OH-Bloomberg: Trump +5/+5

[Wiz in Wis]

A couple things

1) This poll isn't exactly an outlier. Red Avatars should calm down. Trump has been up in OH in the last few weeks of polling. Given what we know about the WWC vote, Ohio is a likely candidate state for him to flip, even if he's down a few nationally. It was to the right of the nation in 08 and 12.

2) We have no trend-line to look at with Seltzer here. It may be that their LV screen is too tight, or it may be accurate. We don't know because we don't have RVs.  Even in the Bloomberg link, they seem somewhat cautious: They go out of their way to say their LV screen is shifting a lot of support to Trump:

“Our party breakdown differs from other polls, but resembles what happened in Ohio in 2004,” said pollster J. Ann Selzer, whose Iowa-based firm Selzer & Co. oversaw the survey. “It is very difficult to say today who will and who will not show up to vote on Election Day. Our poll suggests more Republicans than Democrats would do that in an Ohio election held today, as they did in 2004 when George W. Bush carried the state by a narrow margin. In 2012, more Democrats showed up.”

3) Even if you are inclined to think Clinton is a bad candidate, she's outpacing Strickland by 12 and generic house D by 8. Again, it seems as if GOPer's in general are much more "likely" to say they are voting in this poll. That would point to an enthusiasm gap for the Dems, rather than a shift in preferences. The debates can turn that around on a dime

4) If the best Trump can do is +5 in a state that he absolutely needs to win, that is demographically favorable to him, and at Clinton's lowest point (post-deplorables, post convention, mid-pneumonia), that's hardly worth bragging about.

[Wiz in Wis]

Worth reading and keeping in mind in general:

http://www.stat.columbia.edu/~gelman/research/unpublished/swing_voters.pdf

[Wiz in Wis]

Worth reading and keeping in mind in general:

http://www.stat.columbia.edu/~gelman/research/unpublished/swing_voters.pdf

Their results are not generalizable beyond the 2012 election.

the concept of differential response rates based on enthusiasm is not generalizable? Based on what, your gut?

[Wiz in Wis]

It is f**king Selzer! Third best pollster according to 538.

Bull****

Is it really 'bs' to find issue with the sample the poll used? It is basically 2004's electorate, but only slightly less white (per nyt ep), and somehow less Hispanic:

White: 83% (was 86% in 2004 nyt ep) vs 79% in 2012
Black: 11% (was 11% in 2004) vs 15% in 2012
Hispanic: 2% (was 3% in 2004) vs 3% in 2012

https://assets.bwbx.io/documents/users/iqjWHBFdfxIU/r2.771xfmKOI/v0
http://elections.nytimes.com/2012/results/president/exit-polls (2004-2012)

Now, if this is what their polls tell them the electorate might look like, then ok, but before this election ever started I wholeheartedly believed the electorate was never going to look like 2004 again. IF this is what the poll really assumes, then I have to believe it is wrong. I won't presume to say how wrong, though. It's not about being some partisan hack that just wants to "unskew" to assuage my fears of a Trump presidency, but rather about how I hate this idea that because Obama isn't on the ticket, somehow everything is going back to 2004 and minorities will forever stay low-turnout. That theory is ludicrous. So is this indeed the sample of voters they used?

One thing worse than serial unskewers is the people who go around yelling "HAHA UNSKEWER" or the like whenever someone raises any kind of concern with a poll.

And for the record, since I can't rely on you to give any benefit of the doubt versus your desire to insult people, I'm only raising concern with the sample here and not trying to say Trump isn't closing the gap or perhaps even winning in Ohio right now.

This is similar to what happened with Gallup in 2012. They used an antiquated sample and ended up being wrong in the end.

It's not the sample... it's the LV screen. Lean Trump voters are more likely to say they will vote right now than Lean Clinton, at least using their screen. It's not rocket science.

[Wiz in Wis]

A lot of text.

Firstly, turnout might change. The biggest "homogeneous" group in 2012 was non-college-educated Whites. And they had a very low turnout (56%).

