I would just say, for anyone truly interested in the analytics of the election and not just schlong waving when they see a poll or pattern they like, read the cross tabs or top lines on these polls. There’s a ton of really interesting info that isn’t getting talked about almost at all but is giving big hints as to where this race truly is right now in these states beyond the top line numbers, who has more upside, who’s support is weak among their supporters etc. The three fox news polls in AZ NC and PA the last two days have great data for instance. Example, what if I were to tell you that despite a tie in PA, 13 percent of one candidate’s supporters would consider switching, and only 6 percent of the other’s would? That could have huge ramifications on a late race break to one side or the other if they aren’t firmed up. Or that independents are 15 or 20 percent less likely to actually vote, but the only undecideds there are in that group? So how do you prioritize your own core turnout vs. getting those Indy numbers up if you’re the candidate winning that group right now? Do you target the state you’re trailing by 1, or the state you are trailing by 3 or four but with more voters open to possible change? To me this is the real fun of the election, the data and strategy behind what’s really happening. When you see it through that lens the strategies that play out on the trail make sense. I would also say that beyond the obvious fact that poll averages are better than individual polls, if you’re touting a poll that doesn’t release their data, you’re fanboying the wrong poll. And they exist both ways. The reason the modelers rank polls like fox and NYT so high is for their transparency. Their results can be easily analyzed and adjusted if their samples are off, or they can see why their results vary from others. And when you publish your data your more apt to be making an honest effort. Even then, sometimes you can see obvious issues with the poll. Undercounting or overcounting certain groups for instance. I saw one poll from an A pollster that had a sample with only 23 percent of respondents under 40. Another one used the exact same modeling as 2020, despite shifts in demographics, party affiliation etc. in PA. anyway, my (not so small length wise) contribution, back to sword fighting.
Gallup: For The First Time In Decades, More Voters Identify As GOP (dailywire.com) In an average of Gallup polls taken between July and September, 48% of voters surveyed either identified as Republicans or leaned Republican, while 45% of the voters identified as Democrats or leaned Democrat. 46% of voters thought that the GOP was better able to handle the most important problem facing the country, as opposed to 41% who chose the Democrats. Gallup listed the issues that voters thought were most important as the economy and immigration. 50% of voters deemed the GOP better equipped to keep America prosperous, while 46% thought the Democrats would do the job better. A whopping difference came between the 54% of voters who thought the GOP was better able to keep America safe from international threats and the 40% who chose the Democrats.
Another of the modeling sites I've been following, Dactile, has Harris at her highest chances of winning yet, at 63%. This is model is focus only 8 swing states and high quality state and national polling. What is the probability Harris wins? Building a Statistical Model. (dactile.net)
And Trafalgar’s CEO was equally off in 2022. “I think there’s a good chance there’s a red wave. The theory I’ve been putting forth is there are a lot of Republicans who are really, really hesitant to participate in polls. And this is different than in 2016, when they were kind of shy if they were voting for Trump. And it’s different than in ’20, where you just had to work a little harder to get to them. To find the majority of Republicans, you had to just dig deeper; you had to make a lot more calls. They were unwilling to take polls, so many of them.” Will America’s Most Pilloried Pollster Get It Right Again?
That's hilarious ... "We had to keep calling people until someone admitted they were voting republican, they were hard to find, but we managed to get a few and that's how I know the republicans are going to win!"
My issue with this is you go from analyzing the cross tabs of a singular state poll to "poll averages are better than individual polls." I can easily find other polls with cross tabs that look equally as damning for the opposite candidate. Going with the approach of polling aggregates, it's clear this race comes down to the 3 rust belts, with PA being the closest. 8 of the last 10 state polls out of PA have the race either tied or with Trump ahead. Factor that in with WI and MI being toss-ups within 1 or 2 points and those two states have a horrible history when polling Trump support, you can see how tenuous this race is for Kamala Harris. He legit only needs to pick off one of those 3 states and that makes his path so much easier. While we saw the 3 sun belt states break hard for Trump after the debate, the 3 rust belts stayed stagnant. I think he wins PA. He was shot there, GOP voter registration increases, Teamsters non-endorsement of Harris and let's face it, they are equally as concerned about immigration, fracking and the economy. She obviously needs to turn out hard in Philly. But even Bucks county now has more GOP registered voters than Dems. But frankly, she's more vulnerable in all 3 rust belts than Biden or Clinton were.
Based on Okee's prediction of Trump winning PA looks like it's safe to PA to the strong blue category
This headline from RCP says it all: Harris Leads +2.0 Nationally, But Trump Running 4.8 Ahead of 2020
People are perfectly free to have an opinion on who will win, every model has the race close, so who knows what the deciding factor will be or who will be giving a victory speech that Tuesday night. I am only saying that these opinions should be based on statistical understanding, not hopeful thinking. The best minds in polling and statistical analysis don’t know who will win right now, so none of us do either.. On individual polls, most people don’t understand that these polls tend to have a +/- of 3 points or more, which is in both directions. So they can be off by 6 or 7 points total and still be perfectly valid statistically, which is why you need to aggregate polls. And that’s just the 95th percentile, five percent of polls will be outside of that (like the infamous ABC news Wisconsin poll in 2020). Even then in poll aggregation, not only do you still have a decent MOE, about half the time polling fails to pick up momentum in one direction or the other. In the last off cycle the Dems had it, in the last two presidentials Trump did overall, but especially in Wisconsin (PA last cycle the RCP average actually nailed the final vote to the 10th of a point at 1.2). So will the Dems momentum carry from the last 2 years when they overperformed, will Trump’s history of over performing in an important state carry, will those two things offset and the polls will be largely accurate, or have pollsters adjusted from the last few misses and they are largely right as of today? No one can say with certainty, and almost anyone who does think they know is doing so from a biased POV, even if half of them are eventually proven right. If you read enough cross tabs and top lines though a lot of the race’s undercurrents behind the “So and so leads in X” headlines can be seen, along with who is honestly trying to poll accurately and who is trying to make a statement. I was only using the fox polls as an example because they are a good pollster and gave really useful data on polls taken the last few days. That was all.
