Over-Fitting Data and Anecdotes

“Government exists to protect us from each other. Where government has gone beyond its limits is in deciding to protect us from ourselves.” – Ronald Reagan

I recently caught a couple of mildly predictive articles (and conversations) about the upcoming election and thought I’d pass along. In case I forget to mention it, the problem in all of these articles and analyses (including mine) is that they “over-fit” the data and claim a convincing, undeniable predictions about the Obama-Romney outcome.

First, the anecdote. A friend wrote the other day and noted that if you look all the way back to ’80 (possibly beyond, but that’s the first election I remember), this country has always elected the more “likeable” candidate. I think this is probably true. Reagan-Carter, Reagan-Mondale, Bush-Dukakis, Clinton-Bush, Clinton-Dole, Bush-Gore, Bush-Kerry, Obama-McCain … policy preferences aside, the winner in each case probably was the more likable guy. (This is similar to the Karl Rove “have a beer” metric – that Americans vote for the candidate they’d most like to have a beer with.)

Now, the notion has some truth to it, but it possibly only picks up the game after the campaign has done it’s work. That is, it’s quite possible that the candidate who ran the most effective campaign, and made the most effective policy arguments, always appeared more likeable. That said, obviously candidates need to keep themselves open, happy, smiling, and confident to attract would-be voters.

A more interesting and completely unrelated anecdote is the Washington Redskins predictor. The story goes like this: if the Skins win their last home game before the election, the incumbent party keeps the White House – if they lose, the incumbent party loses. The streak is actually pretty impressive. Going all the way back to Roosevelt-Landon back in 1936, the Skins predictor has only missed once – in 2004. Other than that, the Skins strongly represent the side of the incumbent. On the one hand this portends a bad election for Obama – the Redskins are generally awful. On the other hand, it is obviously coincidence.

Now on to some actual polling analysis. Nate Silver put out a piece a few weeks back discussing spring time polling and what it means for predicting results. At the time, Obama led the average of polls by 47.7-43.7. Pulling the results generally shows that there’s not a huge correlation between the raw April numbers and the raw outcomes. I’ve tinkered with the numbers a bit and I can’t quite pull an extremely clear result out. You can sort-of make an argument that the incumbent party had better have a high percentage of the relative poll share between the two, as undecideds break toward the challenger. The problems of course are that George H. W. Bush had a huge lead in ’92 (by this metric) and lost, while George W. Bush was even in April and still managed to win.

Speaking of the George W. Bush victory, Dick Morris points to it as an outlier as he claims that Romney should win in a walk. Now, Morris is a book-seller and has been in this game long enough to know how to peddle his wares. One must recall that in 2008 Morris was selling the story days before the election that McCain had a shot (even though post-TARP and well before the general election he had conceded that McCain was pretty much a goner). His argument has rationale to it – namely that undecideds to break toward the challenger (they’ve had plenty of time to make up their mind about the incumbent). He notes that of all the incumbent elections back to Johnson in ’64, the 2004 election is the only time that an incumbent has broken out from his spring polling numbers. A cursory glance at Silver’s numbers says this may be close to true. Carter went from 38 to 41 in ’80, Reagan went from 54 to 59 in ’84, Bush went from 38 to 38 in ’92, Clinton went from 48 to 49 in ’96, and Bush went from 45 to 51 in ’04. Not much shift in vote count for the incumbent.

The problem with Morris’s analysis is that he cherry-picks the latest and most damning poll for Obama – a Rasmussen tracking number that puts Obama behind by 48-43. Guys who are selling “hope for a change [and buy my book]” will do this sort of thing.

My prediction: there will be a lot more predictions. My advice for the campaigns? For Obama: stop talking about George W. Bush and start talking about Mitt Romney. For Romney: don’t say a word to or about Barack Obama. Just stand there and be an alternative. It’s about the best a non-interesting, non-inspiring guy like yourself can do, and it might just work. Smile, say nice things about the American people and the things we can still accomplish if we just put our minds to it, and don’t go disparaging the president. There will be plenty of folks who’ll do that for you.

Even if Morris’s premise holds true, Romney still faces a major uphill battle – the same one that George H. W. Bush faced. Nobody knows for sure if we’ll have a third party candidate. Perhaps he should ask Ron Paul (or his son Rand) to be his running mate.


This entry was posted in Uncategorized. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s