• Ben
• April 22, 2021
• 86 Comments

A summary of problems/questions

All of the data he based his original analysis on is verifiably false.
He claims that his analysis produces the same result regardless of the data it’s based upon.
He has refused to disclose the “6th order polynomial” that produces his predicted ballots cast.
Improperly compares 2020 registration totals to 2010 Census data.
Arrived in Colorado knowing his data was bad and intending to present it regardless.

Once again, I find myself getting blown up with claims of statistical proof of voter fraud in Colorado. And once again, it’s simply a case of the underlying data being wrong.

Incorrect Data

Dr. Douglas G Frank has been giving presentations claiming he’s uncovered an algorithm that can precisely replicate turnout in each of Colorado’s counties during the 2020 election. Unfortunately, every single one of the underlying data points he presents is wrong. He may as well be analyzing March Madness scores for all the relevance his data has to the Colorado Election.

In an attempt to put this to bed once and for all I’ll break it down piece by piece below.

Below is a screenshot from Dr. Frank’s appearance on LindellTV. You can see from the screenshot this happens at the 17:32 minute mark.

Here, Dr. Frank is claiming that his algorithm almost perfectly predicts the number of ballots cast in Broomfield County in 2020. His prediction is in the green box, he predicted that 51,260 ballots should have been cast. In the red box is the number of ballots he claims were actually cast — 51,613. Dr. Frank believes 51,613 ballots were cast in Broomfield County in 2020. There’s just one problem — he’s the only one that thinks that! Both the Secretary of State and the Broomfield Clerk report that 47,103 ballots were cast. Here’s the link to the Broomfield Clerk report.

His estimate isn’t just off by nine percent, the number he’s using for verification is too! This isn’t an isolated incident, he’s using incorrect ballots cast figures as the basis for his entire analysis. You can’t get a real result when the underlying data you use is wrong.

At the 41:01 minute mark in the same video, they show Jackson County. Dr. Frank claims 1,004 ballots were cast there. But in reality, it was 889.

And again with Bent County. Dr. Frank claims 2,815 ballots were cast when in reality it was 2,295. This is a 22% error not just in his estimate, but in the data he’s basing his analysis on.

Cheyenne County — he claims 1,281 ballots were cast when it was actually 1,146.

Like they say “garbage in, garbage out.”

Even more bizarre is the fact that even the raw data Dr. Frank supposedly used (sorry the host removed the link after this blog was published) for his analysis doesn’t produce the figures he has been presenting.

The second act in this drama is all too predictable as well. It goes something like this; “Oh but Ben, we corrected the data and the analysis still holds!”

And that right there is what should really scare anyone who understands data. The fact that the analysis would work on fake data, real data, any data you give it, is proof that it never had any interpretive power to begin with.

Misrepresenting the Census

The “Census” data that Dr. Frank displays in the above graphs does not come from the real Census. The data that he shows is his own interpretation of the numbers. He is applying his own smoothing method to the 2010 Census to get estimates of population by age because the census does not produce this level of detail in their data.

Smoothing out the data isn’t so bad in itself. However, he fails to account for both inter and intrastate migration. People moving within and between states is a huge population driver and, as I’ve discussed before, accounts for almost half of Colorado’s voter registration growth in recent history.

As some observant commentators pointed out, this means that the “fraudsters” in Dr. Frank’s world weren’t using the Census data to control the registration as Dr. Frank claims. But, rather, they must have miraculously stumbled upon and used Dr. Frank’s exact same method of smoothing out the census data by age to perpetrate their fraud. That strikes me as pretty unlikely. …unlesss… he as in league with them from the start!

Once again, even if you give Dr. Frank the benefit of the doubt we still have some problems with his assertions.

Census Methodology

In thinking about the claim that some counties appear to have more registered voters than census population start by asking yourself this: why do you think that the Census data is innately more accurate than the voter file?

Wouldn’t a disagreement between voter registration and census data point to the census data being incorrect and not vice versa? After all, every voter registration record is tied to a named person whereas the Census is just a glorified poll.

You can read the methodology for the ACS survey here: https://www.census.gov/programs-surveys/acs/methodology.html
The more interesting elements are in the covered geography and sample size sections.

The Census doesn’t even produce real estimates for most counties in Colorado every year. As you can see from the Geography section of the above link, they only conduct polls in geographies with populations greater than 65,000 annually. That’s only the 12 largest Colorado Counties.

When you look at the Census data and see that Routt County had a population of 25,072 in 2019 what that actually means is that at some point between 2015 and 2019 a few people in Routt County answered the survey and they just used an algorithm to estimate the population for all those years in between.

Right off the bat, this is a problem for Dr. Frank because he’s using 2020 registration data and comparing it to population estimates that are guesses from 10 years in the past. If Dr. Frank were to compare apples-to-apples in the case of Routt he would need to be using a total registration number that’s about 20% lower than the 2020 registration totals he’s using. In a high population growth state like Colorado, this is a particularly bad problem.

The other interesting thing you can find on the methodology link is the sample sizes for each state. You can see that the sample the Census used for Colorado is around 30k surveys. That’s pretty good for a whole State. If that was all done in one shot that’d give a margin of error of less than 1% for total population. Unfortunately, when you consider that the Census only surveys the largest 12 counties every year that leaves a lot of margin for the small counties.

To return to the Routt county example, even if the Census got a representative sample of those 30k responses to come out of Routt in a single year, their best recent estimate would still be something like +/- 2,000 people for Routt County alone. Really little more than an educated guess.

In short. If you compare the real Census population estimates and voter registration for the same point in time registration, doesn’t exceed the population estimate’s error margin in any counties I’m aware of.

Presentation Inconsistency

Anyone who has actually watched Dr. Frank’s presentation on this will note that he’s careful never to show the actual equation that he uses to produce his turnout estimates. The equation is central to his claim because he states that a single “6th order polynomial” can reproduce the turnout rates by age for each county. You can watch his explanation here. This would be damning because it suggests that some ratio describing voter turnout behavior is identical across all counties. But Dr. Frank’s own presentation shows this isn’t even true. The graphs he shows clearly use different equations for each county.

Consider his claims for Eagle and Park County:

Does the distribution and gaps between those lines look the same to you? He keeps most of the data hidden but we can see that in Park he’s showing 18 year-old registration of 113, Ballots Cast of 110, and his prediction is 101.7. In Eagle County he’s showing 18 year-old registration of 488, ballots cast of 481, and his prediction is 493.36. As he explains, his “6th order polynomial key” only produces one number as an output and that number is the key to convert registration to ballots cast. But in his own presentation, he doesn’t show a single consistent ratio of… anything to anything. He’s clearly using different ratios for each county which would have had to been produced by two different “6th order polynomials.” He’s lying about how his forecast is derived.

The below table shows the numbers taken directly from the above screenshots. Next to those, I’ve calculated the ratio for just about every number to every number. Dr. Frank claims that the ratio of… something… should be the same across counties. But, as you can plainly see in the below table not a single one of these number relationships matches.

Deliberate Deception

The most troubling piece of all this is that it seems clear that Dr. Frank arrived in Colorado knowing that his data was bad and intending to present it anyway.

I’ve thoroughly documented the fact that I made Dr. Frank and his crew aware of the errors in his data two weeks prior to his April 23rd Colorado tour. However, the account of events from his supporters is that he didn’t even bother to begin correcting his data until this blog post was widely disseminated.

Why would Dr. Frank present data that he knew was false?

Claim: Turnout by Age was Uniform Across Counties

Dr. Frank is not claiming that the turnout rate (ballots cast / registered voters) is the same by age across counties. I initially thought that was the point he was making because many people who follow him more closely than I told me this is what his work showed. If you listen to his explanation it does sound like he’s saying turnout rate by age was uniform across counties. However, what he’s actually doing is more complicated so I don’t fault him nor his listeners for the misunderstanding.

For those who do think that Dr. Frank is claiming the turnout rate by age is uniform across counties, I leave this section:

This is quite simply not true. And it’s easy to prove this isn’t the case. Below is a graph you can play with that shows the turnout rate by age, by county in Colorado’s 2020 General Election. As you can see, there’s a ton of variation across counties. Just as we’d expect.

All of the data and code to recreate this graph are available here. And if you don’t believe me, well you can always get your own data from the Secretary of State here.

Data

• Raw data Dr. Frank used for CO his analysis. (Sorry, it was removed by the host after this blog was published)
• Raw data Constellation used for CO registration, ballots cast, and turnout.
• Python code Constellation used to compile data for turnout by age graph.
• Colorado Secretary of State data request page.