THE BOOK--Playing The Percentages In Baseball

I’d be interested in hearing thoughts on pitcher batting positional adjustments as well and then how to handle that at the team level, and if you should take positional adjustments into account at all when looking at team totals.

What I have done is use the average batting production of pitchers each year to set the position adjustment.  So that the average hitting pitcher comes out to zero WAR.

This is exactly what not to do at the other positions, where if shortstops hit about as well as 2B or even if center fielders hit as well as 1B, we still want to give the CF/SS credit for playing tougher defensive positions.

But for pitchers, I didn’t have any better ideas.

David, does your batting runs number include SB/CS?  I noticed a huge difference in batting runs for Luis Aparicio between Fangraphs and my site, but a lot of that is explained if you put his steals in the bat column while I’ve got them in his baserunning column.

For pitchers, that’s exactly the right thing to do.  The whole idea is to think of “pools”.  We have these pools:

A. Pitchers
B. Catchers + Pool A
C. Infielders + Pool B
D. Outfielders + Pool C
E. Firstbasemen + Pool D
F. DH + Pool E

In each pool, we compare to the average.  So, for pitchers, we compare off and def runs to the average for pitchers only.  From there, you get to a replacement level.

For Pool B, it really is only catchers (since no pitchers really qualify to be catchers).  So, again, perfectly fine to compare to average.  From there, you get to a replacement level.

For Pool C, that’s where it gets tricky.  You have to baseline the 2B, SS, 3B, and then compare to the average.  From there, you get to a replacement level.

The replacement level of Pool B and Pool C don’t have to be the same.  Ideally, it would be set so that the catcher who becomes a 3B has the same value above replacement whether in Pool B or Pool C.  If he stays in Pool B, he should have more value in Pool B.  If he moves to Pool C, he should have more value in Pool C.  For example, Brandon Inge probably has more value in Pool C than Pool B.

So, when you set up the replacement level, it should have that kind of logical underpinning.

Same thing as you proceed down the line.  If you have Frank Thomas as a DH in Pool F, he should have slightly more value there than in Pool E.

And so on.

The big problem is that the pools are not constant year-to-year.  Well, they kind of are year-to-year, but 5 or 10 years at a time, and things start to change. 

It gets kind of dicey.  In the end, we take our best shot, and hope to get something that we can at least be able to justify reasonably.

Going back to David’s point, I guess we have to PoolA:

PoolA1: Pitchers who bat
PoolA2: Pitchers who don’t bat + PoolA1

Same thing applies.  If you have Ben Sheets, you can’t have it that his poor batting reduces his overall value as a ballplayer.  That’s because he really should be an AL pitcher.

Same thing for Micah Owings if he were an AL pitcher.  He should count in PoolA1.  The problem is that if Owings were an AL pitcher, we wouldn’t know how much value to give him as a hitter.

Anyway, that’s how I think about the pools of players.

The interesting ones are like Scott Hatteberg and Jeff Clement and other C/1B, as they are either in the top pool or the bottom pool.

It provides a certain baseline point to make sure that as you compare pool to pool that when you compare Pool B to Pool E, that it still works out.

So, guys who year after year keep bouncing back and forth between C and 1B provide a good group of players to show the relative value between C and 1B as pools.  To the extent that you can accurately measure their fielding at C and 1B anyway.

There are some giant differences between Rally’s Batting category and Fangraphs’ Batting category.  For instance, in 1970, Rally gives Johnny Bench 25 batting runs; Fangraphs gives him 45 batting runs, for a 20-run difference.  Why is there such a large difference?

Thanks! That makes sense to me for pitchers.

Rally: Yep, SB and CS are included in the batting section.

Alex: For Johnny Bench, I’m not sure why there would be such a difference.  For players who steal a lot of bases, that makes sense, but I’m not sure what could be causing it in this case.

Rally, correct me if I’m wrong, but don’t you do something on the team level to get the team baseruns for individual players to add up to the actual number of runs scored?  I haven’t looked, but maybe that team’s non-team-custom linear weights outproduced its actual run scoring by quite a bit?

