By James D. Agresti
December 15, 2016
The issue of voter fraud was one of the most heated sources of controversy during the 2016 presidential election, and it continues to be so.
After Hillary Clinton’s campaign announced that it was supporting recounts in several states won by Donald Trump, Trump responded with a series of Twitter posts accusing Clinton of hypocrisy for refusing to accept the results of the election after she insisted that he “must.” He then tweeted, “In addition to winning the Electoral College in a landslide, I won the popular vote if you deduct the millions of people who voted illegally.”
Several major media outlets pounced on Trump’s comment. The New York Times, for example, reported that “virtually no evidence of such improprieties has been discovered.” The Times editorial board then called Trump’s statement “a lie,” and the Washington Post’s Fact Checker declared “this is a bogus claim with no documented proof.”
These media reports and Trump’s comment are all misleading. There is material evidence of substantial vote fraud, though it does not prove that Trump would have won the popular vote if such fraud were prevented. It only shows that this is a possibility.
This evidence is documented in a 2014 paper published by the journal Electoral Studies. Based on survey data and election records, the authors of this paper found that the number of non-citizens who voted illegally in the 2008 election ranged “from just over 38,000 at the very minimum to nearly 2.8 million at the maximum.” Their “best estimate” is that 1.2 million or “6.4% of non-citizens actually voted.”
As detailed below, this paper has uncertainties that the authors readily acknowledge, but attempts to dismiss it have been flawed and deceitful. Moreover, there are several reasons why significantly more non-citizens may have voted in the 2016 presidential election than in the 2008 election.
The Electoral Studies Paper
In 2014, the academic journal Electoral Studies published a paper by three scholars who analyzed the results of a large survey conducted by a group at Harvard University. This study also made use of polling data from YouGov and voter registration and turnout data from Catalist, a firm that equips “progressive organizations” with data to help them “persuade and mobilize” people.
In this 2008 survey of 32,800 respondents, 339 identified themselves as non-citizens, and 38 of these non-citizens checked a box that said “I definitely voted” in the 2008 general election or were recorded in the Catalist database as voting in that election. At face value, this means that 11.2% (38/339) of non-citizens voted in the 2008 election.
Applying this 11.2% figure to the Census Bureau’s estimate of 19.4 million adult non-citizens in the U.S., this amounts to 2.2 million non-citizens who voted illegally in the 2008 election. After weighting these results and accounting for margins of error, the authors estimated that a maximum of 2.8 million non-citizens voted in 2008.
On the low side, the authors noted that only five non-citizens who said they voted were recorded in the Catalist database as voting. If these were the only people who voted, it would mean that 1.5% (5/339) of non-citizens voted. Applied to 19.4 million adult non-citizens, this amounts to 290,000 votes. After weighting these results and accounting for margins of error, the authors estimated that a bare minimum of 38,000 non-citizens voted in the 2008 election.
Using other data from the survey, the authors refined their high and low estimates to produce a “best guess” that 6.4% or 1.2 million non-citizens cast votes in 2008. The survey also showed that 81.8% of non-citizen voters reported that they voted for Obama.
As the authors explain, these figures are “quite substantial” and “large enough to change meaningful election outcomes, including Electoral College votes and Congressional elections.” More specifically, they noted that “non-citizen votes could have given Senate Democrats the pivotal 60th vote needed to overcome filibusters in order to pass” Obamacare. This is because Democrat Al Franken of Minnesota captured this 60th seat:
with a victory margin of 312 votes. Votes cast by just 0.65 percent of Minnesota non-citizens could account for this margin. It is also possible that non-citizen votes were responsible for Obama’s 2008 victory in North Carolina. Obama won the state by 14,177 votes, so a turnout by 5.1 percent of North Carolina’s adult non-citizens would have provided this victory margin.
In the 2016 election for North Carolina’s governor, the current Republican governor recently conceded defeat based on a shortfall of about 10,000 votes. The Census Bureau’s estimate for the adult non-citizen population of North Carolina is 479,000 people. Hence, if 2.1% of them cast added votes for the Democrat, this supplied the margin of victory.
Trump currently trails in the popular vote by about 2.6 million. Hence, in order for his statement to be true, 12.6% of the 21 million non-citizen adults in the U.S. recorded by the Census Bureau would have had to cast added votes for Clinton. This is within the realm of possibility given that the study also found that “roughly one quarter of non-citizens were likely registered to vote” in 2008 and 2010.
