With the last set of polls factored into the model, my final prediction is Obama to win 332 electoral votes, with 206 for Romney. This is both the median and the modal outcome in my electoral vote simulation, and corresponds to Obama winning all of his 2008 states except Indiana and North Carolina.
The four closest states – and therefore the most difficult to predict – are Florida, North Carolina, Virginia, and Colorado. Of these, my model only expects Romney to win North Carolina; but Florida is a true toss-up, with just a 60% chance of Obama victory. I would not be surprised if Florida ended up going for Romney. If that happens, Obama would win 303 electoral votes, which is the second-most likely scenario in my simulation. The third-most likely scenario is that Obama wins 347 electoral votes, picking up North Carolina in addition to Florida.
It’s been interesting to watch the forecasts of other poll watchers converge on the 332 estimate. Sam Wang, at the Princeton Election Consortium, also sees 332 as the modal outcome. So does Simon Jackman at the Huffington Post, and Josh Putnam at FHQ. Nate Silver, at his FiveThirtyEight blog, reports the mean of his electoral vote simulation at 313 – effectively splitting the difference on Florida, which he currently rates a 50.3% chance of an Obama win. But his most likely outcome is still Obama 332, followed by 303 and 347, just like me. Update: both Wang and Jackman revised their forecasts slightly downward this afternoon, based on late-arriving polls.
There will be plenty of opportunities to evaluate all of these forecasts once the election results are known. I’ve already laid out the standards I’ll be using to check my own model. This is how quantitative election forecasting can make progress, and hopefully work even better next time.
I’ll add, though, that on the eve of the election, averaging the polls, or running them through any sort of sensible model, isn’t all that hard. We are all using the same data (more or less) and so it doesn’t surprise me that we’re all reaching similar conclusions. The real challenge is producing meaningful and accurate forecasts early in the campaign. My model is designed to be robust to short-term fluctuations in the polls, and converge in a stable and gradual manner to the final, Election Day estimates. It appears that in this regard, the model has worked as intended.
But from a broader perspective, my model has been predicting that Obama will win 332 electoral votes – give or take – since June. If all of us are correct today, the next question to ask is when each model arrived at the ultimate outcome. That’s a big if, though. Let’s start with how the votes come in tonight, and go from there.
I think I’ll just take a long nap, relax and check back tomorrow if your forecast was right. If it was, you are my own personal hero. :)
Drew – I have followed your site for about two months – thanks for sticking to your guns – great intellectual and academic integrity on your part throughout the process.
So, if the model works and election outcomes can be determined early on, the next obvious step, for the losing team mostly, is to figure out what it would to take to move those numbers and to build a campaign around that strategy. Truly fascinating stuff this year from you and your peers. Makes me wonder how different future campaigns will be now that the public is becoming a bit more sophisticated. I think two things are becoming clear from this cycle: big money isn’t the factor we thought it would be and most Americans are tired of being bullshitted. Both things are good in my opinion.
Drew, even odds for you to be branded a damn psychic :)
let’s hope your forecast is correct… and I’ll drink to that!
Found your site through the bbc – very interesting thank you. Looking at the margins of error and other things, I still don’t think Obama will get over 300. I’m going for 297 – 241 but hey, living in the UK, what do I know? But as Sascha said, if you are right I will be in awe!
What is the actual 95% confidence interval?
Wow, bonus points for consistency and integrity of the model.
And if it’s on the money tonight, I predict you’ll be a very popular guy in 2016! :)
Yay, them’s the same numbers I posted on my Facebook page last night. Great minds think alike!
There is a point of whether your 332 electoral votes was prudent as early as June. I think it was, and I point to the success of Lichtman’s “Keys” model. Personally, I believe elections only swing when the background conditions start to move. However, there is no probabilistic way of determining such a thing. What your June prediction means is that the fundamentals of the election have not changed since then. But when Carter lost to Reagan, and even when Obama beat McCain, there is an argument that the fundamentals changed close to the election and significantly altered the final vote. McCain still would have lost without the financial collapse, but it would have been a lot closer.
