A recent conversation with a friend led me to investigate, with some specificity, the origins and sources of some of the voter fraud claims from the 2020 presidential election. Isolated instances of fraud exist in every election. Occasionally, more systemic fraud exists, like North Carolina’s Ninth Congressional District in 2018. But to find such widespread fraud at scale, in so many states, strains credulity.

But this is a hard point to raise. Many cite “irregularities” as evidence of fraud. But I think “irregularities” are a starting point to investigate fraud. And what’s happened in our elections discourse is that, too easily, “irregularities” are treated as proof of fraud, and the investigation into those irregularities receives much less attention.

I drilled deep into one specific claim with a friend to make this point—a claim that had a thorough judicial examination (as many such claims never even make it to court for standing concerns, laches, or other procedural defenses of bars). Did 42,000 people vote more than once in Nevada in 2020?

Here’s the most recent popular origin of the claim, courtesy of testimony before the Senate Homeland Security and Governmental Affairs Committee, as relayed by Jesse Binnall, December 16, 2020:

The election was inevitably riddled with fraud and our hotline never stopped ringing. While the media and the democrats accused us of making it all up, our team began chasing down every lead. Our evidence came both from data scientists and brave whistleblowers.

Here is what we found:

Over 42,000 people voted more than once. Our experts were able to make this determination by reviewing the list of actual voters and comparing it to other voters with the same name, address, and date of birth. This method was also able to catch people using different first name variations, such as William and Bill, and individuals who were registered both under a married name and a maiden name.

This is a specific claim, and it allows us to trace its origin. Binnall’s lawsuit in Nevada state court, Law v. Whitmer, relies on the expert report of Jesse Kamzol. (An expert report is something you file with the court and with the opposing counsel to give them a chance to determine if your expert is reliable and so one, meeting all the rules of evidence.)

Here's an excerpt from Kamzol's report:

Duplicate Voter Registrations:

I identified 42,284 voters who appear to have voted twice in the 2020 general election. This list includes individuals that have the same name, address, and birthdate, and that voted multiple times. All these voters had multiple voter registration numbers associated with substantially the same name, birthdate, and address. There are a few variations within this list because of first name deviations, i.e. Bill and William, but all 42,284 voters have multiple data point matches including birthdates from which I conclude with a reasonable degree of certainty that these duplicate voters are each one and the same individual."

It’s worth opening that this allegation of double voting comes from duplicate voter registrations, which has two complications worth considering. First, are the voter registrations themselves actually duplicates? Second, is there evidence that those who appeared to vote twice actually do so?

At the outset, however, Binnall wrongly characterizes Kamzol's report. Kamzol qualifies with "appears," while Binnall does not. (Nevertheless, Kamzol does conclude with a "reasonable degree of certainty that these duplicate voters are each one and the same individual.")

Kamzol says that the list "includes" individuals "that have the same name, address, and birthday," while Binnall states that the determination is made "comparing it to other voters with the same name, address, and date of birth." Kamzol says that "multiple data point matches" are his parameters. It's not clear what "multiple data point matches" mean.

It’s not clear, because Kamzol's report is very brief and lacks explanation about his methodology. (Go check it out.) A good study published this year in the American Political Science Review recounts some of the difficulties in state data in trying to identify actual duplicates and the rigor that goes into assessing irregularities in what appear to be, on their face, duplicates. Dig into this for rigorous methodology--methodology not explained in Kamzol's report.

In short, Kamzol's report offers a beginning for investigation. It is not the end in itself. And when pressed under cross examination, he could not defend even this conclusion.

*

When the court issued its order, it did not reject Kamzol's testimony outright as a non-expert, as the defendants had requested. The court deemed him an expert. But the court did qualify that there was a problem of methodology:

The Court questions Mr. Kamzol's methodology because he had little to no information about or supervision over the origins of his data, the manner in which it had been matched, and what the rate of false positives would be. Additionally, there was little or no verification of his numbers. [Here the court cites Kamzol's deposition, which I cannot find online, which apparently concedes the lack of methodology, supervision, matching, false positive rate, or verification of numbers.]