Before the 2014 paper was officially published, two of its authors wrote an overview of it for the Washington Post. Criticism was swift and intense, and the Post placed links to four critiques of this article over the top of it, along with the authors’ reply to three of them.
Most of these criticisms were formalized in a paper published by Electoral Studies in 2015, which accused the authors of the original paper of “cherry-picking.” In the context of public policy, cherry picking means to selectively choose only the data that supports a certain conclusion while ignoring any data that does not. It is the equivalent of lying by omission.
This 2015 paper was written by three scholars, two of whom are managers of the Harvard survey cited in the study, and the third a manager with YouGov.
The central argument of their two-page paper is that all of the people in the survey who identified themselves as non-citizen voters either did not vote or were actually citizens. This argument rests on two flawed assumptions.
First, the critics assume that people who stated “I definitely voted” and specifically identified a choice of candidate did not vote—unless Catalist verified that they voted. This is illogical, because Catalist is unlikely to verify respondents who use fraudulent identities, and millions of non-citizens use them.
This is shown in a 2013 investigation by the U.S. Social Security Administration, which found that about 1.8 million illegal immigrants worked in 2010 by using a Social Security number “that did not match their name.” Furthermore, the study found that another 0.7 million illegal immigrants worked in 2010 with Social Security numbers that they obtained by using “fraudulent birth certificates.” Notably, a Social Security number is a common requirement for voter registration.
The Harvard survey and Catalist data evince such identity fraud, because 90% of all survey respondents were matched by Catalist, while most non-citizen respondents were not. In the 2008 and 2012 surveys, only 41% and 43% of non-citizens were matched by Catalist respectively. These low match rates are revealing given that the Catalist database contains reams of data on “more than 240 million unique voting-age individuals.” This amounts to 98% of the 245 million adults who live in the U.S.
Hence, to ignore all votes not matched by Catalist will ensure that most non-citizens are excluded. This is especially true of those who fraudulently use a Social Security number, who are the very same people who have an open door to voting.
Their second irrational assumption is that some citizens in the Harvard survey misidentify themselves as non-citizens, but non-citizens never misidentify themselves as citizens. Hence, they dismiss all votes by people who don’t claim to be non-citizens in two separate surveys. This has the effect of narrowing the field of non-citizens to only those who took the survey in both 2008 and 2010. It also excludes anyone who stated on one survey that they are a non-citizen and stated on another that they are a citizen.
The critics make a legitimate point that random errors by survey respondents will overcount non-citizens. This is because far more citizens were sampled in this survey. For instance, if a survey sampled 100,000 citizens and 100 non-citizens, and 1% of them misidentified themselves, this would mean 1,000 citizens called themselves non-citizens, but only one non-citizen said he was a citizen.
Such logic makes sense in a vacuum where all other evidence is ignored, but the reality is that misidentification of citizenship is not just a random phenomenon. This is because illegal immigrants often claim they are citizens in order to conceal the fact that they are in the U.S. illegally.
This is proven by a 2013 study published in the journal Demographic Research, which compared Census Bureau survey data on citizenship to the number of naturalized citizens recorded by the U.S. Office of Immigration Statistics. The study found that certain major groups of immigrants—including Mexican men of all ages, Mexican women aged 40 and older, and immigrants who have been in the U.S. for less than five years—frequently misrepresent themselves as citizens.
As a worst-case example, the study found that “the number of naturalized Mexican men with fewer than five years of U.S. residence is nearly 27 times higher” in the Census data than the number recorded by the Office of Immigration Statistics. In other words, only about 4% of Mexican men who claim to be citizens and have been in the United States for less than five years are actually citizens.
Now watch how the critics employ their flawed assumptions to claim that “the rate of non-citizen voting in the United States is likely 0.” Again, 38 respondents in the 2008 Harvard survey said they were non-citizens who “definitely voted” in the 2008 general election or were recorded in the Catalist database as voting in that election. Yet:
- instead of examining the 2008 presidential election, the critics focus on the 2010 mid-term election when the presidency was not at stake, and turnout was lower. In 2010, 489 people identified themselves as non-citizens in the survey, and 13 of them said they voted or were recorded in the Catalist database as voting. This cuts the number of voters from 38 to 13.
- then they dismiss anyone who did not also take part in the 2012 survey, which narrows the pool of non-citizens from 489 to 105, or by 79%.
- then they dismiss anyone who did not say they are non-citizens in both 2010 and 2012. This further narrows the pool of non-citizens from 105 to 85, leaving only 6 voters.
- then they dismiss anyone who did not appear in the Catalist database as voting, which cuts the number of voters in 2010 from 6 to 0.
The critics do this without spelling out the implications of their assumptions, without providing a comprehensive and transparent accounting of these numbers, and without mentioning that this was a mid-term election.
They also analyze the 2012 presidential election, and their methods are even more problematic. In this case, 695 people identified themselves as non-citizens in the survey, and 61 of them said they voted or were recorded in the Catalist database as voting. Yet:
- they dismiss anyone who did not also take part in the 2010 survey, which narrows the field of non-citizens from 695 to 105, or by 85%.
- then they dismiss anyone who did not say they are non-citizens in both 2010 and 2012. This reduces the number of non-citizens from 105 to 85. Note that the survey only asked 15 of these non-citizens if they voted in 2012, and 10 of them said they did.
- then they dismiss all 10 of these people, because they do not appear in the Catalist database as voting. Moreover, they do this while failing to reveal that all of these people specifically identified their choice for president—nine for Obama and one for Romney.
- then, buried in a footnote, they mention that one person who identified herself as a non-citizen in both the 2010 and 2012 surveys was matched by Catalist as voting in 2012. They say that this “appears to be the result of a false positive match with Catalist,” because the person “stated in both the 2010 and 2012 survey that she was not registered to vote.” This argument is based on the unspoken assumption that non-citizens would never lie about voting, even though such an admission could expose them to criminal penalties.
Throughout the body of their paper, the critics consider Catalist to be the only valid measure of voting, but when this does not serve their purpose, they dismiss Catalist in a footnote. Such duplicity pervades their analysis. They level the charge of cherry picking even as they engage in it.
Beyond all of the evidence above, the authors of the 2014 Electoral Studies paper have written a working paper that debunks their critics with many more facts.
The authors of the 2014 Electoral Studies paper repeatedly explain that there is room for uncertainty in their results. To that end, they provide a wide-ranging estimate for the number of non-citizens who voted illegally in the 2008 election.
However, one major aspect of their analysis does not quantify margins of error, even though it could be a large source of inaccuracy. This is the fact that the Harvard survey does not provide a truly random sample of the American public.
The Harvard survey uses data from an internet poll conducted by YouGov. The weakness of internet polls is that they are extremely vulnerable to selection bias or non-response bias. This occurs because certain people are far more likely to participate in these polls.
As explained in the textbook Mind on Statistics, “Surveys that simply use those who respond voluntarily are sure to be biased in favor of those with strong opinions or with time on their hands.” In other words, such polls are not based on random samples of people, and they can be misleading.
The Harvard survey attempts to correct for this flaw by using a process called “matching.” This entails selecting a portion of the survey respondents that “mimics the characteristics” of the target population. These characteristics include “age, race, gender, education, marital status, number of children under 18, family income, employment status, citizenship, state, and metropolitan area … religion, church attendance, born again or evangelical status, news interest, party identification and ideology.”
Matching is a common and generally accepted procedure for turning non-random samples into random ones. However, as the Harvard survey points out, matching relies on the “assumption” that there is “no difference” in how people would answer the survey if they have the same characteristics (like race, age, and news interest). This assumption may be false in some cases, because people can differ in ways that transcend such characteristics.
The authors of the 2014 Electoral Studies paper acknowledge this limitation by writing that their conclusions apply “if” the weighted survey data “is fully representative of the non-citizen population.” This is a big “if” given that the underlying data comes from an internet poll, even though it has been matched.
Another source of uncertainty is the fact that the study uses survey data from the Census Bureau to measure the number of non-citizens in the United States. As detailed above, this will produce an undercount of non-citizens, because many illegal immigrants conceal the fact that they are non-citizens. In the words of the Congressional Budget Office, figures for the number of illegal immigrants in the U.S. “are subject to considerable uncertainty.”
Non-citizen voters have incentives to misrepresent either their citizenship status or their voting status. After all, claiming to be both a non-citizen and a voter is confessing to vote fraud, and the Federal Voter Registration Application specifically threatens non-citizens who register with a series of consequences. … This possible penalty would tend to reduce the proportion of non-citizen voters who would report having voted.
