Usage Should Be Pretty Easy to Change

The Wages of Wins folks had a discussion on usage recently, and in the meantime I’ve had some thoughts percolating on if there should really be a usage-efficiency tradeoff or not.  This isn’t meant to be an all-encompassing treatise on usage (although I considered it), but hopefully more of a few notes and a demonstration.

First, let’s start with the definition.  Usage is an estimate of how many possessions (usually scaled out of 100, so it can be a percentage) a player uses while he’s on the court.  Players can use a possession by taking a shot (which includes some percentage of free throws) or turning the ball over.  Taking a shot doesn’t always use a possession because someone could get an offensive rebound, but as far as basketball-reference is concerned, player usage is turnovers plus ‘true’ shot attempts.

Next, let’s talk theory.  Hypothetically, there must be a usage-efficiency tradeoff due simply to game theory.  Let’s say that I knew (or quickly caught on during the game) that LeBron was going to shoot the ball every time; he was not going to pass and he was always going to get the ball.  The defense would quickly key in and he would face all five defenders, none of which were worried about him passing and potentially messing up his 100% usage rate.  His efficiency would be lower than if he were willing to pass here and there.  At the other end, let’s say that Tyson Chandler is never allowed to touch the ball unless he’s wide open at the rim for a dunk.  Because he rarely has the ball, and in fact may even be convinced by teammates to not look directly at it, Chandler will have a very low usage rate.  But when he does get the ball he will be a very efficient scorer because all he has to do is dunk the ball before someone swipes it or fouls him.  These two endpoints define a spectrum that seems to point pretty directly at a tradeoff existing.

So if this theoretical discussion is so convincing, why do I not particularly believe in the tradeoff (as can be seen here, for example)?  The main reason is because players play against actual opponents, and so game theory has already restricted the kind of usage rates we see.  For guys who played at least 10 minutes per game for at least 5 games last year, usage fell somewhere between 35.7% (Kobe) and 5.1% (Andris Biedrins).  These are presumably pretty representative numbers for the range of usages that will actually occur in a reasonable sample of NBA games.  So for all practical purposes, the question isn’t if a usage-efficiency tradeoff exists, it’s does the tradeoff exist in this range.  After all, NBA decision makers have to use information that applies to the game their players play, not the game that happens in thought experiments.

Now I’m going to change gears a little bit.  Instead of trying to prove whether the tradeoff exists or not (and you can find links to people who have tried in the WoW link above), I’m going to try to demonstrate that changing your usage up or down should be very easy.  If changing your usage is very easy, I think it follows that any potential effects on efficiency should be pretty minor.  But we’ll see how it goes.

For my example, I’ll pick on Kobe because he makes it so easy to do so.  Let’s start with the Lakers as a team.  Last year they had an average pace of 90.5, which means that per 48 minutes they would be expected to use 90.5 possessions.  Team possessions includes an adjustment for offensive rebounds that doesn’t happen at the player level for usage, but that’s ok.  The main point is, we can expect the Lakers to have a combination of shots, free throws, and turnovers that add up to the neighborhood of 90.5 per game.  Now if we look at Kobe, we see he took 23 shots per game, another 7.8 free throws, and turned the ball over 3.5 times.  That means Kobe used 29.9 possessions per game (roughly speaking; bball-reference doesn’t list the exact formula for player usage).  Knowing that Kobe uses about 30 of the Lakers’ 90 possessions tells us that Kobe should have a usage of around 33%, which is close to the actual number of 35.7%.  Getting it spot-on isn’t the point, so much as demonstrating that we can use team pace and player shots+FTA+TOV to get a decent estimate.

So let’s say we can convince Kobe to take one fewer shot per game by passing it to someone else or just taking a play off, so to speak (not by turning the ball over).  He would use 29 out of 90 possessions, and his usage would be around 32%.  In general, since usage is scaled out of 100 possessions and there are around 90 possessions in an NBA game these days, shifting a possession per game from one player to another should move their usage by about a percent.  Kobe could have moved his career-high usage of last season to his (still high) usage in the 2009 championship season by taking three fewer shots per game.  He could also just let someone else handle the ball more, since lowering his turnovers would also lower his usage.

Shifting possessions around shouldn’t take a giant effort.  Big men can flash through the paint or roll harder after a pick a couple extra times per game.  Shooters can have an extra screen play or two called for them.  The point guard could just pass to a different guy here or there.  It wouldn’t take a change in offensive philosophy, just a tiny bit of an effort.  If that sounds reasonable, then you could have fairly dramatic changes in usage.  We tend to talk about the outliers, like Kobe or Biedrins, but a lineup of five average guys would put each player at 20%.  If it doesn’t take much effort to shift usage by, say, 2 or 3% in either direction you could take a guy like Pau Gasol (22.1% last year, 98th in the league) and make him Corey Brewer (20%, 161st); you could take Brewer and make him Marvin Williams (18%, 220th).  A couple plays a game, the tiniest change in emphasis, and usage shifts across 80 players.

So changing usage should be pretty easy.  What would it do to efficiency?  Obviously that’s the tougher question.  If Kobe could be bothered to give up some of his 7.7 long jumpers per game his efficiency would definitely increase (although Kobe has a remarkably flat eFG% across shot distance).  If he tries to avoid a bad shot but throws a bad pass instead, his efficiency would actually go up a little (his usage would stay the same, but his shooting accuracy would increase).  But if he gave up the opportunities he got at the rim, his efficiency would go down.  Similarly, if a guy gets stuck with the ball as the shot clock runs down an extra time or two per game, his usage will go up and his efficiency likely down.  In short, there’s no easy way to predict which way efficiency ‘should’ go.  If a team just tried to get a guy his preferred shot (the one he’s good at, not necessarily the one he likes taking) an extra time per game, usage and efficiency could go up together.  If a team just dumped him the ball and said “bail us out”, his usage and efficiency would head in opposite directions.

That, in a nutshell, is why I think the necessity of a usage-efficiency tradeoff is overblown.  Changing usage is easy.  What happens when usage changes is not.  Efficiency definitely does not need to decrease just because a guy has the ball more; the guy could be getting more shots that he’s good at.  A guy who gets the ball less could have his efficiency go down because he gets antsy and starts jacking up shots when he does get an opportunity.  If someone like Kobe wants to increase his usage, he will probably see a decrease in efficiency because he already uses so many that eking out another probably has to happen in a less-than-optimal situation.  But anyone in a more reasonable range should have a fairly easy time.  Does anyone doubt that Pau Gasol couldn’t get the ball one or two more times without suffering?  What if Jeremy Lin didn’t have to just hand the ball to Carmelo, but could run a play and get a shot (Lin was 21st in usage last year, at 28.1%)?  It could go the other way; it’s easy to envision Amare Stoudemire’s role decreasing while he simultaneously relies more on jumpers, lowering both usage and efficiency.  Could Monta Ellis get less of the green light in Milwaukee but take even more long jumpers?  These all seem plausible, if not likely, and go counter to a tradeoff.

So in general I don’t take a tradeoff for granted.  I won’t go so far as to say it absolutely doesn’t exist, because again in some situations it must.  But I solidly believe that the conditions are so variable that any effects are going to be very noisy at best.

About these ads
This entry was posted in Uncategorized and tagged . Bookmark the permalink.

66 Responses to Usage Should Be Pretty Easy to Change

  1. dsmok1 says:

    Good article, Alex!

    When I ran my ASPM regression on RAPM, I expressly put in a term for USG–expecting there to be a significant increase with high usage.

    In the end, there was not a huge effect–USG^2, while positive, has a very small coefficient.

    The basic question that ASPM also looked at was–what is the threshold in points per possession at which point more usage is good? I found it to be below the league average, but not far so.

    Finally, the other place where usage is in the ASPM regression is multiplied times AST%. This seems like an odd place for usage, but this term absolutely dominated any other term for assists.

    Perhaps much of the value in usage is that the assists generated by that player are more valuable?

    • Alex says:

      I’ve always found it odd that usage doesn’t include assists at all, so having an interaction makes a certain amount of sense to me just to help assists ‘catch up’ so to speak. Game theory-wise, I think it would also be advantageous to have some uncertainty in if your players are going to shoot or pass, especially if they have the ball a lot.

  2. Guy says:

    Alex: I think discussions of the usage-efficiency tradeoff often suffer from a failure to define precisely what claims are being made. You make a useful distinction, which is to narrow the range of usage rates under discussion to those that occur in the real world (i.e. the NBA). So let’s limit the discussion to usage between 5% and 35%. But we also need to say what we mean by a meaningful tradeoff. I assume you don’t believe the curve is literally flat from 5% to 35%, but rather that it’s close to that (and perhaps literally flat for some players). But what does “flat” mean? Eli found a tradeoff in which a 1% increase in usage rate would result in a decline of 1.25 in ORtg (points per 100 possessions used). I would say that is a very significant tradeoff in basketball terms. It means, for example, that a 5% usage player would see his ORtg fall by 18.75 (on average) if forced to take his “fair share” of his team’s shots, while Kobe would see his ORtg increase by 18.75 if forced to do the same. In my view, those are meaningful differences, and if true would materially affect our assessment of these players’ value.

    There are many, many reasons to think the tradeoff is real (beyond Eli’s study). But before we debate that, would you agree that this magnitude of tradeoff — in the 5% to 35% range — would be significant? And as a corollary, are you arguing that the real tradeoff, if it exists, is much smaller than this? How much smaller?

    • Alex says:

      I don’t think that a 5% guy could jump straight to 20% in most cases, no. I also doubt that a 30%+ guy could drop to 20% and be dramatically better, a la your Kobe estimate (as it happens, Kobe’s usage in his first four seasons was at least five points lower than his typical usage since then, and his shooting efficiency as well as ORtg are below his career averages).

      I believe the connection to be extremely noisy, as I’ve said in the past, which Eli has also said, I believe, and which virtually everyone ignores. I do think (and I think my article makes clear) that virtually any player should be able to move his usage by maybe 3 or 4 points with no change in efficiency at all if it happens under the proper conditions. Whether or not teams try to make this happen, I don’t know. But I believe that 100%. If a player tries to move his usage more dramatically, I think some amount of a tradeoff will probably occur (again, with doubts as to the specificity). This would be particularly true from low to high; I have mixed feelings about from high to low. How much of a tradeoff I hazard to guess, but my gut tells me less than 1.25 ppp. Do you really think Kobe could play at 20% usage and increase his ORtg by nearly 19 points?

  3. Guy says:

    I’m familiar with the idea of noisy data, but not with a “noisy” relationship. Are you saying that you think the efficiency/usage tradeoff curve is not linear, and is more/less steep at different usage levels? And/or that the curve will have different shapes for different players? If so, I agree with both of those notions. And that could easily mean that any given player, under certain circumstances, might be able to slightly improve his usage without a decline in efficiency — which seems to be your claim. But that doesn’t refute the idea that there is in general a tradeoff, and that on average the impact will be of the magnitude Eli found.

    Interestingly, your notion that the tradeoff may be larger for players with low usage fits very well with Eli’s findings. And this makes sense: players allowed to have high usage rates have presumably been selected precisely because they can increase usage with a smaller than average drop in efficiency, whil low usage players are the reverse: selected because they will pay a price for higher usage. But keep in mind that high usage players are helping improve their teammates by lowering their usage rates, even if the high usage player himself might not (as you believe) perform much better at lower usage rates.

    • Alex says:

      I think the curve is definitely different for different players. I made the comment on low usage players for just that reason. If you look at the lowest usage guys, like Biedrins, Ben Wallace, Joel Anthony etc (from last year), they are uniformly centers with virtually no offensive ability. If you asked them to up their usage, they would quickly run out of the ability to do anything and their efficiency would suffer compared to their current diet of put-backs and wide-open dunks.

      Of course, there are only about 20 guys in the league like this and then you hit double-digit usage. Would these guys all suffer at higher usages? Just looking at a few I’m more familiar with, it isn’t obvious. Chuck Hayes has had higher usages with higher efficiencies; Kapono and Battier did the same before aging set in. Of course, some others have not. I believe that players with a general level of offensive competence could increase their usage with little to no effect on their efficiency (or even an increase, depending on how those extra possessions are deployed). I think it would be easy to take even a fringe player and ask him to do whatever his single talent is one more time per game and defy the tradeoff.

      I’m not sure I agree with your characterization of Eli’s study. Players aren’t selected to have a high usage because they can increase their usage and (relatively) maintain efficiency; they’re selected to have a high usage because they are capable of having a high efficiency at that usage. You can see this in the general shape of the first graph from Evan’s post on usage (http://www.d3coder.com/thecity/2012/03/30/visualization-the-outer-limits-of-the-usage-efficiency-relationship/). Eli’s claim would be that *from that point*, they should be less efficient if they attempted to use more possessions.

      If the question at the end of the day is ‘do I believe in Eli’s study and think it’s generally true?’, I guess the answer is sort of? I find it a little odd that no one has tried to replicate or extend it with a larger data set in the meantime, and its predictive use is tiny at best (which is virtually always what I mean by ‘noisy’). I think my examples make it plain that it doesn’t have to be true, but perhaps it is anyway. If that’s the case, I think my next question would be why coaches (or players) ask players to increase their usage in such a way that they play worse instead of playing to their strengths? Why not increase their usage in ways that would allow them to maintain or possibly increase their efficiency?

  4. EvanZ says:

    Kobe’s 3-year adjusted mid-range shooting rating is +0.12. If you have some faith in this metric, that means that when Kobe is on the floor, the team roughly takes the same number of mid-range jumpshots as when he’s on the bench. So the question to me, really, is not whether Kobe can cut down on the mid-range jumpshots. The real question is whether he is the best player on the Lakers to be taking those shots (I think the answer is probably “yes”). In effect, I think the “usage-efficiency” debate has been kind of misleading until now. We really should be looking at what types of shots account for usage and whether that player should be taking those kinds of shots.

    • Alex says:

      I guess I have two thoughts. One is that my intuition is certainly that if you took your average player and just told him to up his usage, he would try and do so by taking shots he’s either uncomfortable with or shots that are easy to get, presumably because the other team doesn’t mind if he takes them. Jumpers would certainly fall into those categories for a lot of guys (uncomfortable for most big men, easy to get for pretty much any one else). That would be my intuition as to why we generally see a tradeoff, although it certainly may not be true and need not be true. But I agree that looking at what kinds of shots players are taking would inform the debate; if most usage-increasers are just taking more jump shots, it’s a fairly uninteresting result (although knowing would make it potentially correctable).

      My second thought is, why do you assume the Lakers as a team (or any team) have to take however many jumpshots they do? Unless you think NBA offenses (and defenses) have evolved to some manner of perfect equilibrium, I think the real question is why they take those relatively inefficient shots as opposed to putting the ball in the post or taking more threes (or just taking a closer jumper). Given that some teams have made a conscious decision to emphasize the three, I lean towards thinking my question is the right one to ask. (And yes, I’m willing to concede that some number of long-ish jumpers is probably necessary from a game theory/strategy point of view, but that doesn’t mean that teams don’t take too many right now.)

  5. Guy says:

    Exactly right. Alex, you are approaching this in terms of “if player X takes Y more shots, how much worse a shooter do I believe he will become?” It’s better to think of this in terms of how a team is going to allocate shooting opportunities. The first choice is always to get the ball to the rim (preferably in the hands of a tall player), to get a high-percentage shot or draw a foul. But we don’t believe that teams can do that more often simply by willing it to be so. So they must also take jump shots of varying lengths and difficulty. THAT is why, as we all agree, Ben Wallace couldn’t increase his usage without reducing his efficiency — because he would be taking a different (more difficult) mix of shots.

    And I see no reason to believe it doesn’t work in reverse with high usage. If Kobe was told to get to 20% usage by not taking any 3 point shots or long 2s (except very late in the clock), I would expect his efficiency to climb. Not because lower usage magically makes Kobe a “better” shooter, but because the remaining shots he now takes are his higher-percentage shots (including lots of FTs). As Evan says, that change may or may not be good for the Lakers — it depends on who takes those jump shots instead and under what conditions. But it will certainly be good for Kobe’s efficiency rating.

    And how else can we explain the fact that low-usage/hi-efficiency players are actually less talented shooters than hi-usage/low-efficiency players (e.g. they have worse FT%)? They clearly are taking different kinds of shots.

    • Alex says:

      I tend to think of it that way because that is exactly the claim made by the analyses done so far (or you can substitute ORtg for shooting if you like Eli’s). If you’d like to move outside the data, we’re just having a philosophical debate. Which is fine, but then we can’t go throwing those analyses around as strong supporting evidence.

      Has anyone systematically looked at when high usage players come together to play on the same team? It isn’t exactly an apples-to-apples comparison, but I imagine the rough prediction should be that every player on the US Olympic team should be shooting at a ridiculously high efficiency, right? Only Chandler and Iguodala have a usage rate under 20%, and five or more of the guys have rates near 30%. Or perhaps more likely, guys who are traded together should improve their efficiency; recent examples would be the Heat, the Celtics when they got Allen and Garnett, the Knicks when they added Stoudemire. Do any of these predictions pan out?

  6. Guy says:

    I’m moving “outside the data?” Your entire post boils down to “it seems to me, intuitively, that players could increase their usage without reducing their efficiency.” Sheesh. In any case, I’m not suggesting a different empirical reality than what Eli’s data show. I’m suggesting a different and better way for you to think about why players’ ORtg changes with their usage. Consider this analogy: suppose the NBA allowed the Wizards to pick in advance 20 games next season which wouldn’t count for their season record. They would presumably pick their 20 strongest opponents, and we’d expect their 2012-2013 winning percentage to improve considerably. But we wouldn’t say the Wizards “got better,” because we know their improved win% is a function of their easier opportunities. Changing usage is largely (not entirely) about changing the degree of difficulty of a player’s scoring opportunities. Those opportunities aren’t fixed, the way the Wizards’ schedule is, but they are significantly constrained. Some teams take more long jumpers than others, but every team has to take a significant number.

    I also think you’re missing a key point about the value of high-usage players. It’s not only (or even mainly) about how much better they would shoot with less usage, it’s about how much better they make their teammates by reducing their usage. So even if you want to believe the tradeoff only exists on the low side, the high-usage players are contributing a lot of value by increasing the efficiency of teammates. (In fact, if the tradeoff exists mostly on the low side, then it makes even MORE sense to have high-usage players.) And again, the high-usage players are presumably selected in part because they are unusually good at maintaining production as usage grows (their curve is relatively flat). It stands to reason that if an average player were forced to increase his usage to 35%, he would see his efficiency fall far more than Kobe/Melo/etc.

    • Alex says:

      Guy, I didn’t say you were wrong or the world was going to explode or anything. I just said that if we want to talk about shot locations or choices as the reason for tradeoffs, we have moved outside of all of the data and analyses that I’m aware of. I also never said I was ‘inside the data’ in my post; I called it a demonstration or a thought experiment. Let me know if you still feel offended.

  7. Guy says:

    Alex: fair enough. But I don’t agree that talking about location is “outside the data,” at least not completely. Even if we don’t have precise location information, we know a lot about both the high usage and the low usage players that provides powerful clues. If you look at high usage/low ORtg players (HULO) and compare them to LUHO players, you find that the HULO guys are actually better shooters (higher FT%, many more 3PAs). Now, why would these superior shooters put up lower ORtgs, if they are not shooting from greater distances?

    More fundamentally, there is no way to tell a plausible story about NBA decision making without a usage-efficiency tradeoff. NBA teams give more minutes to high ORtg players. Yet there is virtually no correlation between ORtg and usage. How is that NBA teams value efficiency when handing out minutes, yet pay zero attention to it when allocating possessions? That makes zero sense.

    To believe in no U-E tradeoff (and I realize that’s not quite your position), is to believe that every NBA team operates in a massively inefficient way, and one that is aburdly easy to fix. ” Just give more shots to the efficient guys” — what could be easier or more obvious? Take Phoenix: they had an incredibly even usage distribution last year (almost everyone close to 20%). If these players could change their usage with no change in efficiency, they could improve the offense by 3 points a game and still keep all players between 10% and 28% usage. That’s 8 wins! I’m sure this same calculation could be done for virtually every NBA team. So if you follow the WOW gang’s logic, every team in the NBA is leaving tens of millions of dollars just sitting on the sidewalk every year. I think you really underestimate how radical — and wildly implausible — the “no tradeoff” position really is.

    • EvanZ says:

      I’ve been studying a lot of finance lately, and the usg-eff tradeoff seems as obvious to me as the risk-reward tradeoff. You simply can’t look at one of them without taking into account the other.

  8. Guy says:

    Just to provide a little actual data :>), there were 36 players last year with usage rates of 15% or less who were average or better in ORtg (104+). Their average ORtg was an impressive 111. But at the FT line, they shot just 66%! (straight average) There were 23 guys with usage rate of 25% or higher with ORtg of 106 or below (basically, average or worse efficiency), and they had a 78% FT%. Clearly, the latter pool is much more talented at “putting the ball in the basket” given a level playing field in terms of degree of difficulty.

    • Alex says:

      I certainly would never disagree that most guard/wing players are better overall shooters than center/post players. Of course, many center/post players probably haven’t spent their lives practicing shooting from farther away than they’ve ever had to.

      But that speaks little to a tradeoff unless it becomes critical that a player has to add worse shots to increase usage. If it somehow became demanded that Biedrins up his usage from say 5% to 10%, he would likely have to start doing some things he was uncomfortable doing (only in theory; his usage has actually always been above 10% before last year and he’s had three seasons with a better ORtg than last year). Dramatic changes in usage will probably cause changes in efficiency due to changes in role. But what if I just asked Biedrins to work the offensive boards harder and to post up an extra time per game (or whatever he’s good at; Evan would know better than me)? I’m sure he could keep his efficiency just where it is or increase it while getting his usage up to, maybe, 7 or even 8%. Right? So I guess I go back to my question from before, which seems more relevant given your other comment on the optimality of the NBA: why do coaches ask their players to increase usage in ways that make them worse players? Is it really so helpful to the offense to have those marginal shots tend to be poor ones? Why are so many high usage players wings when it feels like a marginal shot from a big man is likely to have a better chance of going in or drawing a foul, or why don’t those wings at least take more threes? Perhaps we see a general trend towards a tradeoff because players don’t know how to make a better choice in their shot selection?

      • EvanZ says:

        “But what if I just asked Biedrins to work the offensive boards harder and to post up an extra time per game (or whatever he’s good at; Evan would know better than me)?”

        Setting aside for the moment that Biedrins isn’t actually good at basketball anymore, even if he could increase USG by a few points, is that really significant? The question is how would a guy like Tyson Chandler get from 13% to 20% USG? Could he do it simply by cutting to the basket more? The answer is almost certainly not. Don’t you think someone would’ve tried that?

        • Alex says:

          His usage prior to this season has always been much higher, so Biedrins is a good example, actually (Chandler’s was higher his first couple years in the league, but has been pretty steady every since then). What was he doing before that he isn’t now? http://hoopdata.com/player.aspx?name=Andris+Biedrins . It looks like, more or less, he stopped getting shots at the rim. Is there a reason he (or someone like him, who presumably is still ok at basketball), couldn’t get those shots back? It seems like it would be worthwhile, even moving him to 8%.

          It seems significant in that moving his usage by 3 points should lower his ORtg by 3.75, right? And increase his teammates’ ratings by a bit? If I did my math right, moving Biedrins from 5% at ORtg 118 (his stats last year) to 8% at 114.25 ORtg would take four guys previously at 23.75% usage and ORtg 100 to 23% usage and 100.9 ORtg. That line-up would go from 100.9 points per 100 possessions to 102. That’s not a lot per se, but a point is nothing to sneeze at, right? Maybe someone should try that.

          • EvanZ says:

            “It looks like, more or less, he stopped getting shots at the rim. Is there a reason he (or someone like him, who presumably is still ok at basketball), couldn’t get those shots back? It seems like it would be worthwhile, even moving him to 8%.”

            Scouting reports? That’s what they pay coaches to figure out. It’s why many people are concerned about Jeremy Lin’s future. It’s also why guy’s like Kobe Bryant, who can score in so many different ways, become super-stars.

          • Alex says:

            Biedrins was getting 7 or 8 shots at the rim for three years (at least, that’s as far back as hoopdata goes) before his shots dropped. I guess it took that long for the league to catch on?

