Per Minute Productivity Analysis: Don't Forget Opponent Quality and Different Scoping of Starter and Back Up Roles
-
In a recent thread about likely starters for the KU/Holy Cross game started by @benshawks08, some discussion about per minute productivity occurred that caused me to recognize at least part of why I tend to interpret who should be playing and who should be sitting somewhat differently than quite a few board rats. It caused me to try to distill the issue of analyzing per minute productivity.
Per minute based productivity stats are fine for analysis, but they have to be indexed for the differences in starting and rotating in. The starter, if he were a cornerstone player, has to start against all kinds of opponents. He has to go against the guys ready for the NBA and the green wood; against the 1ADs, 2ADs, 3ADs, 4ADs and 5ADs; against the best of whatever the opponent has to start. The guy rotating for the starter is often going up against the opponent’s lesser players, or going up fresh against the opponent’s tired starter. Against a long stack, like Duke, UK, and UNC, you can probably compare starter per minute productivity somewhat with rotating back up per minute productivity, because those long stacks are so deep, your back up is playing against a pretty good player. But against a mid major, or against an unranked major, there is probably no unadjusted comparison worth making, because the unranked major and the mid major are usually very thin after the first five, maybe first six guys.
The other adjustment that has to be made to get to a common denominator in comparing per minute production stats among starters and relievers is the situational nature of many relievers. When they are sent in, they are often sent in with a very narrowly scoped purpose. Self doesn’t want Diallo going in and being a first, second, or third scoring option right now. He wants him rebounding and trying to block and alter everything in sight. It makes his per minute scoring deceptively less than he is probably capable of. It is the same at many positions. The sub has a much narrower job description. It inflates some of his per minute productivity numbers and deflates others.
What most board rats fail to take into account when arguing for more time for back ups based on their per minute productivity numbers are: quality of opponents encountered by back ups; and the differently scoped roles of back ups.
-
@jaybate-1.0 It was either last year or two years ago Spurs head coach Greg Popovich was asked about analytics. He laughed and said he doesn’t have time for that and doesn’t believe in them. That someone who isn’t trying to save his job doesn’t versus a guy trying to create a job. Something on those lines, much like what Barkley said.the article continued that even though Pops didn’t like the analytics and such the Spurs were statically saber metrics wise the best team in the NBA. This is from a coach who doesn’t care about the stats as he will pull his starting 5 and out in 5 subs in not only for rest but to save a timeout and coach his team. Show them what they are doing wrong. His system seems to work pretty good.
-
@JRyman I think Self would love the comparison to Pop! Barkley? Maybe not so much!
@jaybate-1.0 I think the roles of the bench players is a huge point. It often seems like Self subs them in to get something done the starter isn’t doing. And if they cannot impact the game that way then back to the starter.
This seems to be the focused analysis HEM observes in Self of certain players. To me, Self has an idea of what role suits each player best based on what he sees in games and practice. He puts them in we he needs that role filled. If they don’t perform that role when called upon, they miss their opportunity.
Self has acknowledged he needs to provide more opportunities but clarifies that he won’t give up wins to do it. So if players want PT, know your ROLE and EXECUTE it.
-
@benshawks08 said:
@JRyman I think Self would love the comparison to Pop! Barkley? Maybe not so much! .
Self has acknowledged he needs to provide more opportunities but clarifies that he won’t give up wins to do it. So if players want PT, know your ROLE and EXECUTE it.
Self would probably like a Barkley type on his team though.
Know your ROLE and EXECUTE it. Well said. Well put.
-
Aww…come on…Chucky Cheese is a hoot…
-
@JRyman we love Barkley too!
-
@jaybate-1.0 So what you are saying is what everyone assumes (I would think) – that the bigger the sample size, the more valid the stats?
-
@HighEliteMajor That is generally true about all statistics but I don’t think that is the point @jaybate-1.0 is making. To me he is arguing that there are significant complexities in measuring the worth of bench players and starters that limit the usefulness of statistical comparisons of players in different roles. That was my interpretation.
@JayHawkFanToo Yeah, he is kinda funny! He sticks his foot in his mouth a little too much for him to be anything like Self who is so calculated in his communication with the media.
