Finding the Most Clutch NBA Player of the 2020–21 Season Using Python
A game-winning or -tying buzzer-beater usually allows a player to come into the conversation of whether or not he’s ‘clutch’. However, being clutch is much more than that. It’s more about stepping up in the final, pressurising moments in the game and leading your team to a win. I previously sought the most clutch NBA team of the 2020–21 season using advanced game stats. I now want to look at individual players and see who’s been able to display his ‘clutch-gene’ this regular season. The definition of clutch I would be following for the rest of this project is: a player’s ability to perform offensively (through points and assists) or defensively (through blocks, steals or rebounds) in the last 5 minutes of a game when the point differential is within 5.
In the previous article, I started out with finding the most-scoring team in Q4 of the regular season. However, over the course of this article, I will be evaluating certain stats — such as the offensive and defensive rating, as well as the True-Shooting Percentage — and further developing my own strategy in order to judge a player’s clutchness. I will only be using stats from the regular season, as the Playoffs haven’t been wrapped up yet.
The GitHub repo of this project can be found here.
Data Used
As always, I’ll be using the Stats NBA API for this project. You can check out data.py in the repo to see how I came across the basic and basic+advanced clutch data. I’ve also included .csv files showing the data I pulled. The further .py files were used to find advanced stats, as can be seen.
Stats to Use
Over this section, I’ll be going over my process and thoughts in evaluating the most fitting metric to find the clutchest player of this season.
Percentage vs. Volume
A perennial debate is that which concerns what kind of statistic is a more truthful representation of a player’s abilities: the percentage of a metric or the sheer volume. For instance, consider the following figure:
On the x-axis are the average points scored by certain players. On the y-axis is the true-shooting percentage, which attempts at measuring a player’s efficiency at shooting the ball. A clear-clutch player, Damian Lillard leads the chart both in average points made (4.8)and TS% (71.6). Judging the other players, how would one compare Bradley Beal to Jamal Murray? Beal gets more points (4.7), but Murray is more accurate with his shooting (68%). Who would then be more clutch? This disparity between the percentage and volume is something I’d have to answer.
A few stats I found that could be useful:
- Offensive and Defensive Rating (OFFRTG, DEFRTG)
- True-Shooting Percentage (TS%)
- Effective Field-Goal Percentage (eFG%)
Offensive/Defensive Rating
Perhaps the most convincing metrics, these are complex instruments to understand a player’s points generated and points defended. I, however, will not be using the offensive rating in this project. I’ll probably write an article later on talking about why these aren’t super accurate, as the same could take a lot of time. The defensive rating seems simple enough, and will be used further.
True-Shooting Percentage & Effective Field Goal Percentage
I will also not be using these (perhaps this is an overdue rebellious phase), for reasons that can be outlined easily:
- The formula for the former is:
The issue, clearly, is that this doesn’t take into account three-pointers.
- Formula for the latter:
This does take three-pointers into notice, but I don’t really understand how they carry a scalar value of 0.5. Further, it does not refer to free-throw attempts. I do think clutch free-throws are super important (Simmons, George showed that this postseason).
Weighted Field-Goal Percentage
The solution to this is the Weighted Field-Goal Percentage, which I found here:
This takes into account all the points that a player is able to produce through scoring. It does not, however, look into account assists. I’ll try to adjust that later. (Note: the metrics in the numerator are the made goals/throws)
Here are the top 10 players this season based on Weighted Field Goal Percentage (further multiplied by points). These could also be termed as the top 10 offensive players this season in the clutch.
A few surprises, Gilgeous-Alexander and Sexton made it onto this list. I will be using the WFG x PTS metric as the offensive score to evaluate a player’s scoring ability in the clutch.
Defensive Score
For this, I’ll simply be using the already prevalent Defensive Rating stat. Other than for its simplicity, this stat takes into account only a player’s steals and blocks — exactly what I’m looking for. Contrastingly, the Offensive Rating had the same limitations that other offensive stats I evaluated did.
As done before, here are the top 10 2020–21 defenders in the clutch, based on their defensive rating:
Two Clippers? I verify the validity of this list and metric.
Top Overall Players
Now that we have the stats ready, and have found both the top offensive and defensive players, it’s only logical to find the top players overall. A few ideas I had to compute the total score:
- Multiply/Add the two scores: The issue with this solution is that these two scores have a very different range, and we would thus have to give them weights somehow. I tried to divide the Defensive Rating by 100 and add it to the Offensive Score, but wasn’t really satisfied.
- Taking an average of the two scores: Same issue as before. I found a similar pattern when I found the sum, product and average of the two scores.
- Weighted Scaling of the two scores: I finally decided to scale the two scores using the MinMaxScaler. This basically gave each particular score a rating between 0 and 1, and I then added them together. However, I took the unilateral decision (it is, after all, my blog) to give the offensive score a weightage which is double that of the defensive score:
overall_score = offensive_score * 0.66 + defensive score * 0.33And so,
Conclusion
Pretty happy with this list. When I think of ‘clutch’, I think of players like Harden, Lillard, and Murray. Gilgeous-Alexander is probably the only player in this list that I didn’t anticipate being on there.
The offensive scorers were the ones that I knew about for the most part. The defenders, not so much. Thus, I’m glad to have found that out through this research. The overall score giving more weight to offense might not seem like the most mathematical idea, especially since offense being twice as significant as defense can be debated. But, as noted, that was a judgment call for a problem that didn’t have an easy solution.
Future Work
I had the idea of including an ‘impact score’ in this clutch score. This would inculcate the win/loss ratio and usage of a player. This would’ve helped in assessing a player’s leadership in the clutch. However, this score could be so interesting that I’ll go over it in another article.
