View Full Version : Win Probability, Run Expectancy, Linear Weights: 1957-2006, except 1999
10-08-2007, 12:09 PM
It's in the main comment, plus in the posts that follow.
10-08-2007, 01:35 PM
Great job, Tom!
Would it be possible to get Leverage Index in there, or am I asking too much?
10-08-2007, 01:36 PM
I have similar data Tango except I chose not to exclude home 9th, tie games or anything else for that matter other than forfeited games. My results are almost identical though it seems I have ever so slightly lower values of just about everything (probably because I didn't have a selection bias like you did...when you exclude the bottom of the 9th, you're excluding a lot of really good pitching and not as much really bad pitching).
10-08-2007, 02:00 PM
Right, it's a tradeoff in biases. When I exclude the bottom halves of the 9th and later innings, I do so because... well, you know the reason. Let me explain for those who don't.
If the game ends with men on base and less than 3 outs, the men on base, essentially, are treated as if they were putout. If the bottom of the 9th DOES end with 3 outs, then that means that you know how many runs scored (at least 1 fewer than a win would have needed). So, you've got a selective sampling problem.
On the other hand, by excluding those innings, I am intentionally throwing away legitimate performances of a higher pitching quality than normal (the BA, OBP, SLG is lower in the 9th than in the other innings).
I've done it both ways, and I prefer the former.
It should be noted however that the win probability chart does not exclude anything other than tie games. We don't have the selective issue I described because it doesn't matter by how much you win.
Matt: good point on the forfeited game. I didn't think about that. I'll have to see which field on the games file Retrosheet denotes those games, and remove them from the list.
10-08-2007, 02:52 PM
There's a forfeit field in the gamelog table, Tom...if there's anything other than a blank in it, the game was a forfeit.
10-08-2007, 03:12 PM
Thanks for the alert.
If you've been following the threads at the Retrolist, note that the "generic out" event is not necessarily "out". I make a note of it in the link on my site, preceding the LWTS values post.
10-08-2007, 03:35 PM
When is a generic out event not an out?
10-09-2007, 07:49 AM
When the "receiver" of a thrown ball fails to make the putout (i.e., error). Look for the string "E" in the second, third, fourth and later positions in the event string.
When "E" is the first character of the string, it's marked as an "error" event. But, when it's the 2nd or later, it's marked as a "generic out", even though there were no outs on the play.
You can use the outs on play field to find all those.
You have a similar situation with "pickoffs", where it'd be better if the "pickoff error" were used in the cases where the runner was safe.
Nothing some post-parsing can't handle.