Saturday, September 20, 2008

Presidential Forecast - 9/20 - Cross-currents

WE'VE MOVED! Democratic Convention Watch is now at http://www.DemocraticConventionWatch.com

Obama makes a nice comeback, 278-260, an increase of 8 EVs from our last forecast. But the gain is mostly from one source, 538.com, which increased from 258 to 306.5 EVs, moving from our worst projection to our best. 538 is basically following the national polls, where Obama has moved from a 2 pt deficit to a 3 pt lead, on average, in the last few days. But Open Left, who just 3 days ago was basically in sync with 538, now differs by almost 40 EVs. Why?

Obama's national poll surge has not--repeat, not--resulted in a clear electoral college lead. There has been enough state polling over the past few days that it would have been picked up by now. - Open Left
Except that we've seen consistently that state polls seem to lag 1-2 weeks behind the national polls. I would expect the state polls to start being more favorable to Obama next week, assuming his national lead holds.

Which leads me to the following conclusion: I think the projections which include some "special sauce", especially NBC and CNN, are beginning to show their value by not bouncing all over the place. CNN especially has shown the most consistency, being at 273 in June, moving slowly up to 291 after the Democratic convention, and has since moved back down to 283. Sure, they were maddening when they were showing NM as McCain-Lean when the polls were clearly showing otherwise. But there is something to be said that they have had 7 states always as Tossups throughout: CO, FL, MI, NV, NH, OH, VA.

Colorado remains our tipping point state.

Map changes: Towards McCain: WI: OL-> T. Towards Obama: FL: ML->T; MI, NM: T->OL; NC: M->ML.

Please also check out our Senate Forecast and House Forecast.

The sources are sorted by each projection's estimate of Obama's Electoral Votes (Algorithm at bottom). The states are sorted from Obama-Strong to McCain -Strong. The right column shows a running total of Obama's EVs. States in 3 or more categories: IN, ME, MI, MN, MO, MT, NM, NC, OH, PA, WV, WI - 12 states, the highest number since we've been tracking this.

DCW Presidential Forecast
State
EVs538
.com
FHQCNNEV.
com
RMNBCOpen LeftRCPElect. Proj.EVs
Date

9/20.9/19.9/19.9/20.9/19.9/15.9/19.9/20.9/15.
Obama-Strong (O)

260157160153193172164157154
Obama-Lean (OL)

336563626661554543
Tossup (T)

45156115139327885120141
McCain-Lean (ML)

301175385867885926
McCain-Strong (M)

170149125146189160146157174
Obama Total

293222223215259233219202197
McCain Total

200160200184247227234216200
Obama Est.

307289283280273273268265264

California
55OOOOOOOOO55
Conn.
7OOOOOOOOO62
DC
3OOOOOOOOO65
Hawaii
4OOOOOOOOO69
Illinois
21OOOOOOOOO90
Maryland
10OOOOOOOOO100
Massachusetts
12OOOOOOOOO112
New York
31OOOOOOOOO143
Rhode Island
4OOOOOOOOO147
Vermont
3OOOOOOOOO150
Delaware
3OOOOOOOOOL153
Maine
4OOOLTOOOOO157
New Jersey
15OOLOLOLOOOLOLOL172
Iowa
7OOLOLOLOLOLOOLOL179
Oregon
7OOLOOLOLOLOLOLOL186
Washington
11OOLOLOLOOLOLOLOL197
New Mexico
5OOLOLOLTOLOLOLT202
Minnesota
10OOLTTOOLTTT212
Michigan
17OTTOLOLTOLTT229
Pennsylvania
21OTOLTOLOLTTT250
Wisconsin
10OOLTTOLTTTT260
NH
4OLTTTOLTTTT264
Colorado
9OLTTTTTTTT273
Ohio
20OLTTTMLTTTT293
Nevada
5TTTTTTMLTT298
Virginia
13TTTTTTMLTT311
Florida
27TTTTMLMLMLMLT338
Indiana
11MLTMLTMLMLTTM349
Missouri
11MLTMLMLMMLMLMLML360
Montana
3MLTMLMLMMLMLMLM363
N. Carolina
15MTMLMLMMLMLMLM378
W. Virginia
5MLMLMLTMMMLMML
Louisiana
9MMMLMLMMMLMM
N. Dakota
3MMLMMMMMMLM
Alaska
3MMLMMMMMMM
Arizona
10MMMMMMMMML
Arkansas
6MMMLMMMMMM
Georgia
15MMMLMMMMMM
Alabama
9MMMMMMMMM
Idaho
4MMMMMMMMM
Kansas
6MMMMMMMMM
Kentucky
8MMMMMMMMM
Mississippi
6MMMMMMMMM
Nebraska
5MMMMMMMMM
Oklahoma
7MMMMMMMMM
S. Carolina
8MMMMMMMMM
S. Dakota
3MMMMMMMMM
Tennessee
11MMMMMMMMM
Texas
34MMMMMMMMM
Utah
5MMMMMMMMM
Wyoming
3MMMMMMMMM


