Is his location to lefties different?
This graph shows Mariano’s pitch location to left handed batters in 2010 and 2011, where red is 2011 and blue is 2010. Each contour represents the same proportion of Mariano’s total number of pitches to lefties. The graph is from the catcher’s perspective, and the dotted box indicates the left-handed batter’s strikezone.
According to this graph, his location was pretty similar in both years. Much like the visualization presented in Simon’s article, it seems that his cutters in 2011 are leaking a little more over the middle of the plate. But how much of this is an artifact of the plotting system? Consider the graph below, which uses the same exact data as the graph above, in addition to the same kernel density smoothing method.
Pitch location now looks identical! What’s different? I reduced the number of bins by a factor of 10. Am I implying that the graphs in the article were made incorrectly? No. The point of me showing these two graphs is that you can often reach a different conclusion with one small, arbitrary tweak in your plotting method if the differences are small to begin with.
We must also consider the role of PITCHf/x calibration error. PITCHf/x works by using three cameras that take a series of images of the pitched baseball while in flight. Nine parameters are recorded: position, velocity, and acceleration in each of the three dimensions. From these parameters we can then extrapolate location, movement, and break. However, these cameras are not always completely accurate. Consider the work of Max Marchi and Mike Fast, who both show that the cameras often incorrectly estimate pitch location by small, but significant amounts. Given the pervasiveness of these small errors it’s hard to confidently say that Mo’s pitch location in 2011 is significantly different from 2010. We should also consider that I have not accounted for differences in the context of these pitches. By this I mean that Mo throws to different locations in different counts (0-0 vs. 0-2), so if Mo has had a different distribution of counts this year then he could have a different pitch location graph while still throwing to the same locations in each count! This is similar to Simpsons’ paradox.
For the above reasons it’s pretty hard to say that there is a notable difference in his pitch location to lefties. However, lets assume that his location is actually different in a significant manner, for the purposes of this post.
Is his new location better or worse?
To evaluate this, I created a model that predicts the quality of Mo’s pitches based on location, using all of his pitches to lefties from 2008-2011. This is 1880 pitches, which is unfortunately kind of small for this kind of analysis (full seasons for starting pitchers are usually over 3000 pitches).
Red indicates locations where Mo performs best, blue indicates locations where he performs the worst. Like the other graphs, it is from the catcher’s perspective. The dotted black box indicates the strikezone for the average left-handed batter. The most blue areas are very far away from the strikezone, because in these areas pitches are always called balls.
As you can see, he is effective pretty much every where in the zone. Does this mean that he should forget about location and just throw it down the middle? No (this is kind of a game theory thing), but he can certainly miss his spot without getting hurt. We can use this model to look at the quality of his pitch location in 2010 and 2011. After normalizing my run values for differing run environments (linear weights made from a few years ago average in the negatives for 2010 and 2011), the model predicts that his pitch locations in 2010 should result in -2 run per every 100 pitches. This is an excellent value, as negative values are very good for pitchers (opposite of Fangraphs). When we use the 2011 data, the model predicts -2.2 runs per every 100 pitches – even better!
However, some caution is needed given the small sample used to create the model. The best conclusion that we can arrive at is that there is no reason to believe that his pitch locations in 2011 are worse than in 2010.
It’s hard to say that his pitch location to lefties is any different in 2011 than 2010. This is reflected in both visualizations of his pitch locations and a regression model used to predict run value. According to the PITCHf/x data, there really isn’t anything worrisome at all. In terms of velocity and movement, the pitch is identical to 2010. In a near perfect career, Mo has had a few blips every once in a while, and this is just another one of those blips. It’s not like he just rolled out of bed a few days ago without the cutter working after it being effective the entire season.
References and Resources*PITCHf/x data from MLBAM via Darrel Zimmerman’s pbp2 database and scripts by Joseph Adler/Mike Fast/Darrel Zimmerman *Mo’s player page on Fangraphs You can follow Josh on twitter @J__Stock.