STUMP » Articles » Prep for Pointing and Laughing: Some Pre-Election Polling » 26 October 2016, 12:27

Where Stu & MP spout off about everything.

Prep for Pointing and Laughing: Some Pre-Election Polling  


26 October 2016, 12:27

If I’m going to suck at predictions, then I’m going to grab a bunch of people’s predictions right now knowing that at least one of them is hideously wrong.


First, let’s go to the guy who achieved fame with his predictions: Nate Silver.

Now he has a nifty prediction tracker and you can go to his site for the bells-and-whistles. I’m just doing a screengrab for posterity:

Now, 538 and other sites sometimes give probabilities that Clinton will win (as I look at 538 right now, they give her a 85% chance of winning), but I can’t measure that. Either Clinton will win or she won’t. But I can compare the popular vote and electoral vote results.


Let’s go to Real Clear Politics, which does its own poll averaging:

Unsurprisingly, both 538 and RCP give similar results.

So let’s go looking for some extreme results, for both Clinton and Trump.


This is the most recent result from the ABC Tracking poll.

I picked this one from the RCP list, because it had the widest margin between Trump and Clinton.


Now, on the Trump-positive poll side, the most extreme one I can find is the LA Times Tracking poll:

One percentage point. Not quite as extreme as Clinton’s lead, eh?

Thing is, these can’t all be right, and I don’t just mean in the obvious way. All these polls tend to claim a confidence interval of +/- 3 percentage points. Now, that does mean some of these results do overlap, but the extreme polls are really pushing it.


So now I’m getting away from the polls, and looking for what specific people say.

Professor Norpoth claims Trump will win:

Norpoth, who has accurately predicted the last five presidential elections, uses two models. The first, he told Fox News’ Tucker Carlson, is the primary model, which tracks how the candidates are performing, and “it usually turns out the candidate who does better in his party’s primaries or her party’s primaries beats the other guy who does less well.”

He used the primary races in South Carolina and New Hampshire for the 2016 election, where Trump won and did better than Clinton did in the Democratic race.

The second model, he explains, is called the Swing of the Pendulum, which is the tendency for a change after two terms of a party being in the White House. This, of course, gives the prediction that a Republican will win in 2016.

Norpoth is so convinced by his models that he’s putting his money where his mouth is. He bought shares in the Republican candidate in the Iowa Electronic Markets a long time ago and he’s sticking with it, he said.

Let’s check out what the betting markets say!

Iowa Electronic Markets have two bets going on: one with a payoff determined by percentage share of popular vote, and the other is winner-take-all. So the prices for VS should indicate popular share prediction and WTA should indicate probability of win.

That’s some odd movements… and I think the prices are based on %age share excluding all the non-Dem/non-Rep candidate vote counts. Fair enough.

The Good Judgment Project has a bunch of open propositions, but unlike the IEM, no money is at stake. Just bragging rights.

Alas, they don’t guess at how the popular vote will break out, just who will win. They currently give Clinton a 76% chance of winning. Less than 538, obviously.

Michael Moore, back in July, gives reasons why Trump will win. (there’s no date on the piece at his site, but he links to this alternet piece from July, which is how I date this.)

Finally, Don Surber questions the quality of operations for these polls:

When I am polled

As one of the last landline owner (I have no cell phone) I take a lot of calls from pollsters, mainly local races. When I was working, I politely told the caller of my newspaper work and got excused. Retired, my opinion is viewed as worthy. My experiences offer why I am skeptical of polls.
The second thing I noticed is the calls now are all automatic. You respond by pressing numbers. That is a vast improvement because it eliminates one source of human error. You get the same questions asked the same way each time.

But punching numbers is a problem. I make mistakes and hang up. Why not use voice recognition? Expense? Seriously? ABC, CBS, CNN, Fox News, and NBC have so much money, they blow their noses in fifties.

I have no idea how well these polls are conducted. They can’t get at me or my husband, because we screen all our calls. The only time I answer unknown calls is when I’m feeling a bit mischievous.

At least I used to. Now I am more likely to go mess with my kids’ devices or something.

Anyway, it does seem the consensus is that Clinton will win and Trump will lose. Even if that’s the case, the question will be how far off the vote counts the predictions are.

I’m just an onlooker here, as I live in New York. It really doesn’t matter who I vote for. If Trump (or any non-Clinton) wins, New York will not be part of that win.

To find the last time New York went for a non-Democrat, you have to go back to Reagan’s landslide in 1984. When I was 10 years old.

So the easiest prediction I have to make is I’m just paying attention to my local elections for actual electoral activity on my part.

Related Posts
Memento Mori, or, Happy Birthday to Me
Foolery for 2019: I Hate the NY State Legislature (and more)
Meep in 2020: A Few Notes on Where to Find Me