NWR: Election Statistics

I

Ian Fitzsimmons

Ian Fitzsimmons
Apropos the bored's occasional interest in things statistical, I've been engrossed by Nate Silver's mathematically-grounded 538 column in the NYT's coverage of the presidential election. It's been a nice way to separate conjecture from evidence, and he is pleasingly transparent in his reasoning.

If he's accurate, it's a shame the extent to which commercial news is attached to the tie scenario, presumably for commercial purposes.
 
The problem is that so few people in the media have any appreciation for statistics or the pitfalls of poorly done statistical analysis. The parties have no reason to look at the numbers honestly, so it falls to the mathematically minded people like Nate Silver to do the job.

Mark Lipton
 
Nate Silver bases his projection on all the polls available to him as well as his analysis of the historical performance of various polls under various situations. It is a very impressive meta-analysis. News media just report whatever poll is in front of them. And, of course, media with greater or lesser party ties--Wall Street Journal, Fox, CNBC, etc.--report polls that favor their side, when they can. There really is no reason to expect them to do what he does, which is why he does it. Since 538 is an NY Times blog, by the way, it is commercial news as well.
 
originally posted by Ian Fitzsimmons: he is pleasingly transparent in his reasoning.
Click to expand...

Transparent?

He's running simulations for elections based on models with tons of assumptions and of course he won't release the algorithms because they are proprietary. Fair enough, but it's not exactly adding signal to noise. Especially when his models show you the likelihood of Obama or Romney winning if there were thousands of elections, and in reality there is only one election with determining factors that he likely cannot capture.

Not to say that I'm opposed to him getting coverage. It's a step in the right direction. But he's also involved in a lot of oversimplification.
 
originally posted by Rahsaan:
originally posted by Ian Fitzsimmons: he is pleasingly transparent in his reasoning.
Click to expand...

Transparent?

He's running simulations for elections based on models with tons of assumptions and of course he won't release the algorithms because they are proprietary. Fair enough, but it's not exactly adding signal to noise. Especially when his models show you the likelihood of Obama or Romney winning if there were thousands of elections, and in reality there is only one election with determining factors that he likely cannot capture.

Not to say that I'm opposed to him getting coverage. It's a step in the right direction. But he's also involved in a lot of oversimplification.
Click to expand...

I'll bow to the arguments of statisticians, who know more than I do, except for the bit about modelling thousands of elections. He's estimating the probabilities of thousands of different situations occurring in order to predict the odds of one thing finally occurring. I would have thought that that's what one did--unless, of course, we're talking about Schroedinger's cat, which of course we're not.
 
originally posted by Zachary Ross:
Sam Wang has been even more accurate than Silver, albeit without news-organ support.

Loading…

election.princeton.edu
Click to expand...

I don't know whether this has been more accurate or not (and since what date?). But the blog is much less transparent about its reasoning than Silver's and writes with a certitude that seems inappropriate for the activity in which he is engaged. Neither of these things, of course, makes his analysis weaker on its own terms. It just makes me less likely to follow him--even if he is telling me what I want to hear.
 
originally posted by Rahsaan:
originally posted by Ian Fitzsimmons: he is pleasingly transparent in his reasoning.
Click to expand...

Transparent?

He's running simulations for elections based on models with tons of assumptions and of course he won't release the algorithms because they are proprietary. Fair enough, but it's not exactly adding signal to noise. Especially when his models show you the likelihood of Obama or Romney winning if there were thousands of elections, and in reality there is only one election with determining factors that he likely cannot capture.

Not to say that I'm opposed to him getting coverage. It's a step in the right direction. But he's also involved in a lot of oversimplification.
Click to expand...

I thought all he was doing was meta-analysis of the public polling?
 
originally posted by Jonathan Loesberg:
[...] Since 538 is an NY Times blog, by the way, it is commercial news as well.
Click to expand...

I thought of this after posting, and you're right. Despite featuring his column, though, most NYT coverage is along the same lines you see in the Post, the LA times, CNN, ABC ... that the race is a tie.

Some of the stuff Silver does can hardly be so sophisticated as to exceed the grasp of regular professional statisticians - like aggregating and averaging data to reduce margin of error. I believe he runs multiple trials to achieve better estimates under the theory of large numbers, which also isn't terribly esoteric. I can't believe that, say, Chuck Todd couldn't replicate this kind of analysis. So my inner conspiracy theorist is wary. I imagine NYT embraces him (as long as he doesn't make bets on TV) partly because his confidence is provocative, partly to hedge their bets, while their headlines are all about the cliffhanger tie.

originally posted by Rahsaan:
originally posted by Ian Fitzsimmons: he is pleasingly transparent in his reasoning.
Click to expand...

Transparent?

He's running simulations for elections based on models with tons of assumptions and of course he won't release the algorithms because they are proprietary. Fair enough, but it's not exactly adding signal to noise. Especially when his models show you the likelihood of Obama or Romney winning if there were thousands of elections, and in reality there is only one election with determining factors that he likely cannot capture.

Not to say that I'm opposed to him getting coverage. It's a step in the right direction. But he's also involved in a lot of oversimplification.
Click to expand...

I take your point; what I mean is that he explains with decent clarity what he's doing, why, and how, at a technical level appropriate to a general NYT reading audience (with maybe a slight stretch). This standard of transparency naturally falls short of a scientific one, where the point is to enable reproduction of results. Even if he were not protecting proprietary algorithms, a thorough technical 'methods' section would drive off readership, don't you suppose?

Re: multiple trials, as above, isn't this a conventional technique to improve statistical estimation?

Re: simplification, all modeling and estimation is simplification - therein lies its value. Imo, Silver strikes a better medium between simplification and complexity than much of the other analysis I'm taking in.
 
"The situation is that many of Nate Silver’s attackers don’t really know what the hell they are talking about. Unfortunately, this gives them something in common with many of Nate Silver’s defenders..."

Loading…

www2.macleans.ca
 
That's actually a pretty weak article, especially his discussions of the weakness of Pecota and where it has failed. Statistical models fail with outliers almost by definition.
 
originally posted by Jason D:
That's actually a pretty weak article, especially his discussions of the weakness of Pecota and where it has failed. Statistical models fail with outliers almost by definition.
Click to expand...

Exactly, statistical models aren't about outliers at all.

Also, the probability of any single event is either 1 or 0.
 
Unfortunately, Nate Silver isn't competing in the space of "statistical models."

Anyways, my own--completely uninformed--opinion is that Silver engages in far more analysis than is warranted by the quality of the raw data. But if you can get paid for mathematical masturbation, you might as well go for it...
 
originally posted by Arjun Mendiratta:
Unfortunately, Nate Silver isn't competing in the space of "statistical models."

Anyways, my own--completely uninformed--opinion is that Silver engages in far more analysis than is warranted by the quality of the raw data. But if you can get paid for mathematical masturbation, you might as well go for it...
Click to expand...

Oh, sure, but he isn't putting these things out for peer review. They're in a blog.

I've been surprised at how well such simple measures have worked.
 
You must log in or register to reply here.
Back
Top