6

Donald Trump has tweeted

“REPORT: DOMINION DELETED 2.7 MILLION TRUMP VOTES NATIONWIDE. 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.” @ChanelRion @OANN

Twitter has marked the tweet "This claim about election fraud is disputed"

Is there evidence that voting machines from Dominion Voting Systems miscounted millions of votes for Trump?

Oddthinking
  • 140,378
  • 46
  • 548
  • 638
Mawg says reinstate Monica
  • 4,642
  • 4
  • 28
  • 41

2 Answers2

34

TL;DR: this claim is based on updates in the real-time election night data reported by news outlets, not the official vote counts reported by states and counties. There are discrepancies in the news outlet's data feed in the vote counts for all candidates, not only Trump's vote counts, and in many states, not only those using Dominion. In a similar instance, this kind of discrepancy has been attributed to a data entry error (by people working for the news/media feed, not by election officials!)

Here is a video of the claim he is referencing (via OAN). The claim is:

According to an unaudited analysis of data from Edison Research, states using Dominion voting systems may have switched 435,000 votes from Trump to Biden. Another 2.7 million Trump votes appear to have been deleted by Dominion.

OAN cites the work of an "unaudited" data analysis by an unnamed author. Here is the "data analysis" that they are citing.

The claim is based on the New York Times' election site. The New York Times uses Edison Research's data feed for this site, but they don't make the full Edison Research feed available. However, some of the data underlying the tables and graphs on their Election site is made available (unofficially) in machine-readable format. This data is aggregated by the New York Times, and possibly processed in other ways, not a direct feed from Edison Research.

The data used here is the time-series data that shows the reported vote share of each candidate and the number of votes counted in a state over time, as shown on the New York Times Election website.

The claim is based on anomalies like this one in the NYT data:

Image from NYT election night data feed

In this example, we can see that the total number of votes increased by only 54, yet the vote share changed from 0.566 to 0.42, to 0.56 to 0.4266. The implication is that:

  • As of that "187" entry shown above, the vote is approximately 1689209 Trump to 1253477 Biden,
  • In the "188" entry, after a 54-vote update, the vote is approximately 1671332 Trump to 1271406 Biden.
  • Somehow, the update has "removed" votes from Trump (who has fewer votes than before the update) and "given" them to Biden (who has more "new" votes than the change in the total number of votes).

("Approximately" because of rounding.)

The author of the claim wrote a script to find all such instances of votes "switching" from Trump to Biden (or to Jorgensen).

  • They found instances of "switching" in states that use Dominion voting systems, and in states that use other systems, or where it's not known what voting system is used.
  • They noted some states that use Dominion voting systems where they did not have any instances of this "switching".
  • While the original "data analysis" only counted votes "switched" from Trump to Biden, it does occur the other way around (i.e. there are updates in which Biden "loses" votes). (When users pointed this out in the original forum thread e.g.: "I was looking through Wisconsin the other day trying to catch fraud and in fact Biden lost significantly more votes than Trump did in that state.", they were dismissed with responses like "Cope harder, Biden boy" and "Get out shill").
  • And, there are also instances of vote counts decreasing in the Edison feed for downballot candidates.

Incidentally, OAN (and Trump, who referred to the OAN report), misrepresented the part about "lost" or "deleted" votes. The "data analysis" counts "lost votes" any time "the total amount of votes counted decreased" from one update to the next, but unlike OAN and Trump, it doesn't say those are "Trump votes".

The entire claim is based on data that is not official election data! Here's some more information about how Edison Research collects the data for the data feed used by New York Times and other sites on Election Night:

  • There are "reporters at county elections offices who call results" into the Edison phone center
  • They use "data feeds provided by some states and counties"
  • They have people who "scour state and county websites for results" to enter into the system
  • They have people who monitor "results sent from counties, cities, and towns via email or fax"
  • They have people ("chasers") who monitor other news sources for results not yet in the Edison system.

In summary: they have many, many, people, manually entering data as fast as they can into an online system, with great potential for error. In fact, we already know of one major error that affected their Election 2020 feed.

AP has a similar process for reporting the vote on election night. A similar vote "switching" anomaly that was reported to the county that was allegedly affected, was attributed to an error in the AP election night feed:

Rock County Clerk Lisa Tollefson is pushing back against claims of vote tampering in Rock County during the Nov. 3 election.

A Nov. 9 article published by The Gateway Pundit, a conservative news and opinion website, claims that a source in Rock County observed vote totals being changed from President Donald Trump to President-elect Joe Biden, citing roughly 10,000 votes that were allegedly changed in Election Night reporting.

The article shows screenshots of Rock County results that were part of Fox News election night reporting, which were submitted by the anonymous source. The article does not include figures from the Rock County election website.

“It looks like the news outlet reported incorrect information,” Tollefson said. “That’s not the county’s website and we’re trying to figure out where the information came from and where the discrepancy came from.”

ff524
  • 10,181
  • 6
  • 47
  • 60
  • 1
    There are several good and coherent analyses of the *apparent* vote switching anomalies in the NTY data that point out they all occur because of *rounding* in the NYT % share numbers. Thus the entire issue is a result of misapplying an analysis based on approximate numbers and is simply not present tin the raw vote counts. A good example of why partizan confirmation bias leads to the dumbest math errors being ignored. – matt_black Dec 04 '20 at 14:06
  • @matt_black Some of the discrepancies are too large to be rounding errors, so it's not accurate to say that they *all* occur because of rounding. – ff524 Dec 06 '20 at 01:08
  • Did you check that, or was that part of the original claim? In in the original claim source based from data published in the NYT it clearly was as more than one analyst has pointed out (see [this youtube dissection](https://www.youtube.com/watch?v=IxElBD67u2Y&list=LL&index=5&t=8s)). The fact that the claim gained any traction is strong proof that even basic critical faculties go out the window when you *really* want to find something because of your partizan position. The basic math error is embarrassingly naive. – matt_black Dec 06 '20 at 15:01
  • @matt_black The example in my answer is one that cannot be explained by rounding error. Also, in addition to vote "switching" the original claim refers to votes being "deleted", which is what they call it when the integer total "votes" field decreases from one entry to the next. No rounding error there. – ff524 Dec 07 '20 at 02:42
  • The example you quote is explained *entirely* by rounding error. The source quotes *exact* total votes but *approximate* % for each candidate. If those % numbers are rounded then a rounding switch in the last decimal place moves about 6k votes in the *calculated* votes from one candidate to the other. The apparent anomalies simply would not exist if the raw votes for each candidate were reported rather than the rounded % shares. – matt_black Dec 07 '20 at 11:32
  • @matt_black I don't see how it is explained by rounding error? The vote share is reported to 3 decimal places. In the first entry, where the trump vote share is reported as 0.566, the possible range of Trump votes counted is 0.5655*2984468 to 0.5665*2984468, i.e. 1,687,716-1,690,701. In the second entry, where the Trump vote share is reported as 0.560, the possible range of Trump votes counted is 0.5595*2984522 to 0.5605*2984522, i.e. 1,669,840-1,672,824. The total number of Trump votes counted has to be less in the second entry than in the first. – ff524 Dec 07 '20 at 19:41
  • You are overinterpreting the precision and the numbers. We *don't* know the exact vote numbers for each candidate from this source. Nor do we know the rounding method generating the vote shares. The method to test the anomaly is to start with exact vote numbers and calculate the rounded share to test whether anomalies happen. Two sources explaining this: [a detailed blog](http://www.knowingandmaking.com/2020/11/how-to-debunk-electoral-fraud-claim.html) and a [long twitter feed](http://www.knowingandmaking.com/2020/11/how-to-debunk-electoral-fraud-claim.html) . – matt_black Dec 07 '20 at 20:22
  • In fact, as far as I can tell from the references, the entire anomaly effect is entirely explicable by modelling the rounding error and applying it to the actual vote count (which is now available) and none of the anomalies appear when the real raw count is used instead of the simplified NYT feed where rounding is done. – matt_black Dec 07 '20 at 20:29
  • @matt_black The [blog post](http://www.knowingandmaking.com/2020/11/how-to-debunk-electoral-fraud-claim.html) you just shared says that the vote share fields are rounded to the nearest 0.001 and so the true vote share could be plus or minus 0.0005. That's exactly what I computed. And, as I mentioned above, part of the claim is that the integer "votes" field sometimes decreases from one entry to the next. It's just not accurate to say that *all* of this can be explained by rounding. – ff524 Dec 07 '20 at 21:27
  • Read the linked analyses to see why the specific original clams compound the rounding problem and why none match that actual vote tallies anyway (which don't show the problem). – matt_black Dec 07 '20 at 22:05
4

All of Pennsylvania uses some form of paper ballot:

As of the June 2, 2020, primary election, all 67 of the Pennsylvania's counties have deployed voting systems that produce voter-verifiable paper records and meet 21st-century standards of security, auditability and accessibility.

In April 2018, the Department had informed counties they must select voting systems that meet the new criteria no later than December 31, 2019, and implement the systems by the 2020 primary.

Having paper ballots allows validation by hand counting and hence makes fraud or deliberate miscounting extremely difficult.

Oops! I failed to notice that the following article was about the 2018 election. However, I searched a fair amount and this was the only accessible article I could find that mentioned Dominion equipment in Pennsylvania. There were several articles that mentioned Dominion in other states for the recent election, but nothing I could find for Pennsylvania.

This article details some of the problems that were experienced during voting in Pennsylvania. They were due to accidentally voting for too many candidates, having Sharpie markers bleed through, and the like. The number of problems were reportedly relatively low.

Another question here had to do with some erroneous reporting of early ballot tallies (don't recall if any was in Pennsylvania), but that had nothing to do with the final tallies.

Daniel R Hicks
  • 5,550
  • 3
  • 36
  • 38
  • 1
    I think this argues that the design of the machines would allow mistakes in the transcription of the votes by the machines and mistakes in the counting of the votes by the machines to be manually detected (which is an important step forward, and I am glad to see it). But *were* any mistakes detected through these processes? The article about minot errors is from last year's elections. – Oddthinking Nov 13 '20 at 03:47