Overall Offensive Load Carried (SportVU Aided)
(Update 2/1/14 Revised numbers and thoughts here, google doc with leaguewide stats here)
(Update 2/17/14 Revised again here, google doc with leaguewide stats through ASB here)
Typically, we simply refer to the size of a player's role in an offense by noting their usage statistic. And as far as it goes, this is great - we know that on a "perfectly balanced" team, all 5 players will have a USG of 20%, so we can tell some things about how much or little a player shoots by comparing their usage to this average. Under 20% usage players tend to be either limited spot up shooters or bangers with little game outside of putbacks. Above 25% represents primary options and above 30% are the true ball-dominant chuckers.
However, usage presents a slightly distorted picture of how offense (especially a high functioning offense) tends to work. If a point guard splits a double team on a pick and roll, euro-steps past the help defender, draws the rim protector into the air and drops a nifty pass off to a grinder big man who happened to be cutting to the basket for an offensive rebound, that possession was "used" and that shot "created" by the big man. As with many basketball statistical questions this gets us into the difficult area of parsing individual credit from events which are undoubtedly positive on the team level.
I'm not going to untie that whole knot here. In fact, I'm going to basically skipping the question of whom to credit by saying "why not both?" I think it meshes with simple observation that there is no reason a basketball possession has to be a one man show, though it certainly can be.
This has also allowed me to express turnovers in a rate stat which encompasses both shooting and play making - traditional TOV% tends to overcredit shoot first players as the only "positive" outcome in terms of reducing TOV% is shooting the ball. Merely not turning the ball over (while for example passing to a shooter) is a neutral event which has no effect on TOV%, which has the counter-intuitive effect of making the players who presumably take the best care of the ball (pass-first PGs) look like the most profligate with it. To put it another way, a measure of taking care of the ball that makes prime Steve Nash look bad in terms of handling the rock has serious issues.
Since the data seemed to naturally break down this way anyway, I've separated players into "PG's, Wings, and Bigs". Their is no position adjustment involved, it's simply a grouping of players. There are 43 PGs, 93 wings and 72 bigs included.
For the data, I used all players who through the games of Monday the 6th had played at least 16 games (more or less half the games played) and at least 20 minutes per game. This left 208 players. I think it's probable that this biased the overall league averages are probably slightly lower than the sample I used as I think minute distribution is roughly rational - better players play more. Better players handle the ball more, so the guys not selected in the sample probably handle the ball/shoot less on a per possession basis, but that's neither here nor there as the numbers aren't scaled to any average.
The three columns are
- OffLoad - the percentage of plays the player is "directly involved in" either as shooter, passer or "hockey assister" while on the floor. Average across the sample was just under 35%, which aligns with about 1.73 players "involved" per play on a team level which feels slightly high, but not ridiculous. Would DEFINITELY accept suggestions for a better name for this one...
- TO/Load - Reconfigured TO% representing percentage of turnovers in total plays including shots AND assist attempts. Average for the sample was 8.76%
- PlayEFG - Expressing the efficiency of possessions in which the player is involved in a shooting percentage-style notation. Alternatively, double the number for a PPP/ORTG style number. More on this below, but I want to strongly caveat the tables below that I don't think this number should carry much weight in evaluating individual players.
First the PG's sorted from highest load to lowest
Some analysis and conjecture about the results.
- OffLoad is purely descriptive, in a vacuum a higher percentage is not better or worse based on this data, it just is. Better offensive players tending to carry higher loads is a result of coaches, players and execs having eyes.
- I like the way TO/Load tracks by player category, as it stands to reason that the guys who play PG are generally better at taking care of the ball than guys who play wing than the guys who play big becuase that's part of why they play those positions.
- Raw USG substantially underrates the role of a PG or similar creator in a functioning offense.
- Best guess is that players as a whole are overcredited with involvement and more so in terms of PlayEFG by generous assist giving. TheNick Van Exel homer assist probably doesn't count as a "chance" in SportVU, so it only gets recorded if the shot goes in. So much in the same way getting fouled while shooting only helps a players FG% because it only counts as an attempt if it goes in, I think the EFG's are slightly high because of that factor.
- That said, I don't think it matters terribly because A) it's a relative stat Luke Ridnour's .493 doesn't mean much except as compared to the league average of .536 and/or Nate Wolter's .471. But more importantly B) I don't want to put much stock at all in EFG as far as giving credit to the individual players as it's to a large degree a reflection of teammate quality rather than individual player talent to the extent player's "PlayEFG" diverges from TS%
- Some interesting individual player comparisons, Russell Westbrook and Reggie Jackson jumped out at me:
- Top and bottom 20 players in terms of turnover rate:
| || |
- All kinds of other interesting things show up here - Ryan Anderson has potential assists on just over 4% of New Orleans possessions while he is on the floor. I almost have to think this represents a failure in the design of their offense given how hard teams have to close out on him. EDIT: I actually misread my data, he actually has potential assists on only 2.91% of possessions he's on the floor. The only other players under 3% are Andre Drummond and Samuel Dalembert, so the point still holds I think.
I haven't attempted to perform any advanced statstical methodology on the data, first because my knowledge on that front gets me to linear regression, barely. But beyond that I worry at times that for all but the very few people who both require the exactitude provided by and either have the ability themselves or in their organization to translate the numbers into an understanding of what's happening on the court and vice versa, I worry that basketball analytics at times becomes too much about the numbers, and I wanted to try and come up with a measure that is both broadly understandable for fans with decent numeracy but not advanced mathematical knowledge and as first-pass accurate. Hopefully this has succeeded on at least one of those goals.