Using Similarity Scores to Visualize Shooting in the 2024 Class

What can we learn about the shooting potential of this year's class, from the 599 players who entered the NBA out of college over the last decade?

Alright, let’s do this intro Q&A style so the methodology and justification of this exercise don’t get as boring as they could be.

What is this?

I’m calculating the similarity between the shooting profiles of 2024 NBA Draft prospects and the pre-draft profiles of all 599 players who entered the NBA out of college over the last decade and played at least 200 minutes over their careers.

How?

There are multiple ways to calculate the similarity between two datasets, but for this particular exercise, I’ll be using a method heavily inspired by Basketball Reference’s similarity score, which they use to compare player’s careers.

In their case, they take two different players and sort each player’s seasons by Win Shares from best to worst. Then they calculate the absolute difference for every year and assign different weight to every season (they multiply the best season by 1, the second best season by 0.95 and so on), with the goal of giving more weight to the players’ peak seasons.

In this case, I’m taking the NCAA box score and box score adjacent stats that have a higher correlation with a player’s NBA three-point volume and percentage, and, just like in Basketball Reference’s case, I’m calculating the absolute difference between each stat for each player’s final season before the NBA, assigning different weights for different stats.

The stats I considered and their respective weights are as follows:

3-Point makes per game - 0.9
3-Point attempt rate - 0.9
3-Point attempts per game - 0.9
3-Point % - 0.7
3-Point makes per 36 - 0.6
Offensive Rebounds per game - 0.5
Free Throw % - 0.3
Field Goal % - 0.3
3-Point attempts per 36 - 0.3
College Class - 0.1

Wait, Offensive Rebounds?

Surprisingly, Offensive Rebounds per game at the NCAA had the seventh highest coefficient of determination (or R²) for NBA three-point percentage and the sixth highest for NBA three-point attempts per game.

While it’s true that correlation does not imply causation, the more I thought about it, the more it made sense to have Offensive Rebounds as a proxy for size, athleticism, and overall productivity.

Does it have any predictive value?

Let me put it this way: if there’s someone who works in analytics for an NBA team reading this post, they will look at it and laugh. There are several models out there that are far more complex and overall better at predicting NBA outcomes for college players.

When it comes to its predictive value, and being optimistic, this exercise can only be taken as a better approximation to a prospect’s future NBA three-point percentage than simply looking at their college percentages.

Let’s take the 2021 NBA Draft class as an example. The average difference between a prospect’s NCAA and NBA three-point percentage was 7.31%. However, the average difference between a player’s NBA three-point percentage and the average three-point percentage of the ten most similar shooting profiles was less than half — 3.28%.

Can this be improved in any way?

Yes, there are multiple improvements that I would like to make in the next iteration of this type of analysis. The main stats that I would like to add to the mix are shot types (pull-ups, catch-and-shoot), assisted vs. unassisted field goal volume and percentages, and field goal percentage per zone. I haven’t found that information on a format that I could scrape into a table.

The other improvement area is the type of analyisis that could be done with these stats. At some point I want to try random forest regression, which is not a new type of analysis to the draft space: Sean Derenthal did it around 6 years ago at The Stepien to predict draft prospect’s careers at their peaks. In order to do that, however, I would need to improve massively in terms of my knowledge of Python.

So, why should I read this?

While this doesn’t aim to be an accurate prediction of what a player’s shooting percentage will be at the NBA level, it’s still an interesting visualization of two things: (1) the range of outcomes for every player, and (2) the many factors that go into projecting a player’s shooting.

When trying to project a player’s outcome at the NBA level it can be difficult to avoid biases, and when it comes to shooting, we (scouts) might focus only on the positive statistical indicators for the prospects that we like, while dismissing the prospects that we’re lower on, lasering only on the concerning indicators.

If this exercise is worth anything, it is as a way to gain a historical perspective on a current player’s shooting numbers, reminding ourselves of how previous players with similar shooting profiles coming out of college fared at the NBA level.

Alright, the next part is likely the reason why you’re reading this. Here are the Shooting Similarity profiles for every NCAA player in the top 30 of our big board.

Final Thoughts

After seeing the results of the exercise, the numbers might seem a bit pessimistic, with only three players (Sheppard, McCain, and George) with an average of over 36.5%. However, it’s a good reminder of how very few players develop into that elite-shooter echelon at the NBA level.

At the end of the day, again, take this exercise as a reminder that shooting outcomes, just like the outcome for every other skill at the NBA level, is never set in stone. When we’re talking about projections, we need to realize that not all roads lead to Rome, but different prospects have a different number of roads that lead there. Take this a visualization of the roads that lead there and the roads that don’t.