Reply
 
Thread Tools Display Modes
  #1  
Old 01-30-2020, 09:43 AM
Fotheringay-Phipps is offline
Guest
 
Join Date: Mar 2009
Posts: 12,032

The interaction of individual and team sabermetric measures


Best as I can tell, most of sabermetrics involves putting a value on various statistical measures of players and teams. One issue that occurs to me is that the value of an individual accomplishment can be dependent on what the rest of the team is doing, and can't be assigned a value in a vacuum.

For example, the value of a stolen base will be higher if the team consists of a lot of singles hitters as compared to if the team consists of a lot of low-BA sluggers. Or the relative defensive value of infielders vs outfielders would depend whether the team's pitchers induce more ground balls vs fly balls. And so on for other measures.

Questions that arise are:
  1. Do the WAR and similar measures which apply to individual players adjust for the value of the statistical metrics based on the rest of the team - or at least the situations that the player is in - or are they independent of that?
  2. Do projections of team success which are based on the aggregate of individual measures reflect the interplay between the various individual player measures, or are they independent of that?
  #2  
Old 01-30-2020, 10:48 AM
Jas09 is offline
Guest
 
Join Date: Jan 2007
Location: St. Louis, MO
Posts: 5,740
For #1, they are independent. I have no doubt that teams are using independent WAR-like figures try to determine how individual players will work as a unit, but the typical purpose of WAR is to isolate what the hitter is responsible for from the context surrounding him.

For team projections there are a number of ways to do it, but I'm not aware of any that attempt to fully simulate the interactions between components of individual performance. I do know there are a few metrics that try to measure sequencing and other team-oriented results - typically to attempt to explain why a team may be over- or under-performing what their individual metrics would indicate.

I also have a recollection of a few articles that pointed out that clustering certain types of players would increase value - for example adding a high-OBP guy to a bunch of other high-OBP guys may be more useful than adding a high-SLG guy to the same lineup (or maybe it was the opposite, I'll try to find the article).

For defense, absolutely teams take into account GB/FB pitcher type when choosing defensive personnel. They go way beyond that now using heat maps and hit location charts to shift individual defenders. I haven't seen any team moving OF around between batters, but they probably should consider it for some hitters (getting their best OF defender in the areas of the OF the hitter is most likely to go to). But none of these factors go into calculating a pitcher's WAR, if that's what you are asking. The whole point of WAR/FIP/xFIP/etc are to remove these types of vagaries and inter-dependencies to try to get to underlying talent. Then you adjust based on any context-dependent data you may think is relevant.
  #3  
Old 01-30-2020, 11:51 AM
RickJay is offline
Charter Jays Fan
Moderator
 
Join Date: Jun 2000
Location: Oakville, Canada
Posts: 42,224
Quote:
Originally Posted by Fotheringay-Phipps View Post
Questions that arise are:
  1. Do the WAR and similar measures which apply to individual players adjust for the value of the statistical metrics based on the rest of the team - or at least the situations that the player is in - or are they independent of that?
  1. Generally speaking, no.

    There is no one authoritative version of WAR, but most common sources, like Fangraphs, Baseball Reference, and ESPN, have individual WAR totally separate from the player's teams.

    This is quite problematic. It is a fairly significant weakness in WAR as a measuring tool.

    Quote:
  2. Do projections of team success which are based on the aggregate of individual measures reflect the interplay between the various individual player measures, or are they independent of that?
Well, kind of. The aggregate project doesn't really address the interplay of different skills you're describing. Whether that matters much is a different question.

The main problem with how WAR is done is that it has no relationship to the team's actual number of wins. Suppose the "replacement level" for a team is 45 wins and 117 losses. (It was 45.5-116.5 in the 2019 AL.) One would therefore expect that the Tampa Bay Rays, who won 96 games, should as a group be assigned about 51 WAR (or 50.5, I guess.)

However, the major references don't do it that way. They assign WAR totally based on individual metrics that attempt to measure runs created as a hitter, runs prevented as a fielder, and runs prevented as a pitcher. A team's players are therefore allocated WAR that approximates what their record would be if they had met their Pythagorean projection. So the Rays players are assigned WAR appropriate to a 93-69 team, not a 96-66 team; someone on that team is undervalued by WAR. Cubs players are overvalued; they went 84-78 but the Pythagorean record is 90-72.
__________________
Providing useless posts since 1999!
  #4  
Old 01-30-2020, 11:56 AM
Fotheringay-Phipps is offline
Guest
 
Join Date: Mar 2009
Posts: 12,032
Quote:
Originally Posted by Jas09 View Post
I have no doubt that teams are using independent WAR-like figures try to determine how individual players will work as a unit, but the typical purpose of WAR is to isolate what the hitter is responsible for from the context surrounding him.
It's impossible to "isolate what the hitter is responsible for from the context surrounding him". Outside the context surrounding him, any hit other than a home run is worth zero, for example.

It would seem obvious that the value of any stat is completely dependent on an assumption as to what the other members of the team are likely to do (or to have done). My question is essentially whether that assumed likelihood is based on league averages (or something along those lines) or reflects team averages (or similar).

Relevance for those who are not GMs of teams is in trying to assess the "value" of one player vs another using these sabermetrics. So one guy has 17 more stolen bases and the other guy has 4 more HRs. Which is worth more? I would think the relative value of one versus the other would depend on the team the respective players are on. (Note: the specific example is completely random and intended to illustrate the concept - I have no idea at all whether the two stats are even of remotely comparable value.)
Quote:
Originally Posted by RickJay View Post
Well, kind of. The aggregate project doesn't really address the interplay of different skills you're describing. Whether that matters much is a different question.
Where it would matter is when you're trying to use the individual metrics to project wins for the future, versus assessing the past. So if you up your team's WAR by X on an individual level, does that translate to X additional wins for the team?

Last edited by Fotheringay-Phipps; 01-30-2020 at 12:00 PM.
  #5  
Old 01-30-2020, 12:11 PM
RickJay is offline
Charter Jays Fan
Moderator
 
Join Date: Jun 2000
Location: Oakville, Canada
Posts: 42,224
WEll, originally it was supposed to, but as it is measured now it does not.

Most of the time, though, the difference between individual WAR and actual team effect is very small. When Jacob deGrom won the 2018 Cy Young Award, despite going 10-9 he was credited with 10 WAR. Out of curiosity I went through his 2018 game log, asking myself "If they'd just had some bum pitching these games how many more games would they have lost?" Well, I'll be damned, it was about ten.

However, there are errors. A really good example would be in the 2017 AL MVP Award race, which was a close call between Jose Altuve and Aaron Judge, which had about the same WAR. This thing is, the Yankees as a team were VERY overestimated in WAR; they were like ten games below their Pythagorean. And a big reason for that was that Aaron Judge hit atrociously in the clutch. He was just not good in high leverage situations - in fact, if you do work situational stats into it, he was probably three full WAR worse than the raw number suggests. That was probably a fluke, but it is what it is, and it affected his team. It counts.

The notion of sequencing - how things are arranged and relate from player to player - is a known issue, but it's hard to work out. Bill James noticed 30 years ago that his Runs Created formula exaggerates the performance of the very best players, because, to make a long story short, it is based on how offensive actions create runs at the team level; the formula was calibrated by making sure it worked for a whole team. But when you applied it to a player like, say, Wade Boggs, the formula was kind of assuming that it was a whole team of Wade Boggses interacting with one another (basically an endless stream of singles and walks where there were guys on base all time damn time) and not Wade Boggs interacting with mortals. So far as I know that problem hasn't totally been fixed.
__________________
Providing useless posts since 1999!
  #6  
Old 01-30-2020, 04:15 PM
Jas09 is offline
Guest
 
Join Date: Jan 2007
Location: St. Louis, MO
Posts: 5,740
Quote:
Originally Posted by Fotheringay-Phipps View Post
It's impossible to "isolate what the hitter is responsible for from the context surrounding him". Outside the context surrounding him, any hit other than a home run is worth zero, for example.
Well, I'm not sure I agree with that. Clearly a hit is worth some part of a run. The problem was that a player with 200 hits (but no HR) on a team where nobody else ever got on base was showing up as being worthless (no runs, no RBI) when clearly he isn't. So by using linear weights to determine what portion of a run a single is worth (and similarly a double, triple, SB, moving from first to third on a single, etc) you can at least begin to compare a good player on a bad team to a bad player on a good team.
Quote:
It would seem obvious that the value of any stat is completely dependent on an assumption as to what the other members of the team are likely to do (or to have done). My question is essentially whether that assumed likelihood is based on league averages (or something along those lines) or reflects team averages (or similar).
It is based on "average teammates" basically. Over the entire history of baseball the linear weighted value of a hit is worth something (based on baserunner and out combinations at the time of the hit). You scale all of those weights and generate your "offensive value" metric. You can see some of the guts of that for wOBA here: https://library.fangraphs.com/princi...inear-weights/. The nut of it is: "For wOBA, we have the runs above average for walks (0.29), HBP (0.31), singles (0.44), doubles (0.74), triples (1.01), and home runs (1.39)"
Quote:
Where it would matter is when you're trying to use the individual metrics to project wins for the future, versus assessing the past. So if you up your team's WAR by X on an individual level, does that translate to X additional wins for the team?
At a first-order level, yes, adding a 5 WAR player to replace your 3 WAR player should add 2 WAR to your team wins. I have no doubt that more sophisticated modeling and simulation software could plug that player into an existing lineup and generate some sort of mean/stdev prediction for additional runs generated. Then you have to convert runs to wins, which obviously depends a bit on things like bullpen strength, and possible even your manager's tactical competence.

Predicting baseball is hard...
  #7  
Old 02-04-2020, 07:45 AM
Snarky_Kong is offline
Guest
 
Join Date: Oct 2004
Posts: 8,517
HBP = hit by pitch? Interesting that that is worth more than a walk. Probably obvious to actual fans of the sport, but why would that be?
  #8  
Old 02-04-2020, 08:47 AM
Snarky_Kong is offline
Guest
 
Join Date: Oct 2004
Posts: 8,517
Quote:
Originally Posted by Snarky_Kong View Post
HBP = hit by pitch? Interesting that that is worth more than a walk. Probably obvious to actual fans of the sport, but why would that be?
To answer my own question: apparently the values are based on historic scoring rates. Walks happen more often when there isn't anyone on base (intentional) while HBP happens more often in scoring situations.

So it's not that getting hit is more valuable, it just is more likely to happen in a valuable situation.
  #9  
Old 02-04-2020, 09:01 AM
Damuri Ajashi is offline
Guest
 
Join Date: Jul 2006
Posts: 21,102
Quote:
Originally Posted by Jas09 View Post
Predicting baseball is hard...
Right?!?!? Like what was the war of the trashcan in the Astros dugout?
Reply

Bookmarks

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is Off
HTML code is Off

Forum Jump


All times are GMT -5. The time now is 06:39 PM.

Powered by vBulletin® Version 3.8.7
Copyright ©2000 - 2020, vBulletin Solutions, Inc.

Send questions for Cecil Adams to: cecil@straightdope.com

Send comments about this website to: webmaster@straightdope.com

Terms of Use / Privacy Policy

Advertise on the Straight Dope!
(Your direct line to thousands of the smartest, hippest people on the planet, plus a few total dipsticks.)

Copyright 2019 STM Reader, LLC.

 
Copyright © 2017