A new report from a conservative think tank says as many as 5.7 million Americans voted illegally in the 2008 election, but experts say the survey extrapolates too broadly and there’s plenty of reason to be skeptical of it.

The analysis by the think tank Just Facts is based on Harvard/YouGov data, the same source of data for a 2014 study used to support President Donald Trump’s claim that millions voted illegally in the most recent election. Brian Schaffner, a political science professor at the University of Massachusetts Amherst who helps manage the data source the 2014 analysis was based on, debunked that analysis in November, saying it extrapolated based on answers from respondents that were likely errors.

James Agresti, the president of Just Facts, told The Washington Times the details of his analysis were “technical” but his results were based on a more conservative analysis and were more accurate than those of the 2014 study.

In 2012, Schaffner and his colleagues went back and re-interviewed the 121 people from the earlier study who said they voted in 2010 but weren’t citizens. Of those 121 people, 36 changed their answer and said they were, in fact, citizens. Of the 85 people who maintained they were non-citizens, researchers could not match a single one to a valid voter record.

In an email, Schaffner said the conclusion of Agresti’s study was highly doubtful and readers should be skeptical for the same reasons they should have doubted the study of the 2010 data. 

“The most important is that the new study makes the same error as the old study in terms of ignoring measurement error on the question they use to identify supposed non-citizens when we have in fact demonstrated that many people answer that question incorrectly,” Schaffner wrote in an email.

A survey question asked respondents about their citizenship status in 2008 and whether they and their parents and grandparents were immigrants or were born in the USA. Separate survey questions asked whether respondents voted that year.

“In addition to ignoring the major issue with the original study, they also claim that we should take any supposed non-citizen at their word if they claim to have voted even if we can’t match them to a vote record because they probably used a fraudulent identity,” Schaffner continued. “However, the issue here is why would a non-citizen who is going through the trouble of using a fraudulent identity to vote then admit to voting in a survey and give us their actual name and address?”

Agresti defended his study in an email to HuffPost.

“As I’ve previously documented, claims that the samples are too small and based on response error are rooted in demonstrably false assumptions,” he said. “Likewise, the professors who authored the 2014 Electoral Studies paper thoroughly debunked those and other spurious critiques.”

Eitan Hersh, a political science professor at Yale, said the survey was drawing overly large conclusions based on the answers of just a few people.

“The way that they’re doing this is kind of through a crazy extrapolation,” Hersh said in an interview. “If I do a survey of 1,000 people and on that survey five people say something crazy or non-truthful, that wouldn’t be something hard to imagine, that five people out of 1,000 people might lie, might not have actually read the text very carefully, might click on a button wrong.”

“It’s extrapolating off of a survey item of maybe a few people answering the wrong way,” he continued. “If five people said this instead of three, do they think that’s the difference between 400,000 people illegally voting? Is that how seriously they take each person in a survey answering this question? Even if you’re not a survey researcher, that kind of seems preposterous.”

Schaffner said he was skeptical of Agresti’s claim his analysis was more accurate than the 2014 study. His argument, Schaffner said, “just sounds like some kind of wishful thinking special sauce that would not really stand up to scientific scrutiny.”

Trump has stood by his claim that millions voted illegally in the 2016 election and has convened a probe to look into the issue. Numerous studies have shown that while non-citizens occasionally do vote, it is not a widespread problem.