THE BOOK--Playing The Percentages In Baseball

I left a comment on his site, but this isn’t really about “little things,” is it?  It’s really about timing.

I suppose I’m being dense, but I don’t follow what aspects of game-state considerations remain in WPA/LI.  As I understand WPA, it effectively adjusts fixed linear weights for any particular game event to reflect four aspects of the game-state context: score, outs, inning, and men on base. I thought by dividing WPA by LI you would in effect be removing these four contextual elements from the calculation.  But I guess I’m wrong.  Which elements of game-state context remain in WPA/LI? My apologies if this has been explained in the past.

Unlike WPA, WPA/LI removes the impact of happening to come up to bat in high leverage situations, on the assumption that this is outside the player’s control. So each at bat is normalized in terms of the overall size of the credit/blame that the batter can get. But within that normalized amount, WPA/LI still allots value based on the relative impact that different events have in that given state (e.g. in some cases a single is as good as a home run). You can click the link in my name for a further explanation (though in the context of RE24 vs. RE24/boLI).

birtelcom, you definitely keep context in when you divide by LI.  Just different context. I’ll quote Tango here:

Runs to win is 10… on average.

So, in that situation, you have a .667 chance of winning.  A safe event adds +.333 wins, and an out is -.167 losses.  You multiply by 1.5 (not 10!) to convert to runs, or about +.500 runs and -.250 runs, respectively.

The game-state still matters; it influences the relative importance of each event.  For example, with the bases loaded in a tie game in the bottom of the ninth, a walk is as good as a home run, so the two have equal value.  The “/LI” removes the relative importance of that plate appearance relative to the average plate appearance.  My example is a really really important situation, maybe an LI of 5.  Success or failure here would count five times as much as the typical PA.  By removing that effect, every plate appearance is judged equally, but within its context.

studes/3: yeah, I like that explanation, of using a different runs-to-win converter.