# Is Brian Bannister on to something?

In a recent interview with MLB Trade Rumors, Kansas City Royals pitcher Brian Bannister reveals that he does statistical  studies to help improve his game.  Bannister, in talking about DIPS theory, suggests that one piece of information that is rarely taken into account, when considering statistics such as DIPS, is the issue of the count.  He doesn’t fully develop the argument, but he talks about the fact that on an 0-2 count, the batter doesn’t have the luxury of letting a pitch go by and so he might be forced into a bad decision and a bad swing.  It’s a logical theory.  And thankfully, one that can be tested, and rather easily at that.
It’s known that in general, pitchers batting average against varies by count with general batting average and OBP on a 3-0 count being much better than on an 0-2 count.  I think the reasons there should be fairly obvious.  The question here though is a little different.  Once the ball is hit (and it doesn’t leave the park), does a pitcher have more control over what happens to the ball to an 0-2 pitch rather than a 0-1 pitch?  Or more accurately, are the results at least more consistent on some counts, but not others.
I took my trusty 2003-2006 Retrosheet PBP data base out and selected out all the balls in play and coded them for whether the batter got a hit or made an out.  I calculated each pitcher’s BABIP for each year, broken down by count in which the ball in play happened.  I kept it to those pitchers with at least 200 batters faced total in the year in question.  As per usual when I do DIPS-based analyses like this, I looked at the AR(1) intra-class correlation over those four years to see how stable BABIP was for each count.  The higher the number, the more stable BABIP is at that count from year to year.
0-0 count – .066
0-1 count – .023
0-2 count – .050
1-0 count – .084
1-1 count – .103
1-2 count – .045
2-0 count – .011
2-1 count – .002
2-2 count – .016
3-0 count – .039
3-1 count – .000
3-2 count – .046
First off, those numbers are all still pathetically small.  (There’s also a likely sample size issue, in that we’re looking at sample sizes as small as 20 BIP on some pitcher-year-counts)  Yeah, some are bigger than others, but these are the types of numbers that spawned the original DIPS theory that on balls in play, a pitcher has very little replicable skill in keeping the ball from being a hit, at least on a year-to-year level.  It looks like that conclusion still holds, and it doesn’t seem to matter much, on a year-to-year individual pitcher basis, what the count is.
But, maybe there’s something to Bannister’s theory.  I ran BABIP by count on all balls in play from 2003-2006 for everyone.  The results seemed to support Bannister’s basic premise that the count does make a difference.  Here are the BABIP numbers for all pitchers in the four years in the dataset, in order from highest to lowest.
3-1 .3119
3-0 .3112
3-2 .3066
2-1 .3053
2-0 .3045
1-0 .3027
1-1 .2978
0-0 .2965
2-2 .2932
0-1 .2908
1-2 .2908
0-2 .2856
That’s what you might expect.  The more that the count tilts toward the batter, the higher the BABIP, and the difference between an 0-2 and a 3-0 count is 26 points or so.  So, overall, the average pitcher does benefit from having the count in his favor.  The problem is again, at the individual level, there’s not a lot of stability, so it doesn’t look like it’s something that one pitcher is able to exploit and not another.  At the individual level, the variation around the mean is random.  Rather, the benefit of an 0-2 count is a general benefit that all pitchers get in the aggregate.
Consider this a tutorial in what’s known as multi-level modeling.  In this case, you have individual players who are all members of an overarching group, MLB pitchers.  In this case, most of the effect might descend from being a Major League pitcher rather than being a specific individual.  Now, Bannister says that the best way to control what happens to a batted ball is to try to control the count.  It looks like he’s right.  Pitchers who pitch ahead in the count are more likely to give up balls in play that end up in someone’s glove.
Now, do pitchers generally show some skill in what counts they give up their BABIP?  The answer is actually “sorta kinda yes.”  Again, I went to the 2003-2006 data set and calculated the percentage of balls in play that came during each specific count (minimum 200 total BIP) relative to the total number of BIP.  Then, I ran a series of ICC’s over the four years in the data set.  The ICC’s were generally in the mid .30 range.  Not huge, but not something that can be dismissed out of hand.  There’s more.  The ICC’s for percentage of time getting the BIP off of an 0-2 count was .51 and for a 1-2 count was .41.  The ability to induce a pitcher’s count, and then to get the batter to hit the ball has some decent stability.  And those are the counts with the lowest BABIP.
So, in a two-step process, there is a certain amount of control that a pitcher has over BABIP.  A pitcher has somewhat of an individual ability to control what counts he gets into, especially two-strike counts.  Then, based on that, there’s a league-wide benefit/penalty for working into specific types of count.  It’s not that certain pitchers have a certain ability to leverage a 1-2 count, comparable to other pitchers.  It’s just that some pitchers are better than others at getting to a 1-2 count, and everyone pitches better when the count is in his favor.  So, a pitcher who is good at getting ahead in the count is likely to have a BABIP that’s particularly low, and that’s not a mistake.
I think Brian Bannister is on to something.

About these ads

### 9 Responses to Is Brian Bannister on to something?

1. Anonymous says:

Jacob: I feel the same. It’s not quite “who can throw a breaking ball for a strike”, but who feels confident enough in his breaking ball (not quite the same, because some don’t trust their pitch even though it’s good).
I think Bannister is missing out on cause and effect. Why does the count affect BABIP? I don’t think that it’s because the hitter needs to swing, as much as I think that it’s because pitchers vary their selections. In hitters’ counts they will throw more fastballs/sinkers (which produce worse BABIP) while in pitchers’ counts they will go to their secondary stuff.
The only thing is that I think it has more to do with what they feel they need to do rather than with what they are able to do. Not all pitchers throw their fastball for strikes that much more than their breaking balls, but most will automatically go to their fastballs when they need a strike. This will help hitters generate a higher BABIP. But then take a look at pitchers like Glavine, Litsch, Rogers and Timlin who beat average BABIP easily last season: they all throw their secondary pitches often, much more often than average, despite walking their share (and therefore falling into hitters’ counts relatively often).

2. Ren says:

Jacob: I feel the same. It’s not quite “who can throw a breaking ball for a strike”, but who feels confident enough in his breaking ball (not quite the same, because some don’t trust their pitch even though it’s good).
I think Bannister is missing out on cause and effect. Why does the count affect BABIP? I don’t think that it’s because the hitter needs to swing, as much as I think that it’s because pitchers vary their selections. In hitters’ counts they will throw more fastballs/sinkers (which produce worse BABIP) while in pitchers’ counts they will go to their secondary stuff.
The only thing is that I think it has more to do with what they feel they need to do rather than with what they are able to do. Not all pitchers throw their fastball for strikes that much more than their breaking balls, but most will automatically go to their fastballs when they need a strike. This will help hitters generate a higher BABIP. But then take a look at pitchers like Glavine, Litsch, Rogers and Timlin who beat average BABIP easily last season: they all throw their secondary pitches often, much more often than average, despite walking their share (and therefore falling into hitters’ counts relatively often).

3. tangotiger says:

Granted all that, what happens if you take the MLB BABIP at each count, and apply that to each pitcher’s frequency at those counts?
That is, let’s say that Curt Schilling is the greatest pitcher in terms of getting into a pitcher’s count. His frequency distribution at the 0-2 count might be say 10% if the MLB level is 5% (all numbers for illustration only). Let’s say then that you apply the MLB BABIP level to his 0-2 count frequency.
Do this for all pitchers at all counts. What’s the resulting BABIP for all the pitchers? That is, how much does one side of the equation (controlling the frequency of counts) affect the overall BABIP. I will guess that 1 SD will be less than .005 points.

4. Ken Arneson says:

This is the kind of stuff I’ve been waiting for sabermetrics to get to. Up until now, so much of sabermetrics has been about learning how things work from the GM’s point of view. I’m far more interested in how things work from the player’s point of view.
I heard Rick Peterson say recently that batting average (not sure if he meant BA or BABIP) is much lower down in the zone than up. So you could add pitch location as another dimension to this analysis. Peterson’s point was that you try to get ahead in the count with low pitches, because if the batter does put the ball in play, the average is lower. Then once you’re ahead, you have an extra advantage because of the count, and you can move the ball around elsewhere in the zone.

5. Jacob says:

Cool stuff Pizza Cutter,
I wonder about this part though:
It’s not that certain pitchers have a certain ability to leverage a 1-2 count, comparable to other pitchers. It’s just that some pitchers are better than others at getting to a 1-2 count, and everyone pitches better when the count is in his favor. So, a pitcher who is good at getting ahead in the count is likely to have a BABIP that’s particularly low, and that’s not a mistake.
In a hitter’s count, the pitcher is much more likely to throw a pitch they know that they can get over for a strike. Typically, this gives the hitter an advantage, as the hitter can reliably anticipate a fastball. Right? Ok, so in theory, if you have a pitcher who has 6 pitches, each with different movement that he can throw for a strike with 100% reliability that advantage the hitter has is lost due to a skill that belongs entirely to the pitcher.
I think clearly you can say that some pitchers are better than others at getting to a 1-2 count, but i suspect that in addition to that, some pitchers are better able to leverage the 1-2 count. Also, I think where the greatest difference lies is going to be in the hitter’s counts.
Essentially, the skill is, how much advantage can you take back from the hitter that he’s gained because he knows that you have to through a strike.
I suspect that if you created two buckets of pitchers, ones whose babip by count don’t vary much, and the other group who does, you might find that they represent two classes of pitchers, those masters like maddux and glavine in the first bucket and the unpolished, who have one dominant pitch, and one or two below average pitchers.
It may be as simple a division as who can and does throw a breaking for a strike and who doesn’t/can’t.

6. Pizza Cutter says:

Tom, I just ran the analysis you suggest. (2003-2006, min 200 BIP) All expected BABIP’s were between a range of .295 and .300. Standard deviation (N = 929) was .00068.
Further I ran a correlation between actual BABIP and projected BABIP, done in this manner. The result was a measly r = .065.
The effect size is probably pretty small, but it’s there.

7. RollingWave says:

Brian Bannister is now suddenly the most popular ” not exactly the most awsome pitcher ever” pitcher on the net :P