BBTF's Newsblog Discussion :: Post-Gazette: Stats Geek: Statistic sheds new light on Big Hit Theory

Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand

Tuesday, May 16, 2006

Post-Gazette: Stats Geek: Statistic sheds new light on Big Hit Theory

Brian O’Neill discovers the joy of using WPA from FanGraphs...unfortunately, he has to use them on the Pirates.

The bottom four Pirates have been the out machines. They are Chris Duffy (-88), Jose Castillo (-98), Joe Randa (-100) and Burnitz (-181). Burnitz’s WPA is the lowest in baseball. Ditto for the Pirates batters’ cumulative number of -603. Neither ranking should surprise anyone.

The only WPA that shocked me was a negative 60 for Freddy Sanchez. He’s hitting .319 with a .356 on-base average and .511 slugging average and he has done a good job cashing in limited RBI opportunities. But a scan of his games shows he has done most of his best work in early innings or blowouts such as Sunday’s loss, when hits hardly move the odds needle. If Sanchez keeps hitting, his WPA should improve.

Repoz Posted: May 16, 2006 at 05:13 AM | 101 comment(s) Login to Bookmark
Tags: pirates

Reader Comments and Retorts

Go to end of page

Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.

Page 1 of 2 pages 1 2 >

1. Foghorn Leghorn Posted: May 16, 2006 at 11:37 AM (#2021911)

he has done most of his best work in early innings

That says a lot about the value of WPA.

2. fuzzycopper Posted: May 16, 2006 at 12:17 PM (#2021925)

That says a lot about the value of WPA.

I don't understand the surpise here. If a team takes a lead by one run after one inning, there are eight innings left for the opposing team to score. If a team takes a lead by one run after seven innings, only two innings remain to tie (or overtake) the deficit. Somewhat intuitively, the probability of the leading team winning when they're up by one with two innings to go is greater than if they lead by one with eight left to play. This same correlation exists with player feats during the game. Thus, via WPA, a solo shot in the first usually wouldn't have nearly the same amount of value as a solo shot in the seventh (special cases like blowouts notwithstanding).

I think WPA is more of a fun little gadget than a hardcore sabermetric tool, but what's wrong with that?

3. Foghorn Leghorn Posted: May 16, 2006 at 12:20 PM (#2021927)

I think WPA is more of a fun little gadget than a hardcore sabermetric tool, but what's wrong with that?

Nothing. It's when people use it to justify other things. Like in the Rowand crashing into the fence thread.

It's bad sabermetrics.

4. Slinger Francisco Barrios (Dr. Memory) Posted: May 16, 2006 at 12:25 PM (#2021929)

The only WPA that shocked me was a negative 60 for Freddy Sanchez. He’s hitting .319 with a .356 on-base average and .511 slugging average and he has done a good job cashing in limited RBI opportunities.

But don't let that sway your opinion of the stat.

5. Bring Me the Head of Alfredo Griffin (Vlad) Posted: May 16, 2006 at 02:06 PM (#2022005)

The thing with Sanchez is that a lot of his late-inning opportunities have come as a PH, and while that doesn't change the leverage of the situation, it probably does have a negative impact on his chance of success.

6. Matt Clement of Alexandria Posted: May 16, 2006 at 02:19 PM (#2022018)

I don't quite get the hate for WPA. It's another way of measuring value, with its own set of assumptions for regression.

WPA chooses the regress to the mean everything that happens over the rest of the game. It measures in-the-moment change on the chance of winning. A homer in a tied 1st has a lesser impact on winning than a homer in a tied 7th becuase the other team has fewer opportunities to respond. Given that we've regressed all future events in the game, the score, inning, and base-out state of the game makes a major difference in calculating value.

It seems like you could regress score, inning, and base-out state as well, and get something like VORP or SLWTs. Or you could regress some mix of the score, inning and base-out state to consider certain specific aspects of clutch value. Or you could account for what a player did in the context of how the game progressed after that situation as well.

Of all the above, I find the WPA assumptions the most useful, though I'd love to see what difference it made to make different assumptions. WPA accounts for the entire situation at the moment, and assesses that value in ignorance of future events, but with (assumed) full knowledge of past events. That seems to me to best correspond to how a player experiences a baseball game.

7. GuyM Posted: May 16, 2006 at 02:39 PM (#2022049)

Nothing. It's when people use it to justify other things. Like in the Rowand crashing into the fence thread. It's bad sabermetrics.

I'm not a big WPA fan myself, but this is precisely backwards. WPA is exactly the right tool for analyzing the cost/benefit of something like risking injury on a defensive play. Clearly, a fielding attempt that makes sense in a tie game might not make sense if you're down by 7 with two innings to play. I'm not arguing for/against that particular play, just saying that the likelihood that a given play will cause you to win/lose a particular game -- what WPA tells you -- is relevant for that discussion.

8. The Keith Law Blog Blah Blah (battlekow) Posted: May 16, 2006 at 02:58 PM (#2022074)

WPA would seem to be a good tool for deciding the MVP as it gauges the real-world amount of wins that a player added, but a poor tool for predicting future performance as situations/context are impossible to predict.

9. Bizarro ARod Posted: May 16, 2006 at 03:44 PM (#2022132)

10. GuyM Posted: May 16, 2006 at 03:48 PM (#2022135)

WPA would seem to be a good tool for deciding the MVP as it gauges the real-world amount of wins that a player added

I don't think so. WPA only "knows" what has happened prior to a plate appearance; it doesn't know what came later. So it says a 2nd inning HR wasn't a big deal, even if it turns out to be the game winner; it gives credit to RBI guys for hitting at the "right" time (RISP), but can't credit a high-OBP guy for getting on base at the "right" time (prior to hits). So WPA really doesn't capture a player's "true" value. If you want value, you'd do as well or better by looking at good 'ole R and RBI.

Of course, you're right that WPA is an even worse tool for measuring talent.)

11. GuyM Posted: May 16, 2006 at 03:54 PM (#2022142)

A run scored in the first is not worth the same as one scored in the ninth. If you had two pitchers with the same ERA and innings pitched where one was a starter and one was a closer, the closer would have a higher WPA because his innings were more valuable, which they obviously are. It's not any different if a hitter is getting big hits late in a game, those hits are more valuable.

It's totally different. A closer is more valuable because he's used selectively in close games, not late innings. Hitting late in the game is no better than hitting early, unless you do it disproportionately in 1) close games and/or 2) RISP.

12. Matt Clement of Alexandria Posted: May 16, 2006 at 04:00 PM (#2022149)

I don't think so. WPA only "knows" what has happened prior to a plate appearance; it doesn't know what came later. So it says a 2nd inning HR wasn't a big deal, even if it turns out to be the game winner; it gives credit to RBI guys for hitting at the "right" time (RISP), but can't credit a high-OBP guy for getting on base at the "right" time (prior to hits). So WPA really doesn't capture a player's "true" value. If you want value, you'd do as well or better by looking at good 'ole R and RBI.

I think that's basically right. It all depends on what you want to call "value".

What I like WPA for is decision-making evaluations (ie Rowand, or bullpen utilization) and "the clutch." WPA tracks value as it is experienced in a ball game - the guy who hits the 2nd inning homer doesn't know at hte time that both teams' bats will go cold, but the guy hitting in the bottom of the 8th knows how big it would be to hit a blast - which I think is a pretty cool feature.

13. Bizarro ARod Posted: May 16, 2006 at 04:03 PM (#2022153)

Hitting late in the game is no better than hitting early, unless you do it disproportionately in 1) close games and/or 2) RISP.

WPA gives big credit to hitters that do 1) and 2), which for a hitter is equivalent to performing in a 'closer' situation.

It doesn't make sense to acknowledge the value from a pitcher's standpoint, but disregard it from a hitter's standpoint.

14. Home Run Teal & Black Black Black Gone! Posted: May 16, 2006 at 04:15 PM (#2022166)

WPA has all the same faults as RBI's as a performance metric.

It's silly.

15. Bizarro ARod Posted: May 16, 2006 at 04:16 PM (#2022168)

but can't credit a high-OBP guy for getting on base at the "right" time (prior to hits)

Let's say a batter draws a leadoff walk in the ninth inning of a tied game. WPA will give equal value to each batter. If one batter is followed by Albert Pujols and the other by Neifi Perez, the first is obviously more likely to score. Pujols will get the credit when he gets the hit. The first batter doesn't get extra credit for batting in front of Pujols.

16. Randy Jones

Posted: May 16, 2006 at 04:25 PM (#2022177)

The first batter doesn't get extra credit for batting in front of Pujols.

but Pujols is getting extra credit for batting behind the guy who drew the walk...

17. GuyM Posted: May 16, 2006 at 04:29 PM (#2022178)

It doesn't make sense to acknowledge the value from a pitcher's standpoint, but disregard it from a hitter's standpoint.

Of course it does. Teams can insert their closers in high leverage situations, but cannot do the same with hitters ("hey, let's have Pujols bat for Molina here -- it's 9th inning w/ RISP").

The first batter doesn't get extra credit for batting in front of Pujols.
Exactly. WPA gives extra credit for timely hitting with men on base, but not for getting on base in a "timely" way. A system should either extract a player from what his teammates do (like RC or BR), or take account of how a player's performance interacted with that of his teammates'. But WPA is a weird hybrid when applied to hitters.

18. Bizarro ARod Posted: May 16, 2006 at 04:30 PM (#2022180)

So if he hits the double that wins the game who should get the credit?

19. Bizarro ARod Posted: May 16, 2006 at 04:34 PM (#2022184)

WPA gives extra credit for timely hitting with men on base, but not for getting on base in a "timely" way.

A hitter gets extra credit for getting on base in a timely situation, but not for the fortune of being followed by good/timely hitters.

20. GuyM Posted: May 16, 2006 at 04:46 PM (#2022198)

I think this piece by Studes -- who loves WPA -- does a good job of explaining its limitations.

21. Bizarro ARod Posted: May 16, 2006 at 04:51 PM (#2022207)

Teams can insert their closers in high leverage situations, but cannot do the same with hitters ("hey, let's have Pujols bat for Molina here -- it's 9th inning w/ RISP").

But the reason the best hitters are in the middle of the lineup is because it makes them more likely to come up in these situations, compared to batting 1st or 9th. And when they perform in these situations they would get credit.

22. Crispix Attacksel Rios Posted: May 16, 2006 at 04:53 PM (#2022213)

So if he hits the double that wins the game who should get the credit?

Steroids.

23. Dizzypaco Posted: May 16, 2006 at 04:59 PM (#2022220)

I think this piece by Studes -- who loves WPA -- does a good job of explaining its limitations.

I read the piece, and I don't think its very persuasive if you are trying to point out the limitations of WPA. I think WPA is a very underrated statistic by many analysts. I'm not quite sure about the implementation of it, but I think the idea is an excellent one

24. Bizarro ARod Posted: May 16, 2006 at 05:04 PM (#2022227)

In that Studes writes at one point:

So the ninth-inning slugger receives four times as much WPA credit as the first-inning guy. Is this fair? Probably not.

He doesn't really explain in that piece why he feels WPA is not a good tool, other than it does not treat all hits equally, which I guess he believes would be more accurate.

25. Repoz

Posted: May 16, 2006 at 05:05 PM (#2022228)

Studes -- who loves WPA

Yes, Studes has become a regular Harry Hopkins for the WPA.

26. Bizarro ARod Posted: May 16, 2006 at 05:08 PM (#2022233)

Steroids.

I was talking about Pujols, not Bonds. :)

27. Matt Clement of Alexandria Posted: May 16, 2006 at 05:19 PM (#2022250)

He doesn't really explain in that piece why he feels WPA is not a good tool, other than it does not treat all hits equally, which I guess he believes would be more accurate.

What he means is that both the 1st inning and the 9th inning home runs were game-winners. With one, we know it at the time, and so WPA records in, and with the other, its status as game-winning is only known in retrospect, and so WPA ignores that it won the game. I think there's certainly a case to be made, when you're talking about value at hte micro-level, that WPA chooses not to record certain data that matter for a player's value.

I disagree that WPA is a "hybrid." I think one of WPA's greatest strengths is that it's very clear, methodologically and theoretically. It records the value of each event in a baseball game based on the situation in the game at the time the event occurred - value defined as improved or decreased chance of winning that specific game. It doesn't account for future events, which can be worth considering, but I think it's quite well-defined and internally consistent.

28. GuyM Posted: May 16, 2006 at 05:36 PM (#2022272)

I disagree that WPA is a "hybrid."

It's a hybrid in terms of whether the opportunities provided to a hitter by his teammates impacts the metric. Most sabermetric measures (RC, Lwts, BR) try to extract a player from that context (since we know players can't control it, and at least the vast majority have no ability to raise their performance "when it matters"). It's fine to put the player back in context to measure value, but WPA only does that partially. In yesterday's AZ-SD game, WPA says that CH Park "won" almost half a game (.44), while Chad Tracy "lost" about a third of a game (-.3) for his team. One problem: AZ won the game. So why does it matter greatly how many runs Tracy's teammates had scored and allowed -- and how many of his teammates were on base -- for each of his PAs, yet not matter at all that his teammates managed actually to win the game? I don't get it....

29. Bizarro ARod Posted: May 16, 2006 at 06:18 PM (#2022313)

Well if you took an 'objective' system, Park would have been more valuable than Tracy. Looking at just hitting using lwts, Park was +1.41 runs and Tracy -1.25 runs. Converted to wins Park was +.141 wins and Tracy -.125 wins. In that particular games their at bats came at more critical times, so Park got +.337 WPA and Tracy got -.299 WPA. Arizona won that game despite Tracy's poor play and San Diego lost despite Park's good play.

30. Matt Clement of Alexandria Posted: May 16, 2006 at 06:25 PM (#2022319)

So why does it matter greatly how many runs Tracy's teammates had scored and allowed -- and how many of his teammates were on base -- for each of his PAs, yet not matter at all that his teammates managed actually to win the game? I don't get it....

Because every measurement in WPA is a measurement of the known situation at the time of the event. We don't know what the following hitters will do, we don't know who will go on to win, we only know what happened before.

This measures baseball as it is lived by the players, managers and fans.

I agree that there are aspects of "value" that don't seem to be covered in WPA. It's not a perfect stat. But it can be summed up precisely in one sentence, and it's internally consistent. Your criticisms seem to be demanding WPA to be something it isn't - to take future events into account. That's a valid criticism of the stat, but not evidence of its lack of internal coherence.

31. The Spanish Inquisition Posted: May 16, 2006 at 07:10 PM (#2022387)

A first inning home run is only (usually) a game winner if the pitchers do a very good job of holding down the opposing hitters the rest of the game. Thus the difference in WPA between a 1st inning home run and a 9th inning home run should be given to the other pitchers. Is this how it works? If so WPA is a terrific measure of the value of a player during an interval (but says nothing about talent or future performance of course).

32. Mister High Standards Posted: May 16, 2006 at 07:25 PM (#2022413)

Most sabermetric measures (RC, Lwts, BR) try to extract a player from that context '

This is false. They don't remove anything from anywhere - they assume average context. Since the average single is worth .5 runs, they give all singles that value. The problem with that, from a VALUE standpoint is that context is everything. Things don't even out if they did then you would see players with 3 win differences between their offensive + baserunning WARP and their WPA.

33. Matt Clement of Alexandria Posted: May 16, 2006 at 07:31 PM (#2022422)

A first inning home run is only (usually) a game winner if the pitchers do a very good job of holding down the opposing hitters the rest of the game. Thus the difference in WPA between a 1st inning home run and a 9th inning home run should be given to the other pitchers. Is this how it works?

I don't think this works. A ninth inning home run is only a game winner becuase of the work of the player's teammates in getting the game to that point, as well. In retrospect, the two events are similar.

The cool thing about WPA is that it doesn't do anything in retrospect, but always in the moment in the game.

34. Dizzypaco Posted: May 16, 2006 at 07:36 PM (#2022431)

35. GuyM Posted: May 16, 2006 at 07:39 PM (#2022440)

This is false. They don't remove anything from anywhere - they assume average context.

You're kidding, right? Obviously, I mean extract them from their specific context.

36. Mister High Standards Posted: May 16, 2006 at 07:48 PM (#2022461)

Guy - they aren't extracting anything - I get your what your trying to say - I really do, but extracting from their specific context is more or less impossiable as you can't extract a fastball david ortiz gets with on 2-0 count because Manny Ramirez hits behind him. Just like you can't extract Fenway park from Wade Boggs stats by applying a park factor, because some players take specific advantage of that park beyond the "average" double factor or what have you. These are two VERY different things. I should have explained it from the get go rather than the terse "false".

Also, we've been through this before - people on the site love WPA, until it gives them a number they don't like. Google up last year's MVP threads, I was hammered pretty hard for saying Ortiz was offensivly more valueable than A-rod last year, though the defense, baserunning, and positional advantage made it a wash.

37. The Keith Law Blog Blah Blah (battlekow) Posted: May 16, 2006 at 08:04 PM (#2022506)

A ninth inning home run is only a game winner becuase of the work of the player's teammates in getting the game to that point, as well. In retrospect, the two events are similar.

This criticism of WPA I don't understand. The only argument I can see--and it's not trivial--against using WPA to measure value is that it assumes an average opponent.

38. The Spanish Inquisition Posted: May 16, 2006 at 10:44 PM (#2022805)

WPA does take their work into account. If the teammates have done a great job and built up a 8-0 lead, that solo shot to make it 9-0 isn't worth much. However if that solo shot is a walk-off to win the game 1-0 then there's no disputing that that player was very valuable that game.

39. Backlasher

Posted: May 16, 2006 at 10:49 PM (#2022813)

SI,

Where have you been? Many of us are curious about your fork project.

40. The Spanish Inquisition Posted: May 17, 2006 at 03:17 AM (#2023800)

I abandoned it as it seemed like there was insufficient support and the evolution of the site seemed like it was going in a direction that I could tolerate.

41. David Nieporent (now, with children) Posted: May 17, 2006 at 06:31 AM (#2023942)

This criticism of WPA I don't understand. The only argument I can see--and it's not trivial--against using WPA to measure value is that it assumes an average opponent.

I thought Matt described the argument pretty clearly in posts 27 and 33.

The argument you cite is also valid, of course, and IMO is pretty devastating. WPA factors in game state context -- that is, baserunners/outs/score -- but not any other context. (It's not just that it assumes an average opponent, but it also assumes average teammates. If I walk in front of the pitcher, I clearly haven't increased my team's chances of winning by as much as if I walk in front of ARod.) It's not clear to me that it's any less distorting to factor in one type of context but not another.

But the one described by Matt is also valid. Let's say I hit a solo home run in the first inning, and then Bruce Chen pitches 9 shutout innings (a fan can dream, can't he?), and we win 1-0. Now let's say that in his next start, Bruce Chen pitches 9 shutout innings, and I hit a solo home run in the ninth inning, and we win 1-0. WPA thinks that I did a lot more to help the team in the second game than in the first. But I did exactly the same thing in each game.

42. studes Posted: May 17, 2006 at 03:09 PM (#2024108)

He doesn't really explain in that piece why he feels WPA is not a good tool, other than it does not treat all hits equally, which I guess he believes would be more accurate.

Wow. I must be a lousy writer! I do obviously feel that WPA is a good tool, for certain things. Do I feel it should be the "ultimate" measure of value? No, for many of the reasons outlined here. I agree with Matt's comments about it.

Guy, I understand the nature of your complaints, but I don't understand why you consider those "issues." When a batter draws a walk before a big hit, he doesn't know the big hit is coming. Why give him credit? WPA reflects what everyone knows at any point in time about a game, and the likely impact of each play given the current context. That's its virtue and its weakness. I do agree that the Rowand catch was a fine application of WPA thinking. As I said in that linked article, it's all about narrative, and Rowand's catch was part of the narrative.

It's true that, in its current state, it doesn't reflect the quality of the next batter, or the quality of the current pitcher, or the quality of the pitcher who might be brought in if the current batter gets on base, etc. etc. Of course, if you factor those in, you need to then "debit" the appropriate players to make the overall math work out.

I'll leave that spreadsheet to the next person. :)

43. DCA Posted: May 17, 2006 at 03:31 PM (#2024132)

But the one described by Matt is also valid. Let's say I hit a solo home run in the first inning, and then Bruce Chen pitches 9 shutout innings (a fan can dream, can't he?), and we win 1-0. Now let's say that in his next start, Bruce Chen pitches 9 shutout innings, and I hit a solo home run in the ninth inning, and we win 1-0. WPA thinks that I did a lot more to help the team in the second game than in the first. But I did exactly the same thing in each game.

Exactly. And that's a huge problem with WPA, or rather with its most common application, and how MHS uses it as his uber MVP stat.

It's basically a truism that in the same game, for the same team, all runs have approximately* equal value. If you win 4-3, each of the 4 runs is equally game-winning: take any one away, and the game is equally tied. So the way to calculate value after the fact is to replace the player with an average player, sim the season multiple times, and take the difference between actual team wins and expected team wins, and that's your value.

