THE BOOK--Playing The Percentages In Baseball

I am worried about the staggered effect too.  I’m going to be addressing the issue of fatigue over the next few weeks, so I’ll be looking at it more and more (as I type I’m trying to correct that methodological problem).  As always, I’m very open to suggestions/collaboration on this one.

Another one: I wasn’t clear on how you did it, but when you created your baseline pitcher data, did you only select their performance as starters?  You should if you can.

Btw, CWEVENT has alot more fields, that I highly recommend (includes flag for pitcher starting or relieving, as well as time up through the order).  You can see it all in the wiki.

On the first question, yes.  My first step was to eliminate any performance by anyone who wasn’t starting that particular game.  (So swing guys only had their starts considered) On CWEVENT, I’ve not yet used it.  Perhaps another thing to add to my schedule for the day.

CWEvent, is that a program that can be used instead of BEVENT?

I haven’t checked out your database thread but perhaps I should.  I think I did have a work-around for determining starting pitcher, I created a table of starting pitchers by game ID.  When I want to run a start vs relief comparison I just check if every pitcher is on it or not.

Haven’t had much time to do any database work lately though.

I highly suggest you go to the wiki.  CWEVENT has been tested against BEVENT by Ted and Dave (the two authors/trustees of those programs).  In addition, CWEVENT has 50 additional fields.

Furthermore, if you create your DB with the same name as what is on the wiki, then we’ll all be able to share our SQL.

I went in and fixed those numbers.  They now make actual sense.

Well, those numbers are very very smooth.  It’s 2.3 wOBA points per 10 pitches for every single 10 pitch point you have listed!  And, if we multiply that by 33.33/10 (for 33.33 pitches through each time of the order), that gives us 7.7 wOBA points, an almost identical finding to Table 82 of The Book.

However, seeing that The Book showed a more staggering effect per time through the order, I’d be very interested to see your results with the extra parameter included.

Is there are selective sampling issue that the staggering has to do with the fact that as the pitcher completes another “lap” around the lineup, he begins by facing the better hitters (leadoff first, and then on down)?  It’s been a while since I read your book… did you control for batter/pitcher quality?

Yes, definitely controlled for batter/pitcher quality, likely exactly like you.

The staggering I’m thinking is that the 30th pitch might be thrown to the 7th, 8th, 9th, 1st, 2nd hitter.  And so, rather than see the staggering effect between the 9th hitter and the 1st hitter the second time round, the possibility that the 30th pitch can be thrown to possibly 5 different lineup slot can smooth that out.  And the farther out you go, the more smoothing you get, thereby obscuring the staggering effect for each time through the order.

On the other hand, if you introduce the times-through-order parameter, you may find that the first 30 pitches have NO change in peformance, because any decline has been captured by the times-through-order.

(However, seeing that the decline happens IMMEDIATELY, I might be wrong here.)

In any case, the times-through-order is a bias that is expected and has been shown to be real.  So, might as well control for it.

Nice work Pizza, although you definitely should have cited Tango’s work in the Book, which was very similar and yielded almost exactly the same results.

Since Tango uses a non-regression model (which is fine, as you pointed out, as long as you control for the obvious selective sampling issues), citing his work helps to support the results of your work, if nothing else.

Anyway, why are we calling this “pitchers wearing down” when we have no idea how much is due to wearing down and how much is due to the batters getting used to the pitcher.  In The Book chapter, Tango admitted (I think) that without further research, we have no idea of the magnitude of each effect.  It is probably both, but I for one would like to know the relative strengths.

Pizza, I don’t think you emphasized enough that we don’t know how much of this effect if due to “times through the order” and how much is due to pitcher fatigue.

Can you put in that parameter (times through the order), as Tango mentions, and come up with the strength of each effect, using a logit regression?

If yes, I would love to see the results.

Of course, practically speaking, it does not make that much difference to a manager whether it is fatigue or times through the order that is causing the drop in effectiveness, but sometimes it does.  For example, if I am a manage, and the other team brings in a PH, and my pitcher has thrown 100 or 110 pitchers (or 90, or whatever), I want to know if he is going to be less effective because he has thrown a lot of pitches or about the same as early in a game because the batter is seeing him for the first time.  Similarly, if my pitcher has thrown 100 pitches, but only went through the order 2 times, is that better or worse than 90 pitches 4 times through the order?  Inquiring managers might want to know (yeah right!).

Also, taking pitchers out after 100 pitches or so is like when you buy multiple items at the store.  NO ONE buys 3 items or 7.  Everyone buys 2, 4, 8, 10 etc.  No reason other than when we do something somewhat arbitrarily, we like round or even numbers.

The other day, I was an Frye’s (a cool electronic store).  The guy next to me at the checkout counter bought a bunch of those new flourescent, energy-saving, light bulbs, as did I.  The checkout person counted out 9, and the guy was lamenting that he thought he had picked up 10.  I jokingly said to him, “Yeah, no one buys 9 of anything, unless they need exactly nine, and even then, they might by one more just to make it an even 10!” He laughed, knowingly.  Even I had bought 6 bulbs, not that I really needed any particular number.

Anyway, if you want to preserve arms for the future (and no one really knows the correct way to do that), there is nothing wrong with using 100 pitches as a rough guideline.  What else are you going to use?

Plus, it is probably CORRECT to take your starter out as early as possible, especially when he is not that good of a pitcher in the first place, since relievers are better pitchers in general, and in the NL at least, it gives you more opportunities to pinch hit for your pitchers (see the post above about the Braves game today).

So this move in the last 30 or so years towards fewer pitches and innings for starters is probably correct.  Whether it was correct in the old days, I don’t know.  It probably was them too, especially before the DH, when all pitchers hit, but no one thought about it.

BTW, I LIKE the changes in the format of the web site, Tango!

Something that I’ve had in mind for a long - my belief is that when a pitcher tires it could be shown in a higher pct of pitches being balls, and also a higher pct of strikes being put in play. I’d like to run a test to see if this is true and can be measured.

I had a discussion with Will Carroll last year on this, in that I believe fatigue causes breakdown of normal mechanics, which besides poorer control etc also is more likely to lead to injury.

The Mike Marshall approach is that you can learn to not alter your mechanics even when fatigued, and still stay effective and avoid injury.

The Don Sutton school of thought is you build up arm strength by throwing a lot of pitches.

I think that a manager must be able to spot the point where his pitcher has lost the good mechanics do to fatigue, and get him out then, but work towards building strength to increase the number of pitches it takes to get to that point.

Some pitchers, it might be 60 or 70 pitches, others 120 or so. I don’t believe 100 is the number for everyone.

This can be done… the early data say that a tired pitcher actually walks fewer batters, but it’s because more balls get put into play.  Not sure if that’s a control issue or what, but it can be looked at.

Pizza - the two numbers I wanted to look at were pct of pitches that were balls and pct of strikes that were put in play, not walk pct

Pizza does a followup:
http://mvn.com/mlb-stats/2008/06/03/more-on-pitcher-fatigue/