A final unskewing
Mother Nature is our best teacher and the only one who is always right. – Viktor Hamburger, biologist In yesterday’s Los Angeles Times...
Welcome NPR listeners!
Welcome, Science Friday listeners! (I’m on today, 2:15-2:40pm Eastern with Flora Lichtman and Nate Silver.)
Update: That was fun. We got to get into the basics: how news media love to be contrary, the weirdest polls (hello Gallup!) get the most attention, and aggregating polls is really the best approach. If you missed it, you can listen here (NPR).
and the Meta-Margin…
Topics:
Sam, I have a question and you might have already answered this:
What’s the effect of GOTV/ground game and a potential change on under/overperforming polls? I’m sure it may not make a different nationally, but does it on a series of states? (Early voting would seem the most obvious — a “tied” poll result isn’t tied when one group’s banked a lot of votes and the other depends on the relative uncertainty of their supporters going out to vote 15-20 days later.)
Just curious. Has there been a study on this?
My impression is that globally, GOTV cancels out. The polls are our best guide. I assumed otherwise in 2004, to my regret. That year, polls were dead on.
As for early voting, it’s good for D morale but again, it’s baked into the polls. Pollsters ask that. They are almost entirely not dummies!
Mr.Wang, your response to Chris R’s question compelled me to post for the first time. Bush’s superior GOTV effort in 04 allowed him to overperform in Florida where most polls had an almost exact tie. RCP had Bush by 0.6%. He ended up winning it by 5 points. You could say that his GOTV effort helped him quite a bit there.
Mr Wang,
Why don’t you publish the map(s) that correspond(s) to the median result?
For example: Obama 293.
One possibility for that result is Obama wins VA, CO, WI, and OH but loses NH, NV, and FL.
Another is Obama wins VA, CO, WI, and OH but loses NH, IA and FL.
Another is Obama wins NV, CO, OH, IA, NH, and VA, but loses FL and WI.
Another is that Obama wins NH, CO, WI, and FL, but loses OH, VA, NV and IA.
(Maybe there are other combinations that result in 293, but it seems that we’re getting further away from the realm of possibility here [i.e., going into outcomes with lower probabilities]).
It would be useful to know which one of the outcomes is the median, which one has the highest probability, and which one has the lowest probability.
Is this a choice on your part, not to show the median outcome map(s), because the median is what ultimately matters and not the scenarios that produces it? Or am I misunderstanding something about how you calculate the median outcome via the Meta-Analysis polynomial: is your median somehow not based on the probability of the 50 state-by-state outcomes in a way that shows a specific outcome map or maps that correspond to the median outcome?
Dr Wang, so early voting, which seems to be trending D this election and very high, is already present in the polls? I am just trying to get an idea of how you track this. Thanks
I believe he means that, when a survey is conducted, you are first asked if you have already voted. If so, you are asked who you voted for. I believe these people are placed in “likely voter” counts because, as we’ve seen, people have the tendency to lie about whether or not they voted.
Sam, the early voting database at GMU (http://elections.gmu.edu/early_vote_2012.html) has some great information on voting statistics. It shows that states like Iowa have had ~20% of the number of votes cast in 2008 already. Other states like Ohio, Florida, and Virginia are quickly moving to 10% of the 2008 vote count, meaning there can be some predictive power here. Granted, party affiliation or primary participation is not a perfect indicator of their actual vote but it does provide something. Are you thinking of making a short math post on this that is completely separate from the MM and PEC poll aggregation?
I wasn’t planning that. I believe that early voting is adequately addressed by pollsters, and is therefore “upstream” of the polling measurement. In other words, polls tell us most of what we want to know about the results of early voting.
The Sorkinesque conspiracy theory is clever but flawed, IMO. A blowout Debate 1 for Obama might have led to a 5-t0-6-point win nationally, which should have been enough to hold the Senate (with at least a 53-47 seat edge) and might even have flipped the House. Why jeopardize a shot at winning all three just to prevent big money going down-ticket to Scott Brown, Thompson, Akin, Allen et al?
It also assumes that they think the damage to coattail effects would be smaller than the second-order effect through the motion of campaign funds. I don’t see an intelligent campaign organization assuming that.