In other words, your prediction holding up tells us something about this election but nothing about predicting future election outcomes with any high precision from several months away.
Drew, given that election day wont change and we don’t know the true answer until election day, what’s the value of a prediction model?
The value of the predictions especially in the near and medium term would be to a candidate say in determining on how to spend or allocate resources.
If you’re losing by 20+ points you’re likely not going to throw everything you have into a state vs. a state that is within a few point or less where a good ground game or advertising dollars may help sway the outcome in your favor.
One thing is certain, at the end of this night, one side or the other is going to look completely ridiculous and pretty clueless due to the fact that the poll aggregators election projections from both sides could hardly be larger polar opposites of one another. I’ve seen “projections” that have Romney winning by the margin predicted here for Obama. The complete and total disconnect is amazing. The “Romney is going to win” crowd, however, seems to rely mostly on supposition instead of actual facts. Not surprising, really, since Republicans everywhere seem to think there are actually two sides to facts. The whole “who is right about the polls” question is the most fundamentally interesting thing about election night, as things stand now, right next to the actual election itself.
If your model is as good as I think it is, then maybe the next step could include: (1) measuring congruence in the party platform and the voters. I suspect that the Republican Platform in 2012 was somewhat radical on social issues (e.g. abortion) and perhaps (given the election) that the Platform and campaign promises spoke ineffectively to Independents (who I consider voters suffering from dealignment); (2) changes in the party platform over time considering campaigns and winners according to public opinion for the candidates; (3) whether or not one issue increased a block of independents to move into a party (e.g., one of the parties become against preemptive drone strikes).
@Roy in Omaha
To be honest, both sides are relying on suppositions. The difference is that sites like this one and 538 use solid math on top of suppositions which may or may not be correct, and they acknowledge this. The Romney side combines suppositions with ad-hoc data manipulation (unskewing polls, doing out-of-sample regressions, etc.) and do not acknowledge their conclusions could be wrong, since accepting you can be wrong has a liberal flavor to it.
Many think markets are smarter than polls at forecasting outcomes. The odds of Obama winning on Intrade are only 67%, and they have been consistenly lower than the forecasts made by poll aggregators. Why the discrepancy?
Excellent job Drew. I too have been following you over the last month. It’s nice to see an impartial academic analysis of the election, and with few exceptions the comments here have kept to the analysis as opposed to the politics. I am predicting though that you may need a moderator to filter out the chaff next time around as your site becomes popular. Again, well done!
Even if this model turns out to be deadon I’m not sure how future candidates use it. The results didn’t seem to change a lot over the course of the campaign so is it telling us anything about what Romney could or should have done differently? Do better in the last two debates? Spend a lot more early on to try to take the lead early? Maybe they could have hit lightly polled places like PA and MN earlier and harder? Those are the only obvious thoughts I had but I would have had them regardless of the model.
If the model turns out to be wrong, not even about who wins but about national polls being less predictive than the state polls of the results, then I guess there will be ample arguments about likely voter models and the real composition of the electorate that goes to polls.
Rory,
I agree with you. To me, the PEC prediction is the most interesting, because it is the simplest. He uses only state polls, includes all of them (i.e. doesn’t exclude polls viewed as “partisan”), doesn’t adjust for sample size, partisan lean, past history, etc. Just takes the median of the most recent polls. The approach worked great in 2004 and 2008; if it works great again in 2012 then we know that the aggregate systematic error in polling is usually small enough that you can ignore it. If his method fails this time, then we know that aggregate systematic error is, at least occasionally, large enough that you *can’t* ignore it.
Either way, 2004 and 2008 tell us that aggregate of state polls do a *very* good job at least some of the time. So the main question going forward will be how to better quantify the error bars and/or how to better understand/predict elections where systematic errors will be important.
Can you infer the impact of each campaigns get out the vote efforts? I imagine that, if Obama’s effort were more successful than the traditional campaign, more of the unlikely voters would actually vote. I would love to see someone attempt it. Thanks. Fun stuff here!
@ Steve
Why the discrepancy between Intrade and the forecasters? More than one PhD dissertation can come out of trying to answer that question!
The short answer is the uncertainty in the underlying assumptions of the models (voter turn up composition and late deciders probably the most important ones).
Modelers don’t assign probabilities of their assumptions being wrong (538 might be the exception, not sure. PEC is the worst in this regards, by far.)
Drew (and everyone interested):
Please follow Intrade and the real-time outcomes of EV. You will notice that Intrade will drift up and down from the initial setting of 0.68, but it will NOT jump. What this means is that the information revealed by the voting process is changing, but not surprising and jolting the market. This happens because the market had superior knowledge that goes beyond anything a model can do. That knowledge is adjusted incrementally as time goes on.
Allan Marlow,
Nate Silver reports that the prediction and betting markets are starting to diverge from Intrade even more than they have in the past.
Drew – I agree with others here, regardless of outcome it’s been a treat to visit your site and read your thoughtful comments. Ill be watching the site over the next few days to see what’s next. Many thanks!
Commentor –
The intrade spread is now closing, very interesting.
Very interesting mapping your analysis with real-time updates. Appreciate the research, analysis, and time you take to explain the metrics.
Well done sir. Well done. I raise a toast to empirical knowledge, and it’s triumph over superstition.
Mr. Boyd doesn’t understand: Mathematics and empiricism based upon facts have a well-known liberal bias.
332 appears to be very likely. Virginia shows a Romney lead but the outstanding counties are largely democratic.
So true Stuart, so true…
And the winner goes to… the Abramowitz prior! My instincts were with Drew’s model since way back in August ( I was there, go check). People should stop deifying Nate Silver and his hedge-all-bets concoction and see how a correctly identified prior exerts a remarkably stable gravitational pull on the outcome.
Excellent job Drew
Congrats, Drew, looks like you’re dead on.
@Allan: “This happens because the market had superior knowledge that goes beyond anything a model can do. ” ??? Smoking crack. Intrade picks winners, i.e. who gets 270+ EV. As the numbers came in players adjusted the prices as their inferior knowledge was exposed, so they update their prices along knowledge. Drew’s model had all this incorporated in back in May, but I don’t see the markets incorporating that same knowledge in their prediction: so the players aren’t rational and they are not efficient. If they were efficient, the price would have been settled a couple months back at a much higher level.
332 is looking better and better; Drew Linzer may win the prediction sweepstakes this year.
You magician you. 332 is looking pretty spot on – well played sir well played.
Drew, you are the MAN!!!!!!
Election Day Forecast: Obama 332, Romney 206
SPOT ON, GREAT JOB!
Congratulations, Drew! It looks like you may be the winner of the poll aggregators in 2012. And I would say that second place should go to Dr. Wang of Princeton.
Congratulations! Florida’s results aren’t in yet, but otherwise it looks like you’ve been dead on since June!
Any recommendations on upper-undergraduate/graduate level texts on introductions to forecasting? Some emphasis on Bayesian approaches and in particular the use of R or home-built MCMC samplers would be nice.
Amazing work on calculating the numbers. You’ve statistically proved the model! Well done.
Nailed it, Dog. Good work.
Amazing! Nailed it, and your state predictions were excellent. Awesome work.
Congratulations, Drew! I’m looking to the APSR (or at least Perspectives) article.
What happened to the other famous political scientist, Dr. John Drew, Ph.D.? Why isn’t he on the comment board to explain to us that the only valid pollster is Rasmussen?
Nice work! Though I think the model’s dead-on (likely) success is partly the result of the “coin flip” probability in FL breaking your way, the early call is most impressive! I look forward to reading more, but wanted to add one more voice to the congratulations.
ns538, samwang, votamatic got it perfect – 50/50, time for the benchmarks!
facts vs facts
logical methodology vs logical methodology
gut feelings, out of the window…
these are better days.
Congratulations on the excellent success of your model. At the very least, you predicted the outcome of the other models based on state polls a good four months in advance, with little wavering to speak of during those months. If Obama wins FL, you’ll have predicted the actual outcome with such success.
Your colleagues at PEC & 538 share your victory on the election eve forecast — but you stand apart on predictive stability!
Congratulations, perfect and perfect for a long time. Not because you got the number exactly; rather, because the race so obviously centered around the number you discerned months ago.
Awesome job and great site. I hope you get the kudos and attention you so richly deserve for hitting the EV on the nose.
Take a bow Mr. Linzer. 332 EV seems merely a formality [of a recount in FL]. You, along with Sam Wang and Nate Silver have vindicated the empirical aggregation of polling. Against the wicked welter of pundits crying bias, yours and other projections have proven that number crunching trumps egocentric prognostications based on a whim. We always knew that some would have serious egg on their faces wednesday morning. Thanks to you and others, we know precisely who they will be!
I have just had a new appreciation of political science. You were able to see through all the cloudiness that surrounded the run-up to election day. Impressive.
Congratulations!
You understood that math trumps rhetoric and that the Republicans had to say they were winning to get their base out, ironically, the Democrats needed a close race to motivate their base as well. So both sides were making this seem closer than it was.
You say through the myth and predicted the math. Great job!
I have to say, I’ve been following you, Nate Silver, and Sam Wang throughout this election. I couldn’t believe you’re model was stuck on 332 EV, especially after the first debate. But you win.
Congrats!
Congrats Drew. It looks like you were spot on. I have a feeling come 2016, you’ll be the go to guy.
Dr. Linzer! Between your work and Nate Silver’s work my passon for statistics has been re-kindled. I have an MA in Sociology from U of Oregon and I live here in Atlanta. I pretty much forgot numbers because of my current job. But now, I am going back to my old passion and putting my faith in good old (new) fashioned hard science of numbers. When I read your predictions and the predictions of other numbers crunchers, in my head I knew you guys were right. My heart was swayed by the political pundits on tv. Never again. Thanks again and congrats on getting it right.
Great job to you and most of the other aggregators. It was a huge win for data dorks! Maybe they will start listening to us on other matters thanks to you.
Wow! You nailed it!
Awesome work. Most people seem to want to give the credit to Silver. But you called Florida before Nate did. Not saying he did not do well, only you did AWESOME….
GOOD JOB….
AWESOME Job.
Long live the central limit theorem!!!
Remarkable accuracy. Congratulations.
Wow, you nailed it. Im truly impressed! Keep up the good work.
Super job – prescient and accurate! Congratulations.
Nate’s great and Sam Wang rocks, but you are the one true aggregating god. Wish I had found your site sooner — it would’ve saved me alot of stress. Congrats on winning the night!
Hey, where’s the other Dr. Drew. You know, the objective tea party, RWNJ. He was telling us and Drew Linzer how mistaken we were this whole time. That all the polls were wrong because they were miscalculating voter turnout. Where are you Dr. Drew???
Great job! Your website was such a great resource to alleviate all the anxiety leading to the election day.
Thank you,
Kari
Congratulations, though I’m disappointed with the result, there is no denying your calls were very accurate.
Hey, thanks for making me look like a genius with my friends!
How come John C. Drew, PhD and Allan Marlow aren’t here to crow?
Well done Drew. Your forecasting model was outstanding and very, very precise.
Congratulations, Drew!
Polls aren’t skewed after all :)
Even Republicans might start to respect math, eventually…
There is still the need to reconcile market lags, but some progress is in the offing there…
On Monday, I thought 332 votes was optimistic, but it seems like you nailed it. Although Florida still hasn’t been called because Miami-Dade is a tabulation mess, Obama has a lead of about 46,000 votes this morning, and Miami-Dade should add significantly to that total — perhaps enough to avoid an automatic recount.
As Sam Wang quoted Randall Munroe this morning: “BREAKING: Numbers continue to be best tool for determining which of two things is larger.”
Drew, Congratulations once again – I have seen your name mentioned in BusinessWeek and I hope that result speaks louder than anything :)
I have also been very impressed with the stability of your prediction.
Look forward to your continued contribution in future elections projections!!
Congratulations Drew! Spot on prediction and a well put together website. I don’t have a strong math background, so your ability to not only crunch the data, but also explain it was wonderful! I look forward to following your website in future elections.
The accuracy of your election-eve 332 prediction is remarkable, but that your prior had predicted this number in July is simply astounding!
Great work, many thanks.
NAILED IT! 100% correct. Congratulations. See you in 2016!
Congratulations and well done.
Bravo, Drew. Well done.
One last thought.
Americans must consider themselves fortunate that a man of Obama’s intellectual caliber, integrity, and astuteness is in command. The alternative of a God-in-waiting in the White House was too terrifying to contemplate.
Friends, enjoy it while it lasts!
Truly remarkable Drew. A couple of questions though: was this year’s long-term stability at 332 unusual? Your prediction was based on poll results, so I find it remarkable that the electorate’s basic sentiments were this stable. Are you surprised at that; is this degree of stability unusual? Have you run this model on previous elections and found similar results over the course of those campaigns?
Apparently Miami-Dade County has finished its count, but won’t publish the results until Friday. Reportedly, Broward, Duval, and Palm Beach counties are still counting absentee ballots. So, we won’t know until tomorrow (at the earliest) if 332 really is the magic number.
Drew, congratulations on such excellent work! According to this article, the Romney campaign has conceded Florida and your EV prediction was correct:
http://www.miamiherald.com/2012/11/08/3087995/romney-campaign-we-lost-florida.html
Wish I had had complete faith throughout in your forecast, thereby avoiding a lot of stress.
Exceptional job, Dr.Linzer.
Bravo Drew! Bookmarked your site for the next election!
On Sunday, November 3 evening the CNN & Fox had interviewed Dick Morris that he said Mit Romeny would have landsliding won presidential election, which bothered me till Election Day I read your final prediction my son-in-law emailed me.
Thank you very much for your fair prediction that it had made it true.
Amazing prediction!!!
Dr Drew, congratulations. I was following you from India, for the last few weeks, and I think you deserve the most credit, because of the stability of your forecast. It will be, as you say, interesting to see how early the forecasts that were on the mark started to hone in on 332 and 303 EV as the modal outcomes.
On a lighter note, I hope to get my teenage son interested in statistics through your work!
I live in Atlanta, and I too found you through the BBC News web site.
I’ve already seen Nate Silver on the Daily Show; while it’s hard to blame Messrs Silver and Wang for their last-minute adjustments–conservatives already fear science; the last thing we need is for them to regard it as a partisan tool–I hold you in the highest possible esteem for your courage and your unshakable faith in mathematics.
I will be tending bar at the Highlander this Saturday from 11:00 am to 7:00 pm. Please introduce yourself and let me buy you a drink–I’ll be the longhair pouring himself a tequila (so it’s probably a good idea to come in early).
I don’t think you’re quite getting the recognition you deserve for this perfect forecast but you will. You’re obviously in great company with Nate Silver and Sam Wang, but your model clearly topped theirs this time around. I mean… You’ve pretty much had it right since… what…. June? Fantastic work.
patronanejo: Thanks! Unfortunately I’m out in California right now, on sabbatical, and won’t be able to take you up on the offer. Next time!
On the nose — congratulations!
And thanks for doing the difficult work.
Absolutely brilliant work Drew – you have shown that most pundits and journalists are mere entertainers and propagandists and haven’t a clue about science and stats.
Congratulations. A victory for statistical analysis over all of the talking heads who got it wrong.