« A Confession | Main | A Brief Note »

That Is Interesting

Ruy Teixeira, in the midst of debunking some of the "we wuz robbed" sentiment out there, reveals something very interesting -- unweighted exit poll numbers always overstate the Democratic Party vote share. Why would that be? In light of the fact that their data is now bound to leak out, I would say the folks who conduct the exit polls have an obligation to let slip the details of the black arts so as to avoid this sort of confusion in the future. There may be a perfectly good reason for systematically overstating the Democratic vote and then re-weighting later, but it's not obvious what that reason would be. So let's hear it!

Meanwhile Charles Kuffner takes issue with my debunking of those who claim to have proven that Bush didn't get 59 percent of the Hispanic vote in Texas, arguing that Latino turnout couldn't possibly have gotten as high as 26 percent of the electorate. Perhaps not. I'll have to look into that further at a later date.

November 13, 2004 | Permalink

TrackBack

TrackBack URL for this entry:
https://www.typepad.com/services/trackback/6a00d8345160fd69e200d83421e7fd53ef

Listed below are links to weblogs that reference That Is Interesting:

» Latinos for Kerry after all from chez Nadezhda
Well this should bring a sigh of relief to Matthew Yglesias, who's been fretting ever since Nov 3 about the [Read More]

Tracked on Dec 4, 2004 6:07:38 PM

Comments

I wonder if ballot spoilage, which happens disproportionately in black precincts, has a significant explanatory role here. Maybe a lot of our people think they successfully voted, but it turns out that they overvoted or spoiled their ballot in some other way. They still tell the exit pollster that they voted for Kerry, oblivious to the fact that their votes didn't count.

There are supposed to be 93,000 spoiled ballots in Ohio -- not enough to give Kerry the lead, but enough that spoiled+provisionals would put the outcome seriously in doubt.

Posted by: Ethical Werewolf | Nov 13, 2004 3:02:55 AM

Well, if I had voted for Bush, I would have been too ashamed to tell exit pollsters about it too!

Well, actually, I voted absentee, so I never told exit pollsters anything. Could more absentee ballots be Republican, leading them to talk less to exit pollsters?

Posted by: Julian Elson | Nov 13, 2004 3:20:14 AM

Thanks for your honesty Tex. This is a right-wing talking point on a silver platter. Not knowing the actual weighting, they'll forget about that.

Posted by: es | Nov 13, 2004 5:06:31 AM

Here's a guess:

It is much, much easier to exit poll in urban precincts (they are closer together, more people etc) than suburban which, in turn are easier than more rural districts (e.g. in my area, which is a mix of rural to suburban density we have on the order of two to three dozen polling places for a total population of less than 100K).

As urban areas tend to be Dem strongholds, we have raw data that disproportionately samples dems.

Posted by: Scott Pauls | Nov 13, 2004 8:18:28 AM

I think Scott Pauls has it about right, and I'd also venture that the work patterns of various segments of the population might have something to do with it: Democrats, with a larger share of their base being female, non-white and working class, are more easily able to vote earlier in the day, at time pollsters tend to find most convenient.

Posted by: Abiola Lapite | Nov 13, 2004 8:37:38 AM

Well, if Scott is right about that, they've got bigger problems, because there's no particular reason to suppose that urban voters are representative of rural voters, even AFTER you normalize the results for every variable you can think of. Urban vs rural is a significant variable in it's own right.

Though I know they do SOME rural exit polling, as I got exit polled back in '96.

Posted by: Brett Bellmore | Nov 13, 2004 9:27:39 AM

Unweighted or not, the simple solution to the question is that oftentimes people of conservative views are hesitant to admit they vote in such a manner. Other less vocal conservatives see their vote as private, and nobodies business but their own. They might discuss it with thier spouses or closest friends, but they do not allow their privacy to be invaded by exist pollers.
Liberals on the other hand tend to be more outward with their opinions on the whole. This isn't true of everyone as I chose not to be questioned either at my polling station.
But people claiming to be conservatives have appeared on other boards claiming that while liberals were lining up to have their votes revealed, many, if not a sizeable number of people in wholly conservative enclaves chose not to be counted.

So the only secrets people should be told about exit polling is how many people refused to reveal their choices.

MYOB'
.

Posted by: MYOB | Nov 13, 2004 10:19:55 AM

Such a dishonest post by Ruy.

Ruy says he doesn't know how this year's exit data was weighted. So instead he cites unweighted exit polling data from past elections to claim that exit polling data from this year's election was wrongly weighted.

Hardly 'debunking' if you ask me. Especially when he fails to mention why exit polling in places like Illinois exactly matched the vote result, while the exit polling in swing states consistently showed an average of +5% for Kerry over the vote tallies.

Why did exit pollsters weight correctly in Illinois but mysteriously not in swing states?

Ruy doesn't bother to say.

Posted by: Night Owl | Nov 13, 2004 11:52:21 AM

The point is not whether vote fraud, vote theft, and supression of the vote played enough of a role to change election results; it's whether these activities played any role at all. If we promise the voters that every vote counts, we need to stand behind that promise.

Posted by: janeboatler | Nov 13, 2004 11:57:22 AM

Abiola Lapite: "I think Scott Pauls has it about right, and I'd also venture that the work patterns of various segments of the population might have something to do with it: Democrats, with a larger share of their base being female, non-white and working class, are more easily able to vote earlier in the day, at time pollsters tend to find most convenient."

Wow. Why would non-whites be more easily able to vote earlier in the day, Abiola? Let alone working class people. And the idea that pollsters would find earlier times to be more convenient is bit strange.

There is a question of how data might come in differently during the day, of course (eg., 4-8PM exit poll results not being available for 4PM broadcass).

Posted by: Barry | Nov 13, 2004 3:05:06 PM

Does anyone have complete exit polling data from before the polls were re-jiggered to match the election results?
I can only find the results for a few states.

Posted by: Glaivester | Nov 13, 2004 5:47:54 PM

1) Republicans did a better job getting their base to turnout in 2004, and thus the GOP thus saw bigger increases in its vote totals over 2000 than the Democrats.

2) These voters tend to live disproportionally in rural areas.

3) For logistical purposes (they're easier to get to, usually being situated near interstates), and for their bellweather qualities, exit pollsters usually target suburbs (not rural areas, in other words).

4) Although final weighting (after the polls close) would have picked up the higher overall turnout in rural precincts, and adjusted total voter numbers accordingly, it wouldn't have accurately acertained the increased share of that larger rural vote that went for Bush, because the predictivave modeling references actual results (not exit polling results, of course) from the previous election.

5) One hears about how exit polling is supposedly highly accurate, but it's been a while since we've had an election with this kind of turnout (68? 60?), and it seem to me the modeling was vulnerable to a shift in the geographic distribution of the vote, which, in the case of the election of 2004, obscured a modestly increased total share of the vote for the Republican candidate. It's not that the exit pollsters weren't aware of the increased turnout in rural precincts; they would have been made aware of this after the polls closed, and they would have adjusted their numbers accordingly. Nor is it even that they wouldn't have realized that this increase in the rural vote meant more votes for Bush than for Kerry (they certainly would have realized this, knowing that rural precincts X, Y and Z went for Bush in 2000). Their problem is that they realized the increased rural votes went mostly for Bush, but they underestimated this margin, because their methodology uses the previous election's percentages to as a guide to how the vote breaks in precincts where no exit polling occurs. But if Bush beat his 2000 percentages in rural areas by significant numbers, everything comes undone. Under this scenario we can conclude that the overwhelming majority of the increase in the vote in rural areas broke for Bush (no surprise there). It may also suggest that there were likely good number of rural voters who switched from Gore in 2000 to Bush in 2004.

6) So, it is changes in the geographic distribution of the vote that are the weak link in exit polling. This is not surprising in retrospect, and is rather cenral to the nature of the sampling, and the predictive modeling required to relate the sample to the whole. This is what they need to adjust for 2008. Their methodology doesn't do well with an unstable electorate.

Posted by: P.B. Almeida | Nov 14, 2004 5:37:00 AM

I've read that on civil rights issues, people report being about 10 % more in favor of civil rights than actual voting. That is, if 58 percent answer polls in favor of civil rights legislation, that means probably about 48 percent (at most) would actually vote that way.

Maybe there's something similar going on with republican candidates. Enough don't want to come out as pro-republican, but will vote for certain republicans when it's just them and their conscience in the voting booth to unhinge most polling.

Often enough, dems are perceived as trying to appeal to voters' better natures while repubs are perceived to go straight for baser self-interests.

If anything, this would indicate that in 2008 the dems need at least a five point lead going into voting day to have a credible chance at winning.

Posted by: Michael Farris | Nov 14, 2004 6:22:40 AM

"Wow. Why would non-whites be more easily able to vote earlier in the day, Abiola? Let alone working class people. And the idea that pollsters would find earlier times to be more convenient is bit strange."

Go ahead and make the accusation of "racism", I dare you ... Sheesh, talk about the substitution of vacuous ad hominem rhetoric for actual thought.

Posted by: Abiola Lapite | Nov 14, 2004 9:58:01 AM

Not like I know for sure, but I did notice that the exit polls had a higher percentage of women respondents than ultimately had voted, I think. If men were more likely a) not to be approached or b) to beg off rather than answer questions...?

Posted by: TtP | Nov 15, 2004 10:44:33 AM

Exit poll data keep being revised to show far higher Hispanic support for Kerry. Newest info summarized at chez Nadezhda.

Posted by: nadezhda | Dec 4, 2004 6:25:43 PM

The comments to this entry are closed.