          • EvanZ says:

            In terms of Biedrins, specifically, he just lost it. I think he hasn’t been physically the same player for a few years. I heard he might have an arthritic hip. At any rate, if you watch him play, he’s not the same player he was 4 or 5 years ago. When Keith Smart coached the team a couple years ago, he would specifically run plays for Biedrins at the beginning of games to try to get him more involved, and it just didn’t work at all. They gave up fairly soon thereafter. But if you’re suggesting that the coaching staff hasn’t thought about “increasing his USG”, that’s simply not true.

          • Alex says:

            Biedrins has been a stand-in for the low-usage player because he was so low last year, but good to know there was a reason for the huge decrease.

          • EvanZ says:

            Alex, do you still maintain that “USG should be pretty easy to change”? I mean, if you really believe that, then someone should be paying you millions as a head coach.

          • Alex says:

            I haven’t heard anything to change my opinion. Do you think it would tear a team/player’s efficiency apart to run an extra play for a guy once or twice a game? That’s all I’ve been claiming here. Do you maintain that the tradeoff is an immutable law of basketball and must occur in every instance?

          • EvanZ says:

            “Do you maintain that the tradeoff is an immutable law of basketball and must occur in every instance?”

            Does gravity take a night off? Of course, it goes without saying that teams should try to optimize their entire strategy to be as efficient as possible. Do you honestly believe they don’t know that? It’s not just one or two shots either, because that won’t take Tyson Chandler from 13% to 20% USG. I think it’s being disingenuous to now say you’re just talking about “one or two more shots”, when the title of the post claims how easy it is to change USG. If it’s “easy”, why now put a disclaimer on it? Either you believe it’s “easy” or you don’t.

      • Guy says:

        Alex: I really think it will help if you think about this more at a team level, instead of focusing on the idea of a single player increasing his usage. Instead of saying “Biedrins should increase his usage,” try saying “his PG should start passing to Biedrins in situations where he currently believes the team has better options.” Doesn’t sound as good, does it?

        “Why do coaches ask their players to increase usage in ways that make them worse players? Is it really so helpful to the offense to have those marginal shots tend to be poor ones? Why are so many high usage players wings when it feels like a marginal shot from a big man is likely to have a better chance of going in or drawing a foul, or why don’t those wings at least take more threes?”
        Why do you assume teams can get more open shots under the basket than they currently do? Don’t you think that’s their first priority on every possession? The reason they can’t always do it is, well, because the other team plays defense. Which of these seems more likely: A) wings take a lot of shots because teams cannot always get the ball to the rim, and the wings are better than big men at shooting from distance, or B) NBA teams have not noticed that efficiency tends to increase the closer a player is to the rim? I think the question answers itself.

        I do think the phrase “shot creation” has perhaps confused this debate more than necessary. It seems to me that “shot creation” should mean the ability to develop a high-probability scoring opportunity, either for yourself or a teammate. But high usage players are not always doing that. In many cases, they are taking an average or below-average quality shot attempt, and the argument for doing so is that they are better than their teammates at making such shots. This can be a valid argument, but it can also be an excuse for a guy who’s taking too many low-quality shots. The really great offensive players do both of these things, creating a lot of high-quality shots AND making the most of worse opportunities. I think it’s a mistake to label both of these abilities “shot creation.”

        • Alex says:

          Perhaps focusing on low-usage guys like clunky centers has made us both think too much on getting the ball to the rim. My main question is, why do players seem to increase usage by taking long jumpers? I think that we can agree that’s the main way players add usage. If a player adds usage by taking a jumper, by definition the team is losing usage from other shot areas (or at best trading one long jumper for another, but certainly not reducing the number of long jumpers taken by the team). When I look at where teams tend to shoot from (http://www.hoopdata.com/teamxefg.aspx), I see a huge range in the long 2 category: it ranges from Denver’s 15.3% to Charlotte’s 34.7%. That implies, to me at least, that teams can exercise some decision making in terms of whether they *want* to take that shot, not that they *need* to take that shot. Does that seem reasonable?

          If that seems reasonable, the next question is where they could maybe move those shots to if they didn’t want to take so many of those bad ones. Short and mid-range shots have fairly small ranges, about 10% across teams. At-rim and three-pointers have larger ranges, about 15% and 20% respectively. So if variance in shot type across teams is any indication of how you could influence usage via strategy (and I won’t claim strategy is the only thing to influence shot type, but I have to assume it does to some extent), then it seems like you should be able to shift your long jumpers to shots at the rim or from three most easily. It doesn’t appear you even need a dominant big man to want to take shots at the rim, looking at who’s near the top of the league.

          To make a minor quibble with your first statement, I think you should have emphasized the word ‘believes’ in “believes the team has better options”. If the PG’s current strategy for team usage is to make a jumper a primary option (or a ‘better’ option), he should indeed dump it into Biedrins. Or he could back up and take a three, or pass to someone that will take a three. Or he could drive and try a floater, or something else to get a closer shot at least. Those would be better options.

  9. Pingback: Experimenting with Wins Produced « shut up and jam

  10. Guy says:

    The Hoopdata breakdowns do show that teams vary in their shot distribution. But we don’t know how much that is a choice, as opposed to reflecting the talents of those players. More importantly, I mainly draw the opposite conclusion as you from the data: they show a remarkable uniformity of shooting opportunities. Look at expected eFG%, which tells you the net benefit from a team’s shot distribution. The range is quite narrow, with a SD of just .74. Teams aren’t differentiating themselves much through shot location alone. In contrast, the SD for how well teams shoot given their distribution (ratio) is about twice as high, and has a much higher correlation with actual eFG% (.89) than does expected eFG% (.44).
    So yes, there is variation, but not a lot. And that still doesn’t tell us whether any given team can CHANGE their distribution in a way that improves efficiency. I’m very skeptical of this idea. And that’s why I would not agree with you that players “add” usage by taking jumpers. In general, players don’t seem to change their usage a lot (after their first few seasons), so for the most part, players don’t “add” usage in any way. What is likely true is that high usage players take relatively more jumpers. I make the distinction because you make it sound like players are, rather perversely, electing to “add” difficult shots. But that merely assumes what you need to prove, which is that teams could take better shots simply by choosing to do so.
    Interestingly, getting to the rim frequently has no correlation with eFG%. That surprised me. In part this is because teams that get to the rim take fewer 3PA. But they also tend to be poor shooters both on long jumpers and 3PAs. So there appears to be some tradeoff there.

    • Alex says:

      I’m a little confused now. If we don’t think players change their usage very much after a couple seasons, why are we concerned about the tradeoff? A couple other thoughts: I’m not sure how large I would expect the SD of expected eFG to be, but it does appear to be true that it has a range roughly half the size of actual eFG. If we believe shooting efficiency to be a key factor in winning, it then seems like shot location alone explains a decent portion of what causes a team to win or lose, right? Somewhere in the neighborhood of half of why teams differ in eFG? You’d want to run some kind of analysis to check, but it seems like shot location alone, independent of accuracy, might be as important as any of the remaining three Factors? And it’s true that some of the difference in shot location is due to player talent, but selecting players and developing talents is a choice that teams make. Perhaps Derrick Rose couldn’t just decide to shoot from different locations than where he does now, but the Bulls could certainly choose to have a point guard who shoots from certain locations.

      I thought the general belief (not just my own) is that if a player wishes to increase his usage from wherever it currently may be, he would have to add shots that are difficult for him. If you believe in a fairly optimal NBA (and it always sounds like you do), this has to be true. If the NBA is set at anything close to an equilibrium, players should only be able to add to usage by ‘perversely’ taking worse shots than those they currently get, right? Otherwise players would be able to add to their usage by taking shots that are not difficult for them, which implies there shouldn’t be a tradeoff.

      • Guy says:

        The variance in ExpFG% would suggest that it explains about one-third of the variance in eFG%. However, that’s a bit misleading because 3P frequency is directly tied to 3P eFG% (i.e. a team shoots more 3s if it is good at it, which is not true for other kinds of shots). So the high value of 3s is itself a function of the shooting skill of those players who choose to take them. If low-3PA teams simply elected to take more 3s, they would obvioulsy not get the current expected payoff in efficiency because they lack good 3P shooters.

        In any case, my point is simply that the range of team behaviors is rather narrow. Every team takes a lot of jumpers. Should some teams take more 3s and fewer long 2s? Perhaps — I’m not claiming every team is perfectly optimized. But my guess is that a team’s success at getting to the rim is about 95% determined by the talent of their personnel.

