Jump to content

Providing independent coverage of the Minnesota Twins.
Photo

Pitcher WAR

  • Please log in to reply
13 replies to this topic

#1 amjgt

amjgt

    Senior Member

  • Member
  • 2,774 posts

Posted 13 September 2020 - 06:09 PM

So I was perusing the pitching stats today and was looking at WAR leaders for pitchers (on ESPN, which I think is bWAR)

 

Kenta Maeda is at 1.5, which without context seems just fine. 

 

Then I look above him.

 

Max Scherzer is at 2.1. Hmmmm...

He has and ERA over 4, and a WHIP of 1.4, while pitching the exact same number of innings as Maeda. He's got a decent amount more strikeouts than Maeda but not a TON more, so I chalk up bWAR being focused on K/9 or something.

 

So I look higher.

 

Max Fried 2.8 WAR. Nearly double that of Maeda. K rate is a decent amount lower than Maeda (about the difference between Kenta and Max), but has a lower ERA. That's literally the only stat that he outperforms Maeda on. So why is his WAR twice as high?

 

So.... What the heck does pitcher WAR look at?

 

Update: Maeda is above both pitchers on fWAR. Which really just makes things more confusing


#2 FlauerPauer

FlauerPauer

    Rochester Red Wings

  • Member
  • 1,973 posts
  • LocationSyracuse, NY

Posted 13 September 2020 - 07:04 PM

Who's Max Fried?


#3 spycake

spycake

    Senior Member

  • Moderator
  • 18,169 posts

Posted 13 September 2020 - 07:50 PM

bWAR doesn’t have anything to do with K rate or BB rate or hit rate or HR rate. It is strictly run prevention, with adjustments for context.

Check out the Player Value — Pitching table. That breaks down bWAR a little bit.

They start with the pitcher’s RA9 (like ERA but including unearned runs), then they consider opposition (RA9opp), team defense (RA9def), and park factors (PPFp).

So in B-Ref’s estimation, while Scherzer has a higher RA9 than Maeda, overall he has faced better opposing offenses, has a worse team defense behind him, and has pitched more in hitter’s parks than Maeda. Fried too.

Maeda’s RA9 is 2.43, and B-Ref calculates that the average pitcher would have a 4.04 against the same offenses, in front of the same defense, and in the same parks (RA9avg). By comparison, Scherzer’s RA9 is higher at 3.40 but the average pitcher would have done much, much worse under his same conditions — RA9avg of 6.12. Similar story for Fried: RA9 of 1.98, vs a RA9avg of 5.85.

That’s not to say B-Ref is correct on all of this, but that is the explanation. :)
  • Cap'n Piranha, kenbuddha and wabene like this

#4 Linus

Linus

    Senior Member

  • Member
  • 2,504 posts

Posted 13 September 2020 - 08:08 PM

bWAR doesn’t have anything to do with K rate or BB rate or hit rate or HR rate. It is strictly run prevention, with adjustments for context.

Check out the Player Value — Pitching table. That breaks down bWAR a little bit.

They start with the pitcher’s RA9 (like ERA but including unearned runs), then they consider opposition (RA9opp), team defense (RA9def), and park factors (PPFp).

So in B-Ref’s estimation, while Scherzer has a higher RA9 than Maeda, overall he has faced better opposing offenses, has a worse team defense behind him, and has pitched more in hitter’s parks than Maeda. Fried too.

Maeda’s RA9 is 2.43, and B-Ref calculates that the average pitcher would have a 4.04 against the same offenses, in front of the same defense, and in the same parks (RA9avg). By comparison, Scherzer’s RA9 is higher at 3.40 but the average pitcher would have done much, much worse under his same conditions — RA9avg of 6.12. Similar story for Fried: RA9 of 1.98, vs a RA9avg of 5.85.

That’s not to say B-Ref is correct on all of this, but that is the explanation. :)


Wow. My head hurts. I think I will Just see how many guys he gets out.

#5 spycake

spycake

    Senior Member

  • Moderator
  • 18,169 posts

Posted 13 September 2020 - 09:13 PM

Wow. My head hurts. I think I will Just see how many guys he gets out.


You may be somewhat facetious here, but perhaps I did a poor job explaining it. The basic concept is pretty simple. We all know pitchers face teams of different quality, and that is only magnified in 2020.

The Twins are playing only central division teams this year. The 10 central division teams collectively average 4.35 runs per game. Only 1 team out of 10 (White Sox) above league average in runs per game.

The Braves and Nats face only east division teams this year. The 10 east division teams collectively average 4.96 runs per game. 7 of 10 teams above league average, and 2 more just a hair below. There are 5 central teams who score fewer runs than the worst run scoring east team (Miami).

So you can see how Maeda might be facing inferior offenses than Scherzer or Fried!

And you could repeat the above exercise with park factors, and also compare the Twins team defensive performance with that of Washington or Atlanta. Those are harder to quantify, of course, but the concepts are not foreign to baseball fans.

Obviously you don’t have to go through these exercises if you don’t want to, but the OP was interested enough to ask the question — you don’t have to be.

#6 Brock Beauchamp

Brock Beauchamp

    A Little Teapot

  • Owner
  • 23,980 posts

Posted 14 September 2020 - 08:52 AM

Maeda is likely being punished by his absurdly low BAA (Batting Average Against), which is probably fair... to an extent.

 

I don't know how much of a mirage Maeda's BAA is right now - given the contact he's allowing, it makes sense it should be lower than the MLB average but almost surely not this low - and it's fair to assume there will be some normalization in that regard.


#7 Thegrin

Thegrin

    Senior Member

  • Member
  • 1,123 posts
  • LocationRipon, CA

Posted 14 September 2020 - 09:10 AM

This is why I am a pacifist,(anti-WAR): )


#8 Brock Beauchamp

Brock Beauchamp

    A Little Teapot

  • Owner
  • 23,980 posts

Posted 14 September 2020 - 09:22 AM

This is why I am a pacifist, (anti-WAR) : )

Controlling for aberrant and likely unrepeatable elements in player performance is a feature, not a bug.

#9 Sssuperdave

Sssuperdave

    Member

  • Member
  • 189 posts

Posted 14 September 2020 - 01:06 PM

 

Controlling for aberrant and likely unrepeatable elements in player performance is a feature, not a bug.

 

Soapbox time...:D

 

I like to divide stats into "predictive" stats and "descriptive" stats. Descriptive stats tell the story of what happened... RBI is a great example. It's pretty clear what an RBI represents, and the stat itself tells a bit of a story.If someone has 6 RBI in one game, they may or may not have been in the right place at the right time, but six of their teammates scored as part of their at-bats, and I hope I was watching!  

 

Descriptive stats may or may not be predictive of future performance. RBI is not all that predictive, although it also depends what you are trying to predict. I guess "valuable" is a descriptor I should apply as well.RBI may do an ok job of predicting the next year's RBI, but they are so situation dependent that they don't necessarily say all that much about a players future value.

 

OPS, wOBA, xFIP for pitchers...those are more predictive stats, especially when the sample size is large enough.

 

In that paradigm, WAR has always been a bit of an enigma for me, and I think using this paradigm is helpful for thinking about how the different sites calculate it.It is a counting stat, and most counting stats are more descriptive than predictive.However, fangraphs in particular makes lots of adjustments to WAR that make it more predictive.Baseball References WAR is more descriptive.

 

Personally, I prefer to have WAR be descriptive, so for me controlling for aberrant elements is indeed a bug.I do wish bWAR and fWAR weren't both called "WAR"... I wish they were completely separate statistics, one descriptive and one predictive.

  • ashbury and amjgt like this

#10 amjgt

amjgt

    Senior Member

  • Member
  • 2,774 posts

Posted 14 September 2020 - 01:16 PM

Hopefully I can get my brain to remember fWAR = Future (as in, the Fangraphs version is more predictive)


#11 amjgt

amjgt

    Senior Member

  • Member
  • 2,774 posts

Posted 14 September 2020 - 01:18 PM

Ironically, bWAR (more descriptive in nature) seems to be dinging Maeda on his seemingly unsustainable results more than the predictive fWAR


#12 Brock Beauchamp

Brock Beauchamp

    A Little Teapot

  • Owner
  • 23,980 posts

Posted 14 September 2020 - 01:22 PM

Soapbox time... :D

I like to divide stats into "predictive" stats and "descriptive" stats. Descriptive stats tell the story of what happened... RBI is a great example. It's pretty clear what an RBI represents, and the stat itself tells a bit of a story. If someone has 6 RBI in one game, they may or may not have been in the right place at the right time, but six of their teammates scored as part of their at-bats, and I hope I was watching!

Descriptive stats may or may not be predictive of future performance. RBI is not all that predictive, although it also depends what you are trying to predict. I guess "valuable" is a descriptor I should apply as well. RBI may do an ok job of predicting the next year's RBI, but they are so situation dependent that they don't necessarily say all that much about a players future value.

OPS, wOBA, xFIP for pitchers...those are more predictive stats, especially when the sample size is large enough.

In that paradigm, WAR has always been a bit of an enigma for me, and I think using this paradigm is helpful for thinking about how the different sites calculate it. It is a counting stat, and most counting stats are more descriptive than predictive. However, fangraphs in particular makes lots of adjustments to WAR that make it more predictive. Baseball References WAR is more descriptive.

Personally, I prefer to have WAR be descriptive, so for me controlling for aberrant elements is indeed a bug. I do wish bWAR and fWAR weren't both called "WAR"... I wish they were completely separate statistics, one descriptive and one predictive.

Well said across the board and I agree with pretty much all of this. Different stats are trying to do different things and using the right stat for your needs is important.

#13 Thegrin

Thegrin

    Senior Member

  • Member
  • 1,123 posts
  • LocationRipon, CA

Posted 14 September 2020 - 02:57 PM

On a serious note.I have always wondered about the "replacement" part of WAR.It seems that some positions are more easily replaceable than others.Starters, for instance, would be more difficult to replace.Catchers too. But corner outfielders, however, would be much easier to replace.So how can a measurement like WAR adequately reflect the ease/difficulty in replacing any particular position?:)


#14 Brock Beauchamp

Brock Beauchamp

    A Little Teapot

  • Owner
  • 23,980 posts

Posted 14 September 2020 - 03:34 PM

 

On a serious note.I have always wondered about the "replacement" part of WAR.It seems that some positions are more easily replaceable than others.Starters, for instance, would be more difficult to replace.Catchers too. But corner outfielders, however, would be much easier to replace.So how can a measurement like WAR adequately reflect the ease/difficulty in replacing any particular position?:)

That is addressed somewhat in the way WAR is calculated. Different positions are rated differently both on offense and defense. A replacement player is not a single constant; the replacement level at SS is not the same as replacement level at 1B because the player profiles and performances are different.