THE BOOK--Playing The Percentages In Baseball

Hi, Tango,

Is that directed at me?  I said something on this topic here:

http://sabermetricresearch.blogspot.com/2006/09/study-protection-and-clutch-pitching.html

Can you give me a hint to where in The Book to look?

Oops, forgot the link.  Entry has been updated.

***

As for THE BOOK, Andy has a section in the Clutch chapter, at Tables 43, 44.  I don’t have my book handy.  Is Glavine in that list?  Who’s the #1 guy?

pp. 108-110.  No tables, just the results.  Only relievers are studied.  Clutch pitching is no more than .003 wOBA, and probably zero.

Tango, you want to remove the : from the link and replace it with / .

Thanks.  I was thinking about the “pitching from the stretch or not” tables.

Thanks, fixed the link.

And, you have the book at your office?  Uh, maybe I should do that too.

I don’t mind this article actually, if for nothing else that it brings up some interesting points and mentions some nice references (James, Silver, etc.)

First, let me say that while James is a great thinker and generally a great articulator, that quote is not one of his better ones.  I DON’T think that is why clutch pitching does not get the same attention as clutch hitting.  Not even close.  But that doesn’t matter.

Now, the concept of clutch pitching can be several things.  One, as Tango mentions, if we try and measure it by using such things as performance with men on base or men in scoring position, we are necessarily going to be conflating “clutch” or smart pitching and ability to pitch from the windup versus the stretch.

If we define clutch pitching by inning/score and always measure it and the control at the start of an inning, then we are going to prety much get rid of the stretch/windup bias or problem.

No big deal either way though.  It would be nice to separate the two, but I could live with leaving them together.

Finally (in terms of what we mean by “clutch") I think we are talking about two things.  One or both (or neither) may be a skill.  One is “smart” pitching, which is the ability to throw in such a way that you optimize batter results.  That means walking more batters with runners on base and especially with bases open, throwing more strikes to the leadoff batter, not allowing home runs with runners on base, etc.

Now, this actually applies in two ways.  One way is in general.  Since the value of a result (s,d,t,hr,bb,k) changes depending upon the out/runner, pitchers should certainly change their approach depending on the outs/runners.  It seems reasonable that some pitchers do this better than others for two reasons:  One, because they are smarter than the average pitcher, and two, because their stuff/skills lend themselves to being able to change their approach more than the average pitcher.  Two, the ability to do this based on not only the outs/runner, but the score/inning as well.  IOW, to optimize the win values of the events rather than the run values.  The most obvious example is the last inning of a game when the tying or winning run is on deck or later, how do you pitch?  Clearly you throw lots of strikes since the win value of a walk is roughly the same as the win value of a home run, especially with 2 outs.  Again, some pitchers are are likely more willing or able than others to change their approach in this and other situations.

The other way that “clutch” pitching can be manisfested, measured, thought of, etc., is simply the ability to pitch “better” as the LI goes up (or the ability or desire or whatever to pitch worse I guess when the LI is low).  For some pitchers, maybe for all pitchers to some extent, they may have to do this.  Just like we think that the reason pitchers pitch better when they relieve is that they can throw harder when they reliever and must pace themselves when they start, it would seem to make sense that pitchers in general would/should conserve themselves in loe LI situations and take it up a notch in high LI situations.  It would also seem to make sense that some pitchers can do this more than other pitchers, for whatever reasons.

O.K. given all that, here is some important stuff about trying to identify and then quantify among players a “skill.” This is what is missing in most of these articles and in many of these discussions even in some analyses by otherwise smart guys like Silver.

(To be continued in next post due to space limitations per post.)

...continued from last post

Just showing a best of/worst of list even with a nice methodology (i.e., you like the split) no matter how striking that list looks, tells us NOTHING in an of itself about: one, whether that particular “split” is a skill, and two, what the true talent magnitude of that skill is (if it is a skill in the first place) for each player.

We can create any kind of list we want for splits:  Who are the best hitters/pitvhers on odd/even days, first half/second half of the season, day/night games, clutch/non-clutch, home/road, and they will all pretty much look the same.  There will be lots of players at the top and lots of players at the bottom, and I suppose to the “untrained” eye it will look like the split is a skill, or at the very least, it will be left to the reader to determine him/herself whether he thinks it is a skill or not, based on his/her own common sense ("Sure, odd/even days is just a fluke, but surely day/night means something, right?").

SO what does it mean if these lists mean nothing in and of themselves?  How can we determine if any of these splits are “skills” and how much of that skill each player has?  I can’t explain the methodology or even the logic here, but anyone who does not know can read Andy’s excellent (albeit complicated and not easy to understand) in The Book.

FIRST you have to determine whether something is a skill and THEN we can start to create those lists.  And because something SEEMS like it must be a skill does not make it so!  On top of everything else, and to make matters more complicated, when we talk about something being a skill (like clutch hitting or a pitcher’s BABIP) we are FUNCTIONALLY talking about there being some spread of skill in the population of major league baseball players. That, in and of itself is a critical distinction in all of these discussions, and again, I don’t have the space or time to get any more into that here.

Now, once it is determined that a certain skill exists, bascially be looking at those lists and seeing (mathematically) whether the “spread” in performance in those lists is similar to what we would expect by chance (if it is, then there is NO SKILL no matter how much variation we see in the list - and of course how much variation we see will be a function of hiow much skill there is PLUS the sample sizes of the individual performances), then we must determine HOW MUCH skill there is (again, the major league population if that is what we are dealing with)!

Just knowing that “there is skill” tells us next to nothing.  Let’s say that we create a list like Nate did, and it is very compelling.  Now let’s say that Andy determines (through statistical tests) that there really is no skill.  End of story.  He has just told us and we have to believe him, that the spread in those lists is likely entirely due to chance (of course, we never know FOR SURE, 100%).  But what if he says, “Oops, I made a mistake.  There IS some skill.” Now, all of a sudden we say, “Well, now that’s more like it!  Look at how skillful Glavine is!  And how non-skillful (or skillful in the other direction) Ryuan was!”

Not so fast!  We need to know HOW MUCH SKILL Andy found (the variance in the population of whatever split we are measuring)!  If it is just a tiny bit, then MOST of what we are seeing in the list is by chance and we can only make tiny inferences about how much skill Glavine, Ryan and all the others REALLY have.

This is another critical point.  First we have to determine if there is any skill at all.  If not, these lists are meaningless in terms of actual “true talent” posessed by the players in the list with respect to what we are measuring and meaningless in terms of predictive value also with respect to the “split” we are measuring.

However, if we find SOME skill it is critcial to determine how much.  If it is small, then the lists are still worth next to nothing.  If it is large, then we have to “talk.”

Continued one more time…

...from previous post

What does that mean that we have to “talk”.  It means to make any sense of these lists, we have to employ another statistical teqhcnique, regression towards the mean, and determine for each player, how much their performance deviated from the norm is likely luck and how much is likely skill.  Because there is so much random variation in baseball perormance, be it hitting or pitching or defense, especially in small samples (e.g., one year) even if we find a large (whatever that means, since it is all relative) measure of skill in the population with respect to something like clutch pitching or whatever, it is still LIKELY that the numbers we see in these lists are MOSTLY due to chance and LEASTLY due to skill. Of coruse, the more data these lists are based on (long-career lists are generally much better than one-season ones of course), the more the deviations from the norm among individual players are due to skill than chance (if there is a skill in the first place, which there may not be).

I realize I am mostly preaching to the choir here, but there is so much misunderstanding of these “skill versus luck” discussions, that it really pisses me off.  Somesay I will write a primer on it for the average person or casual analyst, similar to that above.