BABIP Bounce Back

by FreeZorilla on Jun 5, 2009 8:00 AM EDT Comment 48 comments

Earlier this week, I covered Andy Sonnanstine and how he was due for a pretty massive BABIP regression. While that was not to excuse him from some other pitching issues, the .371 BABIP certainly created some exaggerated results. We saw some regression in Tuesday's game. Today, we will look at the Rays pitching staff's BABIP from 2008 versus their expected BABIP based on the league BABIP per batted ball type. We will take a look at the same thing for 2009 and try to forecast where we should be seeing some corrections. Below is the table for the American league BABIPs per batted ball type in 2008 and 2009:

AL BABIP	LD%	GB%	FB%
2008	0.720	0.242	0.139
2009	0.739	0.239	0.139

So we will utilize each pitcher's batted ball data as weights for each batted ball league BABIP. For example below is Andy Sonnanstine's 2008 batted ball data:

LD%	GB%	FB%
17.00%	42.10%	40.90%

So the calculation would be: (.170*.720)+(.421*.242)+(.409*.139)=.281 Ex BABIP

We can compare this to his true 2008 BABIP of .312 to see that Sonnanstine had approximately .031 of bad BABIP luck. We should expect most Rays pitchers' BABIPs to be a bit below their expected BABIP as there is no adjustment made for their outstanding defense.

Onward with the 2008 table of BABIP vs. Expected BABIP

Name	BABIP	ExBABIP	BABIP-ExBABIP
Dan Wheeler	0.202	0.270	-0.068
Grant Balfour	0.233	0.280	-0.047
Troy Percival	0.181	0.223	-0.042
David Price	0.225	0.263	-0.038
J.P. Howell	0.259	0.291	-0.032
Scott Kazmir	0.275	0.289	-0.014
Matt Garza	0.278	0.290	-0.012
James Shields	0.292	0.281	0.011
Andy Sonnanstine	0.312	0.281	0.031
Jeff Niemann	0.325	0.266	0.059

Last year the Rays were silly fortunate when it came to BABIP. Only 3 returning pitchers Shields, Sonny, and Niemann under performed their expected BABIPs. It was truly a blessed year for the bullpen which led to some unrealistic 2009 expectations.

Now let's take a look at 2009 year to date:

Name	BABIP	EX BABIP	BABIP - ExBABIP
Lance Cormier	0.243	0.279	-0.036
Matt Garza	0.245	0.274	-0.029
Dan Wheeler	0.24	0.263	-0.023
James Shields	0.306	0.313	-0.007
J.P. Howell	0.31	0.305	0.005
Troy Percival	0.306	0.301	0.005
Jeff Niemann	0.279	0.274	0.006
Grant Balfour	0.314	0.308	0.006
Brian Shouse	0.323	0.310	0.013
Joe Nelson	0.321	0.304	0.017
Andy Sonnanstine	0.354	0.303	0.051
David Price	0.488	0.423	0.065
Scott Kazmir	0.361	0.290	0.072

Sonny and Kaz in particular have been victimized by bad BABIP luck, compounding their existing issues. Price has a 40+% line drive rate mostly due to so few balls actually being put into play at this point. On the other hand this season's heroes Cormier and Garza are due for a bit of BABIP regression. The good news is overall we only have 4 pitchers who have outperformed their expected BABIP. With our solid defense, we only had 3 do that for all of 2008. The Rays should expect to start getting some bounces and breaks. Notice last year the only regular who underperformed his expected BABIP to any real degree was Sonny, and that was .031. We currently have 3 pitchers underperforming by at least .51. Remember with the Rays defense it should be expected for most to outperform their expected BABIP based on league averages.

One last table, this one comparing year over year BABIP - Expected BABIP for each returning pitcher side by side:

Name	2008	2009
Dan Wheeler	-0.068	-0.023
Grant Balfour	-0.047	0.006
Troy Percival	-0.042	0.005
David Price	-0.038	0.065
J.P. Howell	-0.032	0.005
Scott Kazmir	-0.014	0.072
Matt Garza	-0.012	-0.029
James Shields	0.011	-0.007
Andy Sonnanstine	0.031	0.051
Jeff Niemann	0.059	0.006

This is just another exercise to see the danger in letting short-term emotions get the best of you when watching games with the naked eye. I am becoming extremely optimistic about this season based on expected regressions and our new and improved offensive lineup upon Longo, Burrell, and Bartlett's returns. Zobrist, Joyce and Burrell are big upgrades from 2008.

2 recs | Comment 48 comments | Share

Story-email Email Printer Print

Comments

Outstanding thread, just one question

How is the exBABIP for each pitcher determined, especially with a guy like Price who has no track record?

by Raymondo on Jun 5, 2009 8:12 AM EDT reply actions 0 recs

Based on his batted ball%

Obviously he is a victim of very SSS. His 40+% LD ratio is multiplied by the league avg .739 BABIP for LDs. I used a minimum of 200 pitches thrown.

by FreeZorilla on Jun 5, 2009 8:23 AM EDT up reply actions 0 recs

Gotcha, thanks

I’m surprised we didn’t see this done in pre season

by Raymondo on Jun 5, 2009 8:30 AM EDT up reply actions 0 recs

RJ definitely brought up the expected regression of the staff in the preseason

by td32 on Jun 5, 2009 10:47 AM EDT up reply actions 0 recs

The expected BABIP is found via the league average BABIP on batted ball type correct?

So it is certainly possible that a pitcher has a higher BABIP on certain balls in play than normal which would increase his expected BABIP. Of course it could go the other way. Obviously there wouldn’t be much deviation from the mean, but dispersion does exist. Certain pitchers do get “hit” harder than others. We won’t really know for sure until hit f/x arrives, but with our naked eye we can see some how teams get better contact off of certain pitchers rather than other pitchers. So until we can make that adjustment it is all truly a guesstimate. A good one for sure, but still incomplete. What if hitters for some reason just make stronger contact off Sonny than Wheeler? That would force Sonnys eBABIP upwards and Wheelers downwards which could eliminate the “luck” factor. Or heck perhaps hitters don’t make good contact of Sonny and his eBABIP is lower than what we’d expect and hes had even more bad luck.

by matthan on Jun 5, 2009 8:36 AM EDT reply actions 0 recs

Its based on the league BABIP not the average of all league BABIPs

Dispersion does exist, but it is minimal. Why was BABIP created? Largely to recognize this exact type of thing. Under and overvalued pitchers vs traditional metrics and the naked eye. There certainly are weaknesses to using league average in general. For example we know that IFFB% are essentially automatic outs. So if we knew the IFFB% of the league that could be factored out of the league average (the data may be available). There are some adjsutments that could also be made for park and defense. Its been proven that BABIP is pitcher neutral while hitters have a greater degree of influence (speed among others).

by FreeZorilla on Jun 5, 2009 9:08 AM EDT up reply actions 0 recs

There is just a weakness in using league average. It creates errors for every pitcher

As you said the true BABIP on batted ball type for each pitcher is probably close to the average. However when looking at the eBABIP and the actual it isn’t like the disparities are extremely large. So just a minor increase in BABIP on certain batted ball types (say a .005) will boost by a few ticks.

Given the sensitivity of eBABIP to the BABIP of each batted ball type it is extremely critical to get the correct figures in there to reduce your errors. A BABIP-eBABIP of .03 could diminish to .01 very quickly with just a small increase.

by matthan on Jun 5, 2009 9:24 AM EDT up reply actions 0 recs

What figures are incorrect?

by FreeZorilla on Jun 5, 2009 9:29 AM EDT up reply actions 0 recs

For vets you could use personal career BABIP per batted ball type

For guys with one or two years of service its probably safer to use the league avg. The point is starters with a full seasons work would not see the discrepancy that Kaz and Sonny have seen YTD.

by FreeZorilla on Jun 5, 2009 9:31 AM EDT up reply actions 0 recs

Well that leads to my points

1. If we are underestimating Sonnys personal BABIP then we are overestimating “luck”
2. It is illogical to use career BABIP for a pitcher with a severely diminished skillset such as Kazmir. His eBABIP should be far lower in previous years than this year.

I’d imagine the best way to do something like this is via comps or ideally with a hit f/x type data

by matthan on Jun 5, 2009 9:34 AM EDT up reply actions 0 recs

Well we don't have Hit F/X

Until I see a better method demonstrating, Sonny is worthy of .051 and kaz .072 discrepancy, I stand by the underlying point.

Another simple method that has been used for years is an expected BABIP of LD% +.120. That wold put Sonny at .321 and Kaz at .305. Still pointing very much in the direction of bad luck.

by FreeZorilla on Jun 5, 2009 9:39 AM EDT up reply actions 0 recs

Sure you can stick by it. People thought the earth was flat for awhile too

You yourself said that there is a difference between personal eBABIP and league average BABIP. Sure given the current resources it is impossible to figure a personal eBABIP for a player with a very small sample size or a changed skill set. However it still exists and it DOES create an error. if there error was +/- .015 it would change the results of this exercise dramatically depending on which way it was for certain pitchers.

by matthan on Jun 5, 2009 9:44 AM EDT up reply actions 0 recs

So you disagree that kaz and Sonny are due for BABIP regressings (the world is round)?

