Touching BasesDecember 23, 2009
Aybar vs. Greinke
By Jeremy Greenhouse

Marc Topkin of the St. Petersburg Times on July 18:

Manager Joe Maddon had his reasons for starting Willy Aybar on Saturday.

Some he could explain, such as wanting to keep Aybar fresh for his primary duties as the Rays' top pinch-hitter. And some Maddon couldn't, and wouldn't, derived from extensive research and data analysis by the Rays front office staff that deduced Aybar would be a prime weapon against Royals ace Zack Greinke…

"Free Willy," Maddon said. "This is something we do back at the office, and we really crunch numbers, just so many different things. And Willy came out on top vs. Greinke, so we had to throw him out there."

The research is based on what Maddon called "an esoteric system" and had to be thorough and complex because Aybar had never faced Greinke. And it went beyond the more visual "swing planes" they have discussed before in arranging matchups.

It is also proprietary, Maddon said, joking that revealing it would carry the potential penalty of banishment to semipro ball back in eastern Pennsylvania.

"I would probably end up managing the Japan-Jeddo Stars," he said.

Aybar went 3 for 3 off Greinke.

On December 7, Tommy Rancel of DRaysBay published this exchange he had with Tampa Bay Rays coordinator of baseball operations James Click:

TR: what does Willy Aybar know about Zack Greinke?

JC: Whatever it is, I hope he's told our other hitters.

I’m intrigued.

The Idea

Use pitchf/x data to create a projection system for individual batter/pitcher matchups.

The Qualifications

I have none. The idea is overly ambitious, and I quickly realized I'm not the man for the job.

The Method

Chris Moore rather brilliantly ranked the best fastballs in baseball using five parameters: horizontal location, vertical location, velocity, vertical movement, and horizontal movement. Zack Greinke unsurprisingly came out on top.

Chris only looked at fastballs from right-handed pitchers against right-handed batters. If Chris were to have looked at RHP vs. LHB matchups, I’m sure Greinke would not have come out ahead, and instead Mariano Rivera would have topped the list. But what about RHPs against only Willy Aybar?

So I came up with a way to predict Aybar’s performance given certain pitch tendencies. For example, Aybar does best against slow fastballs around 90 MPH and he likes the ball down the middle. Plots to illustrate these points.

Aybar.jpg


You can't plot all five dimensions together, but the point is that I made a model using all five variables. I then predicted that model onto a data set containing only Greinke pitches. So the model doesn't have any idea how Greinke would pitch Aybar, but it knows that Greinke likes to throw 94 MPH fastballs on the outer part of the plate, and it knows that Aybar likes to hit 88 MPH fastballs down the middle. After some regression to the mean, you have yourself a projection.

The Technical Details

My first data set consisted of all pitches Aybar faced from 2008 to July 18, 2009, and I tried to limit my sample further to only non-sidearming/knuckleballing RHPs. I ran a local regression to predict run values, weighing recent data the most heavily. My second data set contained all pitches from Greinke to LHBs over the same time span. I predicted my model onto that data set. Next, I regressed the expected run values for Aybar against Greinke toward the actual run values of Greinke vs. all LHBs he faced. I then regressed my projection even further to the the average performance of switch-hitting LHBs against RHPs, which I found to be around the league average .330 wOBA.

The Results

I predict Aybar to be precisely league average against Greinke.

My analysis gleans hardly any new insight into player projections. Aybar is below average against RHPs, but Greinke isn’t a world-beater himself against LHBs, having allowed an .824 OPS against LHBs in 2008.

Pretty much, I don’t think you’re going to get enough data from 1,000 pitches from hitters to beat out traditional projection systems. (For pitchers, however, any amount of pitchf/x data adds significant value.) So I guess I'm not on the same track as Friedman, Click, Kalk and the rest of whatever the Rays have going on in baseball ops.

I actually projected Aybar against all RHPs, and for what it's worth, I predict Aybar will do well against Pedro Martinez and poorly against Mariano Rivera. My model tells me Aybar will do surprisingly well against Roy Oswalt and surprisingly poorly against Armando Galarraga. It's not worth much.

The Loosely-Related Tim McCarver Quote

"I said it was Izturis who didn't get the bunt down last year. It was actually Manny Aybar. Excuse me, Erick Aybar, not his younger brother Manny who plays for Tampa Bay."

Comments

Good stuff, Jeremy. Would like to see you continue to pursue this idea and methodology. Loved the bonus quote from McCarver. Thanks.