538
.com
FHQCNNEV.
com
RMNBCOpen LeftRCPElect. Proj.....


















































































Notes:
538 - FiveThirtyEight - Safe and Likely mapped to Strong (O or M), Lean to Lean (OL or ML), Tossup to Tossup (T)
CNN - Safe mapped to Strong, Leaning to Lean, Tossup to Tossup
Elect. Proj. - Election Projection - Solid and Strong mapped to Strong, Moderate to Lean, Weak to Tossup
EV.com - Electoral-Vote.com - Strong mapped to Strong, Weak to Lean, Barely and Tossup to Tossup
FHQ -
FrontLoading HQ - Solid mapped to Strong, Lean to Lean, Tossup (Dem and Rep) to Tossup
NBC - Base mapped to strong,
Lean to Lean, Tossup to Tossup
OpenLeft - Solid mapped to Strong, Lean to Lean, Tossup to Tossup
RM - Rasmussen - Safe and Likely mapped to Strong, Lean to Lean, Tossup to Tossup
RCP - RealClearPolitics - Solid mapped to Strong, Lean to Lean, Tossup to Tossup

The overall projection is just a straight average of each projections' estimate of Electoral Votes (EVs) for each candidate. For each projection other than FiveThirtyEight, we give Obama 100% of the EVs in a state that is solid for him, 80% of the EVs for a leaner, 50% of the EVs for a Tossup, 20% of the EVs for state that is McCain-Lean, and 0% of the Solid McCain states. Exact opposite for McCain. For FiveThirtyEight, we use their overall estimate of Obama's EVs, not the state-by-state categories.

Comments (20)

Loading... Logging you in...
  • Logged in as
Login or signup now to comment.
I would note that both 538 and Rasmussen have their own flavor of special sauce, but for both the special sauce is a mathematical component of their forecast. For NBC and CNN, the special sauce is humans who refuse to accept the data. While they may ultimately be correct in where the race will go, their stablility is more just being stubborn than anything else.

What I would like to see from the sites and 538 sometimes notes it in the commentary would be a list in order of the states with how much they are currently differing from the national numbers. In other words, how much above 50% nationally does one candidate have to reach to swing this state the other way. Given the volatility of the national poll, the "swinginess" of the states is more important than how a state currently plays.
Reply
1 reply · active less than 1 minute ago
I'm not sure Rasmussen has a mathematical formula. What they say is:

"This Balance of Power Calculator aggregates data from a variety of sources to provide a comprehensive assessment of the state-by-state race for the White House. Data inputs include the latest Rasmussen Reports poll in a state, an average of the latest polling from other firms (the “538 Average”), Rasmussen Markets data, Intrade market data, the aggregated rankings of selected analysts, the state’s voting history and national trends."