Could be that, and it could be park factors.

I think Fangraphs is using the wrong runs-to-wins converter for defense.  From Sean’s website (http://lanaheimangelfan.blogspot.com/2009/03/update-on-war-values.html):

“Another correction I made was to have the offensive runs adjusted by a league specific runs to win converter, but the defensive ratings (and position adjustments) to use a standard 10 runs = 1 win. The reason is that the TotalZone ratings use a standard plays to runs value over the last half century. A play saved by Luis Aparicio in the 60’s counts as .75 runs, just like one saved by Omar Vizquel in the 90’s.

This is a conscious decision to better allow comparisons of fielders, in runs saved, across time. Now since little Louie played in a lower run environment, an extra play could be counted at a lower value than one in the high offense 1990’s. But then, the win value of that run would be greater. Maybe Brooks Robinson didn’t save 268 runs, he only saved 230 because runners would have been less likely to score. But his defense was worth about 27 wins.”

Fangraphs appears to simply add in Total Zone and then apply the runs-to-wins converter to the total RAR.  For instance, Joe Tinker has 576 fRAR and 64.6 fWAR, but 490 rRAR and 49.0 rWAR.  Luis Aparicio has 596 fRAR and 64.7 fWAR, but 509 rRAR and 49.8 rWAR.

Yeah, that too.  I’ll have to check what I’m using.  I think that was a particularly strange year where they played half their games in one park and then half in another.

Yeah, I use custom linear weights by team.  So maybe the 1970 Reds scored less than their linear weights.  I also remove pitcher hitting before comparing players to the league average.  Do you take that step on Fangraphs?  If not every player in a non-DH league will have higher batting runs than on my site.

It really shouldn’t make a difference, as long as players are comparable relative to each other.  I don’t care if I have Willie Mays +57 and Duke Snider +45 or Willie +49 and Duke +37.  But then the AL screwed it up by using a DH, so to fairly compare players in different leagues I can’t compare Ripken to only hitters and Dale Murphy to 8/9 hitters and 1/9 pitchers.

Yeah, pitcher hitting is removed from the baselines and the weights in all the linear weight calculations on the site.  I think that was part of the year-by-year wOBA calculations tangotiger posted.

A question for Rally (and others) about the historic (pre-PBP) version of Total Zone.  Total Zone estimates a player’s chances as his outs plus a pro-rated fraction of hits allowed, with the share based on that hitter’s out distribution (and some other factors).  From there, it calculates a success rate, compares that to average, and gives us net plays. 

The problem I have with this is it will hugely understate the value of good fielders (and overvalue poor fielders).  Let’s take a simple example.  Assume that on every 100 BIP, Ozzie turns 2 extra hits into outs.  If SS’s make 20% of the BIP outs on these hitters, TZ will decide Ozzie has a 74% success rate (4 points above average), and credit him with about +.9 plays. So the fielder ends up with less than half the credit for his extra outs. 

The problem stems from using the hit information to estimate chances.  The hits allowed really tell us almost nothing about the opportunities an individual player had.  Most of the benefit of a hit prevented goes to other fielders, with a minimal impact on the rating of the fielder who actually prevented the hit. But trying to incorporate this information, by creating a “success rate,” results in artificially shrinking the variance of TZ.  While Ozzie is really 10% better on the same number of opporunities, TZ estimates him to be just 4% better on slightly more opportunities.

Let me suggest an alternative.  Forget about the hits, and just assign an expected fractional out to fielders for each BIP regardless of the outcome of that play, based on the same factors TZ now uses.  TZ then simply equals actual minus expected outs.  That way, Ozzie gets full credit for his two outs above average.  (I bet this method would be worth at least another 10 wins for Ozzie).  This would basically be a simplified version of WOWY:  We know how many outs a fielder should make given the opposing hitters, base/out situation, etc., so we debit/credit him to the extent he makes more/fewer outs than we expect. 

Thoughts?