This headline from RCP says it all: Harris Leads +2.0 Nationally, But Trump Running 4.8 Ahead of 2020
What does that headline say to you in the context of my post, and in the context of state level polling?
Trump is in a much better position than he was in 2020 at the state and national level. When you consider that he came within 45,000~ votes of tying Biden with 269 EV's, when you consider that he came within 126,000~ votes of winning the race with 289 EV's, you can see why a shift of approx 7.4 million voters his direction would effect things. And that is what current polling indicates. In fact, RCP is comparing today with Sep 28, 2020. The final polling average at RCP on Election Day 2020 was +7.2 Biden. So Trump is as of today polling 5.2 points ahead of the final round of polling in 2020, so a shift of approx 8 million voters his way. It's worth noting also that despite Biden winning the national popular vote by 4.5%, Trump came within 45,000~ votes of tying him in the EC. So you almost have to assume that Harris needs to win the popular by probably at least 4% to win the EC. She's only up by 2 points nationally in the average of polls and that assumes no anti-Trump bias which has been ever present the past two cycles.
So to my point, you’re banking on the polls being wrong again, because right now, the state level polls say she would win the election, albeit in a squeeker. We will see, around a 50/50 shot of being right I guess. Same odds as the Dems making the case that they have been the ones to overperform in recent elections, that they have a money and ground game advantage this cycle, polls show they have an enthusiasm edge, and an AA woman at the top of the ticket sets up the Obama coalition more than Biden did. We will see in 6 weeks or so which side wins the coin flip, whoever turns out to be right will claim they knew all along, and whoever is wrong will likely claim fraud or some other uncontrollable thing. But if the models all have it as a coin flip, anyone claiming they know with confidence how it will land is doing so from their own bias. Anyway, enough from me for now. Gonna go do something productive .
The national polls are a superior aggregate compared to state polling. If you compared the accuracy of polling in 2016 and 2020 national vs. state, national polling did better, even though they did poorly. They still did better than individual state polling. So without assuming any erroring in the polling against Trump (even though that has happened to a significant degree in the past two Trump elections), Kamala is still in a much more precarious situation than both Hillary and Biden were. Kamala winning the popular by 2% isn't going to win her the election and that's what the national aggregate of polls are saying right now. Also, saying Kamala is ahead in state polling is disingenuous at best. In the no toss-up states, Trump has 262 EVs. They are giving Harris PA, but again, 8 out of the last 10 polls out of PA show the race either tied or Trump ahead. So if you want to act like she's ahead, have at it, because I'm telling you she is not ahead. The national aggregate shows you that clearly. In any case, if you're not able to at least admit Trump is in a much better position at this point than he was in 2020, you're whistling past the graveyard. And I don't see anybody respectable making the argument that Harris is in better position than Biden or Hillary at this point. She's way behind where they were.
Silver has a good free blog post on how he adjusts consistently bias pollsters, with some good charts. As expected, Trafalgar and Rasmussen are skewed pretty far right, overestimating republicans by 2.6 and 2.7%. Ipsos and outward intelligence are pretty far left, at 1.9, but not as bad as Trafalgar and Rasmussen. TIPP, YouGov, CNN, NTY, & Quinipiac look pretty balanced. Which polls are biased toward Harris or Trump?
The most accurate pollster from the 2020 presidential election has Trump leading in 5 of 7 swing states. My only concern is AtlasIntel seems to have gotten some squirrelly results in Michigan and North Carolina, seems backwards to me: AtlasIntel is confirmed as the most accurate pollster of the 2020 Presidential Election | AtlasIntel AtlasIntel had the best performance across all pollsters of the 2020 US Presidential Election with an average error of 2.01p.p. Our final national-level poll showing Biden with a 4.7 p.p. advantage is likely to be the single most precise estimate of the US popular vote. AtlasIntel also conducted the most precise polls across all pollsters for the states of Michigan and North Carolina. In all states polled by AtlasIntel, results fell within the margin of error of our estimates. The AtlasIntel average error was not only lower than that of other pollsters, but also lower than that of polling averages and model-based polling aggregators (see figures below): Pollster Error in the 2020 US Presidential Election [Based on difference between the lead of the winning candidate and poll estimates] For all of the complaints about Rasmussen here, they were the 5th best accurate pollster in the 2020 presidential election. Atlas Poll - US Swing States - September 2024.pdf North Carolina: Harris 50.5% Trump 48.1% Georgia: Trump 49.6% Harris 49% Arizona: Trump 49.8% Harris 48.6% Nevada: Harris 50.5% Trump 47.7% Wisconsin: Trump 49.7% Harris 48.2% Michigan: Trump 50.6% Harris 47.2% Pennsylvania: Trump 51% Harris 48.1%
