Staying Out of Trouble: Criminal Cases Against Russian Mayors

Although repression against elites is a common occurrence in authoritarian regimes, we know little about which elites are targeted. This paper uses an original dataset on the prosecution of mayors in large Russian cities to examine the factors that make elites more likely to be arrested. We argue that in electoral authoritarian regimes like Russia, regime leaders are reluctant to arrest popular officials. Such officials command political capital that is useful to the regime, and arrests of prominent officials can produce popular backlash. We examine this argument using an original dataset on all arrests of municipal leaders in Russia’s 221 largest cities between 2002 and 2018. We find that mayors who won their elections by large margins are less likely to be arrested. In addition, we document several other substantively important patterns: (1) a mayor’s professional background is not related to the likelihood of arrest, (2) opposition mayors are four times more likely to be arrested, and (3) mobilization of votes for the regime is not protective against arrest. JEL Classifications P16, P30, R59


Introduction
Repression of political elites is a common occurrence under autocracy. Office-holders in authoritarian regimes are much more likely to lose office via force (i.e. arrest or violence) than are elites in democracies (Svolik 2009). Subnational officials are a particularly frequent target of such repression. Indeed, in the past five years, there have been major arrest campaigns against local officials in settings as diverse as Venezuela,China,Tanzania,Turkey,Uganda,and Russia. 1 In this paper, we examine why some subnational elites are arrested or charged with criminal wrongdoing, while others are not. There is a vast literature on elite repression under autocracies, but relatively few empirical studies examine which subnational officials are actually targeted with repression (Lorentzen andLu 2018 andZhang 2017 are two exceptions). Generally speaking, the literature on elite repression can be classified into two categories. First, there is scholarship on purges-the large-scale elimination of elites from the regime. In addition to a voluminous historical literature describing famous purges, this literature focuses on uncovering the factors that lead top regime leaders to launch a purge campaign, usually against members of the security services or top officials in the central government (Brzezinski 1958;Bueno de Mesquita and Smith 2009;Sudduth 2017). This literature does not make clear predictions about which elites will be repressed, nor does it help us understand patterns of elite repression in 'normal' times, outside of well-defined purges. We take up both tasks in this paper. Second, there is a large and growing literature on anti-corruption campaigns, primarily focusing on China (Chen and Kung 2019;Li 2019;Li et al. 2018;Manion 2016;Pan and Tian 2017;Wedeman 2005Wedeman , 2017. Much of this literature focuses on determining the purpose of these campaigns. Are they really aimed at stamping out corruption, or are they merely a tool for leaders to eliminate political threats, consolidate their power, and strengthen the political legitimacy of the regime? In recent years, a number of innovative papers have made progress on these questions, generally coming to the conclusion that all three factors-anti-corruption efforts, political vendettas, as well as concerns about the public image of the party-drive the pattern of arrests in China (Lorentzen and Lu 2018;. We adopt a different approach in this paper. We start from the the premise that authoritarian politics is opaque. The decision-making processes of the security services are particularly inscrutable (Soldatov and Rochlitz 2018;Wang and Minzner 2016). Thus, the specific motivation for a particular arrest-whether it is punishment for corruption, a product of factional struggles, or an effort to repress 1 See, for example,"Turkey Suspends Three Mayors, Alleging Links to Kurdish Militants" Washington Post 19 August 2019, "Venezuela Arrests One Mayor and Imprisons Another in a Widening Crackdown" Time. 20 March 2014., "Ubungo mayor, two other Chadema leaders arrested" The Citizen. 20 June 2017. the opposition-is very difficult to discern. Rather than trying to determine the proximate causes for arrests, we theorize the factors that are likely to protect officials from being arrested. Whatever the particular "offense"-be it corruption or being on the wrong end of a political struggle-what are the factors that make it less likely that a given official will be targeted with repression?
Drawing on insights from both the elite repression literature and the literature on subnational politics in Russia, we examine several plausible explanations for variation in the arrests of officials. We train our theoretical focus on examining how the personal popularity of officials affects their chance of being arrested. Some research suggests that popular officials pose a threat to regime leaders, and, as such, are targets for repression (Gueorguiev and Schuler 2016;Egorov and Sonin 2011;Zakharov 2016). We advance the opposing view, suggesting that a base of popular support can protect a leader from arrest. In electoral authoritarian regimes such as Russia, regime leaders need popular officials to help them win elections (Reuter and Robertson 2012;Reuter 2013). In addition, arresting popular officials can produce popular backlash, which can seriously undermine the ability of the regime to win elections. Thus, we argue that more popular officials in electoral authoritarian regimes are-ceteris paribus-less likely to be arrested than are less popular officials.
We examine these propositions using an original dataset of 1051 mayors in Russia's 220 largest cities between 2002 and 2018. During this period, over 10% of these mayors (108 out of 1051) were arrested for corruption-related offenses. Thirty-eight percent of large Russian cities saw at least one of their mayors arrested. We present the first empirical analysis of these arrests. The true reasons for these arrests are opaque and, as we show, even local observers and journalists often have a difficult time understanding the motives behind an arrest. In this regard, large-N analysis excels because it can uncover trends and patterns that close analysis of a few cases may miss.
Our data offers a rare opportunity to leverage high quality, comprehensive data on repression of a homogenous set of elites under autocracy. In many of the other settings studied in the literature, the targets of repression vary wildly in position and authority. Our focus on one elite group-mayorsallows us to examine actors of equal statutory authority.
Focusing on the arrests of Russian mayors is useful for theory-building, because it allows us to exclude some prominent explanations in the literature. For instance, while we show that political factors play a major role, we can exclude explanations that focus on repression as a way to prevent coups or conspiracies (Braithwaite and Sudduth 2016), as mayors in Russia are not in a position to pose such an existential threat. We can also exclude explanations based on factional politics in the top leadership. A number of studies of China find that those with personal ties to Xi Jinping were less likely to be targeted in the latest anti-corruption campaign (see, for example, Lorentzen and Lu 2018). Russian mayors are not linked to the top leadership or embedded in national-level factions, so we can exclude this as an explanation. Finally, we can also exclude explanations based on ethnic rivalry, as ethnicity is not an important factor in Russian elite politics (Giuliano and Gorenburg 2012;Rutland 2010).
In addition, because mayors govern concrete political-administrative jurisdictions, we can param-eterize and measure many structural correlates of repression, a task that would be very difficult in an analysis of repression in the security services. Relatedly, since mayors are public officials, there are many aspects of their biographies and backgrounds that can be gathered and analyzed in a way that could be difficult for other parts of the authoritarian elite.
Our analyses reveal a number of substantively important findings. First, we find that popular mayors, as proxied by their margin of victory in their most recent election, are much less likely to be arrested-irrespective of their political allegiance. We interpret this finding as suggesting that the authorities are hesitant to arrest popular mayors. In additional analyses, we find that these results hold among mayors not affiliated with the regime, which suggests that the findings are not just driven by the need to retain popular pro-regime mayors who can help the regime mobilize votes, but also by a desire to avoid popular backlash.
Second, using occupational backgrounds as a proxy for connections to regional officials, the security services, and the Kremlin-all possible sources of protection-we find no evidence that mayors with such connections are less likely to be arrested. It seems that, at least for this set of authoritarian elites, personal characteristics like popularity are more important than having connections.
We also present several other empirical findings that will be of particular interest to those who follow Russian and post-communist politics. First, empirically confirming commonly-held wisdom, opposition mayors are almost four times as likely than regime-affiliated mayors to be arrested. Second, we find that arrests are more frequent in so-called ethnic republics. Third, we find that mayors are more likely to be arrested in large cities, although this finding is not fully robust to model specification.
Finally, we find no evidence that arrests are more likely after a turnover of governor.
These findings highlight the complex nature of repression in autocracies. We find that who you are can be just as important as professional connections or regime loyalty. We advance the literature on the role of personality, charisma and popularity (Bittner 2018;Breslin 2008;Eatwell 2006;Feuchtwang and Mingming 2001;Jones and Olken 2005;Rahat and Sheafer 2007) by emphasizing that personal popularity can matter in the same way that backstabbing maneuvers between political clans or following the party line can. In this way, our study highlights the complicated relationship that regime leaders have with popular officials. On the one hand, their popularity may be a threat, but this also makes them costly to dislodge.

Literature
The literature on elite repression in autocracies can be grouped into two broad categories: a literature on political purges, and a group of more recent studies on the purposes and effects of anti-corruption campaigns in authoritarian regimes. An extensive historical literature has documented the political purges of the 20th century, from the repeated mass arrests in the early Soviet Union (Pipes 1994;Siegel 1954) and Chiang Kai-shek's repression of the Communists in Shanghai in 1927 (Fenby 2003) to purges in Saddam Hussein's Iraq (Aburish 1999), Hafez al-Assad's Syria (Van Dam 2011), or Ethiopia under Mengistu (Wiebel 2015).
Drawing on these cases, an analytical literature has tried to understand why and when dictators carry out purges of the political or military elite. Keeping elites in check is a central concern in modern autocracies, as dictators are more likely to be overthrown in an elite coup than they are to be ousted in a mass uprising (Svolik 2009). Across countries, elite purges are more likely to occur in non-democratic states (Carey 2010;Hill and Jones 2014), when the elite is temporarily weakened (Sudduth 2017), or when the country is still relatively poor and cooptation is too costly (Bove et al. 2017). In addition to coup threats per se, the intensity of political purges also seems to depend on a number of additional strategic concerns, such as rules of political succession (Egorov and Sonin 2015), the existence of multiple parallel threats to the ruler (Bueno de Mesquita and Smith 2017), the need to enforce top-down accountability (Montagnes and Wolton 2019), or the availability of information (see Gehlbach et al. 2016 for an overview).
This literature on purges is illuminating, but it suffers from several shortcomings. While the literature provides a wealth of theory and evidence to explain why a dictator might launch a purge, most studies lack a clear explanation for why certain elites are arrested, and others not. It also does little to help us explain patterns of elite repression in 'normal' times, outside of well-defined purges. We seek to address these questions in this paper.
In addition to the literature on purges, a growing literature on anti-corruption campaigns has tried to understand why and when authoritarian regimes undertake mass arrests of political elites to fight corruption. Conceptually, one might expect a trade-off between tolerating corruption as a tool to co-opt the elite (Gandhi and Przeworski 2006) and the economic harm being done by excessive elite predation (Acemoglu et al. 2004;Dawisha 2014;Pei 2006Pei , 2016Rochlitz et al. 2020). An additional factor might be a desire to increase regime legitimacy in view of excessive elite corruption, independent from corruption being perceived as a direct threat to the regime or not (Burger and Gitau 1958;Dai 2018;.
While fighting excessive corruption may be an important motivation behind recent, intense anticorruption campaigns in places like Russia, China, Saudi Arabia, Vietnam and Iran, it is likely that these campaigns-at least in part-are also a cover for more typical purges of elites by the author- Studies of purges and anti-corruption campaigns, often in China, make it clear that at least some arrests are politically motivated. Ultimately, however, it is very difficult to discern the proximate causes for arrests: fighting corruption and intra-elite political fighting frequently blur together. In this paper, we do not attempt to disentangle these two as such. Rather, we seek to explore the political factors that make some officials more prone to being arrested, whatever the proximate cause. In other words, we are agnostic on whether a particular arrest is driven by true corruption, political infighting, factional struggle, or some mixture of these factors. Instead, we analyze the factors that protect elites from prosecution, focusing attention on one factor in particular: personal popularity.

