IRREGULARITIES IN THE DECEMBER 1993 RUSSIAN ELECTION RETURNS Mikhail Myagkov California Institute of Technology Alexandr Sobyanin Lebedev Institute of Theoretical Physics August 30, 1995 _____________________________ The authors wish to thank all representatives of the Russiaþs Choice party who collected voting data, which are not officially available. We are indebted to Roderick Kiewiet and Peter Ordeshook for important comments and support. We also would like to acknowledge the financial support of the National Council for Soviet and East European Studies. Introduction The December 1993 Russian elections, which were held only two months after the dissolution of the Russian Congress of People Deputies, presented voters with four decisions: 1. Election of an independent deputy to the State Duma (the lower chamber of the legislature). One half of the Duma was to be formed on the basis of 225 single member districts. 2. A party list preferential vote. The other half of the Duma was to be elected in one national district through a party list, proportional representation system with a five percent cutoff level. 3. Choice of two candidates to the newly created upper legislative chamber, the Federation Council. Candidates to the Federation Council were running in 89 two-member districts, and each voter could (but did not have to) cast two votes. The 89 districts corresponded to the 89 subjects (regions) of the Russian Federation. 4. Approval or disapproval of a draft of the new Russian constitution, backed by Yeltsin, which, if accepted, would provide the president with enormous power. The results of the election were very different from what had been expected. The major surprise was the distribution of votes across the party lists for election to the State Duma. All pre-election polls suggested that voters supported reforms and, therefore, that Russiaþs Choice would win a plurality, if not a majority, of seats. Communists and agrarians did not appear to have significant support, and nationalists were below the five percent cutoff level (Shlapentokh 1995). Nevertheless, the final outcome gave a clear plurality to the Zhirinovskyþs nationalist Liberal Democrats. Democrats (Russiaþs Choice) and Communists ended up close to each other in second and third place respectively. A number of hypotheses have been offered to explain the outcome. Some authors have argued that polling techniques were poor, and that the results of public opinion polls can not be trusted to reflect voter preferences (Shlapentokh 1995). Others look to the effects of the electoral law, combined with the inability of democrats to form coalitions (Remington 1994). A third hypothesis is that massive election fraud exaggerated the difference between the outcome and pre-election polls. The basis for such claims is that election results have not been officially published, and voting data, unofficially acquired by representatives of the Russiaþs Choice, suggest various irregularities. A number of articles in Russian and foreign newspapers, motivated by þstrangeþ election outcomes argue that an official investigation should be launched to assess these allegations of fraud (Sobyanin 1994a, Sobyanin 1994b, Sobyanin and Sukhovolskii 1994, Salie 1994, Vizitovich 1994, Piatkovskii 1994). The influential newspaper þMKþ (Moskovskii Komsomolets) called the elections þa brick fallen on the head of Russian democracyþ (Sorokina 1994). Others argue that no fraud occurred since the Central Election Commission approved the results of the elections, and that the issue of fraud is a legal question and that þno mathematical speculations may imply such suspicionþ (Vedeneev and Lysenko 1994). The only official response to allegations of fraud from the office of presidential administration took the same line: þ... those people [Sobyanin and his group] used to analyze election results, then they started to implement some scientific methods. I doubt that their methods are correctþ (Filatov 1994). This paper investigates whether or not the hypothesis of massive fraud can be rejected on the basis of available data. Specifically, we look at these presumed irregularities and ask: do these irregularities imply election fraud, or can they be accounted for by other factors? We should state at the outset that, given available data, it is virtually impossible to conclusively prove or disprove the existence of election fraud. Instead, proceeding on the basis of circumstantial evidence, we show that the hypothesis of fraud is consistent with both observed aggregate patterns in the data and the incentives of key players, and that it provides the most likely explanations for these patterns. The paper is organized as follows. First, the place of election-related issues in Russian politics before and after 1985 is established. In particular, although the importance of elections and the formal rules dramatically changed, the administration of election machinery remained in the same hands. Second, we show a number of apparent voting irregularities in the data that were the source of suspicion as to fraud. Third, we introduce several alternative models that þexplainþ these irregularities. These models are not the only ones consistent with the data and therefore do not exhaust other possibilities, but we believe that the options considered are the most realistic possible explanations of what happened. Fourth, we present evidence in favor and in opposition to each model. Finally, concluding remarks state lessons that should be learned in the future with respect to election administration. Elections : Before and After 1985 Free popular elections were not of importance in Russia before 1985. The monopoly on power of the Communist Party ( Article 6 of the USSR constitution) allowed no other party or organization to offer competing candidates. Nominations were made by party officials, and the general public had little opportunity to influence the process. No more than one candidate was nominated for a seat, essentially no political campaigning was allowed, and no one could express views other than those officially sanctioned by officials. In this regime, elections were organized as a show of public support for the candidates of a þstable block of communists and non-party members.þ Responsibility for staging the show was in the hands of local party authorities, and official results always approximated 99.99% approval for the candidates of the Communist Party and 99.99% turnout. All of the above suggests that the electoral process was far from what free and democratic elections are supposed to be. Moreover, institutions filled by popular elections ( soviets ) played no significant political or executive role. The adoption of a new electoral law in 1987 for the USSR opened a new page in the process of democratization. One half of the members of the new Union Congress were elected in the first multi-candidate elections since 1917, and although the Communist Party kept control of over fifty percent of the seats, ordinary voters had the opportunity to choose between different candidates and to participate in the nomination process. Some degree of free campaigning was also allowed. A number of elections have been held since then. In the parliamentary elections in Russia in 1990 no seats were allotted to communist party or other organizations. All candidates were selected through multi-candidate, single- member district competition. The Presidential election in 1991, which produced six candidates, became the first in Russian history in which mass media were used as a campaigning tool. Finally, parliamentary elections in December 1993 brought a multi- party system into Russian political institutions. As election rules and laws became more democratic, as people secured the opportunity to exercise their right to vote , there remained one factor that had not been the subject of reform, and which concerns this paper. That factor is who control of the actual conduct of elections. Who were the officials responsible for supervising elections and counting ballots? Local authorities, it turns out, were still in charge of the process, as they were ten, twenty, or thirty years before. They have been þrunningþ the regions for many years, and control everything from regional policies and budgets to crop harvesting. The overwhelming power, retained by local authorities with respect to the electoral process, has been a cornerstone of elections in Russia for several decades. That power was robust against any political changes or reforms. Before 1991 regional authorities were officially called þsecretaries of oblast party committeesþ. After August of 1991 they became þheads of local administrationþ. Despite the fact that the Communist party was officially eliminated as a ruling power, virtually all local party bosses were reappointed by Yeltsin as heads of local administration. That meant not only the same style of ruling as before, but also the same personalities playing the game. Thus, even though the election rules have become more democratic, their implementation has remained unchanged. On the other hand, some changes have had differential impact up on urban versus rural areas. Observers at polling places, media coverage of the pre-election campaigns, as well as the relatively high degree of political activity of voters played an increasing role in many big cities. Those factors make it difficult for local authorities to entirely control elections. In contrast, most villages and small cities have not been touched by reforms, and, as many observers contend, it is those places that became a major source of voting irregularities during the December 1993 elections. The December 1993 Election: Actors and Incentives The Central Election Commission, in charge of running the elections, appointed 89 regional election commissions, which in turn appointed thousands of local commissions. Each region in Russia consists of a number of þrayonsþ (counties). In the past, all rayons had their own election commission. Those commissions were eliminated before the December 1993 election (later we discuss an explanation for this change). Local (sub- rayon) commissions were in charge of counting votes at precincts and they in turn reported the results to regional commissions. After the regional commissions received the results, they aggregated the data to the rayon level and prepared official reports for the Central Commission in Moscow. All commissions consisted of working members and observers, and each party could assign its representatives to work at the commissions as observers. Even though the executive structure of elections might look democratic, several features suggest the opposite. First, the working members of regional election commissions - the only people who had access to the actual ballots - were appointed by and worked under the direct control of heads of local (regional) administration. This fact is extremely important, because in 67 of 89 regions local þheadsþ were running for seats in the Federation Council. A total of more than 100 representatives of top local executive officials were elected to the Federation Council. Only eight of them lost. Second, election commissions were not required by law to make voting outcomes available to the public on any level of aggregation lower than the regional one. Even results aggregated on the regional level have never been published officially. Thus, the Central Election Commission released the outcomes of elections after aggregating the results to the national level. Only one of the 89 regions (Niznii Novgorod) published data at the þrayonþ level of aggregation. Third, in most regions members-observers of electoral commissions did NOT have access to the data until after they were aggregated to the regional level. Moreover, the Central Election Commission deprived its observers of the right to see the data until after the results were officially published. Although the above factors constitute evidence strong enough to raise the possibility of election fraud (it is difficult to believe that people would not take advantage of the opportunities presented to them), a consistent set of incentives should be established. As discussed above, local heads were the key players during the elections. What were their incentives? On one hand, almost all of them there running for the Federation Council. On the other hand, almost none of them was an elected official, but were instead appointed by the President. Therefore, their incentives had to have much in common with the incentives of their boss. Moreover, old Communist party traditions certainly suggested that the best way to show loyalty to the president was to prove that the regions were supportive of his interests. The only issue on the ballot vital to Yeltsin was the referendum on the Constitution, which hinged primarily on turnout. By law, the Constitution would be approved only if more than 50 percent of the voters voted þyes,þ and if turnout exceeded 50 percent. All public opinion polls suggested that the draft of the constitution was supported by a very high margin of probable voters. But the question of turnout was up in the air, making the final outcome uncertain. In fact, it would seem that every voter who voted against the Constitution acted irrationally, since the constitution had a much greater chance of rejection through the mechanism of low turnout than by explicit disapproval of the voters. Therefore the second incentive of local heads was to provide at least 50 percent turnout in their regions. Unusually high rates of regional turnout, then, might be a good indicator of election fraud. The election rules, together with incentives, suggest that the expectations of election fraud was plausible ex ante. In other words, the environment was not incentive compatible with running a clean election. In the following section we present the voting irregularities that make the above hypothesis plausible ex post. Voting Irregularities As we noted earlier, the lowest level of data aggregation officially available was the level of region (state), whereas the data we consider are aggregated at the rayon ( county ) level. The information offered here has never been officially published and remains officially unavailable. It was collected by a number of different sources in the regions who were close to local administrations. In total the elections outcomes are available in the form of 893 observations, which cover 31 regions and 46 percent of the Russian electorate. In the previous section we suggested that turnout was a key parameter of the elections. Here we want to suggest that there were significant irregularities in the correlation between turnout and election outcomes. Moreover, these irregularities are not present in regions where local heads were not running, or were running and lost. The notion of an irregularity of correlation is defined as follows. Let voting outcomes on an issue be described by variables Tfor , Tag and T (T=Tfor+Tag), where Tfor is the percentage of the total number of eligible voters who supported the issue, Tag is the percentage of the total number of eligible voters who oppose the issue, and T is turnout. In a regular case, T should be positively correlated with both Tfor and Tag . In other words, the higher turnout, the higher is the absolute numbers of people who support or oppose the issue. Thus we call election results regular if T is positively correlated with the two other variables, Tfor and Tag. In the next section we discuss conditions which must hold for this not to be the case. The data in Table 1 show the coefficients of correlation between turnout and percentages of total number of eligible voters who supported or opposed particular issues. Notice that in the regions where local heads either were not running or lost, all coefficients are significantly positive. In the regions where local heads ran and won, several coefficients are very close to zero. Moreover, this occurs on every issue. Thus we see that every increment in turnout added virtually zero þnoþ votes for the approval of the Constitution, for the block of democratic parties, and for candidates other than local heads. The same increment in turnout added votes against approval of the constitution, for the block of þanti-reformistþ parties, and for a local head. Notice that average turnout was much smaller in the first group of the regions than in the second group. In fact, in the regions where local bosses were not running or lost, average turnout was less than fifty percent - the threshold necessary for the constitution to be approved. Figures 1 - 7 provide visible illustrations of this data. All graphs portray turnout on the horizontal axis and either Tfor or Tag on the vertical. Figures 1 and 2 show absolute percentages of those who supported or opposed the constitution. Figures 3 and 4 illustrate different patterns for party lists. Figures 5 and 6 present typical Federation Council voting patterns for two regions were local heads won the elections. Figure 7 deals with the same issues as Figures 5 and 6, but for a region in which the head of administration lost the elections. Finally, unusually high turnout in rural areas looks suspicious for that time of the year. For the territories with essentially no roads, no buses, and no telephones, with very short period of daylight, and with the temperature falling below -30 (F), it is hard to believe that eighty percent of eligible voters were able to get to precincts by traveling more than 20 miles on average. On the other hand big cities, with much closer polling places and much better transportation, produced only 49 percent turnout. Models: Election Fraud or Something Else? Irregularities in the data suggest that in the regions were local heads of administration won a seat in the Federation Council, every increment in turnout resulted in the same increment in percentage of votes cast against the Constitution, against the block of democratic parties in the party list competition for State Duma seats, and for the local heads. For many observers, the obvious explanation of this pattern is ballot box stuffing. Local heads who ran for seats in the Federation Council and who administered the elections had a clear incentive and opportunity to increase the numbers of votes cast for them in the protocols aggregated to the rayon level. They were able to do this without danger of being caught since rayon election commissions had been eliminated. Also, they did not have to do it for all rayons. Instead, the numbers could be altered only for those rayons where nobody would be likely to check anything. It is not hard to guess that those rayons would be mostly rural. At the same time election outcomes on other issues were also altered to make vote totals consistent with the Federation Council returns. Moreover, by increasing the numbers for the Constitutional issue, the second goal could be achieved - - exceeding the 50 percent turnout level sought by Yeltsin. It is important to note that in the regions were local heads either were not running or just lost, average turnout was less than fifty percent whereas it exceeded 60 percent in the rest of the regions. If the presumption of ballot-stuffing is correct, then the data suggest that in most cases the extra ballots were marked þagainstþ the Constitution and þagainstþ the democratic block of Duma candidates. How can this supposition be explained? First, one should notice that it was easy to alternate numbers in only one column of the protocol, since the numbers had to be consistent across protocols. Second, according to pre-election polls, the constitutional draft was supported by more than 70 percent of the population. Thus if one decides to stuff ballots for the constitution, the final numbers may look unrealistic ( like 90% vs 10%). Third, if, as we discussed above, the numbers were mostly altered for rural rayons, there would be less notice of irregularities because local, rayon authorities were overwhelmingly opposed to reforms. The reason of such an attitude rests on the fact that in rural rayons land was basically the only valuable good that local bosses could control. It is also well known that one of the bills that democrats were trying to pass through the parliament would allow private ownership of land, thus, depriving rayon heads from their control over land distribution and management. This issue turned most of the rural rayon bosses away from the democratic camp. Thus, altering numbers in party lists outcomes by adding votes to an anti-reform bloc was also in keeping with their objectives. Now let us consider some alternative hypothesis that explain the preceding irregularities without presuming significant fraud. We discuss two possibilities. The first presumes deterministic turnout among pro-reform voters and random turnout among anti-reform voters. Let us assume that everybody who supported the Constitution and democratic reforms decided whether or not to vote early in the campaign, and that the numbers of those pro-reform voters who decided to vote divided by the total numbers of registered voters were constant across rayons and oblasts. In contrast, suppose the opposition voted randomly. Everyone against reform would flip a coin in the morning of the election day in deciding whether or not to vote. Therefore, the total turnout can be presented as a sum of a deterministic ( democrats) and a random (communists) components. This implies that any variance in the total turnout is due to the variance in the percentage of anti-reform persons who turned out to vote. Although this model can þexplainþ the correlation irregularities in the data, which are presented in Table 1, and assumes no fraud, it can not account for the increasing variance of votes in support of the Constitution as turnout increases. Second, it is unclear why people with different political affiliations should have different probabilistic voting strategies. Most importantly, it is suspicious that the model works for the regions where local heads won, and fails in the regions where local heads lost or were not running. A second explanation assumes that turnout and number of people who voted with the opposition are traditionally greater in rural areas. For example, if turnout in rural areas were 70 percent, and only 45 percent of voters supported reforms, while turnout in urban areas were 45 percent, and 70 percent of voters supported reforms , then the absolute support of opposition would increase with turnout, as the data suggest. The argument against this hypothesis, though, is illustrated in Figures 8-10. All the figures have percentage of rural population as the independent variable (horizontal axis). Turnout, absolute support for the Constitution and absolute opposition to the Constitution are dependent variables. Even though turnout seems to be correlated with the independent variable, the latter two parameters are not. This is inconsistent with the hypothesis. High variances (Figure 9 and 10) make the hypothesis even less plausible. Finally, this model also fails to accommodate the fact that voting irregularities are present in the regions where local heads won, and are not present in the regions where local heads lost or were not running. Concluding Remarks The main finding of this paper is the fact that election irregularities are present in the regions where local heads won the election and are not present in other regions. That constitutes a piece of evidence in favor of the fact that election fraud was the source of these irregularities. Although our analysis does not prove beyond the reasonable doubt the fact the crime was committed, in a democratic country such a piece of evidence could become a basis for a more profound legal investigation. Unfortunately this is not the case for Russia. The original ballots were destroyed shortly after the election, and the complete official set of protocols is still unavailable for the public. If the hypothesis about fraud holds, then what can be done to prevent such problems in future? Several suggestions can be offered. First, the local electoral commissions should be neither controlled nor appointed by local officials. Moreover, the whole organizational process should be separated from local authorities. This could be done in a several different ways. For example, those commissions can be formed using a jury selection process. In other words, local election commissions can be formed of people who are randomly selected by computer and are not in any way affiliated with local administrations. Those people would count the ballots and report the data to the central election commission and to representatives of political parties and electoral blocs. Second, observers should be allowed to witness the election without any restriction. Third, a law must be written to establish criminal responsibility for election fraud. Finally, election outcomes must be accessible to the public at all levels of aggregation. Everybody should have access to the data. Another way to prevent election fraud is to establish a systematic mechanism of control that would detect election irregularities if any occur. As an example, let us consider a hypothetical a group of observers that would cover a number of randomly preselected precincts (1-2% of the total number). It is clear that if those observers can insure fraud- free election at their precincts, and then the data from those precincts are aggregated, the results should not be significantly different from ones officially announced for the entire nation. On the other hand, if there is significant difference then an investigation should be launched to check for election fraud. Russia will hold next elections in December 1995. Local heads will once again run for their seats. Therefore, we soon have an opportunity to see whether some lessons have been learned, and what kind of new irregularities might result. References Filatov Sergey, Popahivaet Provokatsiei, Independent Newspaper, May 7, 1994 Vedeneev Yrii and Lisenko, þAleksandru Sobianinu - matematiku I grazdaninuþ, Independent Newspaper, July 23, 1994 Vedeneev Yrii and Lisenko Vladimir, þVibory 1993- uroki I alternativi, Independent Newspaperþ, June 28, 1994 Piatkovskii Aleksei, þVibori v zazerkalieþ, Stolitsa ( the capital) , #2(164), 1994 Sorokina Elena, þKirpich na golovu Rossiiskoi demokratiiþ, MK #107, June 06, 1994 Alexandr Sobyanin, þStatistiku ne otmenish resheniem TSIKþ, Independent newspaper, August 08, 1994 Alexandr Sobyanin, þMinesterstvo Demokratii: svoia ruka vladikaþ, Rossia, #17, May 4- 10 1994 Alexandr Sobyanin and Sukhomlinskii, þBili ili ne bili falsifikatsii na viborah 12 dekabriaþ, Independent newspaper, July 07, 1994 Salie Marina, þNomenklaturnaia Izbiratelnaia Sistemaþ, Independent newspaper, July 21, 1994 Visitovich Valerri, þCentrizbirkom-politicheskoe vedomstvoþ, Izvestia, May 04, 1994 Shlapentokh, þPublic opinion polls and December 1993 elections in Russiaþ, Public Opinion Quaterely, 1995 Remington, Thomas, þAnalysis of the effects of the electoral law on the democratsþ fortunes in December 1993þ, Manuscript, 1994 TABLE 1. Simple Correlation Between T and T(.). Regions where local heads won the elections Regions where local heads either were not running or lost the elections. Number of observations 768 125 Average turnout 61.73% 49.36% ISSUE CORRELATION with TURNOUT CONSTITUTION þForþ Constitution 0.071826* 0.80794 þAgainstþ Constitution 0.80477 0.74181 PARTY LISTS Block of democratic parties -0.011352* 0.42028 Communists 0.61617 0.66353 Agrarian Union 0.72214 0.53401 Liberal Democrats party 0.63575 0.76215 Women of Russia 0.18826 0.57296 FEDERATION COUNCIL þForþ Local Head 0.72012 - þForþ the Second Candidate 0.079175* 0.22715 þForþ the Third Candidate 0.094106* 0.48934 * Not significantly different from zero. 1A region in Russia is similar to a state in the US. 2Aleksander Sobyanin, one of the authors, was working in central electoral commission as an observer delegated by Russiaþs Choice, the pro-Yeltsin party. Protocols are the only official documents that contain information about election returns at any level of aggregetion. For example, there are precinct protocols, rayon protocols, etc. The data that we present in this paper were taken directly from copies of rayon protocols.