Group organization, elections and urban political mobilization in the developing world

ABSTRACT

Elections generate incentives for contention and violence. However, collective action problems mute responses to strategic incentives by unorganized individuals, relative to organized groups. Variation in the severity of collective action problems and the degree of strategic behaviour results in distinct patterns of mobilization across these two types of groups that have been overlooked in previous literature. We explore variation in organized and unorganized political mobilization and violence at elections using new event data for over one hundred cities in the developing world from 1960 to 2014. We find that organized groups are more likely to mobilize before elections to influence their outcome, and under permissive opportunity structures at moderate levels of democracy. Mobilization by unorganized individuals occurs at and directly after elections but does not vary by regime type. Distinct mobilization patterns across group type are a major addition to our understanding of the link between elections, democracy, contention and violence.

KEYWORDS:

• elections
• mobilization
• violence
• cities
• urban social disorder
• authoritarianism
• democracy

Introduction

Elections have come to occupy a central position in the literature on authoritarian regime durability, democratization and conflict. Electoral competition is seen, on the one hand, as a path to stable autocratic rule,¹ and on the other hand, as an opportunity for transitions to democracy.² Moreover, a growing literature finds that elections in illiberal regimes are associated with contention and violence.³ Arguments linking voting to mobilization, violence and repression focus on changes in the strategic environment facing incumbents and the opposition. For example, for Fjelde and Höglund, majoritarian electoral systems lead to zero-sum conflict between constituencies and therefore a higher risk of violence.⁴ Harish and Little argue that elections generate incentives for violence aiming to influence political outcomes.⁵ In such accounts, incumbent and opposition actors are depicted as unitary and strategic. However, groups that mobilize in the context of electoral competition face collective action problems impeding their responses to strategic incentives.⁶ These are likely to be particularly severe for the opposition, which encounters potential repression costs.⁷

In this article, we make a theoretical contribution to the literature on electoral mobilization and violence by exploring problems inherent to strategic collective action. We argue that the ability of unorganized groups to make rational cost–benefit mobilization decisions will be muted compared to organizations such as political parties and labour unions. Variation in the severity of collective action problems and degree of strategic behaviour results in unexplored patterns of mobilization across these two types of groups, which challenge previous studies’ underlying assumptions about the ability of groups to respond to strategic incentives.

We relate this theoretical argument to two important, established empirical findings on electoral mobilization and violence. First, to the timing of these events. Second, to the relationship between regime type and contentious collective action. Organized groups can more easily solve the collective action problem to act strategically and influence the outcome of a vote. We argue such organized mobilization will occur both before and concurrently with elections, and under regimes characterized by institutional uncertainty and permissive opportunity structures. Unorganized individuals, on the other hand, are less strategic. Elections serve as focal points for mobilization by supporters of political parties or candidates, and loosely coordinated groups such as criminal gangs or ethnic factions. This mobilization often takes the form of spontaneous protests or rioting. It is sparked by the process of voting and the declaration of election results but does not vary in response to the broader strategic environment. We argue that this type of unorganized mobilization will occur simultaneously with and after elections, but its frequency will not vary by regime type.

We make an empirical contribution by testing these arguments on a new event dataset measuring political mobilization and violence in over one hundred cities in the entire developing world from 1960 to 2014. We model the effects of elections on collective action and mobilization relative to the baseline frequency of these phenomena in the absence of voting. We include both a much longer time period and a greater geographical scope than previous studies, which begin no earlier than 1980 and are predominantly restricted to Africa. Our data also allow us to make a novel distinction between organized and spontaneous, unorganized mobilization.

We find that organized urban social disorder is more frequent in election months and months preceding an election. Unorganized disturbances, on the other hand, are significantly more frequent in election months and the month directly thereafter. Further, we find that there is an inverted-U shaped relationship between democracy and the effect of elections on organized urban social disorder. Mobilization at elections by organized groups is much more frequent at moderate levels of democracy. The effect of elections on unorganized urban disturbances, on the other hand, is not moderated by regime type. This relationship holds across three measures of democracy, though it is strongest when using the well-known Polity index. Our empirical results confirm that previous accounts of elections, mobilization and violence have over-stated groups’ ability to respond to complex strategic incentives with collective mobilization. This ability is concentrated among organized groups, while unorganized gangs’ or ethnic factions’ mobilization is less strategic.

Previous literature

A large literature examines the effects of elections on political instability, mobilization and violence. Few studies differentiate between different types of actors involved in electoral contention, however. Although both Beaulieu and Brancati specifically conceive of these processes being instigated by the opposition, Hafner-Burton, Hyde and Jablonski, Fjelde and Höglund, and Choi and Raleigh are relatively rare in using measures which attempt to differentiate between the actions of incumbents and their challengers, or state and non-state actors.⁸ Authors such as Daxecker go to lengths to isolate events relating to elections for their analyses, but do not differentiate between types of actors, while Salehyan and Linebarger, and Harish and Little, examine elections’ effects on baseline levels of social conflict by actors of all types.⁹

Importantly, no previous studies differentiate between election-related mobilization and violence committed by organized groups such as political parties, governments and labour unions, and that committed by unorganized groups of individuals. This is an important omission, due to the degree of strategic behaviour involved in authors’ arguments linking electoral mobilization and violence. Fjelde and Höglund, for example, argue that politicians can react to complex institutional incentives given by the electoral system when mobilizing organizationally weak groups in protest and violence.¹⁰ Daxecker argues that both state and non-state actors respond to the presence of international election observers by restricting mobilization and violence to the period before elections.¹¹ For Harish and Little, actors have incentives to mobilize at the time of voting to influence political outcomes.¹² Choi and Raleigh argue that state forces and militias respond to election results and target violence in places costly to the national leader or swing areas.¹³ These arguments and others assume that groups of individuals can relatively unproblematically engage in large-scale strategic collective action around elections.

