Tax Return as a Political Statement

Appendix A: Sources of data and definitions of variables

Appendix B: “Profession” and “Business Connection”

We have compiled an extensive data base with biographical data of the Russian governors. The bulk of the data has been extracted from the governor’s individual websites (for an overview, see Governors.ru) and other media sources, systematic reports on business affiliation of governors and their family members provided by Slon.ru and multiple other media sources.

Profession dummy: All governors have been subdivided according to their profession. We have defined profession as the employment a governor commenced after graduation and in which he worked until eventually entering politics. We were able to identify five professional groups: i) military; ii) economics, commerce and business; iii) civil servant; and iv) others. The first category includes governors who previously served in the Russian army and entered politics after their professional retirement. The second category comprises businessmen who previously ran their own companies or held influential positions in national business groups (e. g. board members). The third category includes bureaucrats who filled positions in public administrations (e. g. city administrations, ministries, public authorities). All other professions are grouped in category four (e. g. actors, journalists, and scientists).

Business connection dummy: This variable measures whether a governor has a strong affiliation with a private or public business group. Among others we have identified close business connections in cases where a governor was CEO, president, member of the board of directors, founder, shareholder or executive in a business group. Thus Victor Kress, the Governor of Tomsk, was a former board member of RAO UES, or Mintimer Shaimiev, the President of Tatarstan, was ex officio chairman of Tatneft. The business connection classification is based on the available information from our biographical data set, as well as on informal information from trustworthy sources (e. g. for Shaimiev see www.economist.com/node/15407883). Eventually, all governors with a business affiliation have been marked with “1”, while the remaining governors have been marked with “0”. The business connection dummy can be regarded as an extension of the profession category “business”, which was described above. Consequently, all governors who have been identified as businessmen are marked as having a business affiliation (“1”), with the exception of the cases when business was transferred to their wives (see below) or cases when the governors seem to have lost any direct links to their former assets (as in case of the Leningradskaya oblast, where governor Valeriy Serdyukov had been deputy CEO of the Vorkutaugol’ in the first half of the 1990s; but currently this company belongs to the holding group Severstal’ and no links between Serdyukov and the company seem to be present; the children of Serdyukov seem to be successful businessmen, however, their business is not directly connected to the previous business affiliation of Serdyukov). Beyond that we were able to identify a number of governors with other professions who also maintained close relationships to various business groups.

Business connection dummy for the wives of the governors: The information on the business connections of the wives of the governors was compiled using a similar approach, except with much less reliable data. We classified as having business connections all wives of the governors who are CEOs of profit-oriented companies (non-for-profit organizations are excluded) or hold a significant number of shares of other companies. In some cases governors transferred all their business activity to their wives after accepting the office (as supposedly happened in Primorski krai with the governor Sergei Dar’kin). In this case we assign the business connection to the wife of the governor, not to the governor himself. In the same way, the mayor of Moscow Yuri Luzhkov does not have business connections, according to our typology, but his wife does.

Appendix C: Anecdotal evidence

Appendix D: Selection bias

Difference of means between the regions where governors have and have not been replaced in the year of elections

Appendix E: Robustness checks

Appendix F: Outliers

An unfortunate problem of our data is that due to a relatively small sample and, at the same time, large differences between observations, outliers may be driving our results. To deal with it, we run six tests.

1. We exclude two observations with extraordinarily large income reported by the governors: Tver and Kaliningrad. In terms of the absolute value, the governors ruling in these territories (wealthy businessmen in the past) reported earnings about 10 times higher than the richest governor outside of these two regions.

2. We compute Cook’s distance and exclude observations where the distance exceeds the threshold, typically used to identify the outliers (we use both thresholds suggested in the literature: 1 and 4/Number of observations, but find no outliers according to the first threshold).

3. We compute the DFBETA statistics for the share of votes for Medvedev and, again, exclude observations, where DFBETA is above the threshold (1 or 2/squared root of the number of observations; again, the first threshold yields no outliers).

4. We compute standardized residuals and drop the observations where absolute value of standardized residuals exceeds 2 (the threshold suggested in the literature).

5. We exclude observations identified as outliers based on Mahalanobis distance, a measure suggested by Verardi and Croux (2009) as a better indication of outlyingness in case one has multiple clusters of possible outliers. In this case we had to estimate an equation with a single dummy equal to 1 for business connections for either a governor or his wife; if we include two dummies, as in the main specification, the algorithm computing Mahalanobis distance does not converge.

6. We exclude the 5 % and the 95 % percentiles of the distribution of regions according to the outcomes of the presidential elections of 2008. The sets of outliers differ from test to test, but the results are confirmed after the exclusion of outliers in all cases (the detailed estimations available on request).

Appendix G: Extreme bounds analysis

For the EBA we use the set of 52 variables: share of votes for Medvedev, age, age squared, tenure, tenure squared, dummy business connections, dummy business connections of the wives, professional background: military, professional background: bureaucrat, professional background: economist, salary of an average bureaucrat, regional education, ELF, log territory, log distance from Moscow, dummy Northern Caucasus, log population, dummy republic, oil and gas extraction, regional income, fiscal transfers, number of presidential visits, log expenditures for parliament, expenditures for parliament per member of the parliament, log salary expenditures for parliament, salary expenditures for parliament per member of the parliament, log budgetary expenditures for governor’s office, log salary expenditures for governor’s office, total administrative expenditures, survival index of the governor, experienced corruption, perceived corruption, number of children of the governor, dummy federal connections, dummy local origin, dummy experience in a regional office, dummy for governors never elected, number of electoral victories, number of appointments, dummy for governors studying outside their region of office, dummy Unity members, number of years governor belongs to United Russia, mobility of the governor and 9 indicators of institutional quality of the region out of OPORA survey. We then estimate regressions with all possible combinations of these variables by four (270,275 regressions), compute the Sala-i-Martin CDF(0) statistic (assigning equal weight to all regressions and assuming a normal distribution of estimators) and select the variables for which CDF(0) exceeds 0.95. We find it to be true for ten variables: share of votes for Medvedev, salary expenditures for the regional parliament, dummy Northern Caucasus, dummy business connections of the wife, dummy local origin of the governor, dummy federal connections of the governor, average salary of a bureaucrat, dummy republic, regional income and fiscal transfers. Thus, the share of votes for Medvedev is a variable robust to possible permutation of controls (CDF(0) is 0.9662).

Appendix H: Indicators of electoral manipulation

1. Voter turnout and votes for Medvedev: In each region of Russia, voting is organized within a number of electoral districts (up to 121). If we look at the voting outcomes in each of these districts, theoretically, if the voter turnout increases, the share of votes in favor of any candidate should increase proportionally. Thus, if for all electoral districts in a region one regresses the share of votes of a particular candidate on turnout, the slope coefficient should be approximately equal to the actual share of votes received by the candidate and the intercept should be close to zero. Electoral fraud in favor of a particular candidate should manifest itself in an over-proportional increase of votes in favor of this candidate in electoral districts with higher turnout: the idea is that the additional turnout reported is actually achieved through rigged votes, which are of course all in favor of the incumbent manipulating elections. Then the slope coefficient should be larger than the actual share of votes, and the intercept should be negative. Therefore, the regions with greater irregularities in voting behavior, associated with electoral manipulations in favor of the Kremlin candidate, should, if one regresses the votes for Medvedev on turnout using a sample of all electoral districts, have (a) a smaller intercept (i. e. a negative intercept, which is large in absolute terms) and (b) a larger difference between the slope of the regression line and the actual votes for Medvedev. These two variables for the presidential elections of 2008 by the GOLOS association (an NGO focusing on promotion of free elections). The respective variables are labeled “shift of regression line” (intercept of a regression of voter turnout on the share of votes received by Medvedev in 2008 among all electoral districts of a particular region) and “additional electoral support” (slope of the regression of voter turnout on the share of votes received by Medvedev in 2008 among all electoral districts of a particular region minus the actual votes received by Medvedev in this region). On average the intercept is negative and the difference between the slope of the regression line and the actual votes of Medvedev is positive (roughly equal to 30 percentage points), suggesting the presence of irregularities. Since both variables are highly correlated, we re-estimate our baseline regression, replacing the votes for Medvedev by each of the manipulation proxies separately.

   The analysis of the electoral manipulations is important from a further perspective. As has been described above, we interpret the low support for Medvedev at regional elections as a sign of low effort invested by the governor (due to his strength), but not as a sign of direct opposition to the federal government. Using electoral manipulations data, we can perform one more check of our interpretation. In the set of our regions, we looked for those where at least one of the following conditions holds: either the shift of the regression line is positive or the additional electoral support is negative, i. e. there is evidence of irregularities consistent with manipulations against Medvedev. In this case one could indeed consider low support of Medvedev as a sign of opposition from the regional governors. There are only four regions in our sample for which this condition holds; if we exclude them, the results of the paper do not change. For all other regions, however, we observe either no conclusive evidence of manipulations at all or evidence of irregularities rather consistent with manipulation in favor of Medvedev; hence the governors’ behavior clearly cannot be interpreted as that of opposition.

2. Distribution of turnout and votes for Medvedev: Another approach to identify the electoral manipulation is based on observation of the density plot of turnout or votes for the incumbent candidate across all electoral districts. Manipulations are typically done by bureaucrats who have to fulfill certain targets. These targets are typically set in percent of the votes (or of the turnout) and end with a zero or a five (e. g. a target is much more likely to be 95 % or 90 % turnout than, say, 93 % turnout). Bureaucrats falsifying elections will try to reach these targets, and therefore we will see an excessive number of districts in our sample where the turnout or support for Medvedev is a “round” number ending with a zero or a five. This is, indeed, what many papers (e. g. Nureev 2010) show for Russia and other countries with widespread electoral manipulations: probability density functions have “spikes” at round numbers, which is not the case for countries with advanced democracy. As a result, Kalinin and Mebane (2011) suggest counting the share of regions where the turnout (percent of population entitled to vote participating in elections) or the percent of votes for the incumbent in the total number of votes ends with a zero or five, as a proxy of electoral manipulation. We generate these proxies for each region in our sample for both turnout and votes for Medvedev, thus obtaining two further indicators of electoral fraud (unfortunately, we have to use the data for the territorial electoral commissions at the district and city levels (territorial’naya izbiratel’naya komissiya) rather than lower-level district and local electoral commissions, where the initial vote count happens (okruzhnaya izbiratel’naya komissiya; uchastkovaya izbiratel’naya komissiya), which means that our results for these variables should be treated with caution).

3. Electoral irregularities: A more general proxy for electoral irregularities was published by Oreshkin (2007). It takes into account all regional elections since 1995 and considers the following properties: (a) extremely large or extremely small turnout; (b) extremely large or extremely small share of votes declared invalid; (c) extremely large or extremely small share of votes “against all candidates” (this option existed in Russian elections, but was abolished in 2006); (d) extremely high support of particular parties (e. g. 100 % of votes for United Russia in a certain district) and (e) extreme differences between the votes for a particular party in a region and on average in Russia. While all these features could have been observed in any district at random, the indicator focuses on districts which are systematically characterized by these irregularities throughout various elections in 1995–2007. The indicator was computed for individual districts and then averaged across regions, ranging from 0 to 10,000.