Happy Thanksgiving! Regularly Scheduled Articles Will Resume Monday, December 1
November 23, 2009
Ahead in the Count
Zack Greinke and FIP
If the internet made crowd noises, the sabermetrics section would have belted out quite a cheer when Zack Greinke, upon winning his 2009 American League Cy Young Award, uttered, "That's how I try to pitch, to keep my FIP as low as possible."
For a few years now, many of us have cheered Brian Bannister, Greinke's fellow Royal, and who also introduced the Cy Young winner to his new favorite statistic. Bannister has shown an ability to understand the statistics that describe the game he plays better than old-fashioned statistics. Bannister has repeatedly discussed his attempts to improve his BABIP, citing his knowledge of BABIP by count as an explanation of his .262 BABIP in 2007, and then reminding reporters that his xFIP was 4.97 in 2008 when he ran into some bad luck, and his ERA was 5.76 in 2008. This is all exciting for us to hear, but the main problem is that Brian Bannister is not really all that good at it. He clearly is better for knowing the intricacies of the game, and he seems like both an intelligent and friendly person. However, sabermetricians (myself included) have had to ignore that Bannister is just a back-of-the-rotation starter. With Zack Greinke, we have a pitcher discussing Defense Independent Pitching Statistics with the press following the announcement of his 2009 Cy Young Award. That's a whole different animal.
However, what has not really been discussed much since he said that is whether what he describes is a good idea. It is smart for general managers to look at FIP or xFIP or QERA, etc., when determining which pitchers are likely to improve or regress in the coming year, but it is quite different to say that an individual making in-game baseball decisions should minimize his FIP.
I am not suggesting that Greinke does this-to the contrary, he reported throwing more fly balls to allow David DeJesus to prevent runs (thanks to his good Zone Rating), thus lowering his ERA but not his xFIP or his QERA, which look at ground-ball and fly-ball rates, rather than home runs like FIP does. He also discussed avoiding walks and seeking strikeouts in his characterization of minimizing FIP, too. However, I thought it would be important to consider the difference between minimizing FIP and minimizing ERA.
Firstly, recall that FIP is equal to (13*HR + 3*(BB+HBP-IBB) - 2*K)/IP, plus a constant that is approximately equal to 3.2. The idea is that this is a good estimator of ERA, and it clearly has shown to be over time. What that means is that, given how pitchers currently attempt to prevent runs, their FIP is a good way to estimate their ERAs. It does not mean that if pitchers intentionally minimized FIP, it would automatically follow the same rule that it approximated ERA well.
If a pitcher is facing Ichiro Suzuki, who has a career BABIP of .357 but only 84 career homers in 6,099 at-bats, it is certainly risky to let him put a ball in play if one's goal is to avoid surrendering runs. Pitching aggressively is likely to lead to a single or double, and pitching around him is likely to either lead to a walk or a weakly hit baseball. If there are runners on second and third base with two out and a weaker hitter on deck, pitching around him might be the smart thing to do to either get him to end the inning or to walk him and take a shot at the next guy. However, pitching aggressively could lead to Ichiro knocking in a couple runs. However, it would not hurt a pitcher's FIP to do that, and it would actually help his ERA if Ichiro hits a ground-ball single (something that he is extremely prone to doing).
Alternatively, if the same pitcher were facing a guy like Adam Dunn, who has a .292 career BABIP but hits 40 home runs every year and only hits grounders 32.9 percent of the time, the pitcher attempting to minimize his FIP would probably consider nibbling a little more. Is that the best way to lower ERA? You might want to risk seeing if you can get him out rather than putting him on since his BABIP is not great, but the strategy will be very different than facing Ichiro if your goal is to minimize FIP.
