Georgia RLA Discrepancy Analysis 2024

Date: December 30, 2024

Election Info: Georgia (GA), November 5, 2024 General Election

Author: Staff+ engineer for a decade at various internet scale companies

I analyzed a total of 442 audited batches, as provided by the Georgia Secretary of State, to determine whether observed discrepancies between reported and human-audited ballots could be attributed to chance. A statistical binomial test was applied to evaluate the likelihood of these discrepancies occurring randomly.

In essence, this analysis aimed to assess the probability that vote discrepancies were due to chance alone, helping to reassure voters that ballots were counted fairly in Georgia's 2024 Presidential Election. The p-value from the test indicates how likely it is that such discrepancies could happen by random chance; a very low p-value suggests they are unlikely to be random and may warrant further review.

Important: Demanding high standards for election security is not conspiracy theorizing, and framing it as such is misguided. All information presented here is based solely on rigorous mathematical analysis. All of the data comes from credible government sources, accessed through a .gov domain.

Background

This report references the publicly available data from the Risk-Limiting Audit (RLA) of Georgia's 2024 Presidential Election. The dataset can be accessed on the Georgia Secretary of State's website via this download link[^1].

Definitions

• Batch-Level Discrepancy: A difference between the reported and actual vote counts within a single batch of ballots.

• Margin Effect: The impact of a discrepancy on the winning candidate's margin. A positive margin effect favors the winner, while a negative margin effect favors the runner-up.

• P-value: The probability of observing discrepancies this large (or larger) if they were due to random chance.

• Statistical Significance: If the p-value is below 0.05, it's unlikely that random chance alone produced these discrepancies.

• Binomial Test: A statistical test (with p=WinnerVoteShare) that counts how often discrepancies favor the winner compared to the total and calculates the likelihood of that pattern occurring by chance.

  $$\begin{array}{}\text{(1)}& P(X\ge k)=\sum _{x=k}^n(\genfrac{}{}{0ex}{}{n}{x}){\left(\frac{1}{2}\right)}^n\end{array}$$

• Reported Vote Count: An unaudited tabulation of votes, which may or may not be 100% accurate

• Audited Vote Count: A human-audited tabulation of votes, assumed to be a 100% accurate representation of the votes in a batch.

Methodology

1. Parse the report CSV to find vote discrepancies between the reported and audited vote counts per batch.

2. Calculate the margin effect for each discrepancy to determine whether each discrepancy favored the winner.

3. Perform a statistical binomial test to assess how likely these discrepancies are to occur by chance alone.

4. Compare the resulting p-value to 0.05 to determine if further investigation may be warranted.

This methodology is effective because it uses a straightforward, statistically rigorous approach to identify and evaluate vote count discrepancies. By linking discrepancies to their margin effect and applying a simple binomial test, it adheres to a common scientific method used to assess randomness, like testing drug efficacy in clinical trials. Its simplicity ensures accessibility and ease of implementation while maintaining robust, actionable insights.

Results

Analysis

The audit discrepancy analysis of Georgia's 2024 General Election reveals that, for most categories, discrepancies are likely attributable to random chance, as indicated by p-values above the 5% significance threshold. This suggests that no systematic issues were detected in these areas.

However, the analysis identified two categories with statistically significant discrepancies, which are unlikely to be due to random chance alone, warranting further investigation to determine their cause and potential impact:

1. Advanced In-Person Votes: Discrepancies affecting the margin for advanced in-person voters exhibit a p-value of 3.85%. This indicates a meaningful deviation from what would be expected under random chance.

2. Imagecast Precinct Votes: Discrepancies affecting the margin for votes scanned in local precincts (Imagecast Precinct) exhibit a p-value of 0.77%. This is even more concerning given the decentralized nature of local precinct operations and their potential vulnerabilities.

Implications of Imagecast Precinct Discrepancies

The concentration of non-random discrepancies in the Imagecast Precinct group is particularly troubling given the decentralized nature of local precinct operations. Unlike the central election office, local precincts operate with less centralized oversight, which can result in inconsistent security protocols, varying levels of staff training, and increased exposure to both errors and deliberate interference.

This finding is even more significant because it aligns with Georgia's known vulnerabilities in election technology. The National Vulnerability Database (NVD) identifies 10 security vulnerabilities (CVEs) affecting the Democracy Suite 5.5a voting system, several of which are classified as "HIGH" and pose significant risks requiring urgent mitigation (source)[^2]. These vulnerabilities include weaknesses that could be exploited to interfere with or manipulate scanning processes, which is particularly relevant since Imagecast Precinct scanners are decentralized and may be more exposed to operational risks.

Despite these risks, the Georgia Secretary of State announced in May 2023 that the Democracy Suite system would not be upgraded until after the 2024 election (source)[^3]. The fact that the Imagecast Precinct group is the only category with statistically significant discrepancies suggests a potential intersection between operational vulnerabilities and the technical weaknesses highlighted in the NVD. This underscores the need for immediate action to investigate and mitigate these risks to protect the integrity of the election process.

Conclusion

In my opinion, the statistically significant discrepancies in the Imagecast Precinct group, combined with Georgia's documented election technology vulnerabilities, pose a significant challenge to ensuring public confidence in the integrity of the 2024 election results.

To address these issues and reinforce the transparency and accuracy of the electoral process, Georgia should promptly take one of the following actions:

1. Conduct a full hand recount of all ballots scanned using Imagecast Precinct (ICP) technology. This is the most thorough and reliable method to verify the accuracy of the results.

2. If a full recount is not feasible, perform a targeted Risk-Limiting Audit (RLA) for all precincts employing ICP technology to provide statistical confirmation of the reported outcomes.

In my view, implementing one of these measures is essential to safeguarding the credibility of Georgia's 2024 Presidential Election and maintaining public trust in the electoral process.

Detailed Results

All Discrepancies

All discrepancies, regardless of scanner or voter method.

Batch-Level Analysis

Discrepancy Effects Table

Election Day Discrepancies

Discrepancies affecting margin for election day voters.

Batch-Level Analysis

Discrepancy Effects Table

Advanced In-Person Discrepancies

Discrepancies affecting margin for advanced in-person voters.

Batch-Level Analysis

Discrepancy Effects Table

Absentee By Mail Discrepancies

Discrepancies affecting margin for absentee by mail voters.

Batch-Level Analysis

Discrepancy Effects Table

Imagecast Precinct Discrepancies

Discrepancies affecting margin for votes scanned in local precinct.

Batch-Level Analysis

Discrepancy Effects Table

Imagecast Central Discrepancies

Discrepancies affecting margin for votes scanned in central election office.

Batch-Level Analysis

Discrepancy Effects Table