Don't trust the unweighted exit polls! trust-the.html

Following up on a point I made yesterday -- in a post that just got Instalanched -- as we political junkies giddily await the first hints of what's happening in Pennsylvania today (finally, another primary!! election results tonight!! exit polls!!! colorful maps!!! whee!!!!), it's very important for us, and even moreso the media, to remember that Obama almost always does well in the leaked, unweighted exit polls, and almost always does less well in the final results. For instance:

In New Hampshire, the "first wave" showed Obama leading by 4; he lost by 3.

On Super Tuesday, the "first wave" had Obama winning New Mexico by 6 (he lost by 1) and losing California by just 3 (actual margin: 8). The "second wave" wasn't any more accurate: it had Obama winning Georgia by 50 (actual margin: 35), Illinois by 40 (actual margin: 32), Alabama by 22 (actual margin: 14), Delaware by 14 (actual margin: 9), Connecticut by 7 (actual margin: 4), Arizona by 6 (lost by 8), New Jersey by 5 (lost by 10), Missouri by 4 (actual margin: 1), Massachusetts by 2 (lost by 15), and losing Tennessee by 10 (actual margin: 13), New York by 14 (actual margin: 17), Oklahoma by 30 (actual margin: 23) and Arkansas by 46 (actual margin: 44). So Oklahoma and Arkansas were the only states on Super Tuesday where Obama did better than the leaked, unweighted exit polls suggested.

On March 4, the "second wave" showed Obama winning Vermont by 34 (actual margin: 21), Texas by 2 (lost by 4), Ohio by 2 (lost by 10), and tied in Rhode Island (lost by 18).
Averaging all those numbers together -- and I recognize that this is very unscientific -- you get an average discrepancy of 7 to 8 points. That is to say, Obama generally does 7-8 points worse in the actual results than he did in the leaked, unweighted exit polls.

This is crucially important, because it has the potential to significantly affect the post-primary "spin." That's exactly what happened on Super Tuesday, when Clinton was able to initially spin a "victory" out of her lukewarm performance, largely because the media was expecting Obama to win some "big states" based on those early, favorable numbers. Likewise on March 4, Hillary was able to claim "success" for her Texas and Ohio "firewall," even though she really needed much larger margins to make meaningful delegate progress, in part because the leaked exit polls again conned the punditry into expecting better showings by Obama, possibly including a win in one or both states.

It'll be a travesty and a farce if that happens again. Hillary Clinton needs to win big -- like, double digits big -- and make significant delegate gains in order to claim any kind of a meaningful victory in Pennsylvania tonight. That basic fact will not change one iota if Drudge and The Corner and Talking Points Memo and the Huffington Post (and, er, the Irish Trojan) publish initial, meaningless numbers this afternoon that show a "OMG A DEAD HEAT IN PENNSYLVANIA!!1!" and then Hillary "pulls away" and wins by 6 or 8 points or whatever.

It's important to remember that these leaked exit polls do not actually represent any version of reality; they are not something that a candidate can "come back" from. In those instances where they differ from the actual numbers, they are, and always were, simply wrong. The exit-poll-fueled "seven-hour presidency of John Kerry," for instance, was always an illusion; Kerry was never "ahead," and Bush never mounted a "comeback." That's all pure perception, and has nothing whatsoever to do with reality. Same thing here. If Hillary, yet again, does substantially better than the exit polls suggest, nobody should be surprised, and still less should she get favorable, "expectations"-based spin as a result. Obama's early "lead," in that event, will have been (again) a complete chimera. So please, for heaven's sake, let's not get all excited if history repeats itself again.

As I said yesterday: Fool me once, shame on you. Fool me 21 times, shame on me.

P.S. Politico adds an additional cautionary note, referring to actual returns rather than exit polls:

Don't be fooled by early results. The cities and suburbs usually report their returns first, which gives the candidate favored in those areas a quick - and sometimes fleeting - lead. The conservative-leaning small towns through the center of the state usually filter in much later in the evening.

This tendency has wreaked havoc in past elections: A Democrat goes to bed thinking he or she is the winner, but wakes up the loser. The last time it happened was 2004, when the Associated Press called the state attorney general race for Democrat James Eisenhower and retracted it later in the night as the numbers closed. Some newspapers went to print with the wrong results.

So Obama could show a lead in the early results, but it might be short-lived. If Clinton is ahead at the start, she may never lose it.

Tags: cake, clinton, Common Sense, economic liberalism, exit polls, obama, statistics (all tags)


Advertise Blogads