Electoral Vote Predictor 2004:   Kerry 211   Bush 327

Data in Excel format Battleground states Info about the states Previous report Next report News   Strong Kerry (110) Weak Kerry (63) Barely Kerry (38) Exactly tied (0) Barely Bush (90) Weak Bush (67) Strong Bush (170) Needed to win: 270

Sept. 20

New polls: CA CO FL MI NH NC TN

Welcome	FAQ	Donations	Electoral college
Electoral college graph	Projected final map	Cartograms	Movie
State poll graphs	State poll tables	EVs as bar charts	More data
Political humor	Political websites	Articles and books	Kerry and Bush records
Americans abroad	Foreign nationals	About RSS feeds	Enhancements

News from the Votemaster

Time is running out. While the election is 6 weeks away, the deadline for registering as a voter is almost upon us. In most states it is in less than two weeks. This rapidly approaching deadline is especially important for the 5 million overseas voters who have to send in the registration form by snail mail, although in some states there is now a way to beat the deadline by fax, as described on the Americans abroad page. If you, your family, or your friends are not registered to vote yet, now is the time.

Some voters have a choice where to register. When George Bush picked Dick Cheney as his running mate in 2000, there was the minor matter that Cheney' presence on the ticket might cause Bush to lose Texas' 32 votes in the electoral college and hence the election. Seems the constitution says electors can't vote for both a president and a vice president from their own state. In the spring of 2000, both Bush and Cheney were living in Texas and registered to vote in Texas. Cheney finessed the problem by changing his voter registration to Wyoming, where he had a summer home. This act could be considered creative voter registration because by no stretch of the imagination did Dick Cheney suddenly become a resident of Wyoming. He continued to live in Texas where he was running Halliburton, the oil services company that continued to pay him a salary even while he was vice president of the United States. The latter job doesn't pay very well--only $202,900 per year--so a bit on the side is always helpful. That Halliburton received a $7 billion no-bid contract to help rebuild Iraq's oil industry is mere coincidence.

But even if you are not Dick Cheney, Cheney's first law of voter registration ("Register where it helps most") may apply to you. In particular, college students studying in a swing state may be able to choose choose between their parents' address and their college address. If a student has no plans to ever live with his or her parents again, a good case can be made that the student no longer lives in the parent's state and can thus register in the college's state. Students living in a dorm and paying out-of-state tuition have a weak case, but can argue that they are planning to run for vice president some day. Students living off campus or paying in-state tuition have a strong case. If you are from a solid state and are studying in a swing state (or vice versa), or have a child, grandchild, friend, or neighbor in this category, please see Swing State Voter Project for details. Also, if you are a snowbird and live for 6 months in the North and 6 months in Florida where you own or rent a house or condo, register in Florida. But hurry. Being registered in two states is illegal, so if you register in a new state, be sure to cancel the old one. If George Orwell were alive now, he would say "All votes are equal but some votes are more equal."

There are seven new polls today. In Florida Bush's lead has dropped to 1%. But the strangest result is in New Hampshire, where the Rasmussen 7-day tracking poll puts Kerry ahead by 6% again. The Mason-Dixon poll of Sept. 13-15 giving Bush a 9% lead was just way off. A 15% change in two days in a small homogeneous state in the absence of any major news is unthinkable.

Many new polls are due out today. I might put out a supplementary map late in the afternoon (Eastern time) if I have the time, but I will have everything tomorrow morning. Please DON'T send me mail telling me about the new polls; I know where to find them. Thanks.

If you missed Saturday's update you might want to check it out as there was a long discussion of polling accuracy there. There was a bit more on Sunday. A summary of that material is now on the Articles and books page.

On the subject of polling accuracy, many people have asked me how well the final election 2000 polls did. After a great deal of effort, I was able to find 15 polls taken in the final 3 weeks before the Nov. 7, 2000 election. They are listed below (and also on the More data page for future reference. The bottom line is that Zogby and CBS were the only ones that predicted that Gore would win the popular vote, and CBS can't gloat too much because although its Nov. 4-6 poll was right on the money, its Nov. 1-4 poll predicted a 5% margin for Bush. In reality, Gore won by half a million votes or 0.5%. Harris and Opinion Dynamics predicted ties and the other 11 predicted Bush would win the popular vote with margins ranging from 2% to 6%. I have not found any polls predicting the electoral vote the way this site does. For that reason, I may just keep this site alive until Nov. 2008 so there will be better historical data available next time.

The MoE for each poll depends on the sample size (N). As a quick rule of thumb, the MoE at the 95% confidence level is about 100%/square-root(N), so for N = 500, 1000, 1500, and 2000, the MoEs are 4.4%, 3.2%, 2.6%, and 2.2%, respectively. These numbers are not exact, but are good enough for government work. But remember, the MoE refers only to sampling error. If there were errors in the methodology, the true mean may be way outside the MoE range. All the pollsters except Zogby, about which I am not sure, used their grandma's special secret receipe for predicting likely voters and except for CBS' final poll, all got it wrong. As I mentioned on Saturday, from now on I will use registered voters instead of likely voters when the data are available.

Pollsters judge their results not by whether they picked the the right horse, but by whether both candidates fell within the ranges predicted by the MoE. Using this criterion, let's see how well they did. The poll taken closest to the election was Gallup's poll of Nov. 5-6, the Sunday and Monday before the election. The prediction was Gore 46%, Bush 48%. The sample was exceptionally large (N = 2350) giving a MoE of 2.1%. That means Gallup was predicting Gore's total would fall in the range 43.9% to 48.1% and Bush's total would fall in the 45.9% to 50.1% range. The actual result was Gore 48.4% and Bush 47.9%. Thus in purely technical terms, a poll ending the day before the election wasn't able to get both numbers right. It missed a lot of Gore voters.

Note that the polls above were taken during the final three weeks of the election. This year the pollsters have an even tougher job for several reasons. First, More people are uncallable because their only phone is a cell phone, cable phone, or Internet phone. Second more callable people screen out pollsters with caller ID or answering machines. Third, overseas voters, who are never polled, are registering in droves. And fourth, both parties are putting so much effort into signing up new voters that the statistical models of the electorate used for normalizing the polls are increasingly invalid because they are based in part on 2000 and 2002 exit polls and the electorate may be quite different this time.

This may seem an odd story coming from an election site, sort of like a soft drink company telling you their products rot your teeth. Polls aren't useless, but you have to understand their limitations to interpret them correctly.

More people complained about the new colors than approved of them, including people who are color blind, so I have gone back to the original colors.

-- The votemaster

WWW www.electoral-vote.com