Demographics and the Electorate

[afleitch]

This issue has crossed my mind quite a number of times as well.

If you use the Reuters data for the past two weeks (which takes in Trump's good week and Hillary's good week), Clinton's two party vote share is 43.8 with whites and 96 with blacks. With Hispanics they only made up 2% of the numbers they polled (and less than 1% on an LV model), but on that she's got 78% and 65.1% of Asian's and others.

That gives an Obama 2008 result exactly (including Indiana), with an 11.6 vote lead...

For the record, the whole sample (20,000+ voters) is 45, 92, 81.7, 66.4 respectively, so fairly stable

Now part of the reason why these demographic models were put in place for this year (and indeed in 2012) is that regression modelling suggested a strong correlation. To some extent, based on prior polling, they were used relatively successfully in predicting the primaries this year and for example The Upshot used them on election nights themselves.

As I jokingly said to another poster, Yeah, we can perhaps scratch off Iowa which she was struggling with even when she was at her convention high. We can make Virginia more safe (which is a more than fair exchange) but around the margins, unless you lock up minorities, women and college voters in a room on election day and clone a couple of million non college educated whites, I fail to see how she's going to do particularly worse than Obama did with these sorts of numbers.

[afleitch]

I did a little experiment using the exit poll data for the 2014 House elections and used Real Clear Politics' own demographic widget. It predicted that the Democrats would have won 48% of the two party vote. Instead they won 47.1. Not too bad a margin of error. However the differences in each of the states was stark and varied greatly from the 0.9% difference recorded nationally.

If we set the Democratic national vote to 0, then the worst swing state for the Democrats was Nevada; at -9.1 points worse for the Democrats than the model predicted. The next worse state was Ohio at -8.6, then Pennsylvania at -5.1, Missouri at -4.8 then Iowa at -4.7.

The best swing states for the Democrats, where they over polled based on the estimate was New Mexico at 4.5, Arizona at 3.2, Georgia at 2.1, New Hampshire at 1, Colorado at 0.7 and Maine and North Carolina at 0.4

Not an unfamiliar pattern of states surely....?

The difference between 2014 and now is that we are using polling data, rather than the exit poll. And the results in 2014 were brutal for the Democrats (as was 2010) and turnout was at a 60+ year low and woeful amongst minority groups.

If we apply the 2014 turnout to the 2012 results, then the Democratic share is reduced to 50.9 and Florida flips to the Republicans. If we apply the state variance we found in 2014, then the states currently with Trump on 538 flip to him as well as Virginia (which we can perhaps exclude this year) and Pennsylvania.

With the Reuters data, but the 2014 turnout, the Democratic share is better 55.4. Applying the same state variance sees Trump win Ohio, Nevada and Virginia (again with the caveats), but just lose Iowa. Clinton wins all the other Obama 2012 states and adds North Carolina to her tally.

So, the LV screen appears to be suggesting a 2014 style turnout (in ratio if not in actual percentage) and even then Clinton can win, providing she relies on her minority base. The question is, they thought 2012 would be like 2010, when it was more like 2008. If 2016 is like 2012, then she's home and dry.

[afleitch]

Question - is there any analysis on what the polls projected turnout is? Obviously a single poll is unreliable, but if you aggregate some you should get a picture of what pollsters are predicting.

This is Reuters (who I can only praise for their willingness to share data)

[afleitch]

Question - is there any analysis on what the polls projected turnout is? Obviously a single poll is unreliable, but if you aggregate some you should get a picture of what pollsters are predicting.

This is Reuters (who I can only praise for their willingness to share data)

That data makes no sense at all... I mean, wouldn't that suggest that turnout was largely stable 08-10-12? When it wasn't at all. I mean 2014 makes sense in relation to the data, considering 2010 turnout was only slightly better than 2014... Unless I'm completely misreading this.

Is it just me or is this an assumed projection...

The graph is RCP. They haven't used midterm data at all so that's misleading. It's been extended from 2012 to 2016 using Reuters turnout model.

[afleitch]

Question - is there any analysis on what the polls projected turnout is? Obviously a single poll is unreliable, but if you aggregate some you should get a picture of what pollsters are predicting.

This is Reuters (who I can only praise for their willingness to share data)

That data makes no sense at all... I mean, wouldn't that suggest that turnout was largely stable 08-10-12? When it wasn't at all. I mean 2014 makes sense in relation to the data, considering 2010 turnout was only slightly better than 2014... Unless I'm completely misreading this.

Is it just me or is this an assumed projection...

The graph is RCP. They haven't used midterm data at all so that's misleading. It's been extended from 2012 to 2016 using Reuters turnout model.

That's even more ridiculous, then. They're comparing the final results, based on the cross-tabs in one poll, dependent current subjective LV screens?

That's the issue I have. They've effectively upped white turnout to recent historical highs and dropped minority turnout to mid term levels. Which is unprecedented.

[afleitch]

Aren't there better polls than Reuters to use for this?

I can't find any other giving projected turnout rates by different groups. There's a lot of smoke and mirrors when it comes to pollsters stating what the LV screen actually is.

[afleitch]

http://www.bloomberg.com/politics/graphics/2016-poll-decoder/

I was actually doing this myself so I'm glad this does it for me! It takes all the demographic data out of the national, non tracker polls and puts it together. You can compare it with the exit polls.

[afleitch]

Aren't there better polls than Reuters to use for this?

I can't find any other giving projected turnout rates by different groups. There's a lot of smoke and mirrors when it comes to pollsters stating what the LV screen actually is.

Even without turnout rates don't a lot of pollsters publish crosstabs with nr of respondents included? I'm guessing that would allow backing out what they think the shares of the electorate are.

Well for example if we take the Bloomberg poll, they don't even provide demographic breakdowns at all. Monmouth gives me that, but not raw numbers to calculate turnout and so on. A few polls give what they assume the makeup of the electorate is, but not differential turnout. Indeed some of them appear to be doing exactly what they did in 2012 with respect to weighting.

[afleitch]

I think the original question being asked was where Trump is making up ground in the polls relative to what we'd expect from demographics. Is white turnout being driven up like crazy relative to non-white turnout? Is college white turnout depressed? Is male turnout up relative to female turnout? Demographics may not be destiny, but some explanation is in order when the demographics from last time, combined with what the polling tells us of the demographics now, give us such a starkly different result from the polls.

I think pollsters are thinking 'Trump f-ck yeah' is a stronger sentiment than 'Hillary, if I have to', when it may actually bring voters out to the polls in equal measure. There was even a sense of that in 2012, which threw off the LV model.

[afleitch]

Pew did a good piece on party ID trends. Data here:

http://www.people-press.org/2016/09/13/party-identification-trends-1992-2016.

Pew (who IIRC did well in 2012) Have 70% of the 2016 electorate as white, down from 73% in 2012 and 74% in 2008.