superrific
Inconceivable Member
- Messages
- 3,053
I think we should have a different thread for talking *about* the polls and *about* the news. For one thing, having the discussion and the polls in the same place means that people have to look at the polls to follow the discussion as they are in the same place. Second, more concentrated discussion is useful, I think. I will offer my own take on what we are seeing. Obviously there will be some disagreement.
1. Don't look at national polls at all. They don't mean much, given that they aren't measuring anything of importance. National polls are used in forecasts in a specific way that can be helpful in the earlier stages of an election. The idea is that you can predict states by taking a national poll and then applying the state's partisan lean. So, for instance, Michigan has been +2 GOP, IIRC, for the past few cycles. So if the national poll is Harris +1, and Michigan is +2 Trump, then you'd expect Trump to be winning by 1 point in Michigan.
The flaw in this analysis is that the partisan lean of the state changes over time. So let's use that example but add some Michigan polling putting Kamala +1 in Michigan. Now, you might interpret that as a biased poll in favor of Kamala, since we know that Michigan is supposed to be +1 Trump based on national polling. I would interpret that as suggesting that Michigan isn't going to be +2 GOP this cycle. If it was just one poll, as in Arkansas or other states where very few polls are run, fine, use the national + partisan lean model. But we have tons of polling of Michigan and that polling is telling us that Michigan isn't +2 Trump. It's +0 Trump. That could be because Michigan changed, or it could be because other places have changed. If Missouri goes from Trump +8 to Trump +16, that's pushing the national average toward Trump; by extension, it means that every other state will see its partisan lean decrease.
So state polls >> national polls right now. I don't think the forecast models are even looking at national polls right now; if they are, they have little weight.
2. The early voting numbers don't cast much light. I had been thinking that the cross-referenced voting numbers are more informative -- i.e. comparing polls of people who have already voted to the early-voting-by-registration maps. For instance, if Kamala is winning in NC +9 and Pubs and Dems have returned ballots in the same proportions, that +9 means either that Kamala is winning indies big time, or there are more Rs who are voting for Kamala than Ds voting for Trump.
I still think that's directionally true; using the already-voted poll in conjunction with the state maps contains a little bit of information. But I don't think the magnitude is very large. There's no reason to think that independents are any less likely to self-select based on early voting. There are plenty of middle-aged educated people registered as independents. There are also plenty of low-propensity, low-engagement voters registered that way. If the former group is more likely to early vote, then you'd expect Kamala to be winning among them at this stage. I mean, I'd rather be up 9 in already-voted than down 9, but I don't know how much this is telling us. It's not nothing, but it's not a ton either.
3. The EC is not necessarily going to favor Trump. It did in 2016, but barely. He won the blue wall states by tiny, tiny margins -- margins that were nearly random, I'd say. If he had lost those states by tiny margins, we wouldn't be having this discussion. In 2020, it was more pronounced, but 2020 was weird. In 2012, the EC favored Obama. And it very well could favor Kamala. Remember: an EC advantage just means that your opponent is running up the score in safe states. And for Dems, that means California. I think I remember reading that the EC advantage for Trump (at least in 16) was almost entirely a California effect. For Pubs, it means the confederate states.
Well, the way the GOP has so heartily embraced Trump, I suspect that he will do better this cycle in AL, MS, LA, OK than in the past. That's not helping him in the election. It's eroding his EC advantage.
Remember what matters: the swing states. Whoever wins those states will have the EC advantage this cycle (in practical terms, if not fully in theory). We don't actually need to win nationally by X to have a chance. We just need to win MI, PA and WI. What is happening in AL or CA doesn't matter. If, as some data suggests, Florida has gone completely red because of conservative in-migration, that's a great result for us. It will tilt the EC toward us.
4. About polls. This is also making us nervous. Won't Trump outperform his polls again? Don't we need to be leading by 5, as in 2020, to make sure we win? Sure, he might outperform his polls, but there's no particularly strong reason, in my view, to think that's likely.
A. I read a very interesting article last week discussing this a little bit. It talked about the role that landlines played in the 2020 polling -- namely, that landline polls were way more favorable to Biden than to Trump. Why? Well, because Biden voters were cautious about the pandemic, and stayed at home a lot, meaning that they were available when the landline pollster calls and willing to talk because they haven't talked to anyone all day. Trump voters were out there going to rallies and doing stuff during the day and then dying in December and January (there should be at least a minor COVID adjustment from 2020, since most Covid deaths occurred after the election and those deaths were predominantly GOP). They weren't answering landlines.
Well, nobody is doing landline polls any more. I think Quinnipiac is doing random digit dialing, which would tend to capture a fair amount of landlines. Those polls have been good for Kamala, which suggests that the vast majority of polls not using landlines or random digit dialing are not going to be weighted in her favor.
B. Plus, the majority of polls are being weighted by recalled vote. These polls favor Trump more than Kamala; they are also methodologically suspect. There's a reason why nobody was using recalled vote weighting. It's basically being used in an attempt to amplify the disengaged Trump voters, like the pollsters who just added 5 to Trump's totals in 2020 because. But if it was bad methodology then, it's bad methodology now.
For one thing, recalled vote is inaccurate. Secondly, recalled vote weighting misses voter composition changes. If a poll reaches too few 2020 Trump voters, maybe it's because it didn't reach those voters for whatever reason. But it might also be because they don't exist anymore -- maybe they died or moved to a different state. The whole point of a poll is to find some answers to those questions. If you're weighting by recalled vote, you're missing those shifts.
Those shifts are almost certainly going to favor Kamala at least a little, because there's almost no room for those shifts to hurt her. People aren't moving to California. They are leaving California. All else being equal, that's shifting the electorates in other states. People are moving to Florida. They are probably moving there from NC, and the ones who are moving are likely to be conservative. People are also moving to NC; typically they are educated people who are more likely to vote for Kamala.
