were no violent events during election periods, resulting in an indeterminately low risk ratio. √ Countries appearing in the same RR category in each model.

Sources: Author’s estimates based on Hyde & Marinov (2012), updated by the author to 2013; ACLED V.3, see Raleigh et al. (2010); and UCDP GED 1.5, see Sundberg & Melander (2013).

measurements are sensitive to the definitions of an election period and type of incident. As the democratic learning hypothesis expects, however, I also identified states that have a consistent drop in contemporaneous disorder around elections. This pattern seems most closely associated with post-conflict conditions. In the majority of African states, though, I could determine no noticeable change in the relative risk of civil violence in the course of elections. Many of these states are disorderly, but political campaign periods appear to be unrelated to variation in the extent of disorder – which seems to accord with views of how competitive autocracies behave.

By extrapolation, similar outcomes are possible in newly democratizing states generally, although circumspection is warranted given the short time period looked at and the special characteristics of the Africa region with regard to poverty and state fragility. Still, it seems plausible that other regions would experience a similar range of outcomes, with elections in different countries having a positive, negative or neutral correlation with the level of civil violence.

As with any observational study, caution should be exercised in the causal interpretation of the results. This article has not attempted to show why countries follow these contrasting trajectories, with some on a democratic learning track and others on competitive authoritarian and weak institutional tracks. But by proposing a transparent and easily reproduced method for measuring electoral civil violence, it does suggest a practical dependent variable for further research in this important area.

Even without being able to say much about the contributory factors in election violence, however, this article has three obvious implications for policy. First, there is no general reason to preclude or defer elections in Africa as a way to minimize social conflict around elections, although there may be specific reasons to do so in particular countries. Second, and related to the first point, it is possible to identify countries with serious and consistent problems of violent elections, which may help focus policymakers’ attention on distinctive national needs for improving security during voting campaigns. Third, another group of countries exists that seems to have been comparatively successful in preventing political competition from deteriorating into civil violence. These positive experiences may have lessons that can be transferred to other national settings.

Footnotes

Replication data

The dataset for the empirical analysis in this article can be found at .

Acknowledgment

I would like to thank Emily Goldsmith for technical assistance with the statistical analysis in this article.

Notes

References

Anifowose

Remi

Odukoya

Adelaja

(2013) Toward curbing electoral violence in Nigeria. In: Ayoade

John

Akinsanya

Adeoye

(eds) Nigeria’s Critical Election, 2011. Lanham, MD: Lexington, 283–312.

Bekoe

Dorina

(2010) Trends in electoral violence in sub-Saharan Africa. US Institute of Peace (http://www.usip.org/publications/trends-in-electoral-violence-in-sub-saharan-africa).

Brancati

Dawn

Snyder

Jack L

(2012) Time to kill: The impact of election timing on postconflict stability. Journal of Conflict Resolution 57(5): 822–853.

Carey

Sabine C

(2007) Rebellion in Africa: Disaggregating the effect of political regimes. Journal of Peace Research 44(1): 47–64.

Carr

Adam

(2014) Psephos Elections Archives (http://psephos.adam-carr.net/).

Cederman

Lars-Erik

Gleditsch

Kristian Skrede

Hug

Simon

(2013) Elections and ethnic civil wars. Comparative Political Studies 46(3): 387–417.

Cheibub

José

Hays

Jude

Savun

Burcu

(2012) Elections and civil war in Africa. Unpublished manuscript, University of Illinois (http://visionsinmethodology.org/wp-content/uploads/2012/02/Savun_VIM2012.pdf).

Diamond

Larry

(2002) Thinking about hybrid regimes. Journal of Democracy 13(2): 21–35.

Dunning

Thad

(2011) Fighting and voting: Violent conflict and electoral politics. Journal of Conflict Resolution 55(3): 327–339.

10.

Eck

Kristine

(2012) In data we trust? A comparison of UCDP GED and ACLED conflict events datasets. Cooperation and Conflict 47(1): 124–141.

11.

Flores

Thomas Edward

Nooruddin

Irfan

(2012) The effect of elections on postconflict peace and reconstruction. Journal of Politics 74(2): 558–570.

12.

Freedom House (annual) Freedom in the World (http://www.freedomhouse.org/).

13.

Hafner-Burton

Emilie M

Hyde

Susan D

Jablonski

Ryan S

(2014) When do governments resort to election violence? British Journal of Political Science 44(1): 149–179.

14.

