One thing to note with black voters in polls

[ProgressiveModerate]

Black voters are such a lopsided group politically, so in polls, you're mostly only going to have errors in one direction, infact it'd be mathematically impossible to have some of these errors in the same magnitude in Biden's favor because he can't receive over 100% of the vote. Furthermore, in polls where a large number of black voters claim "undecided"; that dropoff is disproportionately going to come from Democrats; assuming "undecides" will just break 50-50 for such a politically lopsided group is flawed.

This isn't to say that black voters can't or won't swing right, but this probably explains why cycle after cycle polling tends to underestimate Dems with black voters.

[Amenhotep Bakari-Sellers]

We don't like Trump

[Redban]

conjure any argument to avoid saying that Trump is making gains with a pivotal Democratic group

[Amenhotep Bakari-Sellers]

conjure any argument to avoid saying that Trump is making gains with a pivotal Democratic group

We haven't voted yet he is only polling that way

[RI]

Yes, which is also why people pointing to Phillips overperforming in polls as evidence that polls are wrong are themselves wrong. Behavior at the tails is not equivalent to behavior near the mean.

[ProgressiveModerate]

conjure any argument to avoid saying that Trump is making gains with a pivotal Democratic group

I literally said this doesn't mean that black voters can't or won't swing right; I am personally of the belief they will. I'm just saying getting polls showing stuff like 70 Biden - 12 Trump with black voters isn't on it's own indicative of a giant swing.

[ProgressiveModerate]

Yes, which is also why people pointing to Phillips overperforming in polls as evidence that polls are wrong are themselves wrong. Behavior at the tails is not equivalent to behavior near the mean.

Yeah all 3rd parties suffer from similar errors as well. The only error for a candidate who ultimately receives 1% of the vote is an overperformance in polling. Also as we well know, a lot of people use random 3rd parties are placeholders when they are dissatisfied/undecided but ultimately will vote for one of the 2 major party candidates.

[DrScholl]

conjure any argument to avoid saying that Trump is making gains with a pivotal Democratic group

He's not. You all literally hate Black people and Black people know this. Folks like you are still saying the Central Park Five are guilty just because Trump says they are.

When Trump loses the Black vote you'll be back to using plantation rhetoric and posting articles about how Black people are criminals and lazy.

[Tintrlvr]

This is a real effect and happens every single cycle. A bunch of polls show support for Democrats among black voters at only 80% and everyone freaks out about collapsing support and then it's just a statistical artifact in the end.

[Del Tachi]

This is a real effect and happens every single cycle. A bunch of polls show support for Democrats among black voters at only 80% and everyone freaks out about collapsing support and then it's just a statistical artifact in the end.

Except we have results from 2020 and 2022 that show Republicans doing better with non-White voters, so the polls are part of a trend.

[Amenhotep Bakari-Sellers]

This is a real effect and happens every single cycle. A bunch of polls show support for Democrats among black voters at only 80% and everyone freaks out about collapsing support and then it's just a statistical artifact in the end.

Except we have results from 2020 and 2022 that show Republicans doing better with non-White voters, so the polls are part of a trend.

No it's not all the polls in 22 were wrong, blks don't like Trump

[DS0816]

Black voters are such a lopsided group politically, so in polls, you're mostly only going to have errors in one direction, infact it'd be mathematically impossible to have some of these errors in the same magnitude in Biden's favor because he can't receive over 100% of the vote. Furthermore, in polls where a large number of black voters claim "undecided"; that dropoff is disproportionately going to come from Democrats; assuming "undecides" will just break 50-50 for such a politically lopsided group is flawed.

This isn't to say that black voters can't or won't swing right, but this probably explains why cycle after cycle polling tends to underestimate Dems with black voters.

