Daily question: Who needs the polls anyway?

The American presidential race lives and dies by the polls. Tens of millions of dollars are spent to minutely track every imaginable category of voter tendency. Resulting “predictions” are used to not only inform us of the direction of the horse race but to nudge us to vote one way or the other. After the election is over, one of the very first things we do collectively is to look back and tally whose predictions were off by how much. This is a time honored American sport.

But why bother with all these costly and often inaccurate polls when, apparently, there are far more accurate and cost-free predictors:

  • If the Washington Redskins win their last home game before the election, the incumbent party wins. This has been correct 100% in the last 17 elections, until the last one in 2004.
  • Every time the Los Angeles Lakers play in the NBA finals in an election year (seven times since 1952) the Republican party wins. This year the Lakers were in the finals.

There are of course many other such predictors, from sales of party-color coffee cups at 7-Eleven to Halloween masks at BuyCostumes.com:

presidential2.jpg

If you were modeling a financial or economic system whose outcome depended highly on the outcome of the presidential race, would you include one of these predictors as a strong variable?

9 thoughts on “Daily question: Who needs the polls anyway?

  1. The Redskins lost last night, suggesting the incumbent party will lose.

    We’ll find out tonight if 2004 was just a blip on an otherwise maddeningly good correlation, or whether the Mavericks are really not part of the incumbent team.

  2. Thanks for asking, Kontra.

    I do not model in any factor until I understand it well enough to weight it. Otherwise my model is indefensible, and subject to garbage in / garbage out.

    So I do not promote or demote a factor based on whether it appears silly or voodoo. What might look bloody silly to one person (an Akron rainstorm) might actually have solid causative predictors.

    Computers being the wonders that they are, I can always re-run my model with factors which I do not understand yet (like the Redskins record on the game before the election), using a range of weights to see whether it is worth spending the time to understand them further.

    For me, it all comes down to the pragmatic level of how much resource (client’s interest, time and money) is available to narrow the unknowns and improve the predictions.

  3. Ziad: “In your blog, you asked whether we would include such “predictors” as a STRONG variable. In your comments, you ask twice whether we would IGNORE them.”

    Two sides of the same coin. If a “predictor” has a very high degree of (or even perfect) correlation, higher than common polls and surveys, can you afford to ignore them just because it looks bloody silly? But if you do include them (remember, they are highly correlated) do you then assign a weak weight (because they look silly) or treat them as you’d any highly correlative factor?

    It’s one thing to say, OK I’ll just ignore any natural non-causative factor because that’s just voodoo. But if you do model it in, do you downgrade its weight simply because it looks like voodoo even though its correlation is much higher than other conventional factors?

  4. Hi Kontra. You’ve changed the question.

    In your blog, you asked whether we would include such “predictors” as a STRONG variable. In your comments, you ask twice whether we would IGNORE them.

    I would look into whether any such correlation was causative, and for other factors that would explain them. I would not even elevate them to the status of “predictor” until then. Just something interesting to study, like whether the Akron rain affected certain voters like farmers more than others in that swing state.

    Venn diagrams are twirling in my head, about proportion of voters who are Redskins fans, and how a recent win or loss affects their voting behaviour. Also whether any believe it is a self-fulfilling prophecy.

    I don’t know many clients who would pay me to study these phenomena, so you can reach me through Kontra. I have an excellent statistical background ;-)

  5. Jared, yes I’ve been following Intrade, but as you know, there have been attempts to game it in favor of the Republican candidate. The curious aspect of this is that some “predictors” may be open to manipulation, others not. For example, what if some natural phenomenon was observed to have high correlation? While even large scale behavior such as Halloween mask purchases may be gamed, simply observable facts can’t be. What if, say, the last Sunday before the election it rained in Akron, OH 99% of the time the Republicans won? Would you ignore it?

  6. Timm and Ziad, thanks for the interesting links.

    I’d like to go back the issue of modeling in my question though. As frightening as the prospect of elections-without-elections may be, what exactly do you do if in fact there’s an indicator/predictor that has a very high degree of correlation? In the two cases I cited, nearly perfect correlation!

    Obviously we could easily dismiss these as superstitious, lucky, non-causal, etc., and yet we haven’t dispensed with routine political polls either, which happen to have lower predictive powers in certain areas. It does sound silly. But in financial and economic models, one makes a lot of real-world assumptions (without which it’d be impossible to model real-life phenomenon) that are not an order of magnitude more reliable than these, as the recent credit crisis has shown.

    Sure, one can easily assign lower weights to these but can you ignore them completely? Beyond being silly, it’s a bit of an existential challenge. Halloween, after all, is just around the corner. :-)

  7. I think Halloween costumes might actually be a pretty good predictor of voter sentiment, and consumer confidence. You might even discern the topic du jour, any hidden tendency to vote against a black man, from any caricatures of Obama that overdo African features, or portray him as Muslim, and the reactions of fellow partygoers to those caricatures.

    Kontra, if you could extend the hyperlink to also cover the (short_story), you would disambiguate Timm’s reference, as tortured as that sounds.

    Another nice link on Franchise:

    http://www.associatedcontent.com/pop_print.shtml?content_type=article&content_type_id=911017

Comments are closed.