Toward a Pitch Arsenal Score Statistic

You’ve heard me yammering about pitch-type peripherals for two years now, and we’ve made some advancements along the way. We established some good pitch-type peripheral benchmarks, and we took a first look at properly weighting each pitch. We’ve started to get a sense of how these things interact when it comes to the shape and speed of pitches. We’re making progress.

It’s worth stepping back and figuring out what the aim is at this point. Because we aren’t trying to rank the best starting pitchers overall, really. We’re trying to find undervalued pitchers before the market realizes that they’re good. So we have to move in the smallest possible samples. And we want to have a list of great pitchers that has some weird names on it as well. Those names, we hope, will soon start to make sense.

So, to that end, I’ve taken each pitch type and looked at only those pitchers that have thrown 100+ in each of those types. I’ve summed the ground-ball and swinging strike rates for each pitch, and then found the standard deviations. I’ve given each pitcher a z-score for his ground-ball rate and swinging strike rate on each pitch type. Then I’ve summed the z-scores for each pitch type, and then for each pitcher.

What we should be looking at is an Arsenal Score. With this way of looking at things, it’s possible to have one dominating pitch and still score well. Or a group of lesser pitches that are all positive.

What we haven’t done yet is nail down what the smallest sample for each pitch is. Or how to weight the pitches. Or how to weight the whiffs versus the grounders. So this may look different once we weight each pitch differently, and if we find a way to weight grounders and whiffs more correctly.

But at least we have a first attempt at it here.

So here we go. Here are the top 30 pitchers by Arsenal Score, or the sum of their pitch-type z-scores for grounders and whiffs.

Pitcher Total
Carlos Carrasco 8.93
Carlos Martinez 6.81
Felix Hernandez 6.78
Gavin Floyd 6.77
Marcus Stroman 6.12
T.J. House 5.58
Jaime Garcia 5.54
Brett Anderson 5.24
Zack Wheeler 4.93
Corey Kluber 4.88
Clayton Kershaw 4.69
Francisco Liriano 4.07
Cole Hamels 3.74
Trevor Cahill 3.55
Brandon McCarthy 3.53
Homer Bailey 3.51
Gerrit Cole 3.51
David Hale 3.43
Garrett Richards 3.40
Jacob deGrom 3.39
Tyson Ross 3.23
Wade Miley 3.19
Allen Webster 2.76
Gio Gonzalez 2.75
Sonny Gray 2.71
Jake Arrieta 2.58
Yordano Ventura 2.48
Wily Peralta 2.39
Tyler Skaggs 2.37
Jeff Samardzija 2.36

This is limited to starters — Bryan Morris was actually at the very top of this list before we culled the relievers. And that might be a bit of a problem. If you only threw two pitches, but both were really good, you’d do well here. Should you? Or should you get a demerit for not throwing more pitches?

Carlos Carrasco doesn’t care. Either way, he was there in the top five, the lone starter among relievers. Each of his pitches has rock star peripherals, and he blows the field away. But Carlos Martinez was a reliever last year, and maybe he shouldn’t even be here. He didn’t throw 100 changeups, so we’re judging him solely on the basis of his breaking ball and two fastballs (the sinker was second-best to Aaron Sanchez with a 3.2 score, minimum 240 thrown, and the curve was third-best overall to Brett Cecil and David Robertson with a 2.5 score, minimum 350 thrown). Which are awesome. Does it matter that his change that has diverging peripherals (17.5% swinging strikes, 42.5% ground balls, 40 thrown)?

Felix Hernandez is great, so nice that he’s near the top. Ditto Clayton Kershaw, Cole Hamels, Gerrit Cole, Jacob deGrom, Gio Gonzalez, Sonny Gray, Yordano Ventura, and Jeff Samardzija. Call that the Affirmation Wing of the list. You knew they were good.

