Old Comments (starting July 2004)

Saturday, July 31: For now I will start primarily reporting results based on the last three polls. The distribution of outcomes is tighter, suggesting that using six polls introduces effects caused by variation over time.

Tuesday, August 3: Because of a remarkably strong swing in Tennessee and South Carolina towards Kerry in July, mostly after the selection of John Edwards for the Democratic ticket, Tennessee is now included in these calculations. [bounce chart]

Monday, August 9: The format at RealClearPolitics has been changed to show only one recent Zogby poll on the summary page. For now I will accept data selected by the new criteria, whatever they are.

Monday, August 9: Michael Lee of Van Nuys, CA asked me about the likelihood of an exact electoral tie (269-269 EV). Right now that event is very unlikely (0.7%). If Kerry's margin dropped by 2.5 points the election would be extremely close. In that case the probability of an even electoral split is still only 2.5%. (Whew.)

Wednesday, August 11: In Colorado, voters may have the chance to split their electoral votes. If this passes their nine votes would probably split 5-4 for Bush and Kerry, instead of 9-0 for Bush. Until polling data are available, this will not be included in my calculation on a regular basis. As of today it would change the probability from 98.6% to 99.4%.

Sunday, August 15: I am thinking ahead to the election, wondering how to test the predictive value of this analysis. An obvious comparison is the number of electoral votes. But what about more detailed predictions? For instance, polls can be used to predict the rank order of margins of victory (or loss). At present the order is (Kerry to Bush)
This prediction has the advantage of being independent of overall poll bias.

Wednesday, August 18: Colorado has been added because a recent poll reports a tie there. For now I will ignore the effect of the ballot proposal to split their electoral votes.

Friday, August 20: A change to the calculation. When the calculated SEM is less than 1% (true now in five states) I will force it to 1% to give a more conservative probability.

Friday, August 20: For margin of error junkies. Rachel Findley pointed out this comparison of polls to election outcomes, which finds that only 84% of election outcomes fall within the reported 95% confidence interval. This discrepancy allows a way to estimate polling errors that go beyond sampling error. If the additional error is normally distributed, an appropriate correction would be to increase the reported margin of error by a factor of 1.4. However, this correction does not apply to my calculation because instead of relying on reported MoE, I use inter-poll data to make an independent estimate of variance.

Saturday, August 21: Given where polls have fluctuated in the last few months, for Kerry to get 270 EV a key event will be winning FL, OH or MO. Even without any of those states, a second permutation with high relative likelihood is to win TN and WV.

Monday, August 23: You can use the bias calculation to estimate where things are headed. If you think turnout will boost your candidate by N points, add that. If you think that one candidate will gain X points at the expense of the other, add 2*X. For instance, if turnout will increase Kerry's vote by 2 points, but Bush will pick up 1.5% of voters from Kerry, then the bias is 2-1.5*2=-1%, or 1% to Bush.

Tuesday, August 24: Despite the reasonable showing by Kerry in the new Zogby polls his expected EV and >=270 EV probability have decreased. This is caused by several Florida polls showing a near-tie.

9:00PM: I have added North Carolina and Virginia to the calculation. Because of the large jitter in Zogby polls it's not time to get excited yet, but this is an interesting development. The median EV jumps from 297 EV (95% 269-333 EV) to 310 EV (95% 279-347 EV), not because of NC or VA, but because Kerry appears to have tiny margins in Nevada and Colorado. This could get computationally laborious; with 21 states the maximum unpruned number of permutations is over 2 million.

Thursday, August 26, 2004: The first effects of recent events - the EV count and probability have declined. This could be a rough ride!

Friday, August 27, 2004: I was experimenting with using polls from other sources, but it's a mess. RCP likes likely voter results, but other sites often report the registered voter result. The new numbers in the box above reflect this correction for OH, MO and WI. The only truly new poll is for PA.

Monday, August 30, 2004: Now that many states are in the midrange, it may be time to focus on voter turnout operations. How important this is depends on (a) how accurate state polls are and (b) where things are headed. For what it's worth, I think that one should add a bias of +2% towards Kerry because of Democratic turnout/get-out-the-vote/motivation and because undecided voters tend to break for the challenger. Where things are headed is a question I can't answer.

Tuesday, August 31, 2004: Note that the input data are largely from likely voters, which usually gives a result more favorable to Bush than counting all registered voters. See this explanation of how likely voters are identified.

Sunday, September 5, 2004: I won't update the calculation until after the weekend, when new state polls are out. In the meantime, polls that cover a period mostly on or after the date of Bush's acceptance speech are Bush +11, +4, +10, and +2 (average 6.8) and the four polls before the convention are tie, tie, Kerry +1, Bush +3 (average 0.5), implying a six-point bounce. The bias calculation can be used to infer what this means.

Wednesday, September 8, 2004: There are new polls available in nearly every battleground state (see code). Note that bounces from the selection of Edwards and from the Democratic convention took 2-3 polling cycles to reach their maximum.

Thursday, September 9, 2004: The median electoral vote count for Kerry has dropped below 270 for the first time since June. The key event here was the probability of a Pennsylvania win dropping below 50%. Is the Bush post-convention bounce complete? It's hard to say, but considering what a radical effect comes from a 1-point swing in either direction, whenever the probability is in the 5%-95% range this election is about turnout. This is the situation in which your efforts mean the most!

Monday, September 13, 2004: We now have a nearly full measure of the post-convention bounce - and a likely rebound. Kerry's very slim lead in the Electoral College has stabilized for now. Note the mismatch between the electoral calculation and Bush's lead in national polls. Evidently, it is indeed possible for battleground states to diverge from overall national opinion. Finally, in terms of electoral votes, the Democratic convention bounce was three times the size of the Republican convention bounce - and longer-lasting.

Thursday, September 16, 2004: Today's calculation is unambiguously bad for Kerry. The median is lower than it was before the release of Fahrenheit 9/11. However, overall bias is an unknown factor, and at this stage a 1-2% difference in turnout or of voters changing their minds would be a key determining factor for either candidate.

New Jersey has been added to the list. I am surprised at this need.

Wednesday, September 22, 2004: For the second day in a row, the median EV counts have moved away from Bush and toward Kerry. The Swing Index is near zero, indicating a tie. Today's new polls come from a variety of sources and a large crop from ARG. A nice tabulation can be found at www.race2004.net, which is more convenient to use than electoral-vote.com and more complete than RealClearPolitics.

Here is an intuitive measure of how close the election is: the Swing Index. This is the across-the-board percentage shift in opinion that would be needed to make the win probability 50%. This answers the following question: Given current polls, what fraction of voters would have to change their minds in order to make the electoral college a toss-up (>=270EV probability equal to 50%)? This takes advantage of the bias analysis. Bias occurs if (a) polling organizations give skewed results or if (b) one side turns its voters out better or worse than predicted by the polling criteria. The Swing Index tracks most national polls except for Gallup. On that note, recall that on average, Election Eve 2000 national polls predicted a margin of Bush +2.0%, while the actual outcome was Gore +0.5%, a swing of 2.5%.

So...why aren't you out registering voters? Here are forms.

Friday, September 24, 2004: As has been the case for much of this campaign cycle, three critical big states are PA, FL and OH. The candidate who wins two of these states is likely to win the election. A possible exception occurs for Kerry if he wins one big state plus a handful of smaller states such as WI, CO, and AR. Two of these states currently have polls that favor Bush much more than other polls taken around the same time (WI, Badger; CO, Ciruli). Are these outliers?

Sunday, September 26, 2004: The history over time is updated. Note the sharp change from Sept. 10-15, which might be caused by the Sept. 11 anniversary. Relevant to possible poll biases, I have played around with removing various data to see how the history is affected. Relative to the average, Survey USA and Mason-Dixon favor Bush, while Zogby/WSJ favors Kerry. The Zogby one is easiest to see because a full update occurred on Sept. 20, generating a spike. Zogby is the only pollster in 2000 (out of 15) who said Gore would win the popular vote.

Monday, September 27, 2004: This week's focus is registration of overseas voters. Seven million American citizens live abroad, and most do not vote. This year they could be decisive. The first deadline for applications to be received by county registrars is October 2, and thirty states have deadlines October 5 or earlier. So people must hurry!

Any US citizen living abroad can register and vote, even if it's been decades since they lived in the US, even if they never voted, even if they were born to US citizens and never lived in the US. For them, here is a registration and application form. Other useful web sites are www.overseasvote2004.com and the Pentagon's site. Finally, expatriates who do not receive an absentee ballot can use a Federal Write-In Ballot. Otherwise they must depend on the state and the mail services to get them the ballot in time.

New voter story: A positive shout out to Greg in Tanzania, who was told his absentee ballot will not likely get to him in time. He will do Federal Write-in through the embassy in Dar. He is getting other Americans there to do the same. Go Greg! Wednesday: Good thing, too. Evidently our brilliant government is hard at work helping military overseas vote, but nobody else. [story]

Is the race at an equilibrium? Kerry is effectively 2 points back, as measured by my Swing Index. This is narrower than nationwide popular polls, which show an average 4-point gap. As has been the case for much of this campaign cycle, three critical big states are PA, FL and OH. The candidate who wins two of these states is likely to win the election. In polls completed Sept. 19-25, the average margin is Bush by 3.2% in OH (5 polls), Bush by 1.2% in FL (6 polls), and Kerry by 1.6% in PA (5 polls). Assuming these data, the likeliest deciding state is...Florida. Just as in 2000.

Thursday, September 30, 2004: I am thinking about more sophisticated (and automated) methods of outlier identification. This is likely to have some kind of effect. For instance, the results with Gallup excluded are median Kerry 242 Bush 296, Kerry 95% confidence band 214-274 EV, win probability 10%, Swing Index 1.5%. In addition to the difference in median result, note that leaving out those polls leads to a narrower 95% confidence band and a large change in Swing Index. These are consistent with the idea that this organization is an outlier. However, a rational approach requires an algorithm for looking at individual data values. Developing...

Friday, October 1, 2004: Thanks for your suggestions about how to deal with outlier data values. To reduce disproportionate effects of outliers (but still give them weight), I am considering switching to using the median rather than the mean. I am now experimenting with this approach, as well as with increasing the number of polls used per state. If I implement changes, I will make sure to provide enough information so that you can fairly compare pre-debate and post-debate polls. Thanks to Alan Cobo-Lewis and E.J. Chichilnisky for the suggestion.

Monday, October 4, 2004: In a national poll, Newsweek shows Kerry ahead by 3% among registered voters, a swing of 8 points from their previous poll - and three times the current Meta-Margin. A USAToday/CNN/Gallup poll also shows an 8-point swing compared with their previous poll. [National polls] Now here is an NJ poll that shows a 6.7-point jump over the last three polls taken before the debate. Based on this I venture a guess that Kerry will get a significant bounce this week.

Wednesday, October 6, 2004: Possible clues to whether the bounce has occurred in key states: Florida, Iowa, and Ohio polls show Kerry ahead for the first time in weeks, though by tiny margins. Other polls are generally in the direction of Kerry. Nationally, Kerry's bounce is 5 +/- 2 points (mean +/- SEM), comparing the six polls immediately after the debate with the six polls before. [National polls] This is much larger than the current Meta-Margin, suggesting the potential for a large change in EV standings.

I am now setting the floor for standard error of the mean at 2 points instead of 1, in order to more realistically capture uncertainty in cases where polls happen to be near one another. This does not change things much, except for the probability calculations. Which you are supposed to take with a big grain of salt, remember?

Usually over the last few months, these indices have been mismatched: at a certain point in opinion swing, Bush could very well win the popular vote but lose the Electoral College. In other words, this year the Electoral College mechanism seems to favor Kerry.

Relevant to the Cheney-Edwards debate: campaign satire, possibly amusing to both sides. Cheney is less gaffe-prone than Bush, but he is still a skillful deceiver.

This site is hard work (not that hard, actually, but it's an excuse to include this musical interlude).

Electoral-vote.com is back to using only one poll per state. Bad idea!

Monday, October 11, 2004: Lucia B., a student in Operations Research and Finance, has started to help with data updates and analysis. Welcome and thanks to Lucia!

Over at Race 2004, Stephen points out that Colorado Democrats tend to support Amendment 36, while Republicans are opposed. He thinks this might become a tricky choice for partisans if sentiment shifts towards Kerry. I disagree. Optimally, Democrats and Republicans should stick with their preferences even if Colorado swings into the Kerry column. Here is why.

States tend to move in the same direction in opinion, and Colorado (Bush +5.7% in recent polls) tends to be a lower-likelihood state for Kerry than Ohio (Kerry +0.02%) and Florida (Bush +2.7%). Therefore, if the national election is close enough for Colorado to make a difference, it is likely to go for Bush. In this case, Amendment 36 will provide four EV for Kerry, votes that might be critical. Conversely, if opinion swings enough for Colorado to go for Kerry, then Ohio or Florida will be the deciding state, and losing four electoral votes will not matter. Therefore, passage of Amendment 36 is most likely to either help the Democrats or have no effect.

Tuesday, October 12, 2004: Although the median is a tie, this is just the midpoint of likely outcomes. The actual probability of an exact electoral tie today is only 4%. The Popular Meta-Margin, Bush +0.1%, quantifies how close things are. As ever, it's all about turnout. A strong predictor of victory is winning two or three of the following states: FL, OH, and PA. Kerry is ahead in PA (6 polls), Bush is ahead in FL (6 polls), and OH is tied (4 polls). [sources]

Anecdote: this site originated in July after a conversation with my colleague Steve Gubser. I contended that the election's probable outcome was 3-to-1 favoring Kerry, since Pennsylvania seemed likely to go for the Democrats, at which point either Florida or Ohio was needed to get over 270 EV. Since each was 50-50, the probability of getting at least one was therefore 75%. A few days later I did the calculation properly and created the site. Now, three months later, it's still Florida or Ohio (though at this moment Florida looks less likely). Putting aside all the fancy math, under near-tie conditions the basic principle of Florida-or-Ohio has been quite consistent.

Ohio is critical for victory - but slightly more critical for Bush. This principle can be expressed quantitatively (thanks to Paul Y. at Goldman Sachs for asking this one). Given today's numbers, assuming a Kerry win in Ohio, the probability that Kerry wins the general election is 89% (8:1 odds). Conversely, if Bush wins Ohio, his win probability for the general election is 83% (5:1 odds). The exceptional outcomes are caused by all the alternate scenarios that Electoral College junkies like to speculate about.

Relevant to the Cheney-Edwards debate: campaign satire, possibly amusing to both sides.

Wednesday, October 13, 2004: Tie conditions continue today. Because the outcome is driven mainly by Pennsylvania, Ohio, and Florida, today I summarize their data below.

Pennsylvania (21 EV): Kerry leads in 3 of 4 polls (average margin 1.5%), and in 7 of 8 polls completed since October 1. Win probability: Kerry 77%.

Ohio (20 EV): Bush and Kerry each lead in 2 of 4 polls (average margin Bush by 1.2%). Bush and Kerry each lead in 4 of 8 polls completed since October 1. Win probability: Bush 71%.

Florida (27 EV): Bush leads in both of 2 polls (average margin 5.0%), and in 7 of 9 polls completed since October 1. Win probability: Bush 88%.

Friday, October 15, 2004, undecided voters. From last week's great interview with Charlie Cook of the National Journal comes a key quote: "I cannot remember ever seeing a race where a well-known, well-defined incumbent won a half or more of the undecided vote. Generally it is at least two-thirds to three-quarters going to the challenger.... That's why it is a mistake for people to focus on the spread between the two candidates. The far more relevant figure is the actual vote percentage of the incumbent in a poll (or better, average of polls).... if President Bush is at 46-48 percent of the vote going into election day, he probably loses..."

Therefore the meta-analysis you read on this Web site needs to be adjusted using the bias calculation. Based on current national polls the bias is at least Kerry +1% from undecideds. To read more about the "incumbent rule" see essays by Guy Molyneux, Mystery Pollster, and Mark Shields.

Monday, October 18, 2004: I am considering several changes. (1) Assigning undecided voters using the range of figures quoted by Charlie Cook and Guy Molyneux. (Noon update: Here's an excellent summary.) Of course, I will continue to report the raw poll analysis, as I have been doing since July. (2) For clarity, I may stop tabulating rankings for safe states: for Kerry, WA, NJ, MI and OR; and for Bush, AZ, NC, VA and WV. For now, I will have an open comment period for the next few days.