STATE BY STATE PROJECTIONS,
ELECTION NIGHT AND BEFORE

On election night, once states have been called by a network for a particular candidate – or once I think that the state should have been called (I expect they’ll be skittish on Florida until 90% of the precincts are reporting and they have Palm Beach County), I will upload a graphic showing who is the projected winner.

Until then, for the next several weeks, I will be posting state by state Federal Review projections and analysis, starting with the easy states and eventually getting around to the battleground states. Below is the summary of the data I will be presenting from each state based on the data I have collected since March 2004. As of today, I have collected national polling data surveying a total over 173,000 voters from more than 120 polls.

I have collected data from over 670 polls from the various states.

You may not agree with my projections, which are based heavily on what polls are saying, but sometimes on other factors. Generally, I will not take into account voter registration changes or perceived motivation of one party or another to vote. I assume that much of this will show up in the polls, ideally through the pollsters' various likely voter screens. Also, I remember in 2000 thinking that Republicans were simply more motivated than Democrats because they were motivated by animosity for Bill Clinton. They called themselves Broker Glass Republicans because they would crawl over broken glass to vote. A similar motivation exists for some Democrats in 2004. I think that is showing up in the polls and there’s evidence it may not be enough, by itself, to win the election for Kerry. We thought they were extra motivated in 2002, too.

We’ll see.

All of the numbers that will be listed will be the numbers available on the day I do the projection, obviously. Thus, variations over the next three weeks probably won’t matter when I'm projecting Alaska or Rhode Island. I will get around to the battleground in the last week before the election, when I’ll have data closer to election day – but still not the last weekend numbers. However, I’ll try to update those close state projections on November 1 and give you my overall election projection that night (or maybe on Nov 2 before the polls close).

Here’s the key.

Projection: Of all the estimates listed, this is the one in which I inject my own best guess based on an analysis of polling within the state, national polling, the progress of the race, the campaigns’ commitments to the state and how the state has performed historically. If there’s a bias, you’ll find it here. Feel free to disagree.

2000 Winner: The winner and actual margin between the candidates in 2000.

Composite Margin: My weighted average analysis of state polling. Each week’s polls for a state are average, and each week is weighted 1.5 times the previous week. Thus, recent polls are weighted more heavily, but one week’s result in favor of one candidate will not necessarily move the state from one candidate to another. Thus, subsequent weeks must confirm that the race has changed. This prevents anomalous results showing non-meaningful movement from changing the predicted winner and reduces the impact of partisan polls.

Current Category:The Composite Margin in English.

Probability of Bush Win: Based on a normal distribution around the mean. The Composite Margin is used as the mean, thus a recent anomalous result will not move the mean arbitrarily. The applied standard error in the calculation is based on all state polling since April, thus, it takes into account the volatility of the race over time and not just polling error. If a state is not listed as 100%, it should not be considered a complete lock for the candidate. UPDATEI'm going to report without rounding. Previously, I reported probabilities over 80% as 100%. Now, I know that is not statistically the right thing to do, but I think it makes a hell of a lot of sense when running an election simulation. For instance, because of the few polls in Illinois showing the race close for a brief time, the stat work I have done shows that Bush has a 7% probability of winning that state. Bah. Never going to happen, and any election simulation that has Illinois going to Bush and not also Michigan and Minnesota is per se unreasonable. Thus, I bump up those numbers. But things are more interesting around the low 80s, so I'll start reporting those probabilities in my state-by-state numbers.

Best Bush: The 95% confidence band of the 2-party vote share based on the Composite Result and the standard error applied to construct the probability.

Best Kerry: The 95% confidence band of the 2-party vote share based on the Composite Result and the standard error applied to construct the probability.

Trend: Based on all elections since 1960, I have calculated how states are trending in comparison to the national popular vote. If a state is designated “Democrat +2”, then the trend is that the 2-party vote share in this state will be 2 points more favorable to the Democrats than the national result. Thus, if Bush wins the national vote by 4 points, then Bush would win the state by 2 points. To calculate this number, I compared each election result since 1960 to the national popular vote result and then used a least squares regression to project 2004’s result. I constructed the regression for each year until the present from 1960 until 1996 and then averaged all projected numbers. This method actually predicted the 2000 result quite well, missing only West Virginia, Louisiana and Arkansas, each of which I would blame on President Clinton’s strength in 1992 and 1996, when these states were otherwise trending Republican.

Safe: If yes, then I see no chance of my projection being wrong. You may disagree.

Can other candidate win? If I see the state as close to a toss-up, I’ll answer “Yes” here. If I believe that a candidate cannot win the state unless there’s a significant swing in the national popular vote in favor of the candidate, then I’ll say that. If I say “Yes”, the other candidate can win, then take my Projected Winner as a projection I’m not willing to put money on. But please, no wagering.

National Popular Vote for Candidate to Win: Based on the correlation of state polling during this election cycle with the weekly national polling results, I estimate the national popular vote spread necessary for the candidate to win this state. For battleground states, I will report the spread for both candidates. For instance, if I report “Kerry +5.3” then that means Kerry would have to win the national popular vote by approximately 5.3 points over Bush in order to have a chance to carry the state. The closer he gets to that kind of national lead, the better chance that the particular state will move to his column. Obviously, state’s can be a little independent and not exactly correlated to the national vote, so just take this number with that caveat.

Best Weekly Polling Result for Bush: Each week, I average all poll results that come out for a state. Here I’ll show Bush’s best result. You may be aware of a single poll where Bush did better than I list here. If that’s the case, then I either missed that poll (though I don’t think I missed many), or there was another poll the same week that, when averaged, pulled the number down. In other words, the polling result you may point to was not confirmed by other pollsters in the same time period.

Best Weekly Polling Result for Kerry: See my answer immediately above.

For regular updates to the Composite State by State Margins, click here.

For regular updates to the Election Simulation, click here.

Note that the map on this page may not be updated as regularly.