THE BOOK--Playing The Percentages In Baseball

Thanks for the link MGL, I’m glad you enjoyed the article.  The “too surprising” part was just me intentionally understating my case.  The location section was the real kicker, and I wanted to build up to that. 

Anyway, I was surprised just how little of a difference I found.  I suppose there could be something going on with deception, or he could have been pitching more to batters weakness in the good starts (that wouldn’t necessarily show up in aggregate location), but in the absence of that, there was practically nothing separating the two groups of starts.

If anyone else want’s to look at the data, here you go.

Good starts

http://spreadsheets.google.com/ccc?key=0AmhtqthzQ8zFdDFnejM3QTRYS0Rkbkp0WHFtVVUyUXc&hl=en

Bad starts

http://spreadsheets.google.com/ccc?key=0AmhtqthzQ8zFdHFtU1lBam1BQ01YNlhuZ2xwc3gyQmc&hl=en

What about if you looked at it the other way… don’t separate stats by the results of the pitches, but by the movement of the pitches themselves.

Find the 10 starts in which his curveball broke the most and the 10 when it broke the least… the 10 starts with his fastest avg fastball and the 10 with his slowest (or actually, include fastball movement too since that’s pretty important).

There may be enough variation in batter guessing and BABIP results to have such a wide range in ER when a pitcher is making the ball move in the same way… but you might find that when the pitcher cannot make the ball move versus when he can, might show you a small but significant difference in ERA outcome (like maybe 3.50 versus 4.50).

It seems like after 90% of losses, you hear the starter say “I had good stuff, but I made (1 or 2 or 3) mistake pitches, and they took advantage.” It actually has become something of a joke, but maybe that is exactly what happened.  Or maybe it was just bad luck on those one or two pitches...it would be hard to needle that out definitively.

My observation: If AJ can get called strikes on his curveball, he will probably succeed.  Some of the difference may actually be the umpiring.  AJ throws that backdoor curve to lefties.  If he’s getting that called a strike, said lefties are in for a long night.  Game 2 of the WS versus game 5.  I didn’t see a whole lot of difference in his stuff.  He was throwing that backdoor curve and either he was throwing it just a bit farther off the plate (than in game 2) or the ump had a different opinion on its strike-worthyness.

/my $.02.

http://www.hardballtimes.com/images/uploads/burnett_good_andda_bad_thumb.jpg

I’d like to see the results based something less influenced by luck and variance than game score. Something like xFIP would be better, and using hit f/x to determine the run value of contact would be the best. And I’d like to see this looked at in the other light. What were his results like on days when he had his best stuff opposed to his worst?

The bottom line is, everybody that has embraced sabermetrics already believes that a pitchers stuff is not going to control how lucky he gets. The question is, how much of an impact does it have on his ability to generate whiffs and grounders while limiting walks?

#4

i was going to say the same thing.  Burnetts stuff was exactly the same.  but the ump wasn’t giving him 6+ inches off the plate in 5 against game 2.

game 2:
http://www.tinyurl.com/yjn5lbq

game 5:
http://www.tinyurl.com/yhw2ccu

i would think with a lot of pitchers.  the only difference between games is how many ‘mistakes’ were made and if they got the corners or not.

Mixed-design ANOVA?  The Burnett example really hammers it home, but what if you took all pitchers with a reasonable min. of starts (say, 30?) and did this for them.  See if there’s a difference overall between “good” and “bad” starts.

This, by the way, could be the most interesting and important idea/study I’ve seen in a long time.

Is it possible to see the pitch/fx for the pitches that were actually hit for damage in the bad starts?  Maybe, like Craig in MN says, those two or three “mistakes” that really do make the difference… or in the good starts, “bad” pitches that were missed or hit to fielders.

More generally, what percentage of good contact (line drives or homers, for example) is Burnett giving up on his good/bad pitches in the good/bad starts?  Where are the pitches located?  That may help tease out the “luck” factor, if we see to what extent balls are just finding gloves in his good starts, while in his bad ones even poorly-hit balls off good pitches are falling in - and where they are when the batters swing at them.

Another factor - in the bad starts, are people laying off pitcher’s pitches to a higher degree, and getting called balls, vs. having them called for strikes?

Nick, I am curious.  In the bad and good starts, what is the average (or median) number of pitches thrown and innings pitched?  Given what you have shown, it would seem silly to take AJ out early in those bad starts.

Since Nick said he sorted his starts based on Game Score, his top 10 averaged 108 pitches, facing 26 batters.

His bottom 10, he faced 27 batters, throwing 101 pitches.

Over all his starts, he threw 105 pitches per start, facing 27 batters.

So, I’d say his manager did a pretty good job of not pulling him early.

A lot of good questions here.  Between the comments here and at THT, I’ll have to write a whole other article on this .  At any rate, when I get back home later today, I’ll check about quality of opposition and umpires, as well as which pitches got hit in his bad starts that didn’t get hit in his good starts.  If I find anything interesting, I’ll publish a short post for THT Live. 

Pizza - That was the goal I had in mind when I first started this; however, I don’t know how I would be able to pull it off.  It took an inordinate amount of time just to look at Burnett (between classifying his pitches, writing queries, etc.) and I don’t know how I would be able to do that for every pitcher with at least 30 starts - although it could make for a nice long term project.  If you, or anyone else, can thing of a way to automate the process, I would love to here it.

"So, I’d say his manager did a pretty good job of not pulling him early.”

Yes, indeed.  I need to start giving managers more credit!

From memory, Burnett had quite a few games where he pitched 5-6 innings of 0 or 1-run ball, and 1 (random) inning where he allowed 4+ runs. Can you check his high-scoring innings vs. normal, or other innings that day?

I looked at Bannister’s good and bad starts in 2008 and came to much the same conclusion as Nick did.  He located his fastball a little more over the middle of the plate, but not nearly enough to account for the huge difference in results.  He also had some sequencing issues that seemed to lead to a few home runs, but that was a subjective observation since we don’t know how to quantify optimal sequencing.

Have you ever thought about expanding that study Mike?  Based on the overwhelming feedback I’ve gotten on the Burnett piece, I plan on doing another case study next week utilizing some of the suggestions I’ve gotten.  Given that I don’t have a good pitch classification algorithm, I have to go game by game to feel confident in my results.  That will inherently limit my ability to do anything on a large scale with this, simply do the ammount of time it takes for each pitcher.

You seem to have an excellent algorithm (I remember you saying it was 99% accurate or something in a THT Live article), so you might be able to take a look at this on a larger scale.