Can We Simplify Blast Motion Pt. 2

In our last article we looked at whether or not it was possible to reduce Blast Motion metrics down into a few different numbers. In doing this we help a team wanting to evaluate players, since just having a few different metrics to look at (one for power and two for contact) is a lot easier than having many very similar metrics. In this article, we continue that process to confirm our results.

These graphs show the relationship between Blast metrics. Positively correlated variables are grouped together on here. For the most part we see power metrics and contact metrics being grouped together. Negatively correlated metrics are on the opposite side of the metrics. Variables closer to the center of the plot are also less important to the first components. On both graphs, we see similar trends, and this is a nice way to visualize the new dimensions we created. Our first dimension is that power grouping you see. The second is mostly focused on the longer arrows with contact/connection metrics and OPE and the third is based on the shorter contact/connection metrics.

While the new numbers we have created don’t inherently have meaning, we have learned a few things from this process. First off it is clear that Blast metrics are very correlated to one another. It’s also possible to combine them into a few different numbers that represent the data well and simplify the process. Secondly, while the numbers they are broken down into don’t have meaning, a team could still benefit from creating a scale for new numbers, one representing power and one or two for contact. A team also can clearly see from these graphs that since so many metrics are similar, they don’t need to focus on all of them. This could help a development or evaluation based department. When developing a player, maybe picking one or two power and contact metrics would. Those metrics could be picked based on a coach’s belief of what matters, what is trainable and what hitters comprehend. Doing so would make understanding and implementing a system for Blast easier and coaches wouldn’t have to worry about missing out on information by simplifying things. For teams wanting to analyze a player, they also don’t need to focus on every metric. They could do something like PCA on their data, create a power dimension, provide a scale for it and have just one power number they evaluate players on.

Before we go straight to using our new dimensions though, we need to confirm that they actually line up with performance. So next, we will take a deeper dive into batted ball data from Cape Cod that was paired with Blast data.  To start off, we’ll look at how some metrics relate to exit velocity on batted balls in play from one team’s Cape Cod data last summer. Bunts were removed here, so there are a total of 307 batted balls we are working with. It is worth noting though that other larger data sets have worked on these relationships. The purpose of today is really to see how our power dimension metric relates to EV.  

Looking at our Dimension 1, a dimension based on power metrics, we see Exit Velocity generally increases as it increases. What is worth noting for this relationship though is that our power dimension is created from the more than 13,000 Cape Cod swings that were in the dataset. It’s possible that if more balls in play or just paired Blast and Trackman swings were gathered that we would be able to make a new dimension with just that data and see a stronger relationship.

              Another thing we can do is look at how multiple dimensions or Blast metrics represent xwOBAcon. xwOBAcon can be found by using a simple KNN model that assigns an xwOBA to a batted ball based on other balls from MLB data that were hit at similar launch angles and exit velocities. The other batted balls being compared would be the average wOBA for each launch angle and exit velocity combination of MLB data. The graph here shows the xwOBAcon for any ball in play for Cape Cod data based on that ball’s Exit Velocity and Launch Angle. While this is a cool graph to look at and goes a nice way to showing the benefit of hitting the ball hard at good angles, it is not the point here.

              This following graph (below) tries to do a similar thing, taking a batted ball’s Bat Speed (EV proxy) and Attack Angle (LA proxy) and seeing how they relate to xwOBAcon. There are a lot less dots in this photo because a lot less swings are paired with Blast. Overall we do see the benefit of good Attack Angles, and higher Bat Speeds seem to help, but the relationship isn’t as defined as I expected it to be.

    Finally, we can try looking at xwOBAcon based on our new dimensions. This would be xwOBAcon based on Dimension 1, a combination of our power metrics, and Dimension 2, a combination of our contact metrics. This is interesting. Balls hit with a higher Blast power dimension do generally have better xwOBAcons, but there still isn’t anything as defined as before.

One last thing we can do is look at simple correlations between metrics in a table to see if there’s any trends we are missing in our graphs. I should stress that much larger data sets have found different correlations, so the point here isn’t as much about the exact number but seeing where our dimensions stack up.

              Although none of these correlations are particularly high, and they certainly would vary based on the data, it is cool to see our power dimension near the top of the correlations in EV. Even more noteworthy is how our Angle Dimension (a combination of mostly Attack Angle and Vertical Bat Angle) related well to launch angle and distance.

              While I certainly wouldn’t recommend swapping out actual Blast metrics for the components we’ve created here, I do think we’ve shown it’s possible to reduce Blast metrics down into numbers that represent them well and potentially relate similarly to performance. As mentioned last time, if this helps a coach gain confidence in picking metrics they feel are useful instead of using everything now, then that could be a worthwhile pursuit. If it helps front offices interpret Blast better by just having three numbers then this also will have been worthwhile. I also want to make it clear that I think it is great Blast Motion offers so many metrics. All the numbers they provide have a value, there are just some situations where one could be better off simplifying things.

Going forward more data is certainly needed, but it would be interesting to see how these trends hold up with other batted ball data, or if models to predict exit velocity or xwOBAcon on a ball from Blast Metrics could be created. It would also be cool to see if a batter’s season performance could be modeled from Blast metrics. Unfortunately, my data set was too small to pursue any of these ideas, but they are worth keeping in mind down the line.

In our next article we will shift gears and aim to get a better understanding of the swing on various pitch types and strike zone locations using Blast data.