NBA Skill Set: What Really Matters

We’re not crashing the boards! You’ve got to take better care of the ball! We’re letting them take too many wide open shots! All NBA fans and especially those who’ve ever played for a team—whether it be in college, middle school or whatever—have heard these words of advice screamed before. Coaches aren’t stupid, they can see when their team’s being beat in a specific aspect of a game and need to make adjustments. But which are the most important?

If you’re team’s not rebounding well, but at the same time not covering the perimeter shooters how should they adjust? A 2-3 zone to protect the boards but remain limited around the arc or a 3-2 zone to shut down the long range shooters but abdicate more offensive rebounds? Screw it, they should probably just play man-to-man, but still the question lingers.

In order to feed my statistical curiosity I conducted a quick study of all NBA teams in the previous six seasons. Taking team statistical data from both espn.com and basketball-reference.com I ran a regression to determine which skills (as a team) played the most important roles in determining the outcome of a game.

My first regression used the five more essential statistics (at least in my mind) for a basketball squad to predict my dependent variable: team’s winning percentage. The five independent variables were: FG%, Opp. FG%, D-Rebounds, O-Rebounds, and Assist/Turnover Ratio. These statistics measure a team’s ability to shoot well, prevent their opponents from shooting well, rebound on both ends of the court, and handle the ball both carefully and effectively. I altered these statistics slightly though, relying on adjusted field goal percentage for both the team in question and their opponents. Also instead of using the team’s average rebounds per game, which can be inflated or deflated depending on the number of shots taken and made by both teams, I used the percentage of offensive and defensive rebounds that a team pulled down with respect to every shot taken. A-T Ratio remained untouched.

Results were promising, generating a healthy 0.678 adjusted R-squared value and all the coefficients were significant and had the right sign. Here are their t-scores (the larger the absolute value, the greater that variables importance to the regression):
Adjusted FG% = 11.52385
Opponents’ Adjusted FG% = -8.005760
D-Rebound % = 4.384378
O-Rebound % = 3.763958
Assist/Turnover Ratio = 5.768974

These results are promising with a team’s overall shooting ability reigning supreme. Considering the strong t-scores from all these variables I decided to throw in three more variables that would complement the current skill set: Blocks, Steals, and Free Throw %. These traits I felt were less essential to a team’s overall success but were still admirable qualities that would improve a team’s chances of winning.

My second and more crowded regression was equally satisfying with all the previous variables and the newly added ones emerging significant and with signs corresponding to what we would assume in theory. Here are the t-scores for the second regression listed in descending order (and importance):
Adjusted FG% = 12.93219
D-Rebound % = 6.807323
Steals = 6.211744
Opponents’ Adjusted FG% = -6.184397
Assist/Turnover Ratio = 4.719551
Blocks = 3.177327
O-Rebound % = 3.082311
Free Throw % = 2.190980

There are some peculiar happenings in this regression compared to the previous one. The most astonishing change is D-Rebound’s leap frogging ahead of Opp. FG% which dropped to a surprising 4th place. Steals is awfully high on this scale, or at least much higher than I had anticipated. FT% and Blocks drift towards the bottom as suspected, but O-Rebounds lies fairly low as well. A possible reason for this is that a team who concerns itself with transition defense rather than recklessly going for offensive boards may be able to succeed adequately and therefore make O-Rebounds less essential.

The only other qualm I have is Adjusted FG %’s dominance over all other variables. I did expect it to be the most important variable, but twice as important as the next most critical statistic? Well, it at the very least signifies that it is very hard to compensate for poor shooting by heightening other aspects of your team’s game.

I ran the same regressions as mentioned above but with the team's expected winning percentage (generated via their points scored and allowed) as the dependent variable and found similar results. The adjusted R-squared value was slightly higher and each variable’s t-score increased by a varying margin. Actually it regressed back to my original expectations with Opp. FG % retaking the #2 slot with steals dropping to 4th. Also to my liking, offensive rebounds climbed above blocks. Here are the tallies:
Adjusted FG% = 14.48154
Opponents’ Adjusted FG% = -7.502249
D-Rebound % = 7.297942
Steals = 6.881690
Assist/Turnover Ratio = 4.743004
O-Rebound % = 3.613866
Blocks = 3.230449
Free Throw % = 2.470133

I was interested in seeing how these skills manifested themselves throughout the playoff atmosphere but found myself with a conundrum: there is no system for measuring playoff success (or at least none that I am aware of). WPCT in the playoffs is not apt because of the small sample size and each team does not play the same number of games. The number of wins is also incapable of determining the dominance of a team because it doesn't account for how many times they lost en route to their final ranking. I tried developing a point system for wins and losses but it was purely by trial and error and lacked aesthetic qualities. If anyone is aware of or can derive a system of rating a team's playoff performances (other than laundry listing every possible scenario in a ranked order) please bring it to my attention, as it could be implemented not only in this study but in another one I am working on.

As for this study, I am quite satisfied with the simple, yet useful findings. Sadly basketball is a far more fluid game than baseball so it is very difficult to quantify the cohesiveness of a team and how they interact together. We've all learned from Isiah Johnson that a real basketball team is not just a collection of numbers. Therefore it's hard to draw any further conclusions as to how a team should focus itself on the court or how one should be put together. That's not to say there aren't more inferences that can be made from these results, but we must remind ourselves of its limitations.

Feel free to draw your own conclusions—I’d greatly appreciate any feedback you may have—but remember that this regression is only representative of games played in the past six seasons. Do not use these results as indications that previous notions and standards for how to play the game were in fact wrong; rather they merely represent a shift in how the game is played today. If you have any questions, comments, or would simply like a copy of the data used to run the regression you can post a comment here or contact me via e-mail.