Quantifying Command – Part Two: Edge Percentage

Series Introduction:

With the modern wave of information, every facet of a pitcher’s game can be quantified now better than ever. ‘Command’ – or the ability to locate a pitch – has remained elusive in being properly encapsulated. The goal of this series is to better analyze players’ command solely through the use of statcast data while providing insight on evaluating metrics along the way.

Part Two: Edge Percentage

When you imagine a player with perfect command, what do you think of? Personally, I remember Bartolo Colon toward the end of his career. Sitting between 85 and 90 miles per hour with his fastball, Colon had no ability to blow batters away. Instead he relied on his ability to put the ball in places that would induce called strikes. The catcher would repeatedly set up on the edge of the strike zone and Colon would consistently hit the catcher’s glove.

Subjectively, it was easy to see he had excellent command. Determining how this manifests in statcast data is a more nuanced exercise. Theoretically, if a pitcher is aiming at a particular spot, repetition should yield a circular or oval-like cluster around that location. If pitchers threw each of their pitches in the exact same location, then just measuring the variability in locations would be a decent solution.

Unfortunately, it is not that simple. Especially with fastballs, pitchers mix up their locations. Overlay a few tight clusters near one another and they can look like a larger cluster, deceiving the true extent of a pitcher’s command.

Take Zach Davies’ sinker locations. Cumulatively, they spread across the strike zone, concentrated toward the lower half. By itself, that distribution would suggest he set his target low in the zone and has little control over which side of the plate the pitch finishes. However, filtering by batter handedness shows he consistently keeps his sinker low and away, regardless of the batter’s handedness. Filtering by count further suggests he can precisely command the extent to which the ball is in or out of the strike zone. Naturally filtering through batter handedness and generalized count does not entirely capture his intentions – there may be a 2-2 count (even) where he tries to keep the ball off the plate as if he were ahead – but it illustrates how subdividing situations are fundamental in subjectively analyzing a pitcher’s command with statcast data.

So when subjectively analyzing ‘indicators’ (a term discussed in part one of this series), it is key to filter pitch locations by situation. If a pitcher exhibits tight clusters by pitch type with variation in location by situation, they probably have good command. They may be getting lucky with misses clustering, but increased sample sizes should remedy that concern. On the other side of the coin, a pitcher with a lack of distinguishable clusters and little variation by filter may be sticking with a diverse set of location intentions, but in all likelihood it is that they lack the ability to precisely control their pitch’s location.

The Edge% statistic takes the perspective that pitchers with elite command would want to pitch to the edge of the strike zone. It is a decent idea as the edge of the strike zone is generally a pitcher’s best bet; it has the late count benefits of producing weak contact and whiffs while still being called strikes at a good rate. Additionally, with it being locationally based, it applies to every pitch and is catcher/umpire independent.

Attack ZoneCalledStrike%Swing%Whiff%xwOBAcon

Where Edge% runs into some theoretical pitfalls is with the three assumptions it makes. First, it is assuming pitchers are always trying to throw to the edge of the strike zone. This is faulty in various counts and certainly faulty with various pitch types. In most 3-0 counts, it makes more sense for a pitcher to challenge a hitter with a pitch in the heart of the plate as opposed to aiming at the edge and risk granting a free pass. Likewise, in most 0-2 counts it makes sense to stay in the ‘chase’ or ‘waste’ regions. While the batter is less likely to swing or be called out on strikes, there is a significantly lower chance of them putting the ball in play and escaping such an unfavorable count.

Off that, certain pitches are better at attacking the zone while others are better baiting the hitter to chase. So it will naturally make sense for the latter pitch types to have higher usage ahead in the count and thus lower edge percentages. Furthermore, pitchers with the ability to bait swings father out of the zone have more leeway staying away from the edge of the zone. 

Will Harris’s curveball is a prime example with it yielding the most swings and misses just inches off the ground. While he frequently uses his curveball as a ‘get me over’ when even or behind in the count, his locations when ahead show his ability to command the pitch low. His curveball may have a low edge percentage (30.5%) – even compared to the average MLB rate on curveballs (38.6%) – but his curveball moves a lot and keeping it away from hitters plays to its strengths. It is not that he lacks command with the pitch, it is that he commands it where few others can afford to.

The second faulty Edge% is making is that all pitches on the edge of the strike zone are equal. While the width of the zone does not cause too much concern, how pitches play in different parts of the zone becomes an apparent issue. While both the following Adrian Houser sliders to Paul Goldschmidt were technically on the edge of the strike zone, I have no doubt he knows why one ended up in the catcher’s glove and the other in the centerfield bleachers.

Breaking balls and offspeed pitches generally do better low, breaking away from the heart of the plate while fastballs do better up in the zone. Edge% views all edge locations as the same and thus would score both Houser sliders the same when clearly one demonstrates a lack of command.

Despite these shortcomings, Edge% produces a promising list of indicators. Kyle Hendricks, Zach Davies, Masahiro Tanaka, Marco Gonzales and Dallas Keuchel are all strong positive indicators, but a few names do raise some eyebrows. For one, there’s Noah Syndergaard sitting in third among pitchers. One would generally expect command to be a sort of residual from stuff and performance. So if a pitcher has elite stuff but is struggling, their scoring poorly in command would make sense (Edwin Diaz). However, if a pitcher averages 97.6 mph and scores elite in command, then a 4.28 ERA would be confusing (Syndergaard).

That’s where Edge%’s third and final assumption comes into play: It assumes all non-edge pitches are the same. Gio Gonzalez may have had a 1st percentile Edge% last year, but the metric does not care whether the rest of his pitches were down the middle, thrown behind the batter’s head, or a perfectly executed chase pitch.

In Gonzalez’s case, his non-edge pitches are abnormally shifted out of the zone. It is likely that is why his Edge% is so low in the first place: He avoided the middle of the zone and was constantly begging hitters to take ill-advised swings. It contributed to an elevated walk rate, but also explains how he managed to be an above average pitcher despite his fastball sitting below 90 miles per hour.

Edge% is a leap forward in understanding pitchers, but like walk rate, it captures stylistic tendencies adding noise to the actual command being measured. It measures a generalized symptom of having command without adjusting for context and favoring those who pitch as it expects. Command comes in all shapes and sizes. Like walk rate, Edge% is a valuable tool best used in concert with others to get a complete picture on how pitchers operate.

One thought on “Quantifying Command – Part Two: Edge Percentage

Add yours

Leave a Reply

Powered by WordPress.com.

Up ↑

%d bloggers like this: