Florida, Iowa. Both trending Harris even tho trump still has a tiny lead. If Harris win Florida it is over.
Not at all in the realm of a response to what I posted. Losing states that you were going to lose before by narrower margins, and winning states that you were going to win by bigger ones, is nice. She can achieve a bigger popular vote.
But “increasing her lead” doesn’t mean losing with a larger EC disadvantage. The states that are statistically tied, the ones @Pleonast’s analysis specifically excluded, what does an aggregate of those states tell us about her trend?
I have not crunched it myself. My impression is that the trend in that group is overall positive and that the EC disadvantage is actually decreasing. If that impression is true it is a very meaningful trend.
That’s a subject of Silver’s most recent essay:
“In 16 years of running election forecasts, I’ve never seen such a close election.
Our polling averages in seven swing states — in alphabetical order: Arizona, Georgia, Michigan, Nevada, North Carolina, Pennsylvania and Wisconsin — are within 2 percentage points. A systematic polling error, or a shift in the race in the final six weeks of the campaign, could result in one candidate sweeping all of these states. In our simulations this morning, Kamala Harris swept all seven of these battlegrounds 20 percent of the time, and Donald Trump did in 23 percent of the simulations.”
The rest is for paid subscribers, of which I am not one.
I assume the “half” refers to Nebraska.
So…the “EC advantage” may be decreasing for other reasons, and in other ways…but it may be hard to tell from the results of this particular election, because it’s likely to be so close, the tiny margins could be explained by any number of things. Exit polls with cross tabs might help sort this out, but even they might not end up telling us much.
But…the mere fact that the popular vote will almost certainly be millions of voters different than the effective EC margins (whichever way the latter happens to barely go) shows, to me at least, that we’re still in an era of Republican EC advantage, even in Harris ends up winning.
Long story short: we’re on the same knife-edge as in 2016 and 2020, but in any other country in the world, we wouldn’t be. (Although, Trump does “win the popular vote” in 28% of 538’s simulations. That’s higher than I would have guessed. So, maybe if we were in a normal country, it wouldn’t be a knife edge, but we would still have a real race on our hands. Interesting.)
Again, can we not use a paid tool of the GOP as any kind of real source on anything but being a bought and paid for liar?
I find value in discussing Silver’s analysis and I hope we don’t stop talking about him because one poster has an issue with him.
Not just me.
Moderating:
Drop it now. Whatever your personal issues are with Nate Silver, you’ve made your opinions about him abundantly clear in many threads. Time to stop.
I’ve basically given up watching the polls this season. As far as I am concerned its a coin flip The polls themselves show a near tie, and then when you add in additional uncertainty related to polling mechanisms and turn out, you have to accept that we are in the realm of who the heck knows. Whether the reality is that Trump is ahead with a 60% chance of winning or behind with only a 40% chance, is not particularly relevant on a practical level. There is still a very good chance that Trump will win and a very good chance that he will lose.
So just keep your fingers crossed and on election night have both champagne and bourbon at the ready.
From Silver’s quote:
Isn’t he saying that “his polling is right, unless it’s wrong?” Doesn’t a systematic polling error mean a bias error (as opposed to error from variance)? Errors like the population is mischaracterized or a subsection is consistently underrepresented (as opposed to random sampling noise).
For a close election it comes down to 100k votes in 7 states. < 1% in some states. So easy to over or under represent that demographic. I don’t see how you confidently model that, but to be fair that’s what he is saying too.
He doesn’t have any polling. He’s not a pollster.
Yes. (With the noted correction above noted.) The polls, and the forecasting models based on them, may or may not be subject to systemic errors in how they model the “will actually be voting” populations, and have been in the past, by three percent fairly commonly. Aggregated polls of individual states off by more.
That’s the skewing of the thread title. The pollsters are trying to hit it right but it is squirming moving target and they don’t really know. Could be skewed one way, could be the other, could be a lot, could be near zero. You cannot in advance “unskew” that.
Silver is always straight up that what he, and other forecasting modelers, are selling us is a gambler’s tool, not a prediction machine. The assumption they make is that the systemic skew is unknown and likely of equal probability to be one way or the other. So models based on aggregated polls can give a sense of what is goods odds for a bet one way or the other. (If his model tells him it is 2:1 favoring Harris and he could find a bet that was set at 3:1, he’d take the Trump side of that bet, even if he is expecting to lose the bet more likely than not. He would take Harris though if offered better odds than 2:1.)
My interest is not placing smart bets. For people like me the aggregated polls of subject to systemic error are likely subject to a consistent error within any specific cycle. So as a tool to understand relative positions through the cycle even with various demographics they are still useful.
Yes. I have no idea what the systematic bias is. There are people who don’t want to admit they are voting for trump, and people who don’t want to admit they are voting for abortion, and even more people who just don’t want to answer the phone or talk to pollsters.
But the systematic bias is probably somewhat constant over small increments of time, like several months. So poll movements are probably meaningful, even if the absolute value is off.
@JKellyMap discussed it as well, but the table from Silver’s current analysis has trends (week and month) for each of the swing states.
The general trend, as much as I can make it out from eyeballing, is that Harris is a bit up over the last week but Trump has gained ground over the past month.
You can see it halfway down here: Silver Bulletin 2024 presidential election forecast
If you want to look at RCP, you can see a rough trendline of only swing states here: Top Battlegrounds – RCP Average
They basically just average the polling average for a handful of states (WI, PA, MI, NV, AZ, GA, NC). It’s a bit noisier due to polls coming in and out of its average and no attempt to smooth or anything, but I see no discernible trend there either.
Of course you could argue that the only polls that really matter are WI, PA, and MI - unless the GOP changes the rules in Nebraska.
Or, even more accurately, you could argue that the polls don’t matter at all and what matters is registering voters, getting out the vote, and making sure that every vote is counted.
This is, indeed, where i am figuring my efforts. But to persuade people, but to help people (who are likely to vote D) to vote, and to remind them that downballot votes matter.
But i don’t think polls help us much in understanding how that effort is going.
Actually, ARE there any polls that attempt to measure voter engagement?
I’m not sure how there could be - they would have the same response-rate bias that traditional polls have. Probably even worse since folks willing to answer polls are likely correlated with engagement in the first place. There is a theory that is why Trump seems to over-perform his polls so consistently - he has the least engaged voters of them all.
You can possibly try to tease out voter registration numbers, but that’s a risky game as well. Even worse is trying to discern anything useful from absentee ballot request or return rates. But that doesn’t stop folks from trying to do it (I’ve already seen a few on X).
It took a while, but for the first time … Kamala Harris broke past a 2-point lead over Trump in RealClear Polling’s national general election aggregate:
IMHO, this is a big deal. RealClear leans right, and IMHO will commonly cherry-pick specific polls in their aggregate to inflate the Republican candidate’s number. Typically, this is by quickly aging out Harris-positive polls and leaving in for too long Trump-positive or Harris-weak polls (e.g. the three earliest polls shown above). Plus the decision to reinstate Rasmussen Reports to their average.
Anyway. Fighting a little bit uphill, Harris is ahead by 2.2 points in RealClear Polling’s aggregate. Too close? Within the margin of error? Go to FiveThirtyEight’s comprehensive list of national general election polls and notice how easy it would be to select the right polls over the last 2-3 weeks to build a national poll aggregate lead of 4.0 points – or better – for Harris.
Of course the flip-side of that argument is that Biden never led by less than 6% in the RCP average after July in 2020 and only ended up winning the national vote by 4.5% (with a final poll average lead of 7.2%). They had the same biases then that they do now. A 2.8% miss in the same direction would mean Harris actually loses the national vote, and probably loses badly in the EV.
If you want to cherry-pick individual pollsters, the final NBC/WSJ poll in 2020 was Biden+10. The final Rasmussen was Biden+1. The final FoxNews was Biden+8. All of those are worse for Harris now than they were for Biden 4 years ago. In fact I don’t see a single pollster where Harris is doing better now than Biden was on election day in 2020. And Biden just barely squeaked a win.
On the plus(?) side the most accurate final polls in 2020 were TIPP (Biden+4) and The Hill/HarrisX (Biden+4). Right now those two have Harris+4 and Harris+2, so at least winning if not doing better than Biden.
I’m not convinced that just because something happened in 2020 or 2016 that the same thing must be accounted for again in 2024. I take the 2024 election and the lead-up polling as independent of what happened in the past. I understand that others’ mileage may vary.
EDIT: Put another way – Harris doesn’t have to win the 2020 or 2016 elections. She has to win the 2024 election, which will for certain have many different parameters than the previous two.
In both I see pretty much the same thing: fairly flat in those “swing states” since just before the DNC.
There are six states that are under two points either way. With typical amounts of systemic bias she could sweep all six or lose all six. Or any in between.
What I would love to see is aggregated subgroup trends. I’d specifically like to see the numbers for strong vs mild disapproval among conservative voters. (Not finding on a quick attempt.) I suspect that may translate to “engagement” and if she and Walz are succeeding at decreasing Trump’s turnout.
This is all true, but the pollsters for whom this is their superbowl also know this. So our alternatives are to assume they either a) changed nothing and are once again under estimating Trump b) corrected their mistakes and the polls are now more accurate or c) over corrected their mistakes and the polls are now under estimating Harris. I think b and c are more likely than a.