It’s raining men: How all-male voting “carousels” helped Georgian Dream hold on to power

RU

Traditional methods for detecting electoral fraud analyze anomalous voting at high turnout polling stations and mainly detect ballot box stuffing or other cases of outright falsification of the true count. None of those methods turned up significant red flags in the case of Georgia’s October elections.

However, the inadvertent release of data on the gender composition of voters by the Georgian CEC changed everything. Interestingly, the voters who cast their ballots in favour of Georgian Dream were disproportionately male. Election analysts used the gender imbalance, not the turnout, as the key to uncovering the fraud in this election.

Applying the methodology of renowned analyst Sergey Shpilkin to a pair of variables, The Insider was able to calculate the number of anomalous votes: approximately 300,000 in support of the Georgian Dream. The mathematical explanation for the anomaly is supported by eyewitness accounts of “carousel voting” by predominantly male groups.

Carousel organizers obtained genuine IDs — or at least their numbers — from real voters who had guaranteed they would not go to the polls. Carousel voters then used them to cast multiple ballots. The turnout figures showed no anomalies, as bribed ID holders did not come to the polling stations.

On Dec. 20, the OSCE released its final — inconclusive — observation report on the Georgian parliamentary elections, and both sides interpreted the document in their favor. However, statistical analysis of official CEC data can provide a more objective answer to the question of whether Georgian Dream's parliamentary majority is indeed legitimate.

It’s raining men? Hallelujah!

Georgia's election data is unique. The country’s Central Election Commission (CEC) only publishes only contestants’ votes from each polling station, without indicating turnout or the number of invalid ballots, thereby rendering most conventional methods of detecting electoral fraud inapplicable. However, additional data requested by independent observers was provided by the CEC. Statistics indicating the gender characteristics of voter turnout for each polling station were published as a result, providing unexpected opportunities for analysis.

The main evidence of vote rigging is found in the reported electorate’s gender imbalance, which skews decidedly male. This is a statistical anomaly. In the absence of electoral fraud, the average turnout among men and among women would be roughly the same.

The CEC's gender data immediately raised suspicions. Initially, the CEC reported that 961,751 women and 1,098,661 men had cast ballots — an incredible figure for a nation where women outnumber men by 200,000. In response to the reasonable surprise of local observers, the CEC, in a frantic attempt to cover up the scandal, hid the original data. However, the government body inadvertently issued a document with detailed data for each polling station. The second report attempted to mask the initial anomaly by adding 91,911 female voters and hiding 88,975 male voters, but analysts were not fooled.

The amended CEC document lists 3,088 polling stations, 65 of which do not contain gender data (52 overseas and 13 special polling stations). The remaining 3,023 polling stations are used as the basis of the present analysis. In total, 1,626,171 men and 1,863,223 women are assigned to them, of which 1,006,170 men and 1,052,112 women were recorded as voting.

Such a ratio of male to female voters is inexplicable. If male and female voters showed the same level of electoral activity at each station (the same number of ballots cast per 100 registered voters of each gender), a total of 913,584 male voters would have voted nationwide, instead of the recorded 1,006,170. In short, the CEC’s revised data still shows an anomalous surplus of 92,586 male votes.

Same-sex anomaly in votes for anti-LGBT candidates

Notably, the purported electoral enthusiasm of Georgian men varied widely by region. The distribution of Georgia's “hyperactive male voters” by district is shown in the infographic below, with the largest contribution being made by the Marneuli district (in the southeastern part of the country, on the border with Azerbaijan), where the number of “excessive” male voters reached 9,310.

The inexplicable zeal of Marneuli's male voters inspired The Insider to take a closer look at the possible causes of this surplus of men. Unsurprisingly, this region is notorious in Georgia and beyond for electoral violations: pressure on voters, beatings, and a confirmed case of ballot stuffing, which has become the subject of criminal proceedings.

Overall, anomalous levels of support for Georgian Dream were recorded at polling stations where implausibly high percentages of men turned out to vote. Was this a coincidence, or rather an indication of hidden electoral manipulation, brought to light by an indiscreetly published CEC document?

The Insider's calculations explained

The hypothesis that the anomaly in this election is related to the gender of voters determined our approach to analyzing the available data. We needed to understand how the results differ between stations with higher male activity and those with equal male and female activity. Here is the analysis.

The non-dimensional variable “gender deviation,” which characterizes the comparative relative electoral activity of men and women, is plotted on the horizontal axis.

A positive value indicates that men from the “male voter collective” came to the polls more actively than women registered there, while a negative value indicates the reversed trend. This variable is not sensitive to the ratio of men to women in the polling district, as it compares male activity only to registered male voters and female activity only to registered female voters. Simply put, the “gender deviation” statistic compares the turnout of each group, rather than the absolute numbers of votes cast. If, for example, a polling station located at a site such as a military base were home to100 male registered voters and 20 female registered voters, of which 50 men and 10 women cast a vote, the “gender deviation” would be zero, as the overall turnout of each group would be 50%.

Importantly, this variable is insensitive to possible objective differences in the political preferences of men and women, as well as to the potential objective difference in political activity between urban and rural areas.

The entire observed spectrum of “gender deviations” is divided into 100 equal intervals, and in each interval, the results for the stations within the interval are summarized.

The vertical axis shows the results of Georgian Dream and opposition parties, calculated as a percentage of the valid votes in each bin.

The size of the markers is proportional to the number of voters registered at polling stations. In other words, we can see the stations where gender imbalance made a major contribution to the final election results, and where only negligible, random fluctuations were observed.

The graph shows that Georgian Dream owes its victory to the anomalously disproportionate male voting area. Meanwhile, the opposition parties are decisively leading in the area of conventional “gender-balanced”, or “heterosexual” voting.

Mathematically, the CEC data suggests that for each additional male voter, support for the ruling party drastically increased. How could this have happened?

How did the fraud work?

IIn the experience of Russian observers, such a gender imbalance has been observed in cases of carousel voting that were detected using polling station footage published by the Russian CEC. In such cases, the imbalance can reach 70%, as in St. Petersburg, or 100%, as in Dagestan. But how was it organized in Georgia? Sadly, Georgian polling stations have no CCTV cameras.

The clue came from the second most “hyperactive” male region of Georgia, Gori, which showed a surplus of approximately 5,000 men. On Nov. 25, the Mtavari Arkhi TV channel broadcast the confession of a man called Gocha Chalauri, who shared how he and his associates repeatedly voted for Georgian Dream in the Oct. 26 elections:

“There were four of us in my car, all men, driving as part of a group of five vehicles. We traveled through Gori, from one village to another, and voted for Georgian Dream about 30 times each. In total, 120 or 130 cars like ours were driving around [the Gori region].”

Electoral fraud is a crime in Georgia, so Chalauri's words should be taken seriously.

Still, how significant was the damage? Estimates show that 500 individuals voting 30 times each would produce the apparent effect of 5,000 extra men provided that the “carousel” was one-third female and two-thirds male. Thus, a predominantly male group of carousel voters, casting ballots for Georgian Dream across several villages, could improve the results of the ruling party but would leave visible traces of “gender deviation” — something the CEC overlooked when sharing slightly doctored data with observers.

While this confession fully explains the observed phenomenon, it is still a daunting task to vote every 24 minutes all day long without a break. Therefore, the share of men among the carousel voters would have to exceed two-thirds if it were to leave the observable “footprint” of 5,000 excess men in Gori.

The second caveat to be made is that polling stations record the gender marker of the provided ID, not the gender of the actual voter. In theory, male carousel participants could have cast ballots for absent women, in which case the actual level of fraud would be higher than that which is observable thanks to the available CEC statistics.

Passports for rent

Investigative journalists from Georgia’s leading TV station Pirveli have produced an entire documentary about how voter identification documents are bought — or, more accurately, “rented” — for sums of 100, 150, or 200 Georgian lari (approximately $35 to $70). Alexander Crevaux-Asatian, a high-ranking member of the opposition United National Movement party, has corroborated these reports.

According to Georgian journalists, voters who “rented out” their IDs or personal data promised not to show up at the polls for the recent election. Eyewitness accounts and testimonies from those who took part in the fraud, captured on camera by Pirveli, suggest that this scheme was orchestrated by heads of local administrations and members of the Georgian Dream party.

In a separate Pirveli film, these passports and ID cards are shown ending up in the hands of «carousel» operatives — individuals who are sometimes, though rarely, caught on camera by election observers at polling stations. Unfortunately, as in Russia, “carousel” schemes remain invisible to observers in the field in the absence of official video records. It is nothing short of a miracle that the Georgian CEC allowed a gender data leak, which made it possible to uncover and estimate the scope of the scheme.

Fraud at precinct level

That said, “renting” a passport was often unnecessary. A simple scrap of paper with numbers on it was often enough for the polling station — so long as the individual was part of the scheme. The country’s largest election observation mission, My Vote, put out an official statement detailing the methods of fraud:

“Some of the voters who came to the polling station had a piece of paper with another person's personal number written on it, placed in their passport or stuck on their ID card. Also, the voter is able to vote through a photocopy of the ID card in the phone.

Fraudulent voters had allied registrars at the polling stations who: a) had to quickly enter the personal number written on the paper into the verification machine; b) and should not have been marked or should have been marked poorly.”

The CEC also introduced a new rule for this election without any prior discussion: precinct election commissioners responsible for admitting voters were no longer selected by lottery on election day. Instead, they were appointed when the presence of observers or other commission members was complicated. A new clause specified that “the absence of commission members does not hinder the lottery process.” This adjustment was reportedly introduced to accommodate new electronic scanning systems, introduced in Georgia for the first time.

As investigative journalists were able to uncover, these “registrars” were handpicked from lists sent by the CEC.

Registrars are the commission members responsible for verifying voters' plastic ID cards or paper passports, which then gives citizens access to the polling booth to cast their ballots. Depending on the size of the polling station, there may be anywhere from two to five registrars on duty. However, for the scheme to work, only one “complicit” registrar is needed to process the entire carousel.

At the vast majority of polling stations, registrars relied on an electronic voter database, eliminating the need for regional or alphabetical divisions in the voter lists.

Occasionally, the system would malfunction, and a voter would discover that someone had already cast a ballot in their name. These incidents sparked scandals, some of which were captured in the investigative documentary. However, for the most part, the scheme worked under the radar — and if it weren’t for the anomalous gender data from the CEC, the scale of the fraud would have remained a matter of speculation and suspicion.

Swapping vs. stuffing

Georgia's electoral system has not yet deteriorated to the levels seen in Russia, where ballots can be stuffed in bulk. For each “carousel” voter in Georgia, a rented or stolen identification number of an “abstainer” had to be entered into the system.

As a result, every fraudulent vote subtracted one vote from the rightful voter whose passport or personal identification number was “rented” — according to surveys, these paid abstainers, as part of the general public, were 60% likely to have voted for the opposition had they turned up to the polls themselves. The sudden failure of previously reliable polling companies to predict results, defying all known statistical laws (as per statements from Edison Research and HarrisX), was another anomaly that raised suspicions and sparked protests. Interestingly, correcting the distortions caused by “excess male voters” restores the results to those predicted by the pre-election surveys of likely voters.

Each appearance of a “carousel” participant adds one vote for the Georgian Dream and likely subtracts one from the opposition while leaving voter turnout unchanged. As such, it renders fraud detection through traditional methods nearly impossible.

This is evident from the graphs: the sizeable variance in Georgian Dream’s results (top graph) is not accompanied by a corresponding fluctuation in voter turnout (bottom graph).

On the horizontal axis, district election commission and precinct numbers are plotted in ascending order, while the vertical axis shows Georgian Dream’s results (top graph) and voter turnout (bottom graph).

The above findings were corroborated by Londa Toloraia, a representative of the My Vote project, during a briefing on Oct. 28: “This fraud scheme doesn’t create the imbalances typically associated with other carousel schemes.”

Mathematically, the scheme is akin to shifting ballots from one stack to another, rather than stuffing extras into one of the piles.

Could it have been simpler?

Why did the ruling party bother with such a complicated scheme? Why not, for instance, simply bribe the voters who were instead asked to “abstain”? The lessons from neighboring Armenia are instructive. Before its Velvet Revolution of 2018, vote-buying was a specialized business in the country, albeit one with a well-known flaw: a client might pay for, say, 1,000 votes, only to find a mere 500 reflected in the official results — and be left with no means of identifying the “cheaters.” Moreover, large-scale vote-buying is a serious offense that carries significant risks of being exposed, while “renting passports” — often masked as social benefit processing — carries less risk of disclosure.