Organized and unorganized mobilization at elections

We do not argue that existing arguments linking the strategic environment surrounding elections to mobilization and violence are wrong. Instead, we argue that due to unresolved collective action problems, unorganized groups’ response to strategic incentives will be muted relative to organizations such as political parties and labour unions. Therefore, although we analyse collective political mobilization targeting either another group or the government, we distinguish between spontaneous mobilization and that coordinated by a singular political or identity group.

To illustrate, consider the example of the Malaysian Coalition for Clean and Fair Elections, a formal group of non-governmental organizations, civil society groups and opposition parties. This group strategically mobilized hundreds of thousands in street protests in Kuala Lumpur before parliamentary elections in 2008 and 2013. It demanded institutional reforms to reduce the electoral advantage of the authoritarian ruling coalition. Protests were timed before elections, when they had the greatest chance of achieving the organization's stated goals.¹⁴ By contrast, consider events in the evening of the Malaysian 1969 election. Spontaneous riots broke out in the capital as opposition supporters confronted members of rival ethnic groups, leading to several deaths and a national wave of contention. Although politicians were involved in these riots, they were not coordinated by any formal organization. This mobilization was not strategic, but instead was sparked by the announcement of election results and gave voice to pre-existing grievances and tensions which ran predominantly along ethnic lines.¹⁵ Similar riots and violence have repeatedly marked the post-election period in Kenya and Nigeria since the 1990s.¹⁶ We argue that organized mobilization, like that by the Malaysian Coalition for Clean and Fair Elections, is strategically timed to occur before elections and under permissive opportunity structures at moderate levels of democracy. Spontaneous mobilization, like the post-election violence in Malaysia in 1969, is most likely to occur in the immediate wake of elections and does not vary with regime type.

Periods surrounding elections are regularly associated with contentious mobilization and violence.¹⁷ Actors hoping to influence political outcomes have incentives to schedule activities at this time. These include demonstrations or protests supporting or opposing candidates or parties, strikes and terrorist attacks.¹⁸ Polling and the announcement of election results are focal points for spontaneous mobilization, causing revolutionary band-wagons that rapidly escalate into mass contention.¹⁹ Because of the high stakes of elections for both government and opposition supporters, the scope of urban disorder at election time is likely to broaden through cycles of protest, counter-protest and repression. This leads us to our first testable hypothesis,

Hypothesis 1.1: Organized and unorganized urban social disorder will be more frequent in election months, holding all else constant.

Distinct mobilization processes are likely to affect the timing of organized versus unorganized disorder. Elections involve competition between, and mobilization by, political parties and campaigns whose activities naturally peak on election day. However, governments and other organized political and identity groups also lead, cooperate with or join campaigns; seek to affect the constellation of power within parties; and attempt to intimidate opponents and voters.²⁰ These groups aim to influence the outcomes of elections, and they have the organizational capacity to strategically time interventions for maximum effect. They are therefore likely to mobilize supporters in demonstrations, rallies, canvassing and other activities before voting. Returning to the example above, the broad Malaysian coalition of political parties and civil society groups promoting institutional reforms and opposition candidates organized large street demonstrations before 2008 and 2013 parliamentary elections to influence their results. Because organized groups can act strategically to influence the results of elections, we expect that organized mobilization will be most likely to occur before polling takes place.

Unorganized urban disturbances are more likely to occur at the same time as voters go to the polls, and in the direct aftermath of voting. Without organizations to coordinate collective action, mobilization by large groups is challenging and unlikely, particularly when individuals face potential repression costs.²¹ Elections, however, serve as focal points for individual supporters of political parties or candidates who mobilize legitimately, as well as loosely coordinated groups such as criminal gangs or ethnic factions. Polling and the announcement of election results provide sparks that cause mobilization of otherwise unorganized populations, as pre-existing structures of identity, patronage and grievances erupt spontaneously or through manipulation by elites into contention or violence.²² This sort of mobilization often takes the form of spontaneous protests or riots. The distinct logics by which organized groups mobilize to influence electoral outcomes, while unorganized populations act collectively to contest them, generate the following two testable hypotheses,

Hypothesis 1.2: Organized urban social disturbances will be more frequent in election months and months before elections, holding all else constant.

Hypothesis 1.3: Unorganized urban social disturbances will be more frequent in election months and months after elections, holding all else constant.

Broader institutional contexts also have significant effects on mobilization that vary across organized and unorganized groups.²³ Strategic incentives for mobilization and violence at elections are greater under regimes where the stakes of the vote are higher and the costs of mobilization are lower. Many autocracies hold elections but only rarely allow for a fair and free poll or respect the result and step down in the face of an electoral defeat.²⁴ As regimes democratize, however, the likelihood that an election will result in a change of government increases. This gives actors incentives to engage in irregular political mobilization, because the potential payoff to winning the vote is greater than under autocracy. Democratization also mitigates repression costs, incentivizing mobilization around elections.²⁵

Institutionalization of democracy lessens incentives for mobilization by reducing uncertainty. Institutional uncertainty arises from incongruence between formal and informal institutions, the novelty of democratic practices, and the inability of formal rules to guarantee credible commitments.²⁶ In weakly institutionalized democracies, the formal institution of elections conflicts with informal institutions of corruption, violence and coercion; and repression and voter intimidation are commonly used by weak politicians to suppress the opposition vote, leading to contestation.²⁷ There is significant institutional uncertainty regarding the willingness of incumbents to comply with election results and political parties are more likely to mobilize and protest to reject outcomes they consider illegitimate or unfair.²⁸ In fully institutionalized democracies, on the other hand, uncertainty is reduced as all parties prefer to accept the results of elections rather than using extra-electoral measures to contesting them.²⁹

We argue that these mechanisms linking democratization to increasing, then decreasing, urban social disorder are likely to have effects on organized disturbances, rather than spontaneous, unorganized mobilization. Organizations such as political parties or labour unions are more likely to accurately weigh the costs and benefits of collective action, for example through negotiations with the government.³⁰ They are also more likely to control their members and confine mobilization to the times and institutional settings when it is most likely to generate a positive payoff.³¹ Unorganized urban masses or gangs, on the other hand, may face a clear cost–benefit calculation in regards to repression, for example. However, they have less capacity to accurately gauge the costs and benefits of mobilization and may have idiosyncratic, private or unclear goals for collective action. They are relatively difficult to control and repress, making it less likely that unorganized urban disorder at elections is concentrated among regimes with a middling degree of democracy. For these reasons, we expect that the inverted-U shaped relationship between democratization and urban social disorder at elections will hold only for organized disturbances:

Hypothesis 2.1: The frequency of organized urban social disturbances in election months will be greatest at moderate levels of democracy, holding all else constant.

Hypothesis 2.2: The frequency of unorganized urban social disturbances in election months will not vary by level of democracy, holding all else constant.

Data and research design

We analyse an extended and updated version of the PRIO Urban Social Disorder database (USD 2.0).³² USD 2.0 contains details on urban social disorder events in 102 cities in 98 countries in Africa, Asia, Latin America and the Middle East from 1960 to 2014. Urban social disorder is defined as non-routine political activity and political violence which occurs in cities. This excludes collective action such as mass demonstrations of grief at funerals or canvassing before elections and also excludes routine violence such as arrests.³³ Urban areas included in the dataset were national capitals, former capitals and some large urban centres with a population over one million.³⁴ As illustrated in Figure 1, USD indicates increasing rates of urban social disorder since the 1960s, and that disorder has become increasingly concentrated in Asia, the Middle East and North Africa.

Figure 1. Urban social disorder by region, monthly, 1960–2014.

Our models estimate the number of Disturbances in each city-month. The events included in this indicator include protests, demonstrations, armed clashes, riots, government repression and terrorism. The indicator explicitly excludes warfare. We disaggregate disorder events into Organized Disturbances coordinated by a group or committed by an organization or government, and uncoordinated and spontaneous Unorganized Disturbances. The latter are confined to spontaneous demonstrations and spontaneous violent riots.³⁵ Organized events are much more common. The city-month mean of the Organized Events variable is 0.10 and the maximum is 10, while the Unorganized Events variable averages only 0.02 and has a maximum of 3. In one robustness test, we model the binary Unrest (0/1) variable. Summary statistics are presented in the online Appendix in Table A1.

To capture democracy, we rely primarily on the Polity IV dataset.³⁶ This indicator measures constraints on the chief executive; competitiveness and openness of the selection of the chief executive; institutionalization and government restriction of political competition. Its advantage over other measures of regime type is its inclusion of both indicators of electoral institutions and civil liberties. These two facets of regime type most clearly reflect our theoretical interest in the costs and benefits of collective action around elections for organized and unorganized groups, and are less comprehensively captured in alternative measures of democracy. The Polity variable ranges from −10 to 10, where −10 equals a consolidated autocracy and 10 consolidated democracy. The average city-month in the dataset had a Polity score of −0.9, indicating an anocracy.

The use of Polity in conflict research has been criticized because its political participation components capture factionalism, for example, the establishment of rival governments; civil war; or intense, hostile and violent political competition.³⁷ The kinds of predominantly non-state social disorder events we study here are less likely to be associated with factionalism than is civil war, so the endogeneity concern is much less severe in our case. Nonetheless, we exclude warfare-related events from our dependent variable. We also control for both civil conflict and coup attempts in our models, which are the main forms of factionalism captured by Polity.

Further still, factionalism in the form of “intense, hostile and violent political competition” could include very severe or prominent urban social disorder events and be reflected in moderate scores on Polity. To address this concern, our robustness tests include Vreeland's XPolity variable – which excludes the problematic components of Polity related to the competitiveness and regulation of political participation – in place of Polity and are reported in Table A5 in the online Appendix.³⁸ Finally, robustness tests use the V-Dem Electoral Democracy Index (Polyarchy) variable in place of the Polity score.³⁹ Polyarchy measures the extent to which elections affect the composition of the chief executive; are unmarred by fraud or irregularities; suffrage is extensive; political and civil society organizations can operate freely; and there are free and independent media. Models including Polyarchy are presented in Table A6 in the online Appendix.

Election data are from the National Elections across Democracy and Autocracy (NELDA) dataset.⁴⁰ NELDA covers all elections from 1945 to 2012 which allowed mass voting and allows distinctions between types of elections. We do not distinguish between executive and parliamentary elections, with the dichotomous Election variable capturing both types.

We control for irregular leader exits and coups, city population, GDP per capita in constant 2000 US dollars, recessions and ongoing conflict. All models include both contemporaneous controls plus a lag of one month of all control variables. Excluding the lagged control variables does not change our results.⁴¹

Models and results

We estimate negative binomial models of counts of organized and unorganized unrest events at the city-month level. Following Salehyan and Linebarger and Harish and Little, we are interested in the effects of elections over and above the baseline frequency of mobilization events.⁴² We use city and year fixed effects to look only at temporal variation in events within cities. We first test Hypotheses 1.1–1.3 by estimating the relationship between elections and urban unrest while holding the level of democracy constant, $y_{m,c}={\alpha }_1+{\text{β}}_{1}Electio{n}_{m,n}+{\text{β}}_{2}Polit{y}_{m,n}+{\text{β}}_3{X}_{t,c}+{\gamma }_c+{\nu }_t+ϵ,$where $y_{m,c}$ is a count of social disturbances in month m and city c, ${\alpha }_1$ is a common intercept, $Electio{n}_{m,n}$ is a binary indicator of an election in month m and country n, $Polit{y}_{m,n}$ is the Polity score of country n in month m, $X_{t,c}$ is a vector of control variables, ${\gamma }_c$ is a city fixed effect, ${\nu }_t$ is a year fixed effect and $ϵ$ is an error term. In several models, we include leads and lags of Election.⁴³ Following the specification of our hypotheses, we estimate separate regressions for organized and unorganized disturbances. However, as a robustness test, we also estimate models of both dependent variables simultaneously as a system of equations via seemingly unrelated linear regression.⁴⁴ This allows the two models to be linked via their error terms and for us to directly test differences in the effects of elections on different dependent variables.

