Simeon Woods-Richardson was once a very highly regarded (Top 100 overall) young pitching prospect. He was twice traded as a key part of major MLB deals, but some of the shine on his prospect apple has worn off a bit after a perceived stalling of his development in a traditional scouting tool sense and a couple of decent, but unexciting seasons in the high minors.
Woods-Richardson has not achieved the velocity gains that were projected of him as an amateur and now operates with a seemingly pedestrian 90-91 mph fastball, an above-average changeup that he throws nearly 40% of the time, a big breaking, vertical curveball, and a slider.
The scouting report on him at FanGraphs includes this line: “There’s no plus pitch here, but there are four viable offerings that all move in different directions as well as sentient command of them all.”
Though that hardly reads as the makings of an arsenal that would grade well by the increasingly popular stuff quality models, by the 2023 season’s end Woods-Richardson’s Stuff+ mark ranked 10th out of more than 1,200 AAA pitchers. His arsenal also graded out favorably in limited major league action in 2022 and 2023.
That apparent disconnect makes Woods-Richardson a good use case for understanding what goes into pitch quality models and the focus for this next installment in Twinkie Town’s Analytics Fundamentals series.
The makeup of every play in baseball more or less follows the same high-level script. There is an interaction between the pitcher, the hitter, and the defense that happens in this order: the pitcher delivers a pitch, the batter decides to swing (or not), he hits it (or doesn’t), and, if the ball is put into play, the defense tries to turn it into an out (or more).
For a long time, the way we evaluated pitcher performance was limited to the result of that sequence. Did the batter get a hit? Did the pitcher get an out on a ball in play? Did they get a called strike, or a swing and a miss?
If they got good outcomes, our thinking went, then they must have done well for themselves and we parsed credit to them accordingly.
That long-standing foundation of thinking started to be challenged when Voros McCracken introduced the concept of defense-independent pitching, a way of evaluating pitchers based on the factors over which they had greater control — strikeouts, walks, hit-by-pitches, and allowing home runs — and ignoring the ones over which they did not. That meant setting aside batted balls in play, which McCracken showed were largely outside of a pitcher’s direct control because of luck and their dependence on the team of defenders behind the pitcher turning balls in play into outs.
In the twenty-plus years that have passed since that concept became mainstream, all kinds of metrics that you’re probably now well-versed in — like fielding independent pitching (FIP), expected FIP (xFIP), and SIERA — have been spawned.
Although those represent improvements — and they have shown to be more predictive of future performance than actual results — if you dig into those metrics and their inputs you’ll find that they are also heavily dependent on the results end of the typical play sequence. They still don’t tell us very much about how and why the pitcher got the results he did.
Leading vs. Lagging
The Statcast system and the various measurements it makes throughout the play sequence have moved our ability to analyze earlier in the sequence, which has enabled more understanding of the physical characteristics of the pitches themselves — like pitch type, velocity, velocity differential, spin rate, spin efficiency, spin direction, location, movement, and release points.
These are all things the pitcher has much more direct control over than the result of the play. Instead of using the results to evaluate a pitch, we now can measure and quantify the inputs to it and assess which of those tend to lead to good (or bad) results.
That capability has led to a proliferation of metrics and models that aim to understand and explain the pitch characteristics (and combinations of characteristics) that tend to make pitches successful. Those include many of the public-facing “stuff” or “pitch quality” models that have become more mainstream in the past few years.
You’ve likely come across Eno Sarris and Max Bay’s Stuff+, Cameron Grove’s pitchingBot, and Pitcher List’s Pitch Level Value (PLV), among others, and MLB teams and places like Driveline have their proprietary versions, too.
When you take a look at the pitchers who score well by these methods, you’ll see the names you would expect based on how we traditionally think about ‘stuff.” High velocity and ridiculous movement are usually good ways to grade well and tend to lead to good results.
2023 PitchingNinja Award for Filthiest Splitter.— Rob Friedman (@PitchingNinja) November 22, 2023
Winner: Jhoan Duran. pic.twitter.com/Xkic45bHDU
Jhoan Duran and his triple-digit heater, otherworldly curveballs, and physics-defying splitter are at or near the top of any of these models’ lists. Griffin Jax and his big breaking, but powerful sweeper are there, too. The same goes for Pablo López and his array of pitches that move every which way.
But then there are the guys that grade out well for not-so-obvious reasons. Former Twin Kenta Maeda has regularly graded out pretty well by stuff models, despite 10th percentile fastball velocity and a slider with below-average raw movement.
And of course, there is Simeon Woods-Richardson:
Top ten starters by Stuff+ in Triple-A, min 45 pitches:— Eno Sarris (@enosarris) April 10, 2023
Simeon Woods Richardson
The various pitch quality models deviate from each other in terms of methodology and the inputs they use. Instead of dragging you through the nitty-gritty details of each of them, let’s use Sarris and Bay’s Stuff+ as an example of the overall concept.
In June of 2021, Sarris published a high-level explainer of Stuff+ in The Athletic that contained this helpful graphic laying out the model’s primary inputs and their relative importance:
You can see there that the things you’d expect — like velocity and movement — are included, but they are relatively less important than the differential of velocity and movement. In other words, that’s referring to the difference between a pitcher’s breaking and offspeed pitches from their fastball. Similarly, previously thought to be small details about a pitcher’s release point, have shown to be just as important as the raw movement measurements of their pitches.
The relative importance of the different variables is not random or arbitrarily assigned. The Stuff+ model was trained against real run value results using a decision tree model that captures the relationships across the variables. In short, overly simplified terms, the model identifies the variables and combinations of variables that have led to success in the run value results and evaluates their relative importance to those results.
The SWR Use Case
Why does Woods-Richardson grade out well even though his traditional scouting report isn’t as rosy? Let’s go through some of the inputs and find out.
As we go through, keep in mind that in pitching it often pays to be different. Characteristics in the extremes tend to be good, most of the time. That’s illustrated well by these plots, again from Sarris, that show Stuff+ marks for fastball and curveball movement profiles. Notice all the red (high Stuff+) around the outside edges and all the blue (low Stuff+) in the middle.
It’s also worth noting that we don’t have perfect visibility into all the workings of the model. I’ll be using the public data we have about SWR for illustrative purposes to try to flag which model inputs might be the ones that give SWR an edge, but I must caveat that I’m probably mixing a few apples and oranges.
We’ve already covered that Woods-Richardson’s velocity does not stand out. Neither does his spin rate data. Those are all essentially below the major league average. However, he does stand out from the crowd in many of the other categories that go into Stuff+.
Here’s a table summarizing how Woods-Richardson compares to the league average in terms of the top three most important features — velocity differential, vertical movement differential, and horizontal movement differential from his fastball:
I’ve highlighted in red the places where SWR is noticeably different from the league averages. The velocity differentials are not as big as the movement differentials, especially the vertical movement of his changeup and curveball, the large horizontal movement differential of his changeup, and the relative lack of horizontal movement differential on his breaking pitches.
We can take horizontal and vertical release points together and illustrate how Woods-Richardson compares to the rest of the pitchers with a scatterplot from Savant:
I’ve highlighted him in the plot and you can see that he operates with a fairly unique release point, compared to other pitchers. Unlike the large majority of pitchers, he throws from almost perfectly over the top (which FanGraphs described as a trebuchet) and his hand is more or less right on the center line of the mound as he releases his pitch.
We covered movement differential, which gives some insight into this section, but it’s worth covering movement in more detail because Woods-Richardson has some unique raw movement features, especially with his fastball.
Whereas most fastballs move to the pitcher’s arm side 7 or 8 inches, SWR’s average fastball moves to his glove side almost 3 inches. His over-the-top delivery enables him to work straight behind and even a bit around the ball which gives his fastball a unique combination of above-average vertical break (ride) and glove-side cutting action.
What’s more, and largely because of his unique over-the-top release, Woods-Richardson is also differentiated in terms of spin direction and extension. He is the rare right-hander whose spin axis is on the high side of 12:00, which helps create that cut action on his fastball. He also gets extension down the mound just shy of 7 feet on average, which is significantly above average and serves to shorten the reaction time for the hitter and make his velocity play a touch above its mile-per-hour readings.
Although Woods-Richardson does not stand out in a traditional velocity and raw movement sense, he is differentiated (in some cases, significantly so) in many of the other categories that stuff models consider. And, in pitching, being different is usually a good thing. That likely explains much of why those models look favorably on his arsenal.
To wrap this up, it is worth making the usual disclaimers about aggregate statistics and models. All models are “wrong,” but some are useful. And, aggregate statistics are often poor predictors of individual outcomes.
That said, they often have value in helping us understand the things that tend to lead to successful outcomes. That’s what Stuff+ (and other models like it) do.
It’s also shown to be more predictive of future performance and a reliable indicator faster than our other metrics (like ERA, FIP, etc.) Just 80 pitches are required for Stuff+ to become reliable via Cronbach’s alpha, which is an approach that is widely used to determine when a statistic is more signal than noise. Stuff+ is also more sticky across seasons than other factors.
Of course, a high Stuff+ mark does not guarantee a pitcher will find success on the field, and it doesn’t guarantee anything for Simeon Woods-Richardson’s future. But, it might suggest he has a better shot than traditional scouting methods give him credit for.
Pitchers with high Stuff+ marks tend to be more successful than those with lower Stuff+ marks, and not only because they have superior velocity and movement. These types of analytic tools have enabled us to better identify and value the other characteristics that can lead to success.