THE BOOK--Playing The Percentages In Baseball

I’m bringing this thread forward, with a focus on the last section (WPA/LI).

In a recent chat, David Gassko said this:
http://sonsofsamhorn.net/index.php?s=&showtopic=25771

Tom Tango really likes to use WPA/LI, but I don’t really get that—it seems to me like some weird bastard child solution between WPA and pure linear weights.

Now, David is someone who is immersed in things like this.  And, if he doesn’t get it, I must figure most people don’t.

WPA/LI has nothing--nothing at all--to do with WPA.  WPA is concerned completely about the chances of winning, and assigning the difference to the player, even though he had nothing to do with being there to begin with.  This is like you cash in your chips, and you end up being given 10x more than you deserve because there’s a new employee dishing out the change.  Sure, you get to keep the money.  But, you didn’t earn it.  The only thing you deserve is your money, irrespective of things outside your control, but within the context in which you earned it.

If you are in a situation where a HR and a walk are equally valuable, whether that is in a blowout game, or a bottom of the 9th, tie game, it would be insane for me to give you .13 wins for the HR and .03 wins for a walk (which is what a pure LWTS approach would say).

Do you give a QB 12 yards for a completion, regardless if he makes a 5yd screen pass, or a 30yd long pass?  No, of course not!  The QB responds to the context presented to him, and you evaluate him under that context.

It’s the same thing here.  If a HR and walk are equally valuable in a given situation, you give them both +.05 wins.  If a player knows that a walk and HR are equally valuable, he’d be crazy not to change his batting approach.  It’s guaranteed that the pitcher is changing his pitching approach, even more so as he’s proceeding through the count.

The only issue is if the batter can really change his approach appreciably enough.  For example, given situations where a HR and walk are very very close (say a walk is worth at least 80% as valuable as a HR), and you have two groups of hitters (Raineses and Dawsons), do their walk and HR rates change drastically enough?  Do the Dawsons continue to hit HR at the same rate, regardless of the impact of those HR?  Do the Raineses recalibrate their game?  And what about when a HR is worth at least, say, 8 times more than a walk.  What kind of differences do we have?

Certainly, after several years of data, I’ll go with the game-state Linear Weights (i.e., WPA/LI).  Certainly, after a couple of months of data, I’ll stick with context-neutral Linear Weights.  But, there’s a point when one becomes better than the other.

WPA/LI should not be confused, at all, with WPA.  The entire problem with WPA, the leveraged aspect of it, is undone with WPA/LI.

And when evaluating a forecast of “future wins”, you are comparing it to WPA/LI.  That’s what a player contributes.

***

If someone is still confused about game-state Linear Weights, please, elaborate.

Bumping, in view of interest from Fangraphs.

Tom,

Never noticed your post. The problem for me with WPA/LI is that I want to know two things: (1) How much did a player contribute in real terms (WPA), and (2) How good was his performance in a context-neutral setting (Linear Weights). One is good for backward-looking analysis, the other for looking into the future. I don’t see when I would ever want to use WPA/LI (except to understand a player’s “clutch” contributions, I suppose).

What about a game situation when a homerun and a walk give you the same impact to winning (like bases loaded, tie game, bottom of the 9th), but you do NOT want that impact denominated in huge win terms, like +.40 wins added, but in the more normal +.05 wins added?  What do you do?

What you are suggesting is that even though a HR and a walk are worth the same, you want to “count” the HR for more, because it’s more indicative of his talent. 

While the HR/BB is an extreme case, what about a K and a flyout, with runner on 3B and less than 2 outs?  You are again suggesting that even though the K is far more costly than a flyout, you will count them the same, giving no allowance for the fact that a batter (and pitcher) approach that PA differently.

The reality is that batters and pitchers do in fact change their approach based on the game state.  They, without question, change their approach by count.  I don’t see why we can’t evaluate their true talent based on the game conditions at hand.

Surely someone, Ichiro, Erstad, Pujols, Vlad, Bonds… somebody must have a skill appreciable enough to change his batting approach.  On the pitcher side, we can see Glavine, Nolan Ryan and others to likely have this skill as well.

That’s what WPA/LI does: it gives you linear weights by the game state, without the leverage aspect.

Surely someone, Ichiro, Erstad, Pujols, Vlad, Bonds… somebody must have a skill appreciable enough to change his batting approach.

***

Maybe, but then it’s on you to prove it.

There’s no question in my mind that pitchers, certainly, change their approach by game state.  I’m fairly confident that many batters do, too—at least many try to.  Some are more capable than others.

What I always trip on is the difference between WPA/LI and BRAA.  BRAA is based on the base/out situation and denominated in runs, obviously.

WPA/LI is also based on the base/out situation, but also “some” of the game situation, right?  That’s clumsily said—is there a better way to put it?

I don’t see the status quo to necessarily being for me to overcome, rather than you.

For clutch hitting, we don’t know that hitters are “cooler” than others when the game is on the line.  So, we have to prove that clutch hitting exists, and we have to prove how much effect there is.

But, for man on 3b and less than 2 outs, we know that the hitter and pitcher are changing their approach. Since we do know that pitchers and batters change their approach, WPA/LI therefore becomes the correct measure.  You can’t count a K and a flyout the same way here.

Right, BRAA only looks at base/out, but WPA/LI looks at inning/score/base/out.  Whether a pitcher/batter changes their dynamics only by base/out (not aware enough to figure out how inning/score affect things), that’s certainly possible.  Certainly in the earlier innings, that might be true.

And as long as the pitcher changes his approach, the batter by default is facing a new environment.  Whether he changes his approach or not, you have a new “game factor” to consider.  Like Fenway’s Wall, let’s say.

Tango,
This really helps me understand WPA/LI after spending a long time trying to find a good description.  Could you do the same for WPA/pLI please?, and finally, could you explain how the Clutch stat on FanGraphs takes these two into account?