Update 2-26-24

Since the last update a lot has changed (most importantly Marcus Smart is no longer a Celtic, sad face). I intern for the basketball team at Concordia now, I took a class on sports analytics (shoutout professor Joshua Wyatt Smith), and made a minor but impactful change to OVS. The biggest change has just been improving my methodology (I won't expose myself too badly but I was hand-tracking stats for Concordia in my school notebooks).

The sports analytics class is one of the few being offered at the university level, and the opportunity to work towards my goals in an academic setting was an invaluable experience. Course design meant I could go anywhere I wanted with my research, so I turned to tackling the same question that OVS addresses: what is the usefulness of a midrange shot? By using NBA tracking data I had enough information at my fingertips to get into the meat+potatoes of the question. Although the data was not always reliable, my findings pointed in the same direction as my intuition. By tracking shots on subsequent possessions, I observed teams getting better less contested shots on the possession following a midrange shot, and converting those opportunities more frequently. If this holds, it demonstrates that midrange shots have at least some sort of value that other shots don't, and it may prove they are undervalued like I think.

Having access to Josh was probably the most valuable part of the class in truth. He really did an excellent job in designing the course. It meshed well with the thoughts I'd been having independently that analyzing sports should take as much creativity as it does to play them. I think a lot about encoding the logic and intuition of basketball with the things I do on a daily basis for Concordia. The first office hours Josh held I was up there, and ready to get into it. First thing he told me: OVS is bullshit (my words not his). For what it attempts to capture, it is far too narrow in terms of the scope of data it takes into consideration. Incorporating closest defender data is the obvious next step, but in its current form it can still give us some good information.

The change to OVS really was simple: divide the sum by 5 instead of subtracting 5. I think I literally realized this as I was writing the formula on that chalkboard for Josh in that first conversation. It fixed one major problem immediately; drastically re-shaping the distribution of values the stat spits out. Where values previously could get exceedingly high, especially for centers, these values are now much more normalized. Yet it still maintains the mathematical foundation of OVS. When I was first building the stat, I wanted to build it in a way that was robust. Drawing from the data itself without excessive need for user input or weighting; almost as an effort to make sure the essence of the information was able to be felt in the output. At the end of the day, what I'm really doing is conveying information, and in sports particularly it's hard to capture the whole picture. So we break it down into counting stats, or into play-by-play, or some combination of the two (tagged event data). We really need to be on the continuum, we need to use data in a way that is faithful to the inherent truth that it represents. That's my little diatribe for now, suffice to say I'm now more confident in explaining OVS.

Now, I can break down OVS to be even more granular. We could use 10, 15, 24, 100 different zones and still get a sensible output. You could tailor it to any situation, or any combination of zones. I have a particular interest in identifying zones to cluster by using the techniques we learned in class, but between watching film for Concordia and finishing off my degree with some high-level math classes, I've been too busy. Once the season is over I'll have a little more time on my hands to get back into it.

At this point I will mention that my project on midrange shots led me down the rabbit-hole when it comes to "gravity"in basketball. The conclusion I was reaching before I identified flaws in the data was that midrange shots may have greater gravity because of their situation inside the swaths of empty space between the three point line and the rim. Although it's on the back-burner for now, I wish to incorporate data-driven methods for dynamical systems in an attempt to model basketball as a kind of 10-body problem. The goal being to approximate the gravity that players have as they move around the court. Doing this would combine everything I have talked about. The logic and rules of the game need to be baked into the construction of any system aimed at accurately capturing player behavior.

I suppose that's all for now. I'm going back to Boston for Sloan in three days, and Concordia has our RSEQ semi-final game in two days, and hopefully I'll be able to make it back to catch the final if we make it. Shoutout Sami Jahan for 1,500 points, shoutout McGill for coming in last this year.

-Nick