by FreeZorilla on Jun 5, 2009 9:46 AM EDT up reply actions 0 recs

Without analysis on hit f/x data it is impossible to say with certainty. Of course given their relatively high BABIP I would say it is very likely they are due for a regression. However my point has not been whether certain pitchers are due for a regression. A yes or no answer is insufficient. The degree of regression is what is important. Essentially we have no number that we should expect the BABIP to revert to. For example if Dan Wheelers eBABIP is higher than league average than we should see an even more extreme reversion. However what if his eBABIP is actually less than the average? Then his reversion wouldn’t be nearly as severe.

by matthan on Jun 5, 2009 9:57 AM EDT up reply actions 0 recs

W/out Hit F/X

We still don’t know what is the best #. 3 years of Sonny throwing is still a pretty small sample size. You are most likely correct that his LDbabip and GBbabip should be higher than the league average but we do not know the amount. By the same token , he does have typically have the best defensive team in the league behind him. Either way before this week his BABIP was .371, now it is .354. I would be happy to make a wager that he ends the seasno south of the current figure.

by FreeZorilla on Jun 5, 2009 10:05 AM EDT up reply actions 0 recs

I never once said Sonnys eBABIP is over .354

So why would I wager that it is?

by matthan on Jun 5, 2009 11:44 AM EDT up reply actions 0 recs

But yes virtually no matter what the possible error could be Sonny is a victim of bad luck

by matthan on Jun 5, 2009 9:46 AM EDT up reply actions 0 recs

BTW I'm not saying these numbers are useless. They are very valuable and a good benchmark

They should just be taken with a grain of salt.

by matthan on Jun 5, 2009 9:40 AM EDT up reply actions 0 recs

always with several grains of salt

by FreeZorilla on Jun 5, 2009 9:43 AM EDT up reply actions 0 recs

League average vs the individual pitcher numbers

by matthan on Jun 5, 2009 9:31 AM EDT up reply actions 0 recs

There is no way it has been totally proven to be neutral.

I cannot imagine that if plotted the careers of each pitcher with their BABIP against the league average that the differences would equally dispersed around the mean

by matthan on Jun 5, 2009 9:31 AM EDT up reply actions 0 recs

we always hear people claim

that Sonny is hitable. This seems to support that claim. He is working on his second consecutive year (atleast) of giving up a higher % of hits then an average ML pitcher would be expected to allow when the ball is in play. I think someone had listed his BABIP per type in a previous post. Makes me wonder if Sonny consistently gives up hits on ground balls at a significantly higher rate then league average.

by Kevin Cowley on Jun 5, 2009 9:41 AM EDT reply actions 0 recs

Its a good point

Sonny does seem to get hit harder. I did run the #’s for Sonny using career BABIPs per batted ball type.
GB .269
FB .133
LD .799
The ex BABIP is .326

by FreeZorilla on Jun 5, 2009 9:50 AM EDT up reply actions 0 recs

I'm don't know where to look

to find ML average BABIPs for each type but I’d be curious if it’s any one type of ball that is producing the extra hits. Probably ground balls.

by Kevin Cowley on Jun 5, 2009 9:56 AM EDT up reply actions 0 recs

see below reply fail

by FreeZorilla on Jun 5, 2009 9:59 AM EDT up reply actions 0 recs

This is pretty much exactly my point

If our benchmark was league average then Sonny would have to product a BABIP that is far lower in order to hit that benchmark rather than this expected BABIP that you just produced.

by matthan on Jun 5, 2009 9:59 AM EDT up reply actions 0 recs

Type……..AL08…….AL09…….Sonny career
LD%…….720………..739………..799
GB%…….242…………239……….269
FB%………139………..139……….133

by FreeZorilla on Jun 5, 2009 9:58 AM EDT reply actions 0 recs

Time to go there

Lets look at Edwin just for fun
Type…..09%………Career Avg
LD………16.3%……710
GB………37.7%……281
FB……….46.0%……106

BABIP………………………………………261
LD+120 BABIP………………………..283
ExBABIP using Leage Avg………271
ExBABIP using Career BABIP….270

by FreeZorilla on Jun 5, 2009 10:15 AM EDT reply actions 0 recs

Now Sonny vs Edwin Career BABIP by type

Type……Sonny……Edwin
LD%……799……….710
GB%..….269……….281
FB%……..133………106

Does Edwin really get hit harder on the ground but not in the air? Obviously LD% provides the smallest sample size and therefore the largest margin of error.

by FreeZorilla on Jun 5, 2009 10:19 AM EDT up reply actions 0 recs

Plus a walk puts a guy on 1.000% of the time

Rays Win!

by Sandy Kazmir on Jun 5, 2009 11:23 AM EDT up reply actions 0 recs

My question is about luck itself

isn’t it possible that this spans seasons, not just contained within 1. The Rays were “lucky” last, could just be “unlucky” this.

My point is everyone isn’t going to be at the mean – there will be variance, and the expected variance is important in this calculation. Pitchers should be expected to fall within a range, not at a number. Knowing what the standard deviation is is important, as this can reduce the expectation of regression.

I guess I’m saying the numbers last year didn’t regress – the luck held. It’s possible this season the opposite can happen. The ranges described don’t seem much larger than those in ’08, just more on the negative than positive side. How much regression can be expected is not likely defined by the strict mean.

by nyyfaninlaaland on Jun 5, 2009 11:49 AM EDT reply actions 0 recs

Starters throw typically somewhere above twice as many pitches over a season as relievers

So they should be closer to the mean by the end of the season. I am in the middle of looking at last year’s top 10 worst FIPS amongst AL qualified starters and comparing them vs the following xBABIPs:
LD+120
League Batted Ball BABIPs
Career Batted Ball BABIPs
any other suggested methods?

Follow Me on Twitter @FreeZorilla

by FreeZorilla on Jun 5, 2009 12:02 PM EDT up reply actions 0 recs

BABIPS not FIPS

Follow Me on Twitter @FreeZorilla

by FreeZorilla on Jun 5, 2009 12:03 PM EDT up reply actions 0 recs

I don't know what the extremes tell you

unfortunately it seems a plot of all pitchers (maybe broken into starters and relievers, or better, IP groups) BABIP should give you a view of the Bell curve and point to the range of expected deviation from the mean. Not that I’d want to do such a thing, but maybe there are some tools on FanGraphs that could help?

My point being, variation is to be expected. Determining the “expectable range” is more accurate in determining regression. We too often look at hard and fast numbers for projections. All projections should come with a “landing area”, as variability is what statistics is all about. Smaller the data sample, the larger the range. Thus, for a team like the Rays with a lot of players with relatively short careers, a larger range of performance variability could be expected in the near term.

by nyyfaninlaaland on Jun 5, 2009 12:16 PM EDT up reply actions 0 recs

and variance is even more expected and unpredictable

with a stat like BABIP that has trouble correlating from year to year within a given sample.

by Navi's_Navy on Jun 5, 2009 3:28 PM EDT up reply actions 0 recs

Pardon if I'm wrong

but I remember something of similar vein on hardball times a few years back, comparing all the different forms of BABIPs to see which was the most accurate. I’m not sure if I can find the article.

by Navi's_Navy on Jun 5, 2009 3:29 PM EDT up reply actions 0 recs

Was it the Bendix/Dutton piece?

by R.J. Anderson on Jun 5, 2009 3:31 PM EDT up reply actions 0 recs

I think so

sorry for not being able to give specifics, I remember reading it a few months back, but I think it was written a year or two ago. I should probably work on retaining the helpful information. If it turned on a light in your head though you’re probably on the right track.

by Navi's_Navy on Jun 5, 2009 3:32 PM EDT up reply actions 0 recs

I think Derek Carty wrote a piece a while ago on the different types.

Not sure though.

by R.J. Anderson on Jun 5, 2009 3:34 PM EDT up reply actions 0 recs

Any links to any of these?

Follow Me on Twitter @FreeZorilla

by FreeZorilla on Jun 5, 2009 3:45 PM EDT up reply actions 0 recs

Uh-huh.

http://www.hardballtimes.com/main/fantasy/article/whats-the-best-babip-estimator/
http://www.hardballtimes.com/main/article/batters-and-babip/

by R.J. Anderson on Jun 5, 2009 3:58 PM EDT up reply actions 0 recs

god damnit

good job RJ

by Navi's_Navy on Jun 5, 2009 3:59 PM EDT up reply actions 0 recs

I just Googled "Bendix/Dutton + Derek Carty" to get the first link.

by R.J. Anderson on Jun 5, 2009 4:55 PM EDT up reply actions 0 recs

I was looking

but couldn’t find anything

by Navi's_Navy on Jun 5, 2009 3:59 PM EDT up reply actions 0 recs

Good info.

I was thinking about doing a post with the R/R+1500 method and regress towards team average rather than league average. Since not all defenses are created equal, but just eyeballing the numbers, they seem similar.

by R.J. Anderson on Jun 5, 2009 12:42 PM EDT reply actions 0 recs

Just to build on this

is there any way to effectively and consistantly quantify the role of a defense in relation to BABIP. While in 2008 we were silly lucky in terms of bullpen BABIP, is there any backing behind the statement that expect babips should generally be in the pitchers favor when the defense behind them is proven above average? If so by how much? Will this information be too vague and experimental until hitter f/x comes out (and we can look more specifically into the velocity of balls off the bat, trajectories, etc.)?

Also, the unreliability of BABIP from year to year. I remember seeing something about how may plate appearances it takes for certain advanced (and not so advanced) metrics to stabalize ((I think it was on fangraphs)), but am forgetting where BABIP placed (or if it even placed at all). If it is so volatile why do we insist on working with it (nothing better?)

by Navi's_Navy on Jun 5, 2009 3:25 PM EDT reply actions 0 recs