Early Voting
Early voting is set to play an unprecedentedly large role in this year's election, accounting for perhaps a third of the voters. The election, in other words, is already well under-way. After months of speculation, we're starting to get some hard numbers on how people are actually voting. And the numbers suggest that Obama is doing very well, indeed.
I've found six national polls from the past week that have broken out early voters. (If anyone's aware of others, please add a link in the comments and I'll update this post.) And the results are worth a little head-scratching.
On one side of the divide, we have four polls that indicate a massive Obama lead among early voters. A poll released Thursday from ABC/WashPost found that 31% of voters plan to cast their ballots early, and put Obama up 61-35%. The Pew poll, out two days before, had largely similar findings - 31% voting early, and Obama up 58-34%. Yesterday, Newsweek put the split at 60-35%. Zogby variously put Obama 20, 22, and 27 points ahead last week, but his sample size was tiny, and his results should always be treated with suspicion. Still, that's four polls, and a high degree of consensus - Obama ahead of McCain among this group of voters, by a margin in the mid-twenties, more than double his lead among election-day voters in each of these polls.
On the other side are two of the largest and most stable national tracking polls, both of which show Obama doing only marginally better among early voters than among the electorate at large. On Friday, Rasmussen said that 36% of voters will act before November 4, and published numbers that work out to an approximate 55-45% split. And then, of course, there's Gallup, which finds that the 30% voting early are pretty much the same as the rest of the voters.
That's quite a range. First, let's note the agreement - all six pollsters find Obama doing at least as well among early voters as among the electorate at large. In other words, they give him a statistically significant lead. That, in itself, is a notable finding, given that in the last two cycles, Republicans won early voters 60-40%. I should hasten to note that most of this change seems to result from a shift in the demographics of early voters, and not in the preferences of a stable population. As many as half of all African-American voters in some polls indicate that they'll be voting early; youth turnout is also up. Also, states have expanded early voting opportunities in the past four years, so there are some bluer states in the mix this time around.
But the range of Obama's lead in this group is from the high single digits to 27 points. Polling is at least as much of an art as a science, and results have ranged widely this year, but that's still a ridiculous spread. And, more to the point, the underlying results are divergent: four surveys say early voters are a core strength, and two find them pretty much like other voters. So what's going on?
There's a second set of numbers that's worth pondering: state (and district) level surveys. These are useful for the more granular detail that they offer, particularly in states that don't release early balloting totals by partisan affiliation. They come with a caveat - they're self-reported, and include only those voters who have cast their ballots at the time of the survey, not the larger population of all voters who intend to vote early.
SurveyUSA, to its credit, takes the trouble to ask about early voting in its state and local surveys. The results are instructive; here's a sampling. In the Indiana-9 race, for example, the Democrat has a 54-39% lead, but a massive 64-32% edge among the 12% who have already voted. (Although the day before, a statewide survey found a virtually identical 49-45% vs 50-46% split.) In heavily Republican Kansas, Obama trails 41-53%, but actually leads 53-45% among the 15% who have cast ballots. In the Washington-8 battle, the Democrat leads 50-46% overall, but 57-39% in the 20% of early voters. In Maine, it's Obama up 54-39% overall against 61-34% with early voters. In North Carolina, it's an even 47-47% split overall, but a landslide 59-36% among early voters. Wyoming has Obama down 58-37% overall, but by a narrower 55-41% among early voters. And on, and on.
What's interesting here is are not the margins themselves, but the question of relative performance. The Group of Four shows Obama doing much better among early voters; Rasmussen and Gallup disagree. But the state-level polls run by SurveyUSA seem to validate the former results - with the exception of Indiana, they show Obama running much stronger among those who have already cast their ballots than with those who have yet to vote.
The undisputed guru of all things early-voting is Michael McDonald of George Mason University. Bookmark his webpage; from now until election day, it'll give you a better gauge of where the campaigns stand than any tracking poll. As of this morning, his numbers included six states in which partial or complete returns include partisan identification. In aggregate, Democrats composed 47.6% of these voters, and Republicans 40.7%. Of course, that tells us only a little. Most polls have Obama leading among Independents, and pulling more Republicans than McCain is Democrats, so his actual lead is almost certainly larger. And six states do not a nation make, particularly when just two - Florida and North Carolina - account for almost all of the early votes. But comparing these numbers to those of the past two cycles provides some pretty good indications that Obama is showing extraordinary strength among early voters.
But it's some of the fine-grained detail on McDonald's page that is most interesting. In every state, turnout has surged and Democratic participation is way up. In states that report racial data, black turnout is also up dramatically. And, here's the key fact - in states like New Mexico, North Carolina, and Florida, Republicans continue to fare well in absentee balloting, but are getting crushed as early voting returns roll in. That's important because of the SurveyUSA data. Absentee voters have, in most cases, already mailed in their ballots - they should be included in the subset of those who tell pollsters they have already voted. Early voting, by contrast, is just getting underway. So if Obama is doing much better in early voting than in absentee voting, we'd expect to see the SurveyUSA numbers give him a narrower lead (among voters who have already cast ballots) than the national polls (which are looking at voters who will eventually cast ballots early, including both absentee and early voters). That, in fact, appears to be the case using the Group of Four data, which predicts he'll eventually over-perform in this groups by double digits. But it's emphatically not the case if we use Gallup and Rasmussen, which suggest that he shouldn't even be doing as much better with these voters as SurveyUSA suggests he already is.
