Thursday, August 4, 2016


Twitter is . . . a-twit . . . with the latest poll numbers showing a Trump tumble in the polls. For example:
At least this is focused on state-polling. Before we go deeper, let's look at the top-level facts.

  1. There are four states that matter right now: FL, OH, PA, and NC. Trump really, really wants to win the first three and hold the fourth. 
  2. PA Polling (-11) isn't good--but turn-out models matter (and we won't know the real mix of turn-out until election day). PA has had a surge in R voter registrations which may or may not be reflected in the specific polls.
  3. The national polling numbers get a lot of play--but until the state polls reflect the trends, they don't mean so much. Right now we could could be looking at a re-alignment--or just a convention bump that will diminish over August.

That Said

Getting hammered in the polls certainly isn't good for Trump. For one thing, he's made his polling a centerpiece of the argument for him. For another, the specifics of the polls--that he should stop fighting with the Khans, for example--do make it look like his mouth is getting him into trouble. The Omnivore wants to look at this:
This is the "Median Electoral Vote Estimator" from the Princeton Election Consortium (PEC). It uses a specific methodology to generate a very accurate (historically) estimate what what will happen. The Omnivore calls your attention to three points:

  1. The horizontal red line is the Clinton-Wins line. As you can see all the days of aggregation fall above it. The black line is the computed median Electoral Vote--right now every day has Clinton winning. The standard deviation for all the polls is the gray zone within 95% chance. There's like a handful of days thus far that Trump had a chance. A small chance.
  2. The red and yellow "strike zones" are estimations of how far the Median EV count could move between now and the election. The 'elasticity' of the Median EV count is based in prior history of this election and prior elections. The red and yellow lines show 95% probability. So right now there's like an 80% chance of Clinton winning based on what we know--of course for her to lose, things need to trend in Trump's favor--but that could still happen.
  3. The black line is pretty stable. This calculation uses state-polling only. As such, it only reflects substantial changes across multiple key states--big swings in national polls don't show up at all. Buy this metric, the election fundamentals has yet to change much.
It is worth noting that the highly regarded Larry Sabato group has much the same concept and, also, has not changed as much as today's headlines would have you believe:

In our view, it is much more fruitful to focus on the electoral fundamentals and fixed elements of politics that predetermine most votes, especially partisanship, demographics, and strong forces shaping the political landscape. Polarization in this hyper-partisan era means that practically nine of 10 voters are committed, and the unknown is whether they can be motivated to cast a ballot. Presidential job approval, the state of the economy, war and peace, and a few other items reinforce partisanship and turnout, and influence the few truly swingable votes among hard independents.
They also note that big swings, in an era of almost no one taking polls,  may be due to a change in partisan response rates. In other words, if you are a Trump supporter and Trump's news is in the toilet and you get a call from a pesky pollster, you just hang up. If Hillary is "doing great" on Cable TV, you're smugly thrilled to answer the questions--and vice-versa.

This, if entirely true, means that elections are like a couple billion-dollar waste of money.

NOTE: We know this is not entirely true--there are surveys that keep the same group of people throughout an election and track their movement over time. These do show changes in candidate choice to, at least, some degree.

What This Means for Trump-watch

The key to seeing a categorical collapse of Trump would need to (a) appear in state voting (specifically those four key states) and (b) show movement at the top of--or preferably outside of--the normal EV-elasticity range. Secondly, it will probably need to show evidence to / after Labor Day (August is considered a dead-zone in electoral politics).

If we see that then, The Omnivore asserts that people will start thinking of their own political futures in terms of how much longer to "ride the Trump Train." Also keep an eye on on finger-pointing. If internal accusations reach the level of the press (finger pointing at people other than Trump from otherwise stalwart supporters) then it means that exits are being prepared.

We probably have 3 weeks of this before we get to a real "death-watch" condition.

1 comment: