329,614 Election Crimes Were Detected in 2020

[Important note: A perceptive commenter found a mistake in my numbers. I fixed it below. The message that there was lots of fraud is unaffected, but it now looks much less likely that the amount was big enough to have swung the election.

3,180 views, 11/4. See also my posts on suspicious vote patterns in 2020 and absentee voting fraud and the oppressor/oppressed reason Democrats would cheat.]

In the November 2020 election, 90,687,978 ballots were mailed out to voters, of which 70,551,227 were returned. Of those, 0.8% were rejected, which comes to 564,409 ballots. Of the rejects, 32.8% were rejected because of “Non-matching signature”, which comes to 185,126; 12.1% because of “No voter signature”, which comes to 68,293; and 13.5% for “Voter already voted in person”, which comes to 76,195.¹

Thus, while some absentee ballots were rejected for reasons that aren’t fraud,² about 329,614 were rejected for reasons that are fraud at first look. Not all of these are fraud, of course. Some people write their signatures differently at different times; some people forget to sign things; some people forget they turned in an absentee ballot. But suppose you discount the number by 90%. It still comes to more than thirty thousand attempts at fraudulent votes, thirty thousand election crimes committed.

And those are just attempts. Election offices are reluctant to void ballots because of signatures not matching. My guess would be that when election workers see suspicious signatures, they let the least suspicious 50% through. The Michigan Secretary of State’s guidance to election officials in 2020 said,

Signature review begins with the presumption that the voter’s ... envelope signature is his or her genuine signature.

1. If there are any redeeming qualities in the ... return envelope signature as compared to the signature on file, treat the signature as valid. ...

2. A voter’s signature should be considered questionable only if it differs in multiple, significant and obvious respects from the signature on file. Slight dissimilarities should be resolved in favor of the voter whenever possible.³

In some states, judges held that election officials could not reject a ballot even if fraud was obvious. Courts in Indiana, South Carolina, and Pennsylvania said that in the 2020 election election officials could not reject ballots even if the signatures were clearly mismatched.⁴ The Pennsylvania Supreme Court said, “county boards of elections are prohibited from rejecting absentee or mail-in ballots based on signature comparison conducted by county election officials or employees, or as the result of third-party challenges based on signature analysis and comparisons.” The court held that the state’s election code permits signature matching for in-person voters but not for absentee ballots.

It would be nice if states did audits, with input from both parties so we’d know about this.⁵ So really, doing a bit of guesstimating and rounding, we could use 500,000 fraudulent votes as an estimate, though it might be a lot higher, and I’d guess more like a million, personally.⁶ We do have lots of anecdotal fraud evidence of varying reliability too, but I don’t know what to do with that.) I’ll stick with the 329,614 number for the rest of this post, though, because maybe I’m wrong about how reluctant election workers are to void votes.

How many of those 329,614 fraudulent vote attempts resulted in prosecution? Zero.

How could they be prosecuted? If election workers see somebody come in to vote who’s already voted absentee, are they going to arrest them? Most likely that person had somebody steal his absentee ballot, and we don’t want to arrest the victim. Similarly, if a signature is an obvious forgery, are we going to arrest the voter whose signature was forged? Maybe he sold his absentee ballot to someone else without signing it, but how are we going to prove that, even if it’s true? Thus, what this shows is that convictions for voting fraud are going to be extremely rare even if vote fraud is common. Remember: even if you conclude that just 50,000 of the rejected ballots were true vote fraud rather than voter accidents, you’re concluding that since vote fraud prosecution were 0, the authorities missed 50,000 fraudulent votes. That’s enough to show that successful prosecution is pretty much impossible.

Of course, when we have 70 million absentee ballots, 50,000 fraudulent ballots is tiny by comparison. So is our working estimate of 329,614 out of 70,551,227 absentee ballots, an attempted fraud rate of 0.47%. Absentee ballots were 66 out of 203 million total votes, 1/3 of the total, so the overall fraud rate for all votes would be about 0.16%.

Do rates that low make a difference? The 0.16% rate is for attempted fraud, not actual fraud, of course, and we don’t know whether all the fraudulent votes were for Biden, though I bet they were. Let’s take 0.16% for the sake of argument (which is 0.0016), and ask if that many of Biden’s votes were false, whether that would swing the election to Trump. These are not vote switches— Biden votes that should be counted for Trump— just invalid votes, so what matters is Biden’s percentage lead in a state.