The bottom line here is that we have a dispute - and that the state-level data and actual early returns seem to indicate that Obama is doing better among early voters. Given that the three of the four polls that had him doing much better among this group also gave him double-digit leads overall, and that both of the polls that had him performing about the same gave him leads in the mid-single digits, it seems reasonable to conclude that the models used by the first set of polls are proving more accurate and predictive thus far, and that we might do well to assign greater weight to their conclusions. (This is a tentative conclusion. More national polls will begin asking these questions, more state-level surveys will provide their own numbers, and more state returns will become available. As that happens, I'll revisit the available information, and adjust conclusions as necessary.)
One final caveat: voters lie. As Mark Blumenthal recently pointed out in a fascinating post, a validation survey in 1980 found that 88% of the registered voters who didn't vote that year had told a pollster before the election that they intended to vote, and that after the election, 44% reported voting despite the fact they hadn't. This kind of thing makes it very, very difficult to successfully predict which respondents are actually going to go to the polls - and has helped produce the wide range in national surveys that are employing different likely-voter screens. Measurements of early voting compound the problem in three ways: the sample sizes are smaller, both the rules and voter behavior are changing rapidly with each cycle (depriving pollsters of a good baseline), and likely-voter screens are calibrated against the overall likelihood of voting as opposed to the likelihood of early voting and thus may skew the results. So take all of this with a larger-than-usual dose of skepticism.
Dept of Corrections
Finally, I'd like to offer a series of corrections to my last post, and my humble thanks to reader IIOOII for pointing out the errors. I was using a Gallup post that read, in part:
Gallup Poll Daily tracking conducted Oct. 17-22 finds that among those who have already voted or say they plan to vote before Election Day, 31% support Barack Obama and 29% support John McCain.
I took that at face value; in fact, Gallup was actually trying to say that 31% of those who support Barack Obama intend to vote before election day. Since the statement would have meant that 40% (and not 30%, as I wrote) of these voters were still undecided, I should have realized that something was wrong and dug a little deeper. My apologies.
Advertisement
Is this going to screw up the accuracy of exit polling reported on election evening? Have the polling organizations chosen for that by the major news organizations figured a way to take into account early voting in their election evening calls? If not, I can foresee calling a state here and there too close to call, then the scenario of arguing over the accuracy of counting early voting, just as in 2000 there was arguing over absentees.
October 26, 2008 1:59 PM | Reply | Permalink
The way the exit polling consortium handles this is by conducting a telephone survey of early voters in advance of the election, and then rolling the results in with those of the exit poll itself. It's a little more complicated than that - they use screening questions and past data to validate their results - but that's the gist of it.
So, in theory, if the polls remain divided over how the early voters are voting, it could throw the exit poll results. But bear in mind three things. First, the numbers discussed above are national results; exit polling gets conducted at the state level. Second, part of the current spread is undoubtedly that two-thirds of respondents are prospective - as these people actually cast their ballots (or don't) the results should theoretically converge. And perhaps most importantly, remember that exit polls are lousy predictive tools. Their main value is retrospective and explanatory - after their adjusted to reflect the actual turnout and results, they do a pretty good job of helping us understand why things played out the way they did. It's almost always a mistake to pay too much attention to exit polls, particularly to the first two waves - in well-polled races, the consensus track has a better predictive record.
October 26, 2008 7:21 PM | Reply | Permalink
I've had the same thought, artappraiser.
I've also wondered about the strangeness of doing polling altogether if they include both people who have already voted and people intending to vote. Seems to me there's a catch 22 in there somewhere.
October 27, 2008 1:22 AM | Reply | Permalink
That, in itself, is a notable finding, given that in the last two cycles, Republicans won early voters 60-40%.
Just trying to clarify -- does this mean that 60% of Republicans and only 40% of Democrats voted early in the last two elections?
If so, I find it interesting that right-wing blogs are insisting that McCain is behind right now in early voting because Republicans like to wait for election day to vote.
October 26, 2008 8:04 PM | Reply | Permalink
It's stated in terms of preference, not partisan affiliation. In other words, in the last two elections, roughly 60% of the ballots cast before election day (absentee, mail-in, & other forms of early voting) were cast in favor of the Republican candidate for the presidency.
That said, there is a grain of truth to the GOP talking points. What we're seeing right now is an enthusiasm gap - huge numbers of Obama supporters are stampeding to the polls and mailing in their votes, while many unenthused Republicans are going to wait until Nov 4 to hold their noses and pull the lever. There's pretty good data to suggest that early voters of all kinds tend to be committed partisans, and that they tend to vote most years. Put differently, these folks were going to vote the same way irrespective of when they happened to cast their votes. So yeah, you could imagine a situation in which a minority of die-hard supporters rushed to vote, while a less enthusiastic majority waited until the election to back the other candidate. The problem with the Republican argument isn't history, it's statistics - we have very good evidence to suggest that that's simply not what's happening here.
