Andrea Bargnani is Breaking All the Rules

As I was making my way through the interwebs earlier tonight, I saw an article on Hardwood Paroxysm comparing Andrea Bargnani so far this season to Dirk Nowitzki.  Before I read the article I was thinking, man, someone is feeling optimistic.  Then I read the article.  And my mind nearly exploded.

Bargnani is a center/power forward for the Toronto Raptors; he was drafted first overall in 2006, ahead of guys like LaMarcus Aldridge, Brandon Roy (not as impressive now), Rajon Rondo, and…. actually, that draft sucked.  While he scores a good amount, Bargnani doesn’t do much of anything else.  But as the Paroxysm article points out, his numbers so far this year are pretty comparable to Dirk last year, and Dirk was pretty good, right?  Let’s take a look at Bargs (all numbers per-36 minutes when relevant): he’s shooting a little less than last year, but his field goal percentage is at an all-time high; actually 40 points above his previous career-high.  His three point percentage is low for him, but he isn’t shooting as many so it isn’t hurting as much.  His free throw percentage is also a little low for him, but he’s getting to the line more than ever and 80% is still pretty good.  He’s at a career high in rebounding, with most of it coming on the defensive end.  He’s also at a career high in assists and career low in fouls.  These transfer over to his advanced stats; his true shooting percentage is 39 points over his previous career high and his effective field goal percentage is 23 points higher.

This all came as a surprise to me, especially since I read about it while the Raptors were giving the Wizards their first win of the season.  It’s also because Bargnani is the poster boy for one-dimensional, unhelpful scorers according to Wins Produced.  Bargs has in fact led the league in negative wins produced before due to his lack of a desire to do anything besides shoot and the minutes he keeps getting anyway.  WP isn’t alone in that opinion, although it varies somewhat; Win Shares has had him consistently below average, PER never had him as average until last year (and ok, he was a half point over average in 2010), both flavors of RAPM say he’s below average, ezPM says he’s below average, ASPM says he’s below average… it isn’t a question of if he’s bad, but how bad.  So his performance through 9 games so far (that’s what bball-reference currently has, although his game tonight was fairly consistent) is surprising.

How surprising?  Let’s say Bargnani gets to 2800 minutes, like he played in 2010.  At his current Win Shares, he would produce over 10 wins, which is over two-thirds of his career WS production.  His PER has jumped about 7 points, which would be worth 18 wins or more than his career PER production.  ezPM has him as a plus player for the first time ever; RAPM has him either a bit below average or a bit above, depending on which link you click.  Wins Produced isn’t quite as impressed, but his WP48 of .07 would be only the second time he wasn’t negative (with the first time being a rating of .002).  The .07 corresponds to only 4 wins over 2800 minutes, but last year he cost the Raptors 6 wins, so that’s quite a step in the right direction.  I don’t know if I’m ready for a world where Andrea Bargnani is a borderline-average player.

The other half of what’s surprising about Bargnani is his usage.  Or I imagine it should be surprising for some people.  A fair number of people in the NBA stats community believe in a usage-efficiency trade-off, and commonly point to a post by Eli Witus.  The post claims that when a player has to increase his usage (technically, increasing his shots taken, free throws taken, and/or turnovers) he becomes less efficient.  I’ve noted in the past that the regression is pretty weak, and we should assume that perhaps usage has an influence on efficiency but there are many other things that affect efficiency, and likely affect it more than usage.  But I still see the usage-efficiency trade-off argument appear pretty often.

So the ‘surprising’ part about Bargnani is that he’s doing all this with a career-high usage level.  It’s a 28.9%, which is up only a little from last year’s 28.1%, but both are 5% higher than anything in his first four seasons.  Thus Bargnani becomes an example of the inconsistency of the trade-off: from 2010 to 2011 he fits the bill, increasing his usage and playing worse.  But from 2011 to this season, he increases his usage and plays better.

