Feeding Karl Rove a bug
Today’s PEC news clips: USA Today, the Philadelphia Inquirer, the LA Times, Atlantic Monthly, and the Daily Princetonian. Early on Elect...
Predictions, October 5th: President/Senate
The Meta-Analysis, a snapshot of today’s conditions, has taken a remarkably sharp and large downtick for President Obama. This comes with a massive polling release from three Republican-leaning pollsters: Rasmussen, Gravis, and We Ask America. Just think – what are the odds that they would all come out of the gate so fast and all at once? It’s almost like they planned it. I guess the true size of the change will get sorted out over the next 1-2 days.
The InTrade price for President Obama’s re-election has also dropped sharply. However, our long-term prediction is basically unchanged (see the red strike zone below).
Kevin Drum at Mother Jones asked me to come up with a set of November predictions by today. The midst of a political storm is an unusual time to ask about where we’ll land on shore. But OK, here is where things stand today.
First, a little homily.Without an expression of uncertainty, predictions are parlor tricks. It is not enough to express a win probability. Always ask prognosticators what their uncertainties and/or confidence intervals are. Such numbers can give a very clear view of how strongly to weight the prediction.
The November predictions below come with 1-sigma confidence intervals; approximately two-thirds of the time (68% to be exact), the outcome should be within this range. All calculations are done according to the most recently discussed methods, which are linked.
President: Obama 332 EV, Romney 206 EV. The 1-sigma range (68% of outcomes) is Obama 314-347 EV. The 2-sigma (95%) range is 282-347 EV (95%).
Two-candidate vote share: Obama 52.1+/-0.8%, Romney 47.9 +/-0.8%.
Senate: 54 D/I, 46 R. The 1-sigma confidence interval is 53-55 D/I; 2-sigma, 51-56 D/I.
House predictions are coming soon. I am double-checking a few things. As you know, it’s a more difficult prediction to make.
Topics:
I am a frequent reader, but a first time poster. Thanks for your work Dr. Wang. How far do the polls have to swing to Romney in order for him to overtake Obama in the Meta-Analysis? Do you think the new jobs numbers blunt any impact of Obama’s debate performance? Thank you.
By the Meta-Margin. That is how it is defined.
Sam,
About those Republican leaning polls, I thought the same thing that they would all have favorable upticks for MR so fast. Personally, I feel it’s to keep the base from freaking out. Pundits and News Media Outlets have run with the debates of how well MR performed compared to BO to no end. Game changer? Hell no. Overnight success really doesn’t happen overnight. He’s had a failing campaign all along and one day doesn’t in my opinion change a thing. I’ve noticed too that these polls are in the MUST win states for MR. Coincidental? Yes, you are very smart Sam but it will interesting to see as we get closer to the election what else will be done to bolster MR.
“This comes with a massive polling release from three Republican-leaning pollsters: Rasmussen, Gravis, and We Ask America. Just think – what are the odds that they would all come out of the gate so fast and all at once?”
Are you suggesting a conspiracy?
Republicans think that the BLS is BS!
Tapen
last week we had the poll truthers.
this week we have the jobs truthers.
https://twitter.com/msnbc/status/254340236501463040
Wonkblog (Washington Post) has a good discussion of the supposed BLS ‘conspiracy’. In summary: the BLS is set up to prevent politicians from influencing the numbers.
The President lost about 40% of his lead in the meta-margin in ONE day after his disasterous debate performance. It is hard to quantify psychology, but crowds (see Canetti) tend to follow a “strong” father leader. Romney treated Obama like a chastised son and that sort of humiliation, more than the lies he told or the political issues involved, caused this movement. People assume that the President will come out stronger in the second and third debate but in order to do that, he will have to confront Romney and so far nobody has done that in a debate format. How does one perform a statistical analysis on mob psychology and how does this factor into your November prediction.
Our narratives for the present and the past are very male biased. More than half the electorate are women. I am not a woman, but I do know that a vast majority of Asians and Hispanics would not be impressed with a strong macho image. They want someone who understands them and cares for them. Hence, the famous Bill Clinton phrase, “I feel your pain”.
Romney has left himself open for so many comeback lines. “You want to get rid of the big bird because you don’t want to borrow money from China, but you do want to borrow from China so that the oil companies get a subsidy, or you get a bigger account in the Cayman islands?”
I think Sam’s method may be oversensitive to which polling organizations have released data most recently. There a huge jitter in the meta-margin that I find it hard to believe reflects the real state of the race.
Although I’m dubious about all the juggling in the 538 model, one advantage of trying to correct for house effects is that you can reconcile a more recent Rasmussen poll with a less recent PPP poll. Yes, it all comes out in the wash in the end — but if you’re looking for moment-to-moment trends, perhaps there is value in applying those corrections to smooth out that jitter.
Unlike some of the other models, Dr. Wang’s model does not attempt to account for things like mob psychology. The idea is that such things are baked into the polls. By taking a purely statistical approach — using medians of state polls to neutralize outliers — this model provides an accurate statement of where the race stands right now.
Florida is a big ‘if’ right now, seems like 29 EV would play havok with your 1 sigma numbers.
Colorado is looking ugly, there are at least 3 center-right people here at work that were planning on either not voting or going for Johnson, but the debate has convinced them that Romney is presidential and worth their vote (though they admit they could still change their minds back to ‘no vote’)
“Rasmussen, Gravis, and We Ask America– what are the odds?”
Thanks for this post.
Nothing happening in RAND.
https://mmicdata.rand.org/alp/?page=election
In the past, it seemed like like RAND was very sensitive to changes in opinion.
Could the red house effect polls be outliers?
For one thing, the Rand poll asks 1/7 of the participants each day, so it will be a week before we get the total effect of the debate.
No, that’s not true. With that as the rule, a change should be apparent immediately, then last for a week.
What makes this seem fishy to me is that the respondents’ predictions of who was going to win didn’t even move. Maybe I’m just too plugged-in, but wouldn’t we see a big jump in the conventional wisdom about how will win even if that conventional wisdom is false? All the conservatives I’ve seen are ecstatic.
Maybe the new data in the 10/4 numbers is really from 10/3.
…The 10/5 results are out now, and still no big jump in the poll. I guess the relative stability is real, at least for RAND’s sample.
@Matt
there isn’t much movement at 538 either.
the nowcast is still <5% for Romneys win prob, and the forecast is ~15% for Romney's win prob.
I doubt the debate bounce is going to get him over the 20% threshold where betting Romney becomes cost-viable.
332/206 exactly the same as electoral-vote.com
This addresses questions I wasn’t sure how to formulate. Much appreciated.
I was shocked to see how far the meta-margin changed between 8 AM today and then when I checked in at noon. I was wondering what the cause was. Do these polling firms typically release polls on Friday, or was this an unusual time for them to release a set of polls. The obvious factor for people to jump on to explain this change was the debate, yet I find it hard to believe that the debate could lead to this large a rapid change. It is certainly tempting to wonder about the coordination and timing….but then I do not want to be a “tin foil hat” type who sees conspiracies everywhere, including a good jobs report (all of that has me wondering about the prevalence rate of paranoia in the U.S.). In any event, if this election is at (I think these are words you have used before) a thermostat set point, I would think the next few days would see a slow rise back towards 5 or 6 (a regression to the mean type effect). It will be interesting to watch.
As to why there was a big move in the Meta-Margin, it was in a position where a big downside move was possible, if not likely. Yesterday there were 275 EV at at least O+7, with not much of a floor underneath (15 EV at O+4, and Virginia and Florida below that). Two polls moved Ohio from O+8 to O+2.5, and another dropped Nevada from O+7 to O+2. Those two states (24 EV) were the reason for most of the change in the Meta-Margin.
As far as the timing for the poll releases, it only makes sense for pro-Republican pollsters to rush into the field after Romney “won” the debate on Wednesday. My guess it that we’re seeing nothing more than that.
Good point about the conservative-leaning pollsters rushing to be first in reporting a bounce in MR’s favor. They wanto believe they have a hope of winning.
Sam,
I’ve been following your site for some time and finally decided to comment. As someone who has worked in politics, international relations, and have taught politics as an adjunct, I find your analysis and predictions quite refreshing and straight forward.
My question is as follows: does your Meta-margin model allow for changes/variation in polling methodology, especially when it’s the same pollster? My observation, as of late, is that many of the pollsters are seemingly self serving by changing their methodology.
You’ve done a very good job, as others, in outlining the variables from the standard polling as compared to robocalls, yet I notice that in many recent polls I’ve seen multiple instances when a pollster has not disclosed methodology or has changed from their previous approach; in some cases, seemingly going ‘backward’ and utilizing a less accurate methodology.
Looking forward to your reply. In the meantime, keep up the great work!
Thank you for writing. I have set up the code to be purposely blind as to pollster identity. No house effects and so on. If likely-voter and registered-voter are available from the same pollster, we take likely-voter data.
We considered having it accept at most one poll per state for a given pollster, which would be a justifiable move. However, at this point it would require redrawing the graph, something I don’t want to do midseason. In practice, pollster duplication is a major factor only in the early part of the campaign.
Dr. Wang, Intrade is back up.
[W]ithin the first half-hour of the debate on Wednesday, Obama’s InTrade odds fell from 71% to 67%, just on the back of his weak performance. And then, this morning, they spiked right back up again, from 65% to 71%.
Eric Fisher, please keep your Barry Manilow statistics to yourself. They have no place here.
Dr. Wangs forecast is founded on mathematics and the Central Limit Theorem and the Holy Law of Large Numbers.
He has never said the polls are “stable”.
Au contraire, he has said the polls have a-plenty of jitter, but they will converge.
And empirically, that is what is happening.
I think there actually is an important assumption embedded in his November prediction (call it a hypothesis, since he’s stated it more or less explicitly).
It’s that there’s some kind of hidden “setpoint” that the polls will continue to meander around, instead of the campaign being more like a drunkard’s walk in which only the very recent state matters for the future.
I think that whether this is the case is still an open question. Some previous presidential elections do seem to behave this way. Personally, I wonder if it isn’t more like a drunkard’s walk hemmed in by fences–that is, there isn’t so much a setpoint as set maxima and minima for one candidate’s advantage, between which it can wander more or less freely.
I suppose that’s more or less what people are thinking of when they model the electorate as partisans and undecided voters.
Or maybe its random walk with fences for the presidential and there are local maxima and minima for states like in simulated annealing or genetic algorithms.
I wouldn’t worry so much. The release appears coordinated, yes, but keep in mind that Florida/Virginia/NC were not very heavily polled as compared to Ohio in the past two weeks. AFAIK, most polling websites are sensitive to long lags in between polls. When polling data resumes for those three states by other firms, it will go back to Obama’s advantage. Gravis/WAA have released oddball results in the past weeks, and Rasmussen is just Rasmussen.
I dunno. PPP has Obama only +2 in Wisconsin now, which is a big drop from +7 a few weeks ago, and Romney is up to +1 favorable over unfavorable.
So is the projection mostly one-sided because it is based on poll data, which has been largely better for Obama lately?
Intrade has some real money in it, enough to look for sucker bets.
Sam, thanks for your work. You noted ahead of the debate that a move like this would be possible. Didn’t we see the same exact thing in 2004, with Kerry, after the first debate? It might be interesting to post the Kerry EV chart and the Romney EV charts together to show that 1st debate pop. Thanks
Sir,
I think someone is trying to manipulate your model, and they have succeeded.
How can the debate which was less than 48 hours ago have such a drastic effect
on your MR. I could see the MM going down a little over time but in two days.
Not every voter watched the debate and not ever voter had the same reaction for it to reduce the MM by nearly 50%
Dr. Gee, consider how far-fetched your speculation about manipulation is. Pollsters have far bigger fish to fry than PEC’s model. Much as we here admire it, few in the media and very, very few in the public are following it or are even aware of it.
Now those R-oriented pollsters might have jumped into the field, and released results quickly, to try to manipulate pubic perception. And that could have an indirect effect short–term effect on PEC’s Meta-Margin. But if Sam is right — and I’m betting he is — the effect of these three quickie R-oriented polls should fade out fairly soon.
The proof is in the pudding, i.e. the numbers. That is, we’ll see soon enough.
Simpler than that: a large volume of small sample polls (some of 300 LV or less) creates a stronger narrative than a small volume of large sample polls.
If certain partisan pollsters work together now to exaggerate the debate effect, within 8 days wouldn’t the other guy seem to sharply gain popularity? Is this what’s being said here?
The second day of data in the RAND poll since the debate suggests no change whatsoever in overall movement between the candidates. Both the ‘swapping’ graph and the overall vote projection say the same thing.
I hadn’t looked at the RAND poll before.