It is notable that the only recorded case of mass ballot stuffing in this election became an international sensation. The incident was reported in multiple languages, including Ukrainian, Russian, and English, and resulted in two arrests.

Unlike in Russia, such overt methods of falsification are impossible to pull off in Georgia, where precinct commissions still include many members of the opposition who cannot simply be “bought off.” But while these honest election officials would have promptly identified and contested obvious cases of ballot stuffing, they were unable to counteract the collusion between the ruling party and complicit registrars, as was outlined in the My Vote report.

Also in contrast to Russia, the results could not simply be “adjusted” overnight. Once completed, copies of official results are promptly uploaded online, where they remain unchanged to this day.

How many votes were stuffed? What were the real results?

By identifying the variable that explains the anomaly and understanding how the carousels operated on the ground, it’s possible to estimate how many votes were fraudulently added in favor of Georgian Dream.

The horizontal axis of the graph represents the variable “gender deviation,” which measures discrepancies in voter turnout between men and women. This axis is divided into bins of eight points. The vertical axis indicates the number of votes within each bin.

In the left and central sections of the graph, where the gender balance is relatively even, results show a consistent preference for the opposition over Georgian Dream. And indeed, surveys that recorded respondents' gender as well as their rural or urban residency, whether in Tbilisi or the provinces, revealed no significant differences between men and women in their views on whether Georgia was moving in the “right or wrong direction.” Both 39% of men and 39% of women believed the direction was correct — the same proportion of voters that supported Georgian Dream at gender balanced polling stations. However, starting at a certain point — around a deviation of 20 [×10⁻³] in favor of men — support for Georgian Dream begins to dominate.

At this point, parallels with election fraud in Russia begin to emerge. Similar anomalies appear in the work of election researcher Sergey Shpilkin, where abnormally high voter turnout often signals fraudulent activity. Similarly, In Georgia, there is an area of abnormally high turnout of male voters where, like in Russia, the ruling party or candidates get a disproportionate advantage. However, aside from this distinction, the graphs show a striking similarity, especially when compared to elections that involved merely moderate levels of fraud, such as those observed during Russia’s 2018 presidential contest.

Using a similar analysis, we break down the results of the presumed beneficiary of the manipulation into “normal” and “anomalous” components. The “normal” part is by definition proportional to the results of supposedly “honest” participants. Following the tradition established by Shpilkin, the remaining part will be given a politically correct label — “anomalous votes.”

We categorize all polling stations into bins based on intervals of the “gender deviation” variable, which reflects the gender imbalance. For each bin, we calculate the number of votes cast for the Georgian Dream and the number cast for the opposition. We then assess the proportion of votes between the ruling party and the opposition in cases of negative or zero gender deviations (when women were more active or when voting activity was relatively balanced).

Assuming that the proportionality between the ruling party and the opposition remains consistent even with high positive gender deviations, we estimate how many votes the ruling party should have received in these bins. The difference between the official number of votes and the expected results for the ruling party is highlighted in red.

On the graph, the horizontal axis represents gender deviation (divided into bins of eight), while the vertical axis shows votes for Georgian Dream per bin. Blue bars indicate “normal” votes, while the red bars represent “anomalous” votes.

Adding up the votes represented by the red bars gives an estimate of close to 300,000 fraudulent votes.

Subtracting these 300,000 “carousel” votes both from Georgian Dream’s results and from the overall turnout would lower the party’s share from 54% to 46%, reducing their parliamentary seats from 89 to 77 — to a tipping point of a majority in the 150-seat parliament.

However, if there had been no “passport rentals,” and if the citizens who were paid to stay home had shown up to vote for the opposition, genuine voters would have contributed to the turnout — and thus to the total denominator. In that case, the ruling party’s percentage would have dropped even further, to 39%. In this scenario, Georgian Dream would have secured only 67 seats in parliament — and ceased to be the ruling party.

In reality, not all of those voters would have necessarily supported the opposition, but calculations based on official Central Election Commission data show that even a slight influx of average voters would have inevitably tipped the scales.