Popularity and Repression
Our main focus is on how the personal popularity of subnational officials affects their likelihood of undergoing criminal prosecution. Many scholars have argued that autocrats fear competent officials who might have the skills, resources, charisma and/or following to challenge their rule (Egorov and Sonin 2011;Zakharov 2016). Popularity may be a particularly threatening form of 'competence.' Dictators may fear that popular leaders will use their influence to plot a coup, rally the masses, or So, the idea that popular officials are particularly vulnerable to repression due to their popularity has an substantial theoretical and historical weight behind it. But there is a flip side to such arguments. First, it has long been recognized that competent officials are an asset as well as a liability for autocrats. Dictators need competent officials who can get things done: implement policies, grow the economy, attract investment, mobilize the masses, win elections, outsmart rivals and so on. As Huber and McCarty's (2004) model of bureaucratic delegation shows, politicians find it hard to secure the implementation of their preferred policies when bureaucratic capacity is low. 2 Another reason personally popular officials may be less vulnerable to repression than unpopular officials is that popular officials can serve more effectively as political surrogates for the regime. Such elites' popularity is an asset that makes them well-positioned to mobilize the masses, quell protest, defend the regime, and generally implement regime policy. They can use their influence over other elites to help shore up elite support, or they may use their public popularity to help win mass support for the regime. This latter factor may be particularly important in electoral authoritarian regimes, where subnational officials are often tasked with mobilizing votes on behalf of the regime (Blaydes

2010; Buckley and Reuter 2019; Reuter and Robertson 2012).
There is a second reason that popular officials may be more protected from prosecution. If an official is popular, jailing them can produce a popular backlash, especially if voters perceive that the 2 Empirically, it is clear that dictators do sometimes privilege competence over loyalty. China's Target Responsibility System, in which merit-based performance criteria are used in the evaluation of cadres, is one such example (Edin 2003;Rochlitz et al. 2015;Whiting 2004). Similarly, the East Asian developmental states have been lauded for their ability to delegate economic decision-making to economic technocrats (Doner et al. 2005;Van Dam 1999).
In Russia, Reuter and Robertson (2012) find that governors who sat atop strong local machines were more likely to keep their jobs in the mid-2000s than weak governors. arrest was politically motivated. This may be especially true for elected officials. Popular elected officials enjoy a mandate to govern and voters may retaliate against the regime if regime officials subvert their will by removing an elected official in handcuffs. Such backlash may come in the form of declining approval ratings for the regime, electoral punishment, or most worryingly for the regime, popular protest. 3 In Russia, mass protests have sometimes erupted in support of arrested mayors. For example, the 2013 arrest of Yaroslavl mayor Evgenii Urlashov sparked large-scale demonstrations that drew together disparate opposition forces. Prominent national opposition leaders from Moscowincluding Ilya Yashin and Irina Prokhorova-traveled to Yaroslavl to address the protestors. 4 Three years later, small scale protest in support of the mayor were still a semi-regular occurrence. 5 In sum, there are two compelling sets of arguments about the relationship between an official's popularity and their chance of being repressed. We argue that the application of a particular logic will depend on two factors: 1) the importance of public opinion to the regime and 2) the type of official.
Beginning with the type of official, the argument that dictators will repress popular elites applies better to national-level elites than it does to subnational officials. Aside, perhaps, from the leaders of capital cities and large provinces, subnational officials lack the platform and audience to effectively challenge the regime on their own. Moreover, subnational officials face significant collective action problems if they attempt to mount a unified challenge (see, for example, Shvetsova 1999). Thus, leaders should be less concerned about the threat posed by popular subnational officials.
The relative importance of public opinion for regime stability also matters. While many things determine the relevance of public opinion for regime stability, one important distinction is the difference between regimes that hold multi-party elections and those that do not. In electoral authoritarian regimes, the regime cannot rely solely on coercion and fraud to win elections. Rather, such regimes need millions of voters to come to the polls and vote for the regime (or at least not vote for the opposition). This has two implications for our argument. First, the backlash caused by arresting popular officials is likely to be more costly in electoral authoritarian regimes than in single-party dictatorships.
Second, because regime leaders in electoral authoritarian regimes rely heavily on subnational officials for vote mobilization, sacking a popular local leader will be particularly damaging for electoral authoritarian regimes (Blaydes 2010;Reuter and Robertson 2012;Reuter 2013).
Thus, when it comes to repression of subnational officials in electoral authoritarian regimes-the 3 Instances of arrests of popular politicians leading to mass protests can be found in many electoral autocracies and hybrid regimes. In Kyrgyzstan, the arrest of popular politician Azimbek Beknazarov in January 2002 sparked mass demonstrations which led to the resignation of the government, and severely weakened president Akayev (Olcott 2005, page 134 empirical scope of this paper-we argue that popular officials will be less likely to suffer arrest. Since electoral authoritarian regimes are by far the most common type of contemporary autocracy and subnational officials are frequent targets of repression in these regimes (Freeman 2018;Gel'man 2016;Kwong 2018;Turchenko 2017), we think this perspective has the potential to generalize to other country cases. In the sections below, we examine and test this argument in the context of Putin-era Russia.

Data
Our study focuses on one prominent example of an electoral authoritarian regime, the Russian Federation. A comprehensive biographical dataset covering the universe of mayors of Russian cities with a population over 75,000 from 2000 to 2018 forms the core of our analysis. This dataset is an extensive update of the dataset used in Reuter et al. (2016), which now includes 1051 unique mayors from 220 Russian cities. A team of research assistants collected a wide-ranging set of variables on these mayors: demographics, career histories, electoral outcomes and term-level details such as partisan affiliation. The data also classifies mayors according to their method of selection (i.e., elected or appointed). Thus, our data includes both appointed and elected mayors (glavy gorodov ), as well as appointed city managers, who serve in cities with dual executive systems. For convenience and clarity, we informally refer to any of these city executives as 'mayors,' even while their formal titles vary.
To this mayor-level dataset, we added information on criminal cases against these mayors. The authors worked with research assistants to identify all instances of a criminal charge being brought against these mayors between 2000 and 2018. We also gathered data on whether the mayor was arrested, convicted, or both after being charged, as well as data on the specific charge and the ultimate sentence. 6 All data was gathered for criminal investigations that occurred while the mayor was in office and, to ensure all relevant cases are captured, for investigations that occurred within 4 years of a mayor leaving office. Not all charges result in arrest or conviction: sixty-seven percent of charged mayors whose case concluded prior to the end of our data were convicted. Being arrested is more likely to lead to conviction: eighty-three percent of those arrested are eventually convicted. As discussed below, we drop from our main analysis cases where the mayor is exonerated. For brevity, though at the cost of some precision, we refer to all cases in our analysis as "arrests" even if they were not formally arrested (arestovan/a) Utmost care was taken to locate every criminal case against these mayors. Research assistants entered all 1051 mayors' names and relevant search terms (arrest, charge, court case, etc.) into Google and leading Russian search engine Yandex in order to gather this information, with every case being coded by two RAs independently. In any case, criminal cases against mayors are high-profile events, certain to make news. For mayors that are arrested, news stories about the arrest almost always appear on the first page of search engine results when searching just on the mayor's name.
Still, it is possible that a few cases escaped our attention, especially in smaller cities or in the early 2000s (more on this below). Nevertheless, we are confident in having captured the overwhelming majority of criminal cases. To the best of our knowledge, this dataset and the extensive qualitative data it draws from are the first time that comprehensive information has been collected on criminal cases against authoritarian elites in Russia or neighboring countries.
In Table 1, we can see that by far the most common charge brought against these officials is corruption or closely-related abuse of office. Economic crimes such as fraud make up much of the remainder. A handful of cases concern violent offenses (murder, kidnapping), and we exclude these from the analysis. Our main dependent variable is constructed as follows. We first exclude all cases where the charges were dropped or the mayors were exonerated. Our only exception to this is that we include 13 cases where the year of the charge coincided with the year that the mayor left office. We make this exception because close examination of these cases revealed that the charge was likely a way to induce the mayor to resign, and the charges were later dropped as a condition of this. We also exclude charges that were filed more than one year after the mayor leaves office. On the other hand, we include charges that occurred just one year after leaving office, because many cases are begun as ways of removing mayors from office, but they are only formally charged after the mayor is removed or resigns. Finally, we also drop all years prior to 2002. The quality of information for this period is much lower, so we cannot be fully confident that we have captured all arrests in 2000-2002.
As shown in the online appendix, results are robust to relaxing these coding rules in various ways.
The data contain some interesting descriptive patterns. Figures 1 and 2 show the geographic distribution of the cities in our dataset, with darker coloration of the shapes the more arrests occurred in that city. As one can see, these arrests are widespread across Russia. There do not appear to be obvious geographic patterns. there was significant discussion among political commentators about an uptick in repression against regional elites (Hale et al. 2019). This pattern is borne out by our data and arrests have remain elevated since then. A second, less recognized pattern, however, is the uptick in arrests that occurred   Table 2 is coded in the following way. All appointed mayors are coded as pro-regime, as are elected mayors who run with a formal United Russia (UR) party affiliation. Elected mayors who ran with an opposition party affiliation are coded as oppositional. Independents who ran against UR candidates are also coded as oppositional. All other independents are coded as such. Of course, in Russian local elections, independent candidates are sometimes pro-regime, even if they lack a formal UR affiliation. Thus, this intermediate category contains a mix of pro-regime candidates, true independents, and some undetected opposition candidates. In our models below, we analyze a different version of this variable that attempts to separate true independents from pro-regime independents.
not purely about targeting regime opponents. Table 2 shows that popularly elected mayors are nearly three times more likely than non-elected mayors to be arrested. This makes some sense, since appointees who are anointed by the regime are less likely to be targeted with repression. This variable, however, is highly collinear with regime affiliation, so it makes more sense to examine these relationships in a multivariate setting, which we do in the following sections.  Table 2 also examines appointed mayors in cities with dual executives. In these cities, a city manager, selected by a local commission and local legislature, is responsible for issues of governance (khozyaistvo), while a head of city (glava goroda) is chosen from among the deputies in the legislature to serve in a mostly ceremonial function as the chief political executive. The table shows that city managers are about three times more likely to be arrested than are formal heads of cities (glava goroda). There are two possible interpretations. First, this could be evidence that actual corruption plays a role in arrests. City managers deal with the details of budgets, state contracts, and regulation, giving them much more opportunity than city heads to abuse office. Alternatively, this could simply be evidence that city managers have the real power in these dual executive cities.

Empirical Approach
This section describes how we set up the data for testing our argument. After combining the above mayor-and arrest-level information, we transform our dataset into mayor-year format whereby the officeholder on January 1 is counted as the mayor for that calendar year. Time-invariant mayor characteristics and term-varying characteristics like electoral margin of victory in each term are included in the appropriate years, as are arrest events and a set of city-and region-level measures. The de-pendent variable, Mayor Arrest, is the arrest variable described above. It takes a value of 1 in years when a mayor is arrested, 0 when they are not. For instances in which the mayor is arrested after leaving office, the variable takes a value of 1 in his or her last year in office.
In order to examine how popularity affects a mayor's chances of arrest, we use data (for elected mayors) on their margin of victory in their most recent election. Unfortunately, public opinion data on the popularity of Russia's mayors is unavailable, so we use electoral margins as a proxy. Thus, we are only able to examine this hypothesis for the subset of mayors who are elected.
Does vote mobilization help protect mayors from arrest? Russian subnational officials are often enlisted to help the regime mobilize votes (Beazer and Reuter 2019;Reuter and Robertson 2012).
It may be that those who perform well at mobilizing votes for United Russia are more protected from arrest. We use data on the share of the vote received by United Russia in that city during the most recent regional elections held under the sitting mayor's tenure. This leads to a large reduction in observations, since we are not able to include data for the years that come after a new mayor enters office but before the next regional election.
In order to get a fuller picture of how mayors' personal characteristics beyond personal popularity may affect their probability of being arrested, we also examine mayors' professional backgrounds. To do this, we rely on a series of binary variables drawn from mayors' career histories. 9 For each type of professional history, we use two dummy variables: one that is equal to one if the given mayor ever worked in a given professional sphere and a second that is equal to one if the mayor's most recent place of work was in a given professional sphere.
First, we investigate whether a professional history in the security services protects mayors from arrest: such connections to the FSB or Interior Ministry might help shield mayors from persecution.
Second, we investigate whether work connections in the federal government help protect from arrest.
Third, we investigate whether a professional background in the regional administration helps protect mayors from arrest. If, as some observers aver, arrests are driven primarily by political conflict between mayors and regional administrations, then professional connections in regional government may help mayors avoid prosecution. Indeed, if a mayor's most recent place of work is the regional administration, it is likely that the mayor is a direct client of the governor. Finally, we also investigate whether mayors whose background is outside the city ("outsiders") are more to be arrested with a dummy variable equal to 1 if the mayor's most recent place of work is outside the city. 10 One narrative about arrests in Russia is that they arise when mayors get involved in conflicts with the local elite.
Outsiders are more prone to such conflicts (Kynev 2018). In all of our main model specifications, we include a set control variables, including: a dummy variable indicating whether there was turnover of the regional governor in that year, a measure of the the non-Russian ethnic composition of the region, the natural logarithm of the city's population, and a variable capturing the level of regional democracy 9 As noted above, we are able to hold constant personal connections to the top leadership because almost no Russian mayors are embedded in national-level factions or have personal links to Putin. 10 In our data, 17% of mayors are outsiders.
(the Petrov-Titkov measure for the region in the most recent year available, Petrov and Titkov 2013).
For all models, we run logistic regressions and include mayor tenure in linear, quadratic, and cubic transformations to produce an equivalent to survival analysis (Carter and Signorino 2010).

Results
How do Russian mayors' personal characteristics like popularity and work experience, the extent of their connections to others in the regime, and performance in achieving regime goals affect their probability of facing repression? Table 3 shows the main results from our analyses. First, we find that work experience in various government structures appears to have little effect on a mayor's chances of being arrested. In Models 1 and 2, we can see that none of the indicators of government or security service connections are significant. Work experience in the private sector appears to increase the risk of arrest, but this result does not hold in Model 2, which looks only at the mayor's most recent work experience.
There are several interesting results on the control variables. We find no evidence that insider mayors are more (or less) likely to be arrested, and we find no evidence that governor turnover is associated with more mayoral arrests. We find relatively robust evidence that arrests are more likely in ethnic republics. In some specifications, city size has a positive effect on arrest probability, with mayors in large cities being more vulnerable. 11 Mayors in large cities are more likely to enter into conflict and pose a threat to regional governors, which may explain this finding. Finally, we find that mayors are slightly more likely to be arrested in more democratic regions.
In Models 3-5 we examine how regime affiliation and being elected (as opposed to appointed) affect the probability of arrest. Model 3 indicates that appointed mayors are significantly less likely to be arrested. However, since almost all appointed mayors are affiliated with the regime, this variable is highly collinear with partisan affiliation. In Model 4, we look at the effect of regime affiliation on being arrested using a dummy variable equal to one if the mayor is regime affiliated and zero if he or she is not. 12 Consistent with the descriptive results in Table 2, pro-regime mayors are significantly less likely to be arrested. The effect size is very large. The predicted probability of a pro-regime mayor being arrested in any given year is 2%. For oppositional mayors, the probability increases more than threefold to 7%.
In Model 5, we enter both Non-Elected and Regime Affiliated in the same model. The latter remains significant, while the former drops just below statistical significance. Thus, we are confident that regime affiliation has an effect on the probability of arrest, but we are less certain about the effect of being elected.
In Table 4, we examine how regime vote mobilization and mayor popularity affect the probability of arrest. In Models 1 and 2, we can see that getting out the vote for United Russia has little impact on a mayor's chance of avoiding arrest. This result holds while controlling for regime affiliation. In Model 3, we find strong evidence that winning one's election by a large margin helps protect a mayor from arrest. The coefficient on Mayor Margin of Victory is statistically significant and substantively large.
A mayor who won his or her election by 77 percentage points (the 90th percentile on this variable) has just a 1.5% chance of being arrested in any given year. However, if a mayor won their election by 8 percentage points (the 10th percentile in this data), the predicted probability of arrest in any given year more than triples to 5.2% In Models 4 and 5 we explore this result further by splitting the sample according to regime affiliation. Model 4 reproduces Model 3, but only uses the subset of mayors who are regime affiliated.
Model 5 shows mayors who are independent or oppositional. The coefficient is negative in both models, but is only significant for non-regime mayors. The fact that the result is significant for nonregime mayors suggests that fear of popular backlash drives at least some of the results on mayor popularity. If popular mayors were not arrested solely because the regime wanted to rely on their political capital, then we would expect that this result would only hold among pro-regime mayors. But the fact that it holds among mayors not aligned with the regime suggests that there is more to the story. Since the regime cannot reliably draw upon the political machines of these mayors, it stands to reason that the regime's fear of popular backlash is what is driving this result.
The weak results among pro-regime mayors is peculiar. One interpretation is that a mayor's margin of victory is a much noisier signal of regime popularity among United Russia mayors than it is among non-regime mayors. For pro-regime mayors, their margin of victory is jointly determined by their own popularity and the use of pro-regime administrative resources. For non-regime mayors, the margin of victory is a cleaner measure of their popularity. In

Conclusion
What protects subnational political elites in autocratic contexts from being repressed by their own regimes? Where scholarly literatures on purges and anti-corruption campaigns find that being connected to the wrong network can put one in the line of fire, we uncover a personal asset that can help shield from danger. Our analysis shows that being popular or having political capital to spare is the only factor that consistently protects Russian mayors from repression. At least in this context, who you are matters more than whom you know or what you have done.
In Russian city halls, neither pedigree, friends in high places, nor dutiful service done for the regime's benefit are exculpatory when a target is drawn on your back. We muster evidence from a new dataset covering all mayors in large Russian cities and all cases of arrest or prosecution of these mayors to show that the only reasonably reliable protection a municipal leader has against repression is to be fundamentally politically strong. Strength may come in the form of personal brand, ability to lead your constituents in your defense, or highly-valuable political capital that you can apply to your own reelection.
Our analyses paint a more complicated picture of elite repression than may be found in much of the work in this area. While purges and similar attacks on officialdom may be largely about intrapalace (or at least intra-regime) intrigue and informal ties in some contexts, this is not universal.
Nor are autocracies solely concerned with their cadres performing well on regime-assigned tasks like getting out the vote. They also struggle with managing-selecting, controlling, and eventually perhaps eliminating-intrinsically powerful individuals within their ranks. In this way, we show how efforts to establish meritocracies around principles of either technocratic performance on the one hand or loyalty on the other can fall flat in the face of personal ambitions. Personal cachet matterseven relatively low down within autocratic regime structures and even if higher-ups may wish it were not so.
This is not to say that whom you know or what you have done for the regime is irrelevant. Nor is personal political capital all-powerful. Outright repression of one's own political elites is a quite extreme step for a regime to take. Given that autocracies very often seem willing to take that step, the fact that they may hold back when their potential target is personally popular highlights the importance of that feature. Future work on elite repression can explore the mechanisms by which popularity acquires this power. Scholars can also work to identify heterogeneity in how connections, popularity, and relative 'rank' within authoritarian regimes interact to determine who is promoted, who is pushed out, and who is ultimately punished.

Online Appendix
Descriptive Statistics

Alternative Codings
This table examines how our main result is affected if we use different operationalizations of the dependent variable: mayoral arrest. Model 1 is the baseline. It is a replication of Model 3 from Table  6 in the main text. To remind the reader, this variable includes all charges against mayors with the following exceptions: • We exclude all cases where the charges were dropped or the mayors were exonerated.
-However, we include 13 exonerations where the year of the charge coincided with the year that mayor left office.
• We exclude charges that were filed more than one year after the mayor leaves office. In Model 2, the dependent variable only counts charges that occur while the mayor is in office. Model 3 uses a dependent variable that is similar to Model 1, but slightly more restrictive: it only counts mayors who are charged while in office. Finally, Model 4 uses a DV that is the same as Model 5 in Table 5, but it excludes the 13 exonerations where the year of the charge coincided with the year that the mayor left office. Results on Mayors Margin of Victory remain statistically and substantively unchanged across these specifications.

Bremen Papers on Economics & Innovation #2013
Responsible Editor: Prof. Dr. Christian Cordes All rights reserved. Bremen, Germany, 2020 ISSN 2629-3994 The working papers published in the series constitute work in progress circulated to stimulate discussion and critical comments. Views expressed represent exclusively the authors' own opinions and do not necessarily reflect those of the editors.