WPA Q & A

[tangotiger]

Yes, it makes Vlad a star, in real-time.

But, in hindsight, he may have as well struck out. That's why, you need to decide:
1. do I care about what happens in real-time
2. do I care about what happened, in hindsight
3. do I care about what the past tells me about the future

Choose one, not all.

[Fat Strat]

Yes, it makes Vlad a star, in real-time.

But, in hindsight, he may have as well struck out. That's why, you need to decide:
1. do I care about what happens in real-time
2. do I care about what happened, in hindsight
3. do I care about what the past tells me about the future

Choose one, not all.

OK. I think I'm getting it now, at least somewhat.

Going back to the Pujols and Polanco example for a moment, just for fun (BTW, I had the OPS's backward and edited that).

It doesn't really tell you that Polanco has been better than Pujols at all. It tells you that Polanco has perhaps added some more valuable runs for the Tigers than Pujols has for the Cardinals at this point in the season.

Is that fair?

OR

(assuming for a moment that this trend continued for the course of 1000 PA's or so) That Polanco's performance might be overall more valuable to his team in terms of actual wins than Pujols, regardless of OPS, etc.

Is that a fair understanding?

And I would like to hear more about why it has to be one of those options and couldn't be all of them. After all, wouldn't a WPA tell you both what has happened in the past, what has happened right now, and something about what might happen in the future?

[Michael]

Thank you tangotiger!

[tangotiger]

Fat: you got it.

WPA *can* be used for all three of those things, but it is *not* the best one for each of those three things. WPA is perfect for real-time.

[Fat Strat]

Fat: you got it.

WPA *can* be used for all three of those things, but it is *not* the best one for each of those three things. WPA is perfect for real-time.

OK. Thanks for posting about this!

[haltz]

Fat: you got it.

WPA *can* be used for all three of those things, but it is *not* the best one for each of those three things. WPA is perfect for real-time.

OK. Thanks for posting about this!

Seconded. Thanks for the insight.

[Popeye_Card]

Defense and WPA. How can we go about starting to do something quantitative there? Right now, the pitcher gets all the credit (or demerit) for defensive outs or hits/walks/runs/errors given up.

1.) Should we start tracking WPA for every defensive position player in addition to the pitcher? I.E. if A-Rod handles a ground ball to 3rd and throws the runner out, he gets the +0.03 or whatever credit? Does this help sort out the DH vs. position player MVP debate?

2.) Should errors and mis-plays start counting against players? If Duncan's errors in the field create some -1.50 WPA hole over the season, does that outweigh the 2.5 he might put up on offense over a different LF that you could trot out there?

I imagine trying to do anything in this regard would be working with a huge tangled mess of credits/demerits, but I think something interesting could come out of it.

--P--

[jim]

Defense and WPA. How can we go about starting to do something quantitative there? Right now, the pitcher gets all the credit (or demerit) for defensive outs or hits/walks/runs/errors given up.

1.) Should we start tracking WPA for every defensive position player in addition to the pitcher? I.E. if A-Rod handles a ground ball to 3rd and throws the runner out, he gets the +0.03 or whatever credit? Does this help sort out the DH vs. position player MVP debate?

2.) Should errors and mis-plays start counting against players? If Duncan's errors in the field create some -1.50 WPA hole over the season, does that outweigh the 2.5 he might put up on offense over a different LF that you could trot out there?

I imagine trying to do anything in this regard would be working with a huge tangled mess of credits/demerits, but I think something interesting could come out of it.

--P--

WPA is a zero sum stat, so you would have to break up the defensive portion of the WPA and assign correctly. I don't think it's possible - or at least worth it. Imagine the ol' 1-3-6-1-5-4 putout on a rundown. Who gets what for that?

[tangotiger]

I once posted the leaders/trailers in WPA for the 1999-2002 time period. (I'll see if I can find it in a sec.)

Anyway, what I did with pitchers was two ways:
1. count all balls in play as 100% responsibility of the pitcher
2. count all balls in play as ZERO% the responsibility of the pitcher

And, guess what? With 4 years of data, the results were virtually the same.

From the pitcher's perspective then, whether I'm able to figure out how to properly assign responsibility for a ball hit in the gap (presuming I have hit location data), it becomes not worth it, in the aggregate sense.

Of course, it matters on a game basis. For example, in that Cubs meltdown in 2003, most of those balls were hardhit, and it didn't matter who the fielders were (except for that SS error). But, we know that, because we all saw that game. If all we have left is a game account of it, we wouldn't be able to really know that, and we'd have to rely on hit location and other things to infer responsibility.

As to the fielder, we never really think of his "game saving" plays, but rather just the play itself, irrespective of the game situation.

For this reason, I prefer, for simplicity's sake, to give 100% responsibility to the pitcher, and stick with UZR for fielders.

[Popeye_Card]

WPA is a zero sum stat, so you would have to break up the defensive portion of the WPA and assign correctly. I don't think it's possible - or at least worth it. Imagine the ol' 1-3-6-1-5-4 putout on a rundown. Who gets what for that?

Yeah, that's part of what I was getting at in terms of the tangled mess.

And I would track it independently from the pitcher--i.e. on a ground ball to third, the pitcher gets his credit, and the fielder gets the same credit. It would have to be completely tracked seperately, and not assumed to be a zero sum.

--P--