Similarly, how a pitcher pitches when behind in the count is another example of the difference between pitching to minimize FIP versus pitching to minimize ERA. From 2005 to 2008, BABIP in non-full three-ball counts was .306 according to an article I did earlier this year, which was higher than the overall BABIP of .299. Of course, this reflects the fact that it is slightly risky to give hitters something to hit on 3-0 and 3-1 counts. Because they are not at risk of striking out, hitters can wait for their pitch, and if they get the pitch that they are looking for, they can hit the ball harder. If they guess wrong, they can lay off it. The end result is a higher BABIP. This fact is well known to pitchers who often throw pitches out of the strike zone to see if the hitters are tricked into swinging, walking them if they do not swing. This does not minimize FIP, but given the higher BABIP with 3-0 and 3-1 counts, a pitcher trying to minimize his FIP might not care about giving up a line drive in that situation if it stays in the park.
The question truly comes down to the reason that pitchers "do not control BABIP." When we say that pitchers don't control it, we really mean to say that pitchers don't exhibit any persistent performance in this particular statistic. In other words, pitchers with low BABIPs do not tend to have low BABIPs the following year, any more often than pitchers who had a high BABIP stay high.
The primary reason that I believe this is true is because line-drive rates are not persistent, and line drives generate a far higher BABIP than ground balls and fly balls. Ground balls produce a BABIP around .240, and fly balls (including popups) around .140, but line drives generate BABIPS around .730. However, although ground-ball and fly-ball rates frequently have persist strongly (their year-to-year correlation is often about 0.70), line-drive rate per ball in play has almost no persistence at all. In fact, using the 2005-08 data, I discovered a correlation of 0.00 for pitchers.
Does that mean that they do not control line drives? Not exactly. Consider the following thought experiment. Suppose that I am a pitcher and I decide to repeatedly throw low outside fastballs on the black for every single pitch. Few hitters will be able to club this pitch out of the park, but if I did it consistently enough that hitters knew that old Swartz always throws outside fastballs in every count on every pitch, then hitters would start taking the ball the other way. The result would be a lot of line drives, as hitters prepare to tee off on my predictable pitch. Of course, the second that I surprised a hitter with a cutter inside, they would be caught off-guard-and would be unlikely to hit a line drive if they swung. If I never did that, then my line-drive rate would be very high and my BABIP would persistently be well above .300. Thus, I would be a pitcher who could control my BABIP by making it high.
I presented this concept in more detail in my first-round article for BP Idol, in which I discussed the game theory of pitchers being unpredictable, and thus not exhibiting any persistence in their line-drive rates. The strategy is very similar to a game of rock-paper-scissors. You want to randomize between pitches. Of course, some pitchers have stronger and weaker pitches, and those pitchers that do might play those more often, but this would only be analogous to a situation in which if you used rock to beat scissors, you got two points. You would probably play a rock a little more often in this case. This all highlights why line-drive rates lack persistence. Hitters can have some skill at producing line drives, but pitchers randomize in such a way that they will have line-drive rates of about 19 or 20 percent on average, plus some luck due to facing hitters who guess right more often.
Of course, what pitcher would want to control his BABIP by being predictable enough to have a high line-drive rate and a high BABIP as a result? No normal pitcher trying to minimize his ERA would do this. However, a pitcher attempting to minimize his FIP might repeatedly throw those outside fastballs. They would rarely walk hitters, and occasionally they would strike some hitters out, and few hitters could hit home runs against them. Thus, there may be a very large difference between pitching to minimize your FIP and pitching to minimize your ERA.
The basic hypothesis here is that FIP is only a good predictor of ERA for pitchers who are trying to minimize their ERA. For pitchers who are pitching to minimize their FIP, they would likely have higher ERAs than FIPs on a consistent basis. Once the subjects becomes aware of an experiment, the data is often useless. I should restate that I do not believe Greinke does this. He clearly tries to minimize his ERA, and he is terrific at doing so. He knows the Zone Ratings of his fielders and attempts to capitalize on this by improving his BABIP. His theories about FIP primarily extend to pounding the strike zone early in the count and making hitters miss once he gets two strikes. That is exactly how to minimize your ERA, and your FIP will improve while you do, even if you could pull some tricks to make your FIP a little lower and your ERA a little higher.
The question then becomes if agents realize that more sabermetrically inclined general managers look at FIP before looking at ERA, they could encourage their clients to focus on FIP a little bit. That would certainly be a shame. ERAs would gradually become higher than FIPs. By how much? Well… not that much. After all, pitchers cannot control BABIP all that much.