Skip to main content

What makes a good NBA player? (part 2: rookie combines)

Hi again!
Pearson's correlation coefficients

In part 2 of this series on what makes a good NBA player, I looked at which combine statistics like max vertical leap, lane agility, and bench press in combination with physical attributes like height, wingspan, hand size, etc. correlate most with rookie success in their first year, as evaluated by advanced metrics like offensive/defensive rating, assist %, rebound %, player impact estimate (PIE), etc. You can find a jupyter notebook with all of these Pearson's correlation coefficients at: github link. Feel free to play around with it to see if you can find any interesting correlations!

Big takeaways:

1. Jumping ability =/= rebounding ability: Conventional wisdom suggests that good jumpers are probably good at getting rebounds. However, I found that max vertical leap has a significantly negative correlation with rebound %, for both offensive and defensive rebounds. In reality, it's the guards that have the best jumping ability in terms of max vertical leap, which has significant positive correlations with net rating and assist %.

2. Strength, height, weight, wingspan, reach, and hand size = rebounding ability: True rebounding ability does not come from jumping - instead it is strongly correlated with bench press, height, weight, wingspan, reach, and hand size. Interestingly, it is also negatively correlated with agility (lane agility, three quarter sprint). Big guys get rebounds, and they are not typically agile -> no big surprises here. Interestingly, the strongest correlation is actually between wingspan and rebounding ability, not height or reach.

3. Strength, height, weight, wingspan, reach, and hand size increases shooting percentage: The same big guy attributes also correlate with effective FG % and true shooting %. This one likely just means that big guys are better at finishing at the rim.

4. Agility, not strength = assist ability: All of the big guy attributes are strongly negatively correlated with assists %. Instead, assists are correlated with lane agility and three quarter sprint. This may just be a product of guards fitting a prototype and having more time handling the ball. Big guy attributes are also negatively correlated with usage rate.

5. Offensive/defensive rating, PIE, etc. have no significant correlations with tracked stats: Last, but certainly not least, the metrics for overall player effectiveness like offensive/defensive rating (net rating) and player impact estimate (PIE) do not have significant correlations with ANY of the tracked stats. In other words, the drills that the draft class perform and their physical measurements have absolutely no bearing on their future success in the NBA aside from correlations with rebounding ability and assist ability (big man and guard stats, respectively).

Maybe Charles Barkley is right - maybe analytics can't capture the intricacies of the game...? Nah, I think I am simply running into the same problem that Daryl Morey and the Houston Rockets first ran into when they started their big campaign into analytics - the stats that are available are not good enough. The current stats are designed around prototypical NBA roles - stats like rebounding and assists only usually apply to forwards/centers and guards, respectively. Stats like shooting percentage are skewed towards the big guys that can dunk and finish at the rim. In today's NBA, where big guys need to be able to pass and shoot 3s, guards need to have court vision and finish at the rim, and forwards need to be able to guard all 5 positions, we need stats that break the meta and capture the potential of players to excel outside of their prototypical roles. In the next parts of this series, I will be exploring more unconventional stats and evaluate their effectiveness in forecasting future player performance. Most of this data will be generated from college players and their on-court events. Stay tuned!


Popular posts from this blog

The best and worst defenders of 2014-2015

Welcome to Money B-ball!

I will be using data on NBA player statistics to model player performance. For the first post of this blog, I used data from the 2014-2015 NBA season to evaluate the effectiveness of NBA players in defending shooters by looking at how much they reduced the normal shooting averages of NBA shooters.

In this analysis, I found that the top 10 defenders in reducing total shooting averages are:
1) Andrew Bogut (-9%)
2) Rudy Gobert (-8%)
3) Tony Allen (-8%)
4) Andre Roberson (-7%)
5) Roy Hibbert (-7%)
6) Draymond Green (-6%)
7) Serge Ibaka (-6%)
8) Anthony Davis (-5%)
9) Joakim Noah (-5%)
10) Trevor Booker (-5%)

There's a lot of familiar names on this list - notable shot blockers like Rudy Gobert and Serge Ibaka. Most of these players are power forwards and centers, with the exception of Tony Allen and Andre Roberson, who are both defensive specialists. Andrew Bogut reduced the shooting averages of players he faced by more than 9%!

Here are the bottom 5 worst defenders:
1) Jo…

What makes a good NBA player? (part 1)

Hi again!
Today I will explore what physical attributes contribute to a successful NBA player. When we watch basketball, we sometimes hear the announcers and analysts talk about stats like wingspan, height, and hand size. How do these attributes contribute to a good player? Are there any that are particularly important? For this part one, I have scraped data from to look at the distribution of physical attributes from the 2010-2017 NBA combine. To use my simple data scraper, check out my github:, or email me for a tutorial (scraping data from can be pretty tricky).
First, I distill the data on these pages (!?SeasonYear=2010-11) to the relevant features, such as by calculating a hand area feature in place of hand length and hand width (units are Imperial, pounds, inches, and inches^2).

Now let's plot a notable correlation:
Wingspan vs. height: both given in inches, red line guides …