MISTER FISTER (or... QMAX and HIT TYPE: BEYOND THE SCIENCE OF ONAN)

Mister Fister himself.

I'm here to tell you what you already know, which used to be a guaranteed way for making said practitioner rich and famous, but that's subsequently been outsourced to a distant realm beyond the ozone layer as a result of Internet intervention. (Who knows where all that extra money goes!!)

If you're following baseball at all, you already know that Doug Fister has been pitching extremely well of late (well, not quite so well in that weird, wacky "Game One" in the Yankees-Tigers series that baseball and the Great Media Conglomerate foisted off on us last week--but prior to that, at a level approaching unconsciousness).

Such a seemingly incongruous development is one of those cockle-warming stories that baseball is especially good at providing--transitory transcendence, long odds put through a threshing machine a la the scene in Border Incident where George Murphy is turned into fertilizer, and the strange-but-true relationship between the underdog and the urge for ululation (a topic that Jennifer Dziura, formerly the Princess of Bagels, is about to tackle while wearing nothing but cream cheese in her next one-woman show).

Funny, ain't it, how this "get yourself off" thingee looks
mysteriously like a jock strap...

Mister Fister's ascendance has a been this season's secret stroke of genius, unexpected but filled with potential for explication (though one would be wise to not wear the device depicted at left while attempting to do so). The tools we have at our disposal seem to indicate that the little world of baseball analysis just might be getting closer to taking its statistical data down to the level of an individual game (the point we made fifteen years ago, when we suggested that full acceptance of the precepts involved in advanced analysis would require some such form of "granular integration").

Which, come to think of it, just might be possible with the assistance of said "apparatus."

Of course, no one named Fister is likely to be much more than a flash in the pan, but, by Cracky, we can hope, can't we? Once we have dispensed with (most of) the childishness in the preceding grafs, an examination of various stats that examine the apparatuses (apparati?) that assist us in describing the shape of pitching quality (the ones that go beyond a mere re-casting of earned run average) will show that some potential for breakthroughs in this area are actually on the horizon.

Back in the little so-called "basement world," we have visual tracking aids (PitchFX) that can characterize the details of pitching. (In fact, these are boardroom tools, that point us toward the need for collectivization within the monopoly world of baseball--but that's a diatribe for a different day.) These tools make it possible to measure batter vs. pitcher stats (the BA/OBP/SLG suite of pitcher stats that finally emerged in the 90s as play-by-play data began its initial ascendancy) as a function of individual pitch type. That's why reforming Stalinist stathead Dave Cameron can utilize such data, rendered in somewhat dubious fashion as "Runs Saved by Pitch Type," as part of a Fangraphs post last month that suggests we do not underestimate the improvements that Fister had been making since being traded to Detroit at the end of July.

Hand in hand with this, we have hit type data, another breakout from an observational overlay on play-by-play data, that permits us to make a crude but promising set of categorizations. The key one, as Bill James and others have noted, seems to be the line drive, when, once we isolate it, seems to be an important trend line for pitcher effectiveness. (Actually, the data as it's being applied now doesn't quite rise to the level where it supports the previous assertion, but that's because it isn't currently being applied at the level of an individual game...it's being aggregated as part of the mostly misguided attempt to leach out "luck" via the BABIP/DIPS model.)

These two tools permit some more educated soothsaying, at least. Cameron was able to note that Doug Fister was mixing his pitches more effectively once he'd left Seattle for Detroit, and had improved markedly in using his change-up as an out pitch.

But what remains lacking in this is a framework that describes the quality in each start from a standpoint of hit prevention. Our original tool for this purpose, the Quality Matrix (QMAX for short), got lost in the stampede to the BABIP/DIPS model. It needed only some refinement in terms of overlaying extra-base hits--and a tie-in to the "hit type" data that has emerged in the past several years.

When we bring QMAX into the picture to look at what's been happening with Doug Fister in 2010-11, we can first see that his improvement this season preceded his trade to Detroit. The QMAX grid, focusing on hit prevention/walk prevention, distributes individual starts into the matrix categories, which possess probabilistic quality designations related to team winning percentage. The three basic charts, shown at right, show how Fister was making marked improvement even before he was traded. The fourth "shape chart," below at right, which provides percentage data for the key regions defined on the matrix chart, puts that data into numerical form.

The shape chart shows that Fister's improvement in 2011 was more dramatic than what any of the other available measures suggested to be the case in late July. While it could not predict that Fister's ability to prevent hits would undergo such a dramatic (and possibly short-term) transformation, it clearly pointed to the fact that Doug's actual quality level was significantly higher than the then-current conceptual consensus. In particular, his success at preventing hits (as represented in the percentage of "top hit prevention games"--the S12 category--and the percentage of "hit hard games"--the HH category) had already been transformed prior to the trade to the Tigers.

What's far more interesting than this, however, is the possibility that hit type data, when merged with QMAX, can give us a much better sense of the integration of performance variables. (The pitch type data has possibilities as well, but it may be limited by the fact that the distribution of pitches thrown may not produce a sufficient quantity of information for each individual pitcher--though it's too soon to be sure of that.) We can map hit type data--particularly the incidence of line drives--to the QMAX "hit prevention" axis. When we do that for Fister, we find out that the percentage of balls in play that are line drives goes up in a linear fashion across the axis.

Actually, the data is more interestingly suggestive than that--though we must temper any rush to judgment by noting that the results described below are likely more individualistic in nature, since Fister  has a pronounced ground ball pattern.

The chart at left shows the traditional pitcher stat results filtered through the QMAX grid--or, more accurately, over key segments of the grid (figure below, at right): the "hit hard" section at the bottom; the "Tommy John" or "TJ" region at the lower left, where finesse pitchers try to strand runners despite giving up more hits than innings pitched; the "top hit prevention" region (top two rows on the chart); and the "elite square" (the four matrix boxes at the upper left corner of the chart, where the very best game performances reside).


What's extremely interesting in Fister's case is that the raw percentage of fly balls allowed (column in pink) in these various breakouts barely changes. It's the percentage of line drives and ground balls that fluctuate (look at the G/L column in orange). In his worst games, Fister gives up a bit more than twice as many line drives as he does in his best games.

Now, as we said, it's far from being this simple. Different pitchers will have varying relationships with respect to the hit types, and they need to be studied with respect to other measures (including the BABIP/DIPS formulations, which remain high-level regression-style modeling approaches, built under the assumption that we can't penetrate into the game level at all). But pitch types and hit types and QMAX may just take us there yet.

There are some other game-level fluctuations that need to be measured on a game-by-game basis. Currently we collect the hit-type and pitch-type data without looking at it in terms of individual games. Forman et fils seems to be of two minds about this data--giving us matchless breakouts of the batter vs. pitcher data for hit type, but at the same time giving us double-counted line drive and fly ball data that problematizes our ability to do any analysis with it.

What could help would be a series of hit type charts that go into the game at the inning by inning level. We would be looking for in-game changes in hit type ratios, and for fluctuations in result vs. the overall average result by hit type.

Some examples of these charts are shown on the right. The first one, which charts one of Fister's fine September performances during the Tigers' stretch run, shows that he induced nine ground ball outs over eight innings (we should number them for you, but the horizontal lines should give you the idea of when the innings change) without giving up a single hit.


He wasn't quite so fortunate with fly balls in this game, though the A's were only 1-for-7 (but the one hit was a homer). With line drives, it appears that Doug might have been a bit lucky: the A's were 2-for-5, but the average BA on line drives is around .700, so he dodged at least one hit, maybe two.

Let's contrast this with a game earlier in the year, when Doug wasn't so much in command of his game. (In fact, this is probably his worst start of 2011.) More line drives than ground balls--we already know that this is bad, and it was this way from the very beginning of the game. The White Sox were 8-for-9 on line drives! They were 5-for-9 on fly balls, with two doubles. We had to add a special column marked with an asterisk (*) to show all of the extra plays that entered into the game due to all the baserunners. (The red numerals, which unfortunately don't show well, indicate a hit where a run is scored.)

Note all that extra shading. The darker shades in each hit type show when the count has one or two strikes in it when the ball is put in play. Black areas indicate a ball put in play on the first pitch. Whole lotta data goin' on here...

Let's look at just one more of these hit type charts...this is the one for the "start" in Game 1 of the ALDS that Doug made on October 1st. This one is even more interesting because it shows the sudden shift from line drive to ground ball to fly ball that occurred over the course of the game.

Fister fights his way out of a jam in the "first" (it's actually the second), then cruises on grounders/strikeouts in the next two innings before he suddenly begins to elevate the ball and the Yankees start to hit him. In the final inning, things fall apart due to control issues, some bad luck (two straight ground ball hits) and, of course, the grand slam that was surrendered by Al Alburquerque right after Doug was removed from the game. The Yankees were able to do some uncharacteristic two-strike hitting (we added that breakout to show how much data can be assembled this way), going 3-for-6.

Everyone has observed this type of trend in a game without actually quantifying it before--a pitcher is effective doing one thing, then suddenly is unable to do that thing anymore, trouble comes in a cluster, and even a return to the type of pitching that was effective earlier doesn't work. As you can see in that final inning, Fister had two outs when the ground ball approach came back to bite him: such fine lines are hard to draw with mere numbers.

Mister Fister and the fist of Joe Louis, his reward from the
fans in Detroit for going 8-1 down the stretch.

The line drive-to-ground ball ratio of this game looks pretty good, but Doug gave up one more hit than he should have in each of the three hit types. We can start to quantify "luck" at the game level, instead of at the BABIP/DIPS "model' level.

So what will happen with Doug tonight? Your guess is as good as mine. But what we know is that he has improved a great deal in his second year, and he has a chance to be a very good pitcher for some time to come. I suspect even a smattering of Yankee fans--at least the male ones, at any rate--will find it hard to root against a man named Fister.