THE BOOK--Playing The Percentages In Baseball

I also can’t tell if there was one weight for each pitcher’s pair of data, or not.

It’s possible that there is some kind of selection bias where the pitchers who pitch well in their small sample size are the ones who win jobs out of spring training and given an opportunity to succeed. For example its guys like Joba Chamberlain who succeed in small sample sizes and are given an opportunity the next year to prove its not a fluke. Jeff Francoeur is the hitter equivalent of that.

I realize that this is not a rigorous study and I don’t really want to pounce on the author (I don’t know who he is), as well as the fact that I have written for BP, but, geez....

As Tango says, without a control group, we have no idea what is going on. None whatsoever.  And without knowing what portion of the sample went from relieving to starting (which is the whole point of the article, right?), we have even less idea of what’s going on (can there be less than none?).

“Jason is showing 4.19, so it’s still above what we’d expect.”

Tango, it is certainly not above what I would expect.  For one thing, I find that any matched pair of seasons for pitchers ends up with around .2 runs worse in the second year, even after regressing to account for any selective sampling effects. Which is why I always say that pitchers get worse with age. For another, as you also say, many of these pitchers went from relievers to starters. We expect a gigantic jump in FIP (and ERA) with that change in role, to the tune of .5 to 1.0. So, 4.19 is not larger than we would expect if the true talent in year 1 was 4.00.  Probably smaller, given aging and change in roles.

The author also talks about a steep aging curve for young pitchers.  I have yet to see evidence of ANY aging curve, other than down, for any group of pitchers at at age. Does he have a reference for his claim?  And he tells us that there is a difference between the under 25 and the under 23 numbers? I see virtually (not exactly) identical numbers. Am I looking at two different charts than he is?

I forgot this:  The author also says he used minor league numbers for year 1. Argh! I don’t know if he used translations for those numbers, but surely even if you do, you don’t want to confound this kind of study with minor league numbers!

I went back to my post on pitcher aging curves, which is a good one, BTW.  Here was the money comment:

“All pitchers from around age 21 to 26 are completely flat in runs allowed (linear weights against, or ERC).  After that, they simply skyrocket in a nice smooth pattern to the tune of around .2 runs per 9 innings per year.”

I definitely don’t have that kind of aging as mgl.  It’s more like .1 runs per 9IP.

As for my comments not being overly harsh: I just tried to point out a couple of things that the author missed to make the point.  I didn’t think I had to hammer it more than I had to.  Sometimes I’m a nice critic!

Sometimes u gotta let someone else hammer an author to preserve your image. I don’t mind being the bad cop.

I’m not trying to preserve my image.

I’m just trying to be tough but fair.

I also don’t want to be too hard on those who put in alot of work, and Jason here obviously put in alot of work.  It’s something I’ve been wanting to do, and, it’s not as good as it should be, but it’s the best game in town right now.  I’d rather have the article than not have it.  It gives us data, and a discussion point.

So, I’m being tough by saying that there’s holes, and more work needs to be done.  I’m being fair by pointing out some redeeming points in the piece.

Hopefully, Jason takes this as constructive criticism and not a personal attack.

No problem. I think I was being fair too. I’m just not the coddling type when it comes to adults that I don’t know.

“All pitchers from around age 21 to 26 are completely flat in runs allowed (linear weights against, or ERC).  After that, they simply skyrocket in a nice smooth pattern to the tune of around .2 runs per 9 innings per year.”

I’m calling a brain cramp on myself, but I’m missing something. A quick look at this implies (to me) that pitchers are as good as they’re going to get when they enter the league, as they stay flat through 26 then get worse.

What am I missing here?

Dana: you aren’t missing anything.  That’s exactly what MGL is implying.

I disagree.

There’s a selection bias issue to account for, which is why we disagree.

I make no claim to have anything like the sabremetric chops either of you has. I know enough to be dangerous (which I think is more than about 80% of baseball fans), and I enjoy the insights the numbers often bring.

No disrespect intended, when the numbers show something so out of line with experience and observation--that pitchers don’t get better--don’t we have to question those numbers, or the methods that got us there?

I don’t mean to be argumentative. I’m just trying to learn.

Thanks.

(My word verification is, appropriately enough, ‘students25’

Well, I’m disagreeing with MGL, so, really, your qualms are with him, not me.

Thanks for the feedback, all. I was planning a follow-up piece to show how a workload increase of 75 innings worth of pitches (~1220 pitches) affects the data.

Which control group do you feel would be most helpful to this angle of the update?

Jason your control group could be any group of pitchers that don’t have an increase in workload and who have very similar numbers in whatever category you are measuring. 

I also would not use minor league numbers. And you really cannot mix starters and relievers unless your question is about relievers who increase their workloads by becoming starters. Of course if you do that you would have no idea how much of any effect you find is due to workload increases and how much is due to change in role. In fact we know that going from reliever to starter adds anywhere from .5 to 1 run per 9 do it would be fruitless to try and find some small workload effect. 

If you don’t have enough of a sample of pitchers who increased workloads and changed roles you are pretty much screwed.

I didn’t focus on a change of roles as much as a change in workload.

Thanks for the direction—I’ll have to comb the data to see how I can extract roles & stat blend to see what kind of sample size I am left with.

"I didn’t focus on a change of roles as much as a change in workload.”

Right.  You have to separate the two things otherwise you won’t be able to find out anything about the effects of workload.

What if your increased workload sample pitched .39 runs per 9 worse in year 2 but half (or any significant number ) went from relieving to starting?  What would that tell us about the effects of increased workload on pitchers?  Nothing, since we expect the change in roles alone to add a run or so to a pitcher’s RA9. 

And like I said, you need a control group to tease out the effects of aging, regression, and anything else extraneous to workload that could affect performance.

For example, if your sample group had an increase in FIP of .2 runs and so did your control group, then you can likely eliminate workload increase as an explanation for the rise in FIP.

"No disrespect intended, when the numbers show something so out of line with experience and observation--that pitchers don’t get better--don’t we have to question those numbers, or the methods that got us there?”

Yes and no. The saying, “Extraordinary claims require extraordinary evidence (Carl Sagan),” has lots of caveats. One, what may be an extraordinary claim to one person may not be to another. Some people think that the notion of evolution or that the Earth is millions of years old is, “Is so out of line with experience and observation...” For example, the idea that pitchers, on the average, have little or no upward aging curve is not surprising at all to me. Two, who is doing the claiming? That makes a huge difference. Is it me, or is it Buster Olney or Ozzzie Guillen (no disrespect to them)?

More importantly, there are two (or more) reasons why my claim or Tango’s claim (and I’m not sure what that is) should be taken with a large grain of salt: One, there are so many sampling and selection bias issues that it is virtually impossible to figure out aging curves unless they are large and obvious (in the data, not in someone’s mind), and two, the statement, for example, “Pitchers see a relatively flat performance curve until age 26 and then a decline after that,” is not even close to being an unambiguous statement.

Are we including injuries and attrition in our “average pitcher” statement? If we are including attrition, how do we handle it? Are we talking about each pitcher weighted equally, regardless of how many innings he has pitched at one age or another? So on and so forth.

It is not easy to figure out pitcher aging curves and what two people figure out may be answering two different questions. That is why you can have two people like Tango and I, pre-eminent saberists, come to two different conclusions.  Because neither one of us is that certain of our conclusions (at least I am not, and I don’t think Tango should be either), and because we are using different methodologies that are actually answering different questions.

This is what everyone needs to get: When it comes to aging studies, it is impossible to be able to separate true aging trajectories from regression and selection bias so take any “aging” conclusions with a large grain of salt. Also, if you hear a conclusion, like the one from Bradbury a year or two ago, be sure that the source explains exactly what question he is asking which usually entails explaining exactly what samples of players they are dealing with and what conditions are present or implied…

Thanks to both Tom and to MGL. This is why I like to read and comment here. I’m learning a lot from people more knowledgeable than myself, and having some fun doing it. Hopefully, I’m not lowering property values by dragging down the level of conversation.

Dana: you are one of the good things about this blog.