Georgia On My Mind
Last night, a memo from a McCain pollster touched off a bit of a kerfuffle. The pollster made a number of claims, but the most incendiary were these: that turnout among young and black voters would largely be offset by generally higher turnout, and that Obama's polling numbers represent not his baseline support but rather his ceiling, because all the black voters are already telling pollsters they support him, while a large percentage of other voters are wavering and unlikely to move in his direction.
There is, of course, a kernel of truth in some of these claims. Interest in the election is generally high, and that will to a degree mute the effects of the extraordinary surge among young voters and black voters. Obama does, in fact, enjoy remarkably high support among black voters, and has somewhat less room to increase that margin than other Democrats have occasionally enjoyed. As for undecideds, I'll let Scott Althaus throw cold water on that one, and then pile on by noting that there are about the same percentage of undecideds with white and black voters.
Of course, the memo was also largely constructed on a shaky tissue of lies and distortions. In the memo, McInturff argues that "In most polls, McCain is losing these African American voters by margins like 97% to 1%." But as I pointed out to Ben Smith last night, those numbers appear to be fabricated, and rather overstate the case. I should also have noted that McInturff was exaggerating in the other direction, too: he claims that Republicans "typically" trail by 78-14% among black voters, and that if a poll says 78-18% he advises clients to expect no more than 8%, because "the Democrat candidate will typically carry more than 90%" of the black vote. I'll leave the insinuation that black voters are uniquely dishonest to one side; let's just point out that the split in 2004 was either 88-11% or 86-14% by the exit polls, but that Gallup put black support for Kerry at 93%. So if there's a switch this year, it may be that the polls aren't overestimating black support for Obama as they did last time around.
In response, Nate Silver put up a handy-dandy chart showing that early voting turnout seems to be surging as a function of black population, and suggested that even if there's less late movement in this group this year, the surge in turnout will more than offset that disadvantage. Of course, a surge in early voting can simply mean that the same voters as before are casting their ballots a little earlier. To get a better sense of what's going on, I decided to take a close look at a single state for which we have particularly good data: Georgia.
Think about that for a moment. If black voters simply account for the same proportion, relative to their registration, as they did in 2004, they'd cast 27.2% of the ballots. So three of these polls show a relative decline for black voters, and one shows their proportion staying constant. On the other hand, we have the data from the primaries - in which blacks turned out at a substantially higher rate than other groups - and from early voting - in which they've comprised 35.2% of voters. (The primary is instructive: Polling in the final days projected black voters at "about half," 38%, and 41% of Democratic voters. Even the adjusted exit poll reported 51%. But the hard count from the Secretary of State's office reports they accounted for 55.4% of voters.)
So where are all the black voters? One answer is provided by pollster Ann Selzer, who reports her belief that blacks this cycle are more likely to refuse to answer demographic questions. So perhaps surveys are picking them up and reflecting their levels of support, but aren't able to report their race. But it seems at least as likely that surveys are simply missing them. I suspect that these missing black voters are, in part, a graphic illustration of the cellphone problem. Through the end of 2007, the government reported that 14.5% of adults lived in wireless-only households; that projects to more than 16% today. Back then, black adults were over 40% more likely than whites to live in wireless-only households (18.2% then, likely 20%+ today). Other variables that correlate with wireless-only households? Youth, poverty, living in metro areas, cohabiting, and renting. If you were trying to construct a profile of Obama's core voters, you couldn't do much better than that. The effect is also most pronounced in the south and midwest. But beyond that, other problems plague estimates of black turnout. One is the use of past voting to model future behavior, which screens out voters who may be particularly excited about Obama's candidacy. Another is non-response bias; communities with historically antagonistic relationships with authority are often less likely to offer information to a pollster. However you account for it, though, it seem fairly clear to me that many polls of this race are dramatically underestimating black turnout.
Now let's assume, for a moment, that this year they'll turn out at the same rates as other voters, and comprise 29.2% of the electorate. With more than nine out of ten black voters supporting his bid, boosting their percentage of the total by just 2-3 points could well be enough to tip the state to Obama. Pollster.com's average shows Obama down by 2.6% in the state; FiveThirtyEight's projection shows him down by 5. And if they actually account for a higher percentage of votes than of registered voters? Well, then Jim Martin will be a happy camper.