The veteran question marks often have the Carlos Martinez question mark behind them. Gavin Floyd‘s changeup doesn’t show here, even if he found something with it this year. Brett Anderson is graded on his sinker/slurve alone. Ditto Brandon McCarthy’s lack of change. Tyson Ross samesies. Even Sonny Gray has the changeup problem.

But there are some names that should go on your sleeper lists because they are listed here. Marcus Stroman has the arsenal of an ace. T.J. House got legit results on his many pitches, and he survived a negative sinker (-.1) this year and thrived. Zack Wheeler also did well despite a changeup that rated negatively by this metric (-0.6). Brandon McCarthy had the best four-seamer by a starter (2.3), and adding that to his two-seamer/cutter/curve mix might be all he needs for an excellent year in 2015. Wade Miley might be the odd pitcher to do better in the American League — all four of his pitches rated well, and his slider (1.7) was a top-fifteen slider among starters.

Then you get to the guys that should be deep league considerations near the end of the list. David Hale will get even more credit if we find a way to reward balanced arsenals better — he has four pitches, and three of them had a combined z-score of one or better. If the Braves get another starting pitcher, though, he won’t have a starting job. If he goes into camp as the fifth guy, though…

Allen Webster has terrible command. And terrible body language. These are things that were said about Carlos Carrasco, but again, they’re there if you watch him pitch. And yet his zone% is only ten percent worse than league average, and his first strike rate is virtually league average, and he had modest walk rates in the minor leagues. And his arsenal includes the fourth-best changeup by z-scores (2.4), an above-average slider (.15), a good sinker (2.0) — and survives a terrible (-1.8) four-seamer. What if he ditched the four-seamer and only walked four batters per nine innings next season? He’d still have to leap over a lot of names on that depth chart.

So there you have it. A first shot across the bow. A name. A first round of sleepers.

Perhaps soon, Arsenal Score will get an update.

With a phone full of pictures of pitchers' fingers, strange beers, and his two toddler sons, Eno Sarris can be found at the ballpark or a brewery most days. Read him here, writing about the A's or Giants at The Athletic, or about beer at October. Follow him on Twitter @enosarris if you can handle the sandwiches and inanity.

Newest Most Voted
Inline Feedbacks
View all comments
8 years ago

So what about weighting the pitches on how often they are thrown?
For example, if a pitcher has an above average fastball that is thrown 60% of the time, it should matter more than a breaking ball that is way above average, but only thrown 10% of the time.

Mark Liu
8 years ago
Reply to  Eno Sarris

And what about finding the mean value of a pitcher’s pitch instead of adding it and giving so much credit for a pitcher with 4 slightly above average pitches that he only uses two of? 😀

Jeff Zimmermanmember
8 years ago
Reply to  Matthew

I think there is some value for weighting and non-weighting.

Weighting will get closer to xFIP (strikeouts and groundballs).

Non-weighted looks to find pitchers who could possibly change their pitch mix to see some improvement.

8 years ago
Reply to  Matthew

This edges even closer to tripping over the elephant in the room, which is sequencing. One can imagine two pitchers with exactly the same repertoire thrown exactly as often, but with very different results because of the order in which they throw the pitches.

8 years ago
Reply to  Joe

Wouldn’t the sequencing factor already show up in the whiff and GB numbers?

Logan Davismember
8 years ago
Reply to  Matthew

Agreed. This would make sense as a counting stat; sum of the z-scores of all pitches thrown. (Or as a rate stat weighted by percentages I guess.) As it exists, I think this system probably overvalues guys who have a lot of pitches and thus more z-scores to sum.

8 years ago
Reply to  Logan Davis

It seems like pitch weighting should be done on the basis of how important of a pitch it is in the game of baseball (which is somewhat dependent on how often the guy throws it, but not totally)

Having a bad fastball is, in my mind, harder to overcome than anything else. So it should be weighted accordingly.

Also, a lousy curve hurts you less if you have a filthy slider. But I would think having both is probably more of a boon than just summing the curve’s value plus the slider’s value.