OUP user menu

When the Great Power Gets a Vote: The Effects of Great Power Electoral Interventions on Election Results

Dov H. Levin
DOI: http://dx.doi.org/10.1093/isq/sqv016 189-202 First published online: 17 February 2016


What are the electoral consequences of attempts by great powers to intervene in a partisan manner in another country’s elections? Great powers frequently deploy partisan electoral interventions as a major foreign policy tool. For example, the U.S. and the USSR/Russia have intervened in one of every nine competitive national level executive elections between 1946 and 2000. However, scant scholarly research has been conducted about their effects on the election results in the target. I argue that such interventions usually significantly increase the electoral chances of the aided candidate and that overt interventions are more effective than covert interventions. I then test these hypotheses utilizing a new, original dataset of all U.S. and USSR/Russian partisan electoral interventions between 1946 and 2000. I find strong support for both arguments.


What are the effects of great power electoral interventions? In democracies, national-level elections are pivotal events, enabling a peaceful change in the makeup of the main decision makers and their domestic coalitions. They often lead to major shifts in a country’s domestic and foreign policies and affect its propensity to experience both domestic and international conflict.1 Even in electoral authoritarian regimes, relatively competitive elections can have significant consequences on domestic and international politics. In some of these cases, competitive elections even lead to the fall of the existing leadership, the rise of a new regime, or a full-scale transition to democracy.

Given these stakes, foreign actors face strong incentives to intervene in competitive elections, which now take place in more than half of all states (Freedom House 2012). Indeed, attempts by a great power to meddle in an election of another country in favor of a particular candidate or a specific party may shape electoral outcomes. Between 1946 and 2000, the United States and the USSR/Russia intervened in this manner 117 times, or, put another way, in about one of every nine competitive national-level executive elections during this period. Their methods ranged from providing funding for their preferred side’s campaign (a tactic employed by the Soviet Union in the 1958 Venezuelan elections [Rabe 1982, 136–37]) to public threats to cut off foreign aid in the event of victory by the disfavored side (as the United States did during the 2009 Lebanese elections [Ghattas 2009, 1]).

Observers often claim that partisan interventions, when known or subsequently exposed, make the difference in election outcomes. For example, in the 2000 Yugoslavian elections, one of the main figures in the successful campaign of the democratic opposition headed by Vojislav Kostunica against Slobodan Milosevic admitted in an interview shortly afterwards that “The foreign support [to the campaign] was critical” to its electoral success (Dobbs 2000, 1). Fifty-two years beforehand, and less than three days after the conclusion of the overt US intervention against the Communist Party in the 1948 Italian elections, Palmiro Togliatti, the then-head of the Italian Communist Party, openly blamed the surprising defeat of his party on what he described as the “brutal foreign intervention” of the United States (Togliatti Accuses U.S. 1948).

Nevertheless, scholars pay very little attention to the effects of partisan electoral interventions. This article provides the first cross-national statistical analysis of these effects. I demonstrate that electoral interventions systematically increase the electoral chances of the aided candidate. I also show that overt interventions are usually more effective than covert interventions.

I begin by briefly reviewing the existing literature on this topic. In the second section, I describe in detail the two hypotheses I propose as to the electoral effects of great power electoral interventions. Then, I explain the method by which I operationalize and evaluate these hypotheses. In the fourth section, I test these hypotheses using a new dataset of US and Soviet/Russian interventions and describe my results. Finally, I conclude by expanding upon the scholarly and practical implications of my findings.

Research on Electoral Interventions

In spite of the ubiquity and possible importance of electoral interventions, they receive very little attention from political scientists. This stands in contrast to their extensive investigation into the effects and effectiveness of other types of interventions and foreign policy tools. Some scholars, for example, conduct significant quantitative research on the effects of external military interventions for the purpose of regime change or democratization (Meernik 1996; Hermann and Kegley 1998; Peceny 1999; B. Bueno de Mesquita and Downs 2006; Lo, Hashimoto, and Reiter 2008; Willard 2012; Downes and Monten 2013). Others focus on economic sanctions (Drury 1998; Drezner 1999; Hart 2000; Letzkian and Spercher 2007; Peksen and Drury 2009; Bapat and Kwon 2015) or on interventions in civil wars (Balch-Lindsay and Enterline 2000; Regan 2002; Gent 2008; Schultz 2010).

We usually find interest in the effects of electoral interventions among scholars in two very different subfields. On the one hand, diplomatic historians note such interventions as part of larger studies on a particular era or bilateral relations between states. On the other hand, scholars in intelligence studies discuss cases of such interventions as part of broader qualitative analysis of the effectiveness of various activities conducted by the Central Intelligence Agency (CIA) and other intelligence agencies. Among these scholars, significant controversy exists as to the short-term effects, if any, of such interventions.

For example, quite a few historians studying particular cases involving electoral interventions either largely dismiss their effects on the results of the relevant elections or view them as counterproductive—as harming the side they were trying to help (DeConde 1958, chp 13–14; Miller 1983, 52–53; Gustafson 2007, 49, 73–74). In contrast, other scholars, usually from intelligence studies, see electoral interventions as both effective and decisive in electoral outcomes (Daugherty 2004, 4–7; Prados 2006, 627; Haslam 2005, 13–15).2

Two recent quantitative studies provide the exception to the overall qualitative approach to the study of electoral interventions by foreign actors.3 Corstange and Marinov (2012, 659, 664–69) conducted a field experiment, the first of its kind, on the effects of an overt foreign intervention on the views of the targeted voters toward the intervener. Conducted in Lebanon two months after the 2009 parliamentary election, it found that an overt intervention in favor of one of the sides contesting the election—in other words, the intervener explicitly declaring its support—polarizes the electorate. Those who support the side favored by the intervener view the intervening power in a more favorable light and vice versa.

Shulman and Bloom (2012, 460–64) conducted a conventional public opinion survey on domestic reactions to electoral interventions in the Ukraine fourteen months after the 2004 presidential election in which such an intervention occurred. They found that the public sees such interventions as universally improper, with US or Western interventions perceived as more improper than Russian interventions.

Given their single-country nature, neither study offers conclusive evidence for the perceptions or effects of electoral intervention beyond the specific context that they studied. Thus, a cross-national large-N study that includes numerous cases of electoral interventions in different countries and periods may prove especially useful for analyzing the direct effects of such interventions.


I argue that electoral interventions usually occur when two concurrent conditions exist. One involves motive, the other opportunity. First, a great power must perceive its interests as being endangered by a certain candidate or party within a democratic target. That candidate or party has inflexible preferences on important issues that diverge from that of the great power. These inflexible preferences are due to that candidate or party being either greatly constrained by its political base on these issues and/or ideologically committed to particular positions. That, in turn, makes many of the conventional policy responses (various forms of carrots and sticks aimed at resolving disagreements) appear potentially ineffective or too costly to the great power. Second, a significant domestic actor must consent to, and willingly cooperative with, a proposed electoral intervention by the great power.4 Without the domestic actor’s cooperation in providing information (or “local knowledge”) about the electorate’s preferences and the best ways to intervene in its favor, the great power will usually see its chances of succeeding as too low to justify an electoral intervention.

In the absence of either one of these conditions, the great power will not intervene in the elections. In fact, the great power will “sit out” an election, even in the face of an intervention by an unfriendly great power and even if it sees such a situation as highly threatening to its interests, unless it can find a significant domestic actor willing to accept its assistance.5

Partisan electoral interventions by the great powers are not the only factor that can affect the results of a particular election, but they nevertheless can significantly increase the electoral chances of the supported candidate or party. This is the result of the process by which a would-be intervener and a would-be client “choose” each other and agree to an electoral intervention.

Thus, a great power will not likely support a potential client if that client will still likely lose the election. Under these circumstances, a great power will usually judge that other means, such as post-electoral efforts to influence (or violently remove) the regime, will better serve its interests.6 Similarly, a potential client will likely reject an offer of electoral aid by an outside power if she believes that she will win the election in the absence of such assistance. In these circumstances, the greater risk comes from the possible medium- and long-term costs involved in receiving such an intervention in her favor.

These costs include harming the client’s electoral position in the longer term by alienating voters who, for a variety of reasons, may resent or fear the influence of the foreign power. Such electoral aid also often includes a quid pro quo in which the candidate enacts policies favored by the intervener in return for electoral support. Such promises impose upon the client “sovereignty costs”: they reduce her freedom of action with respect to those, and perhaps other, policies preferred by the foreign power.7

As a result, we should expect most cases of electoral interventions to occur in marginal elections: those in which the result is highly uncertain or one side lags but remains electorally viable. In such situations, great-power intervention is most likely to have a significant effect on the results of the election.8 Given that, all else being equal, the more resources that a particular candidate or party has, the more likely they are to win (Sudulich and Wall 2010, 1; Benoit and Marsh 2008, 874), we can assume that interventions usually increase the electoral chances of the aided party or candidate.Hypothesis 1: An electoral intervention for a particular candidate or party will increase its electoral chances.

Whether a great power chooses covert or overt forms of electoral intervention likely matters a great deal. Conventional wisdom expects that overt electoral interventions, as other kinds of overt interventions, rarely work as intended. It assumes that public intervention produces a backlash against the intervener and thus harms the prospects of the side that it supports.9 Covert interventions, therefore, should prove more efficacious.

However, this ignores the potential benefits of overt interventions. Moreover, great powers that engage in electoral interventions will take steps to minimize the risks of a backlash: they benefit from the information provided by their client about how best to calibrate their electoral intervention in light of local sensibilities, preferences, and politics. Indeed, if overt interventions always failed, we would have difficulty explaining why they have not become rare.10 And, as noted earlier, the evidence for blowback effects remains uncertain; Costange and Marinov (2012, 664–69) failed to find evidence of a backlash in their study.11

Covert and overt electoral interventions involve different mixes of costs and benefits. Overt electoral interventions allow for more extensive electoral manipulation (and higher chances of success) but carry with them some risk of blowback. Consider the distributional politics model of Dixit and Londregan (1996, 1136–40), in which politicians can win elections by promising the transfer of resources to various “persuadable” voter groups (thus buying their votes). Taken in the context of an intervention, this model suggests that great powers, owing to their resource advantage, will usually enjoy a superior ability to promise the foreign population the transfer of particular resources (or threaten the loss of existing resources) to that of any local politicians. As a result, direct overt messages from the great power conveying threats or promises to the target’s public can produce a significant shift in the public’s voting patterns. However, overt electoral interventions are risky. If the public in the target country dislikes any facet of the overt intervention, it can lead to a backlash against the preferred candidate, hurting rather than helping his or her chances of being elected.12

In contrast, a covert intervention carries far lower chances of a backlash due to the inherent secrecy in the provision of the electoral aid. However, the lower risk comes with reduced effectiveness. This is due to the nature of covert interventions. A covert operation needs to provide enough assistance to the client, so it will have a good chance of winning the elections while being, at the same time, greatly limited in the means (or the magnitude of the means) it can use. This limitation is necessary in order to avoid exposure and to enable “plausible deniability” (for this general feature of covert operations, see (Lowenthal 2003, 173–74)). The chances that this delicate balancing act will lead to the under-provision of electoral aid to the client and a subsequent defeat in the elections are far higher than in overt interventions.

The intervener, knowing the benefits and risks of each subtype, will act strategically when choosing the method of intervention, using the information it has on the target public’s preferences (as usually provided by the client) in order to maximize the client’s electoral prospects. For example, in the US electoral intervention in the 1969 Thai elections, the US government chose to intervene in a covert manner largely because the side that it aided demanded complete secrecy in the provision of the US electoral aid, claiming that “A leak would destroy them” (FRUS 27 1964-1968: Document 398). Likewise, one major reason why the United States decided to intervene in an overt manner in the 1953 West German Elections, despite fears of some US officials of a possible backlash, was because the aided side, Chancellor Konrad Adenauer (and the Christian Democratic Union (CDU)), relentlessly pushed for various overt acts of intervention in his favor (Levin 2013, 22–23).13

As a result, when the intervener knows or receives information from the client indicating that an overt electoral intervention is likely to lead to a backlash, it will choose a covert intervention. However, because of the lower effectiveness of covert interventions, the intervener is more likely to fail in such cases. Alternatively, when the intervener knows or receives information indicating that much of the target public is likely to positively respond to an overt intervention, it will choose this more effective option. As a result of this strategic behavior, when an overt electoral intervention is used, the intervener is more likely to succeed. In contrast, when the intervener uses a covert intervention, it is more likely to fail.Hypothesis 2: Overt electoral interventions are more likely to benefit the aided candidate or party than covert electoral interventions.

Methodology, Variables, and Definitions

In order to investigate these hypotheses, a plausible model of the factors that affect cross-national voting, of the type frequently used in the economic voting literature, is required. Accordingly, I use the approach recently employed by two major scholars in this subfield, Hellwig and Samuels (2007), and then add the relevant electoral intervention variables. As a further check, I also employ (with two exceptions)14 a second cross-national economic voting model, that of Kayser and Peress (2012), with the inclusion of the electoral intervention variables.15

Besides the inclusion of the variables specified for each of these economic voting models (such as Gross Domestic Product (GDP) growth rate, party fragmentation, trade openness, and country wealth), I also add, for further robustness checks, relevant control variables. These controls are, for example, for various kinds of national security or foreign policy factors (civil wars, interstate wars, and major foreign policy crises), the level of democracy in the target (Polity scale), and the time period involved (Cold War era or later). Online Appendix 1 provides further descriptions of the control variables and their construction, as well as of the subsequent robustness checks.

All models analyzed here use the standard tool used in cross-national aggregate studies of the economic vote (Wilkin, Haller, and Norpoth 1997; Samuels 2004; Benton 2005), as well as the two above noted replicated studies, OLS with PSCE (panel corrected) robust standard errors (Beck and Katz 1995).

I define an intervenable/competitive election, or the universe of cases in which electoral interventions can potentially occur, as one that receives 7 out of 7 on the 2010 Database of Political Institutions’ executive electoral competiveness index (Beck, Clarke, Groff, Keefer, and Walsh 2001) with a small modification. For an election to get that score, multiple parties (in parliamentary systems) won seats in the election and the largest party received less than 75 percent of the vote, or, in presidential or semi-presidential systems, multiple candidates ran and the winning candidate won less than 75 percent of the vote (for examples of use, see Brownlee 2009; He 2007; Treisman 2007).16

Following this criterion and extending the coverage of this index back to 1946 using Nohlen’s data (see description below), 938 national level executive elections in countries with a population of above 100,000 have been found. These elections come from 148 different countries.17 Because of various kinds of missing data on the independent and dependent variables, the number of elections (and countries) on which the statistical analysis can be done is somewhat smaller in practice.