We then test Hypotheses 2.1–2.2 by allowing the effects of elections on unrest to vary by the level of democracy. These models are specified identically to those above, but include an interaction term of the Election and Polity indicator, $y_{t,c}={\alpha }_1+{\text{β}}_{1}Electio{n}_{m,n}+{\text{β}}_{2}Polit{y}_{m,n}+{\text{β}}_{3}Electio{n}_{m,c}\times Polit{y}_{m,c}+{\text{β}}_4{\mathit{X}}_{\mathit{t},\mathit{c}}+{\gamma }_c+{\nu }_t+ϵ.$In these models, the parameter ${\text{β}}_3$ estimates the difference between the effect of the Polity indicator on unrest when there is an election to its effects when there is no election. As above, our main models are estimated separately for organized and unorganized disturbances. However, we also estimate models via seemingly unrelated regression, as a robustness test, with results presented in Figure A1 in the online Appendix.

Disturbance timing and type

In Table 1, we present the results of six negative binomial regressions and two seemingly unrelated linear regressions. First, we test whether cities experience more social disturbances in election months, distinguishing between organized and unorganized disturbance events. Second, we estimate three models testing whether cities also experience more social disturbances in the three months preceding, and following, an election. Key results from Models 1.5 and 1.6 are presented in Figure 2.⁴⁵

Figure 2. Results, Models 1.5 and 1.6, Table 1.

Table 1. Elections, timing and urban disturbances.

	(1.1)	(1.2)	(1.3)	(1.4)	(1.5)	(1.6)	(1.7)	(1.8)
	All Evs	Org Evs	Unorg Evs	All Evs	Org Evs	Unorg Evs	Org Evs	Unorg Evs
Election	0.45***	0.41***	0.58***	0.45***	0.41***	0.58***	0.05***	0.01***
	(0.08)	(0.09)	(0.16)	(0.08)	(0.09)	(0.17)	(0.01)	(0.00)
Election (m-1)				0.03	0.04	−0.05	0.00	−0.00
				(0.10)	(0.10)	(0.23)	(0.01)	(0.00)
Election (m-2)				0.13	0.20**	−0.32	0.02*	−0.01
				(0.09)	(0.10)	(0.24)	(0.01)	(0.00)
Election (m-3)				−0.01	0.02	−0.10	0.00	−0.00
				(0.09)	(0.10)	(0.20)	(0.01)	(0.00)
Election (m+1)				0.19**	0.11	0.39**	0.01	0.01**
				(0.09)	(0.09)	(0.17)	(0.01)	(0.00)
Election (m+2)				0.07	0.07	0.04	0.00	0.00
				(0.09)	(0.10)	(0.19)	(0.01)	(0.00)
Election (m+3)				−0.05	−0.01	−0.22	0.00	−0.00
				(0.10)	(0.10)	(0.24)	(0.01)	(0.00)
Polity	0.01	0.01	0.02	0.00	0.01	0.02	−0.01*	0.00
	(0.02)	(0.02)	(0.05)	(0.02)	(0.02)	(0.05)	(0.00)	(0.00)
Polity*Polity	−0.01***	−0.01***	−0.01***	−0.01***	−0.01***	−0.01***	−0.00***	−0.00***
	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Coup	2.09***	2.18***	1.60***	2.09***	2.18***	1.60***	0.51***	0.09***
	(0.08)	(0.08)	(0.17)	(0.08)	(0.08)	(0.17)	(0.02)	(0.01)
Conflict	0.19	0.26	−0.13	0.18	0.26	−0.13	0.05**	−0.00
	(0.17)	(0.18)	(0.32)	(0.17)	(0.18)	(0.32)	(0.02)	(0.01)
Observations	55,392	55,392	55,392	55,220	55,220	55,220	55,220	55,220
AIC	37,131.72	32,109.86	10,351.69	37,014.69	31,995.04	10,342.54

Models 1.7 and 1.8 estimated by seemingly unrelated regression.

Robust standard errors in parentheses.

All models with city and year fixed effects, controls for city population, GDP/capita, recessions.

Analysis years 1960–2011.

* p < 0.10, ** p < 0.05, *** p < 0.01.

The first three models presented in Table 1 include the binary Election indicator of a national poll alongside all control variables described above. These regressions find a significant positive association between elections and urban social disorder, whether organized or unorganized. The largest immediate effects of elections are on unorganized events, however. Model 1.1 takes the total number of events as dependent variable. The average number of events when no election occurs is 0.12, holding all else constant. In election months the average number of disorder events is significantly higher, at 0.19. This is a substantively large, 50% increase in the predicted frequency of disorder. Model 1.2 takes the number of organized disorder events as dependent variable. Elections are associated with a significant and similarly sized increase in the predicted number of these events in a city-month, from 0.10 to 0.15, or around 50%. Our results suggest, interestingly, that the strong contemporaneous association of elections with urban disorder is driven more by unorganized events than organized events, despite the relative infrequency of unorganized events overall. Model 1.3, which takes the number of unorganized disorder events as its dependent variable, bears this out. The increase in the predicted number of events in an election month is very large in this model, doubling from 0.02 to 0.04.

To summarize the results of these three models, all show a very robust and sizable contemporaneous association between elections and urban social disorder, regardless of whether it is organized or unorganized. However, unorganized disturbances play a disproportionately large role in the general association between elections and disorder. Although unorganized disturbances are rare, they are twice as likely during election months. This suggests elections perform a very important coordinating role, facilitating forms of unorganized collective mobilization that are otherwise relatively uncommon.

The next three Models 1.4–1.6 in Table 1 include three leads and lags of the Election indicator, capturing effects of elections on disturbances three months before and after polling, as well as contemporaneous effects.⁴⁶ Significant differences emerge in the relationship between elections and organized versus unorganized unrest events. Model 1.4 takes all disturbance events as its dependent variable. Here, there is a strong contemporaneous association with elections, of a magnitude similar to Model 1.1 above. Election months witness 0.06 more events than non-election city-months, holding all else constant. However, this model also shows significantly greater numbers of disturbances in the month directly following an election. In Model 1.4, city-months directly following elections experienced 0.02 more disturbances than non-election months, and this difference is significant at the p < 0.03 level. The magnitude of elections’ effect in the month following voting is around one-third of the contemporaneous effect. All coefficients on further leads and lags of elections are statistically insignificant, indicating that aggregate urban social disorder occurs contemporaneously with polling, persisting but dissipating over the subsequent month.

However, Model 1.5, taking only organized urban disturbances as its dependent variable, generates different results. It finds significantly more organized disturbances in months before elections. As we illustrate in the left-hand panel of Figure 2, the contemporaneous effect of elections on this type of unrest is similar in magnitude and statistical significance to that found in previous models, at 0.04. In contrast to Model 1.4, however, the results of this regression indicate that organized unrest is also significantly greater two months before an election. The coefficient on the second lead of the Election indicator is positive and statistically significant at the p < 0.035 level in Model 1.5. The second city-month preceding an election witnesses 0.02 more organized disturbances than non-election months, holding all else constant. The magnitude of this effect is only around half of the contemporaneous increase in organized disturbances associated with elections. However, it is statistically significant, and of a similar magnitude to the increase in disturbances in the month directly following polling in Model 1.4. This result is the first evidence of a distinct mobilization process underlying organized disturbances, particularly when we note that there are not significantly greater numbers of organized events in months following elections in this model. The coefficients on all lags of Election are small and statistically insignificant in Model 1.5, in contrast to Model 1.4.

Differences in patterns of organized versus unorganized mobilization are reinforced by the results of Model 1.6, with unorganized disturbances as dependent variable. We graph the results of this model in the right-hand panel of Figure 2. We find no significant increases in the number of disturbances in months preceding an election, and in fact the coefficient on the second lead of the Election variable is negative, though statistically insignificant. As in Models 1.4 and 1.5, elections are associated with a significant contemporaneous increase in unorganized disturbances. However, in contrast to organized disturbances in Model 1.5, here we see a significantly greater number of unorganized disturbance events in the month directly following an election, an increase of 0.01 events which is significant at the p < 0.024 level. This is a substantively large, 50% increase in the number of unorganized events, which are rare overall.

Models 1.7 and 1.8 are estimated as a system of equations via seemingly unrelated linear regression. The results of these models are similar to Models 1.4–1.6. Election months are associated with more frequent disturbances, both organized and unorganized. However, the coefficient on the 2-month lead of the Election variable is positive and statistically significant at the p < 0.09 level in Model 1.7, which takes organized disturbances as its dependent variable, while the same coefficient is negative and statistically insignificant in Model 1.8, of unorganized disturbances. A test of the null hypothesis that these coefficients are equal, with Bonferroni-adjusted p-values for multiple hypotheses, is rejected at the p < 0.03 level. From these models, we can conclude that organized disturbances are significantly more frequent before elections than unorganized disturbances. Model 1.8 also indicates that unorganized disturbances are significantly more frequent in the month after elections (p < 0.04). Model 1.7 of organized disturbances also generates a positive coefficient on the first lag of the Election variable, but this coefficient is not statistically significant (p < 0.45). A test, with Bonferroni-adjusted p-values for multiple hypotheses, fails to reject the null hypothesis that these coefficients are equal.

To sum up these three models of the timing of unrest around voting, election months are associated with all types of urban disturbances. Significant differences in the timing of urban social disorder emerge when we disaggregate disturbances by type. Elections are undoubtedly a significant contributing factor to the restiveness of cities in the developing world. However, they are spurring mobilization and contention by different actors in different ways. Organizations such as governments, political parties, labour unions and terrorist groups mobilize at significantly greater rates before and in the month of elections, but this mobilization does not spill over into the month after polling. Unorganized groups, on the other hand, are likely to engage in collective action in periods directly surrounding and following elections.

Models presented in Table 1 demonstrate a significant, non-linear relationship between democracy, as measured by the Polity indicator, and the frequency of urban social disturbances. The size of the effects of regime type on the frequency of disturbances is relatively large, holding all else constant. In Model 1.1 the frequency of all disturbances increases from 0.09 per city-month at a Polity score of −10 to 0.17 between scores of −2 and 2, before declining again to 0.09 at a score of 10. Increases in the predicted frequency of organized events are similarly sized, from 0.07 at very low or high Polity scores to 0.14 at middling values. Although the size of the increases in the frequency of unorganized disturbances are smaller across the range of the Polity variable, from 0.01 to 0.03, compared to average frequencies of this type of disturbance these increases are large. On average, then, hybrid regimes are more susceptible to all types of urban social disorder. We go on to test whether elections exacerbate this underlying structural vulnerability in the following section of the article.

Other forms of domestic political instability are strongly associated with urban social disturbances across all models. In particular, an attempted or successful coup is associated with significantly more urban disorder – 0.31 more disturbances on average in Model 1.1. The effects of coups tend to persist, with city-months directly after those witnessing a coup experiencing 0.05 more disturbances. Civil conflict is also associated with significant increases in the frequency of urban disturbances, an increase of 0.09 per city-month, on average. Larger cities are associated with much more frequent social disturbances. A shift from one standard deviation below the mean to one standard deviation above on the Ln City Pop variable results in a significant increase in the predicted number of disturbances from 0.06 to 0.23 in Model 1.1. GDP is not associated with the prevalence of urban social disorder in this model, and though recessions are associated with significantly fewer disturbances this result should be interpreted with caution as the Recession variable indicates only year-on-year declines in GDP per capita and not economic variation by month.

Democracy, elections and disturbance type

We now allow the effects of elections on urban social disorder to vary by level of democracy, in models reported in Table 2. Models 2.1–2.3 interact Election with Polity. These models test whether there is a monotonically increasing or decreasing relationship between elections and urban social disorder across the range of democracy. Models 2.4–2.6 interact the Election indicator both with the Polity score and its squared term, testing the prediction of an inverted-U shaped relationship between democratization and urban disorder laid out in Hypotheses 2.1–2.2.

Table 2. Democracy, elections and urban disturbances.

	(2.1)	(2.2)	(2.3)	(2.4)	(2.5)	(2.6)	(2.7)	(2.8)
	All Evs	Org Evs	Unorg Evs	All Evs	Org Evs	Unorg Evs	Org Evs	Unorg Evs
Election	0.47***	0.42***	0.59***	0.67***	0.75***	0.10	0.11***	0.00
	(0.08)	(0.09)	(0.16)	(0.15)	(0.16)	(0.33)	(0.02)	(0.01)
Polity	0.01	0.01	0.03	0.01	0.01	0.02	−0.01**	0.00
	(0.03)	(0.03)	(0.05)	(0.02)	(0.02)	(0.05)	(0.00)	(0.00)
Polity*Polity				−0.01***	−0.01***	−0.01***	−0.00***	−0.00***
				(0.00)	(0.00)	(0.00)	(0.00)	(0.00)
Election*Polity	−0.01	−0.00	−0.01	0.00	0.01	−0.02	0.00	−0.00
	(0.01)	(0.01)	(0.02)	(0.01)	(0.02)	(0.02)	(0.00)	(0.00)
Election*Polity*Polity				−0.01*	−0.01***	0.01*	−0.00***	0.00*
				(0.00)	(0.00)	(0.01)	(0.00)	(0.00)
Coup	2.13***	2.22***	1.64***	2.10***	2.18***	1.61***	0.51***	0.08***
	(0.08)	(0.08)	(0.17)	(0.08)	(0.08)	(0.17)	(0.02)	(0.01)
Conflict	0.24	0.31*	−0.08	0.19	0.26	−0.14	0.05**	−0.00
	(0.17)	(0.18)	(0.33)	(0.17)	(0.18)	(0.32)	(0.02)	(0.01)
Observations	55,392	55,392	55,392	55,392	55,392	55,392	55,392	55,392
AIC	37,226.55	32,175.26	10,391.18	37,130.13	32,107.36	10,349.21

Models 2.7 and 2.8 estimated by seemingly unrelated regression.

Robust standard errors in parentheses.

All models with city and year fixed effects, controls for city population, GDP/capita, recessions.

Analysis years 1960–2011.

* p < 0.10, ** p < 0.05, *** p < 0.01.

Models 2.1–2.3 take all urban disturbances, organized and unorganized disturbances as dependent variables, respectively. All of these models find a significant contemporaneous effect of elections on the frequency of disturbances, but none find that this relationship is moderated by the level of democracy in a country. The coefficients on Election are positive and statistically significant, but the coefficients on its interaction term with Polity are small and statistically insignificant. From these results, we conclude that there is no monotonic relationship between democratization and urban disturbances in election months.

Models 2.4–2.6 interact Election with Polity and its squared term, allowing the effects of elections on urban social disturbances to vary non-monotonically with the level of democracy. We plot key results of these models in Figure 3. Here, the marginal effect of Election is shown across the range of Polity in the left-hand panel, while in the right-hand panel the predicted number of events during election and non-election months is plotted across the range of Polity. These results indicate that total disorder, and organized disturbances, display a large, significant inverted-U shaped relationship with elections across the range of the Polity variable. The effect of elections on unorganized disturbances, on the other hand, is not moderated by the level of democracy in a country.

Figure 3. Results, Models 2.5 and 2.6, Table 2.

Model 2.4 takes the total number of disturbances as dependent variable. Results show the relationship between elections and urban disorder depends on regime type. At low levels of democracy below a Polity score of −8, elections are not significantly associated with contemporaneous increases in the frequency of urban disorder events, holding all else constant. However, as the Polity score increases across its range, the marginal contemporaneous effect of an election on the frequency of unrest increases to a maximum of 0.21 at values of Polity between −1 and 1. These are substantively large and statistically significant effects; they are around three times the magnitude of the contemporaneous effects of elections found in models presented in Table 1 which did not interact the Election indicator with Polity. As the Polity indicator increases further, the effects of elections on the frequency of disturbances decline to be statistically insignificant at Polity scores above 8.

Models of organized disturbances generate strikingly similar results to those of aggregate urban disorder. The results of Model 2.5 are graphed in the upper panel of Figure 3, and show increasing, then decreasing, marginal effects of elections on the frequency of organized disturbances across the range of the Polity variable. Increases in the marginal effect of the Election indicator are not as great as in Model 2.4, peaking at only 0.17 instead of 0.21, and are statistically significant only between Polity scores of −7 and 7. Nonetheless, when looking only at organized mobilization events, election months are associated with far greater numbers of disturbances under moderate levels of democracy than when we estimate their effects while simply controlling for the Polity score as in Table 1.