UPDATE: This morning's numbers have black turnout holding amazingly steady as a percentage of the total, at 35.05%. More than 550k black voters in Georgia have now cast ballots. By the end of the day, more black voters in Georgia will have voted early in the general election than turned out to vote in the primaries. That's just astonishing.
[If I may indulge in a tangent, Nate Silver has a nifty post up this morning suggesting that the conservative base is staying home, mailing it in, not returning ballots in Oregon. He offers a regression line showing that support for Bush in 2004 strongly correlates with lower turnout this year. SurveyUSA's polls seem to bear that out; the most recent one found that half of likely voters have already returned their ballots, and that Obama is up by 28% among those who have but just 10% among those who still intend to vote (not all of whom, inevitably, actually will). Self-reported Republican affiliation among likely voters in their last three surveys has declined: 35-42%, 36-45%, 33-47%. Now, Oregon is hardly a representative state. But to "mute the increase" among black voters, in the McCain camp's memorable phrase, the GOP needs to juice turnout among its own base voters by percentages similar to the increase among Obama's base. Even just a modest increase would be bad news for McCain; an actual decline in turnout among Bush voters from 2004 would be cataclysmically bad. There's more than one way to boost relative turnout, after all.]
If you've enjoyed this, please share it with other readers by clicking the 'recommend this' link. You can find more analysis on my blog, or subscribe by clicking "Follow Me" on the right. As always, I welcome your comments and corrections, and thank you for your feedback.
Advertisement
A familiar "face" to say: "Thanks again, Fly." Always good to see you. :)
October 29, 2008 7:02 PM | Reply | Permalink
Interesting take. Perhaps Georgia is about to turn blue, which would bode well for Martin, I would think.
Of the three Senate races that are close but leaning Republican, which do you think would be the best to support financially: Georgia, Kentucky, or Mississippi?
October 29, 2008 7:39 PM | Reply | Permalink
Well, I'd say Georgia's the most likely to flip, Musgrove could use the cash, and Kentucky would be the most important win. So I'd suggest you look at all three candidates, and donate to the one in whom you most believe. I've never yet regretted aiding a candidate I admire, even when she loses; more strategic calculi sometimes yield remorse, even when successful.
October 29, 2008 8:06 PM | Reply | Permalink
All of them.
Ted Kennedy's office has a five-race donation fund going:
https://pol.moveon.org/give/oct08_6.html?id=14614-9185498-yF2H3Lx&t=3
No time like the present to be present.
October 29, 2008 9:34 PM | Reply | Permalink
Thanks for another great post. Articleman also called Georgia for Obama a few weeks ago for related reasons. I was very skeptical but am growing more optimistic.
A question for you. I often respond to people with theories about why the polls are skewed one way or the other, especially those who don't know what they're talking about, by arguing that however flawed the art of polling may be, the pollsters at least know more about their trade than the person making the claim. Thus, while there is a good chance the pollsters are wrong, they have a better chance of getting it right than your average armchair poll critic.
You obviously know much more about polling than these folks, but I still wonder why pollsters wouldn't be aware of the biases you describe and adjust their likely voter models accordingly. You note and disagree with Selzer's explanation for the underrepresentation of black voters. Is it your sense that her view is representative of other pollsters? Or do you think that they're not aware of the underrepresentation at all. Or perhaps aware but too conservative about adjusting their models?
October 29, 2008 7:43 PM | Reply | Permalink
You pose a great set of questions. I'll float an answer; I hope you and others will let me know what you think.
TPMCafe actually offered a terrific overview of the problems bedeviling the polling industry just last week. I won't recapitulate it here. But with regards to African American voters, I think the pollsters are in a tight spot. Many of them suspect that the past isn't going to be predictive of the present. They can see in the same data I cite strong evidence that their surveys are probably off. But they're struggling to find ways to incorporate rapid changes without any reliable data on how those changes are impacting the electorate. So if you're used to using RDD, and it generally works, but you suspect that this time the folks you're failing to reach are monolithically on one side of the race and represent a significant percent of the vote, what are you supposed to do? You can't just make up numbers. If you don't normally adjust your data to reflect demographic assumptions, you can't start doing it just for this one race. So Gallup is releasing three sets of numbers every day. Pew toned down its likely voter screen just for black voters, allowing more of them into the sample. Selzer is benchmarking against census demographic data, and only then applying the screen. Zogby's reveling in the chaos. Everyone's got their own solution.