WPA takes the real-time expected win value of the event where you don't know what will come after, and adds it up for later use; whereas at that later time, you are capable of assessing the actual value of the event because you know what came after. And those two sets of values are very different, and it is a huge mistake to confuse them: WPA is a useful stat in real-time, but certainly not retrospectively.

*I say approximately here because in a blowout late-game situation or even a non-blowout 9th inning lead situation, there is some defensive indifference that allows some runs to score that wouldn't be freely allowed if they were likely to affect the outcome of the game. I could buy an argument that these runs are worth a bit less than other runs.

44. Ron Johnson Posted: May 17, 2006 at 03:47 PM (#2024157)

It[WPA]'s bad sabermetrics.

Dunno if I agree.

I think that SNWL is by far the best way to measure the value of a starting pitcher and SNWL is conceptually close to WPA.

I think Doug Drinen's old reliever reports were by far the best way I've ever seen for measuring the value of relief pitchers and they were straight WPA.

Position players? My position has always been handle with care. There are big issues relating to team quality (for WPA -- might be interesting to do something like a support neutral WPA. But that's a honking lot of work) and batting order position. (For those interested, check out the discussions on RSB and RSBA about James Tuttle's methods. Tuttle found it neccessary to normalize opportunities. He was merely doing a simple bases gained metric, but the problems will be there in any method based on situations.)

But as long as somebody else will do the heavy lifting I'm interested in the handful of players who do significantly better or worse by WPA or RPA than by (say) runs created (or any decent metric)

45. Best Regards, President of Comfort, Esq., LLC Posted: May 17, 2006 at 04:06 PM (#2024186)

I'm working on a statistic that measures how much a player contributes to a victory and how much they contribute to a defeat. It's a simple concept but a tricky calculation. I'll publish it as soon as I'm done and confident that it's measuring what I intend it to (the methodology I'm currently using should be right).

46. David Nieporent (now, with children) Posted: May 17, 2006 at 04:19 PM (#2024201)

Guy, I understand the nature of your complaints, but I don't understand why you consider those "issues." When a batter draws a walk before a big hit, he doesn't know the big hit is coming.

I agree, but so what?

Why give him credit? WPA reflects what everyone knows at any point in time about a game,

And that might indeed be interesting, but is that the same thing as value, which is what the stat is being sold as a measure of? Is the value of performance based upon our knowledge of the context, or upon the actual context? (If a tree falls in a forest and nobody is there, does the run count?)

and the likely impact of each play given the current context. That's its virtue and its weakness. I do agree that the Rowand catch was a fine application of WPA thinking. As I said in that linked article, it's all about narrative, and Rowand's catch was part of the narrative.

If we're talking about narrative, that's entirely fair. But if you read the thread above, it's not being described as "all about narrative." It's being described as about value, too.

If it's "about narrative," then why bother to calculate it after the fact? In real time, sure. ("That hit just increased his team's chances of winning from 48% to 56%." Assuming an average opponent and average teammates.) But after the fact, why do we care what we knew at the time he got the hit?

Of course, if one believes that knowledge of context has a significant impact on individual performance -- in other words, "clutch hitting" -- then my argument doesn't hold. In that case, we could say that he got the hit because of his knowledge of context, so it would make sense to care about it. But I don't understand people to be making that argument.

47. Dizzypaco Posted: May 17, 2006 at 04:47 PM (#2024235)

There are two ways of looking at value, not one. It is a truism that every run in a one run game is equally responsible for that win -therefore, every run is equally valuable no matter when it happened. It is also a truism that a run scored in the bottom of the ninth of a one run game did more to affect the team's chances of winning that game than a run scored in the first inning of that game - therefore, the run scored in the ninth inning was more valuable than the run scored in the first inning

Of the two ways of looking at value, neither is inherently superior, its just a matter of which angle you want to look at it from. WPA is the perfect tool if you define value in the second way, and a terrible tool if you define value in the first.

48. Mister High Standards Posted: May 17, 2006 at 04:49 PM (#2024241)

And that might indeed be interesting, but is that the same thing as value

And assuming everything is the same isn't value either. Which is what RC or whathave you does. It's a step forward towards value - not the final step.

49. Bizarro ARod Posted: May 17, 2006 at 04:58 PM (#2024256)

In the examples where a home run is hit in the 1st of 9th of a 1-0 game, I'm not sure if people are saying a more accurate system would give equal value to each home run, or if it's just an argument against WPA.

Assuming it would be a better system, what about a game where someone hit a home run in a 5-0 game, that eventually was won 6-5. Since that became the winning run, it would give much more value than WPA, though I don't think it would be as valuable as the first run, or a home run in the bottom of the ninth of a 5-5 game.

50. Bizarro ARod Posted: May 17, 2006 at 05:10 PM (#2024279)

1st or 9th

51. KB JBAR (trhn) Posted: May 17, 2006 at 05:28 PM (#2024308)

Unless I misunderstand SNWL, I don't think it's all that similar to WPA, since it doesn't really take game state into account. Instead it just asks how often an average offensive team wins when a pitcher gives up x runs in y innings. It measures wins on the basis of a pitcher's actual performance, not run support. Unlike WPA, regardless of whether a team score 1 run or 9 runs, the pitcher will get the same amount of credit.

ARP, or whatever the stat is called, is more similar to WPA since it takes into account baserunners inherited for relievers, but ignores game state. WX(RL) is the most similar stat since it compares a pitcher's effect on win expectancy to an average or replacement level pitcher.

52. David Nieporent (now, with children) Posted: May 17, 2006 at 05:53 PM (#2024355)

It is also a truism that a run scored in the bottom of the ninth of a one run game did more to affect the team's chances of winning that game than a run scored in the first inning of that game - therefore, the run scored in the ninth inning was more valuable than the run scored in the first inning.

To be a truism, I think it has to be true. I disagree with this statement, meaning I don't think it's true, meaning I don't think it's a truism.

The reason WPA seems to say that the first is less valuable (or "did less to affect the team's chances of winning") is because it isn't looking at the value of "the run scored in the first inning of that game." It's looking at the value of the run scored in the first inning of an average game. But the run scored in the first inning of that game is just as valuable as the run scored in the ninth inning of that game.

53. studes Posted: May 17, 2006 at 06:03 PM (#2024377)

Unless I misunderstand SNWL, I don't think it's all that similar to WPA.

Right. SNWL removes context, doesn't it? WPA is all about context.

I think Doug Drinen's old reliever reports were by far the best way I've ever seen for measuring the value of relief pitchers and they were straight WPA.

I don't remember Doug's old reports, but I believe BPro publishes essentially the same reports today on their site. They really do have great reliever stats.

Is the value of performance based upon our knowledge of the context, or upon the actual context?

I believe that it can be based on our knowledge of the context, and I believe a good rationale could be stated that one based on "current" context is more valuable than one based on "retrospective" context (though I don't feel strongly about it; I don't see why both aren't useful). Players perform and deliver in the here and now, and evaluating them based on the incremental, current impact of their performance is most in line with their reality and the reality of anyone watching the game.

Take BPro's reliever runs added/lost report. That is based on the "current" base/out situation when a reliever comes into a game and the impact he had on a situation -- relative to what reasonably could have been expected to happen, not on what did happen retroactively. I personally think this is a great insight for evaluating the impact of a reliever.

54. Bizarro ARod Posted: May 17, 2006 at 06:03 PM (#2024378)

Since the concepts behind Leverage and WPA are essentially the same, I assume arguments against WPA would also apply to Leverage?

55. Dizzypaco Posted: May 17, 2006 at 06:15 PM (#2024395)

It is also a truism that a run scored in the bottom of the ninth of a one run game did more to affect the team's chances of winning that game than a run scored in the first inning of that game - therefore, the run scored in the ninth inning was more valuable than the run scored in the first inning.
To be a truism, I think it has to be true. I disagree with this statement, meaning I don't think it's true, meaning I don't think it's a truism.

Do you disagree that a homerun in a tie game in the ninth inning does more to increase a chance of team's winning a game than a homerun in the first inning of a tie game? For example, if a player hits a homerun in a tie game in the ninth inning, he increases the chances that his team will win by roughly 50%, give or take, from around 50% to around 100%, give or take. Do you believe that a homerun in the first inning of a tie game will also increase the chance that his team will win by 50%?

My guess is that what you disagree with is the second part of the sentence - that the run scored in the ninth inning is more valuable than the run scored in the first, which is reasonable. But the idea that a run scored in the ninth is more valuable than a run scored in the first is no more unreasonable than the idea that all runs in a one run game are equally valuable. Once again, it all depends on your definition of value.

56. DCA Posted: May 17, 2006 at 06:24 PM (#2024412)

It is also a truism that a run scored in the bottom of the ninth of a one run game did more to affect the team's chances of winning that game than a run scored in the first inning of that game - therefore, the run scored in the ninth inning was more valuable than the run scored in the first inning

No, it wasn't -- it *is*, but it *wasn't*. It depends on when you're doing the valuing.

The run in the ninth inning only is more valuable if you do not condition on the actual score/events of the game. This occurs in real time, since we do not have foreknowledge of future events. So you can say that a late inning HR caused a greater change in the probability of the team winning the game as perceived at the time. This approach is good to use for real time strategic decisions.

But, when we analyze players performance after the fact, the results of the games *are* known, and so conditioning on the final score is only responsible approach to take. When you condition on the outcome, all runs are equally valuable.

57. Dizzypaco Posted: May 17, 2006 at 06:35 PM (#2024433)

But, when we analyze players performance after the fact, the results of the games *are* known, and so conditioning on the final score is only responsible approach to take.

This is incorrect.

Take the case of "ace" relief pitchers, used only in highly leveraged situations. According to your definition, all ace relief pitchers are vastly overrated, because the ninth innings of one-run games they pitch in are no more important than the first innings of the same games. Who cares if Mariano Rivera in his prime could shut a team down in the ninth inning of a one-run game? It was no more important than if he shut them down in the second game. In the other definition, it does matter, because the extent to which he affected his team's chances of winning the game were greater in the ninth inning than they would have been in the second.

Another way to look at it is that in your definition of value, no player's contribution has any value whatsoever if his team lost the game. If a guy hits four homeruns in a game that his team loses 5 to 4, his homeruns, by definition, have no value, since we know the final result, and they lost. In the other definition they would have value, because each of his homeruns increased the chances that they would win the game.

If a player hits a homerun in the first inning of a tie game, he increases his team's chances of winning by less than if he hits the same homerun in the ninth inning of a tie game. This is a fact. And given that it is a fact, it is still a fact after the game is done.

People on this site keep insisting there is only one valid definition of value, only one reasonable way of looking at this subject. And it is this that I disagree with most strongly.

58. Kyle S Posted: May 17, 2006 at 06:58 PM (#2024469)

I think WPA does a nice job valuing bullpen contributions. The case of Oscar Villarreal is a good reason why.

Let's say there is a crippling unexpected player's strike today that cancels the rest of the season. In 2035, you crack open The Sporting News 2006 Baseball Register to refresh your memory of the season.

If all that book had was oscar Villarreal's basic stats, you might be fooled. "5 wins as a reliever? Wow! Good ERA. Not great peripherals, but he got the outs somehow. Nice little season there by Oscar the Grouch."

Of course, Villarreal has had a bad season and allowed a bunch of inherited runners to score. WPA recognizes that, as he scores a -43 (almost a whole win below average). Contrast that with VORP, where Oscar scores as one of the top 120 pitchers in the bigs because of his low RA/9 number.

I guess some of the other reliever stats will capture this as well, but WPA does a nice job.

59. DCA Posted: May 17, 2006 at 08:17 PM (#2024581)

Dizzy, in 57 you're setting up a strawman for me.

According to your definition, all ace relief pitchers are vastly overrated, because the ninth innings of one-run games they pitch in are no more important than the first innings of the same games.

Conditioning on the result, the ninth innings of one-run games are no more valuable than the first innings. The trick, though, is identifying the first innings of one-run games in real time. Since the 9th is no less important than the first, and he can't pitch every day, why not wait until the ninth (or eighth) to decide if you want to use the player? That's why teams can, should, and do wait until the late innings to use their best relievers.

Another way to look at it is that in your definition of value, no player's contribution has any value whatsoever if his team lost the game. If a guy hits four homeruns in a game that his team loses 5 to 4, his homeruns, by definition, have no value, since we know the final result, and they lost.

Yeah, there is a sample size problem with my proposed stat. Which is why I suggested simming seasons. However, most games are reasonably close, and have at least a chance of turning on one player's universe of possible contributions.

So, I am of two hands on this ... On the one hand, the result would have been the same if he didn't hit those HR. On the other hand, I could see that his HR have some value, but they all have the same value.

People on this site keep insisting there is only one valid definition of value, only one reasonable way of looking at this subject. And it is this that I disagree with most strongly.

I'm not one of them. Never have been. But there are invalid inferences that are made from using certain definitions of value in certain contexts, and adding up WPA to give players relative credit for their contributions after the game is over is one of those. And that is what I disagree with most strongly.

60. DCA Posted: May 17, 2006 at 08:23 PM (#2024589)

Since the concepts behind Leverage and WPA are essentially the same, I assume arguments against WPA would also apply to Leverage?

From me, no. My quarrel with WPA is that it is used out of it's appropriate context, which is real time -- that's actually a quarrel with the inference not with the stat. But leverage is mostly used as a measure of how important the selected innings were when the pitcher was actually pitching. Which is fine, that's a valid inference (mostly). I do object to value = RSAA*leverage, and in one of the Sox Therapy threads there was an interesting issue brought up where pitchers who give up baserunners create their own leverage ... you bring in two guys with a 1-run lead in the 9th ... one K's the side, the other walks one and gives up a single ... the second guy has a higher leverage even though it's his fault and not a managerial decision.

61. Bizarro ARod Posted: May 17, 2006 at 08:31 PM (#2024598)

But, when we analyze players performance after the fact, the results of the games *are* known, and so conditioning on the final score is only responsible approach to take. When you condition on the outcome, all runs are equally valuable.

I'm a little confused by this. If you are valuing runs on the final outcome, they will have different values depending on whether the game is 1-0 or 10-0.

They would be equally valuable if the outcome or their in game context were disregarded. Excepting of course the general rule where the 1st run value > 2nd run value > 3rd run value etc in an individual game.

62. Dan Turkenkopf Posted: May 17, 2006 at 08:36 PM (#2024603)

I'm a little confused by this. If you are valuing runs on the final outcome, they will have different values depending on whether the game is 1-0 or 10-0.

They would be equally valuable if the outcome or their in game context were disregarded. Excepting of course the general rule where the 1st run value > 2nd run value > 3rd run value etc in an individual game.

I think he's saying that all runs in a given game are equally valuable - not all runs across a season.

63. Bizarro ARod Posted: May 17, 2006 at 08:46 PM (#2024609)

Since the 9th is no less important than the first, and he can't pitch every day, why not wait until the ninth (or eighth) to decide if you want to use the player?

The 9th is not 'no less important', rather more important! By the eighth or ninth inning you can get extra value out of your closer by using him in close games. Conversely in blowouts you can use your worst reliever because even if he gives up runs they have negligible negative value.

I do object to value = RSAA*leverage, and in one of the Sox Therapy threads there was an interesting issue brought up where pitchers who give up baserunners create their own leverage ... you bring in two guys with a 1-run lead in the 9th ... one K's the side, the other walks one and gives up a single ... the second guy has a higher leverage even though it's his fault and not a managerial decision.

That's more of a problem with leverage compared to WPA. In those two situations, there would be a difference in leverage, but the WPA for those two pitchers would be equal.

64. Bizarro ARod Posted: May 17, 2006 at 09:04 PM (#2024623)

I think he's saying that all runs in a given game are equally valuable - not all runs across a season.

So does that mean runs in a 10-0 are worth 1/10 of a run in a 1-0 game?

If I wasn't able to choose WPA to value runs, I think it would be more accurate to give the most value to the first run scored in a game, then less to the second run, etc., regardless of outcome.

Of course almost all measures of value today use 'runs across a season.'

65. Matt Clement of Alexandria Posted: May 17, 2006 at 09:10 PM (#2024631)

The 9th is not 'no less important', rather more important! By the eighth or ninth inning you can get extra value out of your closer by using him in close games. Conversely in blowouts you can use your worst reliever because even if he gives up runs they have negligible negative value.

Right. The point is that, on average, a close 9th is a more important situation than a close 1st. Thus, managers smartly save their best relievers for close games in the last innings.

However, every time someone says "on average", they're talking about regression to the mean. What WPA does is that it regresses the value of certain events to the mean value of that event in all games of a certain score-inning-base-out state. Regression to the mean, basically by definition, excludes aspects of value.

(all numbers below are totally made up for the sake of argument, the exact quantity shouldn't matter)

Let's take a hitter, call him A. Rodriguez. No, that's too obvious, say Alex R. He has a unexpectedly low WPA for his VORP-production because he hit 9 HRs in tied, bottom first innings and only 2 HRs in tied, bottom ninth innings.

On average a home run in a tied 1st has less value than a home run in a tied ninth. The latter always wins the game in every case, and so has a WPA of +.5. The former only wins the game sometimes, and often has a relatively small effect, so it has a WPA of +.1.

However, let's further say that of his 9 HRs in the bottom of the first, an astounding 6 occured in games that the Blue Fork Wankees won 1-0. Each of those was actually a game winner, in retrospect an event equal value to the two walkoffs. WPA does not record this value, because it only records the average value of those first inning home runs, not their actual importance to winning in retrospect.

Basically, I'd say that WPA is a pretty good estimate of overall value. However, there will be cases where WPA misses a significant variation in value because of its regression to the mean.

It's not wholly unlike the relationship of VORP and WPA. For the most part, big VORP means big WPA. However, as in the case of the MVP race last year, that was not actually the case, because Ortiz' production was arranged such that it had much more WPA value than would be expected in a full-regression stat like VORP. however, we don't know if the regression inherent in WPA might have missed certain aspects of Rodriguez' or Ortiz' value.

(My guess is that it didn't, in this case, and the MVP race was a dead heat. But I don't know that.)

66. Matt Clement of Alexandria Posted: May 17, 2006 at 09:18 PM (#2024641)

But, when we analyze players performance after the fact, the results of the games *are* known, and so conditioning on the final score is only responsible approach to take. When you condition on the outcome, all runs are equally valuable.

I agree, but I would clarify that they have different values based on different outcomes. IOW, every run in a 12-2 win has different value than every run in a 12-11 win.

The close games in the 9th inning are more important because you know it's likely that the game will end up being close, and so the pitchers you used will have their value maximized.

67. David Nieporent (now, with children) Posted: May 17, 2006 at 09:20 PM (#2024643)

This is incorrect.

No, it's not.

Take the case of "ace" relief pitchers, used only in highly leveraged situations. According to your definition, all ace relief pitchers are vastly overrated, because the ninth innings of one-run games they pitch in are no more important than the first innings of the same games. Who cares if Mariano Rivera in his prime could shut a team down in the ninth inning of a one-run game? It was no more important than if he shut them down in the second game.

Presuming you mean 'second inning', that's right.

If a player hits a homerun in the first inning of a tie game, he increases his team's chances of winning by less than if he hits the same homerun in the ninth inning of a tie game. This is a fact. And given that it is a fact, it is still a fact after the game is done.

It is not a fact. You're mistaken -- or, rather, your position is correct only due to an ambiguity in the language. The two "tie games" you cite aren't the same. One is a game tied at the time, and one is a game tied at the end.

Given that the home run is the difference between the scores of the two teams, it doesn't matter when it is hit. The difference between the first inning home run and the ninth inning home run is that you don't know, when the 1st one is hit, that it will be the difference. It may be a HR in a blowout, which is less valuable than a HR in a close game. But given that we're discussing the close game, it doesn't matter when it's hit.

That's the same as the Rivera situation: pitching a scoreless first inning in a 1-run game is exactly the same as pitching a scoreless ninth inning in that game. The problem is that if you use him in the first, you don't know that it will be a 1-run game.

68. studes Posted: May 17, 2006 at 09:22 PM (#2024648)

However, every time someone says "on average", they're talking about regression to the mean.

I don't mean to get technical, but I think this is a wrong use of "regression to the mean." WPA does use average run outcomes for base/out situations to derive tables, but there is no regression to the mean involved in WPA.

Regression to the mean is a specific technique involving correlation analysis, or a description of a phenomenon in which a subject's stats move toward the overall population average. As far as I know.

Like I said, I don't mean to be technical, but when you use it in this context I don't understand what you mean.

69. Bizarro ARod Posted: May 17, 2006 at 09:31 PM (#2024657)

The widely acknowledged problems with using a pitcher's won/lost record in evaluating a pitcher's effectiveness shows why using outcome distorts value.

I think this can also be applied to hitting. By giving a hitter extra credit for home runs in 1-0 wins, it's similar to, say, giving a pitcher a win when he gives of 6 runs in an 8-6 victory. Or a no decision when a pitcher is pulled after giving up 6 runs in the first inning. The result of the game shouldn't determine the value of hitting or pitching.

To me: WPA > Context Free > Outcome Based

70. DCA Posted: May 17, 2006 at 09:34 PM (#2024659)

I agree, but I would clarify that they have different values based on different outcomes. IOW, every run in a 12-2 win has different value than every run in a 12-11 win.

Thanks Matt. For that clarification and for many others in your last couple posts.

71. Dan Turkenkopf Posted: May 17, 2006 at 09:34 PM (#2024660)

I think this can also be applied to hitting. By giving a hitter extra credit for home runs in 1-0 wins, it's similar to, say, giving a pitcher a win when he gives of 6 runs in an 8-6 victory. Or a no decision when a pitcher is pulled after giving up 6 runs in the first inning. The result of the game shouldn't determine the value of hitting or pitching.

72. 185/456(GGC) Posted: May 17, 2006 at 09:48 PM (#2024678)

I should comment on this thread because WPA intrigues me, but I need to read it and the article first.

73. Ron Johnson Posted: May 17, 2006 at 10:08 PM (#2024708)

Unless I misunderstand SNWL, I don't think it's all that similar to WPA.

What I was thinking about when I wrote that is that there is quite literally only one difference in the inputs to SNWL and WPA for a starting pitcher.

However that one difference (the specific run support) is huge. Classic case of missing the forest for the trees on my part.

74. Bizarro ARod Posted: May 17, 2006 at 10:12 PM (#2024711)

But using WPA you're giving a batter who hits a walkoff homer in a tie game extra-credit for the fact that his team managed to keep the game tied. How is that different from rewarding the pitcher for a win in your example?

Giving the teammates credit for a player's walkoff home run would be similar, in a way, to the benefits a pitcher gets for a win or no decision when he pitches poorly.

If both teams have been keeping it tied, the teammates with the player hitting the home run shouldn't benefit from his home run, he should. Just like the credit in an 8-6 win by a pitcher should go to the people who won the game, the hitters.

Using outcomes you could say a 2-1 win by a pitcher is equivalent to an 8-6 win by a pitcher since they both won, but they aren't.

75. Matt Clement of Alexandria Posted: May 17, 2006 at 10:19 PM (#2024720)

I don't mean to get technical, but I think this is a wrong use of "regression to the mean." WPA does use average run outcomes for base/out situations to derive tables, but there is no regression to the mean involved in WPA.

Regression to the mean is a specific technique involving correlation analysis, or a description of a phenomenon in which a subject's stats move toward the overall population average. As far as I know.

I think that's right - I trust others, my official stats training is one semester my junior year.

But what is it called, then, when a projection uses as one factor the population mean? I guess that's a method of accounting for regression to the mean. What WPA does is that it takes the exact context and replaces it with the context of the mean set of events. What's the term for that?

One thing I'd add is that WPA is good for the butterfly effect. In other words, if a guy hits a homer in the top of the 9th to take a 4-3 lead, and then his teammates proceed to tack on five more runs, I think that homer is worth more than the average run in a 9-3 win, because it's hard to be sure that all those runs would have come across if he hadn't hit the homer.

I wouldn't use WPA exlusively and assume a complete butterfly effect, but I'd rather not assume that every event would happen completely independently of previous events. I don't think there can ever be one perfect number for value, it's good to look at a whole bunch of 'em.

76. Bizarro ARod Posted: May 17, 2006 at 10:29 PM (#2024724)

Using outcomes works relatively well in a system like Bill James' Win Shares, because he is using a whole season.

It would be very distorting to use for 1 game, where all the credit went to the winning team, because terrible play on the winning team would be disregarded, and good play on the losing team would be ignored.

77. Ron Johnson Posted: May 17, 2006 at 10:49 PM (#2024735)

Dave last time I checked the Prospectus reliever reports were closer to RPA than WPA. Even so, pretty useful (and it's been a while since I looked at the specifics of their method)

Doug's reports had a lot more than the WPA, though the WPA stuff was the core.

He also included a great usage chart. Here's the 1999 Yankees

           7th         8th         9th
           0  +1  +2   0  +1  +2   0  +1  +2
Grimsley   1   3   0   1   2   0   2   0   0
Mendoza    2   3   2   4   5   0   5   0   0
Naulty     1   0   0   1   0   0   0   0   0
Nelson     0   2   0   1   0   1   0   0   0
Rivera     0   0   0   0   0   0   4  14  12
Stanton    1   2   1   5   4   3   0   0   0
Watson     1   0   1   2   0   1   0   0   0
Yarnell    1   0   0   1   0   0   0   0   0
Starters  16   8   9   7   1   5   0   0   1
Total     23  18  13  22  12  10  11  14  13

Don't know about you but I think there's an awful lot of useful info packed in there.

He also included the number of times a guy was brought in with runners on, the number of times a reliever went more than an inning and the number of times a reliever entered a tie game.

78. Ron Johnson Posted: May 17, 2006 at 10:57 PM (#2024741)

Gack. Sorry for the terrible presentation. Looked fine in the live preview. *&^%$

I put a fair amount of time into formatting that table. Waste of my time and damned annoying.

Jim, if I can't post tables in a meaningful manner I can't see posting here period. I'm serious.

79. Der Komminsk-sar Posted: May 17, 2006 at 11:33 PM (#2024787)

test

             7th      8th      9th
           0 +1 +2  0 +1 +2  0 +1 +2
Grimsley   1  3  0  1  2  0  2  0  0
Mendoza    2  3  2  4  5  0  5  0  0
Naulty     1  0  0  1  0  0  0  0  0
Nelson     0  2  0  1  0  1  0  0  0
Rivera     0  0  0  0  0  0  4 14 12
Stanton    1  2  1  5  4  3  0  0  0
Watson     1  0  1  2  0  1  0  0  0
Yarnell    1  0  0  1  0  0  0  0  0
Starters  16  8  9  7  1  5  0  0  1
Total     23 18 13 22 12 10 11 14 13

80. Der Komminsk-sar Posted: May 17, 2006 at 11:34 PM (#2024790)

dang.

81. Bizarro ARod Posted: May 18, 2006 at 12:27 AM (#2024903)

Here are the top 5 AL relievers using BP's WX and WPA

Papelbon...+1.815 +1.880
Ray..........+1.433 +1.454
Zumaya.....+1.520 +1.403
Ryan.........+1.322 +1.330
Rodriguez..+1.068 +1.245

I don't know how WX is calculated, but both measures seem to agree about value, a quarter of the way through the season.

If WPA was inherently flawed in concept it would seem it should diverge more. Since I assume WX is based on season totals, it would stand to reason that on an individual game basis, WPA would be more accurate.

82. Harold can be a fun sponge Posted: May 18, 2006 at 01:23 AM (#2025010)

Since the concepts behind Leverage and WPA are essentially the same, I assume arguments against WPA would also apply to Leverage?

No, I don't think so. When you look at a lot of WPA data (like last year's season totals), you start to realize that a lot of the data doesn't seem to make any sense (Tony Clark as 3rd-best batter in the NL last year?!). And the reason is that each PA's outcome is essentially multiplied by its leverage and then summed. When you do that, you lose information both about the leverage and about the context-free outcomes.

I think the WP framework leads to lots of interesting analyses; Leverage is one of them, but I'm increasingly believing that WPA (as a performance metric) is near useless. Well, it's useless in the way that RBIs or fielding percentage is useless: there's good information in there, but lots of noise, and lots that's missed.

83. Harold can be a fun sponge Posted: May 18, 2006 at 01:26 AM (#2025029)

I do object to value = RSAA*leverage, and in one of the Sox Therapy threads there was an interesting issue brought up where pitchers who give up baserunners create their own leverage ... you bring in two guys with a 1-run lead in the 9th ... one K's the side, the other walks one and gives up a single ... the second guy has a higher leverage even though it's his fault and not a managerial decision.

This is a problem specifically with Tangotiger's implmentation of LI, not the concept of leverage. Tango's LI is on a PA-by-PA basis. Doug Drinen measured a similar statistic, but his unit of time was a partial inning (basically, the length of time that a given pitcher is on the mound). He looked at the WPA of a perfect partial inning, and called it "P" (short for "perfection", I think). This implementation isn't perfect (it looks at the value gained by getting outs, but not the cost of runs, which is probably a bigger issue when using this for batters).

84. Harold can be a fun sponge Posted: May 18, 2006 at 01:29 AM (#2025042)

Tango's LI is on a PA-by-PA basis. Doug Drinen measured a similar statistic, but his unit of time was a partial inning (basically, the length of time that a given pitcher is on the mound). He looked at the WPA of a perfect partial inning,

BTW, Tango's counter-argument here is that, in real time, a pitcher can be pulled after every PA, and so the proper unit of time is the PA. If a pitcher does raise his own Leverage, the manager does actually make a decision to stay with him in a higher-leverage situation.

85. Chris Dial Posted: May 18, 2006 at 02:24 AM (#2025202)

It[WPA]'s bad sabermetrics.

Ron,
you over-snipped me. I don't care for WPA as an overall value tool (I understand it's use for relievers), but I meant the specific use of WPA in the Rowand thread was bad sabermetrics.

86. Harold can be a fun sponge Posted: May 18, 2006 at 02:32 AM (#2025233)

I meant the specific use of WPA in the Rowand thread was bad sabermetrics.

Care to elaborate? It seemed like a pretty good use of WPA.

87. KB JBAR (trhn) Posted: May 18, 2006 at 02:36 AM (#2025256)

WX is a win expectancy measure; it basically is win expectancy measured against the average. WXRL is against a replaement level. So esesentially WX and WPA are the same thing.

If one wants a touchstone for reasonableness, compare WX to ARP or RSAA. WX should map on a little better to ARP than straight RSAA, since ARP is sort of a hybrid measure. It includes inherited runners scored against compared to replacement level.

88. KB JBAR (trhn) Posted: May 18, 2006 at 02:43 AM (#2025299)

BTW, Tango's counter-argument here is that, in real time, a pitcher can be pulled after every PA, and so the proper unit of time is the PA. If a pitcher does raise his own Leverage, the manager does actually make a decision to stay with him in a higher-leverage situation.

Also, the extent to which a pitcher creates his own leverage is going to be reflected in his performance record.

89. John Lowenstein Apathy Club Posted: May 18, 2006 at 02:43 AM (#2025305)

I'm increasingly believing that WPA (as a performance metric) is near useless. Well, it's useless in the way that RBIs or fielding percentage is useless

Not useless. Dammit, will you people stop saying "useless"! NOTHING in the world of baseball statistics is useless, or almost nothing anyway. Something may not do what you want it to do, or may not do everything, but that doesn't make it useless.

WPA isn't best compared to RBIs or fielding percentage (both of which are extremely useful, anyway). It's best compared to GWRBI, pitcher wins or saves.

WPA is a "payoff" stat (a very sophisticated one, as opposed to a crude one like Game-Winning RBI).

Tony Clark wasn't the 3rd-best hitter in the NL - anyway, that's not what WPA tries to measure. But if you use WPA to argue that Tony Clark's hitting paid off more than all but two other National Leaguers, I think that's a reasonable (if questionable) argument.

90. Chris Dial Posted: May 18, 2006 at 02:58 AM (#2025356)

Care to elaborate? It seemed like a pretty good use of WPA.

The play was worth 0.278? In the first inning?

91. Bizarro ARod Posted: May 18, 2006 at 03:01 AM (#2025366)

Where is WPA for 2005?

92. Bizarro ARod Posted: May 18, 2006 at 03:07 AM (#2025384)

WX is a win expectancy measure; it basically is win expectancy measured against the average.

WX is calculated summing values for each PA, not a formula using totals?

93. GuyM Posted: May 18, 2006 at 03:30 AM (#2025407)

The play was worth 0.278? In the first inning?

Why do you find that implausible? If 10 runs = 1 W, we'd expect the three+ runs saved by the catch to be worth about this much. Anyway, it is correct. Did you really not even check before calling it "bad sabermetrics?"

94. Bizarro ARod Posted: May 18, 2006 at 04:43 AM (#2025447)

Since Tony Clark was 3rd in the NL in OPS among players with at least 300 at bats, I don't see why Tony Clark is an example of WPA absurdity.

95. studes Posted: May 18, 2006 at 04:43 AM (#2025448)

WX is calculated pretty much exactly the same way WPA is. Same thing, different in some of the details.

Ron, those are great reports that Doug put together. I agree. BTW, BPro does has WPA (they call it Win Expectancy) for relievers.

What WPA does is that it takes the exact context and replaces it with the context of the mean set of events. What's the term for that?

Matt, by this do you mean that it assumes average hitters are coming up to bat next, instead of specific hitters with known performance outcomes? If so, that's just assuming average hitters. :)

96. GuyM Posted: May 18, 2006 at 11:32 AM (#2025492)

Perhaps using run expectancy -- rather than win expectancy -- would capture what WPA fans are looking for here, without some of the shortcomings that some of us have identified with WPA. If you calculated "RPA," it would reward hitters who performed better with men on base. It would also do things that seem pretty dubious, like awarding more value to a hitter who draws a BB with 0 outs than when there are 2 outs -- but WPA would do the same thing. What RPA would eliminate is the influence of inning and score, which for analyzing hitters I think you want to do.

97. Foghorn Leghorn Posted: May 18, 2006 at 11:38 AM (#2025495)

Why do you find that implausible? If 10 runs = 1 W, we'd expect the three+ runs saved by the catch to be worth about this much. Anyway, it is correct. Did you really not even check before calling it "bad sabermetrics?"

Did you read the usage?

98. GuyM Posted: May 18, 2006 at 11:54 AM (#2025498)

Yes. You?

Seriously, what's your argument? That is was worth more than .278? Less than .278? It's incalcuable? It doesn't matter?
I assume you have a point here...

99. Foghorn Leghorn Posted: May 18, 2006 at 12:31 PM (#2025506)

That it would take 100 ABs to equal teh play.

100. GuyM Posted: May 18, 2006 at 02:06 PM (#2025554)

Agreed that 100 PAs is too high an estimate -- it's probably more like 10-12 games, depending on how productive you think Rowand and his replacement are (both on offense and defense). But that's not a problem with the WPA half of the calculation that was made.

Page 1 of 2 pages 1 2 >

You must be Registered and Logged In to post comments.

<< Back to main

Tuesday, May 16, 2006

Post-Gazette: Stats Geek: Statistic sheds new light on Big Hit Theory

Reader Comments and Retorts

BBTF Partner

Support BBTF

Hot Topics