General overview 2012: the state of play

September 6, 2012 by Sam Wang

Welcome to all the new readers. Traffic is booming.

First, a brief update. Several weeks ago I showed that convention bounces (and other fluctuations) are uncorrelated with the final outcome. However, it is fun to look at them. In the next 1-2 days, I will give a final measure of the GOP’s post-convention bounce, which is zero or negative. I note that a “negative bounce” is highly unusual.

Since many of you are new, I will take time today to give a general overview of the state of play. The Princeton Election Consortium’s analysis and prediction approach gives a sharp picture of where we are at: very likely Obama re-election, with House and Senate on a knife edge.

To learn more, my July and August posts lay out the issues with relatively little fluff. (Ignore the FAQ, which I am still updating.) To summarize it all, here are some salient points.

The Presidential race is largely determined. The national media is correct that this year’s Presidential race is close. And voters are polarized: as few as 1-2% of voters are persuadable.

However, the media have failed to clearly spell out the logical consequence that the Presidential race is also very stable. President Obama has kept an electoral lead every single day since May. Based on the statistical behavior of polls in past re-election races, his November re-elect probability is 88%. Conversely, the probability of unusual movement or a black-swan event is 12%.

Both Senate and House control are on a knife edge. We have an unusual situation this year: control of both houses of Congress is up in the air. This recalls the elections of 1994 (GOP takeover during Clinton’s first term) and 2006 (Democratic takeover in GW Bush’s 2nd term). What’s different is that 2012 is a Presidential election year, so voter attention is higher. This will have profound consequences for Obama’s probable second term (and for Romney’s less probable first term).

Campaign funding will affect Congress more than the Presidency. This year the Citizens United ruling will have a major impact. Because money is most effective in marginal cases, spending will be most effective in Senate/House races – not at the Presidential level. This is true whether you are a Democrat or a Republican. It is why Karl Rove’s Crossroads GPS is not focused on the Presidential race. And it is why I have set up the ActBlue page at left. Republicans can use the NRSC.

Now, a few words about the methods by which I reached these conclusions.

The state poll Meta-analysis. The first step in our Presidential analysis is to use state polls to provide a snapshot of conditions today. As in past years, I use outlier-resistant medians to generate a probabilistic calculation using Electoral College mechanisms. The result is two outputs: an electoral vote median with a 95% confidence band, and a Meta-margin. They are updated every day at 8AM, noon, 5PM, and 8PM.

On Election Eve, the EV estimator lands on the final election outcome within approximately +/-5 EV. So the snapshot (in the top line of this site) tells us where the race is at any given moment with very high accuracy.

The Meta-margin is very special. It tells us how far the race would have to shift to create a electoral near-tie. It is like the standard national-poll margin you see in the news, with two differences: (1) It is in the units that matter, the Electoral College, and therefore closely reflects swing state movement. (2) It is more accurate than any poll aggregation you will find elsewhere. Its typical precision is +/- 0.2%. By this measure, today Obama leads Romney by 2.0%.

New for 2012: a prediction. This year I draw upon the statistics of past-year races to make a prediction for the November election. This calculation does not use other variables (unemployment, campaign spending, bounces) because they fail basic tests for good model-building.

From now until September 27th, I am giving a long-range outlook, analogous to long-range weather prediction. After that, I will replace the long-range outlook with a shorter-term prediction. In the meantime, what we have is this:

2012 prediction 27-Aug-2012

As of the end of August, the long-range electoral prediction is 283-353 EV (1 sigma, the red zone) and 250-360 EV (2 sigma, the yellow zone). The two-candidate vote share prediction is Obama 51.6 +/- 1.1% (1 sigma).

Noise that is not included in this analysis. I do not use any econometric variables, on the grounds that these simply add noise. One consequences, as I showed yesterday, is that the Meta-analysis has extremely sharp time resolution to detect bounces and shifts.

Commenters. If you are curious about any of these topics, the comment threads on this site are curated and are of relatively high quality. They are worth browsing.

There is much more to write about. But I’ll stop there.


