Which Offensive Rate Stats Stay Consistent When a Player Changes Teams/Roles?
Posted by Neil Paine on December 23, 2010
You'd better read this post from August if you haven't yet.
In it, I looked at the year-to-year correlation coefficients for various offensive rate stats (TS%, AST Rate, TOV Rate, FTA/FGA, & OReb%) when a player changed his role in the offense. Essentially, I concluded that offensive rebounding and assists are relatively immune to changes in a player's possession usage, foul-drawing & turnover avoidance are less immune, and scoring efficiency is the most prone to fluctuate with a role change.
Today I wanted to expand on that post by adding another variable into the mix: changing teams. Other than the new variable, though, this study's format is basically the same as in the first post -- except I used touches per minute rather than possession % to define a player's role, and I added Dean Oliver's Offensive Rating (ORtg) into the mix.
Here's the setup: Once again, I found every player from 1974-2010 who was between 24 and 34 years old and played at least 1,000 minutes in back-to-back seasons. I then sorted those players by the absolute change in their touches/min, and took the top quartile as my sample of players who changed roles. I also isolated players who played for a different team than they had the previous season, forming four groups: players who changed team & role; players who changed role only; players who changed team only; and players who changed neither team nor role. Finally, I ran correlation coefficients on the year-to-year offensive rate stat performances for each group:
| Year-to-Year Correlation | |||||||
|---|---|---|---|---|---|---|---|
| Type | # Plyrs | ORtg | TS% | AsR | ToR | FTr | OR% |
| Changed Team + Role | 414 | 0.617 | 0.593 | 0.792 | 0.715 | 0.742 | 0.924 |
| Changed Role Only | 717 | 0.695 | 0.686 | 0.875 | 0.756 | 0.828 | 0.943 |
| Changed Team Only | 779 | 0.563 | 0.556 | 0.963 | 0.735 | 0.811 | 0.930 |
| Didn't Change Team or Role | 2611 | 0.719 | 0.706 | 0.974 | 0.802 | 0.842 | 0.944 |
The results:
- Shooting/offensive efficiency is actually far more impacted when a player changes teams than when he changes roles. This suggests that a team's system, coaching effects, and teammate interactions play a much bigger role in determining shooting percentages than "skill curve" effects.
- Assists are more dependent on role than team -- for obvious reasons, how much you have the ball in your hands is a major factor when it comes to how often you set up your teammates.
- Turnovers, fouls drawn, and offensive rebounds are all more team-dependent than role-dependent, but the differences are nowhere near as drastic as those observed in shooting efficiency or assists.
- For players with no major contextual changes, assists and offensive rebounds are relatively stable; foul-drawing and turnover avoidance are less consistent; and scoring efficiency is the least consistent of all. This mirrors the findings of the original study.
- Finally, it bears mentioning that even at its least consistent (TS% for players who changed teams), NBA player performance is way more predictable than that of the NFL quarterbacks Jason Lisk looked at in the football study that inspired these posts.
December 23rd, 2010 at 9:48 am
Great info.
Aren't skill curves based on the assumption that your team/coach is "maximizing" your skill set within the offense? It would make sense that changing teams causes the model to break down in this regard since the player and the team are still making adjustments to make it all work.
December 23rd, 2010 at 10:42 am
I'd like to see DREB% in here, since the common thought is that rebounding is the easiest skill to translate from college to the NBA or whatever, but by guessing I'd assume the rate is similar and slightly lower than OREB%'s correlation.
December 23rd, 2010 at 11:36 am
"For players with no major contextual changes, assists and offensive rebounds are relatively stable; foul-drawing and turnover avoidance are less consistent; and scoring efficiency is the least consistent of all."
This may be true, but the lower correlation for scoring may just reflect the much lower variance in the statistic. If you measured the average absolute y-t-y change in each stat, measured in point value, is it really true that scoring efficiency changes more? It could be that scoring is just as stable, but because players cluster so closely together the correlation is lower.
December 23rd, 2010 at 11:57 am
And a question: when a player changes teams, is there a correlation between the change in TS% and the difference in the two teams' TS%. That is, does moving to a higher efficiency team improve (or lower) the player's efficiency?
December 23rd, 2010 at 3:22 pm
This is awesome! Great work (as always)!
December 23rd, 2010 at 6:33 pm
Interesting to see FT change so much when changing roles and teams. I wonder if this is because players that are more volatile at FTs are more likely to change teams and roles?
December 24th, 2010 at 10:14 am
Re: Guy - Interesting comments/questions, ones which deserve their own post when I get the time. I'll try to give those a shot after the holidays.
December 26th, 2010 at 5:22 pm
I personally think that touches would be affected the most. Take Lebron James for instance. Great Post i never really thought about this issue
December 28th, 2010 at 8:28 am
Neil/7: Great, look forward to the future post.
Let me also toss a question into the mix regarding this point: "Shooting/offensive efficiency is actually far more impacted when a player changes teams than when he changes roles. This suggests that a team's system, coaching effects, and teammate interactions play a much bigger role in determining shooting percentages than "skill curve" effects." This seems very important if true, and may well be right. But I worry that significant changes in a player's role could be the result of, rather than cause of, changes in their shooting efficiency. If improving players are allowed to increase usage, it will mask a "usage tax" on their efficiency. If declining/aging/hurt players reduce usage, the reverse would happen. So I think isolating the impact of usage requires looking at essentially involuntary changes in usage. (Which is what Eli Witus tried to measure in his lineup study.) All that said, your evidence seems pretty strong.
I also wonder whether correlations are the best way to compare your four groups. Because the 4 groups presumably vary in terms of sample size, average age, and, most importantly, the talent variance within each group, correlations might differ for multiple reasons. I think it might be better to measure the average absolute change in each stat (e.g. does TS% change less for role changers than team changers?). And (still in the category of suggestions for "work someone else should do") it might be fruitful to divide role changers into two groups: increased role and reduced role. If sample size permits, that might clarify the picture a bit.