Harish

Little

Andrew T

(2013) The political violence cycle. Paper presented at the Berkeley Political Economy Seminar, October (http://bcep.haas.berkeley.edu/papers/Fall%202013/Little.pdf).

15.

Howard

Marc

Roessler

Philip

(2006) Liberalizing electoral outcomes in competitive authoritarian regimes. American Journal of Political Science 50(2): 365–381.

16.

Human Rights Watch (2011) Perpetual fear: Impunity and cycles of violence in Zimbabwe (http://www.hrw.org/sites/default/files/reports/zimbabwe0311webwcover_0.pdf).

17.

Huntington

Samuel P

(1968) Political Order in Changing Societies. New Haven, CT: Yale University Press.

18.

Huntington

Samuel P

(1972) Civil violence and the process of development. Adelphi Papers 11(83): 1–15.

19.

Hyde

Susan

Marinov

Nikolay

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

20.

Levitsky

Steven

Way

Lucan A

(2010) Competitive Authoritarianism: Hybrid Regimes After the Cold War. Cambridge: Cambridge University Press.

21.

Lindberg

Staffan

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

22.

Linebarger

Christopher

Salehyan

Idean

(2015) Elections and social conflict in Africa, 1990–2009. Studies in Comparative International Development 50(1): 23–49.

23.

Mansfield

Edward D

Snyder

Jack

(2002) Democratic transitions, institutional strength, and war. International Organization 56(2): 297–337.

24.

Marshall

Monty G

Gurr

Ted R

Jaggers

Keith

(2014) Polity IV Project: Political regime characteristics and transitions, 1800–2013. Vienna, VA: Center for Systemic Peace (http://www.systemicpeace.org/polityproject.html).

25.

Matanock

Aila

(2013) Bullets for ballots: Examining the effect of electoral participation on conflict recurrence. New Haven, CT: Yale University, Georg Walter Leitner Program in International and Comparative Political Economy (http://www.yale.edu/leitner/resources/papers/Matanock_2013_Bullets%20for%20Ballots.pdf).

26.

Morse

Yonatan

(2012) The era of electoral authoritarianism. World Politics 64(1): 161–198.

27.

Nunley

Albert C

(2013) African Elections Database (http://africanelections.tripod.com/index.html).

28.

Raleigh

Clionadh

Dowd

Caitriona

(2015) Armed Conflict Location and Event Data Project (ACLED) codebook (http://www.acleddata.com/wp-content/uploads/2015/01/ACLED_Codebook_2015.pdf).

29.

Raleigh

Clionadh

Linke

Andrew

Hegre

Håvard

Karlsen

Joakim

(2010) Introducing ACLED: An Armed Conflict Location and Event Dataset. Journal of Peace Research 47(5): 651–660.

30.

Saideman

Stephen M

Lanoue

David

(2005) The (exaggerated) perils of democracy: Analyzing democracy’s influence on different forms of communal dissent. Paper presented at the Annual Meeting of the American Political Science Association, Washington, DC, 1–4 September (http://profs-polisci.mcgill.ca/saideman/saideman%20and%20lanoue,%20apsa%202005.pdf).

31.

Schedler

Andreas

(2006) Electoral Authoritarianism: The Dynamics of Unfree Competition. Boulder, CO: Lynne Rienner.

32.

Schmitter

Philippe C

Karl

Terry Lynn

(1991) What democracy is … and is not. Journal of Democracy 2(3): 75–88.

33.

Schrodt

Philip A

(2012) Precedents, progress, and prospects in political event data. International Interactions 38(4): 546–569.

34.

Straus

Scott

Taylor

Charles

(2012) Democratization and electoral violence in sub-Saharan Africa, 1990–2007. In: Bekoe

Dorina

(ed.) Voting in Fear: Electoral Violence in Sub-Saharan Africa. Washington, DC: US Institute of Peace, 55–38.

35.

Sundberg

Ralph

Melander

Erik

(2013) Introducing the UCDP Georeferenced Event Dataset. Journal of Peace Research 50(4): 523–532.

36.

Taylor

Charles

Pevehouse

Jon

Straus

Scott

(2013) Perils of pluralism: Electoral violence and competitive authoritarianism in sub-Saharan Africa. Vancouver: Simons Papers in Security and Development, no. 23 (http://summit.sfu.ca/item/14904).

37.

United Nations Economic Commission for Africa (2013) African governance report III: Elections and the management of diversity in Africa. Executive summary. Addis Ababa: UNECA (http://www.uneca.org/publications/african-governance-report-iii).

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB