Dream statistics

Were claims of fraud in last month’s Georgian election widespread enough to have the vote annulled?

When Georgia’s pro-Russian ruling party Georgian Dream received 53.92% of the vote in the country’s parliamentary elections on 26 October, opposition parties refused to accept the results and accused the authorities of wide scale voter fraud, claims that were backed up by multiple observers and electoral analysts. In an attempt to assess the veracity of the claims, Novaya Gazeta Europe decided to carry out its own investigation.

Warning signs

As soon as Georgia’s parliamentary elections were over, experts began to notice that the results deviated markedly from the distribution of votes and voter turnout typically seen in free and fair elections. In fair elections, voter turnout and votes cast in different polling stations should form a bell curve known as Gaussian distribution when plotted on a graph.

One of the first analysts to post his conclusions, electoral expert Roman Udot, found that the percentage of votes for Georgian Dream increased at polling stations with a higher turnout — i.e. the higher the number of voters, the higher the proportion of those among them who supported the ruling party, a correlation not seen in fair elections.

Similar conclusions were drawn by Georgian electoral analyst Levan Kvirkvelia, who showed that the distribution of votes cast in rural districts did not conform to Gaussian distribution, again, another potential marker of compromised electoral integrity.

Kvirkvelia added that approximately 400 polling stations showed “unusually high support” for Georgian Dream, noting that it was “virtually impossible” for such a sharp spike to be produced in a fair vote. “In hundreds of locations big and small, 100 voters show up, and all 100 vote for the ruling party — a scenario that is almost unreal in fair elections,” he wrote.

Statistically, the number of districts in which the winning party leads by a certain percentage should conform to Gaussian distribution. However, this method is not 100% reliable, and the polling anomalies pointed out by both Kvirkvelia and Udot could also be explained by regional differences, as research shows that it is practically impossible to distinguish regional voting patterns from falsifications.

This means that the election irregularities pointed out by experts may have been caused by innocuous factors, such as differences in voting partners between regions. Nevertheless, analyst Maxim Gongalsky told Novaya Gazeta Europe that there simply wasn’t enough data available to say that there had been no falsifications in the Georgian elections whatsoever.

To stuff or not to stuff

While it appears that some electoral fraud did take place in the Georgian election, it doesn’t seem to have been in the form of ballot box stuffing. To look further into this, we analysed Georgian electoral statistics using data and methodology advanced by electoral analyst Ivan Shukshin in his analysis of the vote.

Ballot box stuffing can usually be detected through a method popularised by physicist Sergey Shpilkin, in which the distribution of votes cast for a winning candidate suspected of fraud at polling stations with different turnouts is compared with the same distribution for other parties, if their votes are believed to have been cast honestly.

What is the Shpilkin Method?

To assess how fair the elections in Georgia really were, we employed the same methods we used after Russia’s fraudulent presidential election earlier this year to identify statistical anomalies in the results.

Shpilkin’s method reveals how many votes were added to the winner’s total either by stuffing ballot boxes or by rewriting the final tally. It compares the distribution of votes for the different candidates with turnout at each individual polling station.

If the elections were fair, the distribution of votes for the winning candidate and all other candidates should be in proportion across polling stations and should differ only in absolute value due to the different number of votes cast at each. However, as stuffing ballot boxes in favour of one of the candidates increases turnout, the proportion of votes going to each candidate is also affected.

To assess the extent of ballot box stuffing, the votes received by all the runner-up candidates must be added up and multiplied by such a factor so that their curve coincides with the leader’s curve. Anything that falls outside these two curves is considered “irregular voting”.

In recent Russian elections, the final vote tally for an individual polling station is sometimes completely rewritten by officials with no effort made to alter the number of physical ballots cast. To detect this type of fraud, experts look for polling stations where the winning candidate’s result or the turnout is a statistically improbable whole-number percentage, such as 80% rather than 79.61%.

Ivan Shukshin calculated that if anomalous votes were subtracted, Georgian Dream’s result would have been 49.25% rather than the official result of 53.92%.

In the Georgian election, no obvious inconsistencies were detected this way, as there were no whole-number anomalies to suggest that the final vote tally from each polling station had been rewritten, and a rise in the winning candidate’s vote haul was not accompanied by a rise in turnout. Therefore, if individual polling station results were indeed falsified in Georgia, it was done by other means than those used in recent Russian elections.

Unlike the anomalies detected in Russia, those in Georgia were found primarily at polling stations with a low voter turnout.

Nevertheless, the Shpilkin method still detected minor levels of ballot fraud, according to Ivan Shukshin, who calculated that if anomalous votes were subtracted, Georgian Dream’s result would have been 49.25% rather than the official result of 53.92%. Unlike the anomalies detected in Russia, those in Georgia were found primarily at polling stations with a low voter turnout.

That anomaly is most apparent when polling stations are divided into urban and rural categories. At rural polling stations, there was a minor deviation in vote distribution for Georgian Dream from that which would be expected in a free and fair election. By contrast, the curves almost completely coincided at urban polling stations, suggesting that there were almost no irregularities there, or none that could have been revealed using the Shpilkin method, at least. This analytical method also produced an actual result of 49% for Georgian Dream, the same figure given by Shushkin.

Another method for identifying electoral anomalies involves making a graph comparing polling station turnout with the winning candidate’s result. Free and fair elections should produce a Gaussian cloud of dots representing polling stations, with a dense centre that is fuzzy around the edges. However, in the Georgian elections, the distribution of votes at some rural polling stations may be considered “irregular” according to the Shpilkin method — these stations show up outside of the Gaussian cloud on the graph.

However, as do other methods of election analysis, the Shpilkin method assumes uniform electoral behaviour in differing regions of a country. As such, it isn’t possible to draw any definitive conclusions about electoral fraud by using the Shpilkin method alone, especially as anomalies in the Georgian election were observed with greater frequency at polling stations with a low turnout, rather than at those with a high turnout, as would be expected in elections in which ballot box stuffing was likely.

It is more likely that in different regions of Georgia the scale of electoral fraud varied.

Even in countries that enjoy free and fair elections, the “core” may differ from the ideal Gaussian model — this has been observed in Poland, Germany and Spain, for example. Therefore, additional evidence is required before any conclusions about the scale of electoral fraud can be drawn. In the case of this year’s Russian presidential election, for example, there is a strong correlation between turnout and result, a sure sign of ballot box stuffing.

However, Shukshin noted that in Georgian regions with a higher number of anomalous polling stations, observers also reported violations. We calculated that 72% of voters from anomalous polling stations voted in regions where, according to a report by Georgian NGO My Vote published by Transparency International Georgia, observers reported mass violations.

This means that the anomalies detected cannot solely be chalked up to differing electoral behaviour between regions. It is more likely that in different regions of Georgia the scale of electoral fraud varied, just as it did in Russia.

In many regions where election observers recorded violations, our analysis did not in fact detect significant anomalies, demonstrating the insufficiency of this method to detect all instances of voter fraud, meaning that the true scale of ballot manipulation may actually be greater.

Running interference

Electoral integrity in Georgia has multiple safeguards; the electoral register is posted at the entrance to every polling station, giving the name, address and date of birth of each member of the electorate, which makes stuffing ballot boxes with votes from non-existent members of the electorate a challenge. In addition, some 89% of the country’s polling stations are equipped with automated systems for ballot processing. The automatically processed ballots were then recounted manually.

For this reason, mass ballot box stuffing or rewriting vote tallies were hardly an option, something backed up by Roman Udot, who acted as an election observer, who told Novaya Europe that there were no irregularities in the counting over votes, but that other methods to affect the outcome such as vote buying and the exertion of administrative pressure might have been used.

Mass ballot box stuffing or rewriting vote tallies were hardly an option in Georgia.

“People were given an incentive to prevent them going to the polling station — they weren’t able to vote for Georgian Dream but weren’t able to vote for the opposition either. They had their passports taken away from them, supposedly so that the incentive could be transferred to them. In reality, others simply voted in their place.”

Assuming that votes were not cast by proxies for everybody who had their passports confiscated, this method would have created a low turnout at anomalous polling stations, as we observed in our analysis. Udot alleges that the scheme was implemented by high-ranking officials at the Central Electoral Commission.

In its report, My Vote wrote that a complex scheme for falsifying results was used during the election and that, as both Udot and Novaya Gazeta Europe reported in the run-up to the election, Georgian Dream loyalists confiscated documents from voters in order to put pressure on them. The ruling party also reportedly laid on transportation to take voters to polling stations. Additionally, there were cases in which voters were handed ballots on which the Georgian Dream candidate had already been crossed. Observers were also hindered in their work, and were even physically assaulted.

Ivan Shukshin, who also observed the Georgian elections, alleged that in many cases spoiled ballots were counted as votes for Georgian Dream, and that at some polling stations there were cases of people voting more than once.

A bump, not a landslide

Our calculations show that electoral fraud is unlikely to have given Georgian Dream more than a 5% bump to its vote overall, meaning that while there was likely voter fraud at play in the result, had there been none at all, Georgian Dream would still likely have won more seats than any other party in parliament, even though it got less than 50% of the vote.

Indeed, the party would have required a far bigger boost to its share of the vote to gain any meaningful additional power in the new parliament, with a constitutional amendment in Georgia requiring the support of two thirds of parliamentary deputies.

If these violations can be proven, there may indeed be strong grounds for holding a rerun of the elections.

Under Georgian law, at least 10% of the electorate must be registered at polling stations where voting tallies were declared void due to suspected electoral fraud for an election to be annulled. If the anomalies we identified in our analysis of the results were indeed violations rather than regional differences, and if a sufficient proportion of the electorate was registered at the polling stations in question, a rerun of the elections would be legally required.

However, electoral statistics can only be used to indicate the approximate scale of voter fraud. In order for any evidence to be admissible in court, violations recorded by observers at specific polling stations are required, and as over 10% of the Georgian electorate was registered to vote at polling stations where violations were recorded, if these violations can be proven, there may indeed be strong grounds for holding a rerun of the elections.