Friday, February 27, 2015

Let them handle their business – a case against over-helping

Note: I was first thinking about making it a longer post, but then re-read this post by Nick Restifo and realized that he made already most of my points (btw: @itsastat is a great Twitter handle). So I'll keep it in short form.
Note2: After I finished writing, I realized that it became a longer post :D

Hi everybody,
for the last week, I was playing around with data scraping and visualization. In my opinion, the use of clustering and heatmaps is a great but underused way to get a first impression and overview of data. Especially, as there are no numbers needed to understand the results (yes, I'm looking at you Chuckster!).

Anyhow, I was looking at the NBA.com playtype data to see if there is any rhyme or reason in what makes teams good. The Synergy playtypes are very interesting in those regard that the word playtype is used very loosely. The following is in my opinion very important: Those playtypes that they use can be separated into three and a half main subcategories:

1: Active Playtypes: Isolation, PnR Ball Handler, Post Up, (Misc)
One player receives the ball in a more or less neutral position and tries to get into an advantageous position. It is really hard to prevent the initiation of these plays.
1.5: Semi-active playtypes: Hand-Off, Off-Screen
The defense can sometimes prevent the initiation of the plays (which can be more costly than allowing the play itself). Note that they produce on average slightly more points per possession than active playtypes, but they also happen less often (combined 10% of plays). This my be due to the fact that they are more complex (more actions/players involved)
2: Passive playtypes: Spot-Up, PnR Roll man, cuts, putbacks
These are plays that can be avoided. Note that 'can be avoided' does not necessarily mean that the defense is at fault if they happen. Take this cut by DeAndre Jordan (I guess it's a cut) one of the league leaders in cuts. You can either let Blake dunk directly or hope that he somehow botches the pass to DeAndre
3: Transition. You could argue that transition is more a result of offense than of defense. But that's a different story


This separation is important to understand the basketball side of the analysis. You can hardly avoid for a team to initiate the Pick and Roll, but you can decide if you want to risk keeping the ball in the ball handlers hands (and giving him whatever he gets) or if you want to force the ball out of his hands and risk a spot up shot, or the Roll Man or another cutter to beat you.
I will now show you a plot that will blow your mind. Are you ready Chuck?! No worries, I will explain them directly afterwards (click to enlarge).

What you can see here by playtype is differences in 1. Points per possession for team offenses by plays and 2. play frequency for team defenses. I normalized the plots by subtracting the PPP or respective frequency means for each play. The first column shows you the offensive respectively defensive ratings of the different teams.

What you can see on the first plot:
For Off Rating & Points per possession, green means good. So, here you find almost all the good teams in the lower half. As you can see, these teams are all above average in points per possession for PnR Roller. BUT you can furthermore see that those teams that are not as good in offensive net rating (Toronto, Orlando, Boston and (up to now) San Antonio) do not produce as good with their ball handlers. Furthermore, these teams are mostly good at HandOff and SpotUp shooting (as noted by Nick). Off course this is a chicken egg problem, as good SpotUp shooter allow for easier rim attack and Handjobs are basically more fancy pick and rolls (just wanted to check if you are still awake).
And off course, better PnR handler allow for easier cutting and rolling, so there is most likely a trickle down effect. (Side note: I'm looking at some other stats that indicate that high usage ball handler always have/need a midrange game. So don't worry, the midrange game will never die.)

What you can see on the second plot:
For Def Rating, red means good. For Frequency, green means more. What we can see is a big cluster of mostly good defenses on the upper half. Of course, this separation is not perfect, as 5 of the 13 teams are below average. But three of them (Sacramento, Minnesota and the Lakers) allow a lot of transition baskets, which does not really help. Side note: the same for Houston, Toronto and Phily, only they allow 0.1 points less per transition. I guess that helps...)
Anyhow, in my opinion, the plot shows that teams that focus less on the ball handler and more on avoiding the SpotUp shooter have in general a better defensive net rating.  The correlation between those two options is pretty obvious from the plot (given that Handler frequency and SpotUp frequency colors are always the opposite). The R^2 for PnR frequency and Spot up frequency is 0.654 (for Chuck: That means they are super highly correlated. What, you are asking what correlated means? Never mind...)
So, to conclude: Teams need a good PnR game and optimally ball handler to be good on offense. And teams are in general better on defense if they take away the Spot up shooter I rest my case...

ALMOST: Of course, what I described here is not Houston Rockets science and teams know all this (interestingly, the Rockets seem to disagree with my points). On the one hand a defense if of course bound by its personnel. On the other hand, it is interesting to note that increased frequency of PnR Handler plays strongly correlates with increased points per possession for these plays. At the same time, decreased frequency of Spot up shots, does not lead to that much of a decrease in PPP efficiency (Figures below). So, from a game theory perspective, you surely reach a breaking point, where transform a pick and roll into a layup line is not helpful (duh).

Here are the additional figures (let me know if you want more):

Frequency vs PPP PnR Handler: There seem to be two groups (over and under the red line). You should probably be worried if you are above this line.

Frequency vs PPP Spot Up shooter. No correlation. Interestingly 'stabilizes'  for most teams with a frequency above 20% (which fits the Seth Partnow article that once you shoot the 3 it becomes a coin flip). Also: Milwauuke and their tentacles...

Frequency vs PPP: If somebody knows the difference between LAL/SAC/MIN and HOU/PHI/TOR transition D: I'm listening.

Defensive PPP for different play types. If you find anything of interest let me know. Golden State does not defend the roller very well, but is killer in everything else.

Offensive Frequency for different play types. Also no rhyme nor reason (in my opinion). I guess it's partly your style/preference, partly what the defense dictates.


In any case, I hope you like this way of presentation. It's mostly R heatmap.2 with some polishing. I'll use it more often now that I got more into data scraping :)

Have a good weekend (especially you SSAC15 people!),
Hannes

No comments:

Post a Comment