« J-Pod Nails It | Main | The Morning After »

November 08, 2006

Testing The Exits

Given that Democrats had a very good night last night, it's worth taking another look at the exit polls to see how they fared, this time without the pressure of a vast and demoralizing reversal from earlier predictions.  Now, it's worth remembering that these polls are taken after voting, so the effects of suppression and intimidation (the garden variety methods used to tilt the table on election day) won't show up.  Further, this exit data came out around 5pm Eastern, so polls hadn't closed.  But just for kicks, I've matched the exit data with the final vote tallies to see how the polls performed:

State Exit Poll Voting Results Difference
Missouri +2% +3% 1%
Montana +7% +1% 6%
New Jersey +8% +8% 0%
Ohio +14% +12% 2%
Pennsylvania +15% +18% +3%
Rhode Island +7% +6% 1%
Virginia +7% +0.7% 6.3%
Tennessee -3% -3% 0%

Bottom line?  The polls, early as they were, did staggeringly well.  Their largest misreads in Virginia and Montana, but the average margin of error was an astonishingly low 2.4%.  Whatever problems bedeviled the polls in 2004 -- cue the usual argument pitting error against theft -- was fixed for 2006.

November 8, 2006 in Polls | Permalink


I guess I don't care so much about whether the exit polls tell us accurately who's going to win a seat (time will tell....), but knowing that the exits reflect closely the actual vote is significant when the real value of the exits is considered:

WHY did people vote as they did, and

WHAT segmentation factors may have been at work - demographics.

That said, the exit polls still aren't accurate enough for calling closely divided elections.

Posted by: JimPortlandOR | Nov 8, 2006 9:47:47 AM

So, all of those fixed voting machines somehow did not fix the election and that roumor was as valid as the weather manipulation and locusts. Amazing.

JimPortlandOR, the folks at MSNBC have been going on for almost 18 hours now about how my fellow voters are a bunch of racists for not electing Harold Ford, Jr. Might want to get some insight on how psychics assign motovations to people they do not know on that channel.

Oddly, they don't say that Maryland failed to elect Steel due to their basic racist leanings.

Sorry, that was just a lot of words to say you are never going to find the answer that you seek.

Oh, (others I am guessing) please skip bringing up that ad that Corker wanted pulled about the Playboy party? Ford got a bounce off of that one.

Every one of these folks who lost their seat in the Congress as a whole did it by whatever came out of their own mouth. Plenty of help from the press there, but it was still their own words that defeated them.

Posted by: Guy Montag | Nov 8, 2006 9:58:06 AM

HHmm..yeah thanks for crushing the last remaining hopes about 2004 election validity

Posted by: BigDave | Nov 8, 2006 10:03:17 AM

I guess I don't see how the numbers lead you to be so optimistic. The numbers don't look good for Virginia, Montana and Ohio. Or, do I misunderstand?

Posted by: Bruce Wilder | Nov 8, 2006 10:10:56 AM

Those Montana and Virginia exit polls were terrible. They both reported a 7 pt spread (actually, I remember the VA number being a 5 pt spread), and yet both elections were well within 1%. You can't expect an exit poll to accurately predict a winner in a sub-1% election--when it's that close, all you can realistically expect from an exit poll is that it'll tell you "this one's gonna be close." The MO and TN polls were good: they said one candidate would win by a pretty slim margin, and in both cases, they got it right. But neither of the MT or VA polls did that.

Posted by: Haggai | Nov 8, 2006 11:00:28 AM

OK, I am reading this data a lot differently then you...

Am I the only one concerned that the two closest races in the country, which will decide the Senate, are the only ones showing official results varying significantly from the exit polling?

Posted by: greenvtster | Nov 8, 2006 11:11:10 AM

VA was close, but MT wasn't supposed to be so near -- Missouri's numbers predicted a much closer race. These are, now, the two that will decide control. But that's not necessarily something that was predicted.

Posted by: Ezra | Nov 8, 2006 11:39:58 AM

What greenvtster said. What exactly would you expect to see in a non-problem election?

Posted by: tps12 | Nov 8, 2006 11:41:48 AM

Guy find any ads or flyers Cardin sent out about Steele that even remotely resembled the ads about Ford then we can talk about tapping into implicit racist attitudes in Maryland.

Posted by: ellenbrenna | Nov 8, 2006 11:53:15 AM

I'm with greenvtster.

What these numbers show is that Republicans wouldn't even be close in the close elections if they weren't able to...take advantage of...the extraordinary spread between how the voters say they voted, and how the Diebold machines say they voted.

Cause in the places where it was a clear Democratic win, the machines say pretty much exactly what the voters say.

Posted by: kid bitzer | Nov 8, 2006 12:26:59 PM

I'll start taking the voting machine paranoia more seriously when there is one shred of real evidence. If the Diebold masterminds had been on the ball they would have gone after Missouri and would have carried Montana and Virginia, right?