Here they are, by Trump’s strength:

STATE ELECTORAL DEM REP MARGIN

Georgia 16 votes 49.5% 49.3% 0.2%

Arizona 11 votes 49.4% 49.1% 0.3%

Wisconsin 10 votes 49.6% 48.9% 0.7%

Pennsylvania⁷ 20 votes 50.0% 48.8% 1.2%

Nevada 6 votes 50.1% 47.7% 2.4%

Michigan 16 votes 50.6% 47.8% 2.8%

As you can see, if 0.16% is our fraud estimate, Trump still doesn’t win any of these states. Biden won 306 to 232, so Trump still loses the election so Trump needed to switch 37+ electoral votes. He’d need Georgia, Arizona, and Wisconsin, which would switch exactly 37, making it a tie and sending the election to be decided by the House of Representatives, where most likely Trump would have won.⁸

So do I admit that Biden won? No, though I’m now unsure. I chose numbers that I thought an open-minded Democrat would accept for our estimate of 0.16%, not my best guess as to the number. As I said earlier, I’d put the number higher. It would need to be higher than 0.7% for Trump to win. This reinforces something I said in an earlier post, however: I think reasonable men can differ as to who really won the election, and I would have certified it, unhappily, if I’d been the certifier. What I do insist on, even for my Democratic friends, is that (a) absentee voter fraud is in the ten of thousands of votes, and (b) fraud was what kept the electoral college vote from being close.

The sequel to this article is:

Footnotes

1. “ELECTION ADMINISTRATION AND VOTING SURVEY 2020 COMPREHENSIVE REPORT A REPORT FROM THE U.S. ELECTION ASSISTANCE COMMISSION TO THE 117TH CONGRESS”. See also the MIT-Stanford Healthy Elections Project report and website, which use the EAC data but add more data sources too. Their website has a wonderful interface for looking at data state by state.

2. FiveThirtyEight has an article saying what a wonderful sign it was that absentee ballot rejection rates fell from 2016 to 2020. It’s not wonderful as far as fraud goes. We would expect fraudulent votes to have a zero rate of missing the mailing deadline, since the fraudster is a professional and knows enough to mail his ballots early. If we add a million fraudulent votes to the legitimate ones, we’d expect the rejection rate to fall because none of the fraudulent ones will be late but some of the legit ones will be.

3. On September 29, 2020, the California Secretary of State issued emergency regulations that said, “the voter's registration record shall only be rejected if two different election officials unanimously find beyond a reasonable doubt that the signature differs in multiple, significant, and obvious respects from all signatures in the voter's registration record.”

A July 2020 Nevada rule allowed a signature to be rejected only if it differed in “multiple, significant and obvious respects from the signatures of the voter available in the records of the county clerk.”

4. In South Carolina and Indiana the judge said that unless cure procedures were added to the state rules, it would be unfair to the voter to reject a vote purportedly from him.

The Fifth Circuit Court of Appeals and U.S. district courts in Tennessee and Arkansas, on the other hand, upheld rejecting mismatched signatures without providing a cure process for the voter.

5. Wisconsin did an audit. They sampled 14,710 absentee ballot envelopes, and found that only 3 of these were missing signatures, .02%. If this was done fairly, it says that election officials let only a few ballots through without signatures. It’s easy for an election worker to notice that there’s no signature, so we wouldn’t expect many of those mistakes (or people getting away with voting both in person and absentee) unless the election workers were in on the fraud. The audit does not seem to have checked for mismatched signatures.

It is common for states to do audits of their machinery after an election, to see if the votes are counted correctly, but that’s different from a signature audit. I haven’t heard of them doing audits of signature matches for 2020 though (by 2024 some do), except for Wisconsin, Arizona (prompted by a court), and Georgia. In Georgia there was strong resistance from the Secretary of State, who said an audit would be illegal without a court order requiring it. Eventually he did do the audit, for Cobb County, and he found only two mismatched signatures, a man’s wife signing for him, and one signing in the wrong place. That’s too good a result, and that Secretary of State couldn’t be trusted to do it, and he did it with just state employees, so I don’t believe it. There are always going to be one or two cases of pretty much anything. The Brennan Center has an article on audits where it severely criticizes states for doing anything more than the standard tabulation audits after the 2020 election.

Also, fraud gets more complicated. My guess is that most fraudulent absentee votes would be easy to catch, because imitating 100 different signatures is difficult. The fraudster has to hope the election worker is lazy or wants Biden to win. But laziness and being willing to turn a blind eye are common vices. In some (most?) states, it seems pollwatchers from both parties can observe absentee ballot envelopes being opened and can challenge signature matching. If this were universal, and the parties actually did it, it would allay my concerns greatly. Probably in 2024 the Republicans will be more careful about this.

6. Caveat: I don’t know if some of the rejected ballots were “cured” by voters getting to try again with improved signatures. 18 states allow curing, though a lot of it is, I think, election officials filling in minor missing information such as zip codes. I also couldn’t find any information on how many ballots were cured. If 100% were cured in those states, we could breathe easier. Probably the data is not available. The EASC collects the data. On the 2020 questionnaire it sent to state officials, I don’t see any mention of curing.

7. Pennsylvania doesn’t have signature verification, so we should add 0.3% or so to the 0.5% fraud, but that still doesn’t win it for Trump.

8. Here, I’ll attach as a footnote some emails.

(a) I sent this one to the scholars leading the MIT-Stanford project, cstewart@mit.edu, npersily@law.stanford.edu, mitelectionlab@mit.edu. It may be useful to people wanting to go deep into my sources.

Dear Professors Persily and Stewart and MITElectionLab (can you forward to Declan Chin?),

I’m a retired law-and-econ professor at Indiana U. who wrote a game theory book you might know. I have a couple of Substack posts on 2020 election fraud and why I will vote for Trump that are nonscholarly but might be useful to you for understanding conservative attitudes. I’m writing now, though, because I am writing another post on the amount of fraud in absentee voting, and because I found a link mistake on your excellent website.

(1) Many people say the amount of fraud is trivial. Very little is caught and punished; even the good Heritage website really doesn’t have much. Another approach, though, is to look at failed attempts. Using the 2022 EAC report at https://www.eac.gov/sites/default/files/2023-06/2022_EAVS_Report_508c.pdf, 35,718,700 absentee votes were counted and 549,824 rejected in the 2022 general elections.

26.9% were rejected for signature not matching or incomplete, which comes to 147,902.

3.6% were rejected for “Voter already cast another ballot that was accepted (by mail or in person)”, which comes to 19,794.

Those are reasons which look like fraud. Some of them are false rejections because people wrote their signature wrong or forgot to sign at all or forgot they had already voted absentee. But still, we must have thousands of attempts to cast fraudulent votes. And the numbers above are just from states that do signature matching; for a more precise estimate, the rate in states that do require matching should be applied to states that do not, in which case we also have a partial estimate of successful fraudulent attempts.

Is there anything wrong with this idea? I am surprised not to have seen it used before, by someone like Han von Spakovsky at least (but he is not a data guy, I think).

If I have time, I’ll also apply the idea to the 2020 election. In my unreliable memory of numbers so far, if the unsuccessful signature-matching rate of .30% in states that do matching is used as an estimate of *successful* fraud (though only Penn in the swing states didn’t do matching), then it would make Trump win or come extremely close in Wisconsin, Arizona, Georgia, and Nevada, but not in Michigan or Pennsylvania, resulting in an exact electoral tie if we give Trump those states.

Someone ought to look at the rate of rejections in those states rather than use the national average, and do something with that information and the rejection rate in 2016 and 2012. I haven’t drilled all the way down to find the actual data yet, which you don’t have available in the easy way you’ve done it for so much else (again: wonderful job!).

Anyway, I’ll send you the Substack when it’s done, but if I’m way off base, it would be great if I found out before I posted it.

(2) One defect of the otherwise very good Declan Chin report at https://elections-blog.mit.edu/articles/deep-dive-absentee-ballot-rejection-2020-general-election is that it doesn’t mention “Voter already cast another ballot that was accepted (by mail or in person)”. This is of substantial interest in any discussion of fraud and verification methods, even though it’s a single-digit rejection rate.

(3) Another defect in that very good report, and in the data portion of your site too, and in the original EAC data, is that “rejected because of signature” is ambiguous. Does this mean the number rejected as ballots arrive, or the number rejected after the voter has a chance to cure his ballot in states that allow curing? Also, how many ballots are cured? This is another crucial number for thinking about voting law reform. And, it affects my estimate of fraud, perhaps by a lot, which is why I thought of it.

(4) The bad link. On your website, the Methodology report should be featured more prominently, I think. It is indeed mentioned, but the reader has to drill down pretty far, to the state-level reports from the superduper database page. On every one of those, I bet, the link is wrongly written as https://elections.mit.edu/2016-epi-methodology.pdf, a bad link. It ought to be https://elections-blog.mit.edu/sites/default/files/2020-08/2016-epi-methodology.pdf, which I found only by googling for “2016-epi-methodology.pdf”.

An example of one of those bad state pages is

https://elections.mit.edu/#/data/map?view=indicator-profile&indicator=RABP&year=2022

I know from my game theory book errata that people usually don’t bother to contact the author on things like this, but I wanted to as a small act of supplementary service and gratitude for your public service in providing such a good website.

IHS,

Eric

Prof. Eric Rasmusen (retired)

Rasmusen.org 

Erasmuse61@gmail.com, Cell: (812) 345-8573 (Eastern) 

(b) I sent this one to the EASC, which collects election mechanics data from the states. Their 2020 data is available, and it is clear from looking at their website that they have a smart boss who knows what he’s doing. You can get the data in comma separated form (csv), and they have a Microsoft app version that is supposed to be easy to use.

Dear EASC, 

   I am writing about three things: 

(1) I wrote a Substack article on absentee ballot fraud, arguing that there is a lot of it because EASC reports so much signature mismatch, missing, and double voting. https://ericrasmusen.substack.com/p/329614-election-crimes-were-detected  If I've made any blunders, I'd very much welcome your corrections. 

(2) One ambiguity that troubled me for that article is whether "rejected" means permanently rejected, or also includes some ballots that were cured. Your 2020 questionnaire isn't clear about that. Any ideas? For your 2026 questionnaire, you might want to clarify. 

(3) Also for 2026: it would be useful to have some curing questions. This is a big policy issue for the states, and it helps in knowing how much fraud there is too. 

Thank you for your excellent data. I use it via the MIT-Stanford project.