"A man who’s made a living developing fraud detection algorithms has discovered a curious phenomenon: counties that started using Dominion Voting Systems machines have on average moved by 2 to 3 points from the Republican to the Democrat presidential candidate compared to counties that didn’t adopt the machines. The difference persisted even after he controlled for a number of factors, including county population and various demographic characteristics.

“I recommend we audit the machines,” he concluded.

The man is Ben Turner, who used to be the chief actuary at Texas Mutual Workers’ Compensation Insurance. He now runs Fraud Spotters, a consultancy specializing in detecting insurance fraud.

The Epoch Times replicated his initial analysis using the same data he used and arrived at the same results.

Turner told The Epoch Times he’s not making any specific allegation against Dominion; he’s simply reporting what he found.

Dominion didn’t respond to a request for comment. The company has for years been criticized over alleged vulnerabilities in its system. It has previously denied its machines can be used to switch votes from one candidate to another.

Turner reached his conclusion despite being initially incredulous about the claims that the Dominion machines affected the election outcome.

While a Republican himself, virtually all of his colleagues and peers are Democrats, he said in a phone call. The outrageousness of the election fraud allegations made him question whether he was on the “wrong team.”

“I don’t know how to process these allegations,” he said in a recent blog post. “They sound so extreme and far-fetched that they are easy to blow off as ridiculous.”

But he also saw the very mainstream concerns about election systems security, pointing to an HBO documentary released earlier this year on the topic.

Given his expertise, he decided to look into the claims just to get some clarity for himself.

Eventually, however, he realized he needed to go public.

“I kind of was too afraid to go public because I knew that it won’t help me in terms of, you know, half the population now thinks that I’m a tin-foil-hat person. And I’m aware of that,” he said. “But I just decided that if nobody can prove me wrong, it’s not fair for me to withhold this information and that it belongs in the hands of the public and, you know, if I’ve done something wrong with the analysis, then maybe somebody out there will prove it wrong, and that’s fine with me. I’d rather get to the truth than be right. I felt like if I didn’t publish this, I’m a coward.”

He looked at how, county-by-county, election results changed between the 2008 and 2020 presidential races, measuring whether adoption of Dominion would have any overall effect on the changes. He picked the 2008 election because at the time, only New York State had widely adopted Dominion, according to data from VerifiedVoting.Org. He excluded New York from the analysis, leaving him 657 counties that have adopted Dominion and 2,388 that have not as of 2020.

He found that Dominion use was associated with a 1.55 percentage point decrease in the Republican vote and 1.55 percentage point increase in the Democrat vote in the presidential race.

He controlled for the differences in county population, number of votes cast, urban/rural population split, population growth, international immigration rate, low education population, high “natural amenity” areas, high “creative class” population, and manufacturing dependency.

Later, he added seven more control factors to the analysis, including race, voter preference, and population. Not only did the “Dominion effect” persist, but the probability of his results being a fluke decreased.

The effect somewhat decreased, to a 2.84-point shift, when he also controlled for age.

The probability that the results occurred by chance or due to some unobserved factor was about 1 in 1,000, the analysis showed.

In his estimation, the results are significant enough to potentially flip elections in at least four states: Georgia (Biden margin 0.6 point), Arizona (Biden margin 1.2 points), Wisconsin (Biden margin 4.2 points), and Nevada (Biden margin 5 points).

Although his analysis indicated on average around a 3 point “Dominion effect,” it may be higher in some states and lower in others, he said.

“If you believe the Dominion effect is real, it is not hard to believe that this effect would be greater in swing states and could have swung these four states into Biden’s column, putting the electoral college in his favor,” he wrote.

The effect dropped to less than a 2-point blue shift when Georgia was excluded from the analysis.

“I believe this is indicating that Georgia is the strongest case for auditing,” he wrote in an FAQ article addressing questions about his work.

He showed his findings to about 10 people with statistical expertise, two of whom provided him substantial feedback. Neither could find fault with his analysis, he said, and neither wanted their name associated with it.

The Epoch Times spoke to one of the two, who ran a slightly different statistical model on the same data arriving at nearly the same results. The expert concluded Turner’s model was “reasonable,” given the limitations of the data.

After he went public, more people contacted Turner with further questions, suggesting, for example, more extensive controls.

Ultimately, he ran the analysis with about 100 controls. The effect somewhat diminished, to about a 2-point shift toward Biden. He said that at this point the statistical model started to suffer from “multicollinearity,” meaning many of the control factors were overlapping, which caused the model to provide less reliable results for any single variable, including the “Dominion effect.”

“Once we start to overfit the model it gets complicated to decipher, and it would take some time to sort out the issues, and even then it may become a matter of opinion,” he said via email.

He said that the results would be safer to report as a range.

“I would say the national average effect appears to be somewhere between 1.0% and 1.6%,” he wrote.

Turner used prior election data from MIT Election Data Science Lab, current election data from Politico, Dominion adoption data from VerifiedVoting.Org, and demographic data from the Department of Agriculture.

The data isn’t robust enough to narrow the analysis down to any particular state, he said.

The findings aren’t enough on their own to “win a court case,” he said. A statistical analysis, no matter how sophisticated, can be wrong. But he argued the four states should allow Republicans to audit the machines.

If this was a suspected insurance fraud case, the results would warrant a probe, he said.

“If this were an insurance company operation, we would have all these investigators investigating the people that we thought were behind the fraud.”

The Epoch Times asked Walter Mebane, statistics professor and election fraud expert at University of Michigan, to review Turner’s findings. He declined.

“Sorry, but I have no time for this,” he said in an email. “The assumptions that everything is the same over time except the machines are ridiculous.”

While turner didn’t control for more specific demographic changes over time, such as by age and income, he did control for overall population change.