When it comes to political polling, weight matters.
The pollsters who badly missed the national vote, giving Joseph R. Biden a double-digit lead in the final week, are the ones who generally did not “weight” the sample’s voters or underestimated Republican turnout, an examination by The Washington Times shows.
Instead, some firms went with how the surveyed voters identified themselves by party with no adjustment, or “weight,” based on previous elections. In 2016, Republicans accounted for 33% of the vote. But some 2020 polls predicted less than 30%. The actual 2020 GOP turnout: 36%.
Republicans are in open revolt against news media and university pollsters, saying 2020 inaccuracies hurt their voter turnout and prompted constant negative commentary by a national press obsessed with polls.
“After this election, I think the polling industry needs to take a hard look at what it does,” reelected Sen. Susan Collins, Maine Republican, told Fox News.
All polls had showed her losing.
A new Pew Research Center analysis made this industry criticism: “It’s clear that national and many state estimates were not just off, but off in the same direction: They favored the Democratic candidate.”
For example, Quinnipiac University’s final poll said Mr. Biden led President Trump by 11 percentage points. Quinnipiac’s numbers were not close. Mr. Biden is currently ahead in the popular vote by 3.7%.
Quinnipiac’s stated methodology does not weight party identification: It asked 1,516 likely voters with which party they identify. The poll sample produced 34% Democrat, 28% Republican and 28% independent.
The actual vote, according to news networks’ voter exit polls, was 37% Democrat and 36% Republican — an 8-point GOP increase.
The most accurate 2020 national poll came from a company that weights, or adjusts, voter party loyalty — Investors Business Daily/TIPP and its pollster, the New Jersey firm TechnoMetrica Market Intelligence.
It “weighted” the party identification sample to forecast what it believed would be the turnout. It predicted 36% Democrat and 33% GOP — the exact 2016 exit poll results.
TIPP’s final poll put Mr. Biden‘s lead at what would prove to be a highly accurate 4%, compared with his subsequent 3.7-point edge in the popular vote.
TechnoMetrica President Raghavan Mayur told The Washington Times that he considers weighting one of several key methodologies along with “multimodal” data collection — landlines, cell phones and the Internet — and coming up with an accurate voter-age distribution.
“I weighted party to 36% Democrat and 33% Republican at the registered voter level, based on my estimate of party ID,” Mr. Mayur said. “I am not sure if I will attribute it alone to our accuracy, but do acknowledge it was definitely necessary in a multimodal scenario.”
“I had a good age distribution,” he added. “If you rely only on the phone, you won’t have young people, and if you rely exclusively on the web, you won’t have older people. Landlines are important to get adequate age 65-plus, who won’t do a survey on cell.”
Pollster John Zogby also weighted the party turnout — 38% Democrat and 34% Republican. He pegged the race for control of the House of Representatives at a 2-point Democratic advantage when other polls projected a much larger gap.
The current actual spread: a 2.3% Democratic edge, according to Real Clear Politics. The surprisingly close vote is one reason Republicans gained at least nine House seats, while the major media predicted they would lose five or more.
Quinnipiac did not weight its surveys for party alliance and did not use online contacts, relying instead on landlines and cellphones.
Quinnipiac’s last polls had Mr. Trump losing Florida, 47-42. He won by 3.3%. The university foresaw Mr. Trump beaten in Ohio 47-43. The president won Ohio by 8.2 points.
Quinnipiac says there is a good reason they do not weight for party preference.
“We don’t weight to party ID based on past exit polls as it can change from election to election,” polling director Doug Schwartz told The Washington Times. “For example, in 2004 there was an equal number of Dems and Republicans, but in 2008 Democrats outnumbered Republicans by 7 points.”
Besides Quinnipiac, four other established polling firms said Mr. Biden enjoyed a double-digit lead in the campaign’s last week: YouGov, CNBC, NBC/Wall Street Journal and CNN.
CNN’s firm, SSRS in Pennsylvania, did not weight its polls for party. Its sample of voters broke out with 33% Democrat and 28% Republicans.
Likewise, Hart Research for NBC News/WSJ developed a sample that identified 44% Democrat and 39% Republican.
CNBC uses the northern California-based pollster Change Research, which conducts 100% online polling. On Nov. 1, it predicted a 10-point Biden win.
Change Research also said Mr. Biden would win six battleground states. It racked up mixed results, accurately predicting he would win Arizona and Pennsylvania.
But its calls on Florida (51-48) and North Carolina (49-47) were wrong. It had Mr. Biden taking Wisconsin by 8 points. The difference was much less — 0.7%.
It had him up in Michigan by 7 points. The result was 2.7 points.
Change Research’s stated methodology does not provide a party identification breakdown for its voter sample. The company does not use the word “weight.”
Instead, it says it did after-survey “stratification” based on the 2016 presidential vote.
YouGov’s web pages, which predicted a 10-point Biden win, said it weighted its voter sample based on the 2016 election but does not provide a party breakdown.
Pew Research’s Nov. 13 criticism of the polling industry said there are “systematic” flaws.
“The fact that the polling errors were not random, and that they almost uniformly involved underestimates of Republican rather than Democratic performance, points to a systematic cause or set of causes,” Pew said.
“Under this theory, it’s possible that the traditional ‘likely voter screens’ that pollsters use just didn’t work as a way to measure Trump voters’ enthusiasm to turn out for their candidate,” Pew added. “In this case, surveys may have had enough Trump voters in their samples, but not counted enough of them as likely voters.”
In other words, too few Republicans were included in the probable voter turnout models.
Mrs. Collins is a poster child for systematic failure. Not only did none of 14 polls find her leading, almost all predicted a substantial defeat. She won reelection by 9 points.
View original Post