STORY UPDATED: check for updates below.
Does data analysis show that an election software supplier deleted 2.7 million Trump votes and switched 435,000 votes from Donald Trump to Joe Biden? No, that's not true. The states using Dominion Voting Systems software have reported that minor hiccups in the vote-tallying process were the result of user errors, were corrected by the bipartisan "canvassing" system and would not have changed the outcome in any state. The "data analysis" referred to is a string of internet posts for which no one accepts responsibility and the election data firm whose data an anonymous poster claims to have stolen dismisses the claim as a likely fabrication. Lead Stories has tested the "data analysis," finding flaws in the methodology.
"DATA ANALYSIS FINDS 221,000 PENNSYLVANIA VOTES SWITCHED FROM PRESIDENT TRUMP TO BIDEN. 941,000 TRUMP VOTES DELETED. STATES USING DOMINION VOTING SYSTEMS SWITCHED 435,000 VOTES FROM TRUMP TO BIDEN."
This is what the post looked like on Twitter at the time of writing:
(Source: Twitter screenshot taken on Thu Nov 12 18:13:48 2020 UTC)
Dominion Voting Systems currently provides election technology in 28 U.S. states, including swing states Pennsylvania, Nevada and Arizona, though not in every county. For instance, there are two other voting systems used in Pennsylvania, including Unisyn Voting Solutions and Clear Ballot.
Dominion Voting Systems maintains there are no such errors or fraud in its systems. In a November 7, 2020, email to Lead Stories, the company's spokesperson wrote:
There are no credible reports or evidence of any system software errors in Georgia or Michigan, including erroneous reporting of unofficial results from Antrim County, Michigan.
Edward Perez, an election technology and security expert at the Open Source Election Technology (OSET) Institute said he has worked in election technology and election administration for 17 years and that a vote theft of this scale is not possible, given the multiple cross-checks election clerks, administrators and poll-watchers apply to vote tallies:
I am unaware of any credible evidence anywhere indicating that Dominion Voting Systems, through software defects or mischief or malfeasance are deleting or flipping votes on the scale the president is alleging in his tweets.The allegation that that many votes on that massive scale could be deleted or flipped without being detected is borderline absurd.
In a November 12, 2020, phone interview with Lead Stories, Perez, the global director of technology development at OSET Institute, called the claim nearly impossible and spelled out why:
Voting technology does not exist in a vacuum. It is one of at least three legs in a stool that makes election administration possible....In addition to the voting machines, (there are) complicated multilayered procedures, a complex array of safeguards. Every single one of those election records needs to be reconciled against any number of other methods to check. You're going to cross-check and reconcile the number of votes with the number of people who checked in.
If you're magically going to start deleting votes you're going to create enormous holes in the records of an election.
Perez said errors in the tallies in Gwinnett County, Georgia, and Antrim County, Michigan, were spotted and corrected precisely because of the cross-checking process.
Edison Research, whose elections data purportedly was hacked to prove vote-swapping, rejected the idea that the claims of vote fraud are drawn from any Edison reports. In a November 12, 2020 email to Lead Stories, Edison Research President Larry Rosin wrote:
Edison Research has not issued any "report" nor are we aware of any voter fraud.
Edison Research provides exit polls, vote count, election projections and delegate estimates to The New York Times, ABC, CBS, CNN and NBC, among other news outlets.
The claim made it to the President's Twitter account by the following route: Anonymous posts on a website where Trump is referred to as GEOTUS(God Emperor of the United States) that were then turned into a GatewayPundit.com article that was then discussed live on One America News Network.
Origin of the claim
A poster named "PedeInspector" on TheDonald.win claims he stole "Edison" files from the New York Times and used it to pinpoint times when, he claims, votes were suddenly shifted from the Trump tally to the Biden tally.
In a Nov. 12 email, New York Times Vice President/Communications Danielle Ha said there's no evidence of fraud in the Times' data:
The election results are still unofficial in many places. They continue to be updated and verified by election officials, and results providers continue to update and verify the numbers they report. Sometimes over the course of reporting results officials or results providers make small mistakes, such as mixing up two numbers, and they correct those mistakes as they verify the data. These changes are common in the difficult task of reporting millions of votes across thousands of races, and are not at all unusual.
There is no evidence in our data that votes have been fraudulently switched between candidates.
At one point during the string of posts, "PedeInspector" claims One America News, a conservative website, has mixed-up numbers.
Later, "PedeInspector" notes the "Edison data" may not be authentic and may not show what he has said it shows.
"PedeInspector" at one point in the string of posts claims not to be an "American pede." The nickname -- pede -- may be a reference to a dub-step track "Centipede" that was often played at Trump campaign rallies.
Repetition of the claim by Gateway Pundit
Gateway Pundit, a conspiracy-heavy conservative news website, used "PedeInspector" posts to write "Analysis of Election Night Data from All States Shows MILLIONS OF VOTES Either Switched from President Trump to Biden or Were Lost -- Using Dominion and Other Systems."
Gateway Pundit's November 10, 2020, 6:32 p.m. report by Joe Hoft credits PedeInspector: "Tonight we were led to a site on the Internet thedonald.win where someone who had seen our posts decided to do an analysis himself. In the piece the author claims his work is a full list of votes switched from Trump to Biden or votes erased by Dominion (the vote machines used in many states across the US)."
"PedeInspector" posts did not focus on Dominion, but rather appear to have built entirely on the purported theft or scraping of election night data from the New York Times' website.
Gateway Pundit's article continues:
(The author claims that his work has been verified but we have not verified the results so we currently are labeling his results "unaudited.") The author decided to obtain the data himself and perform an analysis of national votes switched from Trump to Biden and votes erased (the total amount of votes counted decreased by that amount throughout the counting).
The author obtained data that has been passed around that was reportedly captured by the New York Times on election night.
By examining the data the author was able to locate an instance where votes were switched from Trump to Biden so he decided to perform an analysis on the entire data set, starting with the states where Dominion voting machines are used.
The author also claims that the data is from Edison Research and it is the same data that is used for election coverage by at least ABC News, CBS News, CNN and NBC News.
Lead Stories Calculates Flaws in "PedeInspector" Methods
To test PedeInspector's theory, Lead Stories replicated his/her method, scraping the data from the New York Times' site. The JSON files do indeed contain a time series of the situation for each state over time.
At each point in time, the data include the percent of precincts reporting, total number of counted votes, ratio for Trump and ratio for Biden. The ratio is a number between 0 and 1, rounded to three decimal places. (If you convert to a percentage, it is rounded off at one decimal place.) Note the JSON data do not include exact vote counts. Note also, third party candidate data are not included.
This appears to be the data the New York Times was using to update graphs and other 'live" displays on its website, but after its initial responses, Edison Research has declined to further discuss the data or its methods.
Tracking "PedeInspector" posts about methods, Lead Stories concluded that he/she multiplied the total vote count in each jurisdiction by the ratios in the New York Times data to obtain Trump and Biden counts, declaring fraud if the Trump count went down or only the Biden count went up from one point in time to the next in the series.
The flaw in this method: The maximum precision of the ratio is capped at 0.001, which is useful for graphics displaying vote proportions. But that ratio means the "PedeInspector" calculation of absolute numbers will be "wobbly" as the total number of votes grows, because 0.001 represents a bigger and bigger number of actual votes. Flipping a number represents a bigger and bigger number of votes in difference in the calculation but the change can be caused by a small number of votes that puts someone "over the hill".
Given that the rounding on the ratio happens to be 0.001, it means that the actual ratio could be 0.0005 higher or lower, for both candidates. So, a one-vote difference could put someone under or over the rounding threshold, leading to seemingly big jumps or drops in the number of votes "PedeInspector" calculates by applying the ratio to the real vote totals.
If you calculate what the minimum and maximum number of possible votes is, given a certain ratio (i.e: total number of votes times ratio-plus-or-minus-0.0005) then you can check for "impossible" drops between one moment and the next. The maximum possible number of votes for a candidate should never be lower than the minimum possible at the previous moment.
Lead Stories did this with an Excel sheet, using the CSV data from "PedeInspector" on thedonald.win.
In the formula, Lead Stories calculated the min/max amount of Trump & Biden votes given the ratio and the rounding. The formulas in the Excel sheet also calculate how many votes 0.1% is (i.e. the precision) and examines the data in two testing columns.
One testing column checks if the maximum possible number of votes for a candidate is higher than the previous minimum and says "OK" if that is the case. The other testing column simply checks if the total number of votes went up or stayed the same compared to the previous line.
In most cases, things are "OK" and where they are "NOT OK" it is usually because the total number of votes went down, not because one candidate or the other was winning/losing odd numbers of votes.
Gateway Pundit's story was not transparent about the way the rounding of the ratio can, as vote tallies rise, generate seemingly big jumps or drops that are not reflective of the actual votes, but only reflective of the limits of applying relatively imprecise ratios to very large numbers. Lead Stories is developing a means to share its spreadsheet for testing of its assumptions and formulas by readers.
State Officials Reject Claims of Dominion Software Malfunctions
Multiple state officials, including at least one Republican, have rejected claims that Dominion Voting Systems software caused fraudulent or erroneous tallies in the votes for Trump and Democratic nominee Joe Biden, nor that there was fraud of any kind.
"This was an isolated mistake that was quickly rectified," Rochester Hills City Clerk Tina Barton said in a video posted to Twitter in response to Republican National Committee Chair Ronna Barret's tweets about Michigan results. Barton's tweet has since been taken down.
Gwinnett County Georgia's explanation of Dominion issues make it clear that the only problems related to Dominion were not with accuracy of ballot tallies, but with rolling up local data to state computers for production of statewide tallies. Here's an example of one post from the county website's updates on election results.
Update #6, November 8, 2020
Dominion Voting Systems technicians successfully pushed the remaining non-provisional Gwinnett ballots to the Secretary of State's election reporting system at 2:34pm Sunday.
2020-11-12T23:53:38Z 2020-11-12T23:53:38ZUpdated to include, in paragraph 11, New York Times' comment on its election data. Updated a second time to add in Lead Stories mathematical examination of the flaw in "PedeInspector" analysis.