Since 2008, a Democratic pickup of the presidency for Barack Obama, and the beginning of the Rust Belt Trio—Pennsylvania, Michigan, and Wisconsin—having established the longest streak of carrying for presidential election winners … Blacks nationally gave Democrats the following percentage-points margins:

• 2008: D+91
• 2012: D+87
• 2016: D+80
• 2020: D+75

That is something.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

Black voters are such a lopsided group politically, so in polls, you're mostly only going to have errors in one direction, infact it'd be mathematically impossible to have some of these errors in the same magnitude in Biden's favor because he can't receive over 100% of the vote. Furthermore, in polls where a large number of black voters claim "undecided"; that dropoff is disproportionately going to come from Democrats; assuming "undecides" will just break 50-50 for such a politically lopsided group is flawed.

This isn't to say that black voters can't or won't swing right, but this probably explains why cycle after cycle polling tends to underestimate Dems with black voters.

This is correct, and it is an important mathematical point that not enough people understand.

However, that does not mean that polling errors are not (theoretically) predictable, it just means that they are not normally distributed. In order to get normally distributed errors, what you have to do is to apply a logit transformation to the percentage scale. If you ran a large number of polls of black voters, then theoretically the transformed results would be normally distributed around the transformed mean.

An implication of this is that if you ever DO get polling results suggesting ~100% support or something asymptotically approaching that, you should take those results particularly seriously as compared to 99%, 98%, or 95% etc (if you have a decent sample size etc anyway).

The same also applies to other subgroups that tend to have overwhelming support one way or another, i.e. "democrats" or "republicans" for example.



In actual practice though, I would suspect that even after transforming the results to a logit scale, errors will still be biased downwards simply because you never are going to have a perfect sample, and you can get some error from a certain number of respondents lying, data errors, and people just trolling the poll. Though this is more of a speculative point.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

Blacks nationally gave Democrats the following percentage-points margins:

• 2008: D+91
• 2012: D+87
• 2016: D+80
• 2020: D+75

That is something.

That is indeed something.

More specifically, it is something false.

A better estimate would be approximately:

• 2012: D+94
• 2016: D+88
• 2020: D+82

Note also that the difference between D+94 and D+82 is not that 12% of voters switched, it is 6% (a decline from 97% to merely 91% support).

That is a weak enough "trend" (especially following the first black President no longer being on the ballot - the congressional level numbers are more stable) that the aggregate effect on overall vote margin is more or less offset merely by the fact that black population share is increasing relative to white population share over time.

See here: https://www.dropbox.com/s/re0gtn1o57fzwp5/Catalist_What_Happened_2022_Public_National_Crosstabs_2023_05_18.xlsx

[DS0816]

Blacks nationally gave Democrats the following percentage-points margins:

• 2008: D+91
• 2012: D+87
• 2016: D+80
• 2020: D+75

That is something.

That is indeed something.

More specifically, it is something false.

…

My source:

https://ropercenter.cornell.edu/how_groups_voted

Again—what I wrote were the percentage-points margins.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

My source:

https://ropercenter.cornell.edu/how_groups_voted

Again—what I wrote were the percentage-points margins.

And the source of that is:

Sources:

“National Exit Poll for Presidential Results.” CBS News, December 14, 2020. <https://www.cbsnews.com/elections/2020/united-states/president/exit-poll/>.

Exit polls are well known to be a flawed and inaccurate source for this sort of thing. News networks use them for election night TV coverage so that they have something to talk about, not because they are particularly more accurate than ordinary pre-election polls.

[DS0816]

…

Exit polls are …

I am fine.

Thanks.

[Former Dean Phillips Supporters for Haley (I guess???!?) 👁️]

…

Exit polls are …

I am fine.

Thanks.

When I was writing my previous post, I thought briefly about putting in a qualifier saying something like "I realize you probably don't realize the methodological issues with regards to exit polls and are posting them in good faith, but..."

However, I didn't do so, and looking back on my post I can see that without that it comes off as a bit gruff, so I apologize for that.

The main problem with exit polls, however, is that they use stratified sampling. This has a tendency to produce biased results, particularly for small sub-groups which are geographically concentrated in some areas more than others (as black voters tend to be concentrated both in the south and in urban areas). That is why they are not the most reliable data source.