Wednesday, September 19, 2012

Ignore The Electoral College! Follow-Up

I've been saying for months that it's best to completely ignore the electoral college until at least after the conventions. Well, unless you're making decisions about where to put resources if you're a presidential campaign; then you don't have much choice.

A little confirmation that it's good advice comes from Nate Silver's latest big look at the electoral college bias. He's been reporting for months that the electoral college had a slight bias for Barack Obama -- that is, if the election was a 50/50 tie nationally, the odds favored Obama to win the election because his votes were likely to be better distributed across the states. Except's reversed.

I agree completely with Silver that whatever you think of his forecast model, this is exactly the stuff that he's going to get right. So I'm inclined to accept that if he says that's where the numbers fall, based on the polls he collected as of this post (yesterday morning), that's where they fall.

So what value added was there from paying attention to, say, how Florida looked back in May? I just can't see it. You were way better off getting a good estimate of how the election looked nationally (which at that point would have put very little weight even on national head-to-head polling), and then assume that you'll get more-or-less uniform swing. And that if it projected as very close, then the best thing you could say was "very close" -- you couldn't actually get any more accurate by looking at individual states.

What I'm not sure is whether there's any utility even at this point in looking at the individual state polling -- that is, if what you're really interested in is who is going to win the election. (Clarification: it's certainly worth it to look at individual state polling as part of figuring out the national situation; if we see Ohio and Florida and North Dakota and Utah all move two points to Obama, that tells us something about what's happening nationally, and the polling aggregators/modelers do and should use that information). I had been saying that now was about the time to start paying attention to the states, but I really don't know if that's true. After all, think about it this way: Silver had Wisconsin surprisingly close, and therefore it had moved up to the 4th most likely tipping-point state...but there's a new poll out today that puts Obama way up there, and I'm guessing that will be enough to move Wisconsin back to a uniform-swing state -- and far from the top of the tipping point state list.

Look, eventually, we know that we won't get 100% perfect uniform swing, and presumably a fair amount of whatever non-uniform swing will happen will be evident as soon as there's enough state-level polling information. Right? It's not as if we expect Ohio (or New Hampshire or Virginia or whatever) to suddenly veer off in the last few weeks, for the most part; we expect unusual swing to be the product of longer-term stuff than that. So presumably if there were enough state-level polls, we could pick up on some (most?) of it early on. It's just that it's not going to matter unless the race is very close, and if it is we still won't quite know enough to know about electoral college bias until, most likely, a bit farther down the line.


  1. It's beginning to sound like the only trustworthy polls will be those coming one or two weeks after the election.

    1. Well, I could make a crack about the (political science) NES survey...but instead I'll just say that I do think that the national polling right now is meaningful.

  2. Looking at where the states stand relative to each other is still interesting. For example it looks like Obama has more or less locked up Ohio, but Romney still has a chance in Iowa. Minnesota used to be a swing state and now isn't really, but Wisconsin still is, despite the fact that they're very similar states demographically and politically.

  3. Methinks you're being too cautious.

    Once we accept that polling is inherently a bit of a guessing game (particularly with likely voter models differing as they do), then I'm not sure we need additional caution. I've been noticing VA being bluer than I really expected, and IA and WI being more purple than I expected. It's been pretty consistent. And it's odd that CO and NV aren't bluer, given how they'd been trending over the longer term.

    A cautious analyst should always have been interpreting the national in light of the local and the local in light of the national.

    Vis a vis Silver's measure of bias: it's one way to do it. It captures bias, if we define bias only as something that can cost an election. However, I'm perfectly comfortable saying that the EC could be biased in favor of the Rs (or Ds) and that party STILL lost, or it won by more than the bias, so it won "fair and square". And Silver's measure isn't talking about that...which is perfectly fine, it's just a narrower definition of "bias" than I have.

    1. However, if you want to ignore a poll, I'd go with today's Obama +14 in Wisconsin.

      There are over 25 polls today on RCP...ONE of them was likely to be outside the MoE!

  4. First of all, it is fun.

    Second, I live in San Antonio like you, and it is quite possible to live here almost completely oblivious to the fact that there is a presidential race going on. The only times I think there was anything about Romney or Obama before the conventions was during the Republican Senate primary as they all trashed Obama. The conventions got more play, especially the DNC since Mayor Castro was a keynote speaker.

    The reason I bring this up is that the election isn't really happening here. So the fact that Rove et al could dump a bazillion dollars on Ohio and Florida and not move the needle was very interesting to me. In November 2007, what was interesting was not where Clinton, Edwards, and Obama shook out in national democratic polls, but in Iowa and New Hampshire polls because that is where the election was happening.

    What is happening where both teams are fighting is intersting in that it is a preview of what could happen when the campaign becomes a large focus nationally.

  5. Well of COURSE you pay attention to the electoral college, silly.

    Every campaign with designs on actually winning (this would exclude anything involving Shrum) will begin, conduct itself and close with an eye firmly affixed on the electoral college.

    That doesn't mean it's the only data set you're watching, but it may be the only one you're ALWAYS watching. I would tend to ignore national polling during select "dead spots", and these might vary based upon the particular cycle you're in that year. During those "dead spots", I'd rarely make critical campaign decisions based upon national data recorded. But I'd make decisions based upon my electoral college strategy, at any point during the campaign, if I felt they were called for.

    Always remember the supposed "architect" Rove screwed up this fundamental tenet, and it almost cost him. He instructed an underling to ignore West Virginia in 2000, under penalty of termination. The guy ignored Rove and plunged resources into WV... and guess what... those WV electoral college votes put W into the White House.

    It's the electoral college, kids. First, last and always. As desired, you can assign probability and confidence levels onto the discrete state polling data acquired, and let that drive your electoral college strategy, but it's that strategy that must direct the campaign.

    And Rove is an idiot.

  6. With my limited knowledge of how campaigns work, I'm fairly sure that they take a lot of factors into consideration about swing states, because it is the electoral votes that elect the president. They poll a lot themselves in crucial states, and watch other polls as much for their effect on the media and voters than for their content. Polls can confirm and suggest strategies. Though I don't think it took a poll to figure out that talking a lot about saving the auto industry would play well in Ohio.

    The truth is that national and state (electoral college) level campaigning and polling affect each other. When they start to tell the same story you might have more confidence in a trajectory. But 2000 and 2004 showed how crucial individual states and voting districts within them can be. Polls are one tool for strategy. But only one.

  7. In a different issue that Silver has been exploring, it appears that there are (possibly!) two distinct universes of this election, which on the one hand has a very close margin based on robocall polls such as Rasmussen and Gallup..

    .. and on the other hand shows an election which Obama has a 4-6 point lead. This comes from pollsters who do live interviews.

    The difference is posited to be due to laws saying robocalls cannot be made to cell phones, so it's only the latter live-call interview polls that are capturing data from what is said to be 1/3 of households who no longer have a landline phone. (This number sounds high to me, but then I'm a fogey who has never owned a cell phone.)

    So under this analysis, the question becomes is there a difference between the American households who have cut the telephone cords, and those landline phone households that Gallup and Rasmussen use as their entire polling samples. I'm guessing the cell-only users skew young. Could be a consideration.


Note: Only a member of this blog may post a comment.

Who links to my website?