So it's easy for me to identify problems with the numbers they're generating, and even to hazard an educated guess about how accurately they approximate reality. But that's not to say I have a good solution for reaching voters who don't answer their phones, or guessing which voters who've never voted before will this year. I have tremendous sympathy for the pollsters. But I still feel bound to point out these problems.
October 29, 2008 9:13 PM | Reply | Permalink
Fascinating. I haven't been following individual polls and didn't realize that Gallup was releasing more than one number (though I only found two: traditional and expanded). Sure enough, the "expanded" result shows a larger Obama lead.
Unfortunately, I couldn't find Georgia results for any of the polls you mention in your comment. Any idea what the four you reference in the original post, SurveyUSA, GQR, DemocracyCorps, and R2K, are doing? One concern I have is that some pollsters might overcompensate for expected black turnout, in which case the results would be skewed in favor of Obama.
October 29, 2008 10:11 PM | Reply | Permalink
The third set of Gallup numbers is the raw returns for registered voters, before any likely voter screen is applied. The numbers for Obama in that sample are either equal to or one point behind the expanded screen. Releasing their raw numbers, along with weekly demographic splits, is essentially Gallup's way of saying: Here! If you have a better idea of how to weight the results, do it yourself. That's a remarkable thing to do, as they normally keep those numbers private. (We only just found out that in 2004, their registered voter sample had Kerry ahead, while their traditional likely voter screen accurately forecast Bush's win.) I applaud them for that degree of transparency, but wish they'd release not just percentages but raw numbers, so we could really play with their data.
The four surveys I referenced for Georgia were the ones that bothered to let people know the percentage of their sample that was black. SurveyUSA, GQR, and DemocracyCorps projected black voters at 26%, and R2K at 27%, of the total turnout. There are varying degrees of transparency in those polls in regard to methodology. But I know that at least two of them don't re-weight the results to hit demographic targets, and the high level of agreement suggests that even if the others do, it hasn't had a material impact. Any poll projecting black turnout in Georgia at 26% of the total is almost certainly undercounting those voters, whether through sampling, screening, or weighting. (In their defense, I should note that the difference between 26% of the total and, say, 29% of the total is well within the sampling error for subgroups. The problem here is that, in general, we assume sampling errors are fairly random - that they're not introducing systemic biases. But in this case, a wide array of national polls have found that both the voters least likely to be reached by conventional surveys and those most likely to be screened out as less likely voters are overwhelmingly Obama supporters. And when white voters split 7-3 one way and black voters 1-9 the other, any variation in sample composition has an enormous impact.)
October 30, 2008 8:32 AM | Reply | Permalink
Hey Fly.
The Oregon data from Silver do nuke the McCain argument, though I think in the South white turnout will rev better than in Oregon, though nothing like black turnout.
Further to Genghis' point on my post, I actually called GA again, this time in a post that didn't look like a three year old formatted it (e.g., the one of mine Genghis linked) :)
http://www.talkingpointsmemo.com/talk/blogs/articleman/2008/10/obama-effect-update-obama-is-g.php#comments
So, appreciating the sagacity of your judgment, please make a call on Georgia -- clearly it's close enough to say increases will make it close, but what say you, FOTW? And is that pronounced "fatwa"? Just wondered.
Also, I think Martin likely won't win, because Georgia requires a 50% vote for such an election, and the libertarian vote is likely to park both Chambliss and Martin south of 50%, causing a runoff in which fewer black voters will vote as a proportion.
While sticking to the view that Obama will win Georgia (the 35.2% present black vote is well higher than the 32% that I find comfortable, though the number may settle further from its initial peak of 39+), I wrote more briefly last week that Mississippi would be within a point -- in Mississippi 37% of voters are black. Assume they comprise 40% of the total Nov. 4 voters -- then Obama getting 15% of the balance could do it, and internals show him 15-18% in some polls. Your thoughts on that one?