-
Pop isn’t big on analytics, but has always gravitated towards players that are analytically sound - Robinson, Duncan, Parker, Ginobili, Leonard, even LaMarcus Aldridge. Pop may not pay attention to analytics, but his teams aren’t exactly bucking the analytical trend. If anything, they fall right in line with it.
When nobody else in the NBA had use for Bruce Bowen, the Spurs turned him into a player that was an elite defender and hit corner threes. What’s the most valuable shot in the NBA according to analytics? The corner three. Bruce Bowen went from being a non-factor to one of the most valuable role players in the NBA because he could hit the corner three. The Spurs may not have used analytics to acquire him, but he’s an analytics masterpiece.
I think Pop understands what analytics try to get at, which is ultimately efficiency. But Pop’s system is already so efficient that he’s doing analytical things without relying on the specific stats.
I doubt many coaches can say the same. Maybe Phil Jackson in his heyday.
-
Sample size matters, of course.
But sample size is a third element to consider.
The samples, assuming adequate size, also have to be sampling similar universes to make inferences significant between the sample and a member of the universe.
For example, bolts and nuts work together as fasteners. If we want to sample the failure rate of such fasteners, we can sample the universe of fasteners and at the right sample size with randomization of sampling, we can assume a normal distribution (unless something tells us otherwise), we can draw some pretty reliable estimates of how many fasters will fail in a given quantity of fasteners produced.
But if we want to know about the failure rates of bolts, we had better sample bolts to make inferences about bolt failure, rather than sampling nuts to make inferences about bolt failures.
If we were to sample nut failures to learn about bolt failures, it just wouldn’t help to draw a big sample of nuts.
-
@benshawks08 said:
there are significant complexities in measuring the worth of bench players and starters that limit the usefulness of statistical comparisons of players in different roles. That was my interpretation.
That’s it in a nut shell. But I also think analysis that takes into account those differences (e.g., that looks at some aspect of performance that would be expected to be quite similar, or alternatively, that looks at some aspect that one would expect to be quite different and finds insignificant difference, or alternatively a counter intuitive difference) can be more or less useful in adding insight into how a sub might perform in a starting role.
For example if Diallo were scoring a little more per minute played as a sub than a starter is scoring, and KU is playing a team 2-deep with OADs at the position, and Diallo is expected to score as much as the starter, then we could infer that Diallo might do a little better than the starter and so maybe he ought to get a shot.
But if Diallo were scoring a little more per minute played as a sub than a starter is scoring, and Diallo were doing it against a very inferior sub, then we might leave the status quo as is, because we might infer that Diallo would probably score quite a bit less against a better opponent.
Or suppose Diallo weren’t supposed to score against his man much, rather he was just supposed to guard and block. And let’s suppose the guy ahead of him was scoring a lot, but giving up a lot of points and not getting any blocks against guys we think Diallo could at least alter shots on. Well, despite Diallo not scoring much, we might give Diallo a shot, because we think Diallo has good scoring fundamentals, and we think Diallo could guard and alter as well, and he could definitely rebound better.
And so on.
-
@justanotherfan said:
I think Pop understands what analytics try to get at, which is ultimately efficiency. But Pop’s system is already so efficient that he’s doing analytical things without relying on the specific stats.
I thought this observation by you was worth calling attention to.
In achieving efficient systems, one is iterating between what one hypothesizes is an efficient design, and at the same time sampling indicators of its efficiency. It is a reinforcing loop. Adjust the system. Read the stats. Adjust the system. Read the stats. The iteration helps you get familiar with the dynamics of the system. Hopefully you keep moving toward more efficiency, assuming you are correctly conceptualizing efficiency. At the point of diminishing returns in system adjustments, without sufficient winning effect, then you have to start thinking about adopting a new system.
-
ESPN has an article where Pop is talking about the 3 pointer and it “like a circus.” Ha.
-
@Second-Prize saw that. “Why not have a 5 point shot or a 7?”
-
So long as the proportionality is sufficient to disperse shooting back across the floor, whatever works.
-
@jaybate-1.0 That was part of Popovich’s quote about the 3 being a circus.
-
Popa doesn’t worry about the big numbers in his contract turning the NBA into a circus.
-
BG averages .95 PPM, Cheick averages .70 PPM. I am pretty sure BG gets our opponents BEST defender the second he checks in to the game. Cheick also demands plenty of attention from opposing coaches and players. Lets assume these 2 guys received starting minutes and lets also assume they could sustain their averages with the increased minutes? BG would be averaging 28.5 PPG on like 10 shots lol, and Cheick would be sitting on 21 ppg in tip in’s and transition dunks! It could happen LOL
-
Of. Course it could happen, but until the fouls and TOs come down he would only be netting about 11.
-
An important consideration in per minute stats is whether the player is playing rotation level minutes. That’s what’s made reading Mick so difficult – until the WUG.
When a player plays regularly, it all balances out. They get the Michigan States and the Loyolas.
This season we have a pretty nice cross section of teams so far – from Chaminade to MSU.
But really, I think what everyone will see when you study the per minute stats and opponents over the last few seasons is that the opponent is largely irrelevant. A guy might do great vs. Iowa State, but suck against Texas Tech.
I looked randomly at Selden last season – best scoring games were vs. Florida, ISU twice, and Baylor. Bad game vs. TCU. A good one vs. KSU and TT, and a bad one vs. both. I think you’ll see that the inconsistency is consistent throughout. Three of Perry Ellis’ worst games were TCU, Lafayette, and Temple. I recall Wiggins having a pretty bad game vs. Texas Tech and a great one against Iowa St. Why? Who knows? But again, you will see that throughout.
That is not to suggest that stats accrued vs. crème puffs might not have less value than the good teams.
But look at our dilemma. Harvard was apparently the Boston Celtics to our Kansas squad, where Holy Cross was a pud. Which team is worse? Is Harvard really a crème puff?
Most players are going to have ups and downs. The per minute stats, I think, are great evidence for rotation level guys, and better indicators with each successive game and season (coupled with he eye test). For my money, the only guy really arcing upward on the eye test is Wayne Selden - the only guy where the pattern seems to have changed. Otherwise, past performance is most indicative for future performance. That does not mean they won’t improve, which players obviously do, it just means we’re in relatively the same ballpark as we were in prior seasons.
-
The anecdotal cases prompt me to these additional hypotheses.
Hypothesis 1: Against the lesser teams, Self doesn’t rely nearly as heavily on his foundation players, and instead tries to develop his lesser starters by playing through them more.
Hypothesis 2: Lesser teams have to pick which poison to let kill them, and opposing coaches decide if they can’t win the game at least they will try to shut down a star (not entirely rational basketball strategy, but rather subjective value driven).
Regarding Hypothesis No. 2, I recall reading about the head coach of Sienna, Jim Patsos, who used to be the head coach at Loyola Maryland, before G.G. Smith. On his wiki page, there is a fascinatingly frank quote by him about why his then Loyola team double teamed Davidson’s Stephen Curry all over the floor and despite being beaten by 30 points.
“On November 25, 2008, Patsos and the Greyhounds double-teamed Davidson All-American Stephen Curry for the entire game, leaving his other three players to face the 24th ranked Wildcats in a four on three game. Davidson won by 30, while Curry stood in the corner during most possessions. Commenting afterwords, Patsos said: “We had to play against an NBA player tonight. Anybody else ever hold him scoreless? I’m a history major. They’re going to remember that we held him scoreless or we lost by 30? I know the fans are mad at me, but I had to roll the dice as far as a coach goes. I’m not some rookie coach,” said Patsos, a former longtime assistant at Maryland. “I won a national title as a top assistant coach to Gary Williams. For 13 years I spent on Tobacco Road. I coached a couple of No. 1 picks in the draft. And we scored 48 points. That’s the problem that Loyola basketball had today.” (implying that his team’s offense cost them the game, not their defense).[5”–John Patsos
My point here is that there is SOME rhyme and reason to the outcomes you describe. Its not ALL random chaos. The drivers are often multifactorial.
One factor probably IS the quality of the opponent.
Another factor probably IS subjective value systems of the opposing coach in some cases.
