BGonline.org Forums

OLM Statistics & Match File

Posted By: Jason Lee
Date: Friday, 27 June 2014, at 9:38 p.m.

First, the match file. If you're having trouble downloading, drop me an email and I'll email it to you. All plays that were voted on have been rolled out. Many (but not all) comments from various OLM threads are included as annotations in the match file.

Now comes the data that I've been meticulously gathering for four months. I've created a metric that measures how well you've voted relative to your team, as judged by XG rollouts.

On each play, I measure the average error (measured in units of 0.001 EMG, henceforth milligames) of the votes cast. Call this number E.

For example, let's say that five votes are cast for play A, which wins the XG rollout, two votes are cast for play B, which is judged by XG to be 0.015 behind, and three votes are cast for play C, which is judged by XG to be 0.060 behind. Thus, E = 1000*(5*0 + 2*0.015 + 3*0.060)/10 = 21.

Now, each player is assigned points. Those who didn't vote get 0. Those who did vote get assigned E-X, where X is the error of their play in milligames. So those who voted for play A, which won the XG rollout, gets 21-0 = 21 points. Those who voted for play B get 21-15 = 6 points. Those who voted for play C get 21-60 = -39 points.

Note that the assignment of points is zero sum. Five people get 21 points, two get 6 points, and three get -39 points, and note that 5*21 + 2*6 + 3*(-39) = 0.

That describes your score for an individual vote. Your score for the entire match is your sum of your individual scores. Since each individual vote is zero sum, so are the totals.

Think of a positive total score as meaning on average, you voted "better" than your teammates did. A person who never votes, or who only votes on unanimous plays, will get a total score of zero. Because the scores are created using the data relative to your team, it wouldn't make a lot of sense to compare scores between players of opposite teams.

Long story short, here are the scores:

The Dilly Builders

Possible
Total
Score
neilkaz 51/51 377.4
willywonka 51/51 209.4
Cheryl Andersen 26/28 181.8
Mr Majestyk 21/51 143.1
mandm1951 49/51 69.4
MishyPoo 50/51 -67.6
Konstantin 15/51 -123.8
David Rennie 50/51 -154.6
AbbycazNZ 39/41 -180.8
Keene 47/51 -226.4
ah_clem 35/51 -291.8

The Prime Factors
Possible
Total
Score
Casper Van der Tak 51/51 608.7
Igor 50/51 478.7
Stick 33/51 414.0
lenny 48/51 329.6
Daniel Murphy 39/51 233.7
garyo 35/51 22.6
kruidenbuiltje 24/27 -34.8
Bob Koca 37/51 -87.5
Taper_Mike 51/51 -194.3
Fatboy 44/51 -321.6
Steve Mellen 51/51 -423.3
Seth 39/39 -464.7

I would caution you to not read too much into these scores. First of all, it's data culled from ONE match. We all know how unreliable the data of one match is for predicting how well you played. Secondly, there's a new set of "noise," which is that you're now being judged against how well your teammates played, rather than against an absolute standard. Sadly, in order for you to do well, your teammates have to err. It's like measuring your batting average based upon the difference between your average and the team's average. Finally, there's a new source of luck, which is where you vote and where you don't vote. If you were busy on a day and miss voting on a play where you'd get the right play, you lost a chance to earn positive points.

The other bit of data in there is participation numbers -- how many times you voted out of how many votes you were on the team. You'll note that four players joined late, so their possible participation numbers were lower. Notice that this doesn't change how their score is computed. A person who does not cast a vote gets a zero score for that vote, so those who joined late just started with a whole bunch of non-votes, getting zero for all of those.

JLee

Post Response

Subject:
Message: