In 2012, a wide swath of conservative pundits — Karl Rove, Dick Morris, Michael Barone, Jennifer Rubin, among others — insisted that, polling evidence to the contrary, Mitt Romney was the clear favorite to win the presidential election. These arguments, which had varying levels of plausibility, flamed out in the most humiliating possible fashion as Rove raged on camera against the Fox News decision desk. The right-wing impulse to deny polling was one of the seeds that eventually sprouted into full-bloom Trumpian election denial.
Disturbingly, some liberals are now engaging in similarly delusional thinking about opinion polls.
Earlier this month, when a Wall Street Journal poll found Donald Trump and President Biden tied, liberal activists questioned the results on the grounds that one of its sponsors was being paid by Trump:
The New Republic published a story, headlined “That Big Poll Showing Trump and Biden Are Evenly Matched? Trump Helped Pay for It,” validating this complaint. These critiques treat the fact that the Journal poll was conducted in part by Tony Fabrizio, a Republican pollster who worked for Trump in 2016 and who is working for Trump’s super-PAC now, as some kind of discrediting scandal. The more banal fact is that the poll was jointly conducted by a Democratic pollster (Michael Bocian) and a Republican one (Fabrizio). It is very normal for pollsters to do polling work for political candidates as well as media organizations, and in no way indicates the poll was biased.
A couple days later, when CNN released a poll showing Trump leading Biden by a point, liberal accounts began questioning the results because the poll contained an “oversample” of Republican voters. In a Vanity Fair column, Molly Jong-Fast attacked this as transparently biased:
It was polling that put Clinton ahead of Trump by one to seven percentage points in the popular vote. It was polling that told us there would be a “red wave” in 2022, whose nonexistence was summed up by the New York Times postmortem: “The skewed red-wave surveys polluted polling averages, which are relied upon by campaigns, donors, voters and the news media.” And it was polling that told us, even as far back as 1995, that Bob Dole would wipe the presidential floor with Bill Clinton. But nearly 30 years later, like Lucy and the football, we are yet again treating junk polls as gospel.
Take, for example, a recent CNN poll, which consisted of 1,503 respondents who reflected “an oversample of…898 Republicans and Republican-leaning independents.” This poll was used as the basis of a New York Times story, which proclaimed, “Party leaders have rallied behind the president’s re-election bid, but as one top Democratic strategist put it, ‘The voters don’t want this, and that’s in poll after poll after poll.’”
Now, it’s worth asking how a poll like this even makes it into the paper of record. Well, it’s simple: Republicans work the refs.
The “oversampling” critique is even sillier than the accusation that the Journal used a Trump pollster. Polls routinely employ oversamples of groups whose views they want to examine. The reason is that they need a large enough sample to obtain a statistically valid number of respondents. Suppose they want to analyze how, say, Latinos (or Midwesterners or college-educated women) plan to vote. A national poll of the entire electorate won’t usually produce enough Latinos to yield a valid finding. So the poll will obtain extra results from Latinos to give it a valid subsample.
Crucially, the national poll will re-weight the subsample. So even though the poll will have more replies from Latinos, they will not have extra weight in the national result. The entire purpose of oversampling is to have a valid analysis of the selected cohort.
CNN’s poll conducted an oversample of Republican voters in order to analyze how they intend to vote in the presidential primaries. There was one national poll testing the general election, and a Republican oversample testing the Republican primary. The general-election poll was re-weighted to correct for the oversample. It did not have excess numbers of Republicans in the national sample. The reason this poll makes it into the paper of record is not, as Jong-Fast claims, because Republicans have bullied them into it, but instead because the poll uses valid methods.
Jong-Fast’s column tosses in fuzzier complaints about polling that have also gained wider purchase in the media. She attacks polls in 2016 for having “put Clinton ahead of Trump by one to seven percentage points in the popular vote.” That finding wasn’t terribly inaccurate, given that Clinton won the popular vote by 2.1 percent, a result that is in fact between one and seven points, albeit not in the dead center of the distribution of predictions.
She likewise blames pollsters for having falsely predicted a red wave in 2022. This has become conventional wisdom, not just from liberals like Jennifer Rubin, now on the blue team, who claims polls “misled voters about the fictitious red wave in 2022,” but even, as Jong-Fast points out, the Times news pages.
The reality is that polls were highly accurate in 2022. It was journalists who predicted a red wave in 2022, using history and on-the-ground reporting, ignoring the polls.
It is true that polls are far from perfect. Individual polls are frequently outliers. More than a year away from an election, polling has somewhat limited predictive value, and even on the cusp of an election, the polling averages can sometimes significantly miss the mark (as was the case in 2020).
But the idea that mainstream polling is corruptly slanting its results because it’s being paid for by Trump, or deliberately polling too many Republicans, is ludicrous. To see that kind of paranoia that has long since overtaken conservatism taking root on the mainstream left is a disturbing sign.
