MyDD :: Chances of a Kerry Victory: 98%?

Chances of a Kerry Victory: 98%?

by Chris Bowers, Wed Jul 21, 2004 at 02:27:51 PM EST

Professor Sam Wang of Princeton University has conducted an interesting analysis of the 2004 Presidential Election that is based on a higher level of statistics than I can quickly grasp. Still, I think I get the gist of it, and I can't find any real holes in his analysis (emphasis in original):

Counting the last six polls the current probability of a Kerry win is 98 percent.

Counting only the last three polls the probability is 99.98 percent. 270 electoral votes (EV) are needed to win. The current 50th percentile outcome for Kerry is 296 EV. The 95 percent confidence band is 270-322 EV. (...)

Current probabilities by state (rounded to the nearest percentage point): AR 15, AZ 0, FL 98, IA 100, ME 98, MI 100, MN 100, MO 11, NV 25, NH 96, NM 100, OH 26, OR 100, PA 89, WA 100, WV 18, WI 88. The calculation is currently dominated by the large states FL (98%), PA (89%), and OH (26%). In most high-likelihood scenarios, two of these are needed to get over 270 EV.

His methodology, explained on his page, is simply to calculate all of the possible outcomes in these seventeen states based on the percentages listed. Kerry wins in 98% of those scenarios. The percentage chance to win a state is determined by state polls.

Good news? Sure, but this is based only on recent polls, and polls can change. Right now, few people would argue that if the election were held tomorrow that Bush would win. However, I tip my hat to such solid analysis, and look forward to future calculations.

Tags: General 2008 (all tags)

Permalink :: 15 Comments :: Post a Comment

I read it as 98% chance Kerry'd win today (none / 0)

based on the fact that most polls have him ahead and the odds that he would really be behind today is extremely low. But I read it as more of a snapshot of where we are right now, than a prediction of the actual outcome.

by Wayne in Missouri on Wed Jul 21, 2004 at 03:20:25 PM EST

Re: I read it as 98% chance Kerry'd win today (none / 0)

You are correct. But hey, its still nice.

by Chris Bowers on Wed Jul 21, 2004 at 04:17:58 PM EST
[ Parent ]

Re: I read it as 98% chance Kerry'd win today (none / 0)

I didn't think such inanity could come out of Princeton, but this is completely ridiculous. The polls themselves are pretty suspect with larger-than-hoped margins of error.

If you think that there is a 98% chance he would win today, being behind by 15 points or so on the terrorism question, you really are living in a dreamworld.

by Anonymous Citizen on Thu Jul 22, 2004 at 08:29:47 AM EST
[ Parent ]

Nice work. (none / 0)

As a math teacher myself, I'm fascinated by stuff like this...

I'm confused by one small portion of the analysis, though:

4 points: 26%, 279 EVs (222-297).

I'm not seeing how Kerry's 50% threshhold can be at 279 EV -- 9 more than needed to win, but the chances of winning are only 26%. These seem to contradict one another. Is it a typo or is there something going on that much more subtle than my very little brain can handle?

by Anonymous Citizen on Wed Jul 21, 2004 at 03:27:09 PM EST

Re: Nice work. (none / 0)

That does seem contradictory. If the 50th percentile is above 270, then by definition the probability of being above 270 is > 50%. I think the 279 is a typo since the other 3 numbers and the mention of a threshold effect in the text are all consistent with each other but not with 279.

by froggywomp on Wed Jul 21, 2004 at 05:45:21 PM EST
[ Parent ]

Re: Nice work. (none / 0)

Sam Wang here. I agree that the 269-EV thing is an apparent contradiction. I think there may be a roundoff error at the 50th percentile - I am tracking it down in the code. Because the result was so extreme, I let this slip by. Thanks for the sharp-eyed look at the results.

by Anonymous Citizen on Fri Jul 23, 2004 at 06:50:10 PM EST
[ Parent ]

Confidence (none / 0)

I'm not sure what "confidence" means, in mathematical terms. It sounds like the joke from Airplane! "There's a 50% chance they'll make it but only a twenty percent chance of that."

by Rob C on Wed Jul 21, 2004 at 04:03:01 PM EST

Re: Confidence (none / 0)

Confidence means exactly what you'd assume it to mean. Most modern polls are done with a 95% confidence level, meaning that if the election was held that day, the actual results would be within the margin of error 95% of the time.

by Anonymous Citizen on Wed Jul 21, 2004 at 05:22:35 PM EST
[ Parent ]

Re: Confidence (none / 0)

It is important to remember however that it is standard practice to give confidence intervals assuming only random error--NOT systematic error such as would come from underrepresentation of people with cellphones or imperfections in deciding who a "likely voter" is.

Random error is what prevents you from getting sixes exactly 1/6 of the time when you roll perfectly balanced dice. Systematic error would happen if the die is loaded.

by froggywomp on Wed Jul 21, 2004 at 05:49:45 PM EST
[ Parent ]

G-O-T-V (none / 0)

What counts when we cast our ballots is what we need to worry about.

by Anonymous Citizen on Wed Jul 21, 2004 at 04:11:40 PM EST

Still not quite so optimistic (none / 0)

I have yet to see a state-by-state poll that shows any state going from red to blue, except Florida and New Hampshire, both of which go back and forth. Wisconsin, Oregon, and New Mexico keep going back and forth as well.

That said, expect for oregon his ordering of polls seems about right. And it says Kerry will get anywhere betweenlikely scenario at 291-247. Which I think is much more realistic than the various projections that have shown Kerry getting like 322 electoral votes.

by Anonymous Citizen on Wed Jul 21, 2004 at 04:54:13 PM EST

can electoral college polls predict for congress? (none / 0)

Jerome:

Have you (or has anyone) done any analysis of electoral college poll data, with respect to coattail effects on downballot candidates, and made predictions for congressional races accordingly?

In other words, on a state district-by-district basis, is there any data which suggests a predictive relationship between a given state's electoral college count (predicted by a given state's poll data stream) and the ideological shape of its simultaneously elected congressional delegation?

(I hope that's a coherent question)

by bz on Wed Jul 21, 2004 at 07:25:35 PM EST

Re: Chances of a Kerry Victory: 98%? (none / 0)

I can't fault the analysis (I got similar results when I ran some much more basic calculations), but whatever the polls say, I find FL (98%) hard to believe. We have a 98% chance of winning in FL if an election were today?

Why not read Things I've Seen

by tis on Wed Jul 21, 2004 at 10:39:09 PM EST

polls (none / 0)

Polls might take on added significance this year, given the considerable anxiety about the possibility of electoral shenanigans. If the GOP is hoping to pull a Diebold scam or a Florida 2000 replay, they need the polls to be within the margin of plausibility going into the election. For this reason it's important that Kerry/Edwards have very strong numbers as we head down the stretch.

by Anonymous Citizen on Wed Jul 21, 2004 at 10:44:30 PM EST

Kerry Victory? (none / 0)

Nice to know there is some optomistic news for Kerry - I just got a poll from Zogby giving Bush 285 electoral votes, and only 253 for Kerry.

by Anonymous Citizen on Thu Jul 22, 2004 at 01:36:28 PM EST

You are not logged in.

In order to post a comment, you must be logged in. If you have a member account, please log in to comment.

If not, you can make an account right here. It's quick and free.