The “election forecast” graph suggests that the people they called Friday swung toward Obama almost as much as those they called Thursday had swung to Romney. But the “shift between candidates” graph looks like good news for Romney both days. (One explanation that occurs to me is that Friday’s sample said they were more likely to vote that they had a week earlier. And actually, there it is! The “intention to vote” graph of Friday shows an uptick for Obama and a big downtick for Romney.)
This is a brilliant poll.
Are these 3500 people polled weekly on the election only–the 3 questions shown–or are they collecting “American Life” data from the same people in a larger weekly questionaire?
And how will they statistically deal with people who vote early?
I tried their FAQ; it was not aimed at me.
Yeah, there’s been some small movement toward Romney ever since Monday or Tuesday. Basically, Obama’s long post-DNC rise finally leveled out and started to subside a little. But there wasn’t much change in that situation after Wednesday; the debate didn’t make much of a difference.
Keep in mind that each ‘dot’ on the RAND poll represents the rolling average for the last seven days, not the shift in support for a single day. So because the ‘gap’ between the candidates hadn’t really shifted fron pre debate levels, it seems to me that RAND is telling us that nothing has changed. But as the split between them is about 0.25%, with a 500 sample daily base, the diffence is literally only 1 or 2 people changing their minds – so like the text says, it’s very noisy. But, what it is not showing is the type of free fall that appeared in the meta margin after the floodgates opened from Rasmussen et al…
You’re dreaming. Romney will win Florida, Ohio, and Virginia, which means he will win the presidency. I’m eight for eight since 1980, and this should make it nine for nine.
@jvnch
Romney winning those states in addition to states where he is favored would only net him 266 electoral votes. That is a losing number.
If he won all those states he’d probably be in a national position where he’d be winning Nevada or Iowa too, which would be a win. But I’m far from convinced he’s there.
The proof’s too large to fit in the margin, huh?
Normally I moderate trash-talking that lacks any evidence to back it up. There is something funny about this one. I’ll save it for November 7th.
I am a very strong Obama supporter and think a Romney victory will be a catastrophe for the country. Having said that I believe the President’s performance in the debate was the worst I have ever seen and I’ve watched them all since the first Kennedy-Nixon debate. Nobody remembers what was said in that debate, the only thing recalled was Nixon’s haggard appearance in contrast to Kennedy’s youthful vitality. Obama looked terrible and his last two minute speech was basically a concession speech. I think the entire Meta-margin for Obama will be gone by the middle of next week. Hopefully he can recover but people who think he can automatically turn it on in the second debate are in for a rude awakening.
Sam, a hypothetical question…
Let’s say that Romney or Obama “win” by 1–5 electoral votes. Is there any chance of sufficient number of Electoral Collegiates being disloyal to swing the election the other way? And if so, could they be stopped, their vote overturned, or is their vote final?
Olav:
See “Faithless Elector” article at Wikipedia:
http://en.wikipedia.org/wiki/Faithless_elector
I believe it is legally possible, and that there’s no recourse if it happens, beyond the state punishing the individual elector.
However, it is extremely unlikely.
Dr. Wang,
I was wondering if you could speak to how to interpret the most current polling data in a tracking poll. A lot of folks think that it’s difficult to understand how any new data impacts the tracking poll until later in the cycle, because of the older poll data rolling off the tracker. However, you seem to suggest the opposite when you wrote the following yesterday responding to the claim that the full impact of the debates won’t be known for a week: “No, that’s not true. With that as the rule, a change should be apparent immediately, then last for a week.”
Can you clarify how best to understand tracking polls as a snapshot of the moment? Thanks!
That is what I thought — that there is no recourse for changing the result.
I read that Wikipedia article some months ago, and was most surprised. For it did, indeed, appear that such a fate of the nation had not been made a hypothetical impossibility.
I do realize it is “extremely unlikely” – but this hardly seems reassuring. For it appears that we are living in an extremely unlikely day and age.
Who would have thought, for instance, a decade or so ago, that Presidential appointments would be blocked time and time again — not because senators consider the person in question to be unqualified, but as a systematic partisan political tactic?
Hi Sam, I’m an amateur stats hobbyist and I’ve compiled an apples-to-apples comparison of poll-based EV aggregators using your exact method on this site:
http://evmodcast.wordpress.com
Do take a look!
Sam,
Came across this and thought this was an interesting read. Passing it on. Don’t know how much weight it will get but who knows. Also, why are my comments still waiting for moderation?
http://www.dailykos.com/story/2012/10/07/1141340/-Gravis-Marketing-Exposed-As-A-Fraud-via-Democratic-Underground
That’s a really interesting article Iseefuture. It makes you wonder if Gravis actually does any polling at all, or if they just make stuff up and put in a press release.
Moderation does not satisfy everyone. Sorry. An alternative is to have no comments section.
I was suspicious of Gravis Marketing when I first saw it appear on RCP. Here’s an Angry Bear discussion of the firm:
http://www.angrybearblog.com/2012/09/gravis-marketing-polls-and-reporting-ii.html
So I Googled this phrase:
“doug kaplan” + florida + republican,
and the first hit was this:
Doug Kaplan – Orange County Orlando Florida Republican Party …
http://www.meetup.com/republican-626/members/14406273/
Altamonte Springs FL – Member
Ron Redlich Meetup Organizer 407-886-8460 webmaster@usavictory.com Local Clubs and Resources = http://republican.meetup.com/626/about/
When I clicked on it, I got an error message. (That’s still the case, though it’s now the second Google hit.) That link was scrubbed, probably because the same url was posted in a comment on the Angry Bear blog post.
Dr. Wang:
I also came across the challenge to Gravis Marketing on Democratic Underground and want to bring it to your attention. This is not the standard complaint about biased polls – I think they have a good case that Gravis is a fraud and presenting made-up numbers.
If this is so, than a statistical detective might be able to test whether their poll numbers look legitimate or not.
I did my own mini-test: I pulled up a poll from the Gravis web site, that gave a number of 3-significant-figure percent values. I took only the last digit of each poll number (which should have a uniform distribution) and it looked strange. I counted the occurrences of odd and even digits (32 odd, 10 even) and applied the Monobit frequency test, and found that this distribution has a probability of just .001. Monobit rejects data as nonrandom if the probability is below .01.
It is true that I ‘polluted’ this test because I first noticed the non-random pattern and then picked a test that was geared to the non-randomness that I found. Also, Monobit requires more than 100 entries; I just had 42.
But in my defense: this was a quick first effort, from a polling sample picked at random, and a more sophisticated analysis would probably identify an even higher non-randomness. Perhaps you or your associates can add to it?
DU article:
http://www.democraticunderground.com/10021489250
Gravis poll used for evaluation:
http://www.realclearpolitics.com/docs/2012/Gravis_FL_1001.pdf
Last significant figure of each entry:
307173 94359 07797 0559
37 0394517 4913950 532372.
Frequency of occurrence:
0. =5
1. =3.
2. =2.
3. =8.
4. =3.
5. = 6
6. = 0.
7. = 8.
8. = 0.
9. = 7.
(end)
As Daniel Patrick Moynihan, the esteemed senator from New York said:
“Everyone is entitled to his own opinion, but not to his own facts.”
We have long seen how some politicians make up their own facts. The new development is that the press is either unwilling or appears to lack the resources (of integrity, intellectual capacity, time and analysis) to arrest them with sufficient force when they do so.
Dr Wang, are we now seeing that pollsters (and perhaps more than one) are even inventing the opinions of the electorate, as they please?
Perhaps it’s not importunate with a second Moynihan quote:
“The central conservative truth is that it is culture, not politics, that determines the success of a society. The central liberal truth is that politics can change a culture and save it from itself.”
I do not think there is enough evidence for fraud in any of these cases, even Gravis. Anyway, one could get results like that simply by not stratifying.
On further reflection, and looking at some of the other results that have been trickling out, it appears to me that Romney definitely helped himself in the polls, but not to the extent that we should expect the whole country flipping to Romney in the near term.
While Rasmussen, Gravis and WAA all have Republican house effects, they’re not big enough to explain the huge shift in the Friday swing-state poll releases all by themselves. Especially not WAA, whose house effect the Votamatic site calculated as being very close to the average pollster.
It’s possible that their results are completely faked and they’re carefully husbanding their big Romney shifts for moments like this to control the narrative.
But I think it’s more likely that they got this result by the simple expedient of doing a flash robopoll in the hours immediately after the debate, when its direct effect on viewers ought to be strongest. With conventional pollsters holding off and taking surveys over a longer period, that means their startling flash estimate gets to dominate media talk about state polling for several days.