Frequently Asked Questions

• Swing State Status: Nevada was one of 7 broadly acknowledged 'swing states' in the 2024 U.S. Presidential Election. ​ ​ 
  • Swing states generally determine the results of U.S. Presidential Elections. As such, swing states were a priority focus for the ETA.​ ​ 
  • Nevada has the smallest population of the 2024 swing states, meaning there was less raw voting data to work through.
    
• Quality and Availability of Data: Robust, comprehensive, and publicly-available election result data from the Clark County Election Department enabled ETA's analysis to progress quickly and efficiently.
  • “Cast Vote Record” (CVR) data for Clark County, Nevada was made publicly available on the Clark County Election Department shortly after the election.
  • The CVR file shows, on a ballot-by-ballot basis, how votes in Clark County were recorded. It does not include any information that can be traced back to individual voters. 
    
• Preliminary Analysis Raised Concerns: In preliminary analysis of the 2024 swing states, the ETA identified several concerning indicators of potential vote manipulation – specifically in Clark County Early Voting data. 
  • These indicators appear to not be present in election result data for Clark County Mail-In or Election Day data.

As a newly-formed organization, the ETA is still in the process of formalizing our data analysis process.

Broadly,  the process is as follows:

1. Identification of Priority Area: Determine the state/county to prioritize for analysis.
2. Acquisition of Source Data: Download source data from official government websites. 
3. Data Cleanup: Convert data to a usable format.
4. Data Analysis
   - Exploration: Assess available data. Look for expected and unexpected trends.
   - Interrogation: Other analysts review preliminary findings and provide input.
   - Replication: A different analyst ensures original analyst's findings are replicable.
   - Validation: Further interrogation, testing, and refinement of reproduced data analysis outputs.
5. Data Package Development:
   - Structuring: Determine the structure of the data package.
   - Drafting: Draft contextual and explanatory content.
   - Graphic Design: Finalize charts/graphs, titles/labels, attribution, refine coloring.
   - Compilation: Insert charts/graphs and re-order content if needed.
   - Refinement: Ensure source list is complete, links are functional, archive websites and add archived links, check for typos. 
6. Web Publication: Add new content to ETA website and format appropriately. Re-check hyperlinks. Share new data package with others via social media.
7. Formalization: Convert web package into a formatted PDF document using InDesign. Ensure formatting is correct. Distribute formal report to state/county officials with cover letter.

Our timelines for releasing data packages for specific states and counties largely comes down to human capacity and availability. Please note that:

• Our entire team is made up of volunteers with varying amounts of available time. 
• Emergent developments in one state may necessitate shifting our focus temporarily from another. 
• It takes time to interrogate our findings, ensure they are replicable, and present them in a way that is accessible and clear to a layperson audience. 

Our overall priority remains the seven 2024 swing states. Our next target state is Pennsylvania. As we bring on additional data analysts to our team, our goal is to release further analysis within a shorter time frame. 

The election result data used in our analysis generally comes from one of three places:

• Publicly Available on County/City Websites (Public): Counties (and, in some parts of the country, cities) are broadly responsible for administering elections, including U.S. Presidential elections. Whenever possible, we download our source data directly from a county/city website.
• Publicly Available on State Websites (Public): Some states collate election result data from some or all counties and release that data publicly on their website. This data may be more or less granular than county/city data, but depending on the formatting, may be easier for our analysts to draw from.
• Election Atlas (Paid): For historical result data in particular, the U.S. Election Atlas is a reputable source that allows licensees to download particular datasets for a fee. 

Because election reporting varies state-by-state, county-by-county, there are some exceptions to this – for example, some state or local governments direct their constituents to data compiled by third party organizations. Any such data will be specifically cited in a given analysis and included in the data package list of sources.

The type of data available and the format it comes in (e.g. .csv files versus .pdf files) can make a big difference to how much manual effort it takes to prepare the data for analysis, and the extent to which there may be risk of human error in moving data from one format into another. 

Our team is made up of individuals with expertise from multiple fields and perspectives. 

This includes but is not limited to:

• Bachelor’s in Data Management and Data Analysis | Master’s in Data Analysis
• Master’s in Computer Engineering | Master’s in Bioengineering
• Bachelor’s in Network Security and Information Systems
  25B Computer Technician Specialist | 25A Signal Officer
• Chartered Accountant; Certified Information Security Auditor | Certified Risk Management Professional
• Bachelor’s in Political Science | Master’s in Politics and International Relations
• Bachelor’s in Data Analysis and Data Science
• Bachelor’s, Master’s, and PhD in Human Development and Family Studies
• Bachelor’s of Sociology | Partial PhD in Sociology

There is always an opportunity for bias in any kind of analytical work. While the ETA is a non-partisan organization, in practice everyone has private political beliefs. 

We seek to minimize the impacts of our internalized bias through:

• Using Public Data: The datasets that ETA uses are sources from county or state government websites and archived to ensure retrievability by others. 
  
• Replicability: We seek to be as clear and transparent as possible about the process steps we took in reviewing, processing, analyzing, and drawing conclusions/theories from data.
  
• Keep It Factual: Be mindful to avoid opinionated language or supposition. Include reputable sources when introducing new concepts. Avoid inflammatory language. 
  
• Multi-Party Review: Ensure that the analysis and findings are reviewed by multiple individuals with different specializations and skillsets.

 The ETA welcomes (and enthusiastically encourages) peer review of our work, as well as constructive debate about our methodology and findings. Replicability and transparency are critical to our work. 

• Source Data: Our source data is taken from public sources (generally state or county/city websites). 
• Process Transparency: We seek to be as transparent as possible in communicating the tools we use to undertake analysis and why we have made the decisions we’ve made about ‘what’ we’re measuring and why. 
• Replicability: The information we are sharing is not secret or hidden knowledge – it is derived from public data, and you can do it too! We want others – experts and hobbyists alike – to feel empowered to not just check our work, but also to undertake their own election analysis efforts if interested. 

One of the biggest challenges the ETA faces are the barriers and complexities involved in getting access to high-quality, reliable election data in a format that can be used for analysis. Most common challenges include:

• Variability of Data Reported:  The specifics of exactly ‘what’ data must be published, the format of that data, and how granular vs broad the data is varies significantly from jurisdiction to jurisdiction. 
  
• Historical Data: The further back in time you area seeking to pull election result data, the patchier and less available online it is likely to be.
   
• Vote Totals vs. More Granular Data: Subdividing election results into the way the votes were cast (e.g. Mail-In versus Election Day) is extremely helpful for our analysis, as is access to ballot-level data and data associated with specific polling stations or electronic voting infrastructure. The availability of this data is variable.
  
• Format: Some file formats are less easily used for analysis than others, and may require very labor-intensive work in order to get the data into a usable format. As this increases the risk of human error, it may be necessary to undertake this process twice with two different individuals in order to ensure information accuracy. 

Voter suppression involves tactics used to discourage or prevent specific groups of people from voting or registering to vote. Some tactics used to suppress voters are unlawful; others, however, use state and federal laws as the vehicle through which voter suppression is undertaken. Per the American Civil Liberties Union (ACLU), voter suppression in the United States disproportionately impacts people of color, students, the elderly, and people with disabilities.

Voter suppression is not the ETA’s area of expertise, as ballots that were suppressed cannot by definition be included in election result analysis. We encourage those interested in learning more about voter suppression to look into the works of individuals and organizations like Greg Palast (author and investigative journalist), Protect the Vote, the ACLU, and the Brennan Center for Justice.

While the term ‘vote manipulation’ can refer to less direct practices like having one’s vote influenced by disinformation, the ETA primarily uses the term to mean direct manipulation of votes that have been cast in a fraudulent manner. Vote manipulation can be done with paper ballots, or with electronic records of ballots cast.  

Examples of vote manipulation include:

• Stuffing Ballot Boxes / Adding Extra Votes: Creating fraudulent paper or electronic votes for one or more candidates and adding them to vote totals for a given candidate
  
• Destruction of Ballots / Deleting Votes: Physically destroying paper ballots for one or more candidates, and/or deleting the electronic record of votes cast for one or more candidates.
   
• Vote-Relabeling / Vote-Flipping: Changing votes for one candidate to instead be for a different candidate. This method is easier with electronic ballot records than paper ballots.  

There are a lot of ways to influence the outcome of elections. Many of these are legal and normal parts of the campaign process; others, such as heavily gerrymandered electoral boundaries or suppressing the vote of a particular population, may be legal but undertaken for immoral reasons. 

Other ways to influence outcomes – such as vote manipulation – are both generally illegal and may leave traces in election result data. These traces may be detectable using statistical methods. 

Detecting vote manipulation through statistical methods opens a path to undertake targeted recounts, audits, and/or investigations into concerning results. This work can either verify the results as accurate, reassuring voters of the integrity of their vote, or identify fraudulent vote manipulation in a way that can empower officials and constituents to push for change in order to protect future election integrity.

Statistical analysis of election results can be particularly helpful in jurisdictions with electronic voting infrastructure, as such machinery is often proprietary and how they function is rarely transparent to the public.

Yes, though our ability to do so depends on whether comparable data is available. 

For example, in the United States, there is a lot of variation in how election results are reported across different counties/cities. After 2020, some counties/states began posting election results by vote type – e.g. sharing which votes were cast via Mail-In ballot vs. Early Vote vs. on Election Day. The ETA would not be able to compare 2024 Mail-In Voting data to 2016 Mail-In Voting data in such a scenario, because there is no 2016 Mail-In Voting data to compare to. Other counties continue to only publish vote totals without any indicator of vote type. 

Historic election results tend to be harder to find the further back in time you go. For example, in the U.S., there is a significant shift in data quality and availability between 2008 compared to 2004, likely due to how widespread internet usage was at the time of that election.

Yes, although whether fraud/manipulation can be definitively ‘proven’ often depends on the jurisdiction’s willingness to verify results in good faith. Dictatorships wherein a ruling party manipulates votes in their own favour are unlikely to investigate their own misdeeds. Regardless, uncovering manipulated election results in such scenarios can still be critical for communicating the legitimacy of the regime at home and abroad. 

There are  studies focused on elections in countries that were previously members of the Soviet Union, such as Russia, Georgia, Ukraine, Armenia, etc. A technical study detailing one set of methods for analyzing election result data can be found here. 

Independent election observer, data analyst, and forensic expert Roman Udot describes the change in Armenian election results pre- and post- Velvet Revolution:

“I gave lectures [in Armenia] on how rigged elections look. The authorities used to claim, ‘You don’t understand, that’s just how people in the villages vote — high turnout, and they love the government.’ Then the ‘Velvet Revolution’ happened, and the election commissions didn't follow Pashinyan’s orders, so he came to power without fraud. As a result, the [Russian Tail in the election result data] immediately disappeared, like a dog’s tail falling off. Now, there are no more unusually high turnouts in the mountain regions. Before, they were just fabricating the numbers, reporting inflated figures to please their superiors, and taking advantage of the fact that no one went there to monitor.”

The “Russian Tail” is an anomalous deviation from normal statistical distribution in election result data that may indicate falsification, particularly if only one candidate appears to benefit. This term was coined by Russian data analysts when reviewing their own elections.

The Russian Tail has appeared in elections wherein Russian election manipulation is suspected. This pattern has appeared in election result data for Russia’s own 2020 constitutional referendum (regarding whether the constitution should be amended to allow Vladimir Putin to serve additional terms as President), and, more recently, has been identified and cited as evidence of Russian meddling in the 2024 Georgian parliamentary elections.

A bell curve is a very common data pattern. 

One way to analyze data is to represent that data visually – for example, through a chart or graph. When data forms a bell curve, most ‘values’ (the things that we’re measuring) fall close to the average (the ‘middle’ of the bell curve). The further away from the average a value is, the less likely it is. 

When representing visually, whether or not the data you are representing ‘fits’ a bell curve can be an indicator of whether something unusual or unexpected happened. 

With election result data, data that does not align with normal distribution could be explainable and innocent – such as a specific local dynamic. However, certain deviations from normal distribution can also be an indicator of interference – particularly if the pattern is seen in multiple places, and if only one candidate benefits.

No.  Almost all election result data is an aggregation of ‘secret ballots’ cast – meaning there is no identifying information that could tie an individual to their vote. In situations where identifying information may be present (like in jurisdictions where voters are required to sign their name on the ballots themselves), the ETA will never distribute that information further than the official source. 

At the ETA, our #1 priority is, and always will be, the safety, security, and wellbeing of our members. We stand by the principle that, no matter how urgent or critical our work is, we will not be able to do it effectively if our team has their personal safety compromised. Threats to jobs and fears of political or personal retribution are also significant considerations. While our work is non-partisan and deals in public data, anything that threatens the legitimacy of a preferred or ruling political party may be interpreted as a political threat.

As such, the ETA’s policy is that members can either choose to be public-facing or not public-facing. This allows individuals to participate and lend their skills to the ETA while limiting the  risks associated with doing so. Every member of the ETA goes through an interview and onboarding process to confirm identity and to ascertain their intent. As an added safety precaution, additional vetting is required for any member allocated to work on more sensitive or confidential subject matter. 

 Once a member of the ETA produces an analysis that the ETA then shares publicly, the analysis ‘belongs’ to the ETA even if the analyst is credited. This is primarily for the analyst’s protection with respect to legal liability. 

One of the key reasons the ETA was established was to provide the organizational structure to best protect individual members from potential retaliation in response to posting data. While such lawsuits would likely be deemed frivolous, they can still cause stress and financial hardship. If the information we post belongs to the ETA, any retaliation should be directed to the organization rather than the individual analyst.