October 26, 2008 8:37 PM | Reply | Permalink
Thanks, Fly, for clarifying that. It makes sense that unenthusiastic people will tend to drag their feet.
October 26, 2008 11:25 PM | Reply | Permalink
And possibly they get disgusted with the level of crap flying around and just end up blowing the whole thing off.
October 26, 2008 11:48 PM | Reply | Permalink
The 60/40 number usually comes from the Kate Kenski paper
http://www.allacademic.com//meta/p_mla_apa_research_citation/0/4/1/4/7/pages41477/p41477-1.php
which is in turn based on the Annenberg data from U Penn
http://www.annenbergpublicpolicycenter.org/
One multivariate analysis of the same data shows that those identified as Democrats had decreased log odds of 38% for voting early in 2004.
What does all this mean?
It means that things look very good for Senator Obama this election cycle.
October 27, 2008 5:26 AM | Reply | Permalink
Thanks for the data / analysis! You're giving Nate Silver a run for his money.
October 26, 2008 9:33 PM | Reply | Permalink
Haven't even read the post yet, but welcome back. It was a pleasant surprise to see your handle in the rec'd list. I remember when you were something of a celebrity around here, and for good reason.
October 26, 2008 9:57 PM | Reply | Permalink
And now, having read the post, thanks again. I hadn't really thought much about this dynamic. Since early voting was such a big part of Obama's GOTV strategy, I guess that partly explains these numbers (also gives a good indicator of the effectiveness of his ground operation). Also, this tells me (as you say in your response above) that McCain's support is obviously much softer. If enough of those people waiting to hold their nose and pull the lever on election day decide it's a done deal, this election could really be a blowout, beyond what any poll predicts.
October 26, 2008 10:18 PM | Reply | Permalink
I'm thinking that highly committed young people, many of whom are cell-only, are going to blow previous year's data models out of the water. As I understand it, all these polls depend on weighting that is based on historical turnouts as categorized by various demographics. What happens when a hundred-year-flood comes along that doesn't fit any previously studied models?
October 26, 2008 11:53 PM | Reply | Permalink
The major (and good) pollsters will adapt their models.
Quite a few of the larger and more reputable polling organizations already attempt to take account of cell-only voters.
You can even graphically see the difference between what Gallup defines as its "traditional" likely voter (LV) model versus its "expanded" LV model, the latter attempting to capture 'higher turnout among groups of voters traditionally less likely to vote, such as young adults and minorities.'
October 27, 2008 5:37 AM | Reply | Permalink
I'm thinking that "highly committed young person" is an oxymoron. I hope we're not really depending on them.
October 27, 2008 8:28 AM | Reply | Permalink
i wonder about party identification of 'Rush' democrats. I suspect most don't tell pollsters that they are dems, but an early vote will be marked at a dem voting. i wonder if the effect is enough to make things look significantly more one-sided than reality.
October 27, 2008 8:29 AM | Reply | Permalink
This is a nightmare year for pollsters, because the electorate is so changed it is impossible to build a model with any degree of confidence.
And any survey of early voters that's worth anything MUST be an exit poll. Phone surveys won't cut it, because you miss all the students and others with unlisted numbers or no land line.
And if 60 percent of the early voters in one location are - say - university students, how would you know to skew your telephone model to that fact?
You are right. Pollsters are all over the place on this, because they do not know who to poll this year or how to reach all of those people.
It's more a guessing game than ever.
October 27, 2008 8:44 AM | Reply | Permalink
80% of the votes in this country are cast and/or tabulated on equipment built by Diebold/Premier or ES&S.
In DuPage County IL where I live all early voting is done on Diebold TsX touchscreen machines. The DuPage County Election Commission is appointed by the local Republican party which holds all 18 seats on the County Board. There are 14 early voting places in the county. With TsX machines you can cast your vote for your local races at any of them, even 30 miles away from home near where you work if you like. Because of that these early votes will never be used in the mandatory Illinois 5% audit or most likely in any recount. Republicans would fight anything but a machine recount tooth and nail and down ballot votes spread out over 14 different polling places make it prohibitively expensive for most challengers to even try. In any case the flimsy paper poll tape back ups on these machines are a nightmare to work with.
Who you vote for may just depend on who controls your local county's election machinery, or who can get next to any of the voting machines or tabulators for a minute with their own memory card programmed to flip votes. Do you know all the employees of the private company your county has undoubtedly farmed out running the election to?
If you can always take a paper ballot, never use a touchscreen if you can avoid it. Chances are even the paper ballot is counted on a computer tabulator which is very easy to manipulate too. But at least there is a real paper record that we can fight to have counted.
You shouldn't be surprised by this. Read Bradblog for more.
October 27, 2008 9:05 AM | Reply | Permalink
Already not surprised. But thanks for the link - hadn't stumbled across that one yet.
October 27, 2008 11:23 AM | Reply | Permalink