The Morning Jolt


The Polls Aren’t Always Wrong

(Nick Oxford/Reuters)

The autumn begins in earnest, your life is about to be inundated with pumpkin-spice everything, and the presidential election kicks into a higher gear. Today we dispel the “the polls are always wrong and don’t mean anything” argument, looking hard at the aggregate polling in the swing states in the 2018 midterms. You can find a few surprises here and there, particularly in the Sunshine State, but by and large, pollsters knew what they were doing two years ago.

Psst! The State Polls Were Pretty Darn Accurate in 2018!

When we discuss the presidential polling in 2020, what should interest us the most is the state polling for Arizona, Florida, maybe Georgia, Iowa, Michigan, Minnesota, North Carolina, probably Ohio, Pennsylvania, Wisconsin . . . and that’s about it. Maybe New Hampshire and Maine’s Second Congressional District.

And perhaps the most important question is, do the state polls in aggregate give us a good sense of who is going to win in these key states? In a lot of the big statewide races in these states in 2018, the aggregate polling did give us a clear view of who was ahead.

In Arizona, the gubernatorial race two years ago was never that competitive; incumbent Republican Doug Ducey was always on course for a smooth win. Ducey led in the final RealClearPolitics average by 16.3 percentage points, and he won the election by 14.9 percentage points. The Arizona Senate race was much more competitive, with Martha McSally and Kyrsten Sinema trading leads, and the margins were small or were tied in the last few polls. McSally led the final RCP average by one point, and Sinema won the vote by 2.3 points.

Were the final polls showing a miniscule lead for McSally “wrong”? This is where the public’s idea of a poll being “right” and a pollster or statistician’s idea of a poll being right often differ. Every poll comes with a margin of error, which describes how close a pollster can reasonably expect a survey result to fall relative to the true population value. For more on the margin of error, see here. If a pollster says, “our final survey before the election shows a close race, with candidate X narrowly ahead of candidate Y,” and the final result is candidate Y beating candidate X over a narrow margin . . . that isn’t, statistically speaking, a wild outlier or “bad” poll.

Returning our attention to the state polls in 2018 . . .

Take a look at the polling in Georgia’s gubernatorial election in 2018. Other than a Trafalgar Group survey that put Brian Kemp up by twelve points, every other poll in the autumn pointed to a close poll — either Kemp up by a few points, Stacey Abrams up by one or two points, or a tie. Kemp won by 1.4 percentage points.

In the Michigan governor’s race, Gretchen Whitmer led every poll conducted in 2018, and the final RCP average gave her a lead of ten points. When all the votes were counted, she won by 8.5 points.

In the Michigan Senate race, incumbent Democrat Debbie Stabenow led every poll taken that year, although her lead shrank from a 20-some point margin to the upper single digits as autumn progressed. The final poll, by Mitchell Research, had her ahead by just three points; the Gravis poll, taken about a week earlier, had her ahead by 13 points. Her lead in the final RealClearPolitics average was 8.3 points . . . and she won by 6.4 points.

In Maine, the polling was pretty sparse last cycle. The Senate race was not expected to be competitive, and Angus King won by 19 points — in between the 13-point lead of the Emerson poll and the 27-point lead of the Pan Atlantic poll. In that state’s gubernatorial race, Janet Mills won by 7.6 percentage points — right around the eight-point lead she had in the Emerson and Pan Atlantic polls. In Maine’s Second Congressional District, the final three polls showed two ties and Democrat Jared Golden ahead by one point. Golden won by about one point.

In Minnesota, incumbent Democratic senator Amy Klobuchar cruised to reelection. In the governor’s race, Democrat Tim Walz led all the polls, with margins from five to 17 points, and he won by 11.5 percentage points.

In New Hampshire’s gubernatorial race, Republican Chris Sununu won by 6.3 points — but it’s worth noting that his four-point lead in the RCP average was a split between the University of New Hampshire showing a tie and Emerson showing an eight-point lead.

In Ohio’s Senate race in 2018, the polling had two distinct phases. Until early October, incumbent Democrat Sherrod Brown led by double digits, with the margin between him and Republican Jim Renacci often in the high teens. But the final two polls showed Renacci closing the gap; Emerson put Brown ahead by six, and Gravis put him ahead by nine. On Election Day, Renacci kept the margin respectable, a mere 6.4 points. The Ohio gubernatorial election makes a stronger case for the argument that some polling persistently underestimates the support for the GOP candidate. Democrat Richard Cordray enjoyed leads of three to six points in the final three polls, and Republican Mike DeWine won by a surprisingly comfortable 4.3 percentage point margin.

In Pennsylvania, Governor Tom Wolf led in the final RCP average by 19.6 points and won the final vote by 17.1 points. In the Senate race, incumbent Democrat Bob Casey led the final RCP average by 14.3 points, and he won the final vote by a 12.9 percentage point margin.

In the Wisconsin governor’s race, Democrat Tony Evers won by 1.4 percentage points over Scott Walker. The two factors in polling jump out here: first, how little polling there was — just four polls conducted in October — and how widely they ranged, from Evers leading by ten to Walker leading by one. But if you average out the margins of the last three polls, you get Evers ahead by . . . two points. Wisconsin’s Senate race always looked on track for a solid win by incumbent Democrat Tammy Baldwin. Her lead in the final RCP average was 10.6 points; in the election, Baldwin won by 10.8 points.

As you can see, in 2018, pollsters were generally in the ballpark in a lot of this year’s key swing states. Any “house effects” of one pollster were generally balanced out by the “house effects” of other ones. The defensive argument, “the polls are always wrong!” is a nonsensical denial put forth by those who aren’t willing to take the time to look at the numbers.

But there were a few races where the last couple polls were pretty far from the final results. In the 2018 gubernatorial race in Florida, only two polls put Ron DeSantis ahead of Andrew Gillum during the entire autumn. Most had Gillum ahead by a few points, and the final Quinnipiac poll put Gillum up by seven. DeSantis won by four-tenths of a percentage point.

In that year’s Senate race in Florida, the polls bounced around quite a bit, but incumbent Democrat Bill Nelson generally held modest four or five point leads ahead of Republican Rick Scott. In the final RealClearPolitics average, Nelson led by 2.4 points. Rick Scott won by two-tenths of a percentage point.

In Iowa, Democratic gubernatorial candidate Fred Hubbell generally enjoyed small leads over Republican Kim Reynolds — emphasis on small. His lead in the final RCP average was seven-tenths of a percentage point. On Election Night, Reynolds won by three points.

Could there be states where a Republican candidate can count on shy voters who are reluctant to talk to pollsters to put them over the top? Sure, the examples of DeSantis, DeWine, Reynolds, and maybe Scott point to cases where it can happen, erasing a gap of perhaps up to four percentage points or so. But it certainly isn’t a consistent effect found in every swing state, and a campaign would be foolish to assume it will finish well ahead of its final average in the late polls.

And now we come to this year’s state polling in those swing states. The news isn’t all bad for President Trump. He’s still ahead in the RCP average in Georgia, and he’s led every poll in Iowa this year. North Carolina looks neck-and-neck, a lot like last cycle. Ohio looked like a small but consistent Biden lead until the latest CBS News poll. Minnesota might be winnable. The notion that Texas is ready to flip to blue appears illusory.

But Trump leads in Arizona are rare, and the situation is the same in Florida and Michigan. New Hampshire doesn’t look close. Trump hasn’t led a poll in Pennsylvania since May, although one Rasmussen poll last month had Trump and Biden tied in that state. With the exception of one Trafalgar poll, Biden has enjoyed consistent leads in Wisconsin.

You can fiddle with the Electoral College map and still get Trump to 270 — if Trump gets just a little bump in Florida, Ohio, and Michigan, he’s almost there, and then he just needs one other decent-sized state perhaps Arizona, or Minnesota, or Wisconsin. But as autumn begins — and voting starts in ten days! — no one should have any illusions about the scale of the challenge facing the Trump campaign right now.

ADDENDA: You must read Kyle Smith’s thorough deep dive into Joe Biden’s record — full of details that even a consummate political junkie like me had either never encountered before or had long forgotten. I had forgotten Biden subsequently denied opposing the bin Laden raid, just how thoroughly he flip-flopped on crime legislation he had previously boasted he had written, or that Biden has not released any medical records in twelve years.

Kevin Williamson on Democratic efforts to lift the cap on state and local tax deductions: “a bunch of champagne socialists trying to sneak out of happy hour without paying their tabs.”

Our Kathryn Lopez with a surprising anecdote about Senator Ted Kennedy: “For a period of time in college, I would frequently be at Mass with Ted Kennedy at St. Joseph’s on Capitol Hill. He would quietly sit in the back and not go up to Communion. I always prayed for him when I saw him and respected the humility I saw there that wasn’t necessarily what I’d see on the Senate floor or a campaign rally. He was a man in his sunset years and only God knows what might have been happening in his heart.”


The Latest