There’s a lot of basketball left to play obviously, so Bargnani could return to form.  And I would be pretty comfortable with that.  But in the meantime, he is full of all sorts of surprises.  Some people might even think he’s making ‘the leap’, which I don’t have any numbers on but I bet is pretty rare for guys who have already played five years and 11,000 minutes.  We’ll have to see how his season turns out.


This entry was posted in Uncategorized and tagged , . Bookmark the permalink.

16 Responses to Andrea Bargnani is Breaking All the Rules

  1. Very nicely done addendum to my HP piece. It will be interesting indeed to see if Bargs’ arc continues or regresses to his career mean.

  2. Guy says:

    Alex: The notion of a usage-efficiency tradeoff does not require that all or even most players who increase their usage from one season to the next will see a performance decline. The idea is that a player would, on average, see such a decline IF everything else were held constant. However, when you compare two different seasons a lot of things can change — teammates, offensive schemes, the player’s own talent. There are a lot of other factors to account for. For example, we would in general expect young players to see their usage increase, and their performance to improve, over their first few seasons. And we’d expect the opposite for old players. So that alone will tend to disguise the usage-efficiency tradeoff.

    To actually see the tradeoff, we ideally would want players to be randomly required to change their usage. But there’s no way to carry out this experiment in the NBA, of course. (Eli’s study was, IMO, a clever approximation of such an experiment.) In the NBA, in contrast, the factors that cause a player’s usage to change will often, perhaps usually, impact his performance as well.

    • Alex says:

      I don’t believe that Eli’s post, or anything I have seen about the usage/efficiency relationship, is similar to what you said. In a previous comment I think you’ve even called it a big effect. It seems hard to reconcile such a big effect with a tradeoff that doesn’t apply to most players. Of course, that analysis was run within a season as opposed to across seasons, so there could be more extraneous factors in Bargnani’s case as you say. But I never hear the tradeoff discussed this way. We all know that the tradeoff exists, and we should expect to see it in any applicable case, even though it doesn’t show up in every case. It was just in the APBR thread for this year’s projections. Everyone is happy to run with the tradeoff idea until someone else butts in and says “I’m not so sure about the tradeoff…”.

      None of that is to say that what you said is wrong. I find it much more reasonable, although I haven’t seen any hard data to that effect. I’m just saying that I never hear about the tradeoff in such a measured tone.

      Looking specifically at Bargnani, what do you think the other factors are? Rasual Butler is the only new starter compared to last year. The Raptors have a new coach but I only hear about his defensive plans, not his offense. And as I said, it seems unlikely that Bargnani is making a jump 5 years into his career. Hoopdata says he’s only shooting a tiny bit less from 3-9 feet, and his shots are assisted a bit more this year compared to last, but the biggest change is in the 3-9 feet range where he’s shooting less and worse. I don’t watch the Raptors much at all, but I have to imagine there were some pretty glaring changes for his efficiency to jump so much, assuming it isn’t just a blip.

  3. Guy says:

    Alex: I won’t claim to speak for anyone else’s view of the U-E tradeoff. I think my view is similar to that of many others in the APBR community (including Eli), but I could be wrong. Maybe Evan or Daniel will drop by and offer thoughts.

    And I certainly didn’t say that the tradeoff “didn’t apply” in most cases. I said it wouldn’t be observable in most cases where a player’s usage suddenly changes. As you know, an effect being hidden by confounding variables doesn’t mean the effect isn’t present.

    In any case, my view is that the ideal usage for each player is an extremely complex game theory problem. I think each player has a distinct U-E curve, which depends not only on his talent but also on his teammates and on the offensive scheme/strategy employed by the coach. It will also be impacted by the decisions of the defending team. I don’t imagine the tradeoff is literally a straight line for all players. So if you told me, for example, that Bargs has a plateau such that he can maintain the same efficiency at 20% or 25% usage, I wouldn’t consider that proof there is no tradeoff. On the other hand, if you told me most players had equal efficiency at 10% or 30%, then I would absolutely reconsider my view.

    If you think about it, I think you will agree that there must be a tradeoff. If the Heat had James and Wade take 100% of the shots, do you doