(No, this isn’t what I was hinting at earlier.)
Assuming you live in the United States, you’re probably used to races being called virtually the instant the polls close. Networks, not wanting to deal with – heaven forbid! – uncertainty (or losing the scoop to a rival network), use exit polls to “cheat” and declare the winner of a race certain without having any actual results to go by. No doubt you may have been confused in 2000 when Gore was called as winning Florida when Bush was consistently leading.
I believe I have a better system to call results based on one thing and one thing only: the results themselves. But it appears complicated at first glance because, as it’s evolved over the years, it involves four different methods of calling a race – four different levels of certainty.
Projection was developed originally as a way for me to avoid having to wait for validation of a foregone conclusion. Used when one candidate leads another by a statistically significant margin consistently, it’s most akin to the networks’ approach but “projection” isn’t really the right word. It’s really more of an expectation. I think as of late I’ve been drifting towards using this as a reflection of what the networks call or aping the AP’s calls.
Auto projection and the other automated methods assume all precincts have an equal number of voters, which isn’t necessarily true but it’s good enough. If Candidate A leads Candidate B by A% to B% with P% of the precincts reporting, then with all percentages expressed as fractions of 1, if A%>B%+(1-P%), the race is autoprojected to A. In other words, A must lead B by at least the percentage of precincts not reporting. This one’s in here for its simplicity and the ability to provide some satisfaction before the really significant one.
Confirmation is a result of the implications of the above assumption, which indicates that A has really won A%*P% of all the votes in play. (Similarly, B has won B%*P%.) Thus, this test involves multiplying A% and B% by P%, and repeating the auto projection test: A%*P%>B%*P%+(1-P%). If A passes this test, and the assumption above is true, it is mathematically impossible for B to pass A. B has been “eliminated” and, if B was second, A is no-doubt-about-it first. A network using this system might still say A “has been auto projected” to win, but once A crosses that confirmation threshhold, you don’t say A “has been confirmed” – you say A has “won”, no doubt about it.
Majority confirmation is one I’m considering dropping. In a two-man race it’s the same as regular confirmation. In large or tightly contested races it might not occur, as I’ve found out in the early presidential primaries. In all races it’s meaningless because the confirmation threshhold has already sealed A’s victory, unless having a majority is meaningful in some way. It basically puts A up for confirmation but against the 50% threshhold instead of B’s reporting-adjusted maximum: A%*P%>.5.
I have tried to keep track of all of this in the past on the general election day, but with 435 House races, 33 Senate races, and 50 Presidential races, I have often lagged behind, which gets worse because I get hooked to what the networks are saying. I’d like to be able to get a constantly updated feed of results that I can plug in easily. The more effective solution at the moment is to enlist any of you who may wish to volunteer; e-mail mwmailsea at yahoo dot com or leave a note in the comments if you’re willing and able to take the challenge November 4. What I’d really like is for some way for a web page to automatically pull up results from a central file and I would only have to make the human projection at most, but even if it were possible I don’t have the requisite knowledge in stuff like JavaScript. Still, I do intend to hold a test of my own abilities to handle the system relatively free from distraction on Super Tuesday, February 5.