  11. Alex says:

    As a side note, Evan and Guy (and anyone else reading through these many comments), my main complaint with a large portion of basketball discussion (and most discussion about anything, with the current state of the world) is that there is little nuance. Evan’s last comment seems like the perfect reflection of this with respect to the usage discussion. “Does gravity take the night off?” Of course not, because gravity is a physical law of nature. The tradeoff, on the other hand, has a shaky history of even being found (from what I know; people were unhappy with early attempts at looking year-to-year or even game-to-game, right?), and about the only solid evidence for it that I’ve seen is Eli’s study. Eli’s study was pretty well-done; I have no big complaints about his method. But people take it as bible whereas the tradeoff is clearly not. Regressions spit out estimates with errors, not facts. The error in Eli’s regression pretty clearly suggests you shouldn’t expect a player to have a tradeoff all the time or a tradeoff of the same size for every player. You can tell because a) the R squared is pretty small b) he estimated what an individual player’s tradeoff should be from a lineup-level analysis and c) if you wanted to know any particular player’s tradeoff you would need to run some kind of fixed effects analysis, which he didn’t do (and which I know Evan is familiar with because he’s posted them). And so my largest complaint with the usage discussion is that for the most part it is completely polarized: you can claim to believe it happens all the time to everyone or you can claim it never happens. Both are stupid and neither are what I have in my article, or at least are certainly not what I was trying to convey.

    So let’s try some nuance. Guy seems to believe that the tradeoff can certainly vary by player and by circumstance. I’m on board with that. As my post says, I certainly think that players could increase their usage by a couple points with little ill effect. It would have to happen in certain ways, and perhaps those ways simply don’t happen in the NBA. Maybe coaches believe it’s better to let Rose or Kobe or Nash create an extra time instead of dumping into the post or running Korver or whoever off another screen for a three. Maybe they’re even right; I’m sure there’s a reason I’m not an NBA coach, as Evan would like to remind me. But as a guy who’s played and watched plenty of basketball, I’m fairly certain that there are some wasted shots and lack of effort over the course of a game. That being said, do I think that the Knicks could force-feed Tyson Chandler the ball an extra 10 times a game next year and have good things happen? No, that sounds to me like a situation where you would definitely expect a tradeoff. Could he get it an extra one or two? Sure, why not? Would that only help the team a little bit, as opposed to a whole lot? Sure, but little helps add up if you make enough of them, right? I think it’s reasonable that teams could attempt to avoid some of this tradeoff by shifting shot locations. Guy doesn’t think that’s plausible or a big factor if it is; I don’t mind that. It’s only my opinion, and it’s his opinion. But I think the important part (and I’m going to speak a little bit for Guy here) is that we both recognize that we’re voicing opinions with numbers that give equivocal evidence and that there isn’t an immutable law at work here. When we talk about usage, and most things in basketball, we’re talking about things that happen in degrees and change in different circumstances. That means that even if it’s “true” that something happens generally, it certainly doesn’t happen all the time, and there might even be times when it’s “untrue”.

    • EvanZ says:

      “The tradeoff, on the other hand, has a shaky history of even being found”

      It’s so deeply embedded in the game (theory) of basketball, that it’s nearly impossible to find. You can’t do an experiment to find it. You can’t tell teams to dramatically change usg patterns just so us stat nerds can figure this thing out. It’s implicit. It’s fundamental to basketball. There are hints of it here and there, but mostly, you just have to use domain expertise and your own basketball knowledge. If it doesn’t make sense to you, so be it. I think NBA coaches (and players) would laugh you out of the building if you told them it was “easy” to change usage without giving up efficiency.

    • EvanZ says:

      “Could he get it an extra one or two? Sure, why not? Would that only help the team a little bit, as opposed to a whole lot? Sure, but little helps add up if you make enough of them, right?”

      How do you know that whatever he’s doing now isn’t already including that “one or two” extra shots?

  12. Guy says:

    Evan makes exactly the right point: if a player can add “an extra one or two,” then logically he can keep going until he hits 35%. At each point, if we rely on your intuition we will think he can add another “point or two.” And by posing the question that way, it’s impossible to prove your intuition wrong, because the change in the player’s ORtg will be too small to measure.

    But whether a few players could “add one or two” is not a very interesting question. The interesting questions are 1) why does usage vary from 10% to 35%?, and 2) why aren’t usage and efficiency more correlated? That is, why does basketball look the way it does? And a U-E tradeoff explains that reality, while the absence of one makes it incomprehensible. You began by conceding that the lack of usage rates below 5% or above 35% were likely rational outcomes, reflecting teams’ adaptation to a usage-efficiency tradeoff outside those boundaries. But if the curves are roughly flat from 10% to 35%, why don’t we see a standard distribution of something like 35-30-25-5-5, with the three highest ORtg players getting the most usage? It must also be true that the current usage practice evolved because it has proven functional.

    And Eli’s study is not the only evidence. Pelton did a study showing that Melo’s teammates shoot better when he is on the floor. I think someone found the same thing for Kobe. We have good evidence that when players are forced to increase their usage — for reasons completely independent of their performance — they lose efficiency. But as Evan notes, these natural experiments are limited.

    I find that it helps to think in terms of the MARGINAL shot a player takes. If teams are acting optimally, the ORtg for an additional shot taken by each player on the team (or really, lineup) would be roughly equal. For some guys, that’s their 35th shot, for others it’s their 10th — it depends on their talent. Your intuition tells you that teams aren’t coming close to this distribution today (and the WOW crowd would have us believe that any .500 team could become the top offense in the league if they would only read Dave Berri — not an exaggeration of their position). What your position means in practice is that most high-efficiency players should take at least a few shots in each game at a time they now pass to a teammate. Even if we think your basketball knowledge is the equal of these players (and their coaches), they have one enormous advantage: they see the situation in real time. They see the spread of the players, who is covered and who isn’t, and so on. How can you be confident that even with that knowledge, these players are SYSTEMATICALLY passing the ball away when they should shoot? If that’s going to be your position, I think you need stronger evidence than “I think they can do better.”

  13. Alex says:

    Gentlemen, it appears I can do nothing to change your minds. The shooting/possession use situation in the NBA is so optimized that any attempt to improve efficiency, even a tiny bit, is a fool’s errand. I defer to your expertise.

    • EvanZ says:

      “The shooting/possession use situation in the NBA is so optimized that any attempt to improve efficiency, even a tiny bit, is a fool’s errand.”

      It’s not a fool’s errand, Alex. The disconnect here is that you seem to believe teams don’t already attempt to do this, whereas myself and Guy essentially take it for granted that improving efficiency is basically all the coaches and players are always trying to do. Now, sure, some do it better than others. I won’t argue with you there.

    • Guy says:

      Well, that’s disappointing. Alex, you are making two very different arguments in this post and your comments, perhaps without being aware of it. Claim A is that at least some high-efficiency players could increase their usage without a significant drop in efficiency. Claim B is that, within the normal usage range, the tradeoff is weak or doesn’t exist at all (“why do I not particularly believe in the tradeoff?”, “changing your usage up or down should be very easy. If changing your usage is very easy, I think it follows that any potential effects on efficiency should be pretty minor,” and “anyone in a more reasonable range should have a fairly easy time.”). Evan and I have been challenging claim B, but you seem to be defending claim A (which neither of us disagrees with).

      Perhaps you think these two claims are the same. They are not, any more than saying “some places on earth are flat” is the same as claiming “the earth is flat.” In the one case you are claiming that NBA teams could make marginal gains in efficiency (a claim for which you have no evidence at all, but it does seem plausible). But your larger claim requires you to believe that every single NBA team could achieve massive gains simply by aligning efficiency and usage better (including the considerable number that employ PhD statistical analysts to look for such gains). It requires you to believe (among other things) that defensive strategy does not change at all based on the usage rates of opposing players — that the knowledge of which opponents shoot the most has zero value to a defense. And it leaves you with no remotely coherent theory to explain the actual behavior we see in the NBA (e.g. why are some players allowed to shoot so much more than more efficient teammates?). You should either offer some arguments for this incredibly radical claim, or acknowledge that in broad terms the usage-efficiency tradeoff must exist.

      • Guy says:

        Sorry, meant to write: “Claim A is that at least some high-efficiency players could SLIGHTLY increase their usage without a significant drop in efficiency.”

  14. Neil Paine says:

    Figured I’d chime in since this is an area I’ve been pretty focused on recently.

    I think I’ve come around to looking at it from a different angle at this stage — instead of thinking the causality goes from the player’s usage to his efficiency, I think efficiency and usage are both just describing how he’s being used by the team/coach/system. If a guy is the team’s designated midrange shooter, for instance, he gets dinged for doing his job whether he’s at a high or low usage. His job might be entirely to absorb low-efficiency possessions as a means of opening things up for teammates.

    In other words, the difficulty/efficiency of that mythical ”marginal shot” might be impossible to predict from either a player’s usage or his efficiency. In fact, the marginal shot theory of skill curves might be completely meaningless for certain types of players.

    Isiah Thomas is a fun example, because his best statistical season is at age 23 on a team that didn’t win anywhere near as much as his later Detroit teams would, teams where he had the same usage (26-27%) but drastically reduced efficiency. Why did this happen? They went from being 4th in pace to 25th (out of 25), a stylistic change that robbed Zeke of easy baskets and made him the designated ”tough shot creator” in the half-court when post-ups to Dantley didn’t work out. Instead of this idea that usage is built in layers, with the first 10% being easy shots, the next 10% being harder, and the next 10% being extremely difficult, Isiah basically cut out the bottom (easiest) 10% and had a mix of nothing but harder and extremely difficult shots making up his 26%.

    If true, this really messes up the skill curve dynamic, which not only assumes that each marginal shot is steadily more difficult than the last, but that going in reverse gets you easier attempts, and that everyone’s offensive game is built in layers going from easier to harder shots. Some players might just take the bullet for their teammates and do nothing but absorb hard shots. And, unfortunately, it’s impossible from the boxscore to distinguish those players from guys who just have horrible shot selection.

  15. Guy says:

    Neil: That’s a very interesting way to think about the issue. In that framework, it’s more important to know the difficulty of shots “assigned” to a given player, than to know how many shots he’s taking. That sounds right, with the caveat that usage must still impose a distinct penalty due to increased attention from the defense. That is, if two equally talented players are assigned the same distribution of shot types, but one is at 13% usage and the other at 26% usage, I’d expect the latter player to be at least somewhat less efficient.

    Of course, most players will still experience a U-E tradeoff, because most players have at least some high-efficiency possessions at the rim (and if they are good 3P shooters, they likely take advantage of those opportunities as well). So any expansion of usage will usually involve taking new shots with expected point values below that player’s mean. But it is possible to imagine players who take such tough shots already that increased usage will have no impact, or even improve their efficiency.

    This points to the need to evaluate players relative to the difficulty of their shots (like an Olympic gymnast), an area I know Evan has done a lot of work. Similarly, to decide if a team is optimizing it’s offense, we would start by asking whether the right players are taking each type of shot. That is, are a team’s best 3P shooters taking the 3s, and the best jump-shot guys taking those shots? That seems to be clearly true of 3PA, at least in general. Is it generally true for 2P jumpers? If so, that’s evidence of efficient use of players (and also evidence that NBA decision-makers understand the value of efficiency). But it still leaves open the important question of whether a given team could take fewer difficult shots, which of course is always to be preferred.

    Interestingly, a similar logic applies to rebounding: we can’t really know the value of an individual player’s rebounds, without knowing his assigned role on the boards (i.e. how many rebounds he’s taking from teammates).

  16. Guy says:

    Addendum: Dre over at WOW provides a good example of how far wrong you will go if you don’t take account of the U-E tradeoff. He concludes that Chandler (13% usage) is the most valuable 2-point shooter in the NBA:
    “Two Point Shooting: Tyson Chandler with +4.6 Wins Produced over Par
    Tyson Chandler was shooting 68% from the field. Basically ever three shots he took, two went in. That’s insane! He was the top earner in this category. I want to stress that players have a quota, so to speak, in the Wins Produced system. To earn wins Chandler had to “Earn” as many points from two point shooting at the average center. After that he started collecting wins. It’s bizarre to think that Chandler’s key contribution this season was offense, but it was. And in terms of the Usage Argument, I have trouble believe Chandler couldn’t miss just as well as Melo with more shots.”

    Consider the claim that additional usage by Chandler would obviously be better than letting Melo shoot. Chandler shoots 63% on FTA, Melo shoots 81%; Chandler is 0% on 3PA, Melo is 32%. Do we really think Chandler is a superior shooter, or might their opportunities be a tad different? The fact that 81% of Chandler’s FG are assisted, compared to 38% of Melo’s, gives us the answer. Clearly, Chandler has almost no ability to create shots on his own.

    As a side note, I think it’s revealing that Dre, who seems to run the WOW site, emphasizes that he is crediting Chandler only with 2-point wins above what an average center receives. But the average player gets ZERO credit for 2-point shooting in WP. That’s one of its biggest shortcomings: players shooting 0-for-0, 1-for-2, 4-for-8, and 12-for-24 on 2PA all are assigned exactly the same value: nothing. There is no “quota.” The fact that Dre doesn’t understand this basic fact about the metric says a lot……

    • Alex says:

      I think Dre was a little imprecise in his description. I assume by ‘quota’ he means what an average player at the position would do. Obviously an average player still generates wins; you have to have a 0 WP48 to actually generate “nothing”; the same would be true for Win Shares.

      I do agree with you that it’s unlikely Chandler could take as many shots as Carmelo and maintain his sterling percentages. That being said, I do still believe that the team could try to find him an extra opportunity or two without much harm, as long as they came in the same general flow of his current opportunities. And in terms of your claim that Chandler has no *ability* to create shots on his own, I think it’s more fair to say that he had almost no *opportunity* to create shots on his own. If you look at his hoopdata page (http://hoopdata.com/player.aspx?name=Tyson+Chandler), you can see that his percent assisted has gone up every year, more or less, since 2007. True, his true shooting has also gone up that whole time (although it didn’t drop when his % assisted dropped in 2010), but we would assume that to be true for most players. But in 2007 Chandler was assisted on less than half of his shots and still had 62% true shooting, which is nothing to sneeze at. Even those shots were presumably different from Carmelo’s; I’m not arguing Chandler is the same quality of shooter. I’m just saying that the opportunity to take certain kinds of shots works both ways; when Chandler was given the chance (or perhaps just because he was younger) he could certainly shoot accurately while generating his own shots. Carmelo has a lot of opportunity to generate whatever shots he wants, and he’s sort of average at making them because he takes more difficult shots. Were Carmelo (or the team) more focused on having someone else handle the ball and generate shots for Carmelo, he would presumably shoot more accurately and have his % assisted go up because he could take better shots.

      • EvanZ says:

        I think there are always going to be individual cases and inefficiencies where a player *could* have increased usage without dramatically decreasing their efficiency. A good example is Bynum actually. His USG went up from 17.6% in 2011 to 23.8% last season and his TS% only decreased from 60.6 to 59.4. That increase in USG was well worth the slight hit on his efficiency (which was also helped by a 10% jump in %AST). In general, those kinds of inefficiencies are figured out in the first few years of a player’s career. Why it took the Lakers this long to figure out Bynum should be getting more shots is beyond me. Perhaps Chandler should also be given the opportunity to get more shots and just see what happens.

        In any event, what we’re talking about now is just a matter of degree not kind. The USG-EFF tradeoff is real, but the size of the tradeoff varies by team and player. Would you not agree with that Alex?

        • Alex says:

          I’m pretty sure I’ve said in the comments that the tradeoff (including whether there is actually a tradeoff) should differ depending on circumstances, so yes I would agree with that. Whether I agree with the idea that it’s “real” depends on what you mean by real. If you think it’s like gravity, which to me implies a constant force that applies the same to everyone, then I would disagree. If by “real” you mean can exist and does for some number of players/circumstances, then sure.

          > Date: Fri, 10 Aug 2012 16:07:04 +0000 > To: akonkel@hotmail.com >

      • Guy says:

        If Chandler took more opportunities, we have no reason at all to believe his efficiency on THOSE additional shots would be above average. In fact, all the evidence suggests that he is far below average on jump shots from any distance. So your claim amounts to saying that his team should be able to do a better job than it does today of getting the ball to him at the rim. This could be true, but what’s the evidence?

        More importantly, you continue to redefine the real debate here. Your original claim was that between 10% and 30% usage there isn’t much tradeoff. If so, we can ignore it when evaluating players and look only at efficiency, as Dre did here. Is that still your position, or do you agree there is generally a significant tradeoff? Please don’t hide behind this idea that “some players could take a few more shots,” which can never be disproved and no one really disputes. Address the important questions. Can most 13% usage guys become 20% guys with little efficiency loss? Do 30% usage players really not improve the average efficiency of their teammates at all? Berri, Dre, and co. say “yes.” What say you?

        And yes, an average player generates wins in WP. But he generates zero wins from his 2P shooting specifically, which was my point. That is true for any average-efficiency player, regardless of usage. So a league-average efficiency shooter at 5% usage is just as valuable as one at 30% usage. This is absurd if there exists a U-E tradeoff, but that’s what WP says. What’s shocking to me is that Dre seems not to be aware of these important aspects of the metric. And BTW, here’s the link: http://wagesofwins.com/2012/08/09/could-the-nfls-chris-johnson-make-usain-bolt-look-slow/.

        • EvanZ says:

          Guy, I don’t think that was the link you intended, but it was hilarious, nonetheless. When people talk about why analysts should have some domain expertise, I’ll fall back on that article.

          • Alex says:

            As a former track athlete, that was tough to read. They were right to catch trouble in the comments.

            > Date: Fri, 10 Aug 2012 16:38:51 +0000 > To: akonkel@hotmail.com >

          • Guy says:

            Sorry, wrong link. Here’s the correct one: http://wagesofwins.com/2012/08/09/the-biggest-winners-by-stat/.

            That was a funny post. I don’t so much mind that they forgot to account for acceleration time–we all make mistakes. But you have to love this: ” in a land where runners are fighting over hundredths of a second for the honor of being best in the world, is it likely that Johnson could shave almost a full second off his time? It’s possible. It’s also possible that to become a pro-athlete have big egos and like to talk smack.” In a land where runners fight over hundredths of a second, how likely is it the fastest guy in the NFL will finish three-quarters of a second (!) behind Bolt over just 40 yards? When your model gives you an impossible answer, maybe the model needs tweaking. And of course you have the always delightful Berri condescencion….

  17. Johannes says:

    @Alex Interesting discussion. However, it strikes me as a little bit odd that you seam to support the position of the WoW network, even so your claims maid in the comments section are incompatible with the views of Barry and co.

    “I do agree with you that it’s unlikely Chandler could take as many shots as Carmelo and maintain his sterling percentages.”
    You point to the fact that small changes should be possible, however, you must admit that if you are going to analyze the value of a player you cannot just ignore extreme usage stats. I find WP somewhat useful for players with average usage (e.g. 16-24%); however, if it comes to extreme cases like Chandler vs Nowitzki I doubt that WP remains accurate. And these extreme cases are often causing disbelief by people looking at the WP numbers .

    Any example being it super obvious or not, will be dismissed by the WoW network, because it cannot be shown in the numbers. For them Chandler should and could be the top-scorer at any team and anyone who disagrees is fooled by his perception (like all NBA Managers and probably all NBA players including Tyson Chandler himself).

    “If by “real” you mean can exist and does for some number of players/circumstances, then sure.”
    Only because you cannot measure/predict it perfectly, does not mean that you can just ignore it. In particular if the effects can be rather large and effects players very differently. This potentially messes up the ranking for some players (again for the extreme cases).

    I also think the value of a player to a team depends on the circumstances. Having Tyson Chandler playing for the Bobcats would do not much good, because the small number of highly efficient points would not help much (if they are not even offset by even less inefficient shots of his teammates). On the other hand could Carmello Anthony give the Bobcats a real boost with his medium efficient high volume scoring, even so I do not want him on an elite team where Chandler definitely belongs (at best next to an highly efficient, high volume scorer like Nowitzki ;) ).

  18. Justin says:

    Really late to the discussion, obviously, but the numbers support the existence of a significant relationship between usage and efficiency:

    http://ascreamingcomesacrossthecourt.blogspot.com/2013/05/usage-versus-efficiency.html

    • Alex says:

      Sounds like the uncertainty is still pretty high though, right? I followed up on that thought here with simulated data, but you could make actual confidence and prediction intervals with your data: http://sportskeptic.wordpress.com/2013/02/01/accuracy-and-precision/

      • Justin says:

        Still quite noisy, but there are a few reasons for that. One is that the same lineup in another year was treated as another observation. It’ll be interesting to look at the numbers for combining lineups and see what changes. Another is that 100 possessions, or 50, for a lineup is still a small sample size. That leads to noise. However, when you pool the lineup data together for an average usage of 15.5 to 16.5, etc., the R^2 is outstanding, and the relationship between usage and efficiency is quite clear.

        One big issue is lineup matchups. Some lineups are only used against weaker options. For example, Jamaal Crawford can’t guard a potato sack, so he can’t go against elite backcourts. Also, think of a really high usage lineup. That’s like LeBron + Wade + Bosh, Carmelo and JR Smith, Durant and Westbrook, etc. Those are lineups guys fear, so they put in their best defensive lineups, generally. I need to adjust for the strength of the opposing defense. That will cause some of the noise. It’s also a systematic error, but it’s an error that I think makes low usage lineups seem more efficient than they are.

        I’ll look more at your post when I add refinements. Thanks.

        • Alex says:

          My general perspective (which shows up in a variety of places on the blog) is that 1) there is, broadly speaking, a usage-efficiency trade off but 2) it would be very hard to predict for most specific line-ups and 3) small changes in usage, say a percent or two, don’t need to affect efficiency at all if they are implemented in an intelligent way. So I’m not surprised by your results.

          A couple of thoughts though: the R squared becomes outstanding when you bin everything together because you’ve eliminated so many sources of noise. If the question is “can we find evidence of an efficiency-usage trade-off”, then binning is fine. If we care about almost literally anything else in any detail, then it throws away all the information and the uncertainty that comes with it. That being said, I’d be willing to bet that if you found a way to put reasonable error bars on the bin graph, by which I mean the error bars reflect all the uncertainty in those means, the error bars across bins would overlap substantially even if the R squared for your nine points is fairly high.

          I find it kind of interesting that your R squared, if memory serves, is lower than what Eli had. You obviously have more observations than he did, so the noise must increase at an even greater rate. I’m not 100% sure what that suggests, but it might influence your thoughts on combining line-ups across seasons.

          > Date: Tue, 21 May 2013 07:49:38 +0000 > To: akonkel@hotmail.com >

          • EvanZ says:

            Alex, it still boils down to the question: Could Tyson Chandler remain efficient at 30% USG? I think all but the most hunkered down WoWers would have to admit that would be a foolish proposition to believe in. It’s really the rare superstar that can seemingly increase USG without significantly affecting TS%.

            I would encourage you to look at nbawowy for the Heat, and look at lineups where LeBron, Wade, or Bosh are off the court. Look at how the USG of the other 2 remaining on court changes and what happens to their efficiency. It’s truly incredible how good those 3 guys are.

          • Alex says:

            If that’s the question you think it boils down to, then that’s fine. I personally think it’s a foolish question, but maybe I’m wrong. I think the better question would be, does Tyson Chandler have to fall apart if his usage went from, say, 13% to 15 or 16%? What line-ups can I put together when a starter sits, or if I make a trade? The very research that supports the trade-off also supports the idea that there is so much noise that it’s hard to make a claim for any specific instance. Here’s the example from the post I linked to in the previous comment:

            To be a little more concrete and topical: the Pistons just traded for Jose Calderon and got rid of Tayshaun Prince and Austin Daye. What should they expect if they ran out a line-up of Calderon (18.2% usage), Stuckey (21.5%), Singler (15%), Monroe (25%) and Drummond (16.4%)? The equation says their 96.1% usage should take their projected offensive efficiency of 109.5 down to 107.5, but it could really be anywhere from 95 to 119.

            As a side note, as best I can tell from your site that particular line-up ending up scoring 27 points in 20 possessions, so their efficiency was much higher than expected (obviously in a very small sample, although we’re pretty much always talking about small samples, aren’t we?).

            > Date: Tue, 21 May 2013 16:43:06 +0000 > To: akonkel@hotmail.com >

  19. EvanZ says:

    “If that’s the question you think it boils down to, then that’s fine. I personally think it’s a foolish question, but maybe I’m wrong.”