Secondly, exit-polls usually underestimates non-college-educated Whites.

Thirdly, looking at selfreported party identification in a LV-model is exactly what unskewing is about.

Not "unskewing." If the election were held today, I'd bet Trump wins OH. But, the LV screen is definitely picking up differential enthusiasm to answer polls, or self-rate their probability of voting higher among GOPers. That is not the same as voters switching to Trump from Clinton.

[Wiz in Wis]

Worth reading and keeping in mind in general:

http://www.stat.columbia.edu/~gelman/research/unpublished/swing_voters.pdf

Their results are not generalizable beyond the 2012 election.

the concept of differential response rates based on enthusiasm is not generalizable? Based on what, your gut?

No, their conclusions that "vote swings in 2012 were mostly sample artifacts" and from selection bias and that "a pivotal set of voters, attentively listening to the presidential debates and switching sides" doesn't exist are as much artifacts of their own selection bias in choosing the 2012 election as anything.

Differential response rates to opt-in polls are obviously a real thing.

Your suggesting that survey methodology examined in 2012 cannot be used in 2016 because 2016 is not 2012? That's not how generalization works. If a survey design issue is uncovered in 2012, if those same approaches exist in 2016, you can make assumptions that the same phenomena are still occurring. I mean, that's why convention bounces are a thing... enthusiasm for candidates goes up, as does the likelihood of participating in a poll.

I'm saying that differential participation in polls is at least partly responsible for this result, as are shifts in enthusiasm affecting the LV screen. However, these may not be all that strongly correlated to final vote. In other words, fewer people switch their minds day to day than these polls would suggest.

I mean, which of these two scenarios are more likely:

Between August 7 and August 28, 10% of Wisconsin voters switched from supporting Clinton to supporting Trump, or, differential rates of enthusiasm, based on the Dem convention and then subsequent bad news for Clinton, affected both participant self-report on probability to vote and probability to participate in the poll in the first place, while actual voter preference changed by a much smaller amount.

[Wiz in Wis]

Worth reading and keeping in mind in general:

http://www.stat.columbia.edu/~gelman/research/unpublished/swing_voters.pdf

Their results are not generalizable beyond the 2012 election.

the concept of differential response rates based on enthusiasm is not generalizable? Based on what, your gut?

No, their conclusions that "vote swings in 2012 were mostly sample artifacts" and from selection bias and that "a pivotal set of voters, attentively listening to the presidential debates and switching sides" doesn't exist are as much artifacts of their own selection bias in choosing the 2012 election as anything.

Differential response rates to opt-in polls are obviously a real thing.

Your suggesting that survey methodology examined in 2012 cannot be used in 2016 because 2016 is not 2012? That's not how generalization works. If a survey design issue is uncovered in 2012, if those same approaches exist in 2016, you can make assumptions that the same phenomena are still occurring. I mean, that's why convention bounces are a thing... enthusiasm for candidates goes up, as does the likelihood of participating in a poll.

I'm saying that differential participation in polls is at least partly responsible for this result, as are shifts in enthusiasm affecting the LV screen. However, these may not be all that strongly correlated to final vote. In other words, fewer people switch their minds day to day than these polls would suggest.

I mean, which of these two scenarios are more likely:

Between August 7 and August 28, 10% of Wisconsin voters switched from supporting Clinton to supporting Trump, or, differential rates of enthusiasm, based on the Dem convention and then subsequent bad news for Clinton, affected both participant self-report on probability to vote and probability to participate in the poll in the first place, while actual voter preference changed by a much smaller amount.

No, I'm not suggesting that. I'm not really talking about polls at all. We're talking about two different aspects of this paper.

Oh... ok. Well, what in the 2012 approach do you think doesn't apply to 2016. Also, interestingly, it appears that the YouGov/Economist poll uses the same approach that Gelman/et al. did with the X-boxes, in that it also uses ideology as a weighting factor. YouGov has always had the race between +1 and +6 Clinton... they really don't show outliers. Today, they show Clinton +2, same as last week.