Read on Twitter
Since voting fraud is a hot issue at the moment, it's worth revisiting one of the cornerstones of the Trump Administration's claims about it—a peer-reviewed article that found that non-citizens not only vote but do so in droves. A quick thread. 1/n
The article, "Do non-citizens vote in U.S. elections?," came out in the journal Electoral Studies in 2014. https://www.sciencedirect.com/science/article/abs/pii/S0261379414000973 It concludes that "non-citizen voting likely changed 2008 outcomes including Electoral College votes and the composition of Congress." 2/n
As you might imagine, this was catnip for Republicans who wanted greater restrictions on voting. It became the foundation for Trump's claim that, absent massive voter fraud, he would have won the popular vote. 3/n https://twitter.com/realDonaldTrump/status/802972944532209664
Electoral Studies isn't a fly-by-night operation—it's a peer-reviewed journal, and a good one. So it seems as though Trump's claims have received the imprimatur of the relevant scientific community. 4/n
The problem with that argument is that science is a collective effort. No single researcher has a monopoly on truth. The peer-review process is meant to catch most problems, but it can't catch all of them. And in this case, it missed a whopper. 5/n
Fortunately, journals allow response articles. A year later, this one arrived. The flaw that it points out is subtle, but the implications for the original study's conclusions are profound. 6/n https://www.sciencedirect.com/science/article/abs/pii/S0261379415001420
The subtle problem is that, when you're dealing with very small groups of people in a survey (like non-citizens in the Cooperative Congressional Election Study that's used in the original article), even tiny amounts of measurement error can have a huge impact on findings. 7/n
If you're comparing the voting behavior of Republicans and Democrats, for example, a few people mistakenly reporting the wrong political party is no big deal. It'll only change the results by a tiny amount, because there are lots of Republicans and Democrats. 8/n
But comparing the voting behavior of citizens and non-citizens is a whole different matter. The expected number of non-citizens voting is zero (especially since the survey was meant to be just a survey of citizens; the non-citizen question was pro forma.) 9/n
Next, by the standards of surveys, the CCES is quite large—nearly 20,000 respondents. A typical poll that goes into the FiveThirtyEight polling average, by contrast, has 1,000-2,000 respondents. 10/n
And with 20,000 respondents, even a very, very low error rate in answering this question (say, 0.1%) has big implications, given that non-citizens in the sample are very rare to begin with. 11/n
If, hypothetically, you have 100 or so non-citizens in your sample out of 20,000 respondents, an error rate of 0.1% means that 20 "non-citizens" are actually citizens who clicked the wrong response by mistake. 12/n
Maybe that seems like a hard mistake to make. But it's pretty clear that some people did. The CCES is a panel survey, meaning that the same people take the survey across multiple years. And when you compare their responses... 13/n
...you find that, sure enough, 20 people who identified themselves as citizens in 2010 responded that they were non-citizens in 2012. A small number relative to 20,000 respondents, but a significant percentage of non-citizens. 14/n
Now, if misclassified citizens vote at the same rate as correctly-classified citizens, you'd expect to see about 13 voters among the non-citizens. You see far fewer than that in the survey—just four. So misclassification almost certainly explains all of them. 15/n
Four misclassified voters translates into just under a 1% rate of voting for people classified as non-citizens. But multiply that percentage by, say, 10m or so unauthorized immigrants, and boom, you get around 100,000 illegal votes. 16/n https://www.pewresearch.org/fact-tank/2019/07/12/how-pew-research-center-counts-unauthorized-immigrants-in-us/
Indeed, a subsequent study in the same journal concluded that "the expansive voter fraud concerns espoused by Donald Trump and those allied with him are not grounded in any observable features of the 2016 election." /fin https://www.sciencedirect.com/science/article/abs/pii/S026137941730166X
