Slog

News & Arts

Line Out

Music & the City
at Night

« Do You Need Something To Do This We… | Two Questions Regarding Taylor Swif… »

Friday, November 2, 2012

Politics / Science Univac Versus CBS, Redux

Posted by Jonathan Golob on Fri, Nov 2, 2012 at 3:50 PM

The kerfuffle between the political commentariati and Nate Silver (of FiveThirtyEight) is baffling to me.

One side has data, the other doesn't. What's the argument?

The whole thing reminds me of a classic moment, early in the computer era:

Remington Rand (now Unisys) approached CBS News in the summer of 1952 with the idea of using Univac to project the election returns. News chief Sig Mickelson and anchor Walter Cronkite were skeptical, but thought it might speed up the analysis somewhat and at least be entertaining to use an "electronic brain."

Eckert and John Mauchly enlisted their former Penn colleague, mathematician Max Woodbury, to assist. Mauchly and Woodbury gathered data and wrote a program that would compare the 1952 returns to previous elections and figure which way the wind was blowing. The duo worked at Mauchly's home because he'd been blacklisted as pro-Communist and wasn't allowed to work at the company anymore.
....
Pre-election polls had predicted anything from a Democratic landslide to a tight race with the Demo candidate, Illinois Gov. Adlai Stevenson, slightly ahead of the Republican, five-star Gen. Dwight D. Eisenhower, Supreme Commander of Allied Forces in Europe in World War II.

So it was a surprise at 8:30 p.m. Eastern time when Univac predicted Eisenhower would pile up 438 electoral votes to Stevenson's 93. The odds of Eisenhower garnering at least 266 electoral votes — the minimum needed to win — were 100-1.

In New York, news boss Mickelson scoffed at putting the improbable prediction on air. In Philadelphia, Woodbury added new data to the mix. At 9 p.m. correspondent Charles Collingwood announced to the audience that Univac was predicting 8-7 odds for an Eisenhower win.
...

As the evening wore on, an Eisenhower landslide gathered momentum. The final vote was 442 to 89. Univac was less than 1 percent off.

Data, modeling (rigorous, high-quality mathematical modeling) works, particularly when one is trying to cut through loud voices spouting bullshit.

The situation is exactly analogous to how Americans think (often incorrectly) about climate change. There aren't two equally valid ways of looking at the situation: there is a data-based and rigorous way, and then there is superstition. Stick with the guy peddling data, and explaining how he's coming to his conclusions.

Count me in with Nate.

FacebookShare

Comments (16) RSS

Oldest First Unregistered On Registered On Add a comment

It is extremely easy to screw up a data analysis in my humble experience. That said: an experienced and proven statistician in a domain they are expert in using solid raw data as a starting point using tools they are familiar with, is typically worth way more than anyone who doesn't meet all those criteria.

Unfortunately you can be smart and wrong if any of the above doesn't hold. Worse, one can sound expert without much expertise.

I think that last may be the tragedy of our age.

Long winded way of saying I agree with you.

Posted by david on November 2, 2012 at 4:13 PM

On the other hand, GIGO. But in this case you are probably right (see what I did there?)

Posted by Cynic Romantic on November 2, 2012 at 4:44 PM

What's very gratifying is that there is actually a right answer to this question. All we have to do is wait for 4 days, and we will see who was right all along, and who was just blowing smoke. I wish we could add the condition that the losers have to STFU for a set period of time, but I'll settle for the re-election of President Obama as a consolation prize,

Posted by Gurldoggie

on November 2, 2012 at 4:55 PM

I found this Deadspin article to be an interesting take:
http://updates.deadspin.com/post/3478090…

The parallels to his Sabermetrics baseball stats experience are illuminating.

Posted by r.chops on November 2, 2012 at 4:58 PM

Well, Jonathan, here's a dark & cynical theory for your consideration:

Say you want to maintain the capacity to manipulate the vote-tabulation process somewhere.

If the results from a manipulated district deviate from the forecast by more than the margin of error, it's unacceptably conspicuous.

So if you want to maintain a capacity to manipulate vote tabulation, it's important that predictive forecasting methodologies not gain too much credibility. Better that every district should be a "toss-up," where outcomes are influenced by vague, various and subjective influences that cannot be measured.

Posted by Eric Arrr on November 2, 2012 at 5:02 PM

I'm with you....nivac.