However, the effect of elections on organized urban disturbances displays striking differences to the relationship with unorganized events found in Model 2.6 and graphed in the lower panel of Figure 3. The effects of elections on unorganized disturbances are small and do not vary significantly across the range of Polity. The coefficients on the interactions of Election and Polity have the opposite signs in Model 2.6 versus 2.5. Declining from a maximum of 0.06 at a Polity score of −10 to 0.02 between 0 and 3, the effects then increase to reach 0.04 at a score of 10. The effect of elections on unorganized disturbances is therefore relatively unchanged across the level of democracy.

All these model results show a significant contribution of regime type to urban social disorder, independent of the occurrence of elections. However, the effects of regime type are much smaller than that of elections. The average frequency of all disturbances is around 0.08 greater at moderate levels of Polity than at its extremes, when Election is set to zero in Model 2.4. As we show in the upper-right panel of Figure 3, the average predicted frequency of organized events in Model 2.5 increases from a minimum of 0.07 at a Polity score of −10 to a maximum of 0.13 at scores between −3 and 3, before declining to 0.08 at Polity score of 10. As we show in the lower-right panel of Figure 3, the overall frequency of unorganized unrest is significantly greater at moderate levels of democracy, though the effects of elections do not vary by regime type. The predicted number of unorganized events in Model 2.6 is 0.01 at the extremes of Polity but increases to a maximum of 0.03 at moderate levels of the variable.

As a robustness test, we also estimated models of organized and unorganized disturbances together as a system of equations via seemingly unrelated regression. The results of these models are presented as Models 2.7 and 2.8 in Table 2 and in Figure A1 in the Appendix. They are substantively identical to the results of Models 2.5 and 2.6, but because they allow us to directly compare the effects of the independent variables on different dependent variables, they increase our confidence in our results. In further robustness tests, we estimate models including XPolity, reported in Table A5 and graphed in Figure A2 in the Appendix. We still find an inverted U-shaped relationship between elections and organized mobilization events across the range of XPolity. Elections are only associated with significantly more frequent organized mobilization at moderate levels of democracy. However, the statistical significance of this result is not as strong as in models using Polity. We find no moderating effect of democracy on the frequency of unorganized events. Models including Polyarchy reported in Table A6 and Figure A3 in the Appendix generate very similar results to those in Table 2. The effect of elections on organized mobilization is significant at moderate levels of Polyarchy but insignificant at high levels. The effect is similar in magnitude to models using Polity. However, as our theory would predict, because Polyarchy does not capture the costs of mobilization through repression at low levels of democracy, we do not see the same decline in the effects of elections on organized mobilization at low levels of Polyarchy. As in all other models, Polyarchy does not moderate the effect of elections on unorganized mobilization. As a final robustness test, we also estimate models which take the binary Unrest 0/1 as dependent variable. These results are reported in Table A7 and graphed in Figure A4 in the Appendix. They are substantively similar to the count models discussed above.

Conclusion

Elections and electoral institutions have been proliferating in developing democracies and authoritarian regimes since the end of the Cold War. Concurrently, an ever-increasing proportion of the developing world's population lives in urban areas. It is therefore important to consider how elections affect the politics of contention in cities. In this article, we argue that patterns of urban mobilization and violence around elections vary across different types of groups. Our findings add nuance to previous studies and point to promising avenues of future research. Many previous arguments about mobilization at elections could be amended to account for collective action problems facing large groups. The actors most likely to follow strategic incentives to mobilize around voting are those organizations able to choose when to act and when to restrain their supporters. In future studies, greater attention should be paid to disaggregating the types of actors involved in electoral contention and violence. In particular, the organization and mobilization tools and strategies used by groups such as political parties and non-governmental organizations to maximize the payoff to collective action are a promising area for future research.

Beyond the direct association of elections with mobilization and violence, our findings also have implications for broader social and economic outcomes in urban areas. The prevalence of urban social disorder promises to increase as imperfect electoral and democratic institutions proliferate across the developing world. If political mobilization around elections spills over into crime – as unorganized mobilization is especially likely to – positive moves towards democracy may be tempered by concurrent threats to city dwellers’ personal safety. Similarly, if urban social disorder has deleterious effects on investment and employment, or leads to significant regime stability, it could hamper economic development in cities that are often characterized by poor public services, infrastructure and inadequate housing. Further research should explore the links between political institutions, urban social disorder, and broader social and development outcomes.

Acknowledgements

The authors gratefully acknowledge helpful feedback from Ben Goldsmith, Erica Chenoweth, Reed Wood and participants at the American Political Science Association, Peace Science Society International and Australian Society for Quantitative Political Science in 2017; and workshops at Arizona State University and the Australian Defense Force Academy.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the European Research Council (ERC) under Grant number 648291; the Norwegian Ministry of Foreign Affairs under Grant number #QZA 13/065; and Nuffield College, Oxford.

Notes on contributors

Henry Thomson

Henry Thomson, PhD (University of Minnesota, 2014) is an Assistant Professor in the School of Politics and Global Studies at Arizona State University. His research focuses on the political economy of authoritarian rule and democratization. His book, Food and Power: Regime Type, Agricultural Policy and Political Stability was published by Cambridge University Press in 2019.

Halvard Buhaug

Halvard Buhaug (PhD, Political Science) is Research Professor at the Peace Research Institute Oslo (PRIO), Norway, and Professor of Political Science at the Norwegian University of Science and Technology. His recent publications on security implications of climate change have appeared in, inter alia, Journal of Politics, Nature, PNAS, and World Development.

