Obama's Chances in the Electoral College? A New Projection Method

by: tremayne

Tue Aug 26, 2008 at 08:15


Yesterday I wrote about the 112 Obama vs. McCain polls conducted since Obama captured the Democratic Nomination in early June (listed at Pollster.com). Of those, Obama was ahead in 101 and tied in 5 more. If we split the ties between Obama and McCain, Obama has won more than 92% of the match-ups. It would be tempting to conclude, were the most important "poll" held tomorrow, Obama would have a 92% chance of winning it (excluding GOTV issues and assuming an even split among undecideds).

This doesn't work because, in part, the race may be tighter now than it was 2 months ago. But as I pointed out yesterday the few polls McCain has won or tied are spread throughout the summer rather than mostly occuring in August. The race has shown, on average, about a 3-point lead for Obama with occasional drifts up to 6-8 point leads or down to 0-1 points leads. So far, these moves have not been sustained, the margin continues to revert back to the mean. Whether the most recent tightening is temporary or not we don't know but reverting again to the mean is the most likely result.

Another reason we can't give Obama a 92% chance of winning is, of course, that the national vote isn't determinative. Al Gore won a half million more votes than George Bush and it (maybe) wasn't enough.

But we now have enough state polling to calculate Obama's chances in 51 separate contests and estimate his electors proportional to those odds state by state. Short version: Obama is projected to exceed the necessary 270 electors by more than 50. Follow along for the long version.  
tremayne :: Obama's Chances in the Electoral College? A New Projection Method

I used a simple method to calculate Obama's chances of winning a state. If, for example, 10 polls have been conducted in a state since early June and Obama has led in 5 of them, his odds of winning that state are 0.5 (ties are divided between the two candidates). For that state Obama is then awarded 0.5 times the total number of electors for that state. All data came from Pollster.com.

A few locations (CT, DC, HI, VT and WV) had no polls this summer but there is no doubt which candidate will win those states and the electors for each awarded fully to likely winner.

Using this method Obama is awarded all the electors for such unsurprising states as California, Illinois, New York and many others. But there are some surprises. Obama has led in all (100%) of the summer polling for these states which have been close in the past:

Iowa

Michigan

Minnesota

Oregon 

Pennsylvania

Washington

Wisconsin

Here are other states where the polls show Obama sometimes ahead:

New Hampshire: Leads in 89% of polls

New Mexico: 86%

Montana: 75%

Colorado: 71%

Ohio: 71% 

Virginia: 68%

Nevada: 44%

Florida: 33%

Missouri: 33%

Arkansas: 33%

Indiana: 25%

North Dakota: 25%

South Carolina: 25%

North Carolina: 14%

Alaska: 14%

Arizona: 14%

There is only 1 state the Obama campaign has talked about competing successfully in where no polls have shown him in the lead: Georgia.

If you multiply the 51 percentages by the number of electors for each state/district and add it all together you get Obama 321, McCain 217. These figures are fairly stable over time but slightly closer than one month ago when I first calculated it. Then Obama had 337 projected electors.

The point here is that the occasional poll showing McCain ahead should be taken with a grain of salt until a pattern emerges and in most cases it doesn't. For example, the blogosphere was abuzz last month when a USA Today/Gallup poll showed McCain up 4 points. But guess what, the "mean" immediately reasserted itself: Obama was ahead in the next 7 consecutive polls by an average of more than 5 points. Things don't always revert to the mean but that is the most likely result.

Another caveat: while the odds remain strongly in Obama's favor now that could change if events (such as conventions, debates, etc.) and increased attention by late-deciding voters combine to cause sustained movement in public opinion. So, as always, it is best to work on the assumption that every last vote will count. Because, you know, it could. Ask Al Gore. 

 

 

 

Tags: , , (All Tags)
Print Friendly View Send As Email
Obama's Chances in the Electoral College? A New Projection Method | 29 comments
http://www.fivethirtyeight.com/ - 'nuff said (4.00 / 2)
Sorry, but I think the methodology there is just a bit better thought out, and, in consequence, the projection just a tad likelier than this seat-of-the-pants method.
I was thinking the same thing, tdub - (0.00 / 0)
I think this is an interesting exercise, Tremayne - and your point about taking McCain-is-leading polls with a grain of salt is well taken.  

But the bottom line is, your analysis doesn't have a predictive function, just a diagnostic one (how much HAS Obama likely been ahead in each state, over this period?).  

Nate's work at fivethirtyeight is statistically and conceptually a more appropriate way to project election results.  

[ Parent ]
True, But (4.00 / 3)
The relative simplicity and transparency of this method shows rather dramatically how utterly specious the "race is tied" or even "McCain ahead!" narratives are.

There are two important lessons here:

(1) You don't need rocket science to figure it out.

(2) You don't have to depend on dueling "experts", and thus still leave the "question" "undecided".

"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

[ Parent ]
Somehow (4.00 / 1)
I doubt the "race is tied" narrative is a product of the media misinterpreting the polls.  It could be, and certainly most of the cable news hacks strike me as a tad bit intellectually incurious these days, but it's probably far more likely that they are fully aware their narratives are incorrect, but are pushing them anyway to drive ratings.  

It's sort of like Bush, really.  No one elected president can be that abysmally stupid, so he must logically be just evil (more likely corrupt, but you get the idea).  

[ Parent ]
Tools And Fools (0.00 / 0)
I think it's both.

And this shows them both up.

For Bush, see The Bush Dyslexicon.  Written pre-9/11, but still indispensible.

"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

[ Parent ]
Different approaches (4.00 / 3)
538 is focused on guestimating based on prior demographics, etc likely outcomes based on that data toward the future.  The problem with 538 is that its based on modeling which may not be as predictable this year. It depends on demographic assumptions and turn out models.Things they can't answer- What happens if Obama's turnout model is extremely sucessful, which is some states changes to outcome with a small amount of change in new voters? They can't answer that from what i can tell. if they can, I would love to see that argument because I've asked and been told it can't be modeled into a predictive system.

However, this diary's analysis is to gauge whether claims of a tie are accurate given all available actual present polling data. I don't think either is wrong. Just seeking to answer different questions. THey are also both flawed just because of what is trying to be gauged- what will the real effect of race be? What about the fundamentals favoring a democrat this time? What about Obama's GOTV which is 3 times or more better than McCain's? What about McCain's money disadvantage (yes he has the RNC but that leads to coordinating issues that the Dems faced last time in a tight election). Etc. None of those things are there in the polling data.

[ Parent ]
I've seen thoughtful discussion of turnout models on 538... (0.00 / 0)
and in the primaries, they were very successful in predicting heightened turnout in states such as North Carolina and Oregon, though those, too, were abnormal results.

I'll be honest, I've seen the light and am buying what Nate is selling!  But in any case, I have found 538 to be proactive in not committing a fallacy and assuming that what was, will be.  

[ Parent ]
Careful about the effects of the GOTV! (0.00 / 0)
"What about Obama's GOTV which is 3 times or more better than McCain's?"
During the primaries, yes. But in November, the republican machine will be working for McCain, and that's on a higher level than Clinton's efforts. Don't underestimate the GOP's  ability to motivate voters to make the "right" choice, the 2000 and 2004 elections showed that they are very effective. Obama's and the DNC machines combined may be a bigt ahead this year, but this is unproven yet, and relying on this could be a horrible mistake.
[ Parent ]
bush 2004- nuff said for arguing about this- look it up (0.00 / 0)

[ Parent ]
I did, but it only raises doubts about Bush 2000 (0.00 / 0)
Read what the republican GOTV guru said about their operation:
STUTTS: Going into the final 72 hours of the 2000 presidential race, then-Governor Bush had a 5-point lead in the polls. But on Election Day, Al Gore won the popular vote by running a far superior get-out-the-vote operation. GOP leaders vowed to never let that happen again. They created the 72 Hour Task Force (defined as the final three days of the election) and over the next three years, we studied and implemented GOTV programs all over the country. We learned the right way to turn-out voters and we learned some hard lessons. I had the privilege of leading efforts in South Dakota (2002) and Louisiana (2003). By 2004, the Bush campaign knew what worked in regards to grassroots campaigning. We implemented our plan to perfection pushing President Bush to victory by a 51 percent margin and increasing majorities in the United States Senate and House. The program accomplished unprecedented goals, the most impressive being the recruitment of 1.9 million volunteers to execute the Get-Out-The-Vote plan in key Bush/Cheney battleground states.
http://www.mattlewis.org/blogg...

If I understand you right in that you wanted to imply that the Bush 2004 GOTV wasn't superior, this totally contradicts your view. And I remember having read several other pundits that praised the GOP organisation for beeing able to deliver virtually all party base votes. Where do you get the idea that they were lacking in this regard in '04?

[ Parent ]
they aren't projecting the same thing (4.00 / 1)
538 is projecting the outcome on November 4th, so it allows for a lot of uncertainty about what might happen in the future.

Tremayne is just looking at the state of the race today. I actually think it's a pretty good proxy for the overall current state of the race, and it goes to show how McCain really has to run the table in just about all the true battlegrounds to eke out a win.

[ Parent ]
Actually, (4.00 / 5)
it is best to work on the assumption that every last vote will count. Because, you know, it could. Ask Al Gore.

Every last vote didn't count.

That was the whole point.

"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

And Dems didn't learn from that experience (0.00 / 0)
Despite their 2000 and 2004 candidates becoming victims of the Republican "suppress the vote" efforts, idiotic state laws about the "right" procedure to count the votes, strange errors, flaws, and outright suspicious results, plus the inherent problems of the "winner takes it all" and electoral college system in general, the Dems haven't really focussed on implementing fairer federal vote laws that would assure fairer elections this year. Even after they managed to win the majority in 2006, the Dem congress didn't seem to give a vote law reform a high priority. As a consequence, despite some lame approaches to reform the insufficient and wrongheaded "Help America Vote Act" of 2002, no serious change for the better has been achieved. Makes you really want to bang your head against the wall, when you see liberals so totally inept of doing anything for their own favor.
[ Parent ]
This... (4.00 / 2)
...is great news for McCain!!!!
simpler idea (0.00 / 0)
The problem is that all but 2 of the 51 contests are strictly winner takes all. If Obama ends up with fewer votes on November 4 in Colorado, he doesn't get 71% of the electoral votes there. He gets 0%. It would be simpler and more accurate to do like Markos does periodically on Daily Kos and just assign all the electoral votes in a state to whoever is leading in the current polling averages. The end result is similar: this is a race that isn't tied but rather leans Democratic with a fair chance of being an electoral college blowout.
re: simpler idea (0.00 / 0)
it may be simpler but it's much less accurate. it's in fact misleading. Obama simply doesn't have the same chance of winning Colorado or New Mexico that he does of winning Illinois or New York. To pretend he does paints a prettier picture of the race than reality dictates. Read up on 538, it's fun.
[ Parent ]
Well, even simpler than that, if we're still just looking for a back of the envelope thing (4.00 / 1)
would be to allot the EVs of every state where there is a larger than, say 5-7% lead to the appropriate candidate.  Then, allot the remaining EVs based an a weighted average of the polling in all of the remaining states (or failing that, simply the national average).  This way, you account for the winner take all business in the base states, but you also account for the randomness of how the swing states will fall.  

In the end, there are a million different ways to get a snapshot, and they all have their advantages and disadvantages, and the key point is to be honest about wht methodology you are using and what the alternatives could be.

[ Parent ]
Yet a better way (4.00 / 1)
First of all, as many people noted, this method, while simple, may be pretty inaccurate. Indeed: in case there is a state for which 4 polls have been taken, Obama leading in two of them by 10 points and 8 points, and McCain leading in the other two by 1 point and 2 points, then this method would give each candidate a 50% chance of winning the state, while the largest leads of Obama would suggest that he is in fact more likely to win that state than McCain.

Next, fivethirtyeight.com is a really great site, even though the methodology is fairly complicated.

Personally, I prefer the site http://election.princeton.edu, which only examine the current situation, and, from the latest polls, deduces like you a win probability for each state IF the election were held today. He then calculates the median electoral vote expected as well as the distribution of outcomes (without resorting to a finite number of simulations like 538 does, therefore making it more accurate). That site describes in details the differences between his method and fivethirtyeight actually.

neat site (0.00 / 0)
Neat site thanks for the link
[ Parent ]
What's Significant Is That They Generally Agree (0.00 / 0)
The margins may vary, but the overall indicattions are generally in sync.

Okay, so I base this on eyeballing, rather than running a regressional analysis.  We can now try 10 different ways of measuring how closely these different measures agree with one another.

Say, who's got the angels?  I've got the head of a pin!

"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

[ Parent ]
Not so much (0.00 / 0)
Well, they don't really agree: the method presented here gives  a result of 321-217.
Fivethirtyeight.com gives  284.7 - 253.3
election.princeton.edu gives 304 - 234

Sure all three of them show Obama ahead, but the results vary by almost 40 EV, which is far from negligeable, especially if the election becomes much closer.

[ Parent ]
They Are ALL In Agreement vs. Versailles CW (0.00 / 0)
And though I haven't seen this method applied retroactively to July, I daresay they all would show a tightening since then.

That qualifies as "general agreement" for me, and I was a math TV in college who used to tear his hair out over the levels of innumeracy I encountered.

"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

[ Parent ]
history back to Jan 1 (0.00 / 0)
http://election-projection.net has a history chart going back to Jan 1.  It shows Obama hit a high in July after Clinton conceded, and has pulled back from the high since McCain started his attack ads.
[ Parent ]
And So Does Princeton (0.00 / 0)

Same shape at the macro level, less so at the meso, though still fairly decent.

"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

[ Parent ]
differences (0.00 / 0)
The results at http://election-projection.net and Princeton are very similar.  The primary differences are: (1) election-projection shows a broader distribution of EV's.  This is because election-projection has a better model for poll errors.  The Princeton result does not model all sources of poll error, so its results are overly confident (a narrower distribution of EV's).  (2) election-projection does a weighted average of ~6 polls, while Princeton does a median of ~3 polls.  For this reason, election-projection is less sensitive to polling noise.  When there is a major shift in opinion, the results will begin changing immediately, but it will not show the full shift until it is confirmed by several polls.
[ Parent ]
better estimate of poll error (0.00 / 0)
http://election-projection.net is similar to the Princeton methodology, but it includes a better estimate of poll error.  It also ranks the most important states based on conditional probabilities, and includes an interactive simulator if you want to test your own numbers.
[ Parent ]
great (0.00 / 0)
Wow Allen, great site! Didn't know that one.
[ Parent ]
Although (0.00 / 0)
These methods all appear to have a common weakness, in that they are projecting basically smooth functions, when we know that's not the case for at least two reasons.

First, some EV combinations are relatively improbable for purely mathematical reasons.

Second, some are improbable because of how national or regional factors tend to produce coupled results between states that the models simply don't capture well enough.

This is not a major problem, IMHO, for the purposes at hand.   (The smooth functions have a utility that offsets these other factors, particularly since we don't have much of a handle on how to deal with my second point--beyond the demographics that 538 uses, which are certainly a good start.) But it's good to keep in mind, nonetheless, to counter the tendency to imbue these methods with godlike powers.  They are more sophisticated, and they're certainly to be applauded for that.  But their creators are well aware of their limitations, and it's good for us to be, as well.

Remember, models are always intended to isolate certain features of the real world, and give us a superior graps on those features.  This inevitably sacrifices our grasp of other features.  Models are tools, and we get into trouble, one way or another, when we start to forget that.

Hope I'm not being too pedantic here, but these points do seem to get easily lost, or forgotten.

"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

[ Parent ]
computers are good at math (0.00 / 0)
Your major point here is well taken: the models by necessity make certain simplifications, and it is important to know what they are in order to understand what the results mean and what they don't.

On the other hand, when it comes to math, the computers do a better job than the human mind.  Take for example your statement that "These methods all appear to have a common weakness, in that they are projecting basically smooth functions".

If you go to the interactive calculator at http://election-projection.net... and make all but four states 100% for Obama or 100% for McCain, then run sims, you will clearly see the discrete nature of the EV distribution.  The model is factoring into account that some EV outcomes are more likely than others.  It does a better job of this than anyone can do in their head.  You are assuming the curves should be more jagged because in your head you can only keep track of a handful of combinations.  When you start factoring in all 2^20+ combinations, the curves no longer look jagged.  There is no error in the models here--the computer is doing this correctly.

Second, still on the interactive page, crank up the correlation slider at the top.  This simulates certain states rising or falling in tandem.  You'll find that the curve becomes more discrete, just as believe it should.  Again, the computer and the model have this right.  Election-Projection.net's main pages give you the EV distribution if the election were held tomorrow.  This is what some other sites call a "snapshot" of the race.  It includes only a small amount of correlation which is suitable for modeling correlated poll errors.  If you wanted to project the outcome on Nov. 4, it would be appropriate to use larger amounts of correlation.

In summary, the computer simulations are good tools for better understanding how the numbers work.  If you play around with the sims and compare different scenarios, it can help you gain a lot of insight into the probable outcomes, and it might dispel some preconceived notions.  That's what the tools are there for, and that's what they are good at.

[ Parent ]
Obama's Chances in the Electoral College? A New Projection Method | 29 comments





Donate to Open Left




blog advertising is good for you
blog advertising is good for you
LIBERAL INSTITUTIONS
STATE BLOGS
USER MENU

QUICK HITS
SEARCH

   

Advanced Search

Home | RSS/XML Feed | Register | Contact

  © OpenLeft 2007-2010, Powered by SoapBlox