How Close Is The Election Really?

by: Chris Bowers

Fri Jul 25, 2008 at 14:11


Currently, the two best statistically based electoral projection sites out there right now, Pollster.com and fivethirtyeight.com, are showing the electoral college is two very different states of competitiveness. While Pollster.com shows Obama with a very comfortable lead, fivethirtyeight shows Obama with only a narrow lead and a 59.5% chance of winning.

Both websites project the current statistical likelihood of one candidate winning a state. Almost across the board, Pollster.com projects Obama with a higher chance of victory:

Comparative Chances Of Obama Victory, Pollster and 538
State Pollster 538
Arizona 33-49% 10%
Arkansas 5-32% 4%
Delaware 95%+ 88%
Georgia 33-49% 7%
Indiana 33-49% 25%
Massachusetts 95%+ 94%
Maine 95%+ 92%
Minnesota 95%+ 89%
Missouri 33-49% 23%
Mississippi 5-32% 4%
North Carolina 33-49% 22%
North Dakota 33-49% 20%
New Hampshire 68-95% 66%
New Jersey 95%+ 91%
New Mexico 95%+ 56%
Ohio 68-95% 52%
Oregon 95%+ 87%
Pennsylvania 95%+ 80%
Virginia 51-68% 46%
Washington 95%+ 94%
Wisconsin 95%+ 87%

There are only two states, Nevada and West Virginia, where 538 projects Obama's win percentage higher than Pollster. By contrast, the above table lists 21 states worth over 200 electoral votes.

Who's right, and who is wrong? I look into this in the extended entry.

Chris Bowers :: How Close Is The Election Really?
Unlike 538, Pollster.com does not have a chance to win projection of the electoral college as a whole. However, I have to think that if 21 states, most of them swing states, shifted significantly in Obama's favor, then 538's overall win percentage calculation would put Obama way past the 59.5% it currently projects. In fact, even though their specific win percentages for each state are unavailable, I'm pretty sure that Pollster.com's percentages currently give Obama over a 90% chance of victory in the Electoral College.

So, according to current polling, is Obama only narrowly ahead, as 538 projects, or is Obama a virtual lock in the Electoral College? I can see an argument for both websites. In favor of Pollster.com, they only include polling data rather than outside demographic assumptions, and appear to give a more realistic account of base states like Massachusetts (McCain does not have a 6% chance to win the state if the election were held tomorrow). In favor of 538, they don't assign equal weight to the largely absurd Zogby Internet polls.

There probably isn't a way to know for certain which website is more accurate. Also, an individual's projection preference will probably be based on their intuitive sense of how close the election is right now. Personally, my gut tells me that the campaign is only lean Obama, but there is still a large part of my brain that can't make sense of why that is. It feels like it is close, but it also feels like Obama should be able to dominate McCain. From what I can tell in conversations with family, friends and colleagues, that is a widespread sentiment right now.  


Tags: , , , (All Tags)
Print Friendly View Send As Email

Take note of 538's Electoral Distribution Chart (0.00 / 0)
Its interesting that his simulations, along with the expected high spike predicting a close result, also shows a pretty large spike around Obama getting a 360+ electoral vote win. Seems to suggest his models think there's a decent possibility of a blowout too even if the rest of his stats are relatively conservative.

Of course, I'm crap with stats so I might have read that wrong.  


I think the key difference (4.00 / 6)
is that Pollster.com tells us what the polls are telling us about the race right now, while Fivethirtyeight is trying to predict what will happen on Election Day. Nate's theory is that the race will tighten up in the fall, since presidential elections almost always do.

Right, plus... (4.00 / 1)
You don't even need a theory of race tightening to loosen up your projection 3 months out. An infinity of events could occur to realign the country. 60-40 in spite of all that potential chaos seems like a dramatic lead to me.

On the other hand, 538 recently dropped it down from about 67-33. If today's good tracking polls hold, it'll probably go back up.


[ Parent ]
And in this (0.00 / 0)
assumption, he is actually wrong....

[ Parent ]
Could you elaborate? nt (0.00 / 0)


[ Parent ]
Well, it's logical (0.00 / 0)
since just about every election in the past fifty years has done so.  

-New Mexico politics from the local perspective.

[ Parent ]
1980 (4.00 / 2)
1980 is shaping up as perhaps the closest analog.  Many people rejected Carter, but had doubts about Reagan.  Then when Reagan performed well in the debates, he began to seem plausible to a lot more people (although he NEVER was plausible to many of us).  He eventually won decisively, though not by nearly as much as in 1984.

I think we have to face that a great many people are still uncomfortable with Obama, or with the idea of Obama, and much of this has to do with race, but some with age.  People were unsure of JFK and Bill Clinton as well.  They need to get to know Obama better and feel more comfortable with him.  Then his numbers will come closer to the generic Dem numbers.  I think it will happen this fall, and end up pretty decisive for Obama.  But it will take time.

John McCain--He's not who you think he is.


[ Parent ]
key difference: current standing vs projection to Nov. 4 (0.00 / 0)
Yes, this is the key difference.  Pollster.com asks "what would the election results look like IF THE ELECTION WHERE HELD TODAY", while 538 tries to project forward to the actual election date.  There are many factors not included in the 538 model, including where candidates are expending their resources and the myriad of unpredictable events that might occur between now and election day.  For that reason, 538's numbers are speculative.

For charts showing the probability of winning based on the current polls, without including the many speculative factors included by 538, see
http://election-projection.net


[ Parent ]
This pretty much lines up with how I have been feeling lately. (4.00 / 2)
I think 538 is potentially overstating McCain's chances at this point, but I also don't feel entirely sanguine about how the election is shaping up.

The bump in the daily tracking polls today is making me feel a lot better (even though I know those things aren't too reliable).  And if we see some good results in other polls over the next week, I will start feeling much better about Obama's chances.

Since the beginning, though, my biggest concern has been about last minute race-baiting, fear-inspiring anti-Obama advertising.  The moment of truth won't come until November 4th...  For this reason, projections at this point are even less useful than in previous elections.

That said, thanks for your work with projections -- I follow the polls religiously, and I appreciate the info I get on this site.      


I don't think 538 is overstating McCain's chances (0.00 / 0)
Basically, there are two models in which Obama could win a close vote:

The first is that he wins the Kerry states and Ohio.

The second is that he wins the Kerry states plus Iowa, New Mexico and Colorado.

Regarding the first scenario, I could easily see McCain winning in Ohio. Regarding the second scenario, I think that's more plausible, but if something goes wrong in just one of those states, it's curtains (unless Nevada also comes through).

Furthermore, if you look at the polls, McCain could easily poach Michigan from the Democrats, and if he does then Ohio is also almost certainly lost. In that case, not even an Obama victory in Virginia could stop McCain.

What I'm saying is that McCain has plenty of plausible winning scenarios. We're definitely in the lead, but it's not a big one and it's early.


[ Parent ]
I think right now: Obama's winning. (4.00 / 5)
The problem is, I think, with 538's model was excellent during the primaires but what that model doesn't capture is the poltical realignment that has occured after 2006. It can't do that at a presdidential level. That's why I think Pollster.com is more accurate. Obama's winning, but it's not a huge national lead right now. I think when the campaign gets focused and he's back here making arguments against McCain it'll get better.

I would just like to say if you know someone in the campaign Chris: please have them start the "We can't afford four more years of Bush policies, John McCain represents George Bush's thrid term." meme. This has disappeared and I do not know why: you don't hear it on cable and you don't hear it enough from the surrogates and you don't  hear it from the candidate.

If John McCain = George Bush then this will be a blowout and that is the theme that needs to be hit day in and day out on every single subject. Obama was hitting it in every speech before the nomination, at the nomination acceptance speech and after and then they backed off for some reason from making it so explicit.

They need to go back to the winning formula IMO.


Bush's third term (4.00 / 3)

 I hadn't explicitly noticed that, but you ARE correct. The Obama campaign has de-emphasized the McCain=Bush talking point. I don't know who advised the campaign to do that, but whoever it was is either an idiot or a saboteur.

 How well was it working? Well, I'll offer a little story. A couple of months ago, as Obama was wrapping up the nomination, my very wingnutty mother was complaining -- yes, complaining -- that Obama was linking McCain to Bush all the time, and that it wasn't fair. My jaw dropped -- wouldn't my mother see that as a compliment? Wouldn't rightwingers see a comparison to Bush as a positive thing, given all the adulation they've heaped on him over the years? But no, even the far right -- and my mother's as far right as it gets -- knows that Bush is poison. And if the right knows it, where would independents be?

 McCain=Bush is a very effective line of attack -- and a good springboard for OTHER attacks. If Obama abandons that line, he's shooting himself in the foot.

   

"We judge ourselves by our ideals; others by their actions. It is a great convenience." -- Howard Zinn


[ Parent ]
I don't think they deemphasized it, rather (4.00 / 2)
He's been abroad and they are trying to stay on message with the trip and foreign policy dominating.  When he gets home it will be different.  I think he's gotten pretty irritated at McCain impugning his patriotism and constantly hitting below the belt.

I think that will be the highlight of their economic message--the forgotten 85% can't afford another 4 or 8 years of Bush economic policies.  And what does McCain have to offer but more of McSame.

John McCain--He's not who you think he is.


[ Parent ]
I think you're right. (0.00 / 0)
The campaign has recently been engaged in defining Obama in an affirmative, rather than a contrasting, way.

When he begins running against McCain again, and when he turns to the economic argument, which he should do quickly, constantly and insistently, then the McCain = Bush argument will be at its strongest. Accurate or not (and "not" is the correct choice), the McCain as maverick on foreign-military policy line has taken hold pretty strongly. But on the economy, he's just a Bush wannabe.


[ Parent ]
Fivethirtyeight is probably closer to correct (4.00 / 2)
Pollster.com has no way of accounting for movement in the state of the race in states that haven't been polled recently. So, if a state hasn't been polled in 2 weeks, but the national mood has changed since then, the model won't reflect that.

Fivethirtyeight, using state-by-state and national polling data, makes an estimate of the general movement nationwide, within demographics and within particular states based on historical trends and applies that information to its models for states that have less recent polling. So it "sees" movement in states before polls do. As such, it reflects the fairly recent downturn in Obama's numbers in all states (even those without recent polling), whereas pollster.com only sees that movement in states that have plenty of recent polling.

It also has a fudge factor that assumes (as has, on average, been the case historically) that the numbers will get closer as election day gets closer, so it will always show a closer race because of that.

Conduct your own interview of Sarah Palin!


Does 538 publish the details of the model? (0.00 / 0)
I imagine that there's some sort of a model for "local gravity," so if there is a big move in one direction in Colorado, we should assume there is also a movement in a similar direction in New Mexico and Nevada - for example.

But more importantly to the results of the general election: there should be certain Bayesian factors employed for conditional probabilities. This election is not an unconnected sequence of local races, so the win-probabilities are not independent. So the chance that Obama wins Ohio are XX, but the chance that Obama wins Ohio given that McCain wins Michigan should be much smaller.

I wish 538 revealed how they take these things into account and maybe allowed us to tweak these variables. The point about conditional probabilities is absolutely crucial in making result predictions. It doesn't require fancy math, but it does require educated guesses about how nationally coherent this election will be, and also how regionally coherent it will be. But it would be silly to take the probabilities listed by either prediction outfit as being mutually independent.


[ Parent ]
he does publish them (0.00 / 0)
See here and here.

[ Parent ]
538 gets it wrong (0.00 / 0)
Cross-state correlations such as those made by 538 are not only useless, they are dangerous.  They are useless because if the state has not been polled in a while then it is not competitive and will have no bearing on the election results.  They are dangerous because the correlations do not take into account which states the candidates are focusing their resources.  For example, Wisconsin may be demographically similar to Michigan, but if the candidates are campaigning in one of those two states you would not expect them to move together.  Similarly, the national polls say nothing about what is going on in the battleground states, and should not be used to project the election.

[ Parent ]
It's probably going to be like 1980. . . (4.00 / 7)
Where it looks close for a long time, and then the American people go strong for Obama at the end.  Like Reagan, people have the doubts about Obama because he's so young, relatively inexperienced, and he's black. . .which is all a very big change for most voters.  Ultimately, you'll see a tide break his way.  

For McCain, I have to wonder whether this will be his high point.  He's not a good speaker, he is gaffe-prone, he looks bad on television (and worse on HD), and the debates won't have a game-changing moment.

Meanwhile, Obama will have moments galore - inspiring speeches, better than expected debates (imagine seeing Obama next to McCain on HDTV), etc.  He won't make the big gaffe - not after everything Clinton threw his way.


qft (4.00 / 1)
Wow. You took the words out of my mouth. Bad economy, bad occupation of Iraq, hated lame duck President from McCain's party, Obama charismatic while McCain is a black hole of anti-charisma.... it all adds up to a blowout. Many swing voters really haven't tuned in to the election yet and the events that have yet to happen (the convention, the debates) heavily favor Obama.

[ Parent ]
Good point on HDTV (4.00 / 1)
Nixon lost because he looked sweaty and unshaven compared to JFK on the new media of color TV in 1960.

McCain could lose because he looks old and unhealthy on the new technology of HD TV in 2008.


[ Parent ]
Actually, Nixon looked worse (4.00 / 3)
In black & white.  Few people had color in 1960.

John McCain--He's not who you think he is.

[ Parent ]
My five euros... (4.00 / 2)

 Republicans are embarrassed by McCain for the most part, but they'll hold their noses and vote for him. That's the vibe I get from the many wingnuts in my family.

 I think that the "feeling" that Obama should be winning easily comes from the idea that Obama is, frankly, simply a much more appealing figure than McCain on so many levels, and Republicans (and the media) put a lot of emphasis on those kinds of qualities in selecting a candidate -- they're almost never about issues. And they recognize, even if they don't admit it, that Obama is a rockstar and McCain...isn't.

 So the idea that a candidate as arid and sterile as John McCain could possibly be competitive against a young, dynamic figure like Obama goes against logic. But McCain is competitive, because (a) Republicans will still line up behind their guy, and (b) Obama's tacks to the right have been counterproductive, and made him look like a flip-flopper.

  If Obama returns to the politics of distinction that served him so well in the primary, and stops assuming that the media will be "fair" to him (it won't) and adjusts his messaging accordingly, then I think he WILL pull away. But if he keeps on with his muddled centrism, he's going to make this election closer than it needs to be.

  He also let McCain get a step ahead of him on gas prices, which was a serious tactical mistake...  

"We judge ourselves by our ideals; others by their actions. It is a great convenience." -- Howard Zinn


Fivethreeeight is more unstable (0.00 / 0)
Nate Silver admits his model's ratings are wont to change rapidly. It seems to overreact to small changes which might be a pattern but might just be standard statistical error. I suspect this is a hangover from his baseball statistics, which after all lack a MoE.

It's also worth considering that Nate appears to consider toss-up to stretch from around 30-70% and lean to be about 15-30% and 70-85%.

Forgotten Countries - a foreign policy-focused blog


Volatility (0.00 / 0)
Volatility in probability is not necessarily a sign of being wrong though. That does happen in some environments. Like Texas Hold 'Em. And politics :)

[ Parent ]
538 Misconception (4.00 / 4)
Chris, you write "McCain does not have a 6% chance to win [Massachusetts] if the state if the election were held tomorrow."

This is a common misconception about 538 -- and it's one of the key differences between 538 and Pollster. Nate is not predicting the chance of winning if the election were held tomorrow. He's predicting the chance of winning the election that is held in November. His model runs 10,000 simulations for what can happen over the next four months based on current polling, and the model includes scenarios based on historical data, where the election breaks in one direction or the other in a big way. Yes, Obama has a 99.9% chance of winning Massachusetts tomorrow, but that's not the point. Nate's model predicts that in 600 of the 10,000 simulations, something unexpected happens that causes a big swing in the national polls. All he's saying is that a 12% lead four months before the election in a state like Massachusetts is a pretty safe lead. How safe? 94%. Seems right to me.



differences in methodology (0.00 / 0)
As noted, 538's numbers are a prediction of November while Pollster's are "if the election were held today". 538's model has all sorts of widgets and tweaks such as (but not limited to):

1) assumes tightening of the race over time
2) uses changes in national polling to predict movement in states that haven't been recently polled
3) uses all sorts of historical regression to predict movement in states that haven't been recently polled

McCain people were going nuts a while back and Nate told them to wait until McCain got a bounce in the national polls. Now they're happier, but if today's Obama bounce holds for a few days, the supertracker should start going up again.

In my opinion, the most alarming factor this week is the rash of swing state polls where McCain is improving in spite of his campaign being such an embarrassment and him repeatedly coming across as such a senile fool.


Those numbers are already shifting back into Obama's favor (0.00 / 0)
The rash of polls  also has a lot to do with the set of assumptions. Are we looking at a reallignment year or not? Are we looking at 2004 or 2006? Or something far more to the progressive end? Anecdotally, I suspect the answer is we are looking at a shift toward the progressive end. The pendulum eventually swings in American history. I think we are seeing that now. I am not sure how they models are suppose to accurately reflect these shifts that aren't based on prior election outcomes and party identification.  

[ Parent ]
Was it really a "rash" of polls? (0.00 / 0)
Wasn't it only those two Quinnipiac polls (for CO and MN) that showed any swing-state movement towards McCain?  

And contra that, there was the Gallup "Obama increasing his lead in all swing states" poll, but it got lost in the shuffle because it didn't break out the numbers state-by-state.


[ Parent ]
That maybe true (0.00 / 0)
But, I assumed their argument was true in my response rather than question its validity.  

[ Parent ]
That maybe true (0.00 / 0)
But, I assumed their argument was true in my response rather than question its validity.  

[ Parent ]
At TPM, (4.00 / 1)
Josh is claiming the swing state movement is mostly about McCain dramatically outspending Obama on advertising. Once Obama takes to the airwaves, I think we'll start seeing movement in the other direction.

(And yes, there have been diaries on that here, too. Maybe I'm getting my uberblogs confused.)


[ Parent ]
Obama bought $5 million of national time (0.00 / 0)
On NBC, during the Olympics.  No one has used national advertising in the Pres since Bob Dole.  That's what money can do for you.

There is an assumption he will have named his VP by then as well.  

John McCain--He's not who you think he is.


[ Parent ]
Obama Bounce (0.00 / 0)
This dKos diary links to several polls that show a 4-5 point bounce for Obama.

http://www.dailykos.com/story/...


Fuzzy math in Silver's simulations (0.00 / 0)
I'm not sure what it is that he's doing, but 538's projection assign a higher probability than I think is warranted to extremely low probability events. For instance, does anyone think that current polling means Obama has a 3% chance of winning Kansas? Or McCain a 3% chance in New York?

I think Pollster also incorporates data over a longer time horizon.

I would call the election "on the brink of being competitive" but not actually competitive. Obama is on track to hold the Kerry states plus Iowa with no sweat anywhere except Michigan and New Hampshire. He has four outs: Virginia, Florida, Ohio, and Colorado+New Mexico. McCain has only two Kerry states where he's at all competitive: Michigan and New Hampshire. He needs to push one or two more states into competitive territory in order to make things close. In theory Pennsylvania, Wisconsin, or New Jersey are his best bets, but none of them are providing any help.


We have to understand what 538 is doing (4.00 / 4)
Of course there is not a 3% of chance of Obama winning Kansas in an election held today.  Current polls make that clear.

But, if there were similar elections held every 4 years for the next 120 years, yeah, I think the Democrats would manage to win Kansas once.  One of the Republican candidates would collapse completely.  Notice that Kansas never decides the election, since in a scenario where McCain loses it, he loses most states.

Republicans, after all, won New York and Massachusetts only six elections ago.  There must be some chance McCain fails like Mondale.

 

New Jersey politics at Blue Jersey.


[ Parent ]
Umm... (0.00 / 0)
59.5%

If that were 60 - 40 you'd call that a 20 point win which is blowout territory.


538's Problem--An Atypical Election (4.00 / 1)
I've been thinking about this myself for a couple of weeks now, and I think that BlatantLiberal has it exactly right--Nate is not taking into account the actual fundamentals of this election, which is shaping up to be a classic realigning election.  Such elections are quite rare--the last true one we had was in 1932--so it's really not possible to build the sort of data-rich model that Nate uses around them.

Others have correctly pointed out what it is that Nate is trying to do, and Hopeful in NJ get's it exactly right:

Of course there is not a 3% of chance of Obama winning Kansas in an election held today.  Current polls make that clear.

But, if there were similar elections held every 4 years for the next 120 years, yeah, I think the Democrats would manage to win Kansas once.

In short, the bottom line is that Pollster.com and 538 are doing two different things, not just taking different approaches to doing the same thing.  

There are three things that I think are important to keep in mind:

(1) Right now, Obama is solidly ahead on the state level.  He holds the states he needs to win, and has McCain battling to hold onto states that shouldn't be competetive (Virginia, Montana, etc.)

(2) The fundamentals--shifts in partisan ID, levels of enthusiasm, campaign organization, etc.--are even more heavily against McCain.

(3) The press is still heavily pro-McCain in a manner that's going to become incredibly hard to maintain all the way to November, no matter how much they love the guy.  And as the strain starts to show more and more, some of that love is going to wear thin.

(4) The combination of all the above makes it highly likely there will be a breakthrough moment--or possibly two--somewhere in the last 10 weeks of the campaign.  Both an Obama breakthrough and a McCain meltdown seem pretty good bets, either of which could mean a shift of 8-12 points.  Thus, while a 53-47% election is quite likely, a 60-40% one is quite plausible as well.

Of all these factors, 538 really only does a good job of picking up #1.  The other factors reflect modalities of how the larger realignment dynamic filters into the presidential race, which Nate's model simply can't capture, because the data isn't there to do it.

It will be interesting, however, to see how his model responds as we get closer and closer to Election Day.  It's altogether possible that it will become significantly more accurate, once enough of the realignment effects make themselves felt in the polling data.  But that could very well not happen till mid-October or so.

"You know what they say -- those of us who fail history... doomed to repeat it in summer school." -- Buffy The Vampire Slayer, Season 6, Episode 3


Pollster race classification (4.00 / 3)
These are great comments.  As several say, the big difference is that 538 uses demographic and past electoral data to make a projection of votes in the states as of NOVEMBER, while at Pollster we are characterizing where the races is AS OF TODAY.   We aren't trying to make the Pollster classification a projection until November. If we were, then whatever method we used would also show a tightening of the race back toward past outcomes.

We all do this in our heads right? If the current polls show Arizona tight, we discount that. And when recently VERY close states like Wisconsin are showing a 10 point or more Obama lead, respect for history should make you think that is likely to tighten in the fall. So you don't need the statistics to believe that Nate's model is doing something like the right thing in bringing the polls back a bit towards past outcomes.

At Pollster, we just want to classify the state of play today. Even if Wisconsin IS likely to tighten, it is important for the campaigns to respond to what is happening there now-- either for Obama to capitalize and consolidate his lead or for McCain to respond in ways that narrow that.  So I think the current state of the states that we do is important for understanding the course of the campaign and candidate efforts. But I want to be clear we aren't making projections for November in the same way Nate does.  We might sometime, but not now.

The last technical point is that when we say a state is outside the 95% confidence interval, that doesn't mean a 95% change of winning the state. It means a 95% chance that the candidate is ahead there now. They might be ahead by a little or a lot, but we are quite certain they are ahead. That doesn't translate easily into the "probability of winning the state" in Chris' table. So as a result the comparison of our confidence intervals with Nate's probability of winning isn't an apples to apples comparison.

Easy bottom line: If the polls decided the winners today, Obama has a substantial EV lead (our state of the race) but it is likely to be a closer outcome in November (Nate's projection model.)

Thanks for a stimulating post and comments.

Charles

Charles Franklin
Co-Developer, Pollster.com


election-projection.net (0.00 / 0)
election-projection.net uses an approach similar to Pollster.com.  It takes the current polling results and uses simulation to find the probability of McCain or Obama winning IF THE ELECTION WERE HELD TODAY.  This gives you the current state of the race, and allows you to track movement over time.

If you are interested in probability graphs that use the same methodology as Pollster.com, see http://election-projection.net


[ Parent ]
USER MENU

Open Left Campaigns

SEARCH

   

Advanced Search

QUICK HITS
STATE BLOGS
Powered by: SoapBlox