Does Win Shares overvalue relievers?

First of all, a small correction:

<i>PCL-1 is Win Shares code for Marginal Runs Not Allowed. It’s just like Marginal Runs for batters, and it’s based on the number of runs allowed below 152% of the league average, times innings pitched, for each pitcher. It’s similar to Lee Sinins’ “Runs Saved Above Average” or Thorn and Palmer’s “Pitching Runs."</i>

It’s not at 152%; it’s 152% less an amount derived from team defensive quality. Small quibble, but one that helps good pitchers.

Anyhow, PCL-2 is one of those things I never did get exactly why it was implemented in the way it was. Were you to want this, you’d want Wins Above Team. Why? Adding it overall does not increase the total claim points available to the team. I alluded to this in the above paragraph, but anything that increases the total claim points available hurts the team’s best pitchers (aside from just saying Pedro gets 30 extra claim points just because he’s Pedro, or Rocket, or Koufax, or Lefty).

My personal Pitcher Win Shares reform, untested, goes something like this:

PCL-1—Runs Saved (current PCL-1)
PCL-2—Wins Above Team
PCL-3—Reliever ERC Runs Saved (current PCL-3)
PCL-4—Errors better than league average, divided by 4 (we put errors entirely to the fielders, but pitchers make errors too, and the unearned runs adjustments don’t entirely make up for this)
PCL-5—Strikeouts better than team, divided by 60 (accounts for the shifting threshold which would result were we to extrapolate the pitcher’s own stats to the team rates, then figure the threshold with them).

Making as many of these better or worse than average or team keeps the main two claim point sources, PCL-1 and PCL-3, and the others color.

Thanks, Charlie.  I keep trying to allow HTML code in the comments, but haven’t figured it out yet.  You’re right about the PCL-1 calculations.  I never know how much detail to cover in these articles.

When you say “Win Above Team”, I take it you mean winning % above team %, extrapolated to number of decisions?  Great idea about strikeouts, by the way.  I was going in a different direction to address the same issue.  And I hadn’t thought about pitcher errors.  Should we use specific error counts for the pitcher himself, or were you thinking team errors?

Pitcher errors. Basically an adjusted fielding percentage. Those unearned runs coming from your own errors aren’t just half yours, but all yours, plus it’s an individual tweaking of something that was divided up on a team rate. What might be right for you may not be right for some ...

And yes, wins better than team winning percentage, extrapolated for decisions. Personally, I’d prefer a net quality starts, but we don’t have those throughout history. What Bill wanted, I think, was some sort of quantification of pitching randomness, ie, three starts of 7 IP - 3 R isn’t as good in terms of team wins as going 9-1, 7-3, 5-5 (though I’m not really sure that’s true).

I have done a bit of work on the strikeout issue, actually, using theoretical teams. I originally went for all three of the pitcher-only stats, but I found that Bill implemented his thresholds (the reduction for team fielding) in a way that he was taking these into account. This isn’t a huge adjustment, BTW. Basically a extra Win Share a season for the best pitchers, maybe a bit less. I’m all in favor of giving the best pitchers more Win Shares, since the system probably undervalues them to a degree.

One other thing I have done, when working on the fielding formulae I e-mailed you, was relax the thresholds for pitching and defense. I’d eliminate the minimum/maximum Win Shares entirely, and set the percentages based on a deviation from the team percentage using team strikeout rate only. Anyhow, this leads to eliminating the rule (almost entirely for bad teams, FWIW) that the team pitcher threshold must be at least 1 run above the actual team runs allowed rate. This lets bad pitchers on bad teams pick up Win Shares they don’t deserve. One could drop it entirely, or set the minimum to be the team runs allowed rate. It helps good pitchers on bad teams.

Bingo!  I’ve been finding the exact same things.  I was surprised to find that there wasn’t more impact from assigning pitcher-only events to specific pitchers, but that was the result.  Doesn’t mean it shouldn’t be done, however.

Also, I found the exact same thing with the artificial restraints on pitching/fielding breakouts.  It really came to light with this year’s Tigers.  Their fielders only appeared to be a little worse than average, but those pitchers stunk!

studes: Doesn’t mean it shouldn’t be done, however.

Actually, I found for other than strikeouts, it shouldn’t. Pitchers who allow fewer home runs and walks give up fewer runs anyways. I had to go through everything about three times before I was convinced that the adjustment as already applied to the pitcher threshold as a team was adequate. For strikeouts it doesn’t apply, since technically a strikeout is only a smidgen better than a groundout or a flyout, but responsibility should go all to the pitcher.

I think the one thing that struck me was that, working through all this stuff, was that Bill should have had a little more confidence. Many things on this side of the equation seem arbitrary, but are right and they work. My own revised fielding formulae gave anywhere from about 18 to about 60 Win Shares to team defense from 1974-1982 NL, and in general they matched up pretty darn good with Bill’s formulae. The exceptions were those where I took something else into account, like (especially) catchers’ throwing.

And, as for dramatic contrast, check out some of those mid-1970s Padres teams. The fielders weren’t good, but could take half of the team’s defensive Win Shares because the pitchers really sucked.

I dont think there’s any doubt that it overrates relievers. How can Gagne with 82 innings pitched be so much more valuable than Brown with 211? I support the anti-Joe Morgan approach, and so should all serious sabermetricians. 

Situational Credit: The win-save-hold credit always seemed like the most arbitrary part of the win-shares calculation and always has bothered me a little.  It gives pitchers ‘situational credit’ which opens up a can of worms.  If you give pitchers credit for how they perform in certain game situations, shouldn’t hitters, and fielders get situational credit, too?  I guess there aren’t any convenient hitting stats for this (please don’t bring back the GWRBI).  It is common sense that each run is not equal in terms of increasing or decreasing your teams probability of winning a particular game, but Win-Shares treats them all equally.  That’s okay because we believe it all evens out over 162 games and if it doesn’t it’s just luck anyway.  Win-Shares is still the greatest, but it’s the greatest because of its top-down approach.  Adding credit for wins, saves and holds seems like a bottom up approach and that’s why it bothers me.

Thanks for your comments, Rob.  Actually, Win Shares does credit hitters for hitting in the clutch.  Win Shares uses the most recent version of Runs Created, in which James added credit for two things:

- BA with RISP vs. the batter’s general BA
- Percent of Home runs hit with men on base

That’s one of the reasons you sometimes see funky outcomes with batting win shares.  For instance, Manny Ramirez and Carlos Delgado just about tied for total Win Shares, according to baseballreference.com.

However, using the “clutch hitting” version of Runs Created, as Win Shares does, Delgado comes out about 14 runs ahead.

Right, but that clutch credit is still top-down.  It ignores the score of the individual games and looks at the data as a whole.  Wins, Holds and Saves are dependent on the score and inning of each particular game, which is what I think of as bottom-up.  I don’t think either approach is wrong it just bothers my pedantic nature that it’s mixed.