Shortstop Gold Glove Breakdown 2023

While digging through BaseballSavant, I was hoping to find different ways to see how to create defensive spray charts based on different batted ball outcomes. I took some of my knowledge of Pybaseball and used the "spraychart" function to create defensive charts for each player based on the data I could pull from last season. 

The public data for defenders is admittedly pretty spotty and unfair when making a complete evaluation. I have been inspired to make defensive "report cards", much like others have done for pitchers who evaluate Stuff+ models as well as clustering of pitches for each start. Unfortunately, the access to defensive metrics and reliability for that matter make it quite difficult to do so. 

That said, I tried to start an analysis by looking at where players are making (or not making) plays on the field to gain a little more understanding of what types of batted balls each player is seeing.

*It is important to note that the data supplied here is very imperfect and it should not be taken as a sweeping understanding of these specific players complete defensive analysis. For example, Statcast has Dansby Swanson with 5 plays as errors in my data while Baseball Reference credits him with 12 in the 2023 season. Additionally, some hits credited to each of these players may be an unfair assignment, as some of the "hits" are usually unattainable plays to make...but I digress.

The spray charts I have included below were a cool way for me to apply some of what I have learned in Python to the defensive side of the ball. While an imperfect way to view defensive efficiency or tendencies, it still shows interesting clusters of the distances balls are hit and what their ensuing outcomes are. 

In the data below, we get to see a look into the different batted ball metrics that came into play during the 2023 season. Mean and median differential is important in some cases, and I personally think it is best to show both when we look at launch angles, launch speed, and hit distance. 

• wOBA_value ... is not all too indicative as evidenced by the identical median values for each. I initially thought that there could be some discrepancy especially seen in errors but very small impacts across the board. 

• hit_distance_sc ... the hit distance shows an interesting comparison between the two players. Volpe made more errors on balls that were hit shorter on the diamond, while Swanson made more errors on balls that hit first at points further into the field. Video backed up this data as well, as Dansby made a couple of errors with his heels on the outfield grass, or even running for a fly balls towards the outfield. More of Volpe's errors were ground balls that bounced early. Here is another flaw in the data, as hit distance is not terribly indicative of the ability to make a play. It would be better to have the defensive positioning data for that. 

• launch_speed ... Volpe's errors were hit hard. His average launch_speeds on all of his batted balls are higher than Swason in all three outcomes too. 

• launch_angle ... the hits for Dansby were low balls that hit the ground early and got through. For Volpe, the positive (albeit nearly zero) LA means those hits were in the air for just a bit longer. Groundballs hurt Volpe on the errors. Dansby suffered on some of those balls hit in the air. 

• outs_when_up ... this one was a throw in metric for me, though it definitely contributes to the second and third order effects when looking at errors specifically. I am working to have my weighted errors metric incorporate more of these effects and the outs when these plays occur is interesting. Volpe committed more errors with two outs in the inning, while Dansby committed more with zero. Again, nothing to discern from that right now but still a small fun fact.