JWeinCom said:
sundin13 said:
I don't trust this whole "The Republicans are skewing the polls" stuff. There are plenty of non-partisan sponsored polls, especially in areas like Pennysylvania, which give Harris a reason to worry about her odds. The NTY/Sienna poll was a positive outlier for Harris in PA but the same poll also showed her down 5pts in Arizona among likely voters.
Besides, any good poll aggregator like 538 or Silver Bulletin will weight the data coming in by reliability, so some Rep Sponsored polls wouldn't do much to move the average (if they were even outliers, which they really haven't been). Even still, Silver Bulletin shows basically every state that matters (except Georgia) moving towards Trump over the last week.
It will be an incredibly close race if polls are somewhat accurate. That hasn't changed and likely won't. Any movement we see from here to election day will likely affect vibes more than actual mathematical odds. |
I don't think that's the case. There was a video, I don't remember quite by who, that pointed out the flaws in Fivethirtyeight's methodology. But here is one clear example. https://cdn.atlasintel.org/9e0da6ea-7e9c-498c-89e3-511bd7344cd0.pdf
This is the results of an Atlas Intel poll. They have a 2.7 star rating on Fivethirtyeight. According to this poll, Kamala Harris is actually winning men in Arizona by about 12 points, but losing women by 12. She is losing black voters 60 to 40%. In Michigan, the male vote is almost dead even, but Trump is ahead by 4 points because he is winning among women by 10%. Among those who choose not to identify their gender, Jill fucking Stein has 54% of the vote, Trump 48%, and Harris 4%. In North Carolina, Harris is actually winning by 3, and her advantage comes almost solely from men, as women are dead even. In Pennsylvania Trump is winning because he is up by about 5 points among women, and winning 18-29 year olds by 10 points. Trump is also getting 46% of the black vote, up from 7% according to 2020 exit polls. Again, this is a fairly highly rated pollster, the 22nd best according to Fivethirtyeight and should be influencing the models. Whether or not Kamala Harris will win, it is almost guaranteed that these polls are total horseshit. Whether it is deliberate or just incompetence I can't say, but there are definitely problems with the polling. The fact that the crosstabs are so wildly off yet the averages are still somewhat close smells of conspiracy to me. I think the problem with the aggregators is that they rely purely on data and I'm not sure if there is any human just looking through and saying "wait that's definitely fucked". |
I think even for the best polls, crosstabs should be taken with a grain of salt, moreso as you get into smaller and smaller pools of people. You run into small sample size issues real quickly when you start dividing data, but that shouldn't necessarily draw the greater pool into doubt. Like all polling, you need to look at if results are repeatable before you really trust them, otherwise you could just be seeing noise (which is normal).
For that Atlas poll, state sample sizes looked to be around 900, which is a decent sample size, but then if you're looking at a sample size of 450 for each gender, that is getting kind of risky, and if you are looking at a sample size of 10 for those who choose not to identify their gender, that info is a good as worthless.
Idk exactly how much of a red-flag those crosstabs should be though. I agree they don't make much sense, but if the numbers are in line with what we're seeing across the board, there doesn't seem to be much of a reason to try to unskew them.
JWeinCom said:
https://www.nytimes.com/2024/10/06/upshot/polling-methods-election.html#
Here is an interesting article that kind of explains some of what we're seeing. It is about how many pollsters are using weighting by recall results, which is explained better in the article. Essentially, there is a concerted effort among legitimate pollsters to weigh the data in such a way that will prevent a Trump underestimation the likes of which happened in 2016 and 2020.
The problem with this is that the weighing is based on the assumption that turnout will be similar to 2020, and so it skews the data to pretty closely reflect the 2020 results. The weighting process makes it really hard to come up with results that vary very much from the last election. That's why we really are seeing in the most part results that are very similar to 2020. In other words, to my understanding, they are trying to ensure their sample reflects the 2020 electorate. The only way the polls will shift very far from that result is if there are people changing their minds. And... really, it's hard to see that happening much. It's pretty clear that nothing Trump says or does will dissuade many people, or conversely that he's winning over tons of new voters. For example, the Marist poll which does use weighting has Trump at a 4% lead in Florida. Trump won the state by 3.2 in 2020. In the times poll, which does not use weighting, the result is Trump winning by 14%. I don't think Trump is winning by 14%, but I also think it's very unlikely that he's winning by only 4%. The 2022 election results and our eyes and ears would tell us that we would expect Florida to be more republican than it has been in the past. In this case, the weighting actually may be hurting Trump, but in any states that may be getting bluer, the opposite will be true.
So, both methodologies have their flaws. The good news, assuming you are not a Trump supporter, is that there is not likely to be a drastic underestimation of Trump as there was, particularly in certain states like there was the last two elections. Tying into what we were talking about before, it also seems clear that there are low quality polls designed to show Trump performing well. There were also some of those in 2020, but those were kind of washed out by the fact that Trump was being underrated in more legitimate polls. If we assume that the more legitimate polls are not underestimating Trump and there are polls overrating him, then the picture changes quite a bit.
So, at the very least, this explainst why we shouldn't judge 2024 polls the same way we did in 2020. They are far less likely to be underrating Trump. The only way they would do so is if Trump has a massive surge from where he was in 2020, and I just don't see that happening. It is possible though that the polls are underrating Harris. My personal opinion is that Harris is a stronger candidate than Biden was, and that Trump is a bit weaker than he was in 2020. Time will tell. |
I agree with this bit mostly though. I think a lot of pollsters are worried about losing credibility by repeatedly underestimating Trump, so they are compensating for that. I don't know if they are overcompensating, but I think it is a reasonable possibility to consider.
And I kind of disagree with this one, or at least it is a little more complicated than that. Silver Bulletin has probably shown the least movement towards Trump of any of the aggregates of late (at least of the ones I've been watching), but it still looks like Harris' position has gotten slightly worse over the past month and noticeable worse over the last week.
I just want to say, I'm not trying to be a doomsayer. I want to be clear again that I think the race is a toss-up and that hasn't really changed. I just thing some narratives (both the ones that say Trump is running away with it and the ones that say Harris is doing great) are a bit overblown.
