Tuesday, November 19, 2013

Politico Sux

Proving once again that Politico is the news source for political junkies who don't actually know how our political system actually works, the site is touting a poll that it says shows that Mitt Romney would beat Barack Obama if the election were held today. Actually what the poll really says is that Romney would lead Obama in the popular vote 49 to 45.

Except that American Presidents don't get elected by the popular vote. We got this thing called the electoral college. So when Politico writes that "Obama beat Romney 51 percent to 47 percent a year ago to win a second term" that is not true. Obama beat Romney 332 to 206 in electoral votes. The WaPo-ABC poll Politico cites doesn't give a state-by-state breakdown because pollsters didn't try to do one.

That's the thing about the last Presidential elections that Politico and many other members of the political commentariate don't seem to ever get: Romney never stood a chance. Sure, there were periods that polls of the popular vote showed Mittens ahead. But he never was ahead if you did an electoral vote count using the various state polls. See this post I wrote in September 2012, the height of the horse race season. The lines in those 538 charts I posted never crossed even when Romney was ahead in the popular vote polls. Despite the fact that it was apparent from shortly after the end of the GOP primary that Obama was going to be reelected, Politico and others kept referring to polls that measured the wrong thing to show that the race was "tightening" when it never really got all that tight.

Coverage of this one year after poll is falling into the same trap that made Politico's coverage of the race so sucky last year. On top of that, Politico shows us that it does not understand what "margin of error" means. At the end, it mentions that the poll has a MOE of +/-3.5%. That means that Romney's 49-45 "win" in the popular vote could be a 46-48 loss. The MOE means that Romney's "real" score is somewhere between 45.5 and 52.5. Obama's is somewhere between 41.5 and 48.5. Because the two ranges overlap, the poll alone can't tell us who is really ahead. There just isn't enough resolution in the image drawn by the poll to say who is really ahead.

If we had more than one poll, you could average them out and minimize the MOE in the hopes of making it so they don't overlap. But we don't have that. Note this is not as big of a problem with a state-by-state analysis because so many polls are involved (at least 51) the noise will tend to average out (not perfectly so because different states have different electoral college values, but it also helps that several states that have high EC value  also tend to poll so lopsidedly that the range is often not within the MOE. For example, CA, TX, and NY)

So what I'm saying is Politico sucks. No wonder Nate Silver, gay wizard of poll analysis, hates them so.

(via Memeorandum)