Quant geeks handicap the election
October 29, 2004 3:49 AM   Subscribe

An highly quantitative approach to state by state poll analysis. This is a meta-analysis directed at the question of who would win the Electoral College if the election were held today. Meta-analysis provides more objectivity and precision than looking at one or a few polls, and in the case of election prediction gives a more accurate current snapshot. Backup site here. These calculations are based on all available state polls, with an emphasis on likely voter data that include Nader where he is on the ballot. Three or more recent polls (up to seven days old) for each state are averaged and the standard error of the mean is used to calculate the probability of every combination of possible state results. The map is not identical to the median. Results are defined as not statistically significant (n.s.) if the probability is less than between 5% and 95%. The effects of turnout are not included, but can be calculated using the bias analysis.
posted by psmealey (28 comments total)
 
I'm no statistician, so I can't be sure why Princeton's results differ from the map I've been compusively watching, but suffice to say that they do differ - which should be a reminder that everyone involved in getting out the vote to defeat George W. Bush for a second time still has their work cut out for them on Tuesday, November 2nd.
posted by JollyWanker at 4:44 AM on October 29, 2004


It still doesn't take into account the 5-10% of voters who only have cell phones, who are overwhelmingly young and Democrat-leaning, and who may well turn out in high numbers for this election.
posted by Pretty_Generic at 4:57 AM on October 29, 2004


Yay.
posted by Pretty_Generic at 4:57 AM on October 29, 2004


who are overwhelmingly young and Democrat-leaning

That's my hope as well. However... I know that conventional wisdom suggests that this is so, but I wonder if there's a current statistical basis that reinforces it. Cautiously optimistic as I am, I still don't see why younger voters wouldn't be as bitterly divided as the rest of the electorate.
posted by psmealey at 5:03 AM on October 29, 2004


Primarily because youth likes to rebel.
posted by Pretty_Generic at 5:25 AM on October 29, 2004


Tuesday can't come fast enough. Even if the election is a huge mess, we'll be spared the need to check the latest polls ten times a day...
posted by The Card Cheat at 5:50 AM on October 29, 2004


I wonder if there's a current statistical basis that reinforces it.

I have a hunch there are no good statistics on the political leanings of people who cannot be polled.
posted by smackfu at 6:23 AM on October 29, 2004


Thanks for your searing insight smackfu.
posted by psmealey at 6:28 AM on October 29, 2004


I'm no statistician, so I can't be sure why Princeton's results differ from the map I've been compusively watching, but suffice to say that they do differ

Electoral-vote.com uses "latest poll wins" methodology, and takes polls from just about every firm. Race2004.net averages the last three polls. And this one apparently does some sort of super advanced statistical mathematical weighting. Or something. But it's run by a professor at Princeton, and a Republican -- so the fact that it shows a Kerry lead, even without the undecideds, is quite heartening. (Both others show a Kerry lead, as well, in the "if the election were held today" predictions; we've got an electoral polling trifecta on our hands!)

I wonder if there's a current statistical basis that reinforces it.

Actually, there is (if you trust an internet poll -- I'm not sure of the methodology, but I don't imagine it was just point-and-click).

"A New York-based television consulting firm randomly questioned 1,225 voters over the Internet and concluded that "cell-only" voters favored Kerry by 53% to 38%." (LA Times)
posted by rafter at 7:15 AM on October 29, 2004


(Actually, as of today, race2004 has switched into the Bush column, but that's without any new voters. We'll see in four days, I guess.)
posted by rafter at 7:19 AM on October 29, 2004


Further, I take back what I said about the guy being a Republican. I didn't catch the Actblue link to balance out the RNSC link. He actually says on another page clearly, "I am a Democrat."
posted by rafter at 7:34 AM on October 29, 2004


if you trust an internet poll

I don't really, but it's at least a place to start. There are definitely other methods available to polling these "non-pollable" voters (campus sampling, focus group sampling, etc.). That's why I bristled a bit at the notion that, since they're not reachable via phone, well, we just can't reach them. Not true.
posted by psmealey at 7:50 AM on October 29, 2004


woohoo!

I think the guy at electoral-vote does a great job, but his methods are not all that insanely scientific. He will throw out a poll that's one day older, or the same age but 1 day longer in duration, etc. One poll at a time goes into his "map," which is why it's the kind of site you HAVE to visit EVERY DAY. I've always thought there's got to be a better way to someone factor them all in and age-decay their weight or something.
posted by scarabic at 8:41 AM on October 29, 2004


Except the votemaster tried to do a weekly rolling average and got shellacked by his readers for it. He's really just giving what the public wants.
posted by calwatch at 8:54 AM on October 29, 2004


I don't go to E-V.com much, but I think the real think to watch there is his Predicted Final Results. His methodology there, if I'm not oversimplifying it too much, is pretty basic. He assumes that undecides will break 2 to 1 to Kerry, so any state where Bush is currently polling at less than 48% will go to Kerry. It seems ridiculously simple, but I have heard interviews with Zogby, as well as Wang, over the past year and they both of said that there's something to it. I dunno. Guess we'll all find out on Tuesday.... or, at least what we learn on Tuesday will be a prelude to a long court battle in Michigan, Ohio, Florida, etc.
posted by psmealey at 9:06 AM on October 29, 2004


Jollywanker: I know a lot of people follow Electoral-Vote.com, but I see that site's numbers as consistently biased toward Bush (not necessarily intentionally biased, but nonetheless biased). Right now they have PA tied -- but most polls for the last week show Kerry winning PA handily. MI and MN are closer, but leaning Kerry in most of the polls I've seen.

MyDD.com, a pro-Dem site, has some excellent poll-crunching -- the site's editor subscribes to all the major polls. Yesterday he posted all the major poll numbers from the ten Zogby battleground states, and even averaged the the numbers.

I took MyDD's numbers one step further, and factored in 2/3 of undecideds breaking for the challenger, plus a 0.5% advantage for Kerry based on fired-up voters and massive get-out-the-vote efforts. Even without my augmentations, however, Kerry wins. My chart is here.
posted by Artifice_Eternity at 9:07 AM on October 29, 2004


I've been hearing from everybody that the undecideds will tend to break for the challenger which I've found reassuring. I actually think it will be the "unlikely voters" who will come out in droves and who will blow out the polls in Kerry's favors, but still...

This Guy says conventional wisdom about undecideds always breaking for the challenger isn't supported by historical numbers. He pretty clearly has a reason for wanting to believe that, but has anybody seen any contrary analysis to this position?
posted by willnot at 9:46 AM on October 29, 2004


MyDD (a Kerry supporter) did an analysis, with data, of how undecideds vote and claims that "66-34 [for the challenger] is where the smart money, the house money, should be."
posted by kirkaracha at 10:08 AM on October 29, 2004


He pretty clearly has a reason for wanting to believe that, but has anybody seen any contrary analysis to this position?

I'm not about to crunch the numbers myself, but the (Democratic) mydd.com finds:
"In 28 final pre-election polls in Presidential elections featuring an incumbent from 1976-1996, a whopping 86% of undecideds broke for the challenger, while only 14% went for the incumbent."

It seems ridiculously simple, but I have heard interviews with Zogby, as well as Wang, over the past year and they both of said that there's something to it.

Last night (10/29) on the Daily Show, Zogby himself called the election for Kerry for just this reason. John Stewart didn't ask the, "So are you totally missing all the new voters question?" which I think will actually be more important in tipping the scales.
posted by rafter at 10:15 AM on October 29, 2004


I'm partial to this page, run by an economist at Minnesota. The simular approach he's taking is simple but allows for more nuanced interpretation than the naive approaches of places like e-v.com, and it more closely addresses the question we're interested in: what are the odds on election night?

I really wish it were possible to get the individual-level results from the statewide polls and do something similar there: find probabilities of turnout and then run 100,000 simulated elections to get a sense of how likely each candidate is to take that state.
posted by ROU_Xenophobe at 10:30 AM on October 29, 2004


If one delves into the linked Blogspot site a little more deeply, the author flip flops, err I mean, clarifies his conclusion:

"There finally is some evidence of undecideds breaking for challengers, but not of a consistently strong break."

His numbers under a variety of conditions point to a 1-2% overall break for challengers, which is actually fairly consistent with what the other pollsters have said (this is a large break when you consider the 3-4% or so that are undecided). With the race either dead even or within the statistical margin of error, a 1-2% break in likely voters could very will tip the scales even before factoring in a surge in turnout from "unlikely" voters, underpolled demographics (the much-debated cell-phone crowd, for example), and so on, all of which are in Democratic favor.

Of course, the Republicans still have the 1) "selective" voter registration in Nevada and elsewhere, 2) Diebold in Ohio, and 3) Jeb Bush and other ballot funniness in Florida and cards in their hand, as well as whichever surprise ace Karl Rove is hiding up his sleeve. It ain't over 'til the last Supreme Court justice sings.
posted by DaShiv at 10:49 AM on October 29, 2004


Well, that's all good and fine, except ...

This is a meta-analysis directed at the question of who would win the Electoral College if the election were held today.

... except the election isn't going to be held today, it's going to be held on Tuesday, and anything could change between now and then.

I'm sick to death of polls. That said, I'm remaining optimistic.
posted by chuq at 11:11 AM on October 29, 2004


Regarding the newly registered youth voters with cell phone meme that has rolled through the media and the internet(s): I want to believe, but it smacks of Gov. Dean's "Perfect Storm" in Iowa that turned out to be a drizzle.

So forget the polls and the analysis. This isn't the time. Contribute, work hard this weekend, and let's all wake up Wednesday morning knowing we did what we could do shape our country in this critical moment in our history.
posted by russh at 2:17 PM on October 29, 2004


3) Jeb Bush and other ballot funniness in Florida

Eek. These words just make me cringe. The biggest part of me says "no, there's no way he would try to get away with something with all the attention that's on him," but there was quite a bit of attention focused on him last time around, too.

Uh oh :\
posted by scarabic at 2:45 PM on October 29, 2004


ROU_Xenophobe, a minor problem with that methodology is the assumption of a normal distribution: "I assume that, in each state, the sampled Bush's vote share follows a normal distribution, with mean equal to the share and standard deviation equal to the reported margin of error divided by 1.96". I do not believe this is the case. Bootstrap would work better, but she needs access to raw data, which unfortunately, is not available.
posted by MzB at 4:40 PM on October 29, 2004


Sure, but it's still better thought out than e-v.com and similar.
posted by ROU_Xenophobe at 4:50 PM on October 29, 2004


why don't you think it's normal? he's talking about the error in sampling a large population - it's pretty much the textbook case where such a distributin should apply.
posted by andrew cooke at 7:19 AM on October 30, 2004


I think the original distribution is hypergeometric (or even a “modified multinomial”). Indeed, it converges to normal, but there are errors. It might not change the results much, but for a swing state the marginal effect (all Bush vs all Kerry) might matter. It’s not a big deal, just an academic nitpick! :-) As ROU_Xenophobe said, the method is better than the one used by electoral-vote.
posted by MzB at 11:34 PM on October 30, 2004


« Older Halloween Flash Fun   |   What Every Child Should Know Newer »


This thread has been archived and is closed to new comments