How much do counts affect BABIP?
In my other article today, I mention a discussion that’s going on in the comments section of a post I made to the CardRunners site. In it, RotoWire’s Chris Liss gets to talking about luck, randomness, BABIP, and how what we often consider to be luck may not be luck at all. His most concrete claim sounded interesting to me, and I wanted to test it:
In fact, Todd Zola sent me BABIP data by count—and BABIP goes up reliably as the count gets more hitter favorably—like .315 on 3-0, and .285 on 0-2. It’s .305 on the first pitch. So let’s say a guy like Haren (or Aaron Harang or Dave Bush) gets a rep as an extreme strike thrower—then batters might swing more often at the first pitcher, rather than take a pitch and get behind. So there, the pitcher’s BABIP would change not bad on luck, for example.
The “BABIP by count” argument
While it’s absolutely true that BABIP differs by count, I didn’t think the overall effects would be very large. To test this, I used the MLB GameDay files to pool all the data since 2008 and came up with average BABIPs by count:
+-------+-------+ | Count | BABIP | +-------+-------+ | 0-0 | 0.311 | | 0-1 | 0.300 | | 0-2 | 0.288 | | 1-0 | 0.311 | | 1-1 | 0.310 | | 1-2 | 0.294 | | 2-0 | 0.322 | | 2-1 | 0.314 | | 2-2 | 0.301 | | 3-0 | 0.336 | | 3-1 | 0.312 | | 3-2 | 0.316 | | All | 0.306 | +-------+-------+
Caveat: There may be some selection bias in calculating league average BABIP by count. That is, perhaps the BABIP for 3-0 counts is too high because poor pitchers reach 3-0 counts more often than good pitchers do. This shouldn’t change our overall impression of the effect, though, because even if there is significant bias, all we’d likely see is the BABIP by count clustered closer to the overall BABIP league average of .306.
From here, I came up with an “expected count-based BABIP” (xcbBABIP —that’s catchy) for everyone who’s thrown a pitch in the PITCHf/x era. To do this, I looked at how many balls in play each pitcher allowed by count and assumed a league average BABIP (for that count) for all of those balls. After adding it all up, here are our 2009 leaders (300 BIP to qualify):
+------+-------------+---------+--------+ | YEAR | LAST | FIRST | xBABIP | +------+-------------+---------+--------+ | 2009 | Harden | Rich | 0.3003 | | 2009 | Hendrickson | Mark | 0.3012 | | 2009 | Lohse | Kyle | 0.3014 | | 2009 | Hanson | Tommy | 0.3015 | | 2009 | Tallet | Brian | 0.3018 | | 2009 | Sabathia | CC | 0.3024 | | 2009 | Nolasco | Ricky | 0.3026 | | 2009 | Weaver | Jered D | 0.3027 | | 2009 | Holland | Derek | 0.3029 | | 2009 | Wellemeyer | Todd | 0.3033 | +------+-------------+---------+--------+
And our trailers:
+------+------------+----------+--------+ | YEAR | LAST | FIRST | xBABIP | +------+------------+----------+--------+ | 2009 | Zambrano | Carlos | 0.3125 | | 2009 | de la Rosa | Jorge A | 0.3114 | | 2009 | Meche | Gil | 0.3112 | | 2009 | Bush | David T | 0.3104 | | 2009 | Stammen | Craig N | 0.3103 | | 2009 | Morton | Charlie | 0.3100 | | 2009 | Carmona | Fausto C | 0.3098 | | 2009 | Suppan | Jeff | 0.3097 | | 2009 | Redding | Tim | 0.3096 | | 2009 | Santana | Ervin R | 0.3094 | +------+------------+----------+--------+
So it appears that that the extent of the “BABIP by count” effect is about 0.006 points of BABIP in either direction, at the extremes, and that’s without any regard for repeatability or regression (and if you’d like a quick idea about that, I found a pissantian 0.01 r-squared for pitchers with at least 400 BIP in adjacent years from 2007-2010).
I don’t think we can rightfully claim that this “BABIP by count” effect is to blame for any truly abnormal-looking BABIPs. The enduring effects of it seem minimal at best and completely insignificant at worst.
Sweet article, i’ve been pondering the affect of counts on some advanced stats for a while now, thanks for the leg work.
Also, where does one find good xBABIP data?
Thanks, Kyle. The kind of xBABIP that I mention here, I don’t think, has much value to us. Because it seems to be unstable, once we’d properly regress it basically everyone would find themselves at league average. For hitters, there is a quick xBABIP calculator here: http://www.hardballtimes.com/main/fantasy/article/simple-xbabip-calculator/
For pitchers, I introduced a sort of xBABIP based on batted ball profile here: http://www.hardballtimes.com/main/fantasy/article/introducing-xw-xbabip-xlob-xhr-fb-and-more/, although I haven’t made all the data available yet. What we essentially see is that extreme fly ball pitchers have xBABIPs around .290 and extreme GB pitchers around .315. The rest fall in between.
Derek,
Can you look at the BABIP by count of Just the first pitch?
I ran the R correlation of “Balls” and “Strikes” independently, and both have R values of 0.67 (strikes being negative).
So the # of Balls in the count is JUST AS IMPORTANT as the number of strikes. Intuitively, I would say that this implies 2 things: Hitters are anticipating a strike, Pitchers are forced to throw a pitch more likely to BE a strike.
Wouldn’t that logic carry over to Pitchers that are known to throw first pitch strikes? Or classic “strike throwers”?
A real interesting analysis would be the BABIP on against the likelihood that the next pitch is a strike. I would hesitate a guess that a 3-0 count is the most likely to follow with a strike, and 0-2 the least likely…
If this truly does exist, “bad luck” pitchers should most of their bad luck in counts where strikes SHOULD have been less likely (0-0, 0-1, 0-2), and there should be almost no effect in “throw strikes anyway” counts…
Thoughts?
Blair,
Wouldn’t “BABIP by count of Just the first pitch” just be the BABIP for 0-0 counts? Or am I not following?
What are your running regressions of balls and strikes on?
I’m not sure I understand what you’re getting at in the second of half of your comment. Could you clarify?
Derek,
1) I was refering to running 0-0 count BABIP for all pitchers, against league average. My suggestion is that deviations here should be the biggest indicator of tipping pitches, or throwing “too many strikes”.
2) Regression analysis is simply (# of Balls in the count):(BABIP). The chart looks like a column of this:
Balls | Strikes | BABIP
0 | 0 | 0.311
0 | 1 | 0.300
0 | 2 | 0.288
1 | 0 | 0.311
etc,
And excel CORREL(a2:a12,c2:c12) for balls.
3) My argument is this. If the phenomena of tipping pitches/predictable pitchers exist (like the Haren discussion), it should exist primarily in counts where strikes are less likely to be thrown anyway.
I mean, we know Nolasco doesn’t like to walk people, but ALL pitchers throw high % strikes in 3-0 counts, so check his numbers in 0-0, 0-1 etc, where most pitchers are more likely to throw junk off the plate. This is where the deviations should occur…
If this exists, we should see Haren’s xBABIP at it’s worst during 0-0, 0-1 and 0-2 counts, as he is negating the Pitcher’s advantage of not needing to throw a strike, by Throwing Strikes.
I should clarify:
By saying “throw strikes”, i refer to pitches that would be called a strike without a swing. Ideally, on an 0-2, a pitcher would throw a Ball but the batter is more likely to swing, thus creating a weaker hit, and a lower BABIP.
This is cool, Derek. I’ve been thinking about the effect count has on BABIP for a while now, and it’s great to see it in thoughtful post.
What about Slg% by count?
Career BABIPs
Haren: .305
Harang: .317
Bush: .293
Nolasco: .312