Going Negative in Autocracy: A Field Experiment at the Moscow Mayoral Elections

Abstract

Opposition in autocracies often uses negativism against the regime to frame its principal message. This study is the first to experimentally evaluate the effectiveness of a negative campaign on a regime candidate’s vote share. For the field experiment conducted during the 2013 Moscow mayoral election, we published a newspaper criticizing the incumbent mayor. We distributed approximately 130,000 copies near the entrances of 20 stations on four randomly selected metro lines one month prior to the election. We found that the incumbent’s vote share was 1.7 percentage points lower at the voting stations where the newspaper was distributed. These votes go to other candidates who address issues raised by the negative campaign. Anti-regime campaigning does not suppress turnout or increase disapproval voting.

Introduction

Most autocracies in the world today simulate democratic procedures, including elections (Guriev and Treisman, 2022, Magaloni, 2010, Hadenius and Teorell, 2007). The current research focuses on the incentives for autocrats to opt for elections. They serve to enmesh, coopt, and gather information from elites, ordinary citizens, and opposition (Geddes et al., 2018). Autocrats employ several strategies to control election outcomes (Schedler, 2002).

Fewer studies have been conducted on how the opposition responds to the opportunity of elections and how voters react to opposition moves. The opposition often “goes negative” against the incumbent when framing its principal message. The opposition tries to persuade voters that the regime is incompetent, corrupt, or repressive and to unite them around the idea of political change. This narrative frequently becomes central in opposition rallies, boycotts, or coalition campaigns. Despite its centrality, it is difficult to assess the effectiveness of negative messaging from the opposition to the regime because many other events change election dynamics. In this paper, we experimentally evaluate the effectiveness of negative campaigning on votes for a regime candidate and other election outcomes in autocracies.

One month prior to the 2013 Moscow mayoral election, we published a newspaper that criticized the incumbent mayor Sergey Sobyanin. We handed out approximately 130,000 newspaper copies near the entrances of 20 stations on four randomly selected metro lines. We compare the election results at the voting stations where the newspaper was distributed with the results at stations in the control group. We showed that, on average, the incumbent’s vote share is lower by 1.7 percentage points near the treated stations. We found no effect on turnout or the proportion of invalid ballots (when voters disapprove of all candidates). One opposition candidate, Sergey Mitrokhin, particularly benefited from the intervention. He may have gained votes because he addressed issues raised by the newspaper in his own campaign.

Our analyses of the data from this experiment show that informational campaigns in an autocratic environment where media is dominated by state propaganda and a regime candidate appears to be invincible can influence voting behavior. Negative information reduces voters’ positive evaluations of the regime and subsequent vote shares. Given the limited number of tools available to opposition groups and civil society organizations under authoritarianism, negative campaigns can help advance political change. Importantly, we find that negative information not only reduces the regime candidate’s vote share and improves opposition results but also does not discourage voters from showing up at the polls.

Negativism as a framing approach

Overall, scholars agree that, in the long run, elections make autocracies stronger, not weaker (see Gandhi and Lust-Okar (2009) for a review). Usually, autocrats are prepared for elections with superior administrative, financial, and media resources. As a result, the opposition is marginalized, the incumbent appears to be invincible in the eyes of voters, and the elections lead to continuity rather than change in leadership.

However, the opposition often uses the opportunity offered by elections to organize informational campaigns to weaken the regime. This “information cascade” produces a moment of short-term instability (Knutsen et al., 2017). Such focal points unfavorable to the regime may have short- and longer-term consequences for opposition groups and civil society organizations. Rarely do elections in autocracies lead to immediate democratization. Change occurs in cases when the opposition can implement “sophisticated, intricately planned, and historically unprecedented electoral strategies” (Bunce and Wolchik, 2010). However, poorer election results serve as a signal of citizen dissatisfaction (De Miguel et al., 2015). Sending such a signal may bring positive (such as policy concessions) and negative (more repression) outcomes (Ash, 2015; Miller, 2015). In the longer run, a negative campaign may help the opposition increase its political base. This is important because autocratic elections are a repeated game. Information obtained in one election cycle may be relevant in the next cycle (Magaloni, 2010).

Negativism often becomes an approach to frame a political message from opposition groups to voters in authoritarian elections. Beissinger (2013) discusses two longstanding traditions in the democratization literature that aim to explain liberalizing political change. The first explanation concerns shifts in values that make societies become more committed to democracy. The second tradition underlines shared grievances that bring people together to fight for a better future. Both drivers imply a formation of a “negative coalition” that unites different social groups around common rejection of the existing regime. To construct negative coalitions across diverse social groups, the opposition creates democratic master narratives regarding civil and political freedoms that are suppressed by the regime.

It is problematic to establish the causal effects of negative campaigns on political outcomes in autocracies because negative campaigns take place with many other events. The case of Chile’s democratization stands out and provides an opportunity to evaluate the isolated effect of the negative campaign. The “No” campaign took place in 1988 before a referendum and aimed to determine whether Augusto Pinochet should extend his rule for another 8 years. The campaign had a specific time frame and a concrete form of television ads. The data show that this campaign had an important causal role in Chile’s democratization (Boas, 2015). Gonzalez and Prem (2018) note that the effects of this campaign appear large (10–13% of the persuasion rate) when compared to similar estimates for negative campaigns in the context of democracies.

In an authoritarian context, two factors working in opposite directions can influence the strength of the negative informational campaign. On the one hand, one can expect them to be ineffective because of voter apathy (Peisakhin et al., 2020). When the result of the election is known beforehand, engaging with political information may seem a waste of time in the eyes of voters. On the other hand, in an environment dominated by state propaganda, alternative viewpoints may become illuminating for some voters. Unfavorable information creates negative valence evaluations in voters’ minds, and they withdraw their support for the regime candidate.

Our main hypothesis is as follows:

H₁: Anti-regime campaigns reduce votes for regime candidates

We borrow the term “negative campaigning” from the classical literature on electoral dynamics in American politics. Interestingly, nonpartisan negative information about the incumbent has a similar effect on voters in a democratic setting as it does in an authoritarian setting. Wood and Grose (2022) show in a randomized experiment that US legislators whose independent audits revealed campaign finance violations face more competitive reelections and are more likely to retire. Contrary to the view that voters are unwilling to investigate or respond to political information, Wood (2020) demonstrates that negative information about campaign finances informs voters and changes their behavior. Most often, however, the source of negativism in campaigns is another candidate rather than independent audits. The literature considers the following outcomes: the effect on vote shares for an attacker, a target, third candidates, and total turnout. The overall conclusion of the current literature on democracies is that negative campaigns are not as effective in gaining votes because of the backfire effect (Lau and Rovner, 2009). There might be positive spillover effects for third-party candidates (Galasso et al., 2023). We do not anticipate a backlash in an authoritarian setting – a mighty regime candidate does not seem to be an appropriate target for sympathy.

H₂: Anti-regime campaigns increase votes for opposition candidates

Contemporary studies note that in democracies, negative campaigns tend to be more memorable and mobilize turnout (Barton et al., 2016). We anticipate a mobilizing effect of a negative campaign in an autocracy as well. Autocratic regimes often have tough time making voters turn out for elections because they do not expect their voters to make any difference in election results. Negative campaigns mobilize voters by making the regime look more vulnerable than perceived and, in general, bringing some element of competitiveness to authoritarian elections.

H₃: Anti-regime campaigns increase voter turnout

Empirical context

The 2013 Moscow mayoral election was a unique moment in a contemporary Russian autocracy. After the 2011–2012 Russian protests, triggered by electoral fraud during the 2011 legislative election, the regime was pushed to yield on some of its efforts to consolidate power. Direct elections of regional governors were restored. The Moscow mayoral election of 2013 was the first election for the mayor of Moscow in 9 years, and this election was considered relatively free and fair. ¹

On June 5, 2013, the incumbent mayor of Moscow, Sergey Sobyanin (United Russia), announced his resignation from office and soon confirmed his intention to stand for the election held on September 8. Five other candidates were allowed to participate in this election: Ivan Melnikov (Communists), Nikolay Levichev (a Just Russia), Mikhail Degtyarev (Liberal Democratic Party of Russia (LDPR)), Sergey Mitrokhin (Yabloko), and Alexey Navalny.

United Russia has constituted the majority party in the State Duma and most regional legislatures since 2007. Three other parties are allowed to participate in elections and win some portion of seats – Communists, Liberal Democratic Party of Russia (LDPR), and a Just Russia. The members of all three parties rarely criticize the regime and obediently vote for United Russia’s bills in legislatures at all levels. Another party, Yabloko, has been in constant decline since the 1990s due to both endogenous (organizational and strategic) and exogenous (regime level and societal) factors (White, 2006). After losing the 2007 parliamentary elections, it was never able to return to the national legislature. Party members are allowed to participate in elections because they pose little threat to the established order. At the same time, this party suffers more from electoral fraud than parliamentary opposition parties (Enikolopov et al., 2012).

Navalny was an independent politician whose ambition was to destroy the regime of Vladimir Putin. Numerous attempts to create his own party have failed. The Moscow mayoral election of 2013 was the last election in which he was allowed to participate. In August 2020, Navalny was poisoned with a Novichok nerve agent, recovered in a German clinic, and returned to Russia only to be jailed. In 2022, he was sentenced to nine more years in a maximum security prison. In 2023, his term was extended to 19 years. On February 16, 2024, he was murdered in a remote Arctic prison.

The participation of Navalny in the mayoral election was more a product of circumstances than routine involvement in electoral politics. First, registered candidates are required to pass a “municipal filter.” Every candidate had to gain support from at least 6% (110) of municipal deputies from no less than 75% of Moscow municipalities. Given that United Russia controlled most municipalities, it was especially difficult for opposition candidates to pass the filter. Sobyanin, who wanted to boost his legitimacy among Muscovites and avoid a boycott of the election, decided to help the opposition candidates, including Navalny, obtain enough signatures. Second, in the middle of the race, it was revealed that Navalny has a firm in Montenegro, despite Russian law prohibiting candidates from having real estate and bank accounts abroad. Several days later, this news was recognized as fake. Finally, Navalny was registered as a candidate on July 17. The next day, he was sentenced to a five-year prison term for embezzlement. He pulled out of the race and called for a boycott of elections. A total of 15,000 people went to the center of Moscow to protest against the sentence. On July 19, he was released from jail and continued his campaign. The election results are presented in Table 1.

Table 1. 2013 Moscow mayoral election results

Treatment

The newspaper “The Truth about Moscow” created for this field experiment discusses the past and present of the incumbent mayor Sergey Sobyanin (see the copy of the newspaper in Appendix 1). The front page of the newspaper contains a bio of Sobyanin that emphasizes his political opportunism. Other articles discuss the involvement of Sobyanin’s wife, the owner of a paving slab producer, in city development projects and wasteful spending on urban forestry purchased at 10 times the market price. The final article considers illegal labor for which the city government was the main source of demand. Corrupt officials employ illegal migrants at one-third of the official salary for the position to pocket the difference. Therefore, the overall message of the newspaper is anti-corruption. More details of the newspaper content are provided in Appendix 1.

Two candidates raised the issue of corruption in their campaigns – Navalny and Mitrokhin. Importantly, according to Russian law, it is forbidden for a registered candidate to directly criticize his or her competitors. Their campaign materials were mostly positive, although some criticism was inevitable during media interviews and grassroots communications. Navalny built his political reputation as an anti-corruption crusader. During the campaign, he claimed that Sobyanin made wasteful spending on roads and plantations, that Sobyanin’s daughters owned expensive apartments, and that his family members received public contracts. Mitrokhin, the founder of the Anti-corruption Center within his party Yabloko, also stressed corruption in his public statements.

Experimental design

The distribution of approximately 130,000 copies of a newspaper took place during the 4 weeks preceding the election date near 20 metro stations. We excluded the following types of stations from participation in the experiment: (a) stations within the circle line, (b) stations located outside the city of Moscow, (c) transport hubs, and (d) stations on short lines with fewer than four stations in total (see Appendix 2 for the map of the newspaper distribution). We exclude (a) the downtown stations because many business centers, government offices, company headquarters, and tourist attractions are located there. People who use these stations are less likely to live in adjacent areas. People who live in the areas outside the administrative borders of Moscow ((b) stations) do not vote in Moscow elections. We excluded stations with adjacent train and transit bus stations (c) from the pool because most of the traffic at these stations constitutes people not living in the neighborhood, and the effect of the newspaper would be diluted. The exclusion of stations on the short lines (d) was important for logistical reasons. The newspaper was distributed in adjacent lines. Two managers supervised the distribution, each covering two neighboring lines and constantly moving between stations (see the list of covered stations in Appendix 3). ² The delivery of the newspaper at the same long metro line required much lower managerial costs than managing the distribution at the shorter lines.

We grouped the remaining lines into seven pairs of adjacent lines (see Appendix 3 for the list). We randomly selected two pairs of metro lines to be in a treatment group for logistical reasons. All stations from the selected pool not included in the treatment group served as a control group (61 stations).

Then, we chose up to 15 of the closest voting stations located not further than 2 kilometers from a given metro station. ³ Those who live further away were less likely to use the metro for commuting. Our sample included 1,077 voting stations out of 3,590 voting stations in Moscow (with 233 in the treatment group). Such significant attrition is caused by a few reasons. We excluded (1) downtown, (2) voting districts located outside Moscow but still considered the administrative districts of the city (e.g., Zelenograd, “New Moscow,” Solntsevo), (3) districts without close access to metro lines (e.g., Zapadnoye Degunino, Golovinskiy district), (4) transport hubs, and (5) districts near short metro lines. To track the change in the treatment effect with distance, we split voting stations into three groups: the 5 closest to a metro station, the 6th to 10th closest, and the 11th to 15th closest.

Results

Before we proceed to our main results, we demonstrate that our randomization is not related to the main variable of interest, votes for the regime candidate. To verify this, we compare the vote share for Putin in the 2012 presidential election, which took place 1.5 years earlier than the mayoral election. Table 2 presents simple means for the outcome variables. There is no statistically significant difference in Putin’s vote share between the treatment and control groups. In addition, we estimate the effect of newspaper distribution using the number of votes for Putin as a dependent variable as displayed in Table 3. The coefficients for Newspaper are not statistically significant.

Table 2. Means of the dependent variables by experimental condition

Notes: Means and differences are given in percentages. For the subsample means, the standard deviations are reported in parentheses. For the differences between the treatment and control groups, the standard errors clustered by metro line pairs are reported in parentheses.

Table 3. Votes for Putin and newspaper distribution

Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Errors clustered by metro line pairs are reported in parentheses.

Next, we estimate the effect of the newspaper on Sobyanin’s vote share for different subsamples of voting stations. Table 4 presents the results. We find support for H₁. The distribution of the newspaper has a negative effect on votes for Sobyanin in the 5 closest voting stations. There, his share decreased by 1.69 percentage points (95% CI: [−2.81, −0.57], p < 0.01). We do not observe a statistically significant effect on further stations. It is likely that people who live farther from the point of distribution were less likely to get the newspaper. The coefficient for the newspaper dummy in the subsamples for the 6^th–15^th closest voting stations is more than twice as low as it is in the subsample of the 5 closest voting stations. In the full sample (15 voting stations), the Newspaper coefficient does not reach a conventional level of significance, although it is economically large, with Sobyanin’s share lower by 1.27 percentage points. We report the results of the randomization inference procedure that simulates the treatment effect under different hypothetical realizations of the randomization in Appendix 5. They are consistent with the estimates displayed in Table 4.

Table 4. Votes for Sobyanin and the newspaper distribution

Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Errors clustered by metro line pairs are reported in parentheses.

The number of votes for Sobyanin was closely related to the number of votes for Putin in the presidential election. On average, each additional 1% of Putin’s vote in the presidential election translated into an additional 0.91% for Sobyanin (column (4)).

Where do these votes lost by the incumbent go? Table 5A and 5B reports the estimation of the effect of the newspaper on opposition candidates and other election outcomes. We do not find support for H_3. We observe no effect on Turnout and Invalid ballots. The newspaper does not have any mobilizing effect on voters but also does not discourage them from showing up at the polls. The results for opposition candidates’ vote shares are mixed. The campaign had an economically significant positive effect on votes for Sergey Mitrokhin (Yabloko), who received an additional 0.7 percentage points at the treated voting stations (95% CI: [0.10, 1.30], p < 0.05), given his overall share of 3.5%. This result is consistent across the samples of the 5 and 15 closest voting stations.

Table 5A. Votes for other candidates and newspaper distribution on the 5 closest voting stations

Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Putin, Zyuganov, Zhirinovsky, Mironov, and Prokhorov indicate the percentage of votes received by each named candidate in the 2012 presidential election. Errors clustered by metro line pairs are reported in parentheses.

Table 5B. Votes for other candidates and newspaper distribution on the 15 closest voting stations

Notes: *p < 0.05, **p < 0.01, ***p < 0.001. Putin, Zyuganov, Zhirinovsky, Mironov, and Prokhorov indicate the percentage of votes received by each named candidate in the 2012 presidential election. Errors clustered by metro line pairs are reported in parentheses.

It would be reasonable to assume, given the anti-corruption message of the newspaper, that votes lost by Sobyanin would be mostly absorbed by Navalny, a prominent anti-corruption activist. The numerical estimation of the treatment effect for this candidate is approximately the same as the one for Mitrokhin –0.74 percentage points (t-stat = 2.02, p-value = 0.09). The weaker statistical significance for Navalny can be explained by the scale of his own mayoral campaign. ⁴ Thus, there might be more noise present in our estimations. According to his own polls, 43% of Muscovites saw his campaign materials. Mitrokhin was another candidate who emphasized corruption in his campaign and, for some, may have been known for his anti-corruption investigations before the campaign.

Another candidate who gained from the intervention is the Communist candidate, Ivan Melnikov. He received additional 0.72 percentage points at the 5 closest treated voting stations (95% CI: [0.21, 1.23], p < 0.05). This result does not hold on the sample of the 15 closest stations, although the coefficient remains economically large (0.35 percentage points). Melnikov did not campaign on anti-corruption messages. However, in general, the Communist Party might be perceived as a real opposition to United Russia by some voters.

Interestingly, we observe a modest negative effect of the newspaper on the vote share for Mikhail Degtyarev (LDPR). He seems to have been perceived as a pro-regime candidate by some voters. Despite the presence of their own LDPR candidate Vladimir Zhirinovsky in the 2012 presidential election, Degtyarev’s vote share was positively correlated with the votes for Putin in 2012.

Concluding remarks

Building “negative coalitions” is often inevitable during the process of democratization. How effective is negativism as a framing approach against autocratic regimes? The conditions under which they take place vary across and even within regimes at different stages of their life cycle. Gonzalez and Prem (2018) calculate the persuasion rate of the “No” campaign in the weeks preceding the 1988 plebiscite in Chile as between 10% and 13%. However, this case is, to a certain extent, an exception. The decision to allow televised political advertising was considered a major mistake by Augusto Pinochet, who overestimated his popularity, called a referendum, and failed to manipulate the election successfully (Treisman, 2020). We organized a relatively modest intervention in a regional election following massive protests and found that the incumbent vote share decreased by 1.7%. We believe that our intervention can be scaled up and have a more significant impact on elections and the overall stability of the authoritarian regime. However, the level of repression in an autocracy may be so high that launching a negative campaign can be impossible. In 2018, we were unable to replicate our experiment on a larger scale in different Russian regions during the presidential election. We printed 50,000 copies of a newspaper with negative information and distributed them in Saint Petersburg. The next 50,000 copies were seized in Yekaterinburg. We received neither the newspapers back nor a report with the reasons for seizure. Then, we contacted 42 printing plants in 18 regions. All of them refused to print the newspaper.

Importantly, the design of the experiment allows us to estimate a short-term effect. At the same time, negative information about the regime may be cumulative and lead to gradual erosion of the regime’s popularity.

We chose an anti-corruption message because corruption is a real political issue in Russia. Of course, negative campaigns in autocracies should not be limited by this issue. In other contexts, economic performance, human rights abuses, disenfranchized social groups, or the moral conduct of an autocrat’s inner circle might be important. The diversity of negative messages is an endeavor for future research.

As a final point, we should emphasize that our experiment was nonpartisan. This study was conducted independently from any other campaign in this election. Cooperation with any candidate in the experiment whose aim is to decrease the rival’s vote share would raise ethical concerns (and is forbidden by law). In a real-world setting, negative campaigns are usually partisan. The element of partisanship in the negative message may have a different effect compared to a nonpartisan message. This change can go in both directions. On the one hand, voters may realize that an information sender is biased toward the target. On the other hand, they may not only withdraw their support for the regime but also become opposition followers.