Pitcher Stats Breakdown
Going through several pitch-level statistics to explain what they are, how they work, and how they predict other stats and pitcher performance
We are going to go through some of the pitcher stats I cite on this blog a lot so we can all get a little bit more acquainted with them and learn some stuff about what stat values mean.
First, let’s start off with a scatter plot that shows the up-to-date Ball% vs. Strike% plot for 2023 data (as of June 19th). The main reason for this is to start by pleasing the eye, and to give us a good cover picture for the post.
Now we will get into business. Subscribe to the blog if you’re not already - lots of stuff goes on here every day!
Considerations
When we’re talking about a statistic, there are important questions immediately before going any further.
1. Stickiness
Firstly, we want to ask: is the statistic sticky? Stated differently, we are asking:
Does a player’s performance in a certain statistic up to this point predict what they will do in the same stat in the future?
The clearest example of a sticky statistic is pitch velocity. If a pitcher has thrown his fastball 96 miles per hour on average to this point (with at least like 20 pitches thrown, we could say), it’s incredibly likely that he will continue to throw that fastball 96 miles per hour on average.
An example of a non-sticky stat would be batting average on line drives. Since we’re already controlling for the batted ball type here and player speed doesn’t really factor in to reaching base on a line drive or not, it’s fair to say that your batting average on line drives last year won’t tell us much of anything about what it will be next year, and we should always just more or less expect the league average.
2. Stabilization
SECOND, we want to know at what point can we trust that a statistic is really telling us something useful about the player. If a pitcher comes in for his first appearance of the year and throws 5 pitches and gets two whiffs, he will have a SwStr% of 40%. But it would be silly to think that tells us anything because it’s based on just five pitches. So we want to know at what point is there more signal than noise to the statistic. That’s using statistical terms to explain statistical things, so I’ll clarify that. “Signal” would be the data that actually shows us how good a player is. The “noise” is the randomness. With five pitches thrown, it’s almost all noise. Anything can happen in five pitches, it’s too small of a sample size. The stabilization rate, in this example, tells you how many pitches we need to see thrown before there is more non-randomness than randomness in the resulting statistic. This isn’t always easy to find, and it’s always better to assume we need a ton of data than a little bit.
3. Predictive Power
The third thing we want to know is the ability of a certain statistic to predict another statistic. A good example here would be SwStr% and K%. A high SwStr% means you’re getting whiffs at a high rate, and that tells us a lot about what K% we should expect to see.
Now, if you have a big enough sample, you can just look at the thing you’re more interested in - in this case, that is K%. We care more about K% than SwStr% because K% is telling us what actually happened in the results of the game. A swing and miss on an 0-0 count is good, but not nearly as good as a swing and miss on an 0-2 count.
If the predictor stat (SwStr%) stabilizes more quickly than the output stat (K%), then we have an opportunity to learn about what the K% is likely to be well before the K% itself stabilizes.
SwStr% is a per-pitch stat, meaning we get a new data point every pitch thrown. K% is a per-PA stat, meaning we get a new data point every time a plate appearance ends. We will see SwStr% stabilize way before K%, which means we can get ahead of the field a bit by knowing the SwStr%.
Okay, that is the long setup for the post. We will answer all of these questions about all of these stats as we go forward.
I don’t know myself the best way to calculate stabilization rates, but I’ll give you some decent guesses based on other people’s research. Important note: this will be the time the stat BEGINS to stabilize, which means that’s about when it crosses 50% skill-based. The stabilization will grow stronger and stronger as more data is accrued, so the bigger sample will always be way better.
K%
League Average: 23%
75th Percentile: 26.8%
25th Percentile: 19.9%
Year over Year Correlation: .62 (moderately sticky)
Stabilization: 80 batters faced
Correlation with ERA: -.44
Correlation with WHIP: -.48
The best you can feasibly do at this is the mid-thirties, but there is the outlier of Spencer Strider and his 39% K% since 2022.
The worst you can do and stay in the league is somewhere around 15% (Marco Gonzales, Zack Greinke), but most values fall between that 19% and 27%.
What does it predict?
