Our friend Zach did an analysis of Clayton Richard’s 2010 success over at Gaslamp Ball. He says:

In 2009, Richard threw his 4-seamer 51.3% of the time, and the 2-seamer only 19% of the time. In 2010, he’s dropped his usage of the 4-seamer, throwing it only 23.8% of the time, and upped the use of the 2-seamer, now throwing it 35.6% of the time.
Richard has also increased the velocity of his 2-seamer. Where in 2009 he averaged 87.6 mph, this season he’s throwing it at an average of 91.2 mph.

Indeed, if you look at the PITCHf/x data, located on FanGraphs, that’s what comes out. Check out his usage, in 2009 vs. 2010:

Year Four-seamer Two-seamer Cutter Slider Change Curve
2009 51% 19% 2% 15% 11% 2%
2010 22% 37% 13% 4% 18% 5%

It struck me a bit that he’d change his repertoire so much in one-offseason. What’s also striking is the speed of his two-seamer; it has jumped almost four miles per hour in one season, while his four-seamer – and basically the rest of his pitches – stayed at the same mph.

As I thought about this, it started to click that perhaps Richard’s pitch selection hasn’t changed that much, maybe the PITCHf/x algorithm has. I wanted to look at the raw PITCHf/x data to see if I could find anything.

The first graphs I’m going to show has horizontal movement on the x-axis and vertical movement on the y-axis, and the pitches are a different color based on their pitch type (as classified by MLBAM). Remember, the view is from the catcher’s perspective:

Richard 2010 new

Richard 2009

You can see the two and four-seamers clustered in the top right of the graph. In 2010 you can see the two-seamers have slightly more horizontal movement and slightly less vertical movement. What is revealing is if we remove two-seamers to get a better view.

Richard 2010, no two seamers1
Richard 2009, no two seamers1

Notice the range of the four-seamer, the circled purple triangles, is much wider in 2009. It looks like in 2009, the algorithm was very reluctant in calling a pitch a two-seamer. Most everything in that upper right hand corner was identified as a four-seamer. In 2010, however, it is much more equal – in fact, as you can see by the graph and the chart above, the two-seamer is now being identified more often than the four-seamer.

If we take the 2010 version of the pitch algorithm as the more accurate one, then it appears what happened in 2009 is that a lot of two-seamers were actually classified as four-seamers. So, in reality, Richard’s pitch selection has probably not changed that much – the pitch algorithm has.

And, indeed, it appears that is what has happened. In an email exchange – and here on the Hardball Times – Mike Fast explained to me that Ross Paul, the guy doing some of the behind the scenes work at MLBAM, is now using a neural networks to classify pitches. A neural net is specifically trained to identify pitches from each pitcher, and should theoretically be a more accurate way to classify a pitcher’s arsenal.


I also created spin graphs for Richard, and while they may not add to the point of this article, I think they are still worth sharing.

Richard, 2010 spin

Richard 2009, spin

Interestingly, it appears that the same thing may be going on with sliders and cutters. It appears that a lot of Richard’s pitches that were being classified as sliders last year are now being classified as cutters.