Again, without explaining how you got to 42,000, it is difficult for a court to accept the figure of 42,000. And without even a single individual that they followed up with to determine its accuracy, it was, again, an absence of evidence:

109. Despite two of Contestants' experts testifying to 'questionable ballots' and 'illegal ballots,' Baselice Dep. 52:20-25 ('questionable ballots'); Kamzol Dep. 53:10-14 ('illegal ballots'), neither provided evidence to support Contestants' allegations regarding the presence of illegal votes in the 2020 presidential election. See Herron Dep. 59:22-60:12, 68:13--69:12 (testifying that neither Mr. Baselice nor Mr. Kamzol disclosed the data underlying their analysis); Baselice Dep. 24:7-15 (explaining that he did not participate in compiling the data he used and 'shouldn't even surmise' 'what the original source of the data was'); Kamzol Dep. 58:6-59: 15 (explaining that he did not know how the matching work to enhance the data he used was performed); Baselice Dep. 60:8-61:17 (acknowledging that he could not determine how many 'questionable' ballots were actually counted, contained votes in the presidential election, or were cast for a particular candidate); Kamzol Dep. 92:4-16 (same). Little or no verification of numbers was done by Mr. Kamzol.

So, circling back to Para. 109, in the deposition (read: the opportunity to cross-examine a witness), Kamzol could not determine how many double votes were cast. That would seem to run up against the 42,000 figure in the expert report.

(There was a supplemental declaration, but it was rejected as filed too late. You can see even within that, however, weak methodology:

Birth dates were not provided on the DMV file and name suffix (JR/SR/etc.) presence appeared to be inconsistent, so false positive matches of like-named family/household members are possible within these matches; however and again, these are all high to mid-high confidence matches that are reliable and merit further investigation.

I don’t think the supplemental declaration advances much in this regard.)

When it came to double voting, there was no evidence that "any Nevada voter voted twice." Here's the specific paragraphs from the court's opinion--and by "any voter," it means not a single one (at least, as put forth by the plaintiffs):

93. The record does not support a finding that any Nevada voter voted twice. See Doe 4 Dep. 10:6-13 (testifying that two voters he checked in were not allowed to vote because of record 16 that they already voted).

94. The record does not support a finding that any individuals were sent and cast [note: an important qualification] multiple mail ballots. Cf Negrete Deel. (LAW 001626) (hearsay declaration alleging that she received two ballots, one each for her married and maiden names, but not that she or anyone else cast multiple votes); Finley Deel. (LAW 004944) (hearsay declaration alleging that voter received two ballots, but providing no evidence that ballot was cast or counted).

*

A journalist at the Nevada Independent looked at a version of the double voting list, and it offers a different kind of explanation.

The Washoe voter file from Nov. 5 has 303,989 records; about 3,800 are duplicates. Some duplicates exist because of a change of address, as the following vote trace data shows (for a friend of ours who is building a new home).

They had changed their address at the DMV (which is linked to their voter file) to the one for the new home, but as the home was not completed in the run-up to the election, when ballots were mailed out to everyone, theirs could not be delivered to that address. The family showed up to early vote, and was asked to fix their address by the poll worker. You can see the challenge code “MAIL” in their record, meaning the mail-in ballot was undeliverable. You can also see an Early Vote Code for the ballot that was counted. You can also see the Voter ID duplicated as there are now two addresses associated with the voter. This type of duplicate record is considered a change of address.

Another type of duplicate record occurs because a person voted by mail, then voted again in person. Analysis shows that registered Democrats and Republicans have about the same rate of these duplicate vote entries.

In an effort to trace one of the duplicate votes, I interviewed a person whose mail-in ballot was collected from her home, but who was concerned that the person wouldn’t turn it in. She wanted to be sure her vote was counted, and appeared at a polling place in person, explaining the situation to the poll worker. In the file, I could see that the mail ballot did arrive — and that her in-person vote canceled out the vote by mail (as it should be).

That is, even at times where it was actually a duplicate in the database, a ballot was only cast and counted once within this system. That is, what might appear to be a duplicate vote is not actually a duplicate vote on close scrutiny, because there are checks in the system that prevent double-voting (even if a voter is mailed or receives multiple ballots).

*

It’s tough to get a concise look at a specific claim, given how many free-wheeling claims there are out there. But the origin of this story begins with an expert report that lacked methodology or underlying data; a report that was minimized and in some cases retracted on cross-examination; evidence deemed low value and low reliability by the court; and evidence lacking any specific backing upon scrutiny. Examination of the evidence offers explanations about some of the apparent irregularities or inconsistencies as misreading the database or misunderstanding the controls in place to ensure each voter casts a single ballot.

But when someone testifies before a committee of Congress about a hard figure like 42,000, that’s what gets picked up. Digging deep into even this claim takes hours of research to identify its origin and its weakness. One could do this with dozens of other claims floating out there. But I thought, given that I’d done the effort elsewhere, I’d share the origins of this one claim.