November 1, 2012 Analysis (posted early)
I'm sure with RCP moving into a tie and the National Journal poll being released, people are nervous about what is going on. I figured I would post early and hopefully allay some fears.Polls included:
Tracking Polls
R+2 - Rasmussen Daily Track
R+5 - Gallup
O+1.3 - IBD/Tipp
Even - ABC/WaPo
Others
O+1 - Battleground
Even - Pew
R+1 - NPR
O+1 - CBS/NYT
Even - Fox
O+5 - National Journal
O+0.03% - Current RCP Average
O+0.56% - Average using the 2008 turnout model
R+2.23% - Average using the D+3 turnout model
R+4.46% - Average using the 2010 turnout model
R+4.51% - Average using the 2004 turnout model
R+6.24% - Average using the Rasmussen Party ID turnout model
With the bizarre National Journal poll entering the averages, we see the averages again move slightly toward Obama. Romney has just about given back all of his gains from a week ago, bringing the race back to the steady state we were at for the prior two weeks.
While it would be easy to blame the National Journal poll for bringing down the numbers (and to be clear, in this run of the models, that poll is the sole cause for the drop), over the last 3 days overall averages have dropped. I was expecting Rasmussen to move back to 50-47 today, as the Saturday sample dropped out, but that did not happen. Also, ABC/WaPo showed a one point move toward Obama, without changing their sample. We may well be seeing a slight movement in Obama's direction due to the Hurricane Sandy response.
Having said that, Romney is still far ahead in this race. Obama needs about a D+5 or D+6 turnout to make this a close race. There are certainly plenty of polls out there that are predicting that exact result, but all of the early vote results, partisan enthusiasm measurements, and Independent support (which is still averaging 8 points, unchanged over the last 4 days) do not make such a result very likely.
With Gallup and other polls on hiatus due to the storm, we will need to wait to see if other tracking polls confirm this movement toward Obama, or if they remain stable.
I was feeling good. Now? Not so much, lol. Will be interesting to see what the polls say on Monday.
ReplyDeleteWell, I voted today but living in Oregon, it's more symbolic.
I should say I was half-joking. I'm still dubious on some of these polls.
DeleteI received 2 phone calls today from my nervous aunt and mother regarding some of these new polls out today. I referred them to this website and how all of your analysis really puts these ridiculous samples in prospective.
ReplyDeleteI watch Fox News, and am surprised that more of the pundits don't make the viewers aware of the disproportional sampling in over weighing the democratic turnout this election. The only ones I even hear on TV that are doing exactly what you are in terms of analyzing the polls from a realistic turnout in 2012 are Karl Rove and Dick Morris. It only makes sense that if an equal amount of democrats and republicans turn out this election with each voting for their own candidate, but Romney is up 7-10 points with Indies, he will win the election. So, just imagine what a R+1-3 turnout would do for Romney.
Anyways, thanks so much for your great analysis. I have added you to my regular website surfing, especially when I start to get panicked!!!
Yeah, it was nice to see on O'Reilly Morris and Rove (though especially Rove as Morris gets a bit... rambunctious) dissect the polls where everyone else takes them at face value.
DeleteI totally agree...and I believe Rove predicted McCain would lose by a large margin back in 2008, so I think he tries to remain pretty fair, even though we all know he personally supports the Republicans. I love when he gets the white boards out and really breaks it down!
DeleteYep. He's a smart guy and integral to Bush's re-election. He has a new article coming out tomorrow (November 1st) outlying his prediction for Tuesday.
DeleteMaybe I am being a little too cautious. I try not be too much of a cheerleader, and tell people what I'm seeing in the numbers. There is a small shift toward Obama, but at best it brings us back to where we were 7 days ago. We are 6 days from the election, and who wins the polls is not who wins the election.
ReplyDeleteI really do think we have reached the point where new polls rolling in and out of the RCP just doesn't matter. The early vote polling problems are skewing polls that were already poorly executed.
The most important thing is that the real numbers on early voting, especially the Gallup poll, Pew poll, and Ohio actual votes point to a failure of the Obama early vote strategy. He needed to run up a huge total to make up for being beaten on Election Day. Exactly like he did to McCain. But his numbers are down. In Ohio, he has lost the total of his 2008 margin of victory in early votes alone. He won in 2008 by 260,000 votes, but Democrat early votes are down 180,000 and Republican up 80,000. With his 2008 victory margin wiped out, on Election Day he will be handily beaten. The GOP election day traditional advantage, increased enthusiasm, and Independent preference will result in a significant Romney win.
Romney already has Florida, Virginia, North Carolina, and Colorado locked up. With Ohio, he will have 281 Electoral Votes and win the election. Picking up Iowa, Wisconsin, New Hampshire, and a surprise like PA, MI, or MN makes the total larger, but aren't necessary to win.
Very true.
DeleteI am confused on how some of the polls move the weighting like they do. I mean, what's the justification for Fox News to movie from a D+1 to D+5??? Makes no sense to me.
