🇬🇪 2024 Elections 🇬🇪

Context and Research Background

On October 26, 2024, Georgia held parliamentary elections. The ruling party, Georgian Dream, has been in power for three consecutive four-year terms. Throughout 2024, amid street protests, the adoption of controversial laws, and criticism from Western partners, the party’s approval rating declined to 30–35% by election day. Exit polls at 8:00 PM indicated they had garnered 41% of the vote, suggesting a potential electoral shift in power.

However, within a few hours, the Central Election Commission released preliminary results showing 52% support for Georgian Dream, which increased to nearly 54% by Sunday morning.

Edison Research, an international company that has conducted exit polls in Georgia for nearly two decades, noted that the largest discrepancy between their polls and official results was 5% in 2018. A 13% deviation cannot be attributed to statistical errors.

This study aims to highlight potential electoral fraud, assess its systematic nature, and evaluate its impact on the final outcome.

Key findings

Based on the analysis results:

1. Georgian Dream systematically manipulated the election results, primarily in Georgia’s regions.
2. This manipulation was systematic and contributed an additional 6-8% to their results.
3. Consequently, it can be concluded that unlawful interference had a significant impact on the election outcome.

During the study, scatter plots were created, marking locations that raise serious concerns. All locations were plotted on a geographic map to help the opposition identify and protect votes at specific precincts.

Recalculation of Results Using the Kiesling-Shpilkin Method

Recalculation of Results Using the Kislinsky-Shpilkin Method

For evaluating the results, I employed the well-established Kiesling-Shpilkin method.

This method is based on creating a histogram of votes for the ruling party as a function of the number of precincts.

I made two modifications to the method:

1. I divided all precincts into urban and rural categories. Urban areas included Tbilisi, Rustavi, Kutaisi, and Batumi, where the ruling party lost the election. All precincts in these cities fit a single normal distribution with minimal outliers.
2. I assigned each precincts a weight corresponding to its size, taking into account that in mountainous areas, there are precincts with as few as 100–200 voters.

Using the most conservative vote recalculation approach, considering only the most questionable ballots, Georgian Dream received 47.5%.

Scatter Plot to mark suspicious precincts

To determine which precincts are among the most questionable, it is necessary to create a scatter plot (turnout vs. Georgian Dream result) for each precinct.

I used the same division into two groups (urban and rural). For each group, I calculated the mean values along both axes and the standard deviation. I shaded this area in green. The area with the most suspicious precincts is shaded in red, located in the upper right corner. The remaining precincts are marked in blue.

Both charts are interactive, allowing you to hover over individual points to view key data for each precinct.

Finally we have the following distribution:

Geographical map

For better visualization, I plotted all precincts on an interactive map.

P.S. and Future Plans

This research is not yet concluded. For more urgent publication ahead of a major opposition rally, I have shared the main findings so far.

Several important questions remain open:

1. Based on preliminary data, Georgian Dream appears to have found a backdoor to systematically influence the results from electronic vote-counting machines.
2. For the first time in 20 years, many voters were assigned new polling locations. Voter lists, compared to the 2021 elections, have significantly changed in terms of distribution across precincts.
3. A preliminary hypothesis suggests that multiple votes by the same person might have been organized at certain precincts.
4. This hypothesis is currently under investigation.

All links to the original methods and algorithm details should also be provided.