By itself, K% does tell you something about ERA and WHIP with those correlation coefficients close to .5. If those numbers are negative, that means that as one goes up, the other goes down. A +1 would be a perfect positive correlation (as one goes up, the other always goes up in the same magnitude), a -1 would be a perfect negative relationship (as one goes up, the other always goes down in the same magnitude).
BB%
League Average: 8%
75th Percentile: 10.3%
25th Percentile: 6.7%
Year over Year Correlation: .64 (moderately sticky)
Stabilization: 80 batters faced
Correlation with ERA: .21
Correlation with WHIP: .47
It’s about the same as K%. The distribution is a little higher. The highest BB% that has survived to make at least 25 starts since 2022 is Edward Cabrera at 12.7%, and the lowest you can seem to go is George Kirby at 3.2%, but most values will be found between 6 and 10%.
It is basically not correlated with ERA at all, but it does have a good amount to say about your WHIP.
K-BB%
League Average: 15%
75th Percentile: 19%
25th Percentile: 11%
Year over Year Correlation: .56 (moderately sticky)
Stabilization: 80 batters faced
Correlation with ERA: -.51
Correlation with WHIP: -.67
To me, it is the best statistic to look at if you only have to pick one, but I would typically want to wait until at pitcher faces at least 100 batters to really start trusting it.
SwStr%
League Average: 12.2%
75th Percentile: 13.9%
25th Percentile: 10.7%
Year over Year Correlation: .71 (very sticky)
Stabilization: 250 pitches
Correlation with ERA: -.38
Correlation with WHIP: -.42
Correlation with K%: .78
Correlation with BB%: .04
There is a strong relationship with K% here, but it does a worse job at predicting ERA and WHIP, and it tells us literally nothing about the walk rate.
Now for the more unknown stats.
Strike%
League Average: 47%
75th Percentile: 49%
25th Percentile: 45%
Year over Year Correlation: .60 (moderately sticky)
Stabilization: 250 pitches
Correlation with ERA: -.47
Correlation with WHIP: -.59
Correlation with K%: .76
This stat is just the percent of your pitches that go for a strike, meaning a called strike, a swinging strike, or a foul ball.
This one stat has just as strong of a correlation to ERA as K% does, which is interesting - and it’s even better at predicting the WHIP.
The top strike rates of the last two seasons (20+ GS)
Strider 54.7%
Rodon 52.3%
Gausman 51.8%
Joe Ryan 51.4%
Max Scherzer 51.4%
Bottom:
Dakota Hudson 40%
Senzatela 40.5%
Urena 40.7%
Heasley 41.7%
Brad Keller 41.7%
Notice how close these leaders are to the average of 47%. This is a pretty narrow distribution, which means that even a one point difference is pretty significant.
Ball%
League Average: 36%
75th Percentile: 38%
25th Percentile: 34%
Year over Year Correlation: .61 (moderately sticky)
Stabilization: 250 pitches
Correlation with ERA: .32
Correlation with WHIP: .52
Correlation with BB%: .74
This one, obviously, does a good job of telling us about walk rates. Much like walk rate, it doesn’t tell us much about ERA, but it does tell us a good amount about WHIP. It is a slightly better predictor of ERA than BB% itself - which is also pretty interesting!
The top ball rates of the last two seasons (20+ GS)
Kirby 30.2%
Bundy 30.9%
Urias 31.3%
Gausman 31.4%
Alcantara 31.5%
Bottom
Flaherty 40.9%
Hudson 40.1%
Kuhl 39.9%
Davies 39.8%
Fedde 39.8%
BIP%
League Average: 17.2%
75th Percentile: 18.7%
25th Percentile: 15.6%
Year over Year Correlation: .66 (pretty sticky)
Stabilization: 250 pitches
Correlation with ERA: .22
Correlation with WHIP: .15
Correlation with BB%: -.54
Correlation with K%: -.82
Turns out that this stat kind of sucks. It tells us a lot about the K%, but that’s obvious because a strikeout and a ball in play are two of the only three options of how a plate appearance can end (when you lump walk and hit by pitch together).
Top BIP% Last 2 Seasons
Watkins 22%
Senzatela 21.8%
Syndergaard 21.6%
Bundy 21.6%
Gonzales 21.5%
Bottom
Strider 12.1%
Snell 13.4%
Greene 13.9%
Cabrera 13.9%
Rodon 14.0%
You can see that you don’t want to be at the top of this list, as those guys with high BIP% are almost always bad - but you can get to the bottom of the list in a bad way too with a super high walk rate. I don’t think you would call Edward Cabrera one of the league’s best pitchers with his 14.6% K-BB% the last two seasons, but he gets near the bottom of the BIP% list by striking out or walking most of the hitters he faces.
GB%
League Average: 43.1%
75th Percentile: 48.3%
25th Percentile: 37.7%
Year over Year Correlation: .74 (very sticky)
Stabilization: 50 balls in play
Correlation with ERA: -.14
Correlation with WHIP: -.02
Correlation with BB%: -.06
Correlation with K%: -.17
I would have thought GB% was a little bit of a stronger indicator of ERA, but it’s just not. We need way more information, but it is true if you’re going to allow a ball in play, it’s overall better for it to be on the ground. You will allow a higher batting average on ground balls as compared to fly balls, but obviously you won’t allow any homers and you’ll allow very few extra base hits, which is key.
Application
Okay, so let’s take one example and do some analysis just to demonstrate how we can use this stuff. It’s most useful on smaller samples, since those are the opportunities we have in fantasy - to use small data samples smarter than the field and beat them that way. Nobody is going to be fooled about Gerrit Cole or someone established like that.
The example I want is Andrew Abbott. Here are all of his numbers after three starts
3 GS
17.2 IP
293 pitches
7.2% SwStr%
47.4% Strike%
35.5% Ball%
17.1% BIP%
16.9% K%
12.7% BB%
4.2% K-BB%
32% GB%
He has a zero ERA after three starts, which is really insane.
His SwStr% has been horrible at 7.2%. The worst qualified SwStr% since 2022 from a starter is Adam Wainwright at that same 7.2%. We are not even to 300 pitches for Abbott yet, but that’s more than enough to think there’s some signal here - so right now we can be moderately confident that this guy is not going to have a good strikeout rate in the Majors in the short term.
Some good news here is that he’s still earning strikes at about a league-average rate. Since we know his SwStr% is super low, that means he’s getting a ton of called strikes and foul balls. That’s fine and dandy, a called strike is just as good and a foul ball is just as good except in two-strike counts, so you can live with that. The problem is that SwStr% is much stickier than Called Strike% and Foul%, so we can’t really dependent on the Strike% to stay average - in fact, it’s quite likely it will fall rapidly in favor of a few more balls and probably a lot more balls in play.
Some other good news is that his awful BB% doesn’t really line up with his Ball% which is right at the league average.
So his SwStr% doesn’t match expectations from his Strike% and the BB% doesn’t match expectations from the Ball%. It seems he’s in for fewer strikes, but also fewer walks. That means a lot more balls in play.
Then we can look at the GB% and see that it’s super low at 32% and then we can say that okay there are a lot more balls about to be put in the air against Abbott - and in his case, that’s frightening because he pitches half of his game in the worst park in the league to give up fly balls in.
All of that is to not even mention the K-BB%. If we look at K-BB% rates under 10% since last year, we find almost entirely bad pitchers. Just a sample of those names:
The average ERA of the group is 4.89 - bad stuff. You cannot have success for long if you aren’t getting strike outs and walking a lot of hitters, which is what Abbott has been doing.
Big fat small sample size disclaimer over all of that, and it’s not like the young guy can’t change rapidly on the fly. This is not necessarily an indictment of Abbott, but just a way to show what these stats can teach us about a pitcher even over a small sample size.
I would predict a high HR rate for Abbott moving forward, and probably a better bad ERA as well unless he greatly increases the SwStr% and lowers the BB%, which is possible!
So, we’ll revisit Abbott again in a few months to see if we were onto something here, but again - my point isn’t to make a claim about Abbott, but just to show how to use these stats in general terms.
Alright, that’s it, hope this helped. If nothing else, it can be a resource used to remind ourselves of averages and distributions of what to expect from all of these stats. Thanks for reading!