The size of the Democratic victory in Virginia has taken many on both sides by surprise — myself included.
My last pre-election column on the polls was headlined: “Final polls: Spanberger leads by a lot, Hashmi by a little, attorney general’s race down to the wire.”
Abigail Spanberger did win by a lot, but so did Ghazala Hashmi and, while the attorney general’s race was the closest of the three, it did not go “down to the wire” in the way many expected. Twice this century we’ve had recounts in attorney general elections, and this year was nowhere close to that.
Of the seven state elections in this century, Jay Jones’ margin was bigger than three of them. He won 53.0% to 46.6%, which is almost exactly what Mark Herring won in his 2017 reelection; he took 53.3% to 46.6% for the Republican candidate that year, John Adams. That 53.0% vote for Jones is also higher than four of our past seven governors have received.
So what happened? Why were the polls generally right in the governor’s race but seemingly wrong in the other two? Or were they wrong at all? Let’s explore!
Governor
Here are the last nine polls in the governor’s race. All showed Spanberger ahead, most by double digits. She won by 14.94 percentage points. All but one poll underestimated that. The poll that came closest was YouGov, which had her up by 15 percentage points. That poll was also in the field for an unusually long time, which made it suspect at the time. Short windows are preferred because public opinion can shift over time. Indeed, the whole point of campaigns is to influence public opinion. State Navigate also came close; it had Spanberger up by 13. Hold that thought.
Insider Advantage (Nov. 2-3): Spanberger 50%, Earle-Sears 40% (+10 D)
Quantus Insights (Nov. 3): Spanberger 53%, Earle-Sears 44% (+9 D)
Trafalgar Group (Nov. 1-2): Spanberger 50%, Earle-Sears 43% (+7 D)
Emerson College/The Hill (Oct. 30-31): Spanberger 55%, Earle-Sears 44% (+11 D)
Echelon Insights (Oct. 28-31): Spanberger 55%, Earle-Sears 43% (+12 D)
Atlas Intel (Oct. 25-30): Spanberger 53.9%, Earle-Sears 45.2% (+8.7 D)
State Navigate (Oct. 26-28): Spanberger 54%, Earle-Sears 41% (+13 D)
Roanoke College (Oct. 22-27): Spanberger 51%, Earle-Sears 41% (+10 D)
YouGov (Oct 17-28): Spanberger 57%, Earle-Sears 42% (+15 D)
Lieutenant governor
Frustratingly, many of the national pollsters skipped over this race, so we’ll go with just the final five polls here. All showed Ghazala Hashmi ahead; three of them had the race within the margin of error. She won by 11.26 percentage points. The only pollster that came close — indeed, very close — was Navigate, which had her up by 12 percentage points.
Trafalgar Group (Nov. 1-2): Hashmi 47.5%, Reid 45.9 (+1.6 D)
Echelon Insights (Oct. 28-31): Hashmi 49%, Reid 46% (+3 D)
Atlas Intel (Oct. 25-30): Hashmi 51.7%, Reid 45.6% (+6.1 D)
State Navigate (Oct. 26-28): Hashmi 53%, Reid 41% (+12 D)
Roanoke College (Oct. 22-27): Hashmi 42%, Reid 40% (+2 D)
Attorney general
Some national pollsters skipped this race, too, but not as many as skipped the LG’s race. I’m listing just the polls that came out in the final week but to really understand this race, we need to step back and take a longer view. Republican Jason Miyares was an incumbent, which helped his numbers, but not enough. Jones consistently led until the texting scandal in early October. At that point, the race flipped. Of the next 14 polls, Miyares led slightly in 11 of them (some within the margin of error) while one was tied. Only two showed Jones ahead — both by State Navigate. Then, things shifted again. Of the final four polls, two had the race tied; two had Jones slightly ahead, but within the margin of error.
In the end, Jones won by 6.4 percentage points, more than twice the most favorable polling for Jones in the final week. So what happened?
Insider Advantage (Nov. 2-3): Jones 49%, Miyares 47% (+2 D)
Quantus Insights (Nov. 3): Tied at 47% (Even)
Trafalgar Group (Nov. 1-2): Tied at 46% (Even)
Emerson College/The Hill (Oct. 30-31): Jones 49%, Miyares 47% (+2 D)
Echelon Insights (Oct. 28-31): Miyares 49%, Jones 46% (+3 R)
Atlas Intel (Oct. 25-30): Miyares 48.1%, Jones 46.6% (+1.5 R)
State Navigate (Oct. 26-28): Jones 49%, Miyares 46% (+3 D)
Roanoke College (Oct. 22-27): Miyares 46%, Jones 38% (+8 R)
State Navigate consistently came the closest
It got the governor’s race and lieutenant governor’s race almost exactly, and if you factor in margin of error, you can drop the “almost.” It also came the closest of any in the attorney general’s race. It was showing Jones ahead even when other polls were showing Miyares up. Even more remarkable, this was the first year State Navigate has done any polling. So how did it do so well?
State Navigate is the brainchild of Chaz Nuttycombe, a 26-year-old Virginia Tech graduate who is now based in Richmond. He says the key to State Navigate’s accuracy is its polling model.
All pollsters have to make some guesses about who will vote, but the electorate is shaped differently each election. Some constituencies turn out more heavily for some elections than they do others. Four years ago, Glenn Youngkin benefited from a larger-than-usual turnout of rural voters, which made that year’s electorate more Republican than it ordinarily might have been. Four years before that, Ralph Northam benefited from suburban voters, who often sat out state elections, coming out to voice their opposition to President Donald Trump.
Because it’s impossible to know exactly what the electorate will look like, many pollsters simply weight their samples to the makeup of the previous election — in this case, the 2021 governor’s race that had an unusually high Republican shape. That gave some of the polls a more Republican slant that we saw on Election Day. Others tried to mitigate that by using a mix of the pro-Republican 2021 numbers and the pro-Democratic 2017 numbers, or the 2024 president number, or some other formula. State Navigate simply had a better model that more accurately reflected what turned out to be our election reality. Unlike others, its sample was not weighted to a particular election but was a formula that Nuttycombe and his team — which included some former staffers from the now-defunct Five Thirty Eight data website — devised just for this year’s circumstances.
“I had seen college students register like crazy,” Nuttycombe says, so he bumped up the weight given to younger voters. “The big takeaway for why we beat the rest of the pack is we didn’t weight our poll to any previous election.”
Viewed this way, none of these polls were wrong, they just weren’t weighted to what the electorate turned out to be.
“We modeled the electorate to look like 2021, so we ended up too Republican,” says Harry Wilson, who runs the Roanoke College poll. “If we had gotten that D/R spread correct, then we would have been about dead on for the governor race. Of course, hindsight is always 20/20. Modeling turnout is always the trickiest part of polling, and I have never found a ‘perfect’ way to do it.”
Independents broke heavily for Democrats
We all want polls to be predictive and often they are, but they can only tell us what’s happening now, not what will happen in the days ahead. There’s something that only the election results can tell us: how those late-deciders broke. In this case, they overwhelmingly went toward Democrats.
We saw indications of that. Polls consistently showed that independents favored Spanberger and Hashmi, but those polls left lots of independents still undecided. For instance, the last Roanoke College poll showed independents favoring Spanberger 55% to 32% — but that still left 14% undecided.
In the lieutenant governor’s race, the final Roanoke College poll had independents backing Hashmi 45% to 30% — so 25% undecided.
It appears that those independents ultimately went overwhelmingly to Spanberger, and she pulled the other Democrats along with her. (I’ll deal with the attorney general’s race separately.)
Here’s what Rebecca Bromley-Trujillo, who runs the Christopher Newport University poll, says: “Looking at the polls that came out around the time ours did (12 days pre-election) versus right before show a heavy shift in undecideds to Democrats. In our own polling, our numbers for the Republicans were almost spot on relative to their final numbers, but we undershot Democrats who clearly won undecideds handily at the end there.”
All the polls through October were right about Earle-Sears’ final vote share. Of 18 polls, 11 had her between 41% and 43%; she got 42.32%. The one that was off the most was a Quantus Insights poll taken Oct. 18-20 that put her at 46% — so less than 4 percentage points off her final mark.
Likewise, the polls were pretty darn close to nailing Reid’s vote share. There were two polls, both in mid-October, that got it exactly right — the Washington Post/Schar School poll and the Suffolk University poll. Where the polls were off were with Hashmi; her vote was consistently underestimated. That’s because none of those polls could measure the late deciders who wound up going almost entirely to the Democrats.
Some voters moved away from Jones but then came back
The polls seemed the most “off” with the attorney general’s race; they showed the race close although it didn’t wind up that way. Upon closer inspection, though, the polls were more accurate about Miyares than they were about any other candidate; it was the Jones vote they missed.
Of 18 polls taken in October and November, eight had Miyares at 46% — which is what he wound up getting. Three others had him at either 45% or 47%. The poll that was off the most was a Trafalgar/Insider Advantage Poll taken Oct. 13-15 that put Miyares at 50%, which was taken weeks before the election and wound up only being off by 4 percentage points. Of the final eight polls, every single one had Miyares at 46% or 47%. That seems a pretty perfect call.
Where these polls missed was on the Jones side of the ledger. That seems the byproduct of the text message scandal. Through September, he was polling at 48% to 51%. Once the text messages broke, his support dropped — but didn’t move to Miyares. Those voters simply moved to undecided. Come Election Day, they moved back, however reluctantly.
We saw an indication of this in an Emerson College/The Hill poll. It asked undecided voters who they were leaning toward. They said Jones, by 63.7% to 36.3%. Here’s what I wrote in my final column about the polls: “That tells me there are some Democrats who are uncomfortable with Jones but their natural leanings might bring them around to him when it comes time to vote. Emerson says if those voters are factored in, the result would be Jones 51.3% to Miyares 48.7%.” Jones’ margin wound up being even bigger than that.
Andrew Kim of A2 Insight wonders whether some of those voters were ever truly undecided, or whether they were just reluctant to admit to pollsters they were backing Jones. This is known as the “shy voter” syndrome and we’ve seen it in the past in other contexts — “shy voters” who planned to vote for Trump but were hesitant to admit that. “Regarding Jones’s margin, I believe the miss was caused equally by the aforementioned weighting issue and social desirability bias, i.e. ‘shy’ Jones voters reporting being undecided to pollsters but coming home to roost on Election Day,” Kim says. “Most polls’ crosstabs were showing larger proportions of undecideds among Democrats/Spanberger voters. In our data, even with new weights applied, the Jones miss is notably higher than Spanberger and Hashmi.”
Tromley-Trujillo of Christopher Newport University says that whatever the reason, partisanship overrode whatever concerns voters had about the texts. “I think many voters were wary of the Jones scandal and had initial strong negative reactions, but ultimately partisan lean and polarization kicked in. In political science we have this term called negative partisanship — essentially what often drives the votes of independents [with a partisan lean] is their dislike of the other side. In this case, Trump was such a powerful negative that they were willing to overlook the Jones text scandal.”
Mark Rozell from the Schar School at George Mason University says we’ve seen this happen before on the other side of the political spectrum: “In 2016 (and 2020 and 2024), religious conservatives turned out in big numbers for Trump, even while fully aware of his personal failings. Was that hypocrisy, or was it instead a calculation that the impact of a leader’s policies on millions of people for the next four years outweighs the personal failings of one man? I am guessing something like that happened here for many voters.”
The only difference is that here the beneficiary was a Democrat rather than a Republican.
In the end, all the polls were broadly right; they also underscore that the only poll that really matters is the one on Election Day.
A Roanoke school board member says the ‘blue wave’ was responsible for her defeat. Was it?
That’s one of the things I’ll look at in this week’s edition of West of the Capital, our weekly political newsletter that goes out on Friday afternoons.
Update: Ooops! Yours truly messed up and mistakenly scheduled the newsletter for 3 a.m. instead of the regular 3 p.m. so it’s already gone out. That’s the hazard of a drop-down menu! If you’re not signed up for West of the Capital, but want to see that item, email me and I’ll forward it to you and sign you up for future editions.
Sign up here: