Quote:
Originally Posted by Nuttyman54
Good stuff, Eric!
I'm curious about some of the details of your methodology that I didn't see in your previous posts. Are you taking into account dead points from 3rd plays and un-winnable awards? For example, as far as I can tell, Philomath won't be giving out a Rookie All-Star award, so those 8 points are guaranteed gone from the district.
|
The methodology is based only on how many total points teams have earned at different events. It takes no other data whatsoever into account.
Here's the basic idea:
-If you've only played one event it estimates an upper bound for points earned at your second as the best that any team has done at an event when one of them was equal to or worse than your first event.
-Similarly, it guesses a lower bound for points based on the worst a team has done that has had an event that's at least as good as your first
After it has ranges for each team, it uses those to create ranges that the point cutoff could be. Then for each team, it compares their range to where the point cutoff could be.
So it doesn't explicitly take into account how many points there are left. However having lots of plays left will change the estimate of where the cuttoff will be by making the bounds wider. So for example, a team might go from listed as "~in" to "in" without a change in either the number of points they've earned or the best guess of where the cutoff will be just because more teams have played events.
So for this year there are two things that are really a big deal as far as trying to get an accurate result:
-There are so many teams that haven't played any event yet. At the moment, teams that haven't played at all are assumed to be exactly average, which obviously wrong. In past years there's only been maybe 1 team that didn't play till the last two weeks, so this didn't really matter.
-
The official ranking page's event results are known to be wrong. They haven't fixed they issue about how they account for ties properly. They just give a bonus to the team's overall total rather than correcting the event score. See here:
http://www.chiefdelphi.com/forums/sh...hreadid=146151. This throws off the estimates because the system is based on comparing event results.