With the debates complete, and just two weeks left in the campaign, there’s enough state-level polling to know pretty clearly where the candidates currently stand. If the polls are right, Obama is solidly ahead in 18 states (and DC), totaling 237 electoral votes. Romney is ahead in 23 states, worth 191 electoral votes. Among the remaining battleground states, Romney leads in North Carolina (15 EV); Obama leads in Iowa, Nevada, New Hampshire, Ohio, and Wisconsin (44 EV); and Florida, Virginia, and Colorado (51 EV) are essentially tied. Even if Romney takes all of these tossups, Obama would still win the election, 281-257.
The reality in the states – regardless of how close the national polls may make the election seem – is that Obama is in the lead. At the Huffington Post, Simon Jackman notes “Obama’s Electoral College count lies almost entirely to the right of 270.” Sam Wang of the Princeton Election Consortium recently put the election odds “at about nine to one for Obama.” The DeSart and Holbrook election forecast, which also looks at the current polls, places Obama’s re-election probability at over 85%. Romney would need to move opinion by another 1%-2% to win – but voter preferences have been very stable for the past two weeks. And if 1%-2% doesn’t seem like much, consider that Romney’s huge surge following the first debate was 2%, at most.
From this perspective, it’s a bit odd to see commentary out there suggesting that Romney should be favored, or that quantitative, poll-based analyses showing Obama ahead are somehow flawed, or biased, or not to be believed. It’s especially amusing to see the target of this criticism be the New York Times’ Nate Silver, whose FiveThirtyEight blog has been, if anything, unusually generous to Romney’s chances all along. Right now, his model gives Romney as much as a 30% probability of winning, even if the election were held today. Nevertheless, The Daily Caller, Commentary Magazine, and especially the National Review Online have all run articles lately accusing Silver of being in the tank for the president. Of all the possible objections to Silver’s modeling approach, this certainly isn’t one that comes to my mind. I can only hope those guys don’t stumble across my little corner of the Internet.
Well, I certainly hope you’re right. You can never know just how stupid the American people can be, however. That’s what got me a bit worried.
I would like to be right, too. But it’s not about whether people are stupid or smart. Voters decide for all sorts of different reasons. Best not to judge.
Drew,
I dont doubt your model any more or less than any other model, but when have you tested it to see how accurate it is? For instance, Sam Wang and Nate Silver have a good track record from 2008 and Silver an ok record in 2010 that make me thing they are probably correct. Thanks. Great site.
Thanks. I ran the model on the 2008 state polls and it worked well; the details are in the research paper that the site is based on.
In the Forcast Detail you have Florida listed as going to Obama (barely) but your most recent polling charts show Romney ahead. I thought I noticed similar behavior with North Carolina, which was barely on Obama’s side of the ledger until recently despite several days of Romney being ahead on the poll tracker. Is there a delay built in here or am I not remembering correctly? 332 seems awfully high to me given how close some of these states are.
That’s right – what you’re seeing is my model adjusting the forecast up a bit (about 0.5%) from current voter preferences, in line with historical fundamentals. That difference will go to zero over the next couple of weeks. But for now, the model’s median forecast still has Obama winning all of his 2008 states, minus Indiana and North Carolina.
I also hope you are right. However, normally reliable sites, such as intrade, are collapsing for Obama, nationally and in tight battleground states, such as Ohio. I find this astonishing especially after his strong debate performance last night. It’s as if only the first debate counted.
I’ve been watching you, Nate Silver, Sam Wang, pollster.com, and DeSart and Holbrook. All of you predict that the President should be re-elected, if not handily, at least by enough that we won’t have to stay up much past midnight on the 6th.
However, Obama shares on Intrade have been declining in price all day. Indeed, the “Obama Wins Ohio” shares are now trading lower than the equivalent Romney shares.
I know that the methodology of following a trading market is different from following analysts such as you and those noted above, but why are the disparities so great right now? Any ideas?
I have no idea what’s going on in the prediction markets. There’s neither enough historical data – nor good enough theory about how those markets relate to actual events – for me to get too bothered. I’ll stick with what I know, and what’s worked in the past: the fundamentals and the polls.
Right-wing media are currently making a big push on the “Romney is winning!” meme. This is a plausible reason for Romney’s rise on intrade.
Justin Wolfers offered some thoughts on what might have happened at Intrade. Apparently, this isn’t a difficult market to manipulate.
http://www.theatlantic.com/business/archive/2012/10/should-presidential-campaigns-spend-more-money-manipulating-intrade/264000/
You also have to be careful for manipulation of Intrade similar to those trying to push up McCain last election: http://www.washingtonpost.com/blogs/ezra-klein/wp/2012/10/23/how-to-manipulate-prediction-markets-and-boost-mitt-romneys-fortunes/
As many have noted losing tens of thousands in an attempt to create perceived momentum is way cheaper then spending millions on ads.
I’m not even sure its just the Right-Wing media pushing the Romney momentum story. I’ve seen stuff on Politico and NBCNEWS.com as well. I’m not seeing anything at the state level that indicates a consistent move toward Romney. Obama’s polls in Ohio and other states just don’t seem to be moving much if at all (to my untrained eye anyway) in the last week or so.
I think we’re just at the point where the interests of the media in general diverge from the interests of the public. Horse races sell steady races do not.
In particular, pointing towards the surge in Intrade prices for Romney:
I’ve noticed the presence of “arbitrage” opportunity more often, coinciding with the “Romney surge”. Arbitrage is easy enough to define. When the odds of O winning is X, and the odds of R winning is 101-X, and either O or R must win, this is a “theoretical arbitrage”.
If that were the market prices, you could pick up 100$, put 100*X/(101) on O, 100*(101-X)/101 on R, and pick up 1 $ for free.
Of course, there’s a buy-sell spread so the figures of X and 101-X odds don’t necessarily translate to a price arbitrage. But the larger the total of the two odds gets above 100, the more buy-sell spread you’ll need to square out the arbitrage opportunity highlighted above.
And I’m seeing more frequent existence of theoretical arbitrage, coinciding with the recent Romney “run” on Intrade.
Translation: someone is overbidding, and asking for their pocket to be picked.
This is wishful thinking Drew. Run your model for 2004 and get back to me. The problem that both you and Nate are driving is that you believe one election, the challenger Obama with a historic win, drives a model for his re-election. It’s entirely unfounded to extrapolate Obama’s multiple 47’s and 48’s in the battleground averages into above 50% averages.
I’ll make a prediction here. Your’e going to learn a hard lesson this election– don’t make models that fit your dogma.
Karl Rove’s theory of election end-games is that undecideds pull the lever for the person they think is ahead. Everyone wants to leave the voting booth feeling like they are on the winning team. So, in a razor-tight race, he who controls the narrative, controls the game. And the right wing has done a much better job of controlling that narrative, at least in the last few weeks.
My hunch is that the betting markets are reacting to that, which would be very difficult to price in to analytical models based on polls and fundamentals.
No Tribe: All my methods are transparent, there’s no dogma here. I use data that goes back to 1948. Numbers over 50% are because I’m recalculating everything in terms of major-party support.
” I can only hope those guys don’t stumble across my little corner of the Internet”
LOL. They will now, because Nate just outted you.
In a race that is this close in the states that hold the decisive balance of power…Florida, Virginia, Iowa…isn’t the real story in the “ground game?” Who will turn out the voters on the day (or in the early voting days)? And it seems to me that 1. no site, not even 538, has taken into account what is now some solid evidence about early voting trends, and 2. that there is really no way to assess how the parties will do in terms of turning registered voters into actual (likely) voters. So many polls seem to be based on “likely voters,” and no one talks about how the various pollsters assess whether or not a voter is “likely.” In fact isn’t Romney’s “surge” arguably due to a surge in registered voters who alreDY favored him suddenly identifying themselves as “likely” to vote for him?
This fallacy that undecideds break heavily for the challenger is parroted out alot. However, its definitely not true. I worked for Kerry in 04 and even though I knew we were down I kept telling myself that undecideds will break for Kerry and we win! However, undecideds broke towards Bush not Kerry. Bush was not above 50 in a few states final polls but he still got over 50. Besides, Obama is pretty close to polling at 50 in Ohio. The Quinnipiac poll and the Time poll from today have him at 50 percent. Some polls just dont push leaners that much, like Survey USA.
I’m noticing errors that make me think Intrade is doing a better job that either 538 or Votamatic. Nate’s model is still placing weight on the economy, even though this should be baked into the polling data by now. This produces an unrealistic level of support for Obama by discounting the clear momentum shift to Romney. Nate is also underrating the advantages of Rasmussen’s voter turnout model and under valuing the data coming in from Gallup. His weighting of various polls seems to be entirely based on his personal views – not backed up by any science or experience. As a political scientist, it was clear to me that Romney achieved a realignment of the electorate – especially among female voters – and that this is the big story regarding the election. Your decision to lean on the 2008 turnout model appears very problematic to me. I do not see anything like that level of enthusiasm for Obama today – especially among white voters. You and Nate seem to be missing the story of this election, a story that is so obvious to me and the bettors at Intrade.
Interesting, however, if I strongly believed that the odds of Obama winning are near 6-1, I’d certainly be over at Intrade making big bets on Obama given I can get nearly even money. You won’t get that deal on Alabama-Vanderbilt.
John C. Drew, I assume this ‘realignment’ refers to Romney’s post-debate one ‘surge.’ The reality is, however, that most of that so-called surge came before the debate, the post-debate bounce was only 2 points, more or less in line with previous post-debate bounces obtained by challengers. Moreover, Romney peaked at 48 and started dropping. That suggest his ceiling is 48. Obama has polled regularly at 49. Depending on which poll aggregator you believe, Obama has already cross back over and taken the lead and is trending farther up (TPM Polltracker) or is slowing trending up as Romney remains flat-lined at 47 (RCP).
Intrade is at 60-40 for Obama. Evidence of a continuing Romney momentum no longer exists, but his campaign is trying to manufacture it.
I may have missed something, but why don’t you show us the probability of a given candidate winning the election based on your model, like Nate and others do? Isn’t the 95% credible interval a somewhat arbitrary loss of precision or information that your model produces (the posterior distribution)?
Thanks for putting together food for our brains. I did find you through Nate….cheers
I forgot to include Joshua C. Haskins, Esq., LL.M Now people will listen to me.
For a strong website that illustrates the momentum for Romney, I recommend http://electionmarkets.blogspot.com/ As I see it, the ground game is already factored into the likely voter models. Moreover, Obama now looks weak in states where his ground game is not so highly developed. The first debate cause a realignment which changed Romney’s status in the eyes of female swing voters and has allowed this election to be a referendum on the poor economy. My wife was unemployed for two years during the Obama administration and is currently in a job with fewer hours and less pay. It is unrealistic to think that voters are not factoring in the pain caused by the Obama economy. As they say, there is little interest in going back to the last four years. :-)
@Dr. Drew, the phrase “Obama economy” kinda gives you away. Most reasonable people have factored in the depth of the global economic crisis, its persistence world wide, and how much a President can do in such a situation in only 4 years. In fact the Romney campaign itself allow 8 years for his economic plan to take real effect and he is inheriting an economy growing jobs albeit slowly rather than the 800K per month job loses that greeted Obama. Only right wing shills use the phrase “Obama economy” my good doctor.
Hey, are there any prediction models out there that are predicting a radically different outcome? For example, are there any they predict a Romney victory?
All reading this blog should be made aware that the verisimilitude of the comments posted by “John C. Drew, PhD” are very seriously undermined by his identity as a self-avowed Obama-hater and Tea Party activist. In other words, pay no heed to right-wing trolling.
Thanks for all of your comments. Even if you don’t like the forecasting component of the model, you can still watch the polling trends here. (For now, there’s simply no evidence of a Romney “surge” beyond the first debate.) As we close in on Election Day, the forecasts will become increasingly aligned with the polls. Sit tight, be patient, it’s only two more weeks.
I think Romney’s momentum is more obvious in the Rasmussen polls. I think a lot of the polls showing Obama hanging on have severe problem in terms of being to optimistic about his 2012 turnout forecasts. My Republican friends are fired up, Democrats look demoralized.
John C. Drew says, “I think Romney’s momentum is more obvious in the Rasmussen polls.”
That’s a joke, right? The Rasmussen poll has a widely recognized Republican bias. One need only look at the state polling in which President Obama is tied or has a lead in every single swing state. Just yesterday the NBC poll again showed the President with a 5-point lead among likely voters in Ohio. The Romney campaign and their right-wing shills are bluffing by trying to create a perception of momentum – but the fact remains that Romney has virtually no plausible path to 270 electoral votes.
Who is John C Drew, PhD?
http://grantwriters.ning.com/profile/JohnCDrewPhD
http://www.breitbart.com/Columnists/John-C-Drew
http://frontpagemag.com/2012/jamie-glazov/young-obama’s-dreams-of-a-communist-revolution-in-america
You guys have made math interesting to this historian.
MAT
Rasmussen polls? Give us a break.
Dr. John C. Drew, award-winning political scientist….http://anonymouspoliticalscientist.blogspot.com/2012/10/trust-me-im-from-past.html
My friends are almost entirely Democratic, so that would seem to translate into a big Obama victory, no?
The Rasmussen complaint about being Republican leaning is only relevant if their results are wrong. If they turn out to be right, they’re just more accurate not more biased. My admittedly imperfect memory is that they’ve been reasonably accurate but I don’t know how they compare with other polls.
I’m curious though about how people resolve the disconnect between state and national polls. If you believe both, are you putting a very high probability on a Romney vote win and Obama electoral win? Or do you believe the national polls are just wrong and the state polls will more likely reflect the national vote totals?
Electoral-Vote.com actually DID (today) examine the case of Rasmussin polling – and the results are pretty mathematically conclusive that Rasmussen DOES have a Republican bias:
http://www.electoral-vote.com/evp2012/Pres/Maps/Oct25.html#item-8
Hedgehog is correct: right now, we don’t know which pollster is going to be closer to the truth – only how their results compare to one another, on average. Rasmussen could turn out to be very accurate. And comparatively speaking, they’re not that Romney-leaning, anyway.
No more comments targeting particular individuals, please.
The x factor in this election is the unreliability (lack of verifiability) of voting machine results in some key states coupled with 1. the ease of altering results and 2. ownership of the technology by Republican partisans. The bogus claims of voter fraud by false identification smells like a smokescreen.
New to the site, so I apologize if this is addressed elsewhere. In “How it Works,” it looks like you are predicting that Obama will receive 52.2% of the vote, and this number is used to adjust the state numbers. Is this number an internal number used just in the model or an actual prediction? How well does the formula you used to get the 52.2 number match previous elections?
Obama’s shares are going up on Intrade: http://www.intrade.com/v4/misc/scoreboard/
Well, I’m in Texas so my vote doesn’t matter, but it seems to me that these quantitative guys like Drew and Nate have their reputations to consider, well past Nov. 6, so it’s not in their interest to skew their results for any candidate. After all, they are not in line for government jobs or anything…
ChrisW: The idea is that I start with a national baseline of 52.2% for Obama, based purely on historical factors. I translate this to the state level, then update from the polls as we go along. The model that I use for the 52.2% was originally invented by political scientist Alan Abramowitz, and it’s had among the best track records of any of this type of fundamentals-based forecasting model.
I also hope you’re right, but I won’t rest till the the weight-challenged singer has completed the aria. I love number-crunching and appreciate that you did so, and that you reported on many odds-makers – so my questions & comments :
– What do the real gamblers in Vegas have to say ? (and could a bet pay off better than my 401(k) ?)
– Who said ” No one ever went broke underestimating the intelligence of the American public” ?
– How come none of these pollsters ever contact me ? ( just because I never answer the phone for unknown callers is no excuse )
Hi Drew: Might we expect a model update today? How often will your model be updated going forward?
Wow ! I read more of this stuff and drilled down into some detail. Boy, did I learn a lot ! I’d never heard of InTrade (sounds like I could make some money there) and “Big Gov’t Social Conservative” gave me a lesson in arbitrage – luckily he summarized it before my head exploded.
Keep up the good work.
Update just posted. I’ll try to get the estimates refreshed every day or two from here on out. Nothing about the race should change more quickly than that!
Mark: It’s a big country, 300 million people. The pollsters are only calling 500-1000 people at a time.
The arbitrage argument can be made even simpler.
Tuesday morning, Intrade started off selling Romney at 40% (higher price).
Same day, same time, other websites were selling the *nearly identical* good Romney, at 37% (lower price).
Obvious arbitrage is, buy at the 2nd site, sell at Intrade.
BUYING at Intrade when this obvious arbitrage opportunity existed, doesn’t make sense.
The run on Romney at Intrade (where his price went all the way to 49%) was irrational exuberance, and the market has corrected the price now.
“The idea is that I start with a national baseline of 52.2% for Obama, based purely on historical factors. I translate this to the state level, then update from the polls as we go along. ”
Well, it’s an idea. But really, its nothing more than that in essence. Starting with a baseline like that in terms of history doesn’t take into account that the partisan makeup of this country has changed dramatically. Drew, I don’t want to come off as too much of a wise guy. You might be right and I’ll want to know now before you are famous…
What you are trying to tell here, is that the national polls that came out today (TIPP, WaPost, Gallup, PPP, Rasmussen, AP) , together showing Obama at a 46-47% average and trailing Romney by 1-2% average, are under-polling the end result Obama will get by 5-6 percent?
Is there anywhere to see what the model projects the final popular vote will be?
Josh: the national vote isn’t something estimated or forecasted by my model, sorry.
Here’s a possible explanation of Romney bump on Intrade. Some people, like me, hedge. They vote Obama and bet a 100 bucks on Romney. Or vote Hillary and bet 100 on Obama. The money pays for a scotch to drown your sorrows when your candidate loses.
No Tribe: First of all, I recalculate all the percentages in terms of the two-party vote, so that explains part of the discrepancy you’re seeing. I’m setting aside all the undecideds. Second, it’s not a national race, it’s 50 state races, and so I only look at the state polls. Third, I agree that the choice of the Abramowitz model that starts me off at 52.2% is somewhat arbitrary and if I’d started off at, say, 50-50, the results would be a bit different. But the way my model works, it has to start somewhere, and the Abramowitz model has a very good track record. I guess you will just have to believe me that I didn’t pick it for ideological reasons! My only interest here is testing out my method, and hopefully making accurate forecasts. If you really don’t like the forecasts, there’s only two more weeks to wait.
Objectively, I think the 52.2% starting point looks too optimistic for Obama. Rasmussen shows Obama, nationally, running at 47%. At this point, I don’t think it is plausible to suggest, as this model does, that Florida and North Carolina are swing states. As is, I think Drew’s model is underestimating the impact of the realignment associated with Romney’s first debate performance. This debate made a big difference among female swing voters and was not anticipated by Drew’s model.
Well, it seems arbitrary, but it’s what you got so go with it and we’ll see.
I would be interested in seeing a model that goes after Rasmussen’s very accurate voter ID methodology:
http://battlegroundwatch.com/2012/10/25/what-do-you-know-a-decent-party-id-and-romney-is-pulling-ahead-nationally-abcwashington-post/#comment-17504
I was struck by your statement that you set undecideds aside. It seems that the allocation of them is half the ball game. One theory of course is that undecideds break for the challenger and even if only 3-4% are truly undecided, that would probably be more than enough to give this to Romney. Of course it could be they just break 50-50 but I would think there’d be enough historical data to at least make some kind of estimate.
“Objectively …”. It’s hard to claim objectivity. Sampling error exists, period. There is no objective sampling method, so every poll is wrong. The question is: how wrong? The whole point of poll aggregation is to not look at single polls, like Rasmussen (a poll that tends to run more favorable to GOP relative to other polls; others run more favorable to dems).
Drew does some tweaking, but I’m curious to see whether it holds. Using only state polls makes sense, the electoral college being what it is; and tossing out undecideds is probably OK because they’re a small percentage now. Both these parts of the design I think lead to a bit more objectivity than, say, Rasmussen has.
Incidentally, I have a Ph.D. too. But I don’t append it to my name as it hurts my self esteem.
I read thru all your stuff here and all the comments. Drew, I commend you on your non-partisanbapproach and level headed comments.
That said, I would like to suggest that your model is not constructed correctly, as others have pointed out.
Romney appears to be leading nationally, appears to be pulling ahead in most swingstates, and has a good chance to win the election.
Factors not picked up by your models include the obvious – lack of enthusiasm for Obama among his base, more enthusiasm on the Republican side, the disaffection among independent who gave Obama a shot in ’08 but won’t this time, and the petty, dishonest, vulgar and desperate campaign Obama is running at the end.
Add it all up and you have to conclude Romney has the clear edge with 12 days to go.
if your model is suggesting a big Obama win, then it is obvious that you need to figure out where you are going wrong.
If Obama does win, and he may, it will beva squeaker.
If I had to bet right mow … Romney wins.
gl
No problem Drew. Thanks for answer.
Anyway, two general responses to some critical posts I just read.
1. Look at the huge vote share drops after the first debate, not just here but also on 538 and Election Consortium. I think the model took the debate into account.
2. Critiquing prediction models and using Intrade and Rasmussen as a basis for your points is pretty dubious.
I trust the various quantitative models but am unwilling to bet on Obama on Intrade – nor advise others like me to do the same.
Despite my intuition telling me that Nate Silver’s likelihood of an Obama victory is very much understated, I know that I have an emotional makeup that is unsuited to speculation. The worst thing that could happen to me is if I win. That happened in my first gold futures trade back in the early 1990s – and then thinking I was a genius, I kept speculating until I lost what I had set aside to invest in gold. That cured my speculation forever and now I am very happily early-retired with a financial adviser making the decisions.
After a few election cycles professional money managers soon figure out that easy money can be made in the election prediction markets when a democrat is projected to win the US presidency by quants. The market is biased by the current participants towards the candidate they want to win – and according to their self-selected demographics that is the republican. If the actual expectation of an Obama win is 95%, then someone could make 25% on their investment in less than two weeks. But that person will not be me – no matter how much I like the odds.
Drew, its interesting that your model gives Florida narrowly to Obama and the other quants do not. I have begun watching for new daily polls on Pollster for both Florida and North Carolina to see an indication of movement towards Obama. In addition, I am looking for whether the Obama campaign schedules any events in North Carolina – as confirmation that their internal polls have detected something promising there.
As it happens, people have already voted and are currently voting. Obama has very large leads in the early vote. Those are not poll numbers. Romney’s campaign touting a ‘record number of contacts’ is meaningless. Knocking on the door and leaving a brochure isn’t exactly GOTV.
PBO is tied or ahead in every swing state. Every one. One 90 minute debate against six years of running for office as an ‘extreme conservative’ cannot change the fundamentals of the race and math is proving that out.
Lastly if conservative extremism is so appealing, why did Willard pretend to be a moderate? He’s a plutocrat without conviction.
I seldom see third party candidates factored in. Take, for instance, Virginia. Your poll data adds to 100% for the 2 major candidates, yet two third party candidates may take up to 6% of the vote, most, I believe, from Romney (see http://www.cnn.com/2012/10/25/politics/third-party-candidates/ ). In how many other swing states are results affected similarly?
The big unknown int he polls, it appears to me, is whether the people who say they will turn out for Obama actually will. The Obama campaign is saying that is running scared. I suspect they are saying that for two reasons — that’s the way to turn out more votes, and they really are (and should be) scared.
So stupid MSM reporting about how Romney has “momentum” and is continuing to rise in the polls, as false as it is, may actually be helping Obama.
Drew Linzer, do you have a twitter account so we can all follow you?
This is a stellar analysis that deserves more recognition.
Here’s Dr. Wang.
https://twitter.com/SamWangPhD/status/261811891209388033
I am a coloradoan and I totally do not get the “irrational exuberance” of Rasmussen etc. Marijuana is on the ballot and that will turn out youth, and Rasmussen’s LV model is a proven fail in Colorado.
In 2010 Ras wrongly predicted Buck and Tancredo (both lost), because his LV model doesnt capture the smartphone and cell-only demographics.
Dr. Drew (and every else echoing his point). All you keep repeating is “the national poll data from Rasmussen says Romney is God and we should bow to him.” However, you seem to be completing ignoring what Drew and others are trying to explain to you. The Presidential election is not based on the national popular vote (just ask Al Gore and a few other Presidential candidates of the past), thus the national percentages really don’t matter at all when they are within 1-2%. This can be achieved by Romney by absolutely dominating Obama in the red states, but staying fairly close in the blue and swing states. As a Ph. D I am sure you are aware of the electoral college process. The huge problem for the Romney campaign is not whether or not he can get close to winning the popular vote, it’s his path to 270 EVs. With the current battleground states you can give NH, FL, VA, IA & CO all to Romney and he would still lose. And if you put IA, CO and NH to the Obama side, Obama could even lose OH and he would still win. The math just doesn’t add up.
Q: Do your projections factor in the newly registered voters for either candidate? Also because minorities are typically undercounted, do you somehow reflect this? Thanks.
Wheeler’s cat: I’m on twitter @DrewLinzer.
Steve: According to my model, the candidates are tied in Florida right now and historical factors suggest Obama has room to improve, so that’s why I have it tilting slightly in his direction.
bien merci Drew.
you and Dr. Wang both need more followers.
I adore your site, Dr. Wang introduced me to it.
I’m a Talebian. Mandelbrot over Gauss.
Pack: Thanks again for your comments here on the site. I just don’t see the evidence for what you’re saying. I don’t track the national polls, but the Huffington Post does, and they have it essentially tied. Romney is certainly not pulling ahead; if anything, the momentum has been with Obama for the past couple of weeks. If the election were held today, the EVs add up for Obama.
It’s true that my model doesn’t explicitly consider things like enthusiasm or disaffection, or the nature of what either campaign is putting out there. But I don’t see the problem with that – the polls alone have been pretty accurate, on average, and I’m simply trusting that to continue.
How about we wait until Election Day before we decide that my forecasts are right or wrong? Nothing about what my model is showing right now seems inconsistent with the data so far.
Ok, so this comments section got long. I’m going to close it here and we can move it to the latest post.
Also, I had to delete a few comments that I didn’t think were appropriate for this site. There are other places on the Internet for that type of exchange.
Hedgehog: The “disconnect” between the state and national polls doesn’t mean that either one is wrong – they can both be correct and yet be different, because they measure different things. The national polls try to measure what would be the nationwide popular vote breakdown. The state polls, conducted primarily in the battleground states, are used to forecast the electoral college votes.
For an electoral college vote, two states of approximately the same population that poll respectively, 55-45 and 35-65 both count as solid for their respective candidate, and both deliver exactly the same electoral vote. But in the national popular vote polls, the 35-65 state weighs more heavily for its favored candidate than does the 55-45 state.
The national polling is closer than the electoral college predictions because the margin for Romney in most of the “solid red” states is larger than the margin for Obama in most of the “solid blue” states.
I understand that Nevada has the best record among all states in calling presidential races in the past 100 years. (wrong only in 1976) NV seems consistently leaning toward Obama this time around, so if NV’s bellwether status holds, Obama appears headed toward victory. Rather simplistic, but interesting nevertheless.
I see a really big misconception here in the comments. I haven’t reviewed the entire method, but it appears, like nate silvers, to take state polling data, weigh it for bias, rate of change, historic accuracy and then use that weighted data to predict the outcome. I.e. its a complicated model that uses specific polling data and a calculated weighting system in the interest of accuracy.
People saying “you’re not taking into account x surge with women voters”, completely ignore that the polls are largely trying to account for voter changes like this, so inherently the model gets adjusted for those changes. More so, a personal opinion or a single poll showing a candidate as being in the lead is no reason to discount a statistics model you don’t agree with.
From what I can tell, 538’s weighting system actually favors Romney too much compared to the battleground state poll averages.
I think you are all delusional. The country wants a change just as the Obama promised except the change they want now is Obama out!
The only thing Obama accomplished in 4 years is to drag this country down. The deficit is higher, unemployment higher, government giveaways are higher…etc.
People see the that the direction we are heading is the wrong direction.
I will be laughing as Romney wins.
As to RG and such. Everybody has a right to their opinion but I’m not sure that opinion pieces (probably including mine) are worth the bandwidth. If one gets their ideas from TV, one should realize that TV stations have dog in the fight. They follow individual polls, particularly their own, (as the CNN polls) and want you to stay tuned until it is all over. They are thus biased toward a near draw. That is good for ratings and advertising profit.
You probably already know about Nate Silver: http://fivethirtyeight.blogs.nytimes.com/
I like the site of the Princeton Election Consortium with Sam Wang as head geek.
He gives Confidence Intervals for the present and does predict the CIs (68% and 95%) for the Nov 6 vote. If you go by Nate Silver and Sam Wang you could almost say that the Silver-Wang rules. (just a prosthetic aside)
http://election.princeton.edu/history-of-electoral-votes-for-obama/
The DeSart and Holbrook Election Forecast has a solid model with good graphics.
http://research.uvu.edu/DeSart/forecasting/october.html
They make it convenient to get probabilities state by state.
Some of these sites might be progressive in their hearts (being able to read and write points you in that direction) but they all let the math rule their polls and do their deciding on models..
The present site on which we are writing is excellent. I am very grateful for all these meta sites. If anyone knows of a meta site that has a conservative bias, I would be very interested.
I am sort of proud I was among the first to poke holes in Nate Silver’s wildly optimistic projections regarding Obama’s presidential chances. Here’s a link to a chart which shows how Nate’s model is leading to increasingly bizarre conclusions. http://www.aei-ideas.org/2012/10/a-sign-nate-silvers-fivethirtyeight-may-have-skewed-way-off-course/ Rasmussen, in contrast, is showing a Romney victory in the EC at 271.
John C. Drew is confusing the amount gamblers are willing to put up to get a full share in case of an Obama win with probability. See Intrade.com for a clear explanation of what the Intrade score means.
Binomial logistic regression (and similar) actually yield probabilities with CIs and are thus not directly comparable with gambler’s hunches. I am not knocking Intrade. It has a good history of predicting wins and this morning predicts a win for Obama with the trend line going up.
An earlier post pining for the use of the economy is rather well met by the site:
http://www.washingtonpost.com/wp-srv/special/politics/2012-election-predictor/
This morning, with GDP growth of 1.7% from Q1 to Q3 and with an approval rating of 50%, it gives a probability of an Obama win at 86.5%. To fellow quants, I know that the prob should be written as 0.865 but it is common in columns on polls to use %.
Of the sites using meta-analysis, Silver’s actually gives conservative results compared with other sites. See prediction by the site on which we write and the sites I give a few comments above this one. This is not a criticism of Silver. Quants differ some in their models and I compulsively check at least 5 sites every morning. After the election, I’ll see about getting a life.
I will add my 2 cents on your comment on Intrade.
It is possible to “rig” the betting markets, in a short time-frame.
For example, let’s assume GOD were betting (GOD is omniscient: he/she knows the true odds (but not the outcome, since it takes out human action, and free will exists, even if GOD were to bet in this market).
Today Intrade is at 65:35.
Another betting website has the odds at, say, 70:30.
A rational HUMAN being would either buy R at 30 on the other website and sell R at 35 on Intrade, or buy O at 65 and sell O at 70 on the other website. The prices would be “dragged” to an intermediate point.
Now, let’s take the optimistic scenario for Romney, and argue that GOD knows the true odds to be say 60:40 right now (i.e. the price is depressed on both websites, due to ‘price manipulation’). Then, GOD would buy R at both websites, more at the cheaper website (since GOD knows the market will correct to the true odds, sooner or later).
This has been a 2 billion $ campaign. There exist enough people with the money to undo the artificial depression of the natural odds.
Thus, I posit that it is unlikely that a movement in EITHER website, is “irrational” UNLESS the action doesn’t sit consistent with human rationality.
There IS evidence that this has happened, in this cycle.
For example, Tuesday before yesterday, the price of Romney shot up to 49% on Intrade.
Where, at the same time, the price on the OTHER betting website did not shoot up.
This WAS an example of “irrational human action”. A rational better, in it to make money, would have sold Romney at 49% and bought at the cheaper price on the other website, and squared off when the odds correct on both websites. And it was proven.
If GOD were acting, and the “true odds” were known to GOD as 50:50, that behavior could be explained as “rational”.
As long as the odds do not exhibit such GOD-like behavior, the trend movement of the odds can be trusted to be “rational” and not “manipulative” (Ah, I finally got to the point, didn’t I).
Isn’t turnout — who ends up voting — a much bigger unknown at this stagte than the expressed preferences of those who respond to pollsters? If this is correct, then whomever has the best estimate of the demographics of turnout will end up with the most accurate prediction.
Thank you for the fine extensive work of you and your team. The four or so meta-sites have been a great comfort throughout this long ordeal. See you, hopefully, in four years.
As I write, it is raining in Sarasota. FL and I auger that this may help Obama since the dems have more votes in the bank and rain suppresses votes without prejudice. That idea, of course, is open to future research.