Baseball analytics still exist in a paradigm where the goal for many is to find or develop a “holy grail” singular stat that says whether a player is good or not. This is not limited to the WAR based pursuits that will likely at some point start to include data from things like StatCast to create better defensive metrics or try to better explain what “should” have happened on the baseball field, as opposed to what did happen on a baseball field. This mindset has moved to the minor leagues where data is more unpredictable and the motivations to win are not the same that they are in the major leagues. The most prominent of these models is Fangraph’s KATOH.
KATOH’s creator Chris Mitchell has created a new version of his rankings called KATOH+ which takes in some amount of scouting opinions from Baseball America and Fangraphs’ Eric Longenhagen. For the Phillies on this list there was name to watch not for whether he would make the list, but how high he would be, and that player is Dylan Cozens. KATOH did not disappoint and had Cozens at #26 on it’s scouting adjusted list, and at #1 on the unadjusted list. The problem is that Dylan Cozens is the perfect player to break KATOH.
In 2016 Dylan Cozens hit .276/.350/.591 with 40 home runs and 21 stolen bases in AA. He struck out at a lot, but he also drew walks at a decent rate. The fact that Rhys Hoskins comes in at #54 on the unadjusted list should help show just how much the raw stats help Cozens in a ranking based on stats. KATOH has moderate feel for context, and we will get to the ballpark later, but for the most part it only has a stat sheet to work with. I will note that it isn’t just 2016 stats, and so KATOH does know about Cozen’s previous success on the bases and previous ability to not strikeout at an absurd rate. What it doesn’t really know is the R/L splits and by now it is rather well known that Cozens has a problem with left handed pitchers. Unfortunately for KATOH projecting forward, managers can choose when to deploy certain relievers. Cozens is not going to see LOOGYs in AA, he is going to see them if he reaches the majors. KATOH also does not know that Cozens heavily pulls the ball on the ground and that the shifts against him are only going to increase in volume. Lastly on the speed, we know that Cozens has 40-45 speed, and we know this because we can measure that objectively with a stop watch. That is not a recipe for future stolen base success, but that is not captured in the stats.
The age of a prospect is a good proxy for both relative physical development and experience. In most cases younger baseball players have played less high level baseball and they have not grown in their frames yet. One of these things is very true about Dylan Cozens. Cozens is 22 and won’t turn 23 until May and he has really only be a full time baseball player since age 18. His 2016 is evidence of his continued growth as a player as he started to consistently tap into his mammoth power. On the other hand, Dylan Cozens has been pretty much physically maxed out since he was 18 or 19. In recent years he has shown up faster and more flexible, rather than being bigger and stronger. There is some amount of additional projection for Cozens, but not the same as your normal 22 year old in AA.
Reading is a huge advantage to hitters. In the comments, Mitchel mentions that he adjusts KATOH based on park factors, particularly park factors by handedness. Reading by the metrics that KATOH uses (Stat Corner) has Reading as a much better power park for righties rather than lefties. The gap actually widens if we look backwards, and we can assume that some of the left handed surge is from Cozens (which is another bit of error to this projection). When we say a park is better for power for a lefty or righty what we are referring to his mostly power to the pull side as that is where most power is going to come from. This is what a right handed spray chart taking advantage of Reading looks like (Rhys Hoskins)
If a park being favorable for righties means that the ball should be easier to hit to left field, then I present Dylan Cozens’ spray chart.
No one is doubting Cozens’ power, but his ability to hit the ball in the air the opposite way means that he was able to tap into that right field advantage. Now KATOH doesn’t know that and gives Cozens little penalty for playing in Reading.
This year Mitchell decided to add Clay Davenport’s minor league defensive metrics into his projections. I cannot speak to Davenport’s methodology, but it comes out with a +13 in right field for Dylan Cozens. Major league defensive metrics are notoriously unstable, and I would expect minor league numbers to be even more unstable. The addition of defense adds false precision to the projections, when in fact they add an additional source of error. The fact that Cozens rates out well is really only the cherry on the top of all of this.
I want to talk about one last point made by the creator of KATOH.
The second list is the old, “stats only” version of KATOH, that doesn’t account for any sort of scouting. This is an obvious flaw, as there’s a reason people in the industry always say “don’t scout the stat line.” But by considering the stats exclusively, this version is 100% objective. And since it doesn’t downgrade prospects who lack scouting pedigree, it’s good for identifying statistical standouts who have gone unnoticed.
This seems to come up a lot that stat based models are more objective than a fan. But the truth is to make the model the creator must make decisions on how the model works. We can see right here a few of those decisions such as how much to weight age, or park, or even how much to weight a strikeout at a certain level. Those are subjective decisions in the same way that a scout is subjectively grading a swing vs what they think that swing will look like vs major league pitching.
All of this is colored by the fact that I don’t think a minor league projection system works in a ranking format. It is interesting on a more micro level as to what causes hitters to break the model or what players are pushed by the model and not by scouts. But models continue to apply a major league paradigm (focus on winning and performance, not development) to minor league stats and that just does not work. The majors and the minors are very different games with the talent collection that occurs in the majors and some skills and stats are not translatable. Scouting is not perfect, but treating models like KATOH as substitutes is irresponsible for major sites.