As noted, the dependent variable in all of the models estimated here, as is common in models of economic voting, is the vote share of the incumbent’s party (in parliamentary systems) or of the incumbent party’s presidential candidate (in presidential and semi-presidential systems with direct elections). Given the independent variables used here, this effectively provides an estimate of the difference between the vote share received with the intervention and the counterfactual vote share that would have been received without it. Nearly all of this data came from the edited volumes by Dieter Nohlen and colleagues (1999, 2001, 2005, 2010) on elections around the world. These scholars, over the course of the last two decades, have painstakingly assembled data on national level election results from all independent states from 1946 to the present. The data on election results in different countries is standardized into one common format, making it an ideal source for cross-national comparisons.18

The main independent variables, partisan electoral interventions and, in subsequent models, subtypes of such interventions, are taken from a new dataset constructed by the author, which includes all such interventions between January 1, 1946 and December 31, 2000 that were done by the United States and the USSR/Russia. An electoral intervention is defined as a situation in which one or more sovereign countries intentionally undertakes specific actions to influence an upcoming election in another sovereign country in an overt or covert manner that they believe will favor or hurt one of the sides contesting that election and which incurs, or may incur, significant costs to the intervener(s) or the intervened country. Some examples of acts that are counted under this definition of intervention include public threats and/or promises made by an official of the great power prior to the elections, provision of campaign funds to the preferred side,19 “dirty tricks,”20 pre-election concessions by the intervener and/or benefits of various kinds to the target, and creation of campaign materials for the preferred side.21 (For a complete list, see Online Appendix 1; see Online Appendix 2 for further description of dataset construction).22

An electoral intervention is coded as covert when all of the significant acts done in order to help a particular party or candidate were a secret and/or the connection between those acts and the election was not known to the average voter in the target. An intervention is coded as overt when at least some of the significant acts done in order to aid a particular candidate or party were known to the average voter in the target to have been done for this purpose.23 Following these definitions, 117 US and Soviet/Russian interventions have been found during this period (see Figure 1 for a description of the dataset).

Figure 1.

Electoral intervention dataset 1946–2000 summary description (left bar first subtype, right bar second subtype)

To investigate my first hypothesis, I include in the first set of models an electoral intervention variable (Electoral Intervention). In order to model the fact that electoral interventions can be done in order to help or to harm the incumbent this variable is constructed as trichotomous,24 coded as 1 if an intervention is for the incumbent, −1 if it is for a challenger, and 0 when no intervention occurs.25 If hypothesis 1 is correct, I would expect a positive and significant effect.

In order to test the second hypothesis I include in the second set of models two trichotomous variables, one of overt interventions (Overt Int.) and one of covert interventions (Covert Int.), following my coding of my electoral intervention dataset. If hypothesis 2 is correct, I would expect the overt interventions variable to have a positive and significant effect, as well as an effect larger in substantive terms than the effect of the covert interventions variable. Electoral interventions can include both significant overt and covert components (for example, a public threat/promise and a covert campaign aid). However, most of the overt interventions do not include a covert component.26 Accordingly, I also include a variable (Covert & Overt) for such cases.27


Tables 1.1 to 1.3 present the statistical results for the first hypothesis regarding the beneficial effects of interventions for the aided side. As can be seen from Model 1 in Table 1.1, I am able to replicate Hellwig and Samuels’ (henceforth HS) main result—that the interaction of economic growth and trade openness significantly reduces the incumbent’s vote share. However, under some subsequent robustness checks, such as fixed effects (Table 1.1 Model 2) and fraud limit (Model 4), this result becomes insignificant.28

View this table:
Table 1.1.

Hypothesis 1: Electoral interventions effects in HS model

& Electoral Int.Fixed effectsNo interactionsFraud limit& repeat E. Int.& FP controls
Electoral Intervention3.190**2.976*3.280**3.115*3.194**3.052*
Previous vote0.368**0.373**0.368**0.389**0.367**0.374**
Trade Openness0.315−2.073−0.8290.9450.3020.261
Growth*Trade Openness−0.291*−0.183−0.257−0.291*−0.262*
Presidential Election−1.737−5.509−1.661−2.925−1.679−1.496
Growth*Pres. Election0.03670.1560.04950.03510.0232
Effective num. of Parties (logged)−14.30**−13.40**−14.22**−13.24**−14.37**−14.29**
GDP Per Capita (logged)0.9351.2390.9160.6530.9810.969
Central & E. Europe−4.710*−4.715*−6.211**−4.668*−4.638*
L. America & Caribbean−1.608−1.534−1.659−1.559−1.524
Repeat Int.0.971
Civil War−1.583
Interstate War4.398+
Elections (N)698698698634698698
  • Notes: These and later models calculated using Stata 11.

  • Standard errors in parentheses

  • +p<.10

  • *p<.05

  • **p<.01

  • Omitted when calculating country fixed effects owing to being country invariant.

View this table:
Table 1.2.

Hypothesis 1: Electoral interventions effects- various controls

& Polity Control& interaction w/Polity& Cold War control& Elect. Observers& interaction w/claritySeparate US & Rus. Int.
Electoral Intervention3.329**5.732+3.198**3.192**4.056*
E. Int.*Polity−0.868
E. Int.*Clarity−3.126
US electoral Int.3.111**
Rus. electoral Int.3.324
Previous vote0.374**0.375**0.368**0.368**0.368**0.368**
Trade Openness0.4780.4710.3910.341−0.2350.309
Growth*Trade Openness−0.322*−0.343*−0.291*−0.282*−0.235−0.290*
Presidential Election−1.668−1.684−1.694−1.717−1.637−1.740
Growth*Pres. Election−0.0377−0.03740.03740.0437−0.05270.0361
Effective num. of Parties (logged)−14.10**−14.07**−14.23**−14.34**−12.83**−14.30**
GDP Per Capita (logged)1.381*1.410*1.0020.9230.9050.930
Central & E. Europe−5.682**−5.642**−4.530*−4.746*−4.928*−4.709*
L. America & Caribbean−2.147−2.097−1.508−1.643−1.471−1.611
Cold War0.227
Elect. Observers0.0377
Elections (N)698698698697675698
  • Standard errors in parentheses.

  • +p<.10

  • *p<.05

  • **p<.01.

View this table:
Table 1.3.

Hypothesis 1: Electoral interventions effects (in KP)

Model 4 & E. Int.Model 7 & E. Int.Model 4 & E. Int. & fraud limitModel 7 & E. Int. & fraud limit
Electoral Intervention3.413**2.958*3.252**2.859*
Previous vote0.394**0.386**0.409**0.405**
Global Growth0.6200.5920.4790.558
Local Growth0.147*0.174*0.129+0.147+
Effective num. of Parties (logged)−14.67**−13.92**−13.11**−11.99**
Population (logged)0.3203.6000.1735.415
Elections (N)700700636636
  • Standard errors in parentheses.

  • +p<.10

  • *p<.05

  • **p<.01.

In Model 1, I also include the electoral intervention variable. The effect is in the predicted positive direction and has significant effects both statistically and substantively. A Wald test also indicates (at the 0.01 level) that it significantly increases overall model fit. On average, an electoral intervention in favor of one side contesting the election will increase its vote share by about 3 percent—quite a significant effect. For example, such a swing in the vote share from the winner to the loser in the fourteen US presidential elections occurring since 1960 would have been sufficient to change the identity of the winner in seven of these elections.29 Its effects also stack up quite well in comparison to another well-known major effect on election results: the state of the economy. As can be seen in Model 3 (where the interactions of the economic growth variable with two other variables are excluded), a 1 percent increase in the real GDP per capita would increase incumbent vote share by about 0.4 percent. To illustrate its effects, had Carter in 1980 run for reelection with the economy that Reagan had in 1984 (+6.6 percent) rather than the economy he actually had in 1980 (−1.9 percent), my model would have predicted Carter’s vote share to have increased by about 3.4 percent. An illustration of the effects of electoral interventions in “real life” cases of such interventions is later provided in this section.

The following models include a battery of various robustness checks. My results hold under country fixed effects (Model 2), as well as when elections in which evidence exists that significant election fraud had occurred (in other words, possible “measurement error” on the dependent variable) are excluded from the dataset (Model 4). In the following model (Model 5), I include a control for whether the intervention in a particular case is a repeat intervention on the side of that intervener. This robustness check examines, among other things, the effects of any possible experience accumulated by the intervener from intervening in elections in that target in the past. This control is neither significant nor has any effects on my result.

In Model 6, I check for three other foreign policy or national security variables that may influence the election results: the existence or eruption of a civil war, an interstate war, or an international crisis involving that country in the twelve months preceding the election. The inclusion of these three factors does not affect the results. Interestingly enough, the control for interstate war has a substantive positive effect on incumbent vote share (+4.4 percent), albeit only at the 0.1 significance level. While this latter result loses its significance in some specifications (see Online Appendix 1), it is nevertheless worthy of further study, especially given that there are only 28 cases of countries engaged in an interstate war prior to an election in my dataset.30

In Models 7 and 8 (Table 1.2), I check for the possible effects of the level of democracy of the target on the effectiveness of the intervention. A logged combined polity control variable does have a significant and substantive effect (Model 8) with every one unit increase in the logged polity scale (now ranging from 0 to 3) decreasing incumbent vote share by 4.6 percent. However, this does not significantly affect the electoral intervention variable. Likewise a subsequent interaction between the polity variable and my intervention variable is insignificant (Model 8).

Model 9 examines the robustness of the effects found in Model 1 as to possible temporal patterns in the data. A dummy variable for the Cold War period does not have a significant and substantive effect, and an interaction between this variable and the electoral intervention variable (see Online Appendix 1) is not significant. Similar non-significant results are found when the time trend variable in Kayser and Peress’s models is included (see Table 1.3) or an interaction between this variable and the electoral intervention is added as well (see Online Appendix 1). These results indicate that, at least within the post-World War 2 era, there is no significant relationship between the time period in which the intervention occurred and the effectiveness of the intervention.

In Model 10, I check for the effects of another potentially significant international factor—the presence of international election observers in the country prior to the elections. This factor has no significant effects.31 In Model 11, I check for the potential effects of clarity of responsibility—the extent to which the pre-election political and institutional context increases or reduces the perceived responsibility of the incumbent for her administration’s recent performance. This factor is believed by many scholars in Comparative Politics to affect the willingness of voters to reward or punish the incumbent for her performance (Powell and Whitten 1993, 400–401,410). The interaction between clarity of responsibility and my intervention variable is statistically insignificant, indicating that this factor exerts no significant influence on the effects of electoral interventions.

My theoretical argument about the effects of electoral interventions expects the United States and the USSR/Russia to be equally effective.32 Nevertheless, some may wonder whether these general patterns are indeed what is found in practice given the significant differences between the United States and the USSR/Russia (such as regime type and areas of influence). Accordingly, in Model 12, I disaggregated the electoral intervention variable by US and Soviet/Russian interventions. The results for the disaggregated Russian and US electoral intervention variables are essentially the same as for the aggregated variable although the results for the Russian interventions are not significant. This is probably due to the relatively small number of Russian interventions (as low as 22 cases in some models here and in Online Appendix 1), which have not been dropped owing to missing data on some covariates.33

This finding cannot, of course, foreclose the possibility that the electoral interventions of the United States and the USSR/Russia each differently affected the target after the intervened election. For example, the types of candidates or parties usually supported by the USSR/Russia may have had, on average, less willingness to do certain types of economic reforms or to follow democratic procedures once victorious than the ones usually supported by the United States. As a result, future studies of other effects of electoral interventions must take the differences between the United States and the USSR/Russia into account. Nevertheless, these results do show that both powers had an equal ability overall to affect electoral outcomes in elections around the world.

Finally, in Table 1.3, I attempt to replicate the Kayser and Peress (2012) study (henceforth KP) on the cross-national economic vote and then include the electoral intervention variable. This is done in order to make sure that my finding of a positive, significant, and substantive effect for electoral interventions was not the result of an unusually well-fitting economic voting model. As can be seen in Table 1.3, I am able to replicate KP’s main result—that once the tendency of the public to compare (or benchmark) the economic performance in their country to that in other relevant countries (or global growth) is controlled, the effects of the local economic performance on the incumbent’s vote share will consistently be affected by the performance of the local economy.34

As for the argument presented here, my trichotomous electoral intervention variable (Models 13–16 and Online Appendix 1) continues to have significant effects both statistically and substantively. A Wald test indicates here as well (at the 0.01 level) that the electoral intervention variable significantly increases model fit.35

Table 2 presents my results for the second hypothesis on the differing effects of covert and overt electoral interventions. As can be seen from Model 1, the results support Hypothesis 2, with overt interventions on average increasing vote share by 3 percent more than covert interventions. This is a major substantive difference in the effects of each subtype.36

View this table:
Table 2.

Hypothesis 2: Effects of covert and overt electoral interventions

HS & Covert/Overt Int.& Fixed effects (HS)& Fraud Limit (HS)KP Model 4 & E. Int.KP Model 4 & E. Int. & fraud limitKP Model 7 & E. Int.& No exposed int. (HS)& overt combined w/mixed
Overt Int.5.424*5.423*5.181*5.507*5.243*5.181*5.478*4.381*
Covert Int.2.2551.9212.298+2.559*2.448+2.0242.0232.368
Covert & Overt−10.13+−13.35−9.976−9.953−9.907−11.66−10.23+
Previous vote0.372**0.373**0.392**0.398**0.412**0.388**0.362**0.367**
Trade Openness0.120−2.0030.764−0.1050.249
Growth*Trade Openness−0.265*−0.165−0.233−0.247+−0.278*
Presidential Election−1.867−5.317−3.054−1.820−1.737
Growth*Pres. Election0.008400.1110.0200−0.003510.0382
Effective num. of Parties (logged)−14.32**−13.41**−13.29**−14.68**−13.15**−13.95**−14.51**−14.41**
GDP Per Capita (logged)0.7960.9380.5160.7630.931
Central & E. Europe−4.560*−6.084**−4.346*−4.698*
L. America & Caribbean−1.491−1.570−1.337−1.622
Global Growth0.5890.4470.556
Local Growth0.137+0.1180.164+
Population (logged)0.3360.1873.942
Elections (N)698698634700636700695698
  • Standard errors in parentheses.

  • +p<.10

  • *p<.05

  • **p<.01

  • Omitted when calculating country fixed effects owing to being country invariant.

Models 2 to 6 in Table 2 (and Online Appendix 1) include a battery of robustness checks for this result, similar to those done for Hypothesis 1. These models show essentially the same results as to the effectiveness of covert and overt intervention variables as Model 1. Interestingly, electoral interventions that combine both covert and overt components are insignificant. This result may be due to the nature of the domestic actors receiving assistance of both types. In other words, actors receiving both types of aid simultaneously may be in an unusually weak political situation (which may be why they requested components of both types of aid), making it quite hard for the great power to provide them with useful assistance.37

Some may wonder whether the differences found in effectiveness between covert and overt interventions in favor of the latter are simply due to the cases in which the covert interventions were exposed prior to the election and the resulting political fallout. In order to investigate that possibility, I excluded in Model 7 the small number of cases of US or Soviet/Russian covert interventions for which clear evidence about this activity by the intervener was exposed and became well known to the public prior to election day. As can be seen from Model 7, this had little effect on the results. Likewise, some may wonder whether interventions which combine both overt and covert components should actually be coded and included as overt interventions. Given the nature of the hypothesis investigated here (that overt and covert interventions differ in their effects), separating the mixed category from overt interventions is required in order to fully identify any differences in the effects of these two subtypes. Nevertheless, as an additional robustness check, I included in Model 8 such interventions together with the fully overt interventions. As can be seen from Model 8, while the effect of overt intervention is a bit smaller in substantive terms, the results are otherwise the same.

Diagnostics and Dealing with Possible Selection Bias Issues

Turning to possible issues of multicollinearity, the Variance Inflation Factor (VIF) statistic indicates little multicollinearity between the independent variables. The average VIF in the main models for all hypotheses is 2.22. As for specific variables, with the exception of the interaction terms which by construction are multicollinear with at least two other variables, none of the variables entered into the models show much collinearity and none have VIF scores above 10 (usually seen as the threshold for harmful multicollinearity [Kennedy 2003, 213]).38 Other diagnostics (residual plots for heteroskedasticity, partial regression plots, and Cooks D for potentially influential observations/outliers) indicate no other possible serious problems “under the hood” as well.

A final concern, as to the possibility of selection bias, involves the electoral intervention variables. Two types of possible selection bias may exist in this regard. The first is selection bias due to missing cases of covert electoral interventions. The data collection strategy of this measure was carefully designed to prevent missing cases of such interventions.39 For example, in order to allow sufficient time for evidence on recent covert interventions to come to light, the dataset stops at the end of 2000. Likewise, the electoral intervention dataset focuses on the United States and the USSR/Russia because of the unique availability of relatively complete data on covert electoral interventions performed by these two great powers. The former USSR is unusual among post-1945 authoritarian powers (for example, China) in that summaries of the archives of its secret services for most of the twentieth century (the Mitrokhin Archives) were smuggled to the West by a defector (see further description in Online Appendix 2). As for the United States, owing to a somewhat more relaxed declassification process for many of the relevant archives, the Pike and Church Committees, and greater domestic and international interest, far more information is available on its post-1945 covert activities than for any of the other democratic great powers (for example, France or Great Britain).

The variety of types of reliable (yet, also frequently ”uncontrolled” or ”unauthorized”) sources from which this dataset was constructed for both interveners also make it highly unlikely that only particular kinds of covert interventions (such as only success cases) were collected. For example, the main source of evidence used for coding the unsuccessful covert Soviet intervention in the 1977 Indian elections noted in the following subsection (see later description) came from the above-noted Mitrokhin archives. The Russian government (and the FSB, the successor service to the KGB) had no ability to control the process by which this failed intervention (as well as other interventions noted in this archive) was exposed. Neither was the Russian government able to block its exposure.

Likewise, the main source of evidence for coding the unsuccessful US covert intervention in the 1980 Iranian elections was a subsequent investigation by the Senate Committee on Foreign Relations (reported in Corn 1993), a process of exposure over which the executive branch had very limited control.40 In another case, one of the main sources of evidence for coding the unsuccessful US covert intervention in the 1955 Indonesian elections came from the (unauthorized) memoirs of a former CIA agent named Joseph Smith who was stationed in that country (Smith 1976). Internal data from the nature of the intervention cases in this dataset also provides evidence for the thoroughness of the data collection process and lack of bias.41

A second concern in regard to selection bias relates to which types of competitive elections the great power intervenes in and/or which particular method of intervention it uses. Some may argue that results from the effects of electoral interventions may be largely due to an alternative “helping likely winners” intervention process in which candidates or parties that are highly likely to win the elections anyway are those who are the most likely to receive such backing, perhaps as a way by which the great power can curry favor with the likely would-be winner. Likewise, the differences found in the effectiveness of covert and overt interventions may be simply due to overt interventions being more likely to be used in “easier” situations than covert interventions (such as situations in which the great power’s confidence in the chances of the preferred side are much higher).42

As previously noted, likely winners are quite unlikely to become the recipients of electoral aid, frequently rejecting such aid when it is offered by the great power.43 Nevertheless, to check for the possibility of a selection bias toward likely winners, unduly biasing the results found here, I used matching (Stuart 2010; Iacus, King, and Porro 2011).44 Matching is a data preprocessing technique designed to bring observational data to approximate, as much as possible, data from a randomized experiment. Matching techniques cannot solve selection effects in and of themselves. Nevertheless, by eliminating the impact of the functional form in the regression model, matching allows for local comparisons between cases in which electoral interventions had occurred and cases in which it had not. As a result, matching minimizes the risk that my estimate of the effects of electoral interventions is affected by systematic biases in which the great power chooses to intervene in elections.45 As expected, matching significantly reduced the imbalance in the post-matching sample, with the L1 global matching index (used to measure imbalance in the matching technique used here) showing an overall decline in imbalance from 0.623 to 0.397.46 Then, following the common methodological advice regarding matching (Stuart 2010, 2, 12–13; Iacus, King, and Porro 2011, 4–5), I reran the main regression models of the first and second hypotheses using the post-matching sample in order to deal with any remaining imbalance. For the sake of brevity, Tables 3.1 and 3.2 present only the results on the main treatment variables (see Online Appendix 1 for the full regression tables). As can be seen in Tables 3.1 and 3.2, the main results remain essentially the same with the substantive effect of the electoral intervention variable (in Hypothesis 1) becoming even a bit larger on average.

View this table:
Table 3.1.

Matching results Hypothesis 1-main models

HS & Electoral Int.KP& Electoral Int.
Electoral Intervention3.515**4.128**
Elections (N)139139
  • Standard errors in parentheses.

  • +p<.10

  • *p<.05

  • **p<.01.

  • The models presented here are Model 1 in Table 1.1 (HS) and Model 12 in Table 1.3 (KP) rerun utilizing the post-matching sample.

View this table:
Table 3.2.

Matching results Hypothesis 2-main models

HS & Covert/Overt Int.KP & Covert/Overt Int.
Overt Int.5.259*6.894**
Covert Int.3.170*2.846*
Covert & Overt−8.287−10.03
Elections (N)139139
  • Standard errors in parentheses.

  • +p<.10

  • *p<.05

  • **p<.01.

  • The models presented here are Model 1 (HS) and 4 (KP) in Table 2.1 rerun utilizing the post-matching sample.

These matching results provide no evidence for a “helping likely winners” selection bias driving the main findings.47 In contrast, these results provide further support for the arguments presented here: that electoral interventions have a significant independent effect on the intervened elections and that overt interventions are usually more effective in this regard than covert interventions are.

Estimated Effects in Particular Election Cases

Of course, a question may be raised regarding how much the estimated electoral intervention effects found here apply in practice to specific elections in which such an intervention had actually been done. Accordingly, in order to illustrate some of the real-life effects of electoral interventions, Table 4 gives the estimated effects on election results in four cases of intervened elections. Each one of these elections (and their results) is widely seen as an important turning point, in retrospect, in the nation’s history.48

View this table:
Table 4.

Estimated effects of the electoral intervention on election results-selected real intervention cases (interventions in favor of challengers in bold)

ElectionAided sideIntervenerActual incumbent vote shareEstimated incumbent vote share w/out the interventionDecisive effect?
W. Germany Nov. 1972IncumbentUSSR45.843.6Yes
India Aug. 1977IncumbentUSSR34.532.3No
Israel June 1992ChallengerUS24.930.3Yes
Yugoslavia/Serbia September 2000ChallengerUS38.243.4Yes
  • Note: Table created by deducting from the true election results the predicted electoral intervention effects in that election. The predicted intervention effect was estimated by generating the predicted vote share from my model for that election and then recoding that case as a non-intervention on the intervention variable, generating a second prediction and deducting this result from the first prediction. The predicted results of Hypothesis 2 were used for estimating the effects.

As can be seen in Table 4, in most cases the electoral intervention had an important and decisive effect on the outcome in the “desired” direction. In the 1972 West German parliamentary elections, for example, my model estimates that the Soviet intervention in favor of Willy Brandt and the Social Democratic Party (SPD) was an important factor in its winning a narrow five seat margin (in a 496 seat lower house or Bundestag) over its main rival Rainer Barzel and the CDU (230 to 225).49 Without the increase in vote share due to this intervention, given West Germany’s electoral system, I estimate that the SPD would have narrowly lost the election to the CDU, 216 to 236, probably leading to Willy Brandt’s loss of the chancellorship.

Likewise, the US intervention against the incumbent, then Prime Minister Yitzhak Shamir, in the 1992 Israeli parliamentary elections is estimated, according to my model, to have cost Shamir’s right-wing Likud party the quantity of votes equivalent to about five or six seats in the 120-seat Israeli parliament (the Knesset).50 Given that in this election, the left wing opposition parties won a narrow, one-seat absolute majority in the Knesset for the first time since the 1974 elections, this intervention was likely an important factor in enabling the coming to power of Yitzhak Rabin, following this election, as the head of a center-left coalition.

Of course, like any other domestic or international factor known to affect elections, an electoral intervention in one’s favor does not always guarantee success to its intended beneficiary. In the 1977 Indian parliamentary elections, the covert Soviet intervention in favor of Indira Gandhi and the Congress party is estimated by my model to have done little to prevent, or to even soften, the crushing blow that it had suffered from the Janata party. In this defeat, which led the Congress party to lose power for the first time since India’s independence, the Soviet intervention is estimated to have assisted the Congress party in keeping only eleven or so seats51 from being lost to the Janata party and/or other parties. This is a number too small to have any serious effect on the election results given that the Congress party lost more than 150 seats in this election and the Janata party won 295 seats and a solid 24-seat absolute majority in the 542-seat lower house (the Lok Sabha).

In contrast, the US intervention against Slobodan Milosevic in the 2000 Yugoslav election is estimated by my model to have been decisive in bringing about his final downfall. Without this US intervention, my model predicts that Slobodan Milosevic would have run neck and neck with his main rival, Vojislav Kostunica (43.4 percent to 46.5 percent52). If the first round of the Yugoslav elections had concluded in this inconclusive manner rather than in an outright Kostunica victory (51.7 percent), Milosevic quite probably would have been able, as he had in the past, to “steal” the elections without bringing about the massive wave of demonstrations, which eventually forced him to acknowledge his defeat and resign from the presidency.

Discussion and Conclusion

Vojislav Kostunica’s campaign team had good reason to see the foreign support as critical to their victory. My findings demonstrate that, overall, partisan electoral interventions seem to substantively benefit the aided candidate or party. Furthermore, overt interventions prove to be significantly more effective than covert interventions in swaying elections.

Of course, given the average effect that I find (about a 3 percent change in vote share), electoral interventions will not always assure victory for the great powers’ preferred candidates. However, such interventions often do swing elections. The evidence presented in this article suggests that in the foreseeable future, partisan electoral interventions will continue to be an effective way for great powers to determine the leadership of other states, regardless of whether their targets are governed by “competitive authoritarian,” partially democratic, or fully democratic regimes. These results also provide further—and cross-national—support for the finding of Corstange and Marinov (2012, 664–669) that no popular backlash effect existed in their survey experiment of overt intervention.

Future research should focus on other effects of electoral interventions. Partisan electoral interventions affect a key democratic institution—the national level elections and the process by which the executive is peacefully replaced or retained. As a result, such interventions may have major effects on the target. For example, one important direction for future research in this regard would be to investigate whether electoral interventions have ramifications for the level of democracy in the target over the medium and long term. Another important direction for future research would be the possible effect of electoral interventions on the target’s domestic stability. Research on this question could investigate, for example, whether such interventions may inadvertently encourage various kinds of extra-parliamentary opposition (such as mass protests, general strikes, riots, and terrorism) by the frustrated losing side.

This study shows that even when foreign powers do not use force (whether overtly or covertly) toward a democracy, they can still exert a major influence over the nature of its leadership, and they are frequently willing and able to use this option. Indeed, in a world in which military interventions are increasingly costly and democracies are more common, partisan electoral interventions are likely to become an ever more central tool of the great powers’ foreign policy. For example, had the Arab Spring led to a new, more enduring crop of democratic and democratizing regimes (in addition to Tunisia) in the Middle East, it is quite likely that some of these countries would have become targets of partisan electoral interventions in order to prevent “undesirable” parties or leaders from winning power. Indeed, carrying out electoral interventions for this purpose was an option openly advocated during the initial aftermath of the Arab Spring by some segments of the US foreign policy community.53

Furthermore, given the fact that many of China’s and India’s neighbors and potential peer competitors are full or partial democracies, a future attempt by either power to rise to regional or global prominence may not necessarily lead to warfare as some theorists predict (compare Mearsheimer 2001, 396–402). Instead, either power may choose initially to invest its efforts in replacing foreign leaders strongly hostile to their geopolitical ambitions with “friendlier” ones through a partisan electoral intervention, thus preempting much of the resistance to their rise. Ballots thus may well supplant bullets in the twenty-first century but in a way quite different than usually conceived.


  • Dov H. Levin is a Lecturer in the Political Science Department at the University of California, 4289 Bunche Hall, Box 951472, Los Angeles, CA 90095-1472. Email: dovlvn{at}ucla.edua. His areas of interest include the causes and effects of partisan electoral interventions and interventions in general, the causes of regional war and peace, nationalism, ethnicity and conflict, and terrorism.

  • Author’s note: For valuable comments and advice, the author wishes to thank Arthur Stein, Robert Trager, Barbara Geddes, Giacomo Chiozza, Jeff Lewis, Christopher Tausanovich, Javier Rodriguez, Ron Gurantz, and Chad Nelson. I would also like to thank the participants of panels at the 2013 APSA conference and the 2013 and 2014 ISA conferences in which earlier versions of this paper were presented, as well as the editors of the International Studies Quarterly and the two anonymous reviewers, for all the very helpful comments. Any remaining errors and omissions are the author’s own.

  • 1For some of the growing literature in International Relations on the significant effects that leaders and their various characteristics can have on their countries’ domestic and foreign policies, see (Chiozza & Choi 2003; Potter 2007; Colgan 2013; Horowitz and Stam 2014; Gallagher and Allen 2014; Mattes, Leeds, and Royce 2015).

  • 2As Prados notes in a concluding chapter in which most covert activities of various types are dismissed as unnecessary and costly failures, “The CIA political actions [electoral interventions] were successful within their immediate parameters” (2006, 627).

  • 3For another recent related article by Berger et al. (2013), see Online Appendix 1.

  • 4From my examination of who was exactly aided in such an intervention in the dataset, it is clear that the great powers do not usually have the ability to create significant parties/parties in other independent countries (in other words, to create their own “opportunity”). The most that the great powers have been able to usually do in this regard is to convince some preexisting grassroots parties and/or politicians in the target to agree on a single presidential candidate and/or a common candidate slate for a parliamentary election, both acts naturally requiring quite heavy cooperation with the intervener by the relevant local actors.

  • 5One example is the Eisenhower administration’s decision not to intervene in the 1958 Venezuelan elections when it discovered the above-noted covert Soviet electoral intervention. The United States decided not to intervene because its offer was rejected by the other major presidential candidate who was quite certain of his chances of victory (Rabe 1982, 136–37).

  • 6Given the costs of electoral interventions, the great power will prefer to not waste resources on what it perceives as futile ones. Likewise, an electoral intervention on behalf of a failed candidate will likely undermine the great power’s position with the victorious candidate or party, which, if it has any hopes of using other (costlier) methods to affect its behavior post-election, would make those methods less likely to succeed.

  • 7For related arguments to those given above on the potential downsides of external aid to the recipient in a civil war, see Salehyan (2010, 507). For the way in which, in some newly democratic regimes, existing commitments to a foreign power on military bases can be a major political and electoral liability, see Cooley (2005, 83).

  • 8I expect the various great powers to be equally adept overall in avoiding giving electoral support to “lost causes.” Likewise, as will be seen in the second hypothesis, I argue that the exact method of intervention is determined by the great power based upon the information it receives from the supported candidate or party. Accordingly, I do not expect the identity of the intervener to matter as to the effects its intervention has on the election results. Nevertheless, the possibility of differing effects in this regard is examined as well; see Model 12 (Table 1.2) and the subsequent discussion.

  • 9To give one example of this view, Huntington (1999, 39) claims that “the more the United States attacks a foreign leader, the more his popularity soars among his countrymen who applaud him for standing tall…the best way for a dictator of a small country to prolong his tenure in power may be to provoke the United States into denouncing him as the leader of a ‘rogue regime’ and a threat to global peace.”

  • 10For example, about twenty of the overt interventions in the dataset had occurred during the 1980s and 1990s.

  • 11For a similar recent finding in a fully authoritarian, non-electoral context, see Bush and Jamal (2015).

  • 12Decision makers have long known about this risk; see, for example, (Stout 1997, 1; The Foreign Relations of the United States [hence FRUS] 1952–1954, 6, 499–500). In Dixit and Londregan’s (1996, 1135, 1138–39) terms, this is the situation when the voters have strong ideological preferences vis-a-vis the great power or the relevant issues that overwhelm any other economic preferences, etc.

  • 13When the factors that cause electoral interventions were analyzed, the evidence indicated that once the intervener had decided upon an electoral intervention, the question of any possible cost differentials (for the intervener) between overt and covert interventions was not of significance in determining the exact subtype or methods used. Likewise, the available evidence seems to indicate that the decision regarding which particular intervention method to use is rarely influenced by possible or expected reactions of the intervener’s own public. As is many times the case in the use of other non-military policy tools (such as economic sanctions, see Cox and Drury [2006, 711]), the use of this tool, even when its use is overt and it is well known to the target’s public, is frequently not widely known to the intervener’s public. Accordingly, decision makers, even democratic interveners such as the United States, seem to usually disregard this factor in electoral interventions. To authoritarian interveners, who are largely insulated from internal domestic pressures, this is also not an important factor.

  • 14The first noted variable, unemployment, is unfortunately not available for many countries. Furthermore, the differences in its measurement and reliability in non-OECD countries make it a problematic tool for cross-national comparisons outside of the OECD (see Hellwig and Samuels 2007, 303). Indeed, even in the models of Kayser and Peress, local unemployment has no significant effects in eight of the nine models tested. The second missing variable, coalition size, is unfortunately either unavailable, irrelevant, or an inapplicable concept for many of the presidential and semi-presidential systems included in my dataset.

  • 15I thank Michael Peress for the aid provided in replicating this model.

  • 16Elections to constitutional assemblies, if one of their explicit purposes is to select an executive, are included. Also included are partial or supplementary elections in parliamentary systems as long as the number of seats contested in them is at least 10 percent of the total (usually enough to potentially affect the parliamentary majority of the executive). For further explanation of this choice, see Online Appendix 1.

  • 17Australia, Denmark, New Zealand, Greece, and Japan have the most competitive elections in this dataset (between 18 and 22) while 23 countries have only one competitive election.

  • 18For a small number of parliamentary systems in which vote share was repeatedly missing, seat share was used instead. Cases in which the elections were clearly competitive but the results were invalidated before becoming fully available and/or the data sources indicate that election fraud was so massive as to make the results completely unreliable were excluded from the data.

  • 19Most (but not all) such monetary contributions are covert. However, one must note that many countries with competitive elections did not have until very recently (and some still do not have) any campaign finance laws and/or these laws include large loopholes or are widely seen as “dead letters.” Accordingly, interventions using monetary contributions were not necessarily illegal under the target’s domestic laws in many relevant elections. Neither were the legal issues the main concern in regard to how this method was exactly used.

  • 20These are acts, usually covert in nature, designed to sabotage in various ways the campaign of the “unwanted” side in the intervened election (such as creating and leaking forged documents ostensibly providing evidence of serious misdeeds by that candidate or party or burning down their campaign headquarters). If Russia was indeed behind the (still disputed) case of the attempted poisoning of Viktor Yushchenko in 2004, the main Ukrainian opposition candidate, that is one recent example, albeit somewhat extreme and unusual, of the use of such an electoral intervention method.

  • 21Acts done by private citizens of a great power on their own volition, such as US campaign consultants hired by a candidate or party in another country to give it campaigning advice, are accordingly excluded. Activities by organizations largely funded by one great power, such as the NDI or IRI, are counted as a partisan intervention if the election-related assistance provided in the run-up to an election in a given country is designed to exclusively help only one particular side contesting it rather than being available to all interested parties or candidates (as is usually the case with the above examples).

  • 22These appendices are available at www.dovhlevin.com, as well as the ISQ website.

  • 23To determine whether a certain known intervention was overt, I examined pre-election mass media descriptions of these acts (and/or reliable secondary sources describing these reactions). If these acts are described by the media as being part of a foreign electoral intervention, then it is assumed that the average voter knew about this intervention. Given that overt interventions are designed to affect public opinion in the target, there was rarely any ambiguity in this regard in practice as to the main components of these interventions. See Online Appendix 2 for a further description.

  • 24Electoral interventions for either side by a great power are seen as inherently identical. The operational differentiation here is only due to the nature of the independent variable (incumbent vote share), which requires it in order to estimate the effect of electoral interventions.

  • 25Given that in cases of double interventions (the Soviets helping one side, the Americans another) there are effects in both directions, I exclude these interventions. Double interventions are quite uncommon in practice, with only seven of the intervened elections in the dataset falling into this criteria (7.8 percent of Cold War interventions, 6.3 percent of the entire period). Also excluded are the five cases (nearly all in founding elections) in which the identity of the incumbent is unclear.

  • 26Only ten of the overt interventions (23.8 percent of all overt interventions) in the dataset fall into this category. Given the good data availability in most cases of overt interventions as to any covert activities also being done by the intervener in that election, it is clear that covert components are not an “automatic” part of most overt interventions. See the diagnostics subsection in the main text and Online Appendix 2 for a further description.

  • 27For a description of the way this variable was coded, see Online Appendix 1.

  • 28This also occurs in many subsequent models where the electoral intervention variables are included as well. In a “clean” replication (without the electoral intervention variable) of HS (see Online Appendix 1), this interaction becomes insignificant under fixed effects as well (0.11).

  • 29Assuming, of course, a similar shift in the relevant swing states and, accordingly, the electoral college.

  • 30Likewise, in a different robustness check, I checked an alternative foreign policy effects measure: sanctions. No significant effects were found (see Online Appendix 1).

  • 31Similar results are found when, in an alternate specification, I include only western-based international monitors. This control is significant and in the opposite direction, but this result is also not robust (see Online Appendix 1). The sending of election observers is also, of course, one of the major non-partisan methods by which democracy is promoted (Hyde 2007). Accordingly, this factor, together with the robustness checks for period, etc., should also capture any separate effect, if any, that such efforts have on election results if done in conjunction with a partisan intervention.

  • 32See ft. 8.

  • 33Similar results are found when these disaggregated variables are included in KP’s model as well as to Hypothesis 2 when a similar disaggregation is done regarding covert and overt intervention variables (see Online Appendix 1).

  • 34This result by KP was quite robust to alternate specifications in the various robustness tests conducted for the electoral intervention hypotheses. However, it also, in some subsequent robustness checks, becomes insignificant.

  • 35The inclusion of various additional control variables shown in Tables 1.1 and 1.2 in KP’s model leads to similar results as those in the HS model (see Online Appendix 1). Other robustness checks using HS and KP’s models utilizing, for example, other controls (such as for the target’s relative power [Cinc]) also showed no significant effects (see Online Appendix 1).

  • 36Excluding the few cases in which covert interventions were exposed prior to the election had no effects on these results (see Online Appendix 1).

  • 37The lack of significance of this result is most likely due to the relatively small number of such interventions (10). One example of how weak the supported side was in many interventions that included components of both types is the 1996 Russian election. In that election, the assisted candidate and incumbent (Yeltsin) was so unpopular that in polls done prior to the start of the intervention, Yeltsin was receiving only 8 percent—the least popular of the five major candidates then in the presidential race (Kotz and Weir 2007, 260). Yeltsin’s extremely bad political situation was due to a combination of factors including a painful transition to capitalism, which reduced many Russians standard of living; a bloody, unpopular, and unsuccessful war against Chechen insurgents; and Yeltsin’s deteriorating health. Nevertheless, in this particular case, the United States was able to “save” Yeltsin, but only after one of the most massive electoral interventions recorded in my dataset. In other such cases, the great powers were usually unable to find sufficient ways to help candidates in such weak situations.

  • 38VIFs estimated using the Stata add-on Collin www.ats.ucla.edu.

  • 39See further description of dataset construction in Online Appendix 2.

  • 40These Congressional investigations were frequently composed of members of Congress opposed to such activities either in general or to those targets in particular, caring little as to how the exposure of evidence about these covert interventions would reflect on the CIA’s reputation for effectiveness, etc. Likewise, the newer volumes of another major source, the FRUS (published by the US government), are required by law (P.L.102-138) to note any such covert activities.

  • 41For example, given that overt interventions, by their very nature, are far easier to track down, one would not have expected the high proportion of covert interventions in my dataset (nearly two thirds of all electoral interventions) unless the process of data collection was quite thorough. Likewise, the number of US and Soviet/Russian covert interventions that were clear operational failures (such as exposure prior to the election) is quite small. For a further test for the possible effects of any nevertheless missing cases of covert interventions (a simulation), see Online Appendix 1.

  • 42Alternatively, one could argue that the selection between covert and overt interventions may be due to possible cost differentials between these two subtypes.

  • 43For one example, see ft. 5.

  • 44Coarsened Exact Matching was conducted using the Stata add-on package CEM (Blackwell et al. 2009). See Online Appendix 1 for a further description of the matching procedure used, as well as the full regression table results.

  • 45It also minimizes the effects of most unobserved covariates (Stuart 2010, 3).

  • 46This measure is an index that measures the global imbalance between the treatment and control observations on the matched covariates ranging from 0 (for perfect balance) to 1 (for full imbalance).

  • 47I try here to match cases on aspects capturing the likelihood of victory, such as the pre-election economic growth rate. As a result, matching in this case only eliminates the possibility that my results are largely due to cases of ‘easy’ victories or landslides being improperly compared by the statistical technique to cases where the election was quite closely fought. Accordingly, the matching technique, of course, does not affect the process of choosing where and how to intervene (described in hypotheses 1 and 2). These, as noted, are determined by the great power’s threat perception, the feasibility, and the willingness of a local actor to cooperate (H1) or by the local actor’s knowledge of which method would be best given the local circumstances (H2).

  • 48For some of the sources used to code these four cases as electoral interventions, see (Andrew & Mitrokhin 2005, 329–30; Memorandum of meeting (Moscow) August 2, 1972, PS archive; FRUS 1969–1972 40:Document 380; Ross 2004, 83–84; Dobbs 2000).

  • 49Shift estimated given Germany’s electoral system (and assuming a uniform shift in the PR component) and that most votes shifted from the SPD to the CDU (the two major parties). The seats of (then non-voting) West Berlin representatives are excluded, but their inclusion would not significantly affect this estimate.

  • 50Parliamentary seat share change given this shift in vote share estimated based on the Israeli election law in force during 1992. Israel had (and has) a single district PR system.

  • 51Estimated using the SMD district level results and assuming a uniform swing in all districts. Many districts had more than two significant candidates (18 parties won seats in the Lok Sabha), so no estimate of where the votes could have otherwise gone (besides being lost to Congress party candidates) could plausibly be made in this case.

  • 52Assuming that most of the votes that Milosevic lost went to Kostunica, a reasonable assumption given that Kostunica was the only other major candidate as well as the main beneficiary of the US electoral intervention.

  • 53See Takeyh (2011) and quotes in Carothers (2011) for examples of such calls.

  • 54The models presented here are Model 1 in Table 1.1 (HS) and Model 12 in Table 1.3 (KP) rerun utilizing the post-matching sample.

  • 55The models presented here are Model 1 (HS) and 4 (KP) in Table 2.1 rerun utilizing the post-matching sample.


View Abstract