This Election Integrity Partnership (EIP) analysis was written by Sukrit Venkatagiri, Mike Caulfield, and Jevin West of the University of Washington’s Center for an Informed Public (UWCIP), with contributions from Joseph S. Schafer (UWCIP), Emma S. Spiro (UWCIP), Kate Starbird (UWCIP), Renée DiResta of Stanford Internet Observatory (SIO), and Mishaela Robison (SIO).

***Figure 1:*** A vote count visualization that we found on Telegram. It appears visually similar to those that Lindell used in his livestream. The visualization is purportedly of election results from Michigan’s District 7 House race, featuring Democrat Elissa Slotkin and Republican Tom Barrett.

Seemingly sudden changes or “spikes” in vote counts and shifts in vote composition for candidates are commonplace in elections. Although these vote count “spikes” have specific explanations and are not indicative of voter fraud, they have been incorporated into voter fraud allegations and conspiracy theories in prior elections, including 2020, and now, in 2022. Visualizations that show temporal changes in vote shares (e.g., Fig. 1) can be easily misinterpreted due to errors stemming from data sources and data processing pipelines. These errors enable potentially misleading reporting on data visualizations — both intentionally or unintentionally.

We find many of these same themes in Mike Lindell’s livestream that aired on frankspeech.com on November 8, 2022 [1] — and, as we will describe, have spread widely on social media and messaging platforms. Mike Lindell, is a conservative influencer and businessman who has a history of using supposed statistical and technical evidence to allege election fraud in the 2020 election.

As further detailed below, our analysis of the livestream indicates that Lindell and commentators depicted supposed vote count changes in four potentially misleading ways:

Misinterpreting initial spikes due to early counting of mail-in and absentee ballots
Misrepresenting the release of large batches of in-person votes for counties that may lean Republican or Democrat
Exaggerating random variation, clerical errors, and other outliers
Using misapplied, nonsensical, and uncorroborated statistical terms (“natural” and “F curve”)

However, spikes in time series graphs of vote counts can happen for several reasons.

Vote counts are reported in batches and varying intervals.
There can be delays in counting and tabulating in-person votes.
Some states legally allow absentee and mail-in ballots to be counted prior to polls being closed, and thus those vote counts may be reported en masse.
Voting “spikes” can also be caused due to clerical errors and are not limited to any particular political party or individual.

Spikes in vote counts have been observed for both Republican and Democratic candidates — irrespective of whether they won or lost those elections. Two days after the livestream aired, Lindell even contacted PolitiFact, to clarify that his claims are not a "Democrat or Republican thing" but rather that the voting machines are part of a “uni-party, deep-state globalist, (Chinese Communist Party) manipulation” of U.S. elections. Lindell added, "the end game here is we have to get rid of the electronic computers.”

Unclear source of data for the visualizations used in the livestream

The visualizations used in the livestream appear to be obtained from a third-party website [2]. In the livestream, Lindell and commentators indicated that the visualizations on this website use data from “the Edison Report” [3]. However, Rob Farbman, executive vice president of Edison Research, told AFP that Edison Research did not provide data to Lindell’s colleagues. Lindell also contacted Lead Stories after their independent analysis of the same livestream “to say the data were taken from CNN, not from The New York Times.” We are unable to independently verify the source of Lindell’s data.

Potentially misleading claims and data visualizations from livestream spread widely on social media

By searching for terms mentioned in Lindell’s livestream, we found references to the livestream and photos of visualizations on a number of platforms, including Twitter, TikTok [4], and Telegram, within seconds and minutes after being mentioned. A (now-removed) video posted on TikTok of Lindell advertising the livestream on November 7 had over 198,000 views [5]; and a reposted video of the livestream on Rumble had over 23,000 views [6]. On Twitter, multiple large-follower accounts, including Diamond and Silk, promoted Lindells’ livestream throughout the night and continued to do so in the days after the elections.

***Figure 2.*** *A screen capture of a TikTok post [*7] that appears to originate from Truth Social. The post by “@mikelindell” (inset) claims that an unmentioned group “are currently stealing Herschel Walkers race with the machines!”

Four ways vote count data can be potentially misleading

We now describe the four ways that Lindell’s descriptions and visualizations of real-time vote counts may potentially mislead audiences.

1. Misinterpreted initial spikes due to early counting of mail-in and absentee ballots

Mail-in and absentee ballots can be legally counted prior to polls being closed in many states. Thus, vote counts visualizations may initially show a large spike immediately after the close of polls, as these pre-tabulated results can be reported at this time. This pattern of reporting is commonplace, given that some states may have a large percent of voters who opt to vote absentee or by mail. For example, Washington state adopted universal mail-in voting in 2011, although individuals can alternatively vote in-person.

***Figure 3.*** *Screen capture of the livestream that was later posted to* *Rumble*. In the bottom-left corner of the visualization, a commentator is using their cursor to point out Fetterman’s purported initial lead at approx. 130,000 votes compared to Oz at approx. 22,000. Captured on November 8, 2022, from the livestream video on frankspeech.com.

In the livestream (5:19:49), Lindell and commentators discussed the Pennsylvania Senate race between Lt. Gov. John Fetterman and Mehmet Oz. Without evidence, one commentator referred to an initial spike in the vote count (favoring Fetterman) as potential evidence of election fraud, stating, “it appears that you've got the absentee [ballots], which are gonna [be] pro-Democrat and also are going to contain most of whatever fraud that there is in there.”

A commentator continued to make uncorroborated claims of irregularities in the initial reporting of votes for the Pennsylvania Senate race:

“We have seen this over and over again, race after race after race, across different states [...] Suddenly the establishment candidate ends up with a 500,000 lead [...] And from that point on, as you said, the graph looks very, very normal. But here is what they did. Look here, this very first update at 130,000 votes for Fetterman and 20 thousand plus for Oz. At ground zero, at the first update after 8 o'clock.” (emphasis ours)

Without more information about the underlying vote count data that is being used in these visualizations, we cannot make specific claims about its authenticity or the potential reasons for the initial increase in vote counts for the Pennsylvania Senate race. However, given recent history this is not necessarily evidence of malfeasance.

2. Misrepresented the release of large batches of in-person votes

Vote shares are not equally distributed within a state, and counties may report batches of results at different intervals and with varying batch sizes. For example, when large counties and/or partisan strongholds report vote counts, they often come in batches which can result in a large spike of votes favoring one candidate over the other. This is not clear and convincing evidence of malfeasance or election fraud.

However, in the livestream, Lindell and commentators potentially misrepresented irregular spikes in the visualization as clear and convincing evidence of voter fraud. For example, referring to a jump in votes for both Herschel Walker and Raphael Warnock (Fig. 4), Lindell provided an unsubstantiated explanation for a larger increase in votes for Warnock:

“You can see in the beginning where the Democrat [Warnock] takes the big lead because they set the algorithms, heck they probably preloaded ballots. [...] Well then the real time votes come in, the real votes, and Herschel's passing the opponent up and then they go ‘You ain’t getting ahead of us.’ Bam, up goes the spike. […] if you took that real-time crime spike out [...] Herschel would be up by 150,000 votes.” (emphasis ours)

More generally, a livestream commentator refers to Lindell’s claims as allegations of fraud, stating, “you said that this has to be some type of electronic type of fraud because... You can't load this many ballots in that short of a time frame. This is just not possible.”

Although we cannot independently verify Lindell and commentators’ claims because the underlying source of data is unclear, such increases in vote counts are usually tied to the factors we mentioned earlier. Such spikes may also be caused by varying amounts of votes released at once, coming from different locations or reflecting different modalities of voting. Thus, we would expect that some batches may have an irregular partisan distribution for one candidate or another.

In this case, Farbman, a representative of Edison Research, told AFP that the large increase in votes for Warnock compared to Walker aligns with the reporting of votes from DeKalb County, the largest county in Georgia, that is heavily Democratic. Given recent history and Farbman’s explanation, claiming spikes in vote counts as evidence of election fraud appear to be unfounded.

***Figure 4:*** Screen capture of the livestream from frankspeech.com. Lindell directs viewers to the graph (inset), and describes patterns that he observes as a supposed real-time vote count is updated. Captured on November 8, 2022.

3. Exaggeration of random variations, errors, and outliers

There are hundreds of races happening across the country, each with different vote counting rates, as well as random variation and errors. There may be errors in data entry or processing that are quickly rectified, but still appear in real-time visualizations. While these abnormalities may appear unusual, they do not seem to be clear and convincing evidence of election fraud, as is claimed by Lindell and commentators.

In the visualization for the Illinois Senate race between Tammy Duckworth and Kathy Salvi (Fig. 5), Lindell also refers to an apparent increase in approximately one million votes in favor of Duckworth as “real-time crime spikes” (approx. 1:59:23). Lindell then proceeds to ask a commentator, “That million vote spike could only be done with computers obviously, right?”

In response, a commentator suggests that the source of the error is difficult to trace due to how vote counts are transferred from counties to the “Edison Report” and the media:

“One would imagine it would be difficult to come up with a way that someone accidentally keyed in a million extra votes. We don’t know at what point… because of the mystifying way that the vote counts get from the counties to Edison [Report] and the media… know at what point that this might have happened [...] and that's part of the problem. There's no accountability and we're talking about all these black box systems.” (emphasis ours)

Farbman told AFP that this sudden increase in vote counts for Duckworth was due to a data entry error that was quickly corrected. Farbman explained, "We added a 9 at the beginning of Duckworth's number in Rock Island [...] The vote typo was fixed minutes later."

***Figure 5:*** *At 8:54 p.m. on November 8, 2022, Lindell posted on* *Truth Social* stating “Million vote spike and the only explanation is corrupt electronic voting machines. Watch now for more Real Time Crime [...].” The screenshot appears to be directly from his livestream occurring at the same time. In the livestream, Lindell directed viewers to the graph (inset), and describes patterns that he observes in the supposed real-time vote count. Captured on November 8, 2022.

Data pipelines can be leaky. The comment in the previous paragraph about the source of error, and later admission of a clerical error from Edison Research, is especially pertinent. Visual inspection of a graph of unofficial vote counts is insufficient to allege voter fraud. The data pipeline would need to be thoroughly investigated. Errors can be introduced — such as in Duckworth’s case — when vote tallies move from county officials to sources like Edison Research [8], even if the original data is error-free. Apart from the time series data that came from Edison Research and scraped from CNN as noted above, other types of errors could have been introduced into this process.

Moving right along the x-axis, the visualization also depicts an alleged “drop” in approximately one million votes. Lindell suggests that this drop is additional, supposed evidence of election fraud because the count of ballots can only increase and not decrease (approx. 5:49:54).

However, in this case, the drop can be attributed to the correction of the aforementioned data entry error. Such clerical errors have occurred in past elections, too.

More broadly, there are a number of ways that this pattern might be produced. Unofficial results are just that – unofficial. The Associated Press process, for example, involves thousands of temporary freelancers phoning in results to hundreds of data entry clerks. This process is designed not for certification of results but for news reporting, and errors are caught after data entry. In such a system, it is not difficult to imagine a temporary error emerging that is subsequently corrected.

4. Used nonsensical statistical terms to convey a potentially misleading sense of expertise and to imply abnormality in the vote counts

Finally, we find other techniques used to convey a potentially misleading sense of expertise and legitimacy by using technical-sounding jargon with no well-founded basis in visualization research, misinterpreting statistics, and making uncorroborated claims of mathematical implausibilities.

Nonsensical Technical Jargon. One notable example is the use of the term “F curve” that does not appear to have an equivalent usage in the extant statistical or visualization research literature. To the extent that the term “F curve” refers simply to two lines intersecting in the line graph, it is an iteration of previously mentioned misleading interpretations of changes in vote counts (see points 1–4).