The machine I voted on had a paper ballot that I checked before I was done. What kinds of machines are in the "almost stolen" states?

Posted by: Sanpete | Nov 8, 2006 12:37:29 PM

you know, most smart conspiracy theorists posit conspirators who are smart, too.

Stealing millions more votes than is necessary is dumb.

Accordingly, so is this:

"If the Diebold masterminds had been on the ball they would have gone after Missouri and would have carried Montana and Virginia, right?"

If there was any fabrication of the totals, it would have been part of an incrementalist strategy to chip off a few thousand here through voter intimidation, a few thousand here through sending people to the wrong polling-places, and so on.

That way you keep the straight-forward miscounting of votes within a small enough margin that it doesn't stand out from the background noise.

That also means that it may be hard to calibrate how to steal a number that will be big enough to win, but not big enough to stand out.

How could you determine that it happened, then?

Well, how about if there was a major discrepancy from the exit polls, which otherwise functioned very accurately?

Posted by: kid bitzer | Nov 8, 2006 12:46:34 PM

Wait, I said that wrong: I meant, what would you expect to see in an election that was a "problem," if not this?

Also note that these figures are percentage points, not percentages. I.e., in MT the percentage difference between exit polls and ballots was (56-49)/49 = 14%.

Posted by: tps12 | Nov 8, 2006 1:29:41 PM

And cosign on everything kid bitzer said. A stolen election would not be a 10-point Santorum victory after a surprise GOP turnout in Philly.

Posted by: tps12 | Nov 8, 2006 1:31:12 PM


I suppose you are talking about the ad that gave Ford a boost? The one that only a racist could find racist?

Fliers against Steele? No, but there was plenty of this. Unless you think "Uncle Tom" is a compliment.

Posted by: Guy Montag | Nov 8, 2006 1:37:29 PM

How could you determine that it happened, then?

Some one of the many dozen people involved might talk, or there would be some trace of tampering, or whatever. What have you really got? Were there paper ballots? How was this monkeying done? Why didn't your "smart but not that smart" conspirators go after Missouri, which everyone knew would be tight? I'll take it seriously when there's some real evidence.

Posted by: Sanpete | Nov 8, 2006 1:42:00 PM

I voted in St. Louis on a machine that had a paper "receipt" that you could check. The Secretary of State/Board of Elections director in Missouri is a Democrat – Robin Carnahan – who I'm assuming was monitoring the machines closely. I'm guessing it would have been more difficult to tamper with the machines here.

Posted by: Jumada | Nov 8, 2006 2:03:01 PM

What greenvtster said. In 2004 the exit polls were much more "off" in battleground states, and always in the same direction. And I think Jumada has shown us how Missouri avoids "inaccurate" exit polls.

Posted by: calling all toasters | Nov 8, 2006 5:26:11 PM

tps12 made the good point here. While the methodology for Ezra's study seems fine, the interpretation stumbles because of how Ezra expresses a percentage as "percent." Better to express these numbers the way Wall Street does -- as "basis points" to avoid confusion.

The Virginia results are the most obvious. The difference is a full order of magnitude, but the table expresses the difference as 6.3 percent. Better to say the difference is 630 basis points.

Posted by: Rick | Nov 8, 2006 5:31:04 PM

Paranoia is so easy. Did Montana and Virginia have paper receipts or not? So far you just assume they didn't. You believe first, look for facts later.

Posted by: Sanpete | Nov 8, 2006 5:48:36 PM

Sanpete, I'm not paranoid, I don't believe anything at this point, I'm just questioning this. I don't have access to the answers to the valid questions you asked, but I hoped by asking that someone would find out exactly those kinds of things. Not paranoid (yet), just curious and maybe suspicious...

Posted by: greenvtster | Nov 8, 2006 6:05:33 PM

FWIW, I was doing persuasion calls in Repub territory in MT on election eve and quite a few refused to tell me who they were voting for so I could persuade them. I suppose the exit pollers encounters some tight lipped Burns supporters too.

Posted by: hit_escape | Nov 8, 2006 6:13:31 PM

i'm with greenvtster too. AFAIK, VA doesn't have paper ballots, just electronic. not sure abotu montana.

Posted by: dan | Nov 8, 2006 6:21:35 PM

Fair enough, greenvtster (your name would make more sense if you spelled it out, you know). I'm not suspicious about this, but I was curious enough to check: Montana, like Missouri, has a voter-verified paper trail; Virginia doesn't.

I'm surprised exit polls are as accurate as they are. I've heard of conservative radio talk show hosts telling people to lie to the exit pollsters, and some callers claiming to do that. And, as hit escape suggests, there may be selection bias if conservatives are just less likely to cooperate.

Posted by: Sanpete | Nov 8, 2006 8:01:53 PM

And some Virginians may have been ashamed to admit voting for Macaca Man.

Posted by: Sanpete | Nov 8, 2006 8:04:13 PM

The comments to this entry are closed.