What is feeding Trump's claims of widespread voter fraud?

President Donald Trump will seek a “major investigation into VOTER FRAUD,” the president announced Wednesday on Twitter. The statement came after a Monday meeting with Congressional leaders, in which Mr. Trump reiterated a claim that 3 million to 5 million illegal votes cost him the popular vote.

White House Press Secretary Sean Spicer has defended Trump’s claim, but neither he nor other administration officials have given substantiated evidence of widespread voter fraud.

But in a speech in Wisconsin last October, Trump zeroed in on a 2014 paper by three professors at Old Dominion University, which claimed, “More than 14 percent of non-citizens in both 2008 and 2010 samples indicated that they were registered to vote.” It’s easy to see why those findings would please Trump: In 2015, there were almost 23 million non-citizens living in the United States, none of whom were legally permitted to vote in federal elections. Fourteen percent of those could translate into more than 3.2 million potential fraudulent votes.

But that study has now come under criticism and provides an example of how academics’ debates and disagreements can get amplified in the political sphere.

“[The study's] finding is entirely due to measurement error,” Stephen Ansolabehere, professor of government at Harvard University, told FactCheck.org. Professor Ansolabehere serves as the principal investigator for the Cooperative Congressional Election Survey, a database of voter data used in the study Trump cited.

Two of that paper’s authors – professors Jesse Richman and David Earnest – summarized their methodology and findings in a 2014 Washington Post opinion piece:

“Our data comes from the Cooperative Congressional Election Study (CCES). Its large number of observations (32,800 in 2008 and 55,400 in 2010) provide sufficient samples of the non-immigrant sub-population, with 339 non-citizen respondents in 2008 and 489 in 2010. For the 2008 CCES, we also attempted to match respondents to voter files so that we could verify whether they actually voted.

“How many non-citizens participate in U.S. elections? More than 14 percent of non-citizens in both the 2008 and 2010 samples indicated that they were registered to vote. Furthermore, some of these non-citizens voted. Our best guess, based upon extrapolations from the portion of the sample with a verified vote, is that 6.4 percent of non-citizens voted in 2008 and 2.2 percent of non-citizens voted in 2010.”

They argue that this group, though small, had the potential to decide tight Congressional and Electoral College elections. They argued that non-citizen voters, who tend to favor Democrats, helped secure Al Franken’s victory in Minnesota’s 2008 senate race, giving Democrats the 60 votes needed to override a filibuster.

Along those lines, modest turnout of non-citizen voters would not have been enough to tip the popular vote in Hillary Clinton’s favor in the 2016 presidential election. But, if non-citizens indeed voted at the rates Professor Richman and Professor Earnest found in previous elections, they could have contributed to the narrow margins by which Trump won some swing states, such as Pennsylvania and Michigan. Striking their names from the rolls could boost his legitimacy.

But after reviewing the data, Ansolabehere says that non-citizen voting isn’t nearly as big a problem as Trump makes it out to be.

“Measurement errors happen. People accidentally check the wrong box in surveys. The rate of such errors in the CCES is very small, but such errors do happen. And when they do happen on a question such as citizenship, researchers can easily draw the wrong inference about voting behaviors. Richman and Earnest extrapolate from a handful of wrongfully classified cases (of non-citizens).... We asked people in successive years their citizenship. That minimizes the error. Upon doing so we find NO INSTANCES of voting among people stating consistently that they are non-citizens.”

Academics regularly critique one another's research and clarify erroneous findings. But when razor-thin elections prompt Americans to seek studies on voting, the process of careful re-evaluation can’t always keep up with the news cycle – or political developments.

After seeing his findings caught up in a larger political debate, Richman has downplayed on its significance. In his own statement to FactCheck.org, he argued that “our results suggest that almost all elections in the US are not determined by non-citizen participation, with occasional and very rare potential exceptions.”