Henrik Urdal

Henrik Urdal (PhD, Political Science) is Director and Research Professor at the Peace Research Institute Oslo (PRIO), Norway. His recent articles on political demography have been published in Conflict Management and Peace Science, Demography, Review of Education Research, Political Geography and World Development.

Elisabeth Rosvold

Elisabeth L. Rosvold (PhD, Political Science) is a Postdoctoral Research Fellow in the Department of Economic History and International Relations at Stockholm University. Her recent research on linkages between disasters, aid and armed conflict has been published in PNAS, Scientific Data, and Journal of Development Studies.

Notes

1 Schedler, Electoral Authoritarianism; Levitsky and Way, Competitive Authoritarianism; Seeberg, “State Capacity”; Knutsen, Nygard and Wig, “Autocratic Elections.”

2 Bunce and Wolchik, “Defeating Dictators”; Donno, “Elections and Democratization.”

3 See, for example, Beaulieu, Electoral Protest and Democracy; Brancati, Democracy Protests; Chernykh, “When do Political Parties Protest”; Choi and Raleigh, “The Geography of Regime Support”; Conrad and Woo, “The Differential Effects”; Hafner-Burton, Hyde and Jablonski, “When do Governments Resort to Election Violence?”; Hafner-Burton, Hyde and Jablonski, “Surviving Elections”; Salehyan and Linebarger, “Elections and Social Conflict.”

4 Fjelde and Höglund, “Electoral Institutions.”

5 Harish and Little, “The Political Violence Cycle.”

6 Olson, The Logic of Collective Action, 105–8.

7 Kuran, “Sparks and Prairie Fires.”

8 Beaulieu, Electoral Protest and Democracy; Brancati, Democracy Protests; Hafner-Burton, Hyde and Jablonski, “Surviving Elections”; Fjelde and Höglund, “Electoral Institutions”; Choi and Raleigh, “The Geography of Regime Support.”

9 Daxecker, “All Quiet on Election Day?”; Harish and Little, “The Political Violence Cycle”; Salehyan and Linebarger, “Elections and Social Conflict.”

10 Fjelde and Höglund, “Electoral Institutions”, 303–4.

11 Daxecker, “All Quiet on Election Day?”

12 Harish and Little, “The Political Violence Cycle.”

13 Choi and Raleigh, “The Geography of Regime Support.”

14 Case, “Malaysia in 2012.”

15 Goh, The May Thirteenth Incident.

16 LeBas, “Violence and Urban Order.”

17 Tucker, “Enough!”; Salehyan and Linebarger, “Elections and Social Conflict.”

18 Harish and Little, “The Political Violence Cycle.”

19 Kuran, “Sparks and Prairie Fires.”

20 Wilkinson, Votes and Violence; McAdam and Tarrow, “Ballots and Barricades.”

21 Olson, The Logic of Collective Action; Kuran, “Sparks and Prairie Fires.”

22 Wilkinson, Votes and Violence.

23 Hafner-Burton, Hyde and Jablonski, “Surviving Elections”; Salehyan and Linebarger, “Elections and Social Conflict.”

24 Bunce and Wolchik, “Defeating Dictators.”

25 Conrad and Woo, “The Differential Effects.”

26 Lupu and Riedl, “Political Parties and Uncertainty,” 1346–7.

27 Tucker, “Enough!”

28 Chernykh, “When do Political Parties Protest.”

29 Przeworski, Democracy and the Market.

30 Orji, “Preventive Action.”

31 Wanyama and Elklit, “Electoral Violence”; Fjelde and Höglund, “Electoral Institutions.”

32 Urdal and Hoelscher, “Explaining Urban Social Disorder”; Bahgat, Buhaug and Urdal, “Urban Social Disorder”.

33 Some readers might be concerned that news reports in less-developed countries are superficial, leading to a bias against organized events being recorded in those areas – see, for example, Zhukov and Baum, “How Selective Reporting.” However, analysis of aggregate data shown in the Appendix shows only a very small negative correlation (r = −0.04) between GDP in 1960 and the number of organized events subsequently coded in a city, and an even smaller correlation (r = 0.02) between GDP and the ratio of organized over unorganized events.

34 A list of the cities included in the dataset is presented in Table A2 in the online Appendix.

35 We include government repression in our dependent variable because USD codes the highest level of escalation for each event, and therefore contentious mobilization that attracted a response by the state will be coded as government repression. However, our results also hold when we use a dependent variable which excludes government repression. Events which are not distinguished in the dataset as coordinated or uncoordinated are not included in our coding.

36 Marshall, Gurr and Jaggers, “Polity IV.”

37 Vreeland, “The Effect of Political Regime”, 406.

38 Ibid.

39 Coppedge et al., “V-Dem.”

40 Hyde and Marinov, “Which Elections Can Be Lost?”

41 Gleditsch, “Expanded Trade and GDP Data”; Gleditsch et al., “Armed Conflict”; Marshall, Gurr and Jaggers, “Polity IV”; Melander, Petterson and Themner, “Organized Violence”; United Nations, “World Urbanization Prospects.”

42 Salehyan and Linebarger, “Elections and Social Conflict”; Harish and Little, “The Political Violence Cycle.”

43 We only display results of models with three leads and lags here. Further leads and lags of the election variable are not statistically significant. Our main results are also robust to the inclusion of a lagged dependent variable.

44 One weakness of the SUREG models is that we cannot estimate a model with a functional form such as Poisson or negative binomial regression which is more suited to our count data. However, we also estimate identical systems of equations using bivariate Poisson and bivariate probit regression (results not reported) and these models generate substantively identical results. Multinomial logit models also generate very similar results.

45 Full model results for Tables 1 and 2 including all variables are presented in the online Appendix.

46 Further leads and lags of the Election indicator, when included in the models, are not statistically significant.