Afterword to the past elections
On December 17, a report by the European Platform for Democratic Election (EPDE) was released.
In it, several paragraphs are devoted to statistical methods for analyzing elections. Below is their translation with comments from the researcher who performed the analysis.
The difference between majoritarian and proportional votes.
A statistical analysis of the results of all polling stations showed that in some of them there is a significant difference between the number of votes cast for the “Georgian Dream” party and for their majoritarian candidate. The initial hypothesis of the method is that the difference in votes for a party and for their majority candidate should not be significant. The analysis showed that this hypothesis was confirmed in almost all 3847 polling stations. However, significant deviations were also noted. It is important to note that these anomalies were found precisely in those 3 constituencies in which the majority candidates of the Georgian Dream (GD) party received the minimum advantage to pass in the first round:
- Region No. 13 (Marneuli – Gardabani). The majoritarian candidate from the GM party received 31,078 votes (a result of 51.4%), while the party received 29,171 votes. The difference is 1907 votes.
- Region No. 25 (Martvili-Abasha-Tsalenjikha-Chkhorottsku). The majority candidate from the GM party received 31,973 votes (53.5%), while the party received 29,761 votes. The difference is 2212 votes.
- Region number 29 (Kobuleti – Khelvachauri). The majoritarian candidate from the GM party received 22,369 votes (53.4%), while the party received 29,171 votes. The difference is 1673 votes.
A detailed analysis of the results at the polling stations in these regions gave the following results:
Region # 13 (Gardabani) at station 13.22.69 (Marneuli) GD received only 216 votes, while their candidate received 450 votes (a difference of 234 votes). An important fact is that the official protocol contains 252 votes for the majoritarian candidate, but it was amended on November 1 and the corrective protocol contains the number 450. A similar difference was found in polling stations number 13.22.03 (the GD party won 249 votes, and the majoritarian candidate got 330 votes), 13.22.34 (the GD party got 254 votes, and the majoritarian candidate got 327 votes) and at 13.22.37 (the GD party got 246 votes, and the majoritarian candidate got 365 votes).
In the region №25 at the polling station 25.65.02 the GD party gained 398 votes, and their majority deputy got 506 votes. A similar situation is observed at the 25.65.01 polling station, where the GD party gained 310 votes, and their majoritarian candidate received 431 votes. Likewise, in constituency 29, at precinct 29.81.55, the GD party gained 226 votes, the majority MP got 337 votes.
Correlation between the turnout and the result of the GM game (Kisling-Shpilkin method)
The Kisling-Shpilkin method, although rather crude, is one of the most effective methods for analyzing elections. His original assumption is that the joint distribution of turnout and game result is consistent with the normal distribution. Exceptions are, of course, not direct evidence of fraud, but should be scrutinized with more sensitive methods.
We applied this method to the proportional part of the first round. Preliminary, to exclude the influence of regional characteristics, we divided the results into three groups: Tbilisi (regions 1-8), East (regions 9-16), West (regions 17-30). The diagram below shows the results of the GD party (blue dots) and the opposition (red dots), depending on the turnout (one dot corresponds to one polling station). Trend lines were plotted for the distributions. Without going into the details of the method, which is well described in the specialized literature, its results can be estimated by the slope of the trend line. With a normal distribution of votes, the trend line is horizontal. If the line has a positive slope (directed to the upper right corner), then the hypothesis of normality is violated, which requires additional research.
With the naked eye, you can see that the trend lines for the GM batch in all three groups have a positive slope. As expected, due to the increased monitoring of sites in the capital, the trend line in Tbilisi is almost horizontal. At the same time, both in the east and in the west, there is a significant tilt towards the upper right corner.
Based on the many tests carried out by the Kisling – Shpilkin method for the countries of the West, Eastern Europe, Russia and Georgia, we can conclude that the hypothesis of the normal distribution works very well in countries with well-established democratic institutions. Moreover, in Western European countries, the opposition is usually in the lead at polling stations with a high turnout. The opposite picture is observed in countries with democracies in transition.
Based solely on statistics, it is impossible to draw final conclusions about the external impact on the election results. However, as we noted above, statistical analysis methods are a good tool for identifying the most suspicious sites. We manually studied many of them and checked the data of observers from opposition parties.
The strangest section that the algorithm produced is 28.79.19. On it, the total of votes is almost double the number of registered voters. It should be noted that in this area the opposition managed to challenge the results in court and now the protocol is balanced.
If we assume that deviations from the normal distribution are the result of external interference in the electoral process, then the algorithm estimates the results of this intervention at the level of 2.6 – 3.6%. Taking into account the official result of the “GD” party (48%), after “normalization” there remains 44-45%, which is consistent with the result of the parallel calculation of PVT.
In total, in the first round, the passage of 6 to 8 deputies of the GD party remains a serious question. It is important to remind that the opposition did not take part in the second round, as a result of which the GD won in all constituencies of the second round. However, based on the results of the first round, the Georgian Dream candidate had chances of winning only in 6 districts.
Thus, if the found anomalies are the result of direct interference by the ruling party, then this had a significant impact on the result of all elections, i.e. otherwise, the GD party would have been defeated.
Tamaz Khunjua, GSAC