Oh, I overlooked the hypothetical of Romney money getting diverted to marginal House seats. I can’t buy that either. It seems to me that if Obama’s election were a virtual lock after a knockout Debate 1, his campaign would be free to spend money on House races, too.
I don’t think there was anyway Debate 1 could have been a knockout for Obama. The format was very favorable to Romney — vague, finance-oriented questions (none on social issues), weak moderator, expectations way lower than his actual debate skill. My guess is that the 1st debate is going to pretty much balance out the 47% tape in the end — each being events that angered and disappointed the candidate’s supporters, and inspired glee in their opponents.
Diabolically clever, yes. And the debate #1 fail would have to be of a large enough magnitude that not just media pundits but actual money guys got sucked in, and took their eyes off the downticket races.
Pretend to have a broken wing, rope-a-dope, whatever. I am guessing they have access to detailed internal polling and focus groups and whatnot before pulling something like this.
And of course, big brass ones, from years of Chicago politics.
If Obama wins by 10 EV instead of 30, but manages to get 53 – 54 in the Senate, and a couple of extra House seats, then it will have worked.
Love Amitabh’s conspiracy theory, but I think Debate1 was a genuine bounce with even a small effect on African Americans.
Here is my favorite RAND characteristics graph which I like to call the Demographic Doom of the All-White GOP.
https://mmicdata.rand.org/alp/?page=election#election-forecast
That would have been an incredible risk, and Team O couldnt be be sure of the reaction. Obama seems pretty risk averse to me.
pechmerle, I believe Amitabh is seeding our brains with strangelets. Only theoretical quantum physicists can reconcile brain strangelets.
And since only Q-physics theorists can reconcile brain strangelets, the rest of us will develop expanding mini-black-holes in our brain tissue, leading to the resolution (in the physicists favor) of the Most Important Argument of All Time, which came first, physics or mathematics, and the eventual re-allocation of all university curricula and research funding to physics departments across the land.
We know there will have to be considerable fraction mixed ballots for a Tester or Heitkamp win. Obama does not have to have >50% approval in a region for a Dem win.
What if internal polling suggested that once you get past 225 Dem House seats, the additional fringe seats were less positively correlated with BHO’s approval or enthusiasm scores? The Chicago goal may not just to take back the Tea Party freshmen or redistricted seats, but scoop out some veterans too.
The flip side is Obama’s own fundraising. What if Team O saw a downslide in their rake, and figured a more competitive race would bump it up?
DAWG, Sam also had a post on how after debate#1 the Senate polling had not dropped like the presidential polls.
Amitabh, are you also part of a plot for a CERN-created black hole will swallow the earth?
(Are they putting something new in the water at Rutgers this week?)
“Amitabh, are you also part of a plot for a CERN-created black hole will swallow the earth?”
It already has, which explains our predicament, at least in part.
wheeler’s cat: I’m an experimentalist, so my brain is as confused as yours (and your French is better than mine).
I confess I thought of the “Obama throwing debate#1” more as an Aaron Sorkin West Wing type of TV plot than a Real Thing, and threw it out there because the polls are doing wacky things.
And I wanted to see what you guys would make of it. Wasn’t disappointed.
pechmerle: How did you know about the black holes? You are Dan Brown, aren’t you?
While pollsters ask about early voting, I seem to be reading some pollsters themselves talking about how the increased prevalence/importance of early voting is causing for them:
1) Their estimates of how much of the voting population have voted early are not in line with what the Secretaries of state report receiving.
2) They seem genuinely unsure as to how to treat those who claim to have voted early. They may just count them as a for sure LV, but the pollsters themselves admit that they are in uncharted waters, and are very unsure as to whether this is a good idea.
I’m not sure as to whether simply taking the median poll can solve for this, if there if this is indeed a systematic error, and is shifting a majority of the state polls where early voting is underway.
Note, I have no insight into whether this is really a problem. It could for example turn out that while large numbers incorrectly state that they have already voted, that making such a claim is a good indicator that someone will eventually cast a ballot.
“1) Their estimates of how much of the voting population have voted early are not in line with what the Secretaries of state report receiving.”
The common error with that assumption is that there would be 100% turnout. If the SoS says that 10% have voted early, and the poll says it’s 20%, then it’s not that far off, since only about 60-65% of registered voters actually vote in presidential elections. Like any other crosstab, there is also a margin of error to account for as well.
“The common error with that assumption is that there would be 100% turnout. If the SoS says that 10% have voted early, and the poll says it’s 20%, then it’s not that far off, since only about 60-65% of registered voters actually vote in presidential elections. ”
The secretaries of state are generally reporting absolute numbers of ballots; the percentages that you see calculated are usually percentages of the total 2008 vote, not the total number of registered voters. It is of course possible that turnout will be lighter in 2012 than 2008 which would increase the percentage of the total vote that the current early vote represents, but I don’t see any reason to think that would be the case in the swing states such as Iowa and Ohio, given that this is a much more competitive election with heavy turnout operations on both sides.
I also don’t think a counting/mailing delay can explain the discrepancy, as for example the Iowa SOS gives updates every day and if you like at the rate of ballots being added to the tally each day there would have to be a delay of many days, maybe even a couple of weeks, between the mailing of a ballot and the SOS including it in the daily update for the numbers to add up. This does not seem likely.
Other suggestions, like maybe people claim to have early voted if they requested a ballot or even just plan to, and these are in fact likely voters, seem more plausible. But the size of the discrepancy cis certainly eyebrow-raising.
what’s the prob? Of _course_ more people report voting than ballots received. Theyre IN THE MAIL
I guess my point is mainly, that, from what I understand, the PEC model has baked in the idea that LV models are preferable. However, if LV models for certain swing states have a persistent, unidirectional error now (caused by how they treat those who claim to have voted early) that they have not had historically, it might be a problem worth looking into after this election. I can’t think of how you would investigate it prior to Nov 7.
Heres something interesting I noticed yesterday. Briefly 538 went back above 70% win prob for Obama. That triggered a chorus of howls and wails from the poll truthers on twitter.
for example, http://dailycaller.com/2012/10/19/nate-silver-vs-the-world/ and http://dailycaller.com/2012/10/19/nate-silver-vs-the-world-ctd/
and before the scheduled 10:00 pm update, Obama’s win prob miraculously declined to 67.9%.
I think 70% represents some freakout gate for the poll truthers.
That’s my conspiracy theory– that (at this point, 3 weeks out) Nate has to fiddle with the model to derail conservative freak outs.
freakout threshold i should have said.
wheelers cat, youre a smart person. Surely you can think of a more useful way to channel your concern than seeking a freakout threshhold, funny as the phrase is.
What Schrodinger’s cat was to quantum phenomena, Wheeler’s cat would be to gravitation and general relativity.
After John Archibald Wheeler, 1911-2008.
Apologies if it’s some other Wheeler you refer to.
no Amitabh cher, you are exactly right.
Ms. Jay, legend has it that Einstein and Wheeler lived in the same neighborhood, and Dr. John Wheeler’s cat was passionately fond of Albert Einstein’s house.
So Wheeler’s cat could be found either at Wheeler’s house or Einstein’s house.
Thus the aphorism, Wheeler’s cat…much luckier than Schrodinger’s cat.
And hopefully lucky for the reconciliation of loop gravity as well.
Must admit I am breathing a sigh of relief now that the Meta-Margin has climbed above 2%. Senate aggregators that consistently predict continued Democratic control are also a good, calming tonic.
I’m looking forward to renewed analysis of the down-ticket races. Do we have more precise, poll-base information on what is happening in the House?
Forget early voting (unless you can do so, in which case I suggest you do so right now):
1) Nate Silver in 2008-9 wrote this post suggesting a 10% disparity in voter contact means a 2-3% overperformance from polls (http://www.fivethirtyeight.com/2008/11/contact-gap-proof-of-importance-of.html),
2) WaPo suggests that this election does have a 10% disparity in voter contact in this election (Please note — I haven’t seen a lot of pollsters asking the ? and PPP’s poll of NC showed no disparity): http://www.washingtonpost.com/blogs/the-fix/wp/2012/10/16/can-a-strong-ground-game-help-obama-stop-romneys-momentum/
Look, I have no idea, but I figured I’d ask Sam — he 1) is willing to debunk myths, even half-baked myths from commenters like me — and this is a question from an amateur, not an expert and , 2) actually may have looked at this or know where to look in an academic study.
PS: The IA RCP average in 08 (the last time we saw significant early voting) showed the pollsters *overestimated* Obama’s support, I think.
Thanks once again for answering and participating in comments section, Sam. That is a very remarkable credit to you.
I think we should totally ignore any analysis of what the early voting means and stick to the poll results. The SOS in most states (certainly in Illinois where I live) are inaccurate and behind real time in their counts. I would more trust the median statistical findings of all the polling organizations in terms of how they incorporate early voting into their polls. We simply do not know what the early polling means and who it favors, it is essentially anecdotal evidence while polling taken as an aggregate is much more scientific.
I must say that things are looking better for the President. Since I tend to focus on psychological factors, the moment in the 2nd debate when Obama turned to Romney and face-to-face rebuked him with a cold seething anger over Libya and reminded him who was the commander-in-chief was a decisive moment. And of course, Romney being the putz he is couldn’t resist plunging ahead and fell into a classic military trap. “Proceed governor” was straight from “The Art of War.”
Here is a question for Dr. Wang and/or his wise followers. I understand that the people who respond to pollsters are a very small percentage of the people who are polled. Thus, they are different from most people. If they are unrepresentative of the voting population in that respect, why would we think they are representative of the voting population on the question of how they would vote?
George F, Poll- Au contraire, analysts DO find a huge correlation between polls and voting. If there were none, Sam wouldn’t waste his time. He’s a scientist, and came closer than anyone to predicting the exact outcome of the 2008 election. You’re in the right place.
I just listened to the NPR interview and resisted an urge to turn it off because of the irritating voice of the moderator, who sounded as if she’s 12 years old.
That’s sad. My own voice is like that. I’m a soprano [lowercase “s”]. A possible solution: hang out with more women. We’re smarter than we sound to some.
Ms. Jay,
I’ve hung out with thousands of women in my 78 years and still say that NPR moderator sounds like a little girl. I didn’t say anything about how smart she may be — that wasn’t my point.
I apologize, Dennis.
Anyone else feel like Dennis did about the interviewer?
I observe that except for the second (in time) Obama low in the Meta Margin, each subsequent low has dipped lower still. Do you detect a trend here, and if so, what do you make of it?
Obama’s highs get higher as well, though of course we have no idea how high his current upswing will go before the election.
If I had to guess, I’d say that the numbers are just getting more volatile as more people pay attention to the presidential campaign, and people who formerly identified as undecided start taking sides. For many voters, the debates were really the first time this cycle that they made an effort to compare the candidates.
The only high that is considerably higher than the previous ones came after the DNC. The others were pretty much in the same ballpark. But the lows have trended substantially lower. My guess is that the present climb will peak at a point that is lower than the average of the previous ones. After that?
I don’t see a trend there, but I wonder if your observation has something to do with the difference between local minimums (or maximums) and global minimums (and maximums). There are an awful lot of local minima and maxima that (to my eyes) don’t follow the trend that you’re observing. It seems to me that you’re looking at changes to the global minimum and maximum observed- the problem is, that as time moves forward, the observed global minimum and global maximum will decrease (and increase) until the election date.
There’s not much “after that” to go. We’re three weeks out, and if we’re really analyzing this as a trend, it’s worth noting that the frequency of the swings doesn’t seem to be increasing.
I think it makes more sense to analyze these swings as reactions to specific events. The first time Romney gained a lot, it was when he’d just clinched the Republican nomination and Santorum, Gingrich, etc. supporters were reconciling themselves to him. The second time, it was when Ryan was announced, giving the ticket lots of conservative cred; the third time was after the first debate, when Obama made really his only major mistake of the entire campaign.
Note also, the relative sizes of the dips are somewhat dependent on aggregation procedures. On electoral-vote.com, Romney’s clinching-the-nomination bounce looks bigger than the Ryan bounce, and almost as big as the debate #1 bounce.
I’m not really seeing a conclusive trend line here. If I squint, I can see the the previous three highs getting higher and the previous three lows getting lower, but in each case, the fourth data point was higher and lower respectively than the third data point. I would also note that there is not much time left in the race, and additional squinting will reveal that the large waves (peak to peak or trough to trough) will either have to increase in frequency or, if it follows the pattern so far established, the rise will just be about peaking or just past peaking when the election occurs.
And Matt’s second comment wasn’t posted when I went to add my comment so I ended up more or less saying what he said.
Sam, would you post the data points on an excel sheet?
Matt.
the swings could be oscillation…..but could also be scalable fractals.
Interesting TPM article on the early-voting trends:
http://2012.talkingpointsmemo.com/2012/10/republicans-boast-early-vote-gains—-but-so-do-democrats.php?ref=fpa
Republicans are touting the smaller D-R gap in early voting compared to 2008, but since the party identification in these counts is based on primary voting, the absence of a contested Democratic primary is probably bringing down the “Democratic” numbers.
Hi, can someone help out a poor physicist who can’t count this morning? I’m trying to reconcile the RCP map and the PEC total above. If I flip CO to blue and assign NH to Obama (though both RCP and PEC show it as pure tossup) I get to 290-248 at RCP: http://www.realclearpolitics.com/epolls/2012/president/obama_vs_romney_create_your_own_electoral_college_map.html. But I’m still 7 short to get to 297, and I don’t see a sensible state to flip. Is there a chance that the PEC count uses an old census? Or is the median estimator not exactly the most likely current value? I thought there was no covariance in this model.
Never mind, of course the PEC estimate is doing something like delta_i = poll_difference_i/sigma_i and summing over phi(delta_i)*EV_i. So when Obama was leading slightly in FL and VA the RCP # was an overestimate wrt the PEC number; now it’s an underestimate.
I guess I’d worry about sensitivity to estimating the sigma on each polling mean if I were to try to do this myself.
There is a link on the sidebar that shows the exact formula. Also matlab code.
I love that youre “a poor physicist who can’t count this morning”. My son’s a physicist who’d never pretend to appear that way. There are wonderful create-your-own maps here too. Look under the daily map and select one of 3 links. IClick on a state to change it. First one you click firms up the whole picture into EV totals… But you can keep changing things and RVs add themselves because face it, half your mind is stuck on some earlier, perfect calculation.
I had java added just for PEC and learned t use itunes to listen to Sam on NPR. Just not “entertainment” oriented, because info and books are “all that”.
Sam,
Why not provide someplace prominent the key number, which is the probability of Obama winning if the vote were held today? (That is, the integral of the distribution to the right of 270). This is a number people want to see! It deserves to be in the bar at the top and/or on the distribution.
E.J.
I think Sam justified not providing it earlier in the year on the basis that, in this calculation, it’d almost always be at or near 100%. But it would have actually been interesting to watch through the post-debate-1 saga.
Agreed. This is the most important # to me without a doubt.
Yes, he mentioned that while his long-term estimate was about 88%, the win probability on any given day if the election were held on that day would be almost always >95%. I agree that it would be interesting to see to what extent this estimate may have dipped during Obama’s low after the 1st debate.
Glad to see Matt & Pat’s comments. I was using 1 in 9 chance for Romney, but my efforts to publicize this haved been Scrooged by my ongoing migraine [earned during a successful campaign to create & spread a paypal account for the defunded Ryanized Ohio soup kitchen– google shows over 3100 links, and MANY people are donating! :D] Long story short, my puppy-based infographic is showing and in catch up now reporting a 1 TO 9 chance for a Romney victory. But these figures here have me less worried about spreading disinfo. ::whew: Head really hurts.
I just read the FAQ about how you do your analysis and while it mostly makes a lot of sense to me… I do wonder about the possibility of poll flooding throwing off your #s.
For example… what if a bunch of right wing pollsters flood the market (as I believe started happening after the first debate) aren’t the # of these biased polls going to bias the median poll result?
I am confused. How can the MM be 2.2% when the Ohio poll shows a tie and a Florida poll shows +1 O? What am I missing here?
Click on OH and FL in the right column above on this page and it takes you to the Pollster.com polls for each state. Dr. Wang’s averages are medians of the last week’s worth of poll results. Gravis’s new OH tie and SurveyUSA’s FL O+1 are considered in the averages.
So jitter is expected to be ongoing?
Today’s breakout on RAND [Election Forecast, by Race-Ethnicity] is pretty interesting. And it’s so stable! Almost like regular polls are just measuring noise.
Anyone:
From all the watching I do I should know this, but how high does the meta-margin need to be for the whole yellow zone to be above the red line, and what are the odds of that happening soon?
Can anybody walk me back through what Sam has said about the influence of voter suppression, intimidation, deregistration, and rogie voting machines? For example a voter may poll as very likely having no idea their registration has been done away with. With so many states varying by low percent, this scares me.
rogue not rogie
For your first question, I don’t believe that can ever happen. Here’s Sam’s explanation:
“This plot now contains short-term predictions for the November 6th election. The calculation is explained here. The red zone is a “strike zone” showing the 68% confidence interval of probable outcomes, based on random movement from today’s polls.”
“The yellow zone is a “watch zone” that shows a combination of the 95% random-movement confidence interval and the 95% gray-zone confidence interval. The November outcome is nearly certain to be within this range.”
It’s nearly certain to be in the range of the yellow but the yellow moves just as the headline prediction moves, in accordance with polls
I was curious as to why Sam’s model seems to be the most bullish on Obama. It’s interesting the difference between yourself and Nate on Obama’s chances: 90 vs 70. Both of you tend to be more bullish on Obama than the punditry and even other aggregations such as RCP or Pollster. I would say that Charlie Cook’s latest seems to be fairly bullish on Obama in his latest release as well.
In any case I am particularly curious as to why your model is more favorable to Obama than Nate’s. Is it just a reflection of the better polling Obama has in the states than he does nationally?
“Bullish” suggests preference, but it’s not that.dont forget that Sam’s algorithms gave him an EXTEMELY precise read on 2008. Maybe he’s the most motivated to be accurate rather than “fair and balanced”
Probably a fluke of how his polls are handled: as far as I know, it’s the median of the last week of polling. Nate Silver isn’t very transparent in how his model works, but there is a lot of “poll weighting” and it’s not clear how far back his polls are averaged (does he even median?). PEC’s website is the best because it doesn’t “correct” anything.
Having been directed to the sidebar explanation, I see that the presidential predictor at least starts with a Bayesian prior of the average of the proceeding months. This is going to make the model (if I understand what it’s doing) a bit loath to leave the range where Obama was doing well. Another reason estimates here might be relatively bullish is the rejection of outliers through the median procedure – if the RCP results are driven by conservative polls with significant house effects (e.g. Gravis) this would reduce bias. I think 538 handles the house effect by weighting pollsters by quality, so it would also be less sensitive than RCP to such bias.
Thanks for the ideas. By bullish I really just meant that he has a higher probability than other people are showing. Since I am an Obama supporter, I consider that good news, hence “bullish.” It wasn’t meant to be a comment on preference other than my own, lol.
Another point about 538 is that its model uses economic data, and that it considers the likelihood that state A and B vote together so it will have a different sensitivity to a given poll than the punditry. Maybe for example a lot of swing states being positive for Obama mean it’s more likely that a state marginally for Romney is really blue.
Yes Rilke an, Silver is using weighting which Wang [said “Wong”] has given up, and which Silver himself admits is poor methodology, adding noise to signal.
This year proper weighting re the economy -if there were such a thing- would be a real brain twister.
I’m a refugee from 538 and I diswant his noise.
Yes, I meant diswant, and don’t care that it isn’t a word…yet.
“Silver himself admits is poor methodology, adding noise to signal.”
Link?
It’s just a different but obviously relevant metric. You can’t know a priori whether it adds useful information in a given election or across a set of elections.
Rilkefan, sorry, no link, and f course maybe I’m wrong. But I have the impression this was touched on in the radio discussion this link is in response to. There’s a link to that at top, and it’s worth hearing. I cant listen again nw with this migraine, or at least, I’d really rather not. Or maybe I saw it on this site or his. Very sorry. Sam? Anyone?
Ah, Rilke!
Rilkefan, very sorry, darn migraine, i can still be wrong but what i said was WAY too general. I just read up on Silver and he has all sorts of weighting… im meant adding ECONOMIC weighting which changes your use of the stats according to how the economy is doing at the time. Again, youre right to call me on it, i still can be way off, and please everyone correct with info, but not opinion.