Also, some polls do price in Obama getting 98-100% of the black vote, thinking of recent Rasmussen internals in VA, IN, PA, NC, and GA. But some don't, likely because the black sample within a Ras poll is small and thus less predictive, and I always add back to a level about like 98-100 in trying to project. There are so many of these, and with Kos' African-American internals on his national tracker (95-3 presently), I think national numbers of 98/2, a shade better in the South, seem right. What say you, FOTW?
Best, A
October 29, 2008 8:18 PM | Reply | Permalink
As I noted in my reply to Genghis, it's a heck of a lot easier to point out the flaws in other people's polling than it is to offer one's own predictions. It's generally something I try to avoid. having said that...
Georgia, at the moment, looks too close to call. (Thanks for the runoff info; I'll admit it was news to me.) Mississippi, on the other hand, still seems to lean toward McCain. Let me try to justify those claims.
Here's the thing about black turnout. At this point, we lack good data on the composition of black early voters. Are they the same folks who generally turn out voting earlier, or has Obama succeeded in driving a wave of first-time and infrequent voters to the polls? Intuitively, I lean toward the latter explanation, but I can't demonstrate it empirically. The (admittedly flawed) polling, however, points to the former explanation, suggesting that we're seeing a shift in timing more than one in composition.
It's also important to note that blacks don't "undervote." The last time the Census looked at the question (after the '98 midterms, if memory serves) it found that turnout could generally be predicted as a straight function of socioeconomic variables like education and income. In other words, voters black and white generally vote at the same rates once these other variables are factored in. So for Obama to drive black turnout up even as far as the level of registration would already be a major and historic accomplishment - in essence, decoupling the likelihood of voting from socioeconomic factors, and making black voters (holding all else equal) far more likely to vote in this cycle than voters of other ethnic backgrounds, despite universally high levels of interest and commitment. I wouldn't count on it going much higher than that until I see some data suggesting that it will. I'm also skeptical that any group will ever vote 98 or 99 to 1 for a particular candidate - yes, we see that in some state-level surveys, but those with larger sub-samples have consistently shown wider margins. It could happen, but I'd be hesitant to count on it.
That's a long-winded way of saying that your projections look to me to be extremely optimistic, though not inherently implausible. I hope you're right. And I'll freely admit that I don't really know what will happen on Tuesday. That's the exciting thing about a precedent-shattering election - it's inherently unpredictable.
October 30, 2008 8:56 AM | Reply | Permalink
GA is going blue. I post on it daily as well as NC with details of the early voting stats
http://www.politicalbase.com/profile/jnail/blog/&blogId=5161
I am in Atlanta and we are working hard to build on the enormous early vote.
Articleman was right and we are going to "get 'er done"...Y'all
October 29, 2008 8:44 PM | Reply | Permalink
Word to John Nail.
October 29, 2008 11:50 PM | Reply | Permalink
I loved Nate Silver's post about the low returns from red counties in OR. Let's hope it's a trend.
It's clear that, in this cycle, Dems are much more strongly motivated to vote early than Reps. That could just be a consequence of Obama's emphasis on early voting. Or it could also be a sign of a higher general level of enthusiasm, which will affect turnout all the way through Nov 4.
Anyway, I thought a-man was crazy to call GA for Obama. But I don't anymore. I'd give it roughly even odds.
October 29, 2008 10:29 PM | Reply | Permalink
I love that it looked crazy. I still think it's right. I am very interested in MS, and FOTW's of it. What do you think of MS, based on 37% of the MS voters being African-Americans, Alex?
October 29, 2008 11:52 PM | Reply | Permalink
Wonderful to see you back Fly. There was a great article a couple of days ago over at RCP discussing the difficulty pollsters are having remaking their voter screens and explaining the increase in variance and volatility between polling groups this cycle. It can be found here: http://www.realclearpolitics.com/horseraceblog/2008/10/more_on_the_polls_1.html
October 30, 2008 9:42 AM | Reply | Permalink