Anyone have other points? Or follow-up questions that I might be able to address?
1. Don't look at national polls at all. They don't mean much, given that they aren't measuring anything of importance. National polls are used in forecasts in a specific way that can be helpful in the earlier stages of an election. The idea is that you can predict states by taking a national poll and then applying the state's partisan lean. So, for instance, Michigan has been +2 GOP, IIRC, for the past few cycles. So if the national poll is Harris +1, and Michigan is +2 Trump, then you'd expect Trump to be winning by 1 point in Michigan.
The flaw in this analysis is that the partisan lean of the state changes over time. So let's use that example but add some Michigan polling putting Kamala +1 in Michigan. Now, you might interpret that as a biased poll in favor of Kamala, since we know that Michigan is supposed to be +1 Trump based on national polling. I would interpret that as suggesting that Michigan isn't going to be +2 GOP this cycle. If it was just one poll, as in Arkansas or other states where very few polls are run, fine, use the national + partisan lean model. But we have tons of polling of Michigan and that polling is telling us that Michigan isn't +2 Trump. It's +0 Trump. That could be because Michigan changed, or it could be because other places have changed. If Missouri goes from Trump +8 to Trump +16, that's pushing the national average toward Trump; by extension, it means that every other state will see its partisan lean decrease.
So state polls >> national polls right now. I don't think the forecast models are even looking at national polls right now; if they are, they have little weight.
2. The early voting numbers don't cast much light. I had been thinking that the cross-referenced voting numbers are more informative -- i.e. comparing polls of people who have already voted to the early-voting-by-registration maps. For instance, if Kamala is winning in NC +9 and Pubs and Dems have returned ballots in the same proportions, that +9 means either that Kamala is winning indies big time, or there are more Rs who are voting for Kamala than Ds voting for Trump.
I still think that's directionally true; using the already-voted poll in conjunction with the state maps contains a little bit of information. But I don't think the magnitude is very large. There's no reason to think that independents are any less likely to self-select based on early voting. There are plenty of middle-aged educated people registered as independents. There are also plenty of low-propensity, low-engagement voters registered that way. If the former group is more likely to early vote, then you'd expect Kamala to be winning among them at this stage. I mean, I'd rather be up 9 in already-voted than down 9, but I don't know how much this is telling us. It's not nothing, but it's not a ton either.
3. The EC is not necessarily going to favor Trump. It did in 2016, but barely. He won the blue wall states by tiny, tiny margins -- margins that were nearly random, I'd say. If he had lost those states by tiny margins, we wouldn't be having this discussion. In 2020, it was more pronounced, but 2020 was weird. In 2012, the EC favored Obama. And it very well could favor Kamala. Remember: an EC advantage just means that your opponent is running up the score in safe states. And for Dems, that means California. I think I remember reading that the EC advantage for Trump (at least in 16) was almost entirely a California effect. For Pubs, it means the confederate states.
Well, the way the GOP has so heartily embraced Trump, I suspect that he will do better this cycle in AL, MS, LA, OK than in the past. That's not helping him in the election. It's eroding his EC advantage.
Remember what matters: the swing states. Whoever wins those states will have the EC advantage this cycle (in practical terms, if not fully in theory). We don't actually need to win nationally by X to have a chance. We just need to win MI, PA and WI. What is happening in AL or CA doesn't matter. If, as some data suggests, Florida has gone completely red because of conservative in-migration, that's a great result for us. It will tilt the EC toward us.
4. About polls. This is also making us nervous. Won't Trump outperform his polls again? Don't we need to be leading by 5, as in 2020, to make sure we win? Sure, he might outperform his polls, but there's no particularly strong reason, in my view, to think that's likely.
A. I read a very interesting article last week discussing this a little bit. It talked about the role that landlines played in the 2020 polling -- namely, that landline polls were way more favorable to Biden than to Trump. Why? Well, because Biden voters were cautious about the pandemic, and stayed at home a lot, meaning that they were available when the landline pollster calls and willing to talk because they haven't talked to anyone all day. Trump voters were out there going to rallies and doing stuff during the day and then dying in December and January (there should be at least a minor COVID adjustment from 2020, since most Covid deaths occurred after the election and those deaths were predominantly GOP). They weren't answering landlines.
Well, nobody is doing landline polls any more. I think Quinnipiac is doing random digit dialing, which would tend to capture a fair amount of landlines. Those polls have been good for Kamala, which suggests that the vast majority of polls not using landlines or random digit dialing are not going to be weighted in her favor.
B. Plus, the majority of polls are being weighted by recalled vote. These polls favor Trump more than Kamala; they are also methodologically suspect. There's a reason why nobody was using recalled vote weighting. It's basically being used in an attempt to amplify the disengaged Trump voters, like the pollsters who just added 5 to Trump's totals in 2020 because. But if it was bad methodology then, it's bad methodology now.
For one thing, recalled vote is inaccurate. Secondly, recalled vote weighting misses voter composition changes. If a poll reaches too few 2020 Trump voters, maybe it's because it didn't reach those voters for whatever reason. But it might also be because they don't exist anymore -- maybe they died or moved to a different state. The whole point of a poll is to find some answers to those questions. If you're weighting by recalled vote, you're missing those shifts.
Those shifts are almost certainly going to favor Kamala at least a little, because there's almost no room for those shifts to hurt her. People aren't moving to California. They are leaving California. All else being equal, that's shifting the electorates in other states. People are moving to Florida. They are probably moving there from NC, and the ones who are moving are likely to be conservative. People are also moving to NC; typically they are educated people who are more likely to vote for Kamala.
Anyone have other points? Or follow-up questions that I might be able to address?
