Election Polling, Predictions, and Possible Outcomes

“Politics is an act of faith; you have to show some kind of confidence in the intellectual and moral capacity of the public.” – George McGovern

Last week we took up a brief discussion of polling and predictions that led to some interesting discussion (off the books) with my good friend possiblywrong (who also happens to be a coworker). The discussion was the prospect of winning the popular vote but losing the electoral college, and just how consistent predictions could be that showed both outcomes as likely.

Can a  candidate win the popular vote and lose the election? Of course. Not only is it obviously mathematically valid, it has happened no less than four times in the history of the United States (1824, 1876, 1888, and of course 2000 when George Bush lost the popular vote but narrowly won the election.)

A more pertinent question is “by how much?” … or, “by how much, realistically?” That is, what margin of victory does a candidate need in the popular vote to feel comfortable that he’ll win the electoral college? (Of course, candidates don’t think about it that way – they only think about the electoral college, and rightly so.) The impetus for the question is analysis from Nate Silver that purports to show Obama with a narrow advantage in the popular vote, but a significant advantage in the electoral college. (More on Silver in a moment.)

At a first order, we want to get at just how close recent elections were in both popular vote and electoral college. As a model, we will assume that “undecided” voters chose one candidate or the other by nearly a coin flip, and that voters “on the edge” in any state had about the same likelihood of switching their vote. Thus, if we start dialing votes from the winner to the loser in the general election (as a hypothetical) we would dial them in the same proportion in the states. Thus, a one point shift in the general election can be approximated by a one point shift in every state. (It’s a simple model, so don’t get too hung up on its errors.) Thus, to figure out just how “close” the election was in electoral versus popular vote, we dial in enough vote swaps to switch enough electoral college votes to flip the election, and compute the resultant popular vote margin. For the last 5 elections, here’s what it looks like:

A few notes on the table are in order. First, I compiled the results by hand using electoral results I found online – this isn’t a hard-charging, fact-checking outfit. Second, I actually computed the tipping state using current electoral vote counts per state – which is wrong but not wrong enough to matter. I stopped in 1988 because (i) the kids wanted to play and (ii) adding up the number of electoral votes that Mondale or Carter needed to get the job done was exhausting.

What is interesting here is that the clear “outlier” in the data is the 2008 election, where Obama won the tipping state by a much wider margin than he won the popular vote. (Will his “over-performance” in the electoral college continue in 2012?) Other than that, the margins in the tipping state were all very close to the national margin – less than a percentage point. The implication? That winning the popular vote by at least a point is a pretty good cushion to operate from when it comes to the electoral college. Yes, it’s still possible to lose, but the margins are usually pretty tight.

OK, so why the data pull? Last week we conjectured that Nate Silver was “late to the party” – failing to recognize that the election had turned and that Romney is at least a 50-50 shot to win now. By not means a “lock” – but he’s in the lead. Silver, who is a liberal but also interested in getting it right, not just in his guy (Obama) winning, has been peddling explanations for why the clear momentum shift is meaningless.

First, he claims that “state polls” show Obama is a clear electoral college leader even though the popular vote is slipping away from him (again, “in the polls”). We argue, based on the table above, that the notion of losing the popular vote and winning the electoral college has limits – some fairly tight ones.

Next, faced with a massive deficit in the Gallup poll (Romney 52-45), Silver argued that Gallup is an outlier because it is not in agreement with the average of polls, saying “its results are deeply inconsistent with the results that other polling firms are showing in the presidential race, and the Gallup poll has a history of performing very poorly when that is the case.” Umm, Nate, every poll has a history of performing poorly when it is many sigma outside of the average. That doesn’t mean that the poll is wrong, it just means that it is an outlier. Is it an outlier because it has a better model or because it has a worse model? I don’t know, but obviously it likely that a poll on the fringe is, in fact, on the fringe.

Now, if Mr. Silver wants to discard outlier polls, that’s fine. It’s his model. If he wants to ignore polls altogether, that is also fine. But spinning to find justifications for why bad polls don’t count but good polls do is a bit absurd. As with all polls and prognosticators though, we don’t get to call them on it until election day. They can always “adjust” the results just before the end to get closer to right (if, in fact, they’re fudging the numbers to spin a narrative).

As for me, I see an election that is moving to Romney. I’m not exactly a fan, but I’d be hard pressed to think he’s worse for American than Obama … in the short term anyway. I’m still looking for the “October Surprise” that can flip the momentum back to the president, and it may still be out there, but it’s getting late in the game. If things continue on their current trajectory, Silver and the “dead heat” pollsters will be changing their tune in about  a week.

This entry was posted in Uncategorized. Bookmark the permalink.

One Response to Election Polling, Predictions, and Possible Outcomes

  1. “Now, if Mr. Silver wants to discard outlier polls, that’s fine. It’s his model. If he wants to ignore polls altogether, that is also fine. But spinning to find justifications for why bad polls don’t count but good polls do is a bit absurd.”

    As I have joked before, I confess that I usually find politics about as mathematically interesting as economics or meteorology, all for the same reason– because modeling their behavior is messy. So it took some reading to get a better feel for the current discrepancies between the various national and state polls, and the various election forecasts.

    And I think I have to come to Silver’s defense here… sort of. First, I don’t see any indication that he *manually discarded* Gallup as an outlier poll. That is, his *algorithm* de-weights Gallup (or at least, it currently does)… but– and this is extremely important– that same algorithm has been operating this entire time, without subjective late-October modification, and would presumably de-weight Gallup for “outlier behavior” no matter which candidate was favored. In other words, he wrote the code months/years ago, and has continued to pour the same polling data sources into it, including Gallup. The “spinning to find justifications” reads to me more like an attempt at explaining the seemingly strange output of his algorithm. (Indeed, his post right around that time linked here would seem to directly contradict any accusation of “statistical corruption.”)

    Or at least, I hope so. Unfortunately, although Silver has described his methodology, algorithms, etc., in quite a bit of *prose* detail, he has not actually made the source code itself public, which would allow us to verify whether anything that I wrote above is actually true. I can’t say I blame him, since it’s a moneymaker, but it certainly detracts from the scientific integrity that can possibly be attached to his predictions.

    So, the best we can do (or the best that I care to do 🙂 ) is remember this point in time, and these current predictions, on November 7th, and see how accurate those predictions really were.

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 )

Google+ photo

You are commenting using your Google+ 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 )


Connecting to %s