I’ve spent a decent amount of time over the past few weeks/months thinking of ways to better visualize pitch f/x data especially in the context of an at bat. It just so happens that in my day job I spend an equal amount of time analyzing transportation networks where a common data visualization is via network graphs. I thought there may be some cross applicability there, so I started to create some graphs just to see what they might look like and what they might tell us. I decided to start with Shelby Miller for a couple or reasons 1) His arsenal is reasonably limited 2) It feels like he has lengthy at bats against him frequently 3) He’s really good!
So with that in mind I present the following Shelby Miller network graph (I’ll explain in detail after the graph itself)
You’re probably going to have to click on the graph to enlarge it and see what’s actually going on. Let me spend some time orienting you to the graph. Each node (the blue-ish boxes) represents either a count or an at bat’s outcome where for this particular graph I binned the outputs into BB, K, Hit (H), and Out (O). The color of each line represents the pitch type (legend in the upper left) and the width is in proportion to the number of pitches. Starting at the top we have the first pitch of an at bat and can walk through the counts and look at pitch selection and how we progress through various counts, looking for thick lines along the way to see trends.
Now is probably a decent enough time to talk about the weaknesses of the above. The primary weakness is that this ignores sequencing, so a 1-2 count is a 1-2 count no matter how you get there. Clearly in practice this is a false assumption (although I’ll try and deal with that in a separate post later), but it you want to capture the whole at bat we need to simplify the combinations somewhat.
With that in mind we can probably still glean some things from the visual reasonably quickly. 1) Shelby pounds away with the fastball really no matter the count. 2) He doesn’t appear to waste a ton of pitches (look at the width of lines between 2 strike counts and Ks and compare to the width of lines between 2 strike counts and the next count if he would have thrown a ball). 3) He gives up a decent amount of 2 strike foul balls (the loops that return to the same count).
Now for basically the same graph, only with the type of hit broken out.
HRs come from the fastball primarily, but come from a wide variety of counts (i.e. he wasn’t necessarily behind in the count serving it up per se).
All in all, I think there’s more to get at with these types of graphs (trying to look at sequencing, looking at a couple of pitchers to spot differences), but those will have to wait for another post.