I think the main reason for the difference is, as rilkefan suggested, the prior that Sam uses. If instead of focusing on the predictions, you look at the EV count predicted, Sam is now at 293 (which I expect this to go down a bit in the next few days based on the polls in Ohio) and Nate is at 286 (the now-cast). Now, I don’t know exactly their sigmas, but I’m pretty sure this makes for a difference that is a lot less than the 1 in 3 vs 1 in 9 of their predictions.
The thing to keep in mind, though, is that the prior Sam uses will matter less and less as the election approaches. So if, like me, you find it dubious (compared to the rest of the model, at least), you can just look at the meta-margin the next two weeks.
Sadly to say, the friday swing States numbers were not so good for Obama as I read on Nate Silver’s site.
Has the meta-margin of +2.22 included the friday swing State’s polls?
Silver just isnt as good as Wang [pronounced “wong”]. He adds all sorts of weights, and compared to our neurologist friend, Silver is more haunted by his own mind. My advice: stop reading 538. Seriously.
I’m here instead of there because he adds noise, the very sort of noise he himself admits gives misleading numbers, the sort which Wang gave up.
“he himself admits gives misleading numbers”
You really need to back this up.
oh rats posted these on wrong subthread. trying again-
Rilkefan, sorry, no link, and of course maybe I’m wrong. But I have the impression this was touched on in the radio discussion this link is in response to. There’s a link to that at top, and it’s worth hearing. I cant listen again nw with this migraine, or at least, I’d really rather not. Or maybe I saw it on this site or his. Very sorry. Sam? Anyone?
Ah, Rilke!
Ms. Jay Sheckley // Oct 20, 2012 at 6:28 pm
Rilkefan, very sorry, darn migraine, i can still be wrong but what i said was WAY too general. I just read up on Silver and he has all sorts of weighting… im meant adding ECONOMIC weighting which changes your use of the stats according to how the economy is doing at the time. Again, youre right to call me on it, i still can be way off, and please everyone correct with info, but not opinion.
well that and Nates comments are teh suxxor anymore, Ms. Jay.
Biden with bag full of Dunkin Donuts at campaign office in FL:
http://instagram.com/p/RAruBtmwwS/
We’ll see if PPP’s O+1 and Gravis’ tie will move OH to 2/2.5 and the MM down
I have it dropping to +1, since the previous +5 for PPP should fall of if I understand things correctly:
+3, +3, +1, +1, 0
Matt, I don’t think that the old PPP falls off, since these are different PPP polls without overlapping polling dates, and the old PPP is within a week of the latest poll. Ohio should drop from O+3 to O+2.
Pollster also hasn’t put up an O+3 WI poll released today by Grove Insight (http://polltracker.talkingpointsmemo.com/polls/5082dd0febcabf7ba100004c). This should move WI from O+2 to O+2.5.
If it was O+1 (and I’m not sure who is right) the MM shouldn’t be higher than +1.5 because IA is 1.5 and OH is (theoretically 1)
Amitabh, you and I have an acquaintance in common among the CERN physicists.
“The median poll, which reduces outliers….”
In five years of reading Nate Silver, this is the first time I’ve seen him refer to a median. Did Sam talk him into it during the recent co interview?? 🙂
http://fivethirtyeight.blogs.nytimes.com/2012/10/20/oct-19-after-romney-gains-should-obama-concede-florida/
mirable dictu!
Nate used the “m” word.
jamais de ma vie!
What ticked down today that took 6 away from O?
What about some Latinos who vote do not speak English?
Those people will not be counted in English only polls.
Not freaking out, but I just read R is putting about 60+ staffers back in PA. Has that shown signs of shifting now?
its kabuki.
chillax ma petit choux.
😉
For what its worth, RCP is leaving out some polls that show O leading in swing states. I dont have a complete list, but the latest case is O +2 PPP in VA. But they where VERY quick to post PPP OH poll showing close race, even though it came out AFTER PPP VA poll. So as of this post they have yet to post PPP VA O +2.
Earlier this week, after noticing that Electoral-Vote.com was posting some polls that weren’t listen on RealClearPolitics (RCP), I wrote the latter an email inquiring why this was so. Never did receive a response…
Where do you find the most reliable list of polls, if not on RCP?
