THE BOOK--Playing The Percentages In Baseball

I am not sure why the 4 WAR is problematic.  It appears that the WAR calculation (using FIP) has done a great job of adjusting for bad luck, which Nolasco had plenty of in the 2009 campaign.  I wonder if other pitchers who also had similar bad luck had large differences between the two measurements?
vr, Xei

Only 11% have voted for the 1 WAR choice, including myself. He stressed production during the piece and when he asked the question. If he had asked “was his performance at the 1 WAR level or the 4 WAR level”, I think I would have voted for 4 WAR.

Obviously depends on your definitions of value, performance, production, etc.

His park adjusted ERC (applying BaseRuns to his park and defense adjusted component stats), according to my numbers, was very good, and his park and defense adjusted FIP was even better (because of a poor BABIP).

If you want to simply talk about how he pitched with no regard for likely fluctuation (luck) and defense, then you can use ERA, and his was bad obviously.  Then again, you can also forget about ERA and use win/loss, which was very good of course. If I am pitching and my team scores 8 runs, I might not care about my ERA, and I might allow 5 runs.  Should that be held against me?  Some might say yes, and some might say no, and they would both be reasonable positions.  Of course, you might have to ask, “Held against for what?”

When you try and answer questions that use vague terms or the question itself is simply not definitive enough, you usually have several different reasonable and defensible answers.  Which is why I typically hate questions like that…

MGL, to your point re: definitions, particularly of production, that was sort of the point of the piece I wrote (that was me on BtB). In other words, I asked the question about Ricky to get to a different question regarding the methods these two sources use for calculating WAR, which I see as a stat that tells me what the pitcher did independent of defense.

Both methods discount defense in different ways, and they’re both reasonable. Component stats don’t count things that defenses do, Rally’s method takes off a prorated defensive run total. Both make sense in my definition of production (how many runs a pitcher saved compared to a certain baseline of your choice, converted to wins).

The part of the definition I don’t have a sure answer about is whether a pitcher should receive any more credit/debit for things like timing/context. In that respect, you are right that I did not have a full definition, but that’s because I wanted to know the opinions of other people’s definitions.

I prefer to use context in evaluating what a pitcher has done, but I don’t do the same for hitters.  The difference, in my opinion, is that a hitter has no control over his context, it depends on what the other hitters in his lineup do.  A pitcher does have control of context to some degree.  If the bases are loaded, it’s because that same pitcher put them there (unless he was a mid-inning reliever).

For projecting, I don’t think there’s much value in looking at context or timing for hitters or pitchers.

Rally #5: Absolutely agree on the projecting count. I do also think that pitchers have control to some degree over their context, but I’m still on the fence on how much credit they should get. I’m guessing, however, that there’s no in-between between full credit or no credit.

It’s important to note that if you base it on FIP, it ignores non-HR hits allowed AND it ignores the sequencing of events.  I presume, without even looking at his stats, that this pitcher must have been horrible with men on base. 

Rally is taking both into account: non-HR hits allowed AND sequencing.

Rally is doing a generic fielding adjustment. At the extreme, a team like the Mariners that is +0.5 runs per game on fielding would have that figure subtracted from a pitcher’s runs above average.  So, if Felix is say +1.5 runs per game better than average, and he pitches in 24 “complete” games, then his unadjusted +36 runs becomes +24 runs.  He gets downward adjusted by 12 runs.

FIP of course ignores fielding (and non-HR hits) altogether.

Ideally, the “best” answer is somewhere between the two (with regards to fielding / non-HR hits).

However, the sequencing thing (performance with men on base) is another matter.  It really depends on the specific question.

It is identical however to how a batter does with men on base.  If Ichiro is lights out with men on base, he gets the extra credit for that, if you are going to give pitchers the credit / debit for sequencing.

Tango #7: I may be missing something, but shouldn’t the fact that pitchers generally have more control over their environments factor into whether or not you can give them some credit/debit for sequencing, a la Rally’s explanation?

Again, I don’t think there’s a way to give them credit for something in between “full sequencing” or “no sequencing,” so the point may be moot, but I think it’s worth asking.

You could give the player half credit for sequencing, use an average of FIP and adjusted ERA or something like that.  You may not think it’s the right way to handle the problem, or not want to go through the extra steps to do it, but I see no reason why you couldn’t.

I am not sure I would give treat the pitcher’s sequencing differently from the hitter’s just because he controls his environment more.  All pitchers will allow baserunners and pitch both with men on and with the bases empty.  The same is true of hitters.  If a pitcher puts on a lot of baserunners, then yeah, that’s his fault (to whatever extent you hold him and not the defense or whatever else responsible for them), but that’s already considered.  The real difference in looking at sequencing, as Tango says, is how each performs in the different situations when they do occur.

When a pitcher gives up a HR with the bases loaded, and you decide to include the sequencing and debit him a full 4 runs for that home run, are you doing so with the intention of debiting him for creating a bases loaded environment, or with the intention of debiting him for pitching poorly (or at least getting a poor result) once he was in that environment?  If it’s the former, I don’t think you need extra consideration.  The pitchers who give up more baserunners and put themselves in more potential run-scoring situations are going to already be rated poorly for that.  If you also want to consider the effect of how he pitched once he’s already in that situation, that’s the same as what you would do with hitters, looking at how he performs differently given the unique situation.  The hitter doesn’t create those situations himself, but that doesn’t change the fact that he will inevitably find himself in high- and low-leverage situations or in runners-on and bases-empty situations, just like the pitcher, and that he will sometimes perform differently in each, either because of some skill or random variation or whatever, just like the pitcher.

The pitcher creates his own situations and is already debited if he has a tendency to create a lot of bad situations.  The hitter doesn’t create his own situations and thus is not credited for creating good situations for himself.  So that difference is already handled.  Now, do you also consider what they do in those situations once you have already considered how they got there?  I don’t see much reason to answer differently for pitchers just because you already dealt with how they got there differently.

Suppose a pitcher gives up 3 infield singles.  Then, he gives up a HR.

According to FIP, the responsibility for BIP goes all to the fielders.

How then can you hold the pitcher responsible for 4 runs for being on the mound when three runners were put on base because of his fielders?

I have park adjusted batting stats off each pitcher, expressed in wOBA. I then developed a formula for converting wOBA to expected ERA, which I will probably develop into an article in the near future.

I wanted to convert wOBA to an ER/PA rate stat. For the league as a whole in 2008, ER/PA = .334 * wOBA. Next question, did that hold for all levels of wOBA? - No. I queried all pitchers with 30+ IP in the majors from 1998-2008, calculated non-park adjusted wOBA, grouped by wOBA rounded to nearest .010 (.315 to .325 become .320) and summed the IP and ER for each group of pitchers.

The ratio between wOBA and ER/PA was not the sme at all levels of wOBA. I got a best fit linear regression formula of
ER/PA = ((wOBA-Lg)*.875+.343)*wOBA
The mean error for each group with decent size ws less than 0.04 of ERA, but for individuals there is of course random error, up to 1.30 for pitchers with 100+ IP.

With wOBA in the equation twice, it becomes a function of the square of wOBA. As Rally pointed out, the pitcher is responsible for his context of runners on base. A Lower wOBA results in fewer runners on base, fewer BF/IP, and therefor a lower ERA than just a scalar multiple of wOBA would show.

For Nolasco:
Year wOBA wERA
2006 .345 4.81
2007 .349 5.11
2008 .297 3.39
2009 .311 3.83

2009 was a very solid, above average season when measured by hits and walks, but not so by total actual earned runs. Of course I would say that the 3.83 wERA is more predictive of future performance.

ER/PA = ((wOBA-Lg)*.875+.343)*wOBA

Part of this of course makes perfect sense.  The relationship between wOBA and runs per PA is to take wOBA, subtract the league average, and divided by 1.15.  If you divide by 1.15, that’s like multiplying by 0.87.

Once you do that, you simply add to .12 (which is runs per PA).

Brian is doing .343*wOBA, and when wOBA is .33, that’s like adding .11.

So, I can see how that relationship makes some sense.

I didn’t previously see the 1.15 to 0.87 relationship.

I’m glad my empirical test came up with the same numbers.

For hitters, I’m sure that (wOBA-Lg)/1.15 is fine. That’s what I use for BRAA.

The difference here for pitchers is that it there is a squaring of wOBA, and I am confident that it’s the ‘sequencing’ effect that Rally mentioned above. HRs don’t contribute as much to ERA when there are fewer other runners on base. In 1977 John Candelaria led the NL in ERA while also leading in HR allowed.

((wOBA-Lg)/1.15)*wOBA + .343*wOBA

This gives an r2 of .97 betweeb wOBA and ERA when the pitchers were grouped by intervals of .010 in wOBA.

My coefficients were the ones that produced a mean error of virtually zero while minimizing the rms error, based on all MLB pitchers 1998-2008 30+ IP in each season, grouped by .010 of wOBA allowed.