PA: Insider Advantage: Tie

[kwabbit]

Isn't Insider Advantage generally pretty Republican friendly in their polling results?

Either way, gonna be a long night I guess.

I think they had Trump winning PA by a point in their last poll before Election Day 2020....but I might be misremembering. 

That was for the "Center for American Greatness" IIRC, so it was R-sponsored. This is non-partisan. They've found good results for the GOP this cycle, but it's not like 2020 when they basically served as an alt-Trafalgar.

[kwabbit]

The poll is actually Fetterman +1, not a tie, it's 46.3 - 45.5.

Man, the doomer-ism in here is one thing, but y'all are going off about a crosstab that has 36 (Hispanic) and 49 (black) people in it!

Statewide polling is going to have rough samples with minorities, but I think we can all agree an 85 person sample for minorities is just terrible. Thus, the sample with 36 hispanics gets a wonky result (Oz +60... which obviously is not happening)

Same thing with "Oz having strength with 14% of Blacks" .... it's a tab of 49 people. That's like, what, 6 people? lol

What I find interesting that people are ignoring here is the White vote... it's literally tied, again. If the white vote is anywhere close to tied, it's great news for Fetterman. And the n= is at least strong here, 465.

(not to mention IA is an R-leaning pollster)

Of course as a proportion the White vote share has a small margin of error. However, in absolute terms it doesn’t. A 5% sampling error on the White vote causes as much of a shift in the top line as a 25% sampling error in the minority vote. You can’t hold the White % fixed and correct the minority cross tabs.

Trump won the White vote in PA by 15 points, so if you’re gonna play the cross tab game that should raise some eyebrows. I know the ‘other’ category has obviously inaccurate/weird results, and that a tie in the White vote seems more plausible because the state as a whole is tied, but you have to realize that a 15pt swing wouldn’t happen either.

Most of this can be attributed to the fact that IA is a mediocre pollster. It doesn’t seem like they weight by education, which is why they likely have the Democrats doing so well among Whites. They counter that by finding an extremely unrepresentative minority sample (which is a common theme among R pollsters, such as Trafalgar).

[kwabbit]

The poll is actually Fetterman +1, not a tie, it's 46.3 - 45.5.

Man, the doomer-ism in here is one thing, but y'all are going off about a crosstab that has 36 (Hispanic) and 49 (black) people in it!

Statewide polling is going to have rough samples with minorities, but I think we can all agree an 85 person sample for minorities is just terrible. Thus, the sample with 36 hispanics gets a wonky result (Oz +60... which obviously is not happening)

Same thing with "Oz having strength with 14% of Blacks" .... it's a tab of 49 people. That's like, what, 6 people? lol

What I find interesting that people are ignoring here is the White vote... it's literally tied, again. If the white vote is anywhere close to tied, it's great news for Fetterman. And the n= is at least strong here, 465.

(not to mention IA is an R-leaning pollster)

Of course as a proportion the White vote share has a small margin of error. However, in absolute terms it doesn’t. A 5% sampling error on the White vote causes as much of a shift in the top line as a 25% sampling error in the minority vote. You can’t hold the White % fixed and correct the minority cross tabs.

Trump won the White vote in PA by 15 points, so if you’re gonna play the cross tab game that should raise some eyebrows. I know the ‘other’ category has obviously inaccurate/weird results, and that a tie in the White vote seems more plausible because the state as a whole is tied, but you have to realize that a 15pt swing wouldn’t happen either.

Most of this can be attributed to the fact that IA is a mediocre pollster. It doesn’t seem like they weight by education, which is why they likely have the Democrats doing so well among Whites. They counter that by finding an extremely unrepresentative minority sample (which is a common theme among R pollsters, such as Trafalgar).

You're proving my point why this poll at large shouldn't be taken that seriously.

I would just happen to say though that finding Oz and Fetterman close among Whites has borne out in numerous other polls though, more often than not. A n=450 sample is always going to be a bit more reliable than a n=36 sample. So it's not 'playing the crosstab game.' One sample is clearly more reliable to look at than the other, just in sheer quantity.

My point was a small error in the White cross tab causes as much of an error as the entire messed up Hispanic cross tab. If the Hispanic cross tab was 60-30 Fetterman and the White vote was 49-45 Oz the poll would be the same. This poll isn’t great, but the trendline is still in Oz’s direction. It was conducted with the same methodology both times.