From: "Dean Oliver" <deano@...
Sent: Tuesday, March 30, 2004 2:32 PM
Subject: [APBR_analysis] Re: Clutchness and a game state matrix
>--- In APBR_analysis@yahoogroups.com, igor eduardo küpfer
>> Okay, using your fomat I've updated my matrix to include 186 games now.
>> Here's a section of the matrix, focusing on the last 5 minutes:
>Format isn't quite the same, since we have time/score along different
Er, I uploaded the wrong one. Never mind, you get the picture.
>> You can see how much noise there is, even with that many games
>> the sample. I get the feeling that some smoothing is in order, but I
>> idea how to go about doing that.
>That's where the theory is helpful. No time now, but we can talk
>Does your matrix really have full population? Specifically, do you
>have games where the home team is up 7 with 4:40 to go and they've
>lost them all or do you just have no games like this? (I'm guessing
>the blanks are for when you have no games.)
Yeah, I have a spreadsheet with the full population. I have one game in the
situation above, which the home team lost (LAC @ MIL). You're right, the
blanks are no games. This is beginning to bug me -- I have a pretty large
sample of games, about 5 weeks worth, and there are still many situations
not covered. I think that even with a full season's worth of games, we'd
still have to do some smoothing.
>Let's take your example, tie game 6
>seconds left, man at the line for 2 shots. Let's say he's a road
>player. Unfortunately, who has possession is key here, so the table I
>sent only tells part of the tale.
>Making those two shots changes the nominal odds from 0.49 to 0.75.
>Missing those two shots changes the odds to approximately .25.
>The "clutchness" of this situation is the difference between where
>your odds could be at the high and low ends, or about 0.5. That's a
>huge situation. A more clutch situation is making a field goal at the
>buzzer when down by 1 -- the range there is 1.00 (definitely win) to
>0.00 (definitely lose) or 1.
>So it is defined prior to outcome.
I see -- you are defining the clutchness at time t by the difference in win
probabilities at t + 1. You are, in effect, referring to possible future
events. Makes sense to me, but I can't help but think there's another,
easier way. Or maybe I'm just stubborn.
>> >But even this isn't right. You
>> >need ANOTHER matrix for who has possession in the last minute or two.
>> >I have only built part of that.
>> Working on it.
>I'm gone for a month beginning in a few days. I expect progress by
>the time I'm back!