The 2016 Election
The number of non-citizens who voted in the 2016 election may be significantly higher than in 2008, because:
- Trump campaigned on a promise to crack down on illegal immigration, and this may have driven non-citizens to vote against him.
- the number of adult non-citizens in the U.S. recorded by the Census Bureau has risen from 19.4 million in 2008 to 21.0 million in 2016.
- shortly before the election, Obama publicly stated that election records are not cross-checked against immigration databases and “there is not a situation where the voting rolls somehow are transferred over and people start investigating, et cetera.” This let non-citizens know that they stand little chance of being caught if they vote.
Likewise, early in 2016, the Obama administration supported a court injunction to prevent Kansas, Alabama, and Georgia from requiring people to provide proof of citizenship in order to register to vote.
So-Called Fact Checks
Some of the nation’s most prominent fact checkers have scoffed at Trump’s assertion that he won the popular vote if illegal votes are deducted.
The Washington Post’s Fact Checker dismissed Trump’s claim as “bogus” and attributed it to “a self-described conservative voter fraud specialist” who has “declined to provide any evidence to back it up, even though reporters have asked.”
The Post’s analysis, written by Glenn Kessler, completely ignored the fact that Trump’s statement is supported to a degree by the 2014 Electoral Studies paper. Kessler is clearly aware of this study, because he quotes its lead author and links to an earlier Post fact check that cites the study. Yet, Kessler doesn’t even hint at what the study shows. Instead, he provides a link that says “we’ve previously given Trump four Pinocchios for making a number of bogus claims about alleged voter fraud.”
Worse still, in both of these fact checks, the Post declares that Trump took the study “out of context.” This is a blatant falsehood, but Kessler says it is so because the lead author of the study wrote that “almost all elections in the US are not determined by non-citizen participation, with occasional and very rare potential exceptions.” This does not in any way contradict Trump, who quoted the authors of the study word-for-word as follows:
Non-citizen votes could have given Senate Democrats the pivotal 60th vote needed to overcome filibusters in order to pass health-care reform and many other reforms, and other Obama administration priorities. … It is also possible that non-citizen votes were responsible for Obama’s 2008 victory in North Carolina.
Kessler and his fellow Post reporter had good reason to know that these words are accurate and in-context, for the authors of the study wrote them in the Post, and the Fact Checker linked to their article.
PolitiFact, another popular fact checking organization, also published a misleading analysis of this issue. This pertains to the number of non-citizens who are registered to vote, which is another finding from the 2014 Electoral Studies paper. PolitiFact says that “Trump accurately cites the study” but is still wrong, because the study was “rebutted multiple times for the methodology it uses.”
PolitiFact then gives the distinct impression that the people who conducted the study are nobodies who merely wrote an article for the “Monkey Cage” blog of the Washington Post. PolitiFact does this by failing to mention that the study was published in a peer-reviewed academic journal and by failing to cite any credentials of the study or its authors, even though two of them, Jesse Richman and David Earnest, are university professors.
In stark contrast, PolitiFact touts the study’s critics with phrases like “three experts,” “peer-reviewed article,” “a political science professor,” “an election expert,” “an associate policy analyst,” and “experts who actually gathered the underlying data.”
PolitiFact’s analysis provides no indication that anyone in this organization read the body of the original paper, read the authors’ replies to their critics, or judiciously examined any of the attacks on the paper. It simply portrays the authors as unaccomplished and their critics as reliable.
This appeal to authority is especially deceitful given that two of the three “experts who actually gathered the underlying data” have made donations to left-leaning political causes. These are Brian Schaffner and Samantha Luks, who are among the three scholars who wrote the 2015 paper in Electoral Studies that criticized the original paper.
In 2004, Schaffner donated to America Coming Together, a liberal organization “heavily funded by billionaire George Soros” that was “on the cutting edge of national politics.” In 2016, Schaffner gave $250 to Hillary Clinton, and Luks donated to the Democratic Congressional Campaign Committee.
In sum, PolitiFact neglects the actual facts of this complex issue and makes it seem as if this is a case of “the experts” versus people with no credibility. That is not fact-checking but shilling for a particular point of view.
Contrary to the claims of certain major media outlets and fact checkers, a comprehensive analysis of this issue shows that substantial numbers of non-citizens vote illegally in U.S. elections. However, contrary to Trump, the data does not prove that he would have won the popular vote if this fraud did not take place. Instead, it only shows that this is a reasonable possibility.