My projections are based on past turnout and margin of victory trends for each of Ohio's 88 counties. The Cleveland Plan dealer has an excellent overview at http://blog.cleveland.com/pdgraphics/2008/10/03WGWATCH01.pdf which illustrates quite well how the individual counties in Ohio trend. The accompanying article at http://blog.cleveland.com/openers/2008/11/tracking_ohio_vote_on_election.html is also a very good rundown of the Ohio counties and their trends. I don't buy into their bellwether counties however.
I use similar data from the Ohio Secretary of State website to build customized targets for each county. Then, if McCain or Obama outperform their targets, they start to gain an advantage.
What's the difference between Full and Partial county projections?
The Full County projection doesn't give McCain or Obama an advantage for a county until 100% of the precincts for that county are reported. This provides a much better basis for determining whether or not McCain or Obama is outperforming their target for a county. Full County projections are slower to develop but should be more stable. The Full County projections predicts the net number of votes Obama will win or lost Ohio by.
The Partial County projection doesn't wait until 100% of the precincts for a county are reported to give McCain or Obama an advantage. It uses a combination of past performance and the actual results of reporting precincts to project a vote total for each candidate. This projection assumes all precincts in a county behave the same and this is not correct. For this reason, it will develop quickly but can have a lot more variance. Obviously as more precincts report, the stability increases. It starts out assuming Ohio is a tie until either McCain or Obama begin to outperform their targets and then adjusts their totals accordingly. Absentee ballots may also cause these results to swing.
How the heck can McCain be leading in the raw vote totals and Obama be projected as having a winning margin, or vice-versa?
Go back and look at the Cleveland Plain Dealer link in question #1. There are counties where McCain must win big to have a chance at winning Ohio. It's possible for him to win these counties, but still not win them by enough, and then when the Obama county results come in, Obama will probably take the lead. The same is possible for Obama, although in 2004, the rural county results (which favor McCain) came in quicker than the urban counties (which favor Obama).
What assumptions did you build into your projection model?
I developed my projection model so that it at the start, it projects Ohio to end up in a tie. Then, as results come in and McCain or Obama outperform the model, it projects them to win Ohio. I started the projection using the individual turnout and margin victory ratios for each of Ohio's 88 counties.
Turnout in Democratic counties has traditionally lagged Republican counties but I adjusted it up in my model to the same level as seen in Republican counties. My model assumes a turnout ratio of 72.3% in all counties, which is higher than 2004's turnout of 70.5% but lower than Ohio Secretary of States estimate of "up to 80%." If actual turnout is lower in Democratic counties compared to Republican counties, my model will over forecast Democratic strength until more results come in. This only amounted to 25,000 votes so it won't make much of difference and will self correct as actuals come in anyway.
After I adjusted for increased turnout in Democratic counties, I then reduced Bush's 2004 margin of victory (or increased Kerry's) by 0.85% in each county. At that point, Ohio became a tie. What this means if McCain has to be within 0.85% of Bush's margin of victory in each county to win Ohio. If Obama eats into it more than that, he will win.
How will early and absentee voting impact the projections?
This is the big question mark and why the partial county projections could have large swings. The Columbus Dispatch has early and absentee voting going to Obama 56% - 42%. Estimates are that 20% - 30% of the votes were cast early, so these votes will have an impact on the results.
The county election boards will not be counting the early and absentee ballots in a consistent manner. Some of them will count them first, and then begin reporting other precinct results. Other counties will report the precincts first, and then include the early and absentee ballots at the end. This could cause Obama to take large leads in early results, or it could give McCain a large early lead that he will have to defend as the absentee votes are counted. The problem is, it will be tough to tell whether the absentee votes will be counted first or last for each of the counties. This is why the full county projections may be more stable and telling.
Can't you just project at a precinct level instead.
I could, but there are over 11,000 precincts in Ohio and it would be a nightmare. Besides the same problem with absentee ballots still exist. Also, I can't remember if the Ohio Secretary of State publishes results at a precinct level in real-time.
What impact will provisional ballots have?
Provisional ballots have tracked traditional ballots to within 0.1% of a percent. They should not come into play and my methodology does not do anything special for them.
Why do you have the Full County projections based on Obama's margin of victory/defeat instead of McCain's?
The graph came up with a blue bar and I couldn't figure out how to change it's color. Blame Google Docs.
|