ELECTORAL COLLEGE MODEL PREDICTS ROMNEY WILL WIN EVEN BIGGER THAN PREVIOUSLY THOUGHT IN 2012

Despite the fact that polls still show a dead-heat race (Obama is currently at 48.2 percent, with Romney capturing 47.3 percent of likely voters in the most recent Real Clear Politics average), an updated election model shows an even larger gap between the Electoral College votes that Romney and Obama are projected to win. According to Bickers and Berry, the Republican challenger is projected to take 330 of the 558 votes, while Obama is expected to capture only 208 of them.

http://www.theblaze.com/stories/electora…

Posted by Supreme Ruler Of The Universe

on November 2, 2012 at 5:35 PM

@6 - This method does not even consider polls - just economic data - which is dangerous when trying to predict a win for the most unlikable candidate in the history of likability polling.

(By this method, they predict Romney will win Minnesota! Do you think Romney will win Minnesota?)

Basically, they came up with a formula that retroactively fit the last eight elections. But why stop there? Why not see if your model works against all elections in the 20th century? There is no shortage of economic data to plug into their model.

But they stop at 1980. Odd. My guess is because their model falls apart with a bigger sample size.

Posted by Free Lunch on November 2, 2012 at 6:20 PM

Once again Bailo ignores the old add age: "don't bring a pocket calculator to a super computer fight"...

Posted by COMTE

on November 2, 2012 at 6:26 PM

If Silver were an anomaly—just some unknown guy we never heard of whose numbers were way different than everyone else's—then I could see why people would complain. But (1) Silver has a great deal of credibility, and (2) pretty much all of the other major polling aggregators are coming up with roughly the same predictions. All of them are predicting a narrow Obama win at this point. Pollster.com, electoral-vote.com, Princeton, TPM, etc, are all predicting an Obama win. Even Real Clear Politics (a very right leaning site) is showing Obama with a significant lead in the electoral college numbers.

Nate Silver is probably the best known of the data aggregators, so he has apparently become the lightning rod for complaints. But he's not just sticking his finger in the wind and blindly guessing. He's using a lot of hard polling data, and everyone else who is seriously crunching the numbers is coming up with basically the same answer he is: Obama has a narrow but significant and persistent lead in the electoral college. And Rmoney's bump that he got after the first debate has stalled and reversed.

At this point, I'm more worried about R-74 than I am about Obama.

Posted by Reverse Polarity on November 2, 2012 at 6:31 PM

...lead over Mitt Romney has been cut from eight points last month to three last weekend at 47-to-44 percent.

http://www.piercecountyherald.com/event/…

Posted by Supreme Ruler Of The Universe

on November 2, 2012 at 6:34 PM

I check 538 daily, though I put more creedence on Real Clear Politics, because they were so spot-on last time, plus they won't give me a rosy picture.

So I like 538. But what I found annoying was the way my fellow liberals slavishly transformed him into a sort of oracle. "Nate said it so it is right." Come on, man. I know he's a liberal like us, but that doesn't mean that we have to shut off our skepticism.

On the other side, I do think that much of the criticism is coming from irate conservatives who root for the team and lash out at anybody who disagrees with them. Nate Silver is a young, well-educated (apparent) liberal who works for the New York Times. He represents hipster success to them. That's why they hate him.

Posted by floater on November 2, 2012 at 8:37 PM

@10 - Fair enough. But seriously, you're advocating a methodology that puts Minnesota in the Romney coumn, yet hands Obama Iowa?

Posted by Free Lunch on November 2, 2012 at 10:27 PM

@10 - So you're stating that you do believe Romney will carry Minnesota?

Posted by Christampa on November 2, 2012 at 10:33 PM

I'm a statistician myself. Building these kinds of models is my job, I do it all day.

That said, Joe Scarborough is doing HIS job, being a journalist on a 24 hour news program.

Scarborough's gotta keep his audience in order to keep his advertisers, so he'll say whatever it takes to keep them watching. And right now, he's got to avoid the facts to accomplish that.

Posted by Urgutha Forka on November 2, 2012 at 10:51 PM

If Jeff Renner predicts an 85% chance of rain and it rains 85% of the days he predicts 85% chance: that's good modeling. Let's say he predicts an 85% chance of rain November 6th. Is he wrong if it doesn't rain?

It's an imperfect analogy but it's closer to what's going on here than a lot of what I'm hearing.

Posted by david on November 2, 2012 at 11:49 PM