THE BOOK--Playing The Percentages In Baseball

Can you post the above as a reponse on that website?  Not that the guy is going to change his mind and admit the error of his ways.  People have an incredible ability to dig their heels in even when they are wrong.

The following was posted on SOSH, and I am reposting it here:

==================================

one of the problems being that players on good teams will have higher WPA’s, just by virtue of them having more wins

FALSE!

Noah is completely right that it’s players on .500 teams that will be inflated a smidge.

Well, its been pretty widely proven that the home team wins (off the top of my head) 53.5% of the time. So right off the bat, WPA has a major flaw.

That’s the AVERAGE home team.  In any case, if you want to provide the necessary charts, and perhaps provide different ones for different parks, be my guest.  In the end, it’s a minor issue that’ll come out in the wash, since every team plays the same number of games at home and on the road.

WPA, as a framework, allows you to provide whatever parameters you want.  If certain sites choose to have a different implementation, that’s not an issue with the WPA framework.  That’s an issue with that particular implementation.

The biggest issue with WPA, imo, is that it is almost 100% a retropsective stat, not a predictive one.

But, isn’t OBP, SLG, BA, ERA, and every single stat in existence a “retrospective” stat?  You are implying that WPA doesn’t have the byproduct of being used in a predictive fashion.  I don’t see why that is an issue, any more than ERA would be an issue.

A good (and perhaps necessary) next step for WPA would be to generate a separate set of value tables for each ballpark based on its run environment, thus park-adjusting the stat.

A person who says this has an excellent grasp at what WPA is doing and not doing.  This quote is dead-on, on the absolute necessity to get the right run environment to generate the win probability tables.  90% of the problem with any WPA construction must resolve the run environment issue.  Otherwise, a pitcher and hitter can’t be compared fairly.  Until this is addressed, anything else is nothing more than a tweak.

Adjusting for opposition quality would be a bitch but would be a huge step in measuring actual value. There’s a major problem in that the batter / pitcher matchup ceases to be symmetrical. That is, giving up a walk-off homer to Alex Gonzalez should reduce WPA a lot more than giving one up to Ortiz (or Hafner). Yet obviously Gonzalez and Ortiz should get the same credit for hitting the same walk-off HR off the same pitcher. Basically, for the pitcher, you want to recalculate the run environment based on the current and following batters, but not based on his own quality, and for hitters, you want to recalculate the run environment based on the opposition pitcher, but not based on the quality of the lineup they’re part of. Which of course destroys the zero-sum nature of the metric.

Another excellent quote, and is 9% of the problem with WPA.  In fact, what needs to happen is to look at it from the standpoint of a gambler.  If Santana is pitching, he is “preassigned” WPA before the game even starts.  In fact, all 50 players are preassigned WPA, based on their expectations for that game.  By the end of the season, if all the preassigning was done “perfectly”, then every player’s in-game WPA will total to exactly zero, and you are left with the sum of his preassigned WPA.  Which of course makes WPA itself useless if that was the case!

The larger power of WPA is to represent in-game what is happening.  Its utility on a seasonal basis is more for an historical account, and its use for analysis and prediction would be trumped by other metrics.

WPA is what it is.  Nothing more, nothing less.

***

As for Ortiz/Hafner, Ortiz was unreal in crucial situations, and Hafner, not so much (even with all those grand slams).  There was a 3.3 win difference that is completely attributed to the timing of Ortiz’s performance versus that of Hafner.  That is a huge difference.  Whether the timing was by design is irrelevant.  Redsox got 2 extra wins because of it, and the Indians had 1 extra loss on it.

A followup to my SOSH post:

=============================================

Tango, I’m not sure I understand what you’re getting at here. You could take the starting lineups and starting pitchers and run a large number of simulations (using player stats) to determine the expected WPA for each player. But where does that lead to?

This leads to EXACTLY where you were going with, with your example.  If you have Eckstein hitting in front of the pitcher, or hitting in front of Pujols, the value of his performance is greatly affected.  A walk is far more valuable if you have someone who can leverage it.

Therefore, at its ultimate, WPA must account for all known variables.  Exactly what a gambler would do, and exactly what a fan would do.  The fan is aware of the identities of all players involved, and therefore, the win probability tables should reflect those parameters.

However, once you do that, the in-game WPA of Pujols and Eckstein, over a long period of time, will equal to exactly ZERO!

That is, if Eckstein gets on base in front of Pujols, Eckstein gets a huge plus on his walk.  But he got the huge plus because of Pujols.  Pujols is jipped out of that, because to Eckstein, Pujols is part of the context. 

Now, once Pujols is at bat, the opposing pitcher sees him, and he’s scared to death.  If he gets him out, that’s a huge plus for the pitcher.  However, this has to be a zero-sum situation.  If Pujols only gets the standard minus, who ends up getting the rest of the minus?

Therefore, in order to resolve the Pujols on deck, and Pujols at bat situations, you must treat every player as being in the context.  We are no longer comparing players to the “average player”.  We are now comparing players to our expectation, given the context.  And that means, essentially, comparing players to their own averages.

In the end, if you made a perfect guess as to each player’s average, and the expectant batter/pitcher matchup that would result, then, over a long period of time, the sum of all in-game totals, for each and every player, will be exactly zero. 

All of the value comes in the preassignment of wins, exactly the way a fan and gambler would do it if they see Santana pitching, or knowing if Pujols is playing or not.

What this means then is that the value of WPA comes in looking at games in isolation, to see who performed better than their own expectations.

***

Two things:
1 - It is 1 million times harder to do it the right way, than the quick way.  You will, in effect, get just about the exact same answers anyway.  So, unless you are a real-time gambler, don’t bother.

2 - The preassignment of wins must now decide how a reliever will be used (amount of leverage), and clutch performance (to the extent that it exists).  You can of course choose not to do this, and simply let the in-game numbers capture the differences.  That is, you are not worried about getting the in-game numbers to sum to zero for each player, because, in the end, whatever wins you preassign, and whatever wins results in the in-game, will add up to what they should be, no matter how you handle WPA.

Just want to point out that users can generate different WPA tables for different parks in my spreadsheet.  Plus, there is an option to give the home team an advantage.

I have to say that I’m not a fan of adjusting WPA for the quality of the participants at any point in time.  I understand why some would like to do it, particularly gamblers and in-game managers, but the purer version of WPA also has value in that it assigns WPA to players based on what they do, not what they’re expected to do.  I personally have a preference for the cleaner approach.

It all depends on what you are trying to do, of course.  If the bases are loaded in the bottom of the 9th, down by 2, and Bonds is at bat, with Mayne or deck, a “clean” WPA table will give the same win probability if either Bonds or Mayne would have been at bat, and therefore, the same gain if they have the same result. 

Strategy-wise, manager or fan, or betting-wise, gambler, that’s not the case.  It depends what you want to do.

The only reason I prefer the cleaner approach is stricly because of time.  If I were retired and had oodles of time, I’d opt for the manager/fan/gambler approach.

I noticed this quote from Palmer in The Hidden Game, when discussing the Mills Bros system. The reason I have always missed it is that it’s not in the body of the chapter, but in the ‘notes’ section.

“The major flaw in the Mills brothers’ system is that the Player Win Average weights a few events very heavily, many others quite lightly, so that it effectively has a smaller sample and is therefore less accurate. A combination of overall and situational data would be better.”

It looks like Palmer didn’t really understand win probability at that time.

David,

I think Palmer understood win prob pretty well at the time (and now).  In the Hidden Game, he talks about how ace relievers have twice the leverage than starters.

Without remembering the context of the above quote, his quote seems fine to me.  It sounds like all he is saying is that you can’t get a good enough read on a player’s talent level by looking at his PWA, since you are intentionally introducing noise that the player himself can’t control.

Can you expand on the context of his quote?

Well, my interpretation is that he is starting out by judging WP by the standards of regular Lwts. Then, when he says that a combo of regular info plus situational info should be prefered, it seems like he is implying that the base-out situation should perhaps be included (as in value-added), but not the inning-score context. He doesn’t seem to realize that it is inherent to WP that some PAs will correctly receive a very high weight.

It’s pretty clear to me, unless someone can convince me otherwise, that Palmer was using an inappropriate standard to evaluate WP.