| |
BGonline.org Forums
The right way to measure which player played better in a match
Posted By: Jeremy Bagai In Response To: The right way to measure which player played better in a match (Bob Koca)
Date: Wednesday, 1 June 2011, at 8:28 p.m.
"If you use MWC method you will say that the error in the first match was much less important than the error in the second match but in fact they were equally important."
Delightful example, Bob. But I think you are philosophically off the mark.
The error in the first match swings your chances from 55% to 45%. The error in the second match swings it from 100% to 0%. These are not "equally important". They are equally important in hindsight -- but everything is equally important in hindsight. Probability has no meaning in hindsight. What can it mean when the outcome of an event is determined?
How about we play a series of 11 point matches. You get to start 1,0 for each match. But I get to win any match that gets to DMP. Note that your spot comes into play every match, whereas mine doesn't. But I'm still happy happy about the situation.
The "right" way to measure anything is to measure exactly that thing. So if you want to measure ability to win backgammon matches, then look at win/loss rates. Your elo rating is necessarily a better measure than anything that is derived, like MWC lost or PR.
But an elo rating, like backgammon, has a lot of noise. MWC has much less noise. But it is a measure that is dependent on the strength of the analyzing bot.
An error rate is also dependent on the strength of the analyst, but introduces new dependencies. It does not take into account choking at the end of matches. And it also assumes that the EMG transformation is meaningful and consistent.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.