Electoral fraud and the paradox of political competition

ABSTRACT

Why are some elections more fraudulent than others? While much work has been devoted to understanding the structural conditions under which election quality can suffer, little is currently understood about election-specific dynamics that shape the conduct of polling day. This study assesses the impact of a more proximate determinant of election day fraud: the anticipated closeness of the race. In doing so, the paper sheds light on a potential paradox of political competition; highly competitive elections are seen as a healthy sign of democratic functioning, yet they may also lead to a reduction in the integrity of the process. Using novel pre-election polling data for 109 presidential elections around the world between 1996 and 2016, results suggest that ex ante closeness incites electoral fraud. In democratic contexts, closer elections – and elections in which the incumbent’s prospects are ambiguous – are associated with greater levels of ballot box manipulation as attempts are made to get over the finish line. This is the case largely irrespective of whether the incumbent is marginally ahead or behind in the race, suggesting that it is the mere uncertainty of the election result that can encourage election day fraud.

Introduction

The quality of elections has come under increased scrutiny in recent decades. As the most visible expression of democracy, polling day represents presents an opportunity for the lawful contestation of political power. Yet, elections are often contentious and regularly fall short of “free and fair” status (Bishop and Hoeffler 2016). This is the case not only for what one may call developing or democratizing states but also for more established democracies. Electoral competitiveness poses a problematic dilemma in this regard, as it is precisely when elections are expected to be close that political parties and candidates are incentivized to manipulate the process to tip the scales in their favor, especially in the immediate vicinity of polling day.

Previous research efforts have been able to identify many structural factors that are associated with reductions in the integrity of the electoral process. Electoral institutions such as oversight bodies and electoral rules, as well as the independence of the judiciary and media can ultimately be taken advantage of by willing perpetrators (Molina and Lehoucq 1999; Lehoucq and Kolev 2015; Birch and Van Ham 2017). Other societal factors such as economic inequality, ethno-linguistic polarization, and income levels also provide an electoral environment that obstructs democratization (Lehoucq 2003; Ziblatt 2009; Lehoucq and Kolev 2015; Bishop and Hoeffler 2016). While this field has done much to indicate which structural factors incite various conceptions of electoral manipulation in a broader sense, it can be aided in two primary ways. First, disaggregating electoral manipulation into its constituent elements allows for a more specific analysis of strategies that are employed differently, at different times, and under different circumstances (see, for example, Van Ham and Lindberg 2016; Asunka et al. 2017). This paper focuses on electoral fraud in the immediate vicinity of the election, where incentives to tamper with the process can be entirely different than in the early stages of the campaign, for example. Second and relatedly, the more nuanced conceptualization of electoral fraud taken here allows for the empirical assessment of more temporally proximate factors that may deviate from election to election, under the same socio-economic and institutional conditions.

To that end, this paper contends that the value of an individual vote – and subsequently the incentive to unlawfully capture this vote – is likely to vary significantly in winner-takes-all elections, and that this value is at its highest when polls predict a tight race. The pressure to deliver a positive result amid growing uncertainty on election day can incite political actors to manipulate the ballot box. However, this relationship is also likely to be contingent on the level of democracy. In autocracies, elections are rarely competitive and generally serve a largely different purpose than in democratic contests (Gandhi 2008; Simpser 2013). In democracies, on the other hand, the incentives to manipulate the ballot box become more acute when the transfer of executive power is genuinely at stake, and these incentives become more pronounced as the election approaches. These incentives are also likely to be further amplified when the costs of cheating are lower in less democratic states. For those in power, a possible resource imbalance could provide the means through which to tamper with the voting and tabulation processes depending on their electoral chances relative to the opposition.

This paper examines the degree to which the anticipated closeness of the election incites the manipulation of the ballot box, and how this relationship differs across levels of democracy. Additionally, this paper also addresses the possible asymmetry of this relationship, depending on whether the incumbent is leading or trailing in the race. This is achieved by conducting a quantitative analysis combining original polling data collected from 109 presidential electoral races around the world between 1996 and 2016 with a measure of electoral irregularities committed on the day of, or in the immediate aftermath of the election. Through this analysis, this paper finds that elections which are expected to be close incite more fraudulent activity. This trend appears to be strikingly similar across levels of democracy. Whether the incumbent is leading or trailing, on the other hand, only appears to have a marginal impact on incidences of electoral fraud. The primary contributions of this paper are twofold. First, the collection and utilization of pre-election polling data reduce the bias inherent in many previous efforts to estimate the effects of electoral competition. In doing so, this study is able to separate the state of competition ex ante from the election itself, thus enabling a cleaner analysis of an innately murky subject. Second, by testing the effect of the ex ante closeness across regime types, this paper is able to problematize the relationship between political competition and democracy to the extent that one need not necessarily be synonymous with the other.

Electoral fraud as a method of manipulation

Various conceptualizations of electoral quality have enriched the field in recent decades, although there exists a degree of overlap and confusion with regard to the specific forms of manipulative efforts. Beginning with Lehoucq’s flagship definition of electoral fraud as “clandestine and illegal efforts to shape election results” (Lehoucq 2003, 233), definitions of “electoral malpractice” (Birch 2011), “electoral integrity” (Norris 2015), and “free and fair” elections (Bishop and Hoeffler 2016), have broadened the debate to include non-intentionality and normative issues, as well as the characteristics and timing of the various forms.¹ As studied here, however, electoral fraud will be conceptualized more precisely, as a temporally and substantively distinct form of electoral manipulation.²

This temporal specificity is important to keep in mind in this context, as incentives to use one method over (or in addition to) another are likely to differ depending on the context of the election. The choice of whether to employ ballot box tampering over intimidation and violence has been found to be dependent on the location of observers and patterns of electoral competition, for example (Asunka et al. 2017). Resource availability has also been found to be a key factor in determining whether political actors engage in coercive manipulative strategies or vote-buying (Van Ham and Lindberg 2016). It is also plausible to suggest that the use of one manipulative tactic may preclude another. For instance, an election may be largely pre-determined through the use of large-scale “upstream” manipulation such as stifling opposition movements, media domination, and voter registry manipulation, thus rendering more acute uses of fraud profligate (Birch 2011). This paper, therefore, contends that a more nuanced conceptualization of electoral fraud provides added value in this context by attempting to isolate its incidence from other manipulative techniques. To that end, electoral fraud here shall refer to illicit attempts to manipulate the contents of the ballot box, either in terms of what is deposited into the ballot box (e.g. individual-level fraud or ballot box stuffing), or in terms of what is withdrawn from the box (e.g. misreporting of votes or result tampering). This conceptualization, therefore, speaks specifically to the event of the election itself and its immediate aftermath.

Nevertheless, there exist a plethora of studies that have considered the determinants of election quality in a broader sense, and it is important to consider them in relation to electoral fraud as well as other related forms of manipulation. Bishop and Hoeffler (2016), for example, found that “fair” elections are more likely to occur in states with higher levels of with foreign aid and observation, income, and executive constraints.³ Birch (2011) found political corruption and press freedom to be two key institutional determinants of the occurrence of electoral malpractice. Other studies of a similar mold have also associated clandestine attempts on the electoral process with high levels of social and economic inequality, ethnic polarization, the use of plurality electoral systems, the strength of oversight institutions, and low state capacity (Molina and Lehoucq 1999; Lehoucq 2003; Birch 2007; Ziblatt 2009; Kelley and Kolev 2010; Fortin-Rittberger 2014; Lehoucq and Kolev 2015; Birch and Van Ham 2017). That being said, these covariates are for the most part structural country level factors that do not tend to vary in short periods of time, and consequently, these explanations cannot generally account for the fluctuation in levels of election quality within states from election to election.

There is considerably less well-grounded consensus in the field with regard to the effects of the electoral competition, however. This is understandably the case when we consider that studying the dynamics of political competition is inherently difficult from a purely methodological perspective in the realms of the manipulation of votes and voters. Essentially, most estimations of political competition where fraud is suspected are likely to bear the fingerprints of manipulative efforts, thus distorting typical measures of competition such as election results. Indeed, previous studies have produced findings to suggest that competition can both encourage (e.g. Nyblade and Reed 2008; Ruiz-Rufino 2018) and deter (Lehoucq and Molina 2002) manipulative efforts. In the following section, these apparent competing expectations will be reconciled by speaking of competition in terms of the anticipated closeness of the race, and necessary nuance will be added to the role of regime type. A theoretical argument will then outline the expectations of the study.

Electoral competition and the value of votes

In democracies, there is a significant risk attached to the thought of trying to unlawfully influence elections. In such contexts, Simpser’s “conventional wisdom” (2013, 63) that greater competition leads to more electoral manipulation remains just that; evidence is for the most part anecdotal.⁴ This is puzzling in that the public exposure of intentional wrongdoing during electoral campaigns can negatively impact a candidate or party’s chances of victory. Nevertheless, electoral fraud and other forms of manipulation continue to plague many elections in what are commonly considered to be democratic states.⁵ But if being called out for cheating undermines the legitimacy of results, then why are elections manipulated in democracies? One answer relates to the form and timing of the manipulative strategy. Many forms of manipulation do not become evident to the general public until after the election has taken place. Election day irregularities such as tabulation discrepancies, ballot stuffing, fraudulent voting, and polling station offences rarely interrupt the declaration of results, and allegations pertaining to such events are frequently anecdotal. Without evidence, it is more difficult to hold suspected officials to account through the judicial system than the electorate. Perpetrators, therefore, may well assess the risk as one worth taking.

A second answer relates to the institutional context. As the rules of the game are set, and the degree and nature of political competition vary, so too do incentives to manipulate the process. Patterns of electoral competition can differ substantially across electoral rules, for example. In zero-sum contests such as plurality rules in single-member districts in legislative elections, or direct presidential elections, for instance, there is no consolation for attaining even one vote fewer than one’s rival. Parties and candidates are usually well aware of the degree of political competitiveness well in advance of an election and the campaigning process. Pre-election polls are frequently conveyed through media outlets and are often used by political parties and candidates for campaign strategies, and have been shown to affect the voting behavior of citizens (Bursztyn et al. 2017). The argument put forth here suggests that in zero-sum elections, actors are incentivized to engage in manipulative techniques to attain victory when they are unsure of the electoral outcome ex ante. Yet, this line of inquiry throws up a potential paradox of political competition. Highly competitive elections in which more than one party has a genuine chance of victory usually signify a fervent and healthy democracy in action. Therefore, it is when elections are at their most competitive – i.e. when democracy is supposedly at its healthiest – the integrity of the electoral process is likely to suffer.

Previous efforts to elucidate the effects of competition have produced mixed findings. In an important contribution, Nyblade and Reed (2008) distinguish between the adverse effects of political competition on “looting and cheating”. Whereas electoral accountability in competitive polities can dissuade politicians from engaging in the former because of the fear of electoral repercussions when exposed, the incentives to cheat (and win office) may, they argue, outweigh these same fears. A similar argument is made by Ruiz-Rufino (2018), who argues that the patterns of competition in disproportional electoral systems can encourage incumbents to commit electoral fraud when they fear their position is under threat. Conversely, however, Lehoucq and Molina (2002) argue that high levels of competition can dampen electoral fraud by encouraging credible electoral opponents and civil society actors to be more vigilant and call manipulative efforts out. Indeed, electoral manipulation has also been shown to be more prevalent precisely when elections are at their most uncompetitive (Simpser 2013).

While there appears to be some degree of conflict in these findings, it is of pivotal importance to maintain a view of the regime types to which these expectations pertain. The conduct of a single election does not necessarily define the democratic condition in a given state, and there are several other institutional and societal aspects that determine such a description. The effect of political competition is therefore expected to be fundamentally contingent upon the level of democracy. In non-democratic settings, genuine political competition is systematically stifled from the outset (Ruiz-Rufino 2018). Elections are rarely close, and often serve a purpose entirely alien to that in institutionalized democracies (Gandhi 2008; Gandhi and Lust-Okar 2009; Blaydes 2010). It, therefore, seems illogical to presume that the relationship between political competition and electoral fraud would manifest itself in the same way across regime types. Electoral manipulation is ubiquitous in authoritarian contexts, and often prevents elections from becoming genuinely competitive in the first place (Levitsky and Way 2002).

In democratic contexts, however, an individual vote is deemed more valuable to a candidate or party if it can conceivably play a decisive role in the outcome of an election. In winner-takes-all contests where the anticipated victor is not clear, vote value increases to the greatest extent ex ante, and actors have a greater incentive to capture existing ballots, generate new ballots, or deter potential votes for other candidates. Electoral fraud in parliamentary elections has been demonstrated to be more prevalent under majoritarian rules, for example (Lehoucq and Kolev 2015; Ruiz-Rufino 2018). The indivisible nature of the prize at stake in winner-takes-all contests accentuates the desire to emerge successful from the ballot due to the lack of consolation for losers.⁶ In close contests, and when uncertainty dominates the discourse of election day, tensions regarding the result are heightened to the largest degree. On the ground and at the polling and counting stations, party and candidate activists may feel compelled to break the rules to ensure a positive result. The career prospects that may come with a victorious outcome – or perhaps even the fear of the consequences of a defeat – may compel political actors on the ground to break the rules to push the result over the line. Political actors are therefore expected to commit fraudulent acts on election day to sway the result in their party’s favor out of desperation and/or consolidation, both of which can be seen as a product of uncertainty when alternative options to capture votes (legal or otherwise) are restricted.

$H_1$ Elections expected to be close ex ante will be associated with more electoral fraud.

While this framework is not expected to apply to autocratic states in which elections serve a fundamentally different purpose, “intermediate” democracies provide a context where elections can be and often are highly competitive, while the costs associated with committing electoral fraud may be substantially lower than in more established democracies. Political competition in intermediate democracies can be fierce, and in many instances, these states may lack the strong, clean, and impartial institutional foundations required to withstand such pressure (Diamond 2002; Hyde and Marinov 2012). As these institutional obstacles are removed from the equation (or are more easily circumvented), electoral fraud may become a more feasible course of action in the face of electoral uncertainty on election day vis-á-vis established democracies.

$H_2$ The effect of competitiveness on electoral fraud will increase in less democratic states.

That being said, the relationship between ex ante closeness and electoral fraud is not strictly speaking expected to be symmetrical for the incumbent and opposition. In many states, incumbents benefit from resource imbalances between the respective contestants of an election. The incumbent party or candidate, therefore, may have a greater capacity to commit electoral fraud, as well as additional incentives if their position in office is at stake. Depending on their relative position in the electoral race, incumbents may be more inclined to manipulate the ballot box out of desperation (when behind) or consolidation (when ahead). This can be formalized in the following competing hypotheses:

$H_{3a}$ The effect of competitiveness on electoral fraud will increase when the incumbent is ahead in the race.

$H_{3b}$ The effect of competitiveness on electoral fraud will increase when the incumbent is behind in the race.

Empirical strategy

The hypotheses will be assessed with an analysis of a global sample of presidential elections in democracies between 1996 and 2016.⁷ Presidential elections are chosen as the theoretical expectations established in the above hypotheses expect ex ante closeness to affect levels of electoral fraud to the greatest extent in winner-takes-all contests. Similar trends have been established in majoritarian system parliamentary elections (e.g. Ruiz-Rufino 2018) but presidential elections – where individual candidates are pitted against one another on a national scale – remain unaddressed. Presidential elections are zero-sum games, in which there is only one winner and no consolation for second place (Linz 1990). In such contexts, the incentives referred to in the theoretical framework are expected to be more pronounced than in parliamentary elections, for example, where legislative seat allocation systems – and therefore electoral stakes – can vary considerably. To estimate the relationship between political competition and electoral fraud, the analysis will employ a series of ordinary least squares models with standard errors clustered by country.⁸ The estimation strategy can be summarized by the following Equation 1 where $i$ indexes elections and $X$ denotes the vector of control variables to be included in the analysis:$\begin{array}{rl}& electoralfrau{d}_i=\alpha +{\text{β}}_1(electoralcompetitio{n}_i)+{\text{β}}_2(democracyleve{l}_i)\\ & +{\text{β}}_3(incumbentlea{d}_i)+{\text{β}}_4(electoralcompetition\times democracyleve{l}_i)\\ & +{\text{β}}_5(electoralcompetition\times incumbentlea{d}_i)+X_i+{\epsilon }_i.\end{array}$

Data

Electoral fraud

Electoral fraud is measured using the “election other voting irregularities” variable from the Varieties of Democracy (V-Dem) dataset based on national and regional expert assessments (Coppedge et al. 2018). This measure covers the “use of double IDs, [the] intentional lack of voting materials, ballot stuffing, misreporting of votes, and false collation of votes” (Coppedge et al. 2018, 55).⁹ In addition to bearing similarity to the definition of electoral fraud provided above, this operationalization is preferred as any form of manipulation that takes place prior to the election is likely to become empirically tangled with the public’s perception of the electoral process. The temporally confined nature of this variable, therefore, allows the analysis to separate events that occur in the run up to the election from the event of the election itself. The variable is originally constructed using an ordinal scale by country and regional experts in which possible answers range from 0 (widespread irregularities) to 4 (no irregularities). As employed here, the variable is transformed to an interval scale which ranges from 0 to 6 where higher values indicate higher levels of irregularities.

Polling data

To estimate electoral competition, novel pre-election polling data was collected for the purposes of this study. The inclusion of polls is based on two factors: (a) that they were conducted up to 12 months prior to the election and (b) the source of the poll is deemed credible.¹⁰ The period of one year – with the exception of the month prior to the election – is deemed a suitable period to accurately gauge the level of competitiveness of an upcoming election, whilst ensuring that there is no temporal cross-over between ex ante closeness and the fraudulent phenomena measured by the outcome variable. Polls are preferred to the use of the margin of victory (e.g. Ziblatt 2009) or the previous election results (e.g. Ruiz-Rufino 2018), for example, as such measurements may well be the product of the fraudulent activities that they will be measured against. The use of polling data provides a much more accurate reflection of the information that is available to candidates, parties and voters in the run up to elections. The theoretical foundations of this paper suggest that the driving force behind electoral fraud in close contests is the perception of how close the race appears to political actors.

Electoral competition will be operationalized in two ways. First, the base measure of Ex Ante Closeness takes the absolute percentage point difference between the top two candidates, and subtracts this value from 100. For example, if Candidate A has 35% of preferences and Candidate B has 30%, the marginal difference (5) is subtracted from 100 to create a value of 95. This variable, therefore, indicates the baseline closeness of the race, with higher values indicating tighter elections. To investigate $H_3$, a dummy indicator – Incumbent Lead – is also included which signifies whether the incumbent candidate and/or party is leading in these polls (1 if the incumbent is leading, and 0 otherwise). The second estimation of political competition – Incumbent Margin – integrates both of these elements by taking the polling data and integrating the position of the incumbent relative to the next best candidate. For example, a poll lead of 5 points for the incumbent would score a value of 5, whereas an incumbent deficit of 5 points would score a value of −5.

Democracy

In order to distinguish states according to their level of democracy, Freedom House’s Freedom Rating status will be used, which categorizes states in a given year as either “Free”, “Partly Free”, or “Not Free”. Although democracy categorizations are in some cases somewhat arbitrary, the tripartite categorization of Freedom House is deemed sufficient in this case as $H_2$ predicts an explicitly increased effect in less vis-á-vis more democratic states. Following the exclusion of the “Not Free” cases, the categories of “Partly Free” and “Free” are deemed sufficient to proxy as a distinction between less and more democratic states. Making broad distinctions between what is considered “democratic” and “semi-democratic” is notoriously difficult and controversial (Bogaards 2012). However, it is adequate in this study to simply compare full democracies to intermediate states, where competition is still a defining feature of elections.¹¹

Control variables

Control variables are included primarily on the basis that they may be driving the relationship between electoral competition and electoral fraud, as well as their propensity to deviate from election to election. First, the degree of power exercised by the president – that is, the stakes of the election – is also included in the analysis, to cater for the possibility that electoral competition and fraudulent efforts are more fierce in contests for more powerful positions.¹² Secondly, the presence of international and domestic election observers has been shown to limit the extent of electoral fraud (e.g. Hyde 2011), but may also be sanctioned to elections that are expected to be tight. The partisanship of the media is also likely related to the reporting of pre-election polls, as well as the broadcasting and reporting of suspected irregularities during the electoral process. Each of these variables is taken from the V-Dem dataset (Coppedge et al. 2018). The analysis will also include controls for political corruption, economic development, and region, as well as a lagged dependent variable which accounts for the level of irregularities in the previous presidential election.¹³

Analysis

Figure 1 portrays a scatter plot of the relationship between the anticipated closeness of elections and voting day irregularities, where higher values on the y-axis indicate greater levels of voting irregularities, and higher values on the x-axis indicate closer elections. Whilst only a simple scatter, it is evident to see the initial relationship between electoral competition and fraud in democracies. The graph demonstrates that in free and partly free states there is a positive relationship between the two; as ex ante closeness increases (i.e. when the margin between the leading two candidates decreases), elections appear to be more fraudulent. Initial indications also suggest that the relationship is linear in both instances, and perhaps surprisingly, the magnitude of the association is almost identical.

Figure 1. Scatterplot of ex ante closeness and voting irregularities across levels of democracy. The x-axis ranges from 50 to 100, which is calculated as 100 minus the percentage point margin between the leading two candidates in the pre-election poll. For example, if Candidate A was polled on 35% and Candidate B was polled on 30% (5-point lead), the value given to this race would be 95 (100 − 5).

Unsurprisingly, on the whole, most democratic nations are associated with fewer election irregularities than those in the partly free category. Interestingly, however, this graphic indicates that within the democratic subgroups, the elections with less fraud tend not to be close ex ante. In the party free category, for example, the election with the fewest reported election day irregularities has a marginal lead of 44.6 percentage points (Ecuador, 2013), whereas in the most fraudulent election the margin was just 3 points (Kenya, 2007). The free states present a similar trend: the pre-election poll margin for the leader in the best-performing election in the sample was 16.5 percentage points (Chile, 2009), compared to 3 points in the worst (Philippines, 2004).¹⁴

The results of an ordinary least squares analysis are presented in Table 1. Model 1 presents a baseline model in which only the democracy level is accounted for. While we can see confirmation that an increase in democracy level decreases the extent of electoral irregularities, this model also suggests that when the democratic context is controlled for, ex ante closer elections are associated with more electoral fraud. As additional controls are introduced in Model 2, the magnitude of the effect of closeness decreases, but remains statistically significant (p < .01). As a direct test of H1, we can, therefore, conclude that ceteris paribus, closer elections are indeed associated with greater extents of electoral fraud. To assess H2, Model 3 introduces an interaction term between closeness and democracy level. Although the beta value $\left(electoralcompetition\times democracyleve{l}_i\right)$ is negative – suggesting a suppression of the effect when FH Level equals “1” (free states) – it fails to attain statistical significance, suggesting that the effect of ex ante closeness on electoral fraud is not significantly different across levels of democracy, thus emulating the trend lines reported in Figure 1. This analysis is therefore unable to reject the null in the case of H2, as there appears to be no statistically significant difference between the effect of electoral competition in free and partly free states.

Table 1. Ordinary least squares regression analyses of election day irregularities.

	DV: Election day irregularities
	(1)	(2)	(3)	(4)	(5)	(6)	(7)
Ex ante closeness	0.042	.012	.017	.015	.009
	(.007)***	(.004)***	(.008)**	(.004)***	(.007)
Incumbent						.041	.052
						(.012)***	(.018)***
Incumbent margin^2						−.0004	−.0005
						(.0001)***	(.0002)***
Incumbent lead (0,1)				.162	−.626
				(.094)*	(.852)
Interactions
Closeness*FH level			−.009
			(.009)
Closeness*FH PR
Inc Margin*FH level							−.021
							(−022)
Inc Margin^2*FH level							.0002
							(.0002)
Closeness*Inc lead					.009
					(.01)
Controls
FH level	−1.374	−.294	.542	−.317	−.315	−.298	.271
	(.158)***	(.107)***	(.843)	(.098)***	(.096)***	(.107)***	(.627)
Voting irregularities(t − 1)		.42	.423	.419	.413	.421	.428
		(.134)***	(.137)***	(.133)***	(.134)***	(.136)***	(.141)***
GDP(log)		.014	.019	.019	.022	.017	.023
		(.082)	(.082)	(.074)	(.076)	(.077)	(.074)
Presidential power		−.01	−.016	−.006	−.004	−.002	−.008
		(.055)	(.058)	(.052)	(.051)	(.052)	(.055)
Domestic observers		−.092	−.077	−.156	−.151	−.107	−.101
		(.254)	(.253)	(.247)	(.238)	(.252)	(.243)
International observers		.076	.1	.092	.096	.108	.123
		(.095)	(.096)	(.095)	(.092)	(.097)	(.099)
Media bias		.152	.156	.138	.135	.141	.138
		(.058)**	(.06)**	(.061)**	(.063)**	(.062)**	(.063)**
Corruption		1.018	1.01	.962	.974	1.026	1.026
		(.382)**	(.387)**	(.372)**	(.378)***	(.379)***	(.378)***
Region dummies	N	Y	Y	Y	Y	Y	Y
Constant	−.412	−.149	−.423	−.371	.307	.103	−.292
	(.659)	(1.159)	(1.185)	(1.117)	(1.206)	(1.034)	(1.044)
R^2	.569	.868	.87	.873	.875	.872	.873
N
(elections)	109	109	109	109	109	109	109
(countries)	50	50	50	50	50	50	50

Notes: Robust standard errors, clustered by country in parentheses. Positive coefficients indicate more voting irregularities.

*p < .1, **p < .05, ***p < .01.

Turning to the coefficients of the control variables themselves, the only three variables to achieve consistent statistical significance across Models 1–3 (besides the democracy indicator) are the lagged dependent variable, media bias, and corruption. The results also appear, perhaps surprisingly, to show an inconsistent effect of domestic and international election observers given past findings. There two plausible explanations for these results. First, the inclusion of a lagged dependent variable tends to bias coefficients toward zero, particularly when those variables are likely to have similar values in the previous election. Secondly, international organizations are more likely to send observers to elections which are anticipated to be problematic in democratizing states, for example (Birch 2011; Kelley 2012).

Models 4–5 test the potentially asymmetrical nature of the relationship predicted in H3 by introducing a dummy variable indicating whether the incumbent candidate or party was leading in the polls. Evidence in support of either hypothesis here is limited, and although Model 4 provides some evidence that ceteris paribus electoral fraud is more prevalent when the incumbent is leading (p < .1), this effect appears to be independent of electoral competition. Once the interaction term between ex ante closeness and Incumbent Lead is introduced in Model 5, it becomes evident that there is not a statistically significant difference between the effect of electoral competition when the incumbent is leading as opposed to trailing. This may not necessarily discount either hypothesis entirely, however, as it is indeed possible that this mechanism is at play in both directions: consolidation and desperation. It is possible that incumbents are incentivized to manipulate the ballot box in both instances, but there is no significant difference between the two mechanisms. Nevertheless, this analysis is unable to confidently assert that the effect of ex ante closeness on fraud becomes stronger when the incumbent is leading or trailing. The marginal effects of Model 5 presented in Figure 2 illustrate this finding succinctly.

Figure 2. Illustration of Model 5 from Table 1. Marginal effects of ex ante closeness when the incumbent is leading versus trailing on the extent of voting irregularities. The x-axis ranges from 50 to 100, and can be interpreted as the inverse of the absolute poll margin (100 – margin). For example, a score of 95 would indicate a 5-point margin. Confidence intervals are illustrated with dashed lines.

Finally, Models 6–7 utilize an alternative operationalization of electoral competition that incorporates both the extent of the lead as well as the position of the incumbent in the race. Incumbent Margin is calculated by centering the marginal lead in the polls on the incumbent candidate or party, thereby taking negative values when the incumbent is trailing and positive values when leading.¹⁵ Models 6–7 include this variable along with a squared term to test for the potential non-linearity of the relationship. Whereas Models 2 and 5 can be seen as a direct test of H1 and H3 respectively, Model 6 tests both of these hypotheses simultaneously. The hypothesized relationship expects the standard term to be positive, and the squared term to be negative, resembling an inverse “U” shape. Model 7 then introduces H2 to the equation, by interacting both forms of the Incumbent Margin variable with the democracy level. The nature of the relationship becomes much clearer when the marginal effects of Model 7 are plotted in Figure 3. Elections tend to be more fraudulent when the incumbent is marginally ahead or behind in the race. This seems to be the case, albeit to slightly different extents, in both free and party free states. When the result is not certain, and the incumbent margin is in the range of 0 (dead heat), voting irregularities increase.

Figure 3. Illustration of Model 7 from Table 1. Marginal effects of the incumbent’s margin on the extent of voting irregularities across levels of democracy. Negative values on the x-axis indicate that the incumbent is trailing in the polls, whereas positive values indicate a lead. Confidence intervals are illustrated with dashed lines.

Robustness tests

Given the potential variability of predictive polls in the lead up to elections, robustness tests were undertaken first including the proximity of the polls to the election as a control variable, and then limiting the time-span within which the polls were conducted to 180, 90, and 45 days respectively. Replications of Model 3 and 6 remain largely consistent across specifications in terms of sign and magnitude albeit with increased significance levels for the interactions in Model 3 – suggesting that there may well be support for H2 when poll accuracy is accounted for.¹⁶ Appendix 6 details the full output of these model specifications. Appendix  7 also details a replication of Models 2–3 and 5–6 using a continuous measure of democracy: Freedom House’s Political Rights indicator. Once again, the results of the analysis prove robust with the only notable deviation from the main results being the replication of Model 3, in which the standard error of ex ante closeness rises substantially. These models were replicated once more in Appendix 8 with the addition of year dummies variables to cater for external shocks that may have been driving the results for observations across different countries in the same year. Despite a partial rise in the standard errors of some coefficients, the results broadly hold across each of these tests.

Concluding remarks

Using novel polling data for a global sample of the presidential election between 1996 and 2016, this article has demonstrated that in winner-takes-all elections, increased levels of competitiveness incite greater attempts to manipulate the ballot box. Interestingly, the nature and magnitude of this relationship is similar in “partly free” states as well as more established democracies, producing a potentially troubling implication for competitive democracies; highly competitive elections – often identified as one of the hallmarks of a healthy democracy – can often be detrimental to the integrity of the process. This is seemingly the case whether the incumbent is trailing or leading the race, suggesting that the mere uncertainty is enough to encourage political actors to engage in fraudulent attempts to sway the results in their favor. The hypothesized mechanisms of desperation and consolidation may indeed be both at play here but this analysis has been unable to uncover the extent to which each (or neither) apply.

That being said, there are several other aspects of this phenomenon that cannot be wholly investigated by the approach taken in this study. Regrettably, the form of the data prohibits the investigation of presidential run-offs, for example, as each election is awarded an aggregate rating for all election-rounds taking place within a given year. This is, therefore, one potentially fruitful avenue for future research, given that the theoretical expectations laid out here would anticipate an even more pronounced effect when tensions are heightened, time is short, and candidates are few. Further, the theory laid out here also does not necessarily preclude its application to other forms of elections. Other winner-takes-all contests such as parliamentary elections under majoritarian rules or mayoral elections on the local level, for instance, may also show similar trends.

By focusing on irregularities reported on polling day, this study is able to make two primary contributions. First, by employing novel pre-election polling data in conjunction with a temporally confined measure of electoral fraud, this analysis is able to effectively isolate the patterns of electoral competition taking place prior to election day from the conduct of the election itself. The use of pre-election polls removes problematic election results from the equation, focusing rather on the perceptions surrounding the election that ultimately shape the behaviors of the relevant actors. Important to note here, however, is that elections deemed “clean” in this study may not necessarily be so. It may also be true that electoral competition, as measured in this study, is itself affected by manipulative efforts taking place earlier in the electoral cycle. One strength of the approach taken here, however, is that it is possible to estimate the relationship between electoral competition and electoral fraud regardless of what manipulation may or may not have occurred before. Secondly and relatedly, this study contributes a more nuanced understanding of the circumstances under which perpetrators may attempt to manipulate elections, and why they would resort to election day fraud in particular. Patterns of electoral competition, it seems, can have similarly adverse consequences for the integrity of election day in semi-democratic and democratic states. Greater levels of uncertainty combined with decreasing time with which to reduce this uncertainty can incentivize actors to manipulate the ballot box to sway the vote in their favor. Competitive elections may therefore not be the democratic hallmark many presume them to be.

Disclosure statement

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

Additional information

Notes on contributors

Stephen Dawson

Stephen Dawson is a PhD Candidate at the Department of Political Science, University of Gothenburg. He is affiliated with the Quality of Government Institute.

Notes

1 Here it is important to note the difference between the manipulation of elections and political corruption. Electoral misconduct is distinct in this sense as public interest is substituted for “personal or partisan interests” (Birch 2011, emphasis added). The prize – and hence the incentive – for the manipulation of competitive elections is the acquisition (or conservation) of office itself.

2 For an overview of the various manipulative techniques used to undermine elections, see Schedler (2002, 2013).

3 “Fairness” here refers to polling day itself whereas “freeness” refers to the process leading up to the election (Bishop and Hoeffler 2016). This distinction bears similarity to that of Birch (2011), who differentiates between “upstream” and “downstream” malpractices.

4 In one notable exception, Ruiz-Rufino (2018) shows that patterns of electoral history can incentivize the obstruction of political competition in disproportional electoral systems.

5 Examples include Argentina (Gonzalez-Ocantos et al. 2012), Canada (Christensen and Colvin 2007), Japan (Christensen and Colvin 2007), the UK (Scott 1972), and the USA (Cox and Kousser 1981; Bensel 2004; Campbell 2005; Christensen and Colvin 2007).

6 That being said, the incentives to manipulate the process are not expected to be the same for positions that grant little or no meaningful power, such as ceremonial presidential positions.

7 As elaborated below, “democracies” here will refer to states which are categorized as either “free” or “partly free” by Freedom House. Autocracies are excluded from the analysis.

8 There are several instances in which there only one observation per country, thus discounting other approaches.

9 Descriptive statistics and descriptions for this and all other variables included in the analysis can be found in Appendix 1.

10 For a full list of the polls included in the study, as well as their conduction dates and sources, see Appendix 9.

11 To eliminate the possibility that a cut-off point may be driving the results of the analysis, models are replicated in Appendix 7 using a continuous measure of democracy (Freedom House’s Political Rights score). The results are largely consistent across these alternative operationalizations.

12 Elections for ceremonial presidential positions (e.g. Ireland or Austria) have been excluded from the analysis.

13 The within-country average standard deviation of this variable is 0.4 whereas the overall standard deviation is 1.

14 The descriptive statistics of pre-election poll margins are broken down by democracy level in Appendix 3.

15 As the sign of the value is lost when squaring a variable that takes on both positive and negative values, the Incumbent Margin variable was modified for the analysis (50 + Incumbent Margin) to remove negative values. While this does not affect the model or beta values, it does render the constant in Models 6–7 meaningless. Figure 3 corrects for this and transforms the variable back into its original form.

16 Predictably, these additional analyses reduce the N significantly, so these results should be read cautiously.