For the first time since we began tracking our results in early March, both Democrats are now slight favorites in their prospective races against John McCain.
The cheery news for the Democrats is in Minnesota, where a new Star Tribune poll has Barack Obama leading John McCain by 13 points, and Hillary Clinton leading him by 9. This continues a pattern in which traditionally blue states have begun to turn bluer. The Democrats, and particularly Barack Obama, might be able to get away without a serious investment of resources in states like Minnesota, Washington and New Jersey.
The other poll out today is from Nebraska, where Rasmussen has John McCain 11 points ahead of Barack Obama and 23 points ahead of Hillary Clinton. Although Nebraska hadn't been polled since February, that's about where our regression model had figured it was at.
Nebraska is one of two states (Maine is the other) to split some of its electoral votes by Congressional District. In 2004, John Kerry ran 11 points better in NE-2 (Omaha) than he did in the state as a whole, and 6 points better in NE-1 (Eastern Nebraska) than in the state as a whole. Meanwhile, he ran 18 points worse in NE-3 (Western Nebraska).
What this implies is that if Obama is about 10 points down in Nebraska overall, NE-2 in Omaha should be considered a toss-up, whereas NE-1 may be competitive. There are definitely scenarios where this is relevant. For example, if Obama wins Kerry states + Iowa + Colorado -- one of his more plausible electoral combinations -- he would be sitting on 268 electoral votes. Winning NE-2 in Omaha would get him to 269 electoral votes, at which point the tie would probably be broken in his favor by the incoming House of Representatives. We account for any and all such scenarios in our simulations.
Sunday, May 18, 2008
Today's Polls, 5/18
-- Nate Silver at 9:46 AM
Labels: minnesota, nebraska, today's polls
38 comments
Your simulations make no sense. You have McCain getting 272 EV's and getting 50.2% of the popular vote. Yet, you have Obama having a better chance of winning. Your bias is starting to seep through.
Brandon, a couple months ago, he had Obama with more EVs, but McCain with a better chance of winning. Now, it's for the same reason: One candidate WINS MORE, but the other WINS BY A LARGER MARGIN WHEN THEY WIN. Hence, their AVERAGE electoral votes will be greater, even if they have a lower chance of winning. Pablano's model is just math.
Indeed. Did you know that the programmer who designed EXCEL's random-number generator has donated $200 to Obama's campaign?
We've been seeing results like these for several weeks now, in which a candidate with slightly fewer average electoral votes is nevertheless favored to win. Sometimes this has worked in Obama's favor and sometimes in McCain's favor. It can happen fairly easily if the distribution of electoral vote totals is asymmetrical and a candidate tends to win by a larger margin when he wins than he loses by when he loses, or vice versa.
For example, let's say we get to Election Day. All of the states go exactly as they did in 2004. That's 251 electoral votes for Obama and 286 for McCain. Except Ohio. Ohio is in play, and Obama has a 60 percent chance of winning it.
Under this scenario, Obama will win 251 electoral votes 40 percent of the time, leaving him 19 votes short, and 271 electoral votes 60 percent of the time, just putting him over the 270 he needs to clinch the election.
Obama's average number of electoral votes in this scenario is 263.0 -- fewer than half. However, he would be a 60% favorite to win the election, because he wins the election any time that he wins Ohio. If the election turns out to be as close as we're projecting it, such scenarios are fairly easy to come by.
And in terms of a candidate losing the popular vote but winning in the electoral college, we don't need to go back very far to see an example of that.
I don´t understand why that poll got him so much ground- his chances in MN were good before. His Win% in MN moved up about 10%, when I am right, and his chances in NE down about 15%, so why were this polls so good for him?
Did his regression numbers everywhere improve by that? (I know they improved, maybe except the Interior West, but did they improve by large margins?) I have to update your regression numbers anyway, so I´ll see that in 5 minutes...
Rasmus, I'm pretty sure that Pablano's model has some randomness involved: He takes the candidate's chance of winning each state, puts that into a program, and runs 10000 simulations. That results in some minor day-to-day margin variation noise. Don't worry too much about it, though, I'm pretty sure most polls, which average a sample size of around 400-600 respondents, have a greater margin of Error than Poblano's model.
Also, I should note, MN has a greater effect on his chances than NE, because MN probably has twice as many electoral votes.
Rasmus, I'm pretty sure that Pablano's model has some randomness involved: He takes the candidate's chance of winning each state, puts that into a program, and runs 10000 simulations. That results in some minor day-to-day margin variation noise. Don't worry too much about it, though, I'm pretty sure most polls, which average a sample size of around 400-600 respondents, have a greater margin of Error than Poblano's model.
Also, I should note, MN has a greater effect on his chances than NE, because MN probably has twice as many electoral votes.
sensible person,
I know how his spreadsheet works.
I rebuilt it with Excel MS- I don´t have that things like the Swing State Analysis, and I am weighting some things a little bit different than 538 does, but I also have that poll and regression weighting system and I also run the election 10000x with random numbers.
Nevertheless: He gained a bit ground in MN, he lost ground in NE, his regression values got better in VT and WA and MA and other states, and worse in FL, SC, IN and other states.
But none of this things has enough impact to give him a gain of 1,5% in the Win%.
Variation noise is there, but with 10000x runs it should go down to about 0,05%-0,1% in the popular vote, about 1 EV and about 0,5% in the Win%.
Oh, and in my sheet his Win% actually moved DOWN 0,15% with that changes- the regression changes are more important than the polls, and his IN regression went down to McCain +6,4 from McCain +4,2- no idea why.
That 0,15% is random noise, for sure, but certainly that MN poll did not move up his Win% so much- even if the sample size is big and the poll gets a lot of weight.
Are you sure that the House of Representatives would select Obama in case of a tie? Each state gets one vote in the House, and I could see McCain getting the necessary 26 states.
Oh wait, I've answered my own question. I was thinking in terms of electors rather than the Representatives in the House, who of course will be majority Democrat.
Semblance,
If you count out the states at the moment, there are 27 state delegations with a majority of Democrats, 21 with a majority of Republicans, and 2 tied (Arizona and Kansas).
And there are some seats Democrats look reasonably likely to pick up this fall that would earn them additional states: Don Young's seat in Alaska (which would flip AK from R to D), Rick Renzi's open seat in Arizona (which would move AZ from a tie to D), either of the Heller or Porter seats in NV (flip from R to D), Heather Wilson's open seat in NM (flip from R to D), maybe even Gary Trauner winning the open seat in WY that he almost won last time against Cubin (flip from R to D).
It's not a 100% sure thing, because a few red-district Ds have indicated in previous races that they'd probably vote with their district if the election went to the House -- Gene Taylor in MS (who would now flip MS from D to a tie if he followed through) and Stephanie Herseth Sandlin in SD (who would move SD from D to R) come to mind. But there's some margin for error, especially given that there are currently two tied states, and that if neither side gets to a 26-state majority the Vice President Elect, selected by the almost-certainly-Democrat-controlled Senate, will act as President until the logjam is broken. The pressure on red-district Ds, especially, to vote for McCain will be a lot less if all they're accomplishing is making a Dem VP nominee the acting President instead of making Obama President.
Re Nebraska's districts, in 2004 Kerry only ran 5 points ahead of the state total in NE-02. Maybe you mean 11 ahead of the NE-03? In any case, NE-02 is still fairly solidly McCain territory based on that poll.
Wait - I get it. He ran 5.5 points better, reducing his margin by eleven.
And while I'm at it, I'll use my master degree in stats to make some nerdy points.
Brandon- if you want to read more on why Obama is showing as favourite but getting less votes, the difference here is one between mean and mode.
Poblano - do your simulations take into account the fact that results in states are highly correlated? For example, in the 38% of simulation runs that show McCain winning Pennsylvania, he should win Ohio not 57% of the time, but probably more like 80-90% of the time. I'm not sure how you would take that into account, but just to raise the point.
Good point by the last poster, i was thinking the same thing. For instance when Obama thinks he will perform well in NC, and he could very well win it. BUT there is no way NC will be a swing state, because if Obama does win NC than he will also most likely win ohio and the southwest.
Just out of curiosity, what percentage of the simulations end up with both candidates receiving 269 electors? I know there are a number of plausible scenarios (in addition to the one you mention) for such a result. I think I read somewhere that you lump these into the Democratic total because of the likelihood of Democrats retaining their state-by-state House majority, but I'd be interested if it's more like .1% or 1%.