Sorry, typo - movie --> move
DeleteHere's my worry. I don't honestly think the race has moved because of the hurricane, but boy oh boy, would the MSM love to run with that story! Obama's "9/11" moment!
DeleteAnd I'm sure a few pollsters would happily adjust their models to suit that narrative, giving the illusion that Obama has the momentum.
Romney's margin of victory is important because of the Senate races, if it looks like Obama has been put away, I see Democrat turnout being incredibly anemic across the board.
Anonymous,
DeleteMoney. With the exception of Rasmussen, I don't believe any of the polls explicitly weight their samples by party-ID.
But, they do apply a transform to the sampled data which is functionally equivalent by having a loose Likely Voter screen, which means more of the Registered Voters make it into their final data set. Less strict filtering means a higher proportion of phone calls are useful. Less phone calls means less costs.
So we're seeing very large fluctuations as they sample an underlying voter population which is saying they voted and didn't, a registered voter and not, etc. Pollsters like Gallup who have a more rigorous LV Screen, make many more calls, see stable results -- I don't think it's a coincidence.
Dave is avoiding some of these pitfalls by making a trade-off: he gains stability by transforming his dataset into partisan voter space, but he loses the dynamic data about swings in the partisan ID. (ie. A D+3 split can compose a range of D/R/I's that yield different outcomes).
I find this a worthy trade though given the polarized nature of the election, the shitty polling response rates, and that (barring some massive discontinuity) we can constrain the potential space the partisan ID can exist in pretty well using other sources of information, such as registration, early voting, voter intensity polling, explicit voter ID polling by Ras/Gallup, fundraising, etc.
Uriah - Yes, that was a conscious decision on my part. I thought about adding cross over support to my model, but decided not to do it. What I lose from that is the ability to detect large shifts like the Reagan Democrats, but some of the poll already address that by asking "which party do you identify with" instead of "which party are you registered with".
DeleteI have my theory that there really are only two driving factors (3 if you include undecideds). We will find out in 5 days if my theory is right.
Do you have any data on crossovers, I'm thinking there is a little Romney pick up there as well.
ReplyDeleteI do, but my model ignores them. I do that on purpose, because I don't believe that partisans ever support the other party. Some polls identify partisan affiliation, some party registration.
DeleteIt is easier to just assume Dems vote for Dems and GOP votes for GOP.
And if anything, that probably means Romney's margin of victory is being "undercounted".
DeleteHow many "Obama Republicans" have you met compared to "Romney Democrats"? I'd be willing to be the crossovers favor Romney almost 2 to 1.
I have several family members that are "Dixiecrats", registered as Democrats and still give lip service to the Democrat Party, but usually vote conservative. And they despise Obama.
The correct method of combining the results of multiple surveys is to add the responses, not average the results. Accepting all ten surveys in RCP's current average, the end result is a Romney lead of 0.9%, based on a sample of 12,000+, which would have a 95% confidence interval of +/- 0.9%. So, cheer up. A little. And GOTV!
ReplyDelete(This requires assuming that the surveys are all equally well done, which is definitely not true, but that's the assumption the press has been making.)
Also - crossovers are worth modeling - they can be important, although not as important as they used to be, they're not hard to model. More often than not, they're a net plus for Republicans.
-Expert Witness/Survey Methodologist
You point to one item that I have considered, but ended up rejecting. Using a poll weighting model based on sample size. Larger samples should have more value.
DeleteHowever, my methodology is only designed to "correct" the RCP average, not to fix the problem they introduce through their process.
Looks like the MSM-Beltway Industrial Complex is trying to drag Obama (and the Dem majority in the senate) across the finish line. (shrug) I don't know why anyone would be surprised that many/most of these jokers would be perfectly willing to cook their results to push a political agenda; they do it all the time on non-poll-related stuff. This is of a piece with the "news" coverage for this entire administration...that is, "are you going to believe what we tell you to believe or your own lying eyes?"
ReplyDeleteREAD THIS: http://www.redstate.com/2012/10/31/on-polling-models-skewed-unskewed/
ReplyDeleteLooks like they are still up to it. These were pasted in Iowa local news today.
ReplyDelete----
Obama leads in Iowa 50%-44%, according to NBC News/Wall Street Journal/Marist polls.
The same survey puts the president ahead 49%-46% in Wisconsin, and 49%-47% in New Hampshire.
An NBC/WSJ/Marist poll released Wednesday gave Obama a 5-point lead in Ohio, 50%-45%.
It's time to start ignoring those types of polls. Besides, the NBC/WSJ polls have been terrible.
DeleteI agree, but that's what they are flooding the airwaves with. This is what the LIV sees.
DeleteExcellent way of describing, and pleasant piece
ReplyDeleteof writing to take information on the topic of my presentation topic, which i am going to
present in institution of higher education.
My site: waist to height ratio calculator
It's wonderful that you are getting thoughts from this piece of writing as well as from our discussion made at this time.
ReplyDeletemy blog post; Link Building Service