Guy

How do you get the +0.9 plays - isn’t it more like 0.4?:

100BIP = 70 outs + 30 hits (where I assume you get your 70% out rate from)

Expected outs for Ozzie is 70*.2= 14

If Ozzie on his own turns 2 hits into outs then expected outs from TZ is 72*.2 = 14.4

I could have it wrong, but I think it works like this.  Say that an average SS makes outs on 70% of balls in his zone, defined as his outs plus his allocated share of hits allowed.  Maybe that’s 14 outs and 6 hits (20% of 30 total hits) per 100 BIP.  If Ozzie makes 2 extra outs, he has 16 outs.  Batters now have 28 hits (assuming all other fielders were average), which the SS gets 20% of the blame for, so SS is credited with 5.6 hits.  His new success rate is 16/(16+5.6) = .74.  So Ozzie’s TZ is (.74 - .7) * 21.6 = +.86.

*

Another benefit of just assigning expected outs based on all BIP, is that each fielder’s rating is completely independent of the performance of his teammates.  In TZ, a player’s rating is affected by the quality of his teammates.  This is a minor problem in the pre-PBP version of TZ, but a very large problem in the PBP years where almost half of the credit/blame for plays spills over to adjoining infielders. 

Using the PBP data may make TZ-PBP more accurate in any given year than just assuming a typical distribution of BIP (or it may not).  But over a career, it’s almost impossible that it would be more accurate, given the systemic errors we know it has to make.  So for career WAR, I think the WOWY-like approach I’m suggesting has to be a better choice.  But I may well be missing some important advantage of TZ....

Using TZ for season and WOWY for career can’t work in a presentation form.  Your season totals won’t add up to your career totals.  I don’t think too many people would be able to accept that.

I tried looking at San Diego, 1980.  Take the number of balls in play for the league and the team.  See what percentage of BIP the shortstop typically fields, then see how many extra plays Ozzie made.  I come up with a +95 for Ozzie.  I can get that down some if I have the groundball tendency and left/right split of the pitchers, all the things pre 1990s TotalZone uses.  But I’m pretty sure I’d still have some crazy outliers, lots of seasons in the +50 or -50 range. 

The time it would take to set something like that up would be a lot more time than I have, though you are welcome to try it, and if you can get better ratings than what TotalZone has right now, go for it.

I’ve tried approaches like that in the past and the result has always been too many unbelievable ratings.  TZ doesn’t seem to work on the micro example of a shortstop making one additional great play (and therefore saving one extra hit), but works better at the season level than anything I’ve seen given what it has to work with.

If the typical shortstop makes 350 plays and has 150 hits go by him, and Ozzie makes 450 plays, TotalZone makes the assumption that there were 100 more balls hit to his position, instead of the assumption that Ozzie only had 50 hits in his zone (assuming team hits allowed were similar, the 1980 Padres had a league average DER).  Then TZ will see think Ozzie had more balls hit to him, but made a higher percentage than average, so he comes out at +30 plays.

I’m a lot more comfortable calling him +30 than +95.  Seems like most other people are too.

I started a related thread here:

http://www.insidethebook.com/ee/index.php/site/article/how_lucky_has_scott_rolen_been_with_his_opportunities_to_field/

Rally, is my description of your runs-to-wins converter in #9 correct?  If I understand it correctly, then Fangraphs is inappropriately adjusting Total Zone by the same runs-to-wins converter as batting runs, even though Total Zone should always be divided by 10 to convert to wins, regardless of the league’s offensive context.  This will result in bias towards good fielders from low offense eras (i.e. Luis Aparicio, Joe Tinker).

Yes, you are correct.  And you’ve found the explanation for it that I had forgotten.  Bravo!

If Fangraphs adjusts their runs-to-wins converter, than career rWAR and fWAR should come out extremely close to the same, correct?  Will that be true for pitchers as well when FGphs gets to that point for pre- 2002 years?

They should be similar, but there are going to still be some differences.

The runs to win conversion for the TZ component of pre-2002 WAR should be updated on the site during tonight’s load.