    I wouldn’t ask the question if I thought everyone was in agreement about the answer. Unfortunately, there are some folks who believe regardless of how low Tyson Chandler’s USG is, that he is the “best” scorer in the league. And you know I’m not lying about that.

  20. Guy says:

    I think Justin’s study confirms that Eli was providing an accurate assessment of the *average* impact of usage on efficiency, across players. It’s true that we don’t know how much this impact varies from player to player. But a low R^2 doesn’t prove that the impact varies wildly. This could mainly be a function of the fact that the performance of each individual lineup will have a large amount of random variance in it, simply because it has relatively few possessions. That would be true even if we had 10,000 lineups, or 10,000,000 lineups (as long as most had only 50-200 possessions). Even if every single lineup was effected exactly as the model predicts, I doubt you could get an R^2 above .10.

    Alex thinks that knowing the average impact has little or no value. I think that depends on what question you want to answer. For player valuation, it’s extremely useful. If a player has a 14% usage rate, he is forcing his teammates to increase their usage by 6%, and we can quantify the cost of that to a team. No, we don’t know for sure what the impact is on this particular team. But since the impact is spread across 10 players, it’s likely to be near the average, and it’s certainly better to use our best approximation than to ignore the effect. So I don’t actually care what Chandler would do at 30%, but I DO want to estimate the amount his low usage reduces his teammates’ efficiency. Similarly, I don’t really care whether Kobe would be more efficient at 20%, but I do need to credit him with the increased efficiency he very likely creates among his teammates.

    The question Alex wants to answer — what effect will small changes in usage have on a specific set of players? — is also perfectly valid. And he’s right that Justin’s and Eli’s studies can’t answer it. But the findings have other, valid, uses.

    • Alex says:

      I don’t think that it has little or no value; I think it has little *practical* value to NBA decision-makers. It’s obviously an important topic overall and better to know the result than not know it, but I imagine that a coach, scout, or GM would take so many other things into account, as well as try to think of ways to potentially increase a specific player’s usage without hurting his efficiency, that the information is relatively minor, particularly because of the uncertainty involved. Some guys are obvious: I’m not going to trade for Chandler and then expect him to be the focus of my offense. But what about a guy at 18, 19%, or a little lower? Is it really going to kill me to get Iman Shumpert? To ask him to shoulder a bit more of the load? Probably not 25%, but how about 20? Would you put your foot down, based on this, and say there’s no way you want that guy on your team?

      You seem unconcerned with the poor model fit. That doesn’t set off any red flags for you? It isn’t of interest that the fit for the larger data set is lower than the fit for Eli’s data? If Justin were to combine the same line-up across seasons (e.g. the Thunder starting five for 2012 and the same guys in 2013), which sounds like his next step, that would presumably bump it up a bit but: a) I find it a little worrisome in general that more observations apparently injected more noise than information and b) even if combining helps, you start running into the same interpretation issues that you have with multi-year APM. Those players in 2013 are not the same guys as they were in 2012, and if I’m going to make use of this information, I need to know who I’m getting *now*, not who I’m getting over the average of the last two (or more) years with guys who he may or may not play with.

      Also as a practical matter, Justin (or someone) could run the analysis again but limit it to line-ups with over, say, 200 possessions. That seems like a pretty good number. And for this season, at least, there are over 140 line-ups that fit the bill, so with five seasons of data there should be plenty of observations. You could even make it 400 and there are about 50 line-ups that make the cut this year. Presumably the R squared would be much better and the estimates would be less noisy. But again, practically for the decision-maker, he would then have to make estimates based on what happens for the best, healthiest guys (since they would be the ones that make the cut-off). You’re already asking him to extrapolate to individual players based on a line-up analysis, and now he would have even further to go. At some point the generalizability of the result is going to completely break down.

      Again, I’m not trying to completely dismiss the result or say that it’s false. I am questioning its precision and how actionable it is. I think those are valid questions.

      > Date: Fri, 24 May 2013 19:46:52 +0000 > To: akonkel@hotmail.com >

      • EvanZ says:

        “Some guys are obvious: I’m not going to trade for Chandler and then expect him to be the focus of my offense. But what about a guy at 18, 19%, or a little lower? Is it really going to kill me to get Iman Shumpert? To ask him to shoulder a bit more of the load? Probably not 25%, but how about 20?”

        Nobody disagrees with this. On the contrary, you’ve repeated several times what everyone else seems to already take for granted.

        “Again, I’m not trying to completely dismiss the result or say that it’s false. I am questioning its precision and how actionable it is. I think those are valid questions.”

        It seems like you are trying to dismiss the result…until you start talking about the actual results, and then you seem to be agreeing with everyone else.

        Can we just agree to agree?

  21. Guy says:

    Alex: I think one mistake you are making is to focus only on our ability — or inability — to predict the consequences of a *change* in usage for specific players. But there are many, many reasons we want to know the likely impact of a player’s usage rate on his teammates, even if it remains unchanged. Should I offer a max contract to player X? Should I select player Y in the draft? To answer these questions I want to estimate the player’s value as best I can. And to do that, you need to adjust for his usage rate. Our assessment of both high- and low-usage players will be much more accurate if we account for the impact of their usage on their teammates. That’s extremely actionable.

    And no, the “poor model fit” doesn’t concern me. Why should it? Our goal isn’t to predict the efficiency of individual lineups, but to estimate the coefficient for usage. The larger data set has a lower R^2 because Eli set a higher thresh hold for possessions (IIRC). Your R^2 will be a function much more of the average possessions than the # of lineups. Think of all the noise you are dealing with: your expected efficiency (based on season stats) is itself only an estimate of these players’ true efficiency rates at normal usage. Then the data for each lineup has a huge amount of random error, even at 200 possessions. Plus they may have played mainly at home or away, against bad/good opponents, etc. Just a WAG, but even if every single lineup were affected by usage change exactly as the model estimates, I doubt your R^2 exceeds .1.

    • EvanZ says:

      Guy, as I said above, Alex does agree with us. He just seems to have a hard time admitting it.

      • Alex says:

        Evan – Shumpert is currently about a 15% guy; you think he can get to 20% no problem? I was under the impression that was a jump too far. And I believe that in previous discussions, Guy has argued that the efficiency change for even a 1 or 2% usage change was a big deal. But perhaps I’ve misremembered people’s opinions on the matter.

        Guy, just double-checked Eli’s study. He used cut-offs of 50 and 100 possessions just like Justin. I’m not sure if they both weighted by possession the same way (it sounds like Justin used cutoffs and weighted but Eli did one or the other), so it may not be perfectly comparable.

        And maybe I’m misreading you here, but don’t your two paragraphs contradict each other? In the first one you want to know about player X’s usage so you can account for how it affects his teammates, emphasizing the specific individual. But then in the second you say you don’t want to predict the efficiency of individual line-ups.

        > Date: Fri, 24 May 2013 21:31:34 +0000 > To: akonkel@hotmail.com >

        • Guy says:

          Alex: Eli got R^2 of .02 and .04 at minimums of 50 and 100, while Justin got .02 and .03. This doesn’t seem like a material difference. And in any case, the R^2 has little relevance here, given these tiny sample sizes. (For 100 possessions, the SD of binomial variance will be about .1, or about five times the size of the impact we expect to see from extremely high or low usage).

          No, there is no contradiction. This research is telling me that if I add a 25% usage player to an otherwise average lineup (replacing a 20% player), I gain about 1 point per game on average. Do I know for sure how much that 25% usage player will improve individually? Of course not. But this gain comes from him and *every other player* on the court with him. Some of these players may benefit more than expected from reduced usage, others less. But spread across 9-10 players, and thousands of possessions, there’s a good chance my team will experience something approximating the average gain.

          And even if I add this 25% player to four guys who typically have 75% usage, and no one has to change their game, the 25% guy is still generating additional value. That’s because, other things equal, there are more guys who can achieve X efficiency at 19% usage than at 20% usage. Having a 25% usage player makes it easier/cheaper to fill out the rest of my roster (or, same thing, it allows me to buy more efficiency for any given budget of $Y).

          The only question here, Alex, is whether to use imperfect information or none at all. You seem to be taking the position that ignoring this information is as good a choice as using it (i.e. the information has no “practical value”). That’s a very strange conclusion. A GM who uses this information won’t always be right, but he will be right far more often than one who ignores it.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s