It looks to me like RCP has a policy of not listing commissioned polls. Two recent VA polls by PPP were commissioned ones, by Health Care for America Now and the League of Conservation Voters, and therefore didn’t make it to RCP.
As to where to go for the most comprehensive list of polls, Polltracker (http://polltracker.talkingpointsmemo.com/) does a good job picking them up, but I find the site hard to navigate. Usually I keep track of what they uncover by following the Twitter feed at 538.
Since Pollster provides the data feed for PEC, I pay a lot of attention to them too, but they sometimes miss things. (Twice this weekend I sent them polite emails with links. When the MM drops this morning because Pollster finally listed yesterday’s PPP OH poll, you’ll know who to thank. Sorry for that, but the model is useful only if it gets all the data.)
Aha, you were the reason I had to modify my “Anatomy Of A Bounce” post.
It looks like the race has tightened in Ohio after debate #2. I was not expecting that.
The PPP polls in NH and IA are troubling signs for O campaign. Team Obama is releasing internal polls for PA ( O +7-8%) after reports of staff move by RNC. Here are today’s RCP composites;
State RCP Oct. 3, RCP Oct. 20, Net Change
OH, 5.0, 2.6, -2.4
FL, 2.0, -2.1, -4.1
VA, 3.5 0.0, -3.5
CO, 3.1 -0.2, -3.3
NC, 0.8, 5.6, 4.8
NV, 5.2, 3.0, -2.2
NH, 6.0, -1.0, -7.0
WI, 7.6, 2.8, -4.8
IA, 3.5 2.4, -1.1
PA, 8.0, 5.0, -3.0
MI, 10.0 5.0, -5.0
I think Obama needs to change the narrative if state polls continue their slide. If today’s up stick in national polls for Obama needs to strengthen Sunday & Monday polls. The leaked reports of Iran -US direct negotiations after elections appears to be intended to have O on the defensive in Monday’s debate.
Your concern trolling still betrays your Romney side. Which is common. Actually, if the rumor is true, it would help Obama before and during the debate. How you view this as a problem is beyond me,as this takes one more bullet out of R’s saber rattling stance. As for the poll numbers, some are up, some are down. You look at the data as a whole and follow the trend as Dr Wang has done.
Question for Mr. Wang and Others Here…
When we note that certain Pollsters lean Democratic or Republican, WHY do they lean? What benefit is there to be seen that way for them?
Every time I read insights here or on FiveThirtyEight, and see data and then the qualifying “left leaning” or “right leaning”, I end up just thinking well, doesn’t that inherently de-legitimize them in a small way? I assume they do this for a reason, but it seems a little bit (sorry for layman’s term) ‘unprofessional’. :}
@Jeff
In many cases, ‘lean’ comes down to differences in methodology. Robocallers , for example, can’t legally call cell phones, which may in theory lead to a different survey audience that may lean differently than the population as a whole. Always interviewing the person who answers the phone, rather than randomizing by asking for the adult whose birthdate is closest to ‘x’, is another example, because you will always miss some people. Calling at certain hours vs others may influence the type of person you reach. Etc, etc..
In short, how you contact people may influence the type of people you tend to reach, introducing a bias. For example, if I did a cellphone only poll conducted between midnight and 2 am, it’d probably skew pretty heavily D due to who would answer the phone at those hours.
Nate Silver has had quite a few articles on these matters, if you want to learn more….
Here is the sure sign that Obama is going to win.
Nate Silver just got godwinned by John Nolte.
https://twitter.com/NolteNC/status/259832893990387713
???
Regarding who has “voted”: were I asked if I have voted, I’d respond “Yes”, but my sealed ballot is sitting on the front seat of my car, awaiting transport to the County offices for hand deposit in the next day or two.
I daresay a lot of other early voters might be in analogous situations.
RAND looks bad…
not for the battleground states.
https://mmicdata.rand.org/alp/?page=election#results-by-respondents-characteristics
Why does no one mention that women’s body issues will strongly effect men, especially single men who gets women pregnant.
Someone needs to tell Obama’s campaign that they should put ads that say a man will have to pay child support for a child he does not want until the child is 18.
A lot of men will be paying big bucks for that alone.
I say this as a pro-life advocate but VERY strong Obama supporter.