Billy says:

Sam, have you ever thought of writing a book about this type of approach? Nate Silver has a book coming out soon and it would be interesting to see a counterpoint to that. I understand that most books nowadays require book deals and such, but there are avenues for self-publishing (such as Amazon) and you do have a strong following here.

Matt says:

Andrew Sullivan at The Daily Beast probably sent you a lot of traffic with his blog post linking to your analysis.

Tapen Sinha says:

As Matt noted above, Sully has sent you a lot of traffic. However, I think Sully misrepresented your analysis and lumped it with Silver. This blog is far more definitive about the outcome than Silver is.

Olav Grinde says:

What a wonderfully succinct post!
This almost belongs on your front page.
I’m really looking forward to seeing the chart of this coming week, contrasted with these last couple of weeks.
By then, voters will have hade a chance to hear both President Obama’s and ex-Gov. Romney’s speeches, hear the other speakers at both conventions, and compare the platforms of the Democratic Party and the GOP. Hopefully there will be a sufficient number of new state-level polls to provide a sharp evaluation.

Jacob Hartog says:

Is it possible to use your methods to generate a probability of popular vote victory/defeat? The Meta-Margin shows the degree of popular vote shift that would needed to produce a switch in electoral college winner, (right?) but not who the actual popular vote winner is predicted to be.

xian says:

btw, just eyeballing the meta-margin chart from a shape/geometry point of view, it appears to be trending toward zero before election day. I realize that is a truly unsophisticated way of looking at the chart, but does it in any sense indicate a tightening race overall?

Sam Wang says:

Some rapid replies…
Billy – I’ve thought about writing a scholarly article. But you are right, a single might be a better route. One question is how to reach the academics too. I would like to influence their thinking. From what I’ve heard, Wlezien and Erikson might already be partly there…
Matt/Tapen – Andrew Sullivan’s links gave single-day spikes in traffic. Now we are getting that much every day. Also…did you really think I have not examined traffic statistics in minute detail?
Jacob – The prediction would be the same in either case.
Xian – Too much fitting. If there is no long-term correlation, then fitting to a line or whatever is silly. I saw something on Andrew’s site to that effect too. A tasteless application of quadratic fitting.

Jacob Hartog says:

Jacob – The prediction would be the same in either case.
Is this true? Let’s imagine a scenario where Romney is polling at 80% in all 25 lowest-population states and Obama is at 52% in all 25 highest population states. Wouldn’t the Meta-Margin be 2% in favor of Obama (because a national swing of 2.1% toward Romney would cause Romney to win the electoral college), even though Romney might already be winning the popular vote?

Sam Wang says:

Jacob – Relative to the uncertainties this is in the noise, and there is no net difference.

Matt says:

I think you are probably referring to that blog post by Steve Lambardo at pollster. The fact that he had to squint really hard to find some way to make the data say what he wanted it to say and then just out of thin air correlated it to Republican talking points is pretty much par for the course when it comes to a lot of political commentary.

Matt McIrvin says:

That Lombardo post was particularly ridiculous, since he not only fit a quadratic curve to a noisy dataset, he thought it was significant that the bottom of his parabola was the day of Obama’s “you didn’t build that” speech (amazing how it affected the bend of the curve backwards in time). Friends don’t let friends do that kind of statistics.

AlpsStranger says:

Good riddance to Mittens.

xian says:

Forgive my belaboring this ascientific thinking, but I wasn’t so much talking about why something might be happening (fitting with a line that’s supposed to mean something or with significant events), as whether something might be happening at all.
That is, if I were in a leaky boat and I got the sense that the water level were rising, but I wasn’t sure it was because of waves and splashing and such, I might not have a good theory for why the level might be rising but I still might develop an intuition that it seems to be.
If i stop blurring my eyes and using magical thinking, the only thing i feel comfortable saying is that the most recent negative peak in the chart is the lowest.

Sam Wang says:

Properly deployed, the point of statistics is to turn your broad question into a sharper inquiry. In this process, taste is required. The general approach is to come up with a simple explanation for the data, then ask what the probability is that the data arose by chance under that explanation — or whether something additional is at work. This is “taste.”
Fitting data to a quadratic in the service of some excessively specific hypothesis is an example of bad taste. However, when it comes to the use of statistics, pundits are not bound by any licensing authority. Thus foolishness like that found in Lombardo’s essay.

Eric Fisher says:

You are an excellent Matlab coder, and you have me as an academic hooked on what you are doing. I will try to get some students interested in this kind of modeling. It has broader applications in economics and finance.
Simple technical question: the time series properties of your Meta-margin seem to indicate that one campaign of attack ads (Bain Capital) can move the margin easily by two points. Perhaps some heavy Romney spending in FL and OH could bear fruit. Then the election would be a toss-up. Do you know how often the metamargin has moved 2% in a week in the years for which you have been doing this analysis?
That’s not how I understand Taleb’s notion of a black swan.

Sam Wang says:

Eric Fisher – thank you. I wrote the MATLAB. Andrew Ferguson wrote the python.
A black swan would be a totally unexpected event like a hostage crisis or an economic collapse. Its effect would have to last until the election.
2% in a week: Swift Boat campaign, Sarah Palin VP nomination, Lehman Brothers collapse.

wheelers cat says:

The CU polysci professors, Lombardo’s silly scatter plot, and Silvers phantom bounce are all different examples of wishful time travel….travel back in time to a past electoral landscape where Romney could actually win…
But unfortunately for Romney, time travel to the past is impossible because of closed form time curves. Hawking and Carrol agree.
we can only go **forward**.

Tapen Sinha says:

Slightly off topic: This kind of line would be quoted by generations to come. “A freedom which only asks what’s in it for me, a freedom without a commitment to others, a freedom without love or charity or duty or patriotism, is unworthy of our founding ideals, and those who died in their defense.” by Obama tonight. It is in the same class as JFK in his inaugural address: “Ask not what your country can do for you – ask what you can do for your country.”

Brad Davis says:

What do you think of the usage of emplying some ‘goodness of fit’ metrics such as AIC to compare various ‘competing’ models such as Lombardo’s ‘quadratic fit’ to a more parsimonious description? Or would that be giving too much credit to an otherwise easily discredited method?

Matt McIrvin says:

I don’t know, I’m increasingly convinced that this year there’s a real difference between the popular-vote and electoral-vote situations.
Part of it is the number of national polls that show a *gigantic* difference in Romney’s favor (7 or 8 or 9 points) when likely-voter screens come in. I think it’s why Romney is now tied or leading in most of them. It’s the basis for a million articles about Obama’s looming turnout problem.
The odd thing is that if you look at state polls in battleground states, you don’t see that split at all! It just doesn’t exist. The LV/RV results are about the same, or in some cases Obama actually does better with likely voters.
The only thing I can figure is that either the national polls have some bizarre systematic error in their LV screens (but why wouldn’t state polls by the same company show that?) or, more likely, that the difference is because of things happening in the *safe* states.
I can easily concoct a Just So story in which a lot of lefty progressives in deep-blue states go Naderite because they don’t like Obama’s civil-liberties record and figure their vote doesn’t matter anyway, whereas Tea Partiers in deep-red states are all fired up to eject him from office, yet neither phenomenon has any effect on the states that really matter. But that’s all it is, a Just So story. Nevertheless, I wonder.

Brian says:

@Matt McIrvin – if I read you correctly, you’re suggesting there may be a chance that Romney wins the popular vote while losing the election by a non-trivial margin. It wouldn’t be quite like 2000, in which Al Gore won the popular vote and lost the election by one state. Nor would it be like 2004, in which the EV count was similarly close while the popular vote went handily for the winner, Bush. To see a significant Romney popular vote win alongside a 300+ Obama EV win would truly be something.

Matt McIrvin says:

It’d be really, really unusual. To the point that I wouldn’t have regarded it as possible until now.
I’m just trying to figure out what’s going on. Part of this is probably the general fog of convention time; things are happening fast and different metrics have different delays built in.
It could be that Obama’s EV margin is actually just about to evaporate, at least temporarily; or that the very tight national polls are some kind of statistical fluke and everything is shortly going to return to the relatively stable status quo ante.

Matt McIrvin says:

Note also, I have no good way of testing my weird hypothesis because there’s so little state polling in safe states. It might be possible to approach it by looking at semi-safe states considered longshot battlegrounds.

Sam Wang says:

Matt McIrvin – I am not seeing a reason to cook up such a hypothesis. Look at the current RCP. Recent polls show a median of Obama +0.5+/- 0.7% (median +/- SEM, n=6). Sure, that is 1.5 points different from the Meta-margin. But the national polls are actually more stale, since they span the peak of the Ryan-VP bump. Thus we would expect them to be slightly more favorable to Romney.
So yes, I think it’s just convention fog. Give it a few days.

Hans says:

Just for curiosity, I’m interested in knowing how these methods might be applied in the case of a third party candidate that can affect the outcome of the race. This year it obviously doesn’t matter, but it might in future. The 1980 and 1996 elections would be examples, and possibly 2000 depending on how you think Ralph Nader affected the election.

Sam Wang says:

If the third candidate ever took the lead, one would need to calculate the statewise win probability differently. This is easy. However, third-party candidates tend to fade in the home stretch.

Matt McIrvin says:

I’ve actually been hearing lots of speculation about Romney losing significant votes to Virgil Goode in Virginia. I seriously doubt it’ll really be a major consideration.

Brian says:

@Matt McIrvin – re: Virgil Goode, not likely. The electorate is utterly polarized right now. People have either made up their mind already to vote for Obama, or hate him deeply and want him to lose. There is virtually no middle ground. A vote for a 3rd party candidate right now is essentially a vote for Obama. I predict an insignificant tally of votes for The Honorable Mr. Goode in November.

Michael K says:

re: Virgil Goode — I was curious so I did some digging:
The Constitution + Libertarian party candidates combined got 0.50% of the Virginia vote in 2008 and 0.66% of the Virginia vote in 2004. Which was fairly close to what they got nationally in each case.
In 2008, Barr(L) got 0.7% in his home state (GA) vs. 0.4% nationwide. But on the other hand Baldwin(C) got 0.09% in his home state (FL) vs. 0.15% nationwide.
In 2004, Badnarik(L) got 0.55% in his home state (TX) vs. 0.32% nationally. Peroutka(C) got 0.12% in his home state (MD) which was exactly what he got nationally.
So the history on native-son third party boosts is mixed. On the high-end (Barr 2008, Badnarik 2004) the boost would be ~70% which still realistically amounts to a boost of no more than 0.3% in Virginia.
I have no idea how many of those voters would vote for Romney if he and Obama were the only ones on the ballot. But it could be an interesting question in the event of a razor-thin Obama margin.

Sam K says:

Are you worried at all that there seem to be fewer state-level polls this time around compared to 2008 [1]? If the race is as stable as the analysis here suggests, it shouldn’t matter too much in the long run, but I can see it making it harder to detect temporal effects like convention bounces. Or to put it another way, if the race does change significantly, a state-polls-only approach like yours might lag such a shift.
For example, looking just as swing states, IA, NH, NV, VA, and WI have had no new polls since the Republican convention. A +2% Romney bump in all of those states would certainly have moved the meta-analysis needle, but we wouldn’t be seeing it right now.
Thanks for running an awesome site!

Sam Wang says:

Sam K – thanks. See the more recent post for an answer. It’s probably ok.

Gregg Gibson says:

What about the Gary Johnson/Ron Paul effect? I have seen polls where Johnson is at around 4%, with most of this coming from disgruntled Ron Paul voters (aorund 20% of the national vote), many of whom may not vote for anybody, but just stay at home. So I’m not sure that two-party polls are that reliable. They UNDERSTATE Obama’s actual margin by several percentage points at least. This wouldn’t change things much in the electoral college. But it might make Indiana, Missouri and Georgia close. I personally know a lot of Republican-leaning independents who are furious at Romney, and not at all mollified by the Ryan VP choice. Obama isn’t very popular, but he faces no real opposition.

Leave a Reply

Your email address will not be published. Required fields are marked *