They don't actually say they put that into a formula. Looking at their past shifts, it looks to me like they're really doing the same thing as CNN or NBC, except that they'll explicitly attribute changes to one of the factors. Otherwise they wouldn't have 10 states shift on the same day, as they did on August 22. As I've said before, I like having some of the subjective projections in the DCW list, as they bring in factors that are difficult to quantify.
Reply
538 is not a "special sauce" projection. It's pure mathematics. Yes, they can change the formula, but they always follow the formula. NBC, CNN and Rasmussen have a human making the final decision on each state - so we call them "special sauce" states.
Reply
I think you should consider adding another site to your list:
http://election.princeton.edu/
or at the very least , check it out. They also have an interesting map with the sizes of the states distorted to show their EV weight. I followed this site 4 years ago (different incarnation) and it was quite accurate. (Another nice benefit: they show Obama way ahead.)
Reply
2 replies · active less than 1 minute ago
Oregon Dem's avatar

Oregon Dem · 860 weeks ago

Yes it is an interesting site, providing another way of looking at what would happen if the vote were held today. What was the site that just posts the results based on only the most recent poll? Maybe that should be dropped and this added?
Reply
I don't think any of them post just on the most recent poll. Electoral-vote uses all polls within the same week as the latest poll so sometimes it only has one poll for a state. I believe Open Left uses any polls within the last two weeks but always has at least one poll. Not sure the time frame for RCP, but I think they have a four different pollsters minimum.

I think once we get closer to the election and typically have two or three polls in the major states each week, electoral vote is more reliable because they only consider the "current" polls whereas some of the others assume that "dated" polls still have value.
Reply
We're not going to make any changes to the projections used, so that the historical numbers remain consistent. And as we get closer to the election, and more polls are released, I expect the projections to get more in line with each other.
Reply
I just tried something I've been meaning to do for a while.

I went through Pollster.com and classified states as follows:

Solid for a candidate if the pollster.com average is above 50.

Lean for a candidate if the candidate is the only one to have reached 50 in a poll since McCain became the presumptive nominee in March (Zogby Interactive doesn't count).

Toss-up if neither of the above apply.

The idea is that if a state consistently polls 50-47, regardless of the methodology of the poll or the state of the national race, it's very hard for the trailing candidate to win. But if a state has a lot of polls like 46-40, but the leading candidate never breaks 50, the trailing candidate has a chance.

The results:

Solid McCain:ID, UT, AZ, WY, AK, ND, SD, NE, KS, OK, TX, LA, MS, AL, GA, SC, TN, KY

Solid Obama: CA, IA, IL, NY, VT, ME, MA, RI, CT, DE, MD, DC, HI

Lean McCain: MT, MO, AR, IN, NC, FL

Lean Obama: WA, OR, CO, MN, WI, MI, PA, NJ

Toss-ups: NV, NM, OH, WV, VA, NH

There are a couple of intriguing things buried in there. West Virginia as a toss up is startling. It's true McCain broke 50 in late February, which is just before my cut-off, so it could easily be called a McCain lean. But still, I wonder if Obama should have put some more resources there. Maybe it's not too late.

Also notable is that Colorado is an Obama lean but New Mexico is a toss-up. Obama winning Colorado and losing New Mexico is not a popular parlay, but it's not out of the question.

Those are the most notable surprises, although there are a few others that caught my eye (Maine and North Dakota are safe, New Jersey, Arkansas, and Washingon not entirely so).

Oh, and if you want the bottom line, this method yields 164 Obama-strong, 100 Obama lean, 52 toss-up, 73 McCain lean, and 149 McCain solid, for a total of 264 to 222 and an Obama estimate of 285.

At any rate, it's a little different way of looking at it, so I decided to share. :)
Reply
uplandpoet's avatar

uplandpoet · 860 weeks ago

i am very familiar with west va, having an exwife and a son who live there, and having lived there awhile myself. it is not as white as iowa, they are much poorer, less educated, and generally more disposed to not think highly of black people. not to cast disparagement on the good , open-minded people of wva, but there are quite a few of them that are not as open=minded as say, your average white mississippian.
Reply
1 reply · active less than 1 minute ago
...and yet they don't poll that way recently. A recent poll from ARG (admittedly not the most reliable pollster) gives McCain 49, Obama 45. Obama leads among women, Democrats, independents, and those under 50. If those results were from North Carolina or Montana, we'd be saying the state was in play. Was that poll a fluke? Blankenship recently had it McCain 44 to Obama 39, and THAT poll was taken just after the Republican convention, at the height of the Palin bounce.

I think it's worth it to send Clinton (either one) there for a day, and to run some ads.

Assuming that people are closed minded when they're showing indications of keeping an open mind is a good way to squander an opportunity.
Reply
uplandpoet's avatar

uplandpoet · 860 weeks ago

you know, if he has the people and the money, sure go for it, but not at the expense of colo, ohio, fla, nev, nh and va.

they certainly OUGHT to be giving obama a look in the beautiful state of WVA, and they vote D for most statewide offices, so maybe you are right, but....
Reply
I like your thinking SLCScott. I usually don't put much stock into what people say "they ought to do" on blogs but I happen to be leaning that way with you here.

BTW: WOOT SLC!
Reply
Btw, that is a way cool map! WE TOTALLY NEED SOMETHING LIKE THAT HERE! It gives you a different perspective on America.
Reply
My take is that, of all of the "EV aggregator sites", 538 will reflect trends more quickly and more accurately. Unlike most of the other sites, they acknowledge that (1) national polls matter, and (2) one state's results can change the nature of another state's predictive result if the demographics of the two states are similar. They also seem to look pretty closely at the crosstabs when modifying their state-by-state predictions.

The downside of this methodology is that we won't get a good sense of stability in the predicted result. So we feel jerked around with the poll watching. But applying non-mathematical overlays onto the data will only make the trends more stable and not more accurate.
Reply
1 reply · active less than 1 minute ago
"But applying non-mathematical overlays onto the data will only make the trends more stable and not more accurate."

That would be true if we had a large, controlled data set. But in US Presidential elections we don't. The system has too many variables compared to the amount of data; it's not possible to model it without making lots of assumptions. Those a priori assumptions are just as subjective as the subjective judgement used in the non-mathematical projections. The advantage of the mathematical projections is that they are unbiased, in the sense that once the a priori rules are set, the projection can't be finessed to favor one candidate, make the race look closer than it is, or support earlier analyses.

As for 538's model in particular, I love reading Nate's stuff, but his model seems over-leveraged to data. A poll in say, South Carolina can have the following effects:

--changes the model's perception of the overall state of the race
--changes the model's perception of similar states
--changes the model's perception of how various demographics are leaning

That's a triple whammy that might cause North Carolina results to jump, for instance. And the South Carolina poll that triggered it all might be an outlier. Giving 600 people in a poll the ability to switch the whole country around seems like bad statistics to me.

The test of this is in the 538 results. 538 aims to be different from some other projections by accurately predicting the result on Nov. 4, rather than the state of the electorate now. And yet it has swung from one extreme to the other and back again. That means that at some point it failed badly at its stated goal.

My suspicion is that Nate will seriously retool the 538 model to get rid of this swinginess before 2012, but for now 538 is an exercise in overstating the importance of outlying state polls.
Reply
I agree with most of what you say. And Nate considereed "devaluing" the polling data during and immediately after the conventions, as the predictive nature of the polls was poor. And the swinginess was due to excessive movement during that time.

But (I believe) that most of the "bounciness" of the prediction is reflective of the fact that the national and state polls range from McCain +2 to Obama +4, and that's probably where the final vote will lie.
Reply
OK I want to point out yet another thing about the website and the funky looking map mentioned above http://election.princeton.edu/ You can go there and it gives you a map based off of current polling data AND a map that shows the effect of shifting just 2% in either direction in each state. It is quite remarkable what a an effect that 2% has.
Reply
greenwoodindy's avatar

greenwoodindy · 859 weeks ago

One television station in Indiana which normally I watch for election predictions that have been fairly accurate at this stage in the game as to who is going to "win" Indiana's electoral votes is WTHR. Their polling in conjunction with the Indianapolis Star is actually for the first time that I've ever seen showing a Democrat leading (47% to 44%) the presidential race in Indiana. I know there are still some "undecided" and "others" in the mix, but the poll is worth a look. The survey/poll is at http://www.wthr.com/Global/story.asp?S=9026832
Reply
New forecast is up.
Reply

Comments by