Touching BasesOctober 07, 2010
Searching for Unusual Pitch Selections
By Jeremy Greenhouse

Michael Lewis and Bill Simmons have written that "baseball is an individual sport masquerading as a team one." Some have reasoned that this is why baseball lends itself to statistical analysis, but I don't think that's the reason. Sure, some individual sports, like tennis, are great for analysis, but with others like boxing, I wouldn't know where to begin. I believe that what sets baseball apart from other team sports is that it can better be classified as a sequential game as opposed to a simultaneous one.

Basketball, hockey, and soccer are good examples of simultaneous games, as concurrent player interaction makes it extremely difficult to isolate any single event from the play as a whole. Football is difficult to categorize, as there are ten minutes of high-octane game action which I'd call simultaneous play, but the rest of the game involves more discreet decision-making. Play calling lends itself beautifully to analytics. As for baseball, most of the game is played in turns. Each defender positions himself, the pitcher chooses a pitch type and location, and the batter decides whether or not to swing. The rest is a matter of execution.

David Gassko wrote an awesome article using game theory to explore the batter pitcher match-up. In his analysis of pitch selection, he used Brad Lidge as his example. Lidge throws a fastball and a slider. Really good ones at that. His task is to mix his pitches in such a way that the batter cannot gain an advantage by anticipating one way or the other. That mix will depend on the batter (it's often convenient to assume that pitchers have perfect information with regards to the batter; they do not.), the park, the umpire, and a bunch of other stuff. I'm going to focus on the count. The count should only matter in determining the rate at which he chooses to throw strikes. Now, there's strong evidence to suggest that baseball players don't act rationally with regards to the count. Dave Allen has shown that batters swing more often 3-2 than they do 2-2. But most pitchers will follow the count in the sense that they throw more fastballs when they need strikes and mix in their harder-to-control off-speed pitches when they can afford balls. Here is Lidge's pitch mix for his career, data courtesy of FanGraphs.


That seems fine to me. I ordered the ball/strike count from from highest run expectancy to lowest, which should theoretically follow with highest fastball percentage to lowest.

A.J. Burnett, like Lidge, mainly sticks to two pitches. He might even adhere more strictly to the count than Lidge. When he falls behind, he refuses to throw a breaking ball. He hasn't thrown a 3-0 curveball since 2008. But when he has two strikes, he relies heavily on it.


And the best example of pitch selection based almost entirely on the count comes from Tim Wakefield.


I wanted to find a few pitchers who defy this trend. "Pitching backwards" is a common way to describe such an approach. I looked at a fair number of pitchers, and while some guys depend less on the count in selecting pitches than others, I didn't think I would find anybody who truly "pitched backwards." I e-mailed Rich Lederer, and he suggested I look into Bronson Arroyo. You should too.


I would guess that something funky's going on here. Arroyo's changeup probably isn't like your normal change. But since 2002, Baseball Info Solutions video scouts have been consistent in calling that pitch -- whatever it is -- a changeup. I don't know what to make of that. Still, how can he throw his curveball 30% of the time on a 2-0 count and 8% on an 0-2 count? Has Arroyo ever given an interview explaining his thought process? Are there any other pitchers at all similar to Arroyo?

The other way that pitchers can defy convention, other than by pitching backwards, is by not following a trend at all. Certain pitchers will only employ a certain pitch in certain counts.

Bobby Jenks has embraced the idea of the "out pitch." He’s a fastball-slider pitcher early in the count. When he gets to three balls, he’ll use the fastball exclusively. But when Jenks gets the count to 0-2 or 1-2, he busts out a curve nearly half the time. He neglects the pitch on other counts, but it’s this huge weapon in these scenarios.


Jenks isn’t alone. Another A.L. Central Closer who embraces his curveball as an out pitch is Joakim Soria. Soria, a four-pitch pitcher, mixes his fastball, slider and change regularly. His curveball, however, he keeps in his pocket until he gets to two strikes at which point it enters the hitters mind.


If you can think of any pitcher whose pitch selection puzzles you, please let me hear them.


Fantastic idea to order the counts by run expectancy in the graph, Jeremy. It seems so obvious now that I've seen you do it.

Thanks, Mike. Hope you don't mind me stealing your bar chart concept.

Not at all, Jeremy. I love what you did with it.

Great graphs! And I have to agree with Mike -- the way you've structured the bars (ranked by run expectancy) really brings home the point that there are significant differences in how pitchers approach each count.

Great job, Jeremy. Is there anything unusual about Brett Myers, Chris Narveson, Ted Lilly, or Barry Zito?

Thanks, The guys you mention all throw a whole lot of off-speed pitches, but I think given their diverse repertoires, their mix depending on the count probably fits.

A guy who's known for pitching backward or "randomly": Trevor Hoffman?

Can you show James Shields? How much does he go to his fb ahead in the count?

Arroyo's fastball is a pretty mediocre pitch. At 88 mph without great movement, if a guy knows it's coming, it gets crushed. But his breaking stuff can be quite confounding, in part because he throws mixes up speed and arm angle constantly. Because hitters know this they look slider/curve and react fastball. And in turn, when hitters can really expect fastball, he busts out the changeup. It's a massive game of cat and mouse because he never has the option of just gearing it up.

It might be interesting to look at other guys with slow fastballs. How does Moyer work?

I love the graphs and hope you can make them for all pitchers. The ordering by count makes the information much clearer.

That said, I'm a little wary of reading too much into the variation on the right side of the charts. 1-2 and, much more so, 0-2, are traditionally times to "waste" a pitch. That means you get ahead in the count, you throw a big loopy curve, the batter takes and the umpire calls it a ball unless it somehow passes right through the wheel house. It's some baseball kabuki.

But I don't think the value in the charts lies in any conclusion. They're so valuable as a way to process the information that I want to see more. After that, we can figure out what they mean.

Thank you!

Moyer would be the obvious choice for a looksee here. Maybe Matsuzaka?

Thank you for the comments, gentlemen. I do appreciate it.

Hoffman looks normal enough.

Good call on Shields. 30% FB on 0-1 and 1-1 counts is quite low. Throws curve 28% of the time on 0-1 counts, and no more than 20% on any other. Throws only 19% FB on 2-2 and 52% FB on 3-2. Very interesting.

Moyer is a great find, too. Doesn't throw FB 60% of the time in any count besides 3-0. Throws cutter/slider evenly in all counts. Seems like he has a very random pitch selection.

Peter, just because there's a tradition to waste pitches doesn't mean it makes sense.

I'd checked Matsuzaka before writing the article. So many pitches means it's tough to tell how he mixes, but he does follow understandable trends according to the count.

What are you using to get the run expectancy for each at-bat?

Also, what are the gaps at the top of the graphs (for instance, Brad Lidge at 2-0 and 3-1)?

Sal, here is how count-based linear weights are derived:

I pulled whole numbers off FanGraphs, so they don't always add up to 100%.

How about Pedro? He was famous for "throwing changeups in fastball counts, fastballs in curveball counts, curveballs in fastball counts" according to the inimitable McCarver.