« O'Reilly Hated This... | Main | The Soloist Leaves AFI Fest »
October 22, 2008
Trying to Figure This One Out
(Note: See addition at bottom to know why the survey is skewed)
A new Associated Press-GfK poll says that "McCain and Barack Obama essentially running even among likely voters in the election homestretch."
But as I read the piece about the poll, I was very confused.
"The poll... found Obama at 44 percent and McCain at 43 percent"
But later, it said:
"Polls are snapshots of highly fluid campaigns. In this case, there is a margin of error of plus or minus 3.5 percentage points; that means Obama could be ahead by as many as 8 points or down by as many as 6. There are many reasons why polls differ, including methods of estimating likely voters and the wording of questions."
So the +/- 3.5% means up to a 14 pt swing in the polling numbers?
Would that make anything less than a 15 pt lead suspect in most polls?
Another oddity is interesting...
"The AP-GfK survey included interviews with a large sample of adults including 800 deemed likely to vote. Among all 1,101 adults interviewed, the survey showed Obama ahead 47 percent to 37 percent. He was up by five points among registered voters."
So... trying to figure out this math... 73% of those polled were deemed likely to vote. Working backwards, it seems that the entire remaining 27% were registered, but deemed unlikely to vote.
If Obama had a 10 point lead with all 1101 adults and a 1 point lead with 800 of that 1101, he would have to have a 37 point lead with the 301 "not likely" voters for the math to make sense, no?
301 unlikelies - 204 pro-Obama, 96 pro-McCain
800 likelies - 404 pro-Obama, 396 pro-McCain
Based on the numbers directly above -
Total – 610 Obama / 492 McCain – 55% Obama
Deemed Likelies Only – 50.5% Obama / 49.5% McCain
So I guess the question about this poll is… are people who are “unlikely” to vote so much more pro-Obama than everyone else… or is this polling organization, eliminating 27% of registered voters as unlikely to vote, cooking the books?
And while I know that some will think I am just being pro-bama, it really is the math that caught my eye. The lede and headline says one thing, but every other piece of math in the story seems to bring that headline into question.
Or maybe I just need to go back to school to learn more math...
ADD, 7:20p - Now it makes a little more sense... this is from the poll report by way of The Atlantic's Marc Ambinder.
The highlight that answers the question...!
Posted by dpoland at October 22, 2008 12:20 PM
Comments
Dave, just read 538.com. Nate Silver covers this poll and itsthe best polling site on the web
Posted by: Mulder215 ![[TypeKey Profile Page]](http://www.mcnblogs.com/thehotblog/nav-commenters.gif)
at October 22, 2008 02:26 PM
Seconding fivethirtyeight.com.
Posted by: RocketScientist at October 22, 2008 03:10 PM
Seconding RocketScientist:
http://www.fivethirtyeight.com/
If you type digits, Mulder, it takes you to something that looks convincing, but isn't Nate Silver's site.
Posted by: Not David Bordwell at October 22, 2008 03:44 PM
Better yet, here's the analysis of the poll(s) in question:
http://www.fivethirtyeight.com/2008/10/some-likely-voter-models-are-suspect.html
Posted by: Not David Bordwell at October 22, 2008 03:53 PM
The partisan weighting is problematic as well. For instance, Rassmussen has Obama on top by 6%. What's the party affiliation of those polled? Dems 39.3%, Rep 33% ("unaffiliated" made up the final 27.7%). In other words, Obama is up by 6%, with 6.3% more Democrats represented in the poll.
Now, its entirely possible that more Dems may vote, and by a margin of 6%. But turn-out projections have been misleading in the past.
Posted by: mysteryperfecta at October 22, 2008 04:04 PM
Partisan weighting, mystery, is generally based on the weighting of the two parties in the various states.
Posted by: David Poland at October 22, 2008 04:32 PM
Yes, but that's also based on the nationwide number of larger Democratic party-identification numbers which is something like 40-33 right now.
Posted by: jeffmcm at October 22, 2008 04:33 PM
"Partisan weighting, mystery, is generally based on the weighting of the two parties in the various states."
And it hasn't necessarily translated into votes. Which is why its problematic.
"Yes, but that's also based on the nationwide number of larger Democratic party-identification numbers which is something like 40-33 right now."
There'd be even a larger divide if there were more Disney characters.
Posted by: mysteryperfecta at October 22, 2008 04:40 PM
That gets a big fat whatever. The facts of the matter are, right now, nationwide, Democrats have an advantage in party identification of up to ten points. This is a statistical fact which the various polling groups are taking into account in their methodology.
Posted by: jeffmcm at October 22, 2008 04:50 PM
"This is a statistical fact which the various polling groups are taking into account in their methodology."
Where did I dispute this? I merely said that party identification is not a reliable indicator of turn-out, and turn-out is a vitally important factor.
Rasmussen not only uses state voting history, but national trends and recent polling to determine partisan weighting.
Posted by: mysteryperfecta at October 22, 2008 06:04 PM
You sounded pretty snide earlier, that's what I was responding to.
Agreed, turn-out is an unknown, but based on what's already happening with early voting, there are some trends.
Posted by: jeffmcm at October 22, 2008 06:07 PM
"You sounded pretty snide earlier, that's what I was responding to."
The Disney remark? It was just an ACORN joke. I initially had a smiley at the end of the sentence, but erased it.
I don't know how to quantify early voting; its wide availability this year is unprecedented. Suffice it to say, Obama supporters have to feel pretty good about where things are at, despite these poll aberrations.
Posted by: mysteryperfecta at October 22, 2008 06:37 PM
Allow me to encourage the use of smileys :)
Posted by: jeffmcm at October 22, 2008 06:53 PM
To be honest, I was initially going to use the razz smiley :p, then I changed it to a wink ;), but both of those can come off as patronizing. Should have stuck with the tried-and-true :).
Posted by: mysteryperfecta at October 22, 2008 07:03 PM
Hey, any of those would have avoided what I thought was out-and-out contempt 8|
(that one is what I look like most of the time.)
Posted by: jeffmcm at October 23, 2008 12:44 AM
"Would that make anything less than a 15 pt lead suspect in most polls?"
No, it would rather mean that anything less than a 7.5 pt lead is suspect. Say the polls showed 54-46 for Obama - that could mean anything between 50-50 and 58-42, but it could never never be a lead for McCain.
Posted by: Mikkel at October 23, 2008 10:18 AM
Post a comment
Thanks for signing in, . Now you can comment. (sign out)
(If you haven't left a comment here before, you may need to be approved by the site owner before your comment will appear. Until then, it won't appear on the entry. Thanks for waiting.)