In our analysis, we find that this term is used at least twice, once during the livestream, and once in a seemingly related Telegram channel (Fig. 6). Referring to a similar, perceived spike in votes for the Democratic candidate in Michigan, a commentator claims without corroboration that “this is not something that you would normally see and Jeff can tell you a little more about this.” To this, another commentator adds, “Michigan may be trying to follow up on their big F curve spike with a somewhat smaller spike here, but still very significant.” (emphasis ours)

To the best of our abilities, we did not observe Lindell or commentators explain what an “F curve spike” is. Note that this potentially misleading term was also used during the 2020 elections to make allegations of fraud (see Fig. 7 and 8).

***Figure 6:*** *Screen capture of a post shared to Telegram by “The Lone Racoon” with 2,000 views [*9]. Like Figure 1, this appears to be similar to the visualizations that Lindell and colleagues use in the livestream.

***Figures 7 and 8:*** Examples of simplified versions of the “F curve” which were used as memes to allege election fraud in the 2020 presidential election and in the 2022 midterms. Obtained via Google search for “F curve election.”

Commentators also use the term “natural” to suggest a demarcation between regular and irregular behavior. For example, one commentator asked, “Does that [graph] look natural to you?” to which another responded, “It looks fairly natural.” However, temporary irregularities in vote counts are commonplace given recent history.

Misinterpreting Statistics. On the livestream, between approx. 7:31 and 7:47 p.m. PST, a commentator made a premature claim about the Colorado District 3 results because votes were still being counted. The commentator stated that Lauren Boebert was “redistricted into a 10-point-plus conservative statistical [sic] district so she didn't lose any support, she gained 10 points according to voter registration.” However, they refer to the current visualization depicting Boebert “to be losing by 7 to 10 percentage points right now” as “absolutely insane.”. More recently, despite Boebert lagging behind her opponent as of 10:20 p.m. PST on November 8, The New York Times described the race between Boebert and her opponent as “unexpectedly tight” on November 10.

Premature and uncorroborated predictions such as these can be potentially misleading to audiences, as shown in 2020.

Uncorroborated Claims of Mathematical Implausibility. Lindell and commentators also make multiple, uncorroborated claims of mathematical implausibility. For example, one commentator referred to a spike in votes for Warnock as “mathematically, something impossible” (emphasis ours). Claims that such spikes as mathematically “impossible” can be misleading. As we have mentioned in the previous three sections, there are a number of reasons why such spikes in vote counts are possible — and even commonplace.

We also find that two commentators commentator made uncorroborated statements about mathematical implausibility:

Commentator 1: “That spike up there is pretty inexplicable. That one looks to be a reactionary dump of ballots.”

Commentator 2 (in response): “Absolutely. There is no technical explanation for that spike and there’s no explanation from an election integrity perspective for that kind of a spike. That’s 100,000 votes that comes out of nowhere. That’s just not real.” (emphasis ours)

It is important to not ascribe expertise, authority, or legitimacy simply because a person uses technical-sounding jargon or makes repeated but uncorroborated claims of statistical improbability. It is also not possible to make claims of probability based on inspection and visualization of data alone.

A Final Note: What to Talk About When Talking About Data Science Expertise

While this post details some of the ways in which the analysis presented by Lindell and others is potentially misleading, it is impossible to “pre-bunk” every statistical quirk that can be exploited — or just misunderstood — by voter fraud influencers. Instead, it is worthwhile for communicators and educators to focus on the larger issue as well: why do such presentations, despite potential flaws, feel compelling to viewers?

While confirmation bias and wishful thinking play a role, another reason is that such presentations feel to the public like “data science.” In the public’s mind, this is what data science looks like: you have data, you apply math, you produce a chart, you get a result. But such presentations may differ from real data science in dramatic ways. Understanding what real expertise looks like can better help people understand what research is, and what may merely be an imitation. Here we present some features of data expertise to help the public better understand what they should look for in choosing data interpreters.

First, data experts are sensitive to the myriad ways in which errors can be introduced when collecting and processing data. In this case, we noted that Lindell and his commentators do not seem to have a full grasp on how “Edison Report” — or whomever may be their purported source of data — produces their data, highlighting “the mystifying way that the vote counts get from the counties to Edison.” Data scientists are trained to build reliable data pipelines, but they understand that leaks invariably find their way into these pipelines. Before making claims about data, it is critical to be on the lookout for these leaks and to search for corroborated context before making claims, rather than simply raising questions.

Second, experts often understand — in depth — the specific underlying behavior that the data purports to represent. A common exhortation of the conspiracy theorist is to “just do the math!” But unlike high school algebra, the numbers in an election are typically understood by investigating the processes that produced them. In the field of elections, this behavior is incredibly complex and various. To analyze election data, it often is not enough to be a mathematician who worked on data analytics in another domain. You need to be an election statistics expert. And when supposed experts on YouTube or Rumble do not comprehend or communicate basic differences in election procedure to their audiences — such as when different ballots are counted — that expertise may be missing.

Third, experts deal with elements of scale and randomness. This one is a bit harder for the public to ascertain: experts place local weirdness in the context of global scale. Judging whether a given race’s progression was mathematically unlikely isn’t possible without evaluating that in the context of the number of races run, and the expected variation within them. Hundreds of races across twenty or more time slices can throw up the occasional quirk, and experts have to look at local patterns through that lens.

Most importantly, experts engage with the larger discipline or profession. When it comes to science, one of the most common public misunderstandings is that individuals endeavor to produce good results through an individual process. But what gives science its special power and professions their ongoing improvement is their social process. Data scientists and data professionals engage publicly with other experts and professionals in their specific field, show their work, and have it critiqued by people familiar with the best work in the field.

Since it may be difficult for members of the public to understand what a data reporter (like Lindel and commentators) is missing in terms of background knowledge or mathematical process, this understanding may be the most crucial. Such oversights are surfaced when individual researchers seek out feedback from a broader expert community. If your chosen data reporter does not engage with a relevant academic or professional community, find someone who does. If influencers are not engaging with that community — or if that community is not engaging with them — it may be because the community has determined they are not doing good data science.

While communicating about specific statistical misunderstandings exploited by voter fraud influencers can help in the short term, in the long term educators and communicators must do better at explaining why the surface features of expertise presented in such videos are not substitutes for the expertise needed to make such claims.

***

Citations

[1] The livestream lasted approximately eight hours. We viewed the livestream in real-time between approx. 6:53 p.m. and 10:45 p.m. PST, transcribing the audio to text. Because of the ephemeral nature of livestreams, a recorded version does not appear to be accessible on the same website. A copy of the livestream appears to be accessible here.

[2] https://magaraccoon.com/Midtermresults.asp. Last accessed on November 12, 2022.
[3] Throughout the livestream, we find that Lindell refers to Edison Research as “Edison Report,” though one commentator corrects Lindell stating that the name of the company is Edison Research, but that the data “is actually taken from CNN” (approx. 5:51:25).
[4] Note that a search for “Mike Lindell” on TikTok returns no results, but keyword searches for “crime spike” and “frank speech” do. We did not interact with any representatives of TikTok to write this report.
[5] As of 1:08 a.m. on November 10, 2022. The video is no longer accessible on TikTok.
[6] As of 2:43 p.m. on November 9, 2022.
[7] Accessed on November 9, 2022. This account does not appear to exist on TikTok as of 5:40 p.m. on November 10, 2022.
[8] Throughout the livestream, we find that Lindell refers to Edison Research as “Edison Report,” though one commentator corrects Lindell stating that the name of the company is Edison Research, but that the data “is actually taken from CNN” (approx. 5:51:25).
[9] Captured on November 9, 2022 from https://t.me/s/ALoneRaccoon.

How Real-Time Visualizations of Vote Count ‘Spikes’ Can Lead to Unfounded Allegations of Election Fraud