Friday, November 2, 2012

Election Day and Numbers

It's almost election day, so I feel obligated to write something political.

I'm a numbers guy -- always have been, probably always will be.  When politicians propose policy, I'm one of the people who demand that they show numbers that make at least some sense.  And when I want to know whether my candidates are doing well, I look at numbers: fund raising and polls in particular.  I'm also a pseudo-academic [1], so have been pleased that some real academics look at ways to combine multiple polls to give more accurate results.  The chart to the left is an example from the fivethirtyeight web site from earlier this month that shows estimated probabilities for Obama or Romney winning in the electoral college and in the popular vote.

Nate Silver and Sam Wang, have been under attack lately.  Silver and Wang are only two of the better known aggregators.  There are also sites like Votamatic and Real Clear Politics' No Toss-Up States.  All four of those show Obama with a high probability of winning the electoral college vote; not surprisingly, much of the criticism comes from Romney supporters.  One of the common complaints leveled at Nate Silver in particular is that he doesn't weight all polls evenly.  These attacks seem particularly partisan since, a site that is openly partisan in favor of Romney, and that mangles the reported polling data to show that Romney will apparently win in a landslide, is not given the same treatment [2].

Another form of attack comes from the media.  The Washington Post's Chris Cillizza, whose Fix column rates various races, recently moved Ohio from "lean Obama" to "tossup".  Cillizza's reason for this?  "….the absolute necessity for Romney to win the state if he wants to be president - leads us to move it back to the 'tossup' category."  Not that the numbers have changed, but that Ohio has become the "must win" state for Romney, therefore it becomes a tossup.  Maybe Chris is right.  OTOH, I'm more inclined to the theory that Chris' job #1 is to sell newspapers and pull eyeballs to the Washington Post's web site.  That's a lot easier to do if it looks like a horse race.

And finally, there are attacks based on the hypothesis that polls can be wildly wrong because they don't reflect the secret behind-the-scenes things that only a political insider would know.  Certainly polls can be wrong.  They can be wrong even beyond the margin-of-error numbers that are always included in the press releases [3].  Polls were one of the reasons that the Chicago Tribune printed its "Dewey Defeats Truman" headline.  But the statisticians that design the polls continue to learn their craft and improve their skill.  For example, one hears that cell phones have made polling less inaccurate.  Yep, and you can bet that the poll designers were at the forefront of saying that, and then designing methods to account for the effect.

People like Nate Silver and Sam Wang, even though they may personally prefer an Obama victory, depend for their livelihoods on being accurate and unbiased.  Upton Sinclair famously said, "It is difficult to get a man to understand something, when his salary depends upon his not understanding it!"  It is difficult to get an academic statistician to bias his/her results when their reputation and future salary depend on being unbiased.  I've hung out with academics and former academics most of my life.  And based on that experience, I'm a firm believer that if you can bring real numbers that show Silver and Wang have flaws in their models, they'll be the first to admit it.  So far, the attackers aren't bringing numbers.

[1]  The way I define these things, real academics get PhDs and work for universities.  They teach, do research, speak at conferences, and publish in refereed journals.  I'm only a pseudo-academic.  I stopped at multiple Masters degrees and my research activities were within the confines of the old Bell System and its various derivative parts following the 1984 break-up.  I have occasionally spoken at conferences and did publish a paper in the refereed IEEE Spectrum, but the conferences were, and Spectrum is, aimed at practicing engineers as well as academics.

[2]  Nate attempts to weight polls based on several factors, including historical accuracy.  At least IMO, the manipulations done by and some others lack the same sort of statistical justification that Nate considers.

[3]  Other factors: the Tribune's political insiders also predicted a Dewey win, and working around a year-long printers union strike forced the Tribune to go to press before there were any actual results available.

1 comment:

  1. A follow-up comment: the people with the numbers were right; the people with the gut feelings were badly wrong. Lots of people should be asking themselves, "Why would I ever hire Rasmussen to do polling for me again?" And, "What reality does George Will live in?"