The point is: in games evolving over a time period into emerging complexity, drivers get harder to unsnag, but they are there. It is not all randomness. This was the great revelation of first Chaos Theory and then Complexity Theory in the last quarter of the 20th Century that now shapes policy and strategy in all major fields of science, politics, war, law, and sport. The appearance of chaos in highly instituted, rule based activities, with an array of plays and incentives, may well signal some real chaos, but within that chaos there may well be strange tendencies and predictability of infinite variation within limits. In games with rules the chaos becomes more metaphorical. And game theory dynamics emerge and after a certain amount of play and counter play and exhaustion of the menu of countermoves, an equilibrium strategy, or tendency of play can emerge. Put another way, borrowing one of the early examples used, you may not be able to predict exactly where an eddy will appear downstream from the stick you hold in the current, but you can predict within certain limits where the eddy will form and move. Its fundamentally different than random variance and measurement error. Just gotta think it through and look for it and learn to love the imprecise but still useful predictability of things never quite repeating identically with similar limits. Same for looking for dynamical patterns of play in games. And these tendencies of interplay leave residues of statics that may call attention to them but obscure their process, or obscure them entirely as random noise. Great coaches I believe feel, or perceive, the processes of interplay; they become more than mechanistically cognizant of processes of interplay. They become kind of like savants at pattern recognition
I’m really doubtful that the opponent doesn’t matter. It may not be a tightly definable driver, because of the multifactorial contributions of other drivers, but its there and it is not completely elusive to recognize, just not tightly predictable.
I just presented one XTReme case in Patsos that shows how the opponent absolutely DOES matter.
But I feel (and share) your sense of complexity often being too tough to unsnag on available time, budget and information.
Rock Chalk!
-
I’m not sure I’m intelligent enough to get into this conversation.
Yet the Wizard once said,
“I’d rather have a lot of talent and a little experience than a lot of experience and a little talent.”
The game of basketball isn’t rocket science boys.
-
The Wizard may seem simple to you, but there was nothing too insignificant for him to systematize. The pyramid was maticulously constructed. Every spot on the floor was studied statistically for the best places to shoot. Every player was assigned spots where he was and was not allowed to shoot. Banking vs. swishing was studied. They were taught how to ties shoes and wear their socks. How to fill lanes was studied. Timing and length of outlet passes was studied. He studied the game statistically for at least 20 years. He studied whether straight or curved cuts were most effective. What people saw in his championship run was the distillation of 20 years of research.
-
First I faved you because I respect your knowledge of the game.
You are correct about the Wizard, yet here is one thing you forget about the Wizard. He didn’t worry about what the opponent was doing. He didn’t change his game plan because he was playing this team or that one. He was going run his stuff. The Wizard was a bit before my time but it seems to me he wasn’t a close minded person. In that he adjusted his plan to the players strengths that played for him.
Me personally I’m starting to believe Coach has opened his mind to that truth after the MSU game. That doesn’t mean the high/low is going away. Why should it? It’s one of the best offense schemes out there. It’s just coach realizes that he can’t run it the way he wants.
@jaybate-1.0 I don’t know how you feel about me or think of me, but understand this. Coach has dominated the regular season in college basketball. So I say why cant he be the next Wizard and dominate the post season too?
Me personally I know he can. Especially after how he has changed some things after that MSU game. MSU didn’t win that game KU lost it, and coach knows it
-
I think you are a board rat, like me, trying to understand the greatest game ever invented and advocating what he thinks is best for his beloved Jayhawks.
Self shares some important characteristics in the depth of his apparent thinking with Wooden and seems the only coach that might have a chance in this generation of going on such a run.
But they are, at the same time, quite different, as you rightly note.
Wooden was also quite different from Allen.
Difference should not surprise.
Cat-skinning has many paths.
Thought, as Borges said, is a labyrinth.
Action finds critical path, or loses.
You want Self to approach the game more like Wooden and suspect he is.
I think greatness hews its own path, but passes through similar phases of life that evidence bends in trajectory fitting that phase.
Self seems at one of those phases that Wooden came to and that Coach K and Roy came to…
This we agree on.
The nature of the bend is taking shape before us.
I think Wooden is helpful as an indicator that such bends come at such phases of a career and involve a coach taking another decisive step forward in the way he plays the game…
I do not think the bend will be as similar to Wooden’s in terms of technical solution, as you appear to.
But Self appears to have stepped to the plate to try to make the next bend.
It’s at once momentous AND just another bend.
It depends on how it turns out.
