Felix Hernandez and His Fastball, Part 1
(I know I already have another Part 1 out there without resolution, I’ll get to that eventually, but for the moment this has captured my attention)
Near the end of June in 2007, Dave Cameron wrote an open letter to then-Mariners Pitching Coach Rafael Chaves pleading with him to modify the game plan for Felix Hernandez when it came to starting out games. Namely, Dave pointed out the predictability of Felix throwing a high percentage of fastballs at the start of games. A year and a half later and has anything changed? I decided to take a look at Felix’s 2008 through the perspective of Dave’s letter and his intent.
Frankly if you followed any of Felix’s starts and were paying attention (or happened to participate in the game threads at Lookout Landing where I was constantly harping about it), you already know the answer to the posited question above. The answer is no. Now, I could leave it at that, but a 75-word post is not going to get anyone’s attention and besides which, I love making graphs.
Let us jump right into those graphs then. Here is a chart of Felix Hernandez’s fastball frequency over time; time, for the purposes of a baseball game, being measured in pitches. until it reaches his overall average frequency, about 67%.
If you find a pitch count number along the bottom (x) axis and move upward (y) until you reach the trend line, that value will give you the percentage of pitches, on an average start, that were categorized as fastballs up to that point in the game. So, after 11 pitches, roughly 84% of them had been fastballs. By the time Felix has thrown 40 pitches in a game, that ratio is down to around 74% and it continues to fall
It’s obviously not uniformly descending, but it’s really close and it paints a stark and unmistakable pattern. Felix starts out a game gung-ho about his fastball and slowly begins to work in his other pitches as the game wears on. It’s not even a gentle downward slope, but a rather dramatic curve, suggesting that the set of pitches one through about 20 and 20 through rest of game are rather disparate. Next, we’ll dive a little further into how Felix’s fastball percentage varies on a per pitch basis.
Matthew Carruth is a software engineer who has been fascinated with baseball statistics since age five. When not dissecting baseball, he is watching hockey or playing soccer.
Hey, great study and article. It is very interesting how much of a dramatic and steady decrease in fastball percentage that is. I wonder what it is like for some other pitchers? How hard was that to calculate, I’d love to see some guys like Tim Lincecum, Johan Santana, C.C. Sabathia, Cole Hamels etc.
…and while we’re at it, how about Roy Halladay? I’d love to see, on paper, his tendencies.
VERY interesting read, Matthew. I don’t know exactly the required amount of work for this particular article, but I’m sure it wasn’t exactly a walk in the park.
Thanks for doing all the work!
The initial drop in fastball frequency looks dramatic, but it is partly due to the small number of pitches into the game, so for every non-fastball he pitches, the fastball frequency would fall substantially. On the other hand, from the 20th pitch on, the downward slope of the trend line remains roughly constant, suggetsing that the fastball frequency is actually constant during the span.
What we’re really interested in is the probability that a fastball comes up in Felix Hernandez’s next pitch at a certain point of the game, but not the accumulated fastball frequency up to that point. I think in order to depict an accurate graph, you should take the logarithm of the accumulated fastball frequency.
Maybe my math is fuzzy here but what would the logarithm of that tell you? I think you want the derivative over x of (x*y) as they’re defined in that graph.
Did he really start every game with two fastballs? How did that work out for him? It looks bad but it may be hard to swing at a 96mph fastball with the first swing of the game.
Well, my reasoning is as follows.
X is a function of Y, that is, x = f(Y = y).
Taking the natural logarithm of f, and then differentiating it with repsect to y gives
d lnf(y) / dy = f'(y) / f(y)
which is the relative rate of change of f at y.
However, since f is not a continuous funciton – therefore not differentiable – the method is not applicable here. I didn’t think it through at first.
I’d agree. I’m not sure this chart shows us much. And I think very early on it would be positive to establish a 96 MPH fastball. That doesn’t mean to just throw the fastball (as Dave cites in his letter), but clearly the other pitches work better off his fastball.
Another study might be to see these percentages through each individual batter each time through the batting order. He might throw two fastballs to the lead-off hitter, its just important that he doesn’t do that every time he faces that hitter, or the lead-off hitter for each inning.
To rephrase:
I know what the graph shows, but there isn’t anything to compare it too. It is basically an extended version of Dave’s original chart.
I am going to look at Webb, Carmona, and Beckett (three similar pitchers IMO, less with Carmona tho) and see what they tend to do.
I agree that the chart should either show it in clumps (say per 10 pitches), or should ALSO show “future accumulation”, as in, starting from pitch 2, he’s throwing x% the rest of the way. This way, you have two lines, one that shows his accumulation to date, and the other shows his future accumulation.
As it is, you can’t really tell what he’s doing at pitch 40.
Something similar is coming later on. This is a past-looking chart, and as such, yes, of no real use for showing what he will throw next given a certain set of known variables.
But yes, the idea of showing a future accumulation is one I should have added in.
Perhaps it’d help to show a 10-pitch moving average of fastball frequency. In other words, an average for every 10-pitch span – 1st-10th, 2nd-11th, 3rd-12th pitches, and so on.
Can this be directly attributed to Mel Stottlemeir’s pitching plan of slow, hard and harder, or in this case harder, hard and slower, or were their other factors at play in the pitch calling? Did Felix always have the same catcher? More specifically, is this the same pattern whether Kenji, Burke or Clement were catching and if he did have more than one catcher did it matter? Am I barking up the wrong tree? Either way I suppose Rafael Chaves could alter it if he thought it necessary. I’m just wondering if the pattern in pitch calling was a lack of subtlety in the cather’s part or the coaches part and if Clement being the full time catcher would have an effect on Hernandez’s selection regardless of Chaves’ acknoledgement of the issue this year. It would be interesting to know if Clement caught Felix as well if there were any significant difference in the selection we could expect.
Interesting read. thanks
It would be interesting to see what the data actually shows (breaking it down by catcher) by my recollection was that they were getting their orders from the dugout (to the point that you often saw Kenji glancing to the dugout before pitches) so I wouldn’t expect it to vary much by catcher.