THE BOOK--Playing The Percentages In Baseball

Hopefully, David will come back with some more work on the subject and not leave us hanging.

***

I’m planning on doing a third article, but what exactly do you want to see that I have yet to do? I have some suggestions from Tango and Guy—do you have anything else?

Nah, not really.

However (if I may put in a request, since you are young and energetic )…

I would LOVE to see some research along the lines of whether young pitchers with high K rates but are bad overall (usually because of high walk rates) have a better chance of becoming decent (or better) pitchers in the future, than equally bad pitchers with lower K rates, even after controlling for the “Bill James bias”.

IOW, do something similar to what you did in these studies, but limit pitchers to those who had bad FIP’s for a min number of PA/IP, restrict it to a young age or little MLB experience (but, as I said, some min number of IP, maybe 150), and then look at each group’s chance of having an average or better subsequent year or years, or career, rather than what their average subsequent careers look like.  IOW, test the hypothesis (CW) that young power pitchers (pitchers with good “stuff") have better “upsides.”

Basically, let’s say that I have two pitchers with equal projections, but both projections are bad (and the projections adjust for the high K, low K bias).  One pitcher has a very high projected K rate, but his overall projection is bad because his BB (and/or HR projection) is bad, and the other has a lower (say two other groups, average K rate, and low K rate) projected K rate, and his overall projection (in FIP or whatever) is equally bad.  The CW is that the first pitcher has a better chance of eventually becoming a decent or better MLB pitcher.  Is that true?  I have wondered for a long time if it is.  Generally teams will keep “bad” (overall bad projection) pitchers with good or great stuff, in hopes that they might someday “harness that stuff” and become better pitchers, whereas they have little use for pitchers with bad overall projections and little else (not great or even good stuff).  And the great or good stuff is usually manifested in a high or at least a higher K rate.  Maybe the answer is just a matter of regressing each pitcher to the appropriate overall number (again, say, FIP), and the “good stuff” pitchers should get regressed to a much better “mean ERA (or FIP, or whatever)” than`the “non-good stuff pitchers.”

IOW, if our projections for both types of pitchers include regressing them both to the same mean overall number, then our projection models are just wrong, and in actuality, the “good stuff” pitcher should have a better projection, even if both pitchers have around the same overall sample performance.  Or maybe it is that “good stuff” pitchers genuinely have a “higher upside” (chance of becoming better pitchers) than a “non-good stuff” pitcher, even after adjusting for the different K rates.

I don’t know, but have long wondered.  Again, to avoid the Bill James mistake when comparing low K to high K pitchers, you have to use FIP, DIPS ERA, or something like that (you can also appropriately regress each component).

Good idea, Mitchel. I’ll look into it.

It may also be interesting to look at the flipside of that question - do pitchers with lower K rates have more “downside” risk? That is to say, if we take two equally good (or average) pitchers, using FIP or xFIP or DIPS, are the ones with lower K rate more likely to fall off?

An observation about your method:  when you take a bunch of pitchers and separate them by K-rate (for example), then compare the best low-K pitchers to the worst hi-K pitchers, you’re not just comparing the impact of Ks “other things being equal”.  What you’re really doing is comparing a group of Hi-K/Hi-BB/Hi-HR pitchers (I assume) to a group of lo-K/lo-BB/lo-HR pitchers, and asking which one does better in the future.  It’s not the same thing, because you have multiple moving parts.

In fact, you can’t hold “all else equal” while also making the pitchers equally talented.  So perhaps this is a place where multiple regression would be useful (tho I share Tango’s general aversion to the method for many sabermetric issues).  Take a large group of pitchers, calculate their post-age-26 RAR (or some value metric combining longevity and talent), which is your dependent variable.  Then look at their pre-26 rates—K, BB, HR, BABIP—to see which best predict future value.  If some variables have stronger (or less) predictive value than their inherent FIP value, that tells you something interesting.  If not, then all we care about is pitchers’ overall true talent.  Would that work?

(Side note: James may not have been aware that he was measuring the superior talent of high-K pitchers as well as their durability and/or growth potential.  But in those pre-DIPS days, it was not widely understood that between two pitchers with similar W-L or ERAs, the higher K pitcher was likely more talented.  So his study still told an important story, even if he somewhat misinterpreted the cause of future success for hi-K pitchers.)

(Side note: James may not have been aware that he was measuring the superior talent of high-K pitchers as well as their durability and/or growth potential.  But in those pre-DIPS days, it was not widely understood that between two pitchers with similar W-L or ERAs, the higher K pitcher was likely more talented.  So his study still told an important story, even if he somewhat misinterpreted the cause of future success for hi-K pitchers.)

That is a very good point.  Although later on, he should have repudiated his original findings, or at least qualified them.

That’s funny.

David is back with a 3rd look today:  http://www.hardballtimes.com/main/article/do-pitch-counts-count/.  Interesting findings.  He shows that hi-K pitchers (pre-age-26) improve their walk rates quite a bit, as I thought might be the case, but this is basically offset by their declining K rate. 

His finding that every extra pitch thrown before age 26 reduces future IP by .5 innings is very important if it holds up.  That means a pitcher essentially loses 8 pitches later for every extra one thrown pre-26!  But I think some more work is needed to be sure this is right.  Some of the stats experts here should comment, but I think controlling for Ks, BBs and FIP-ERA simultaneously may be problematic (getting a positive coefficient for BBs is certainly odd).  It also seems to me that his Pitch variable— once you control for IP, BBs and Ks—may be measuring BABIP to a significant extent.  And while much of that is luck, some of it isn’t, and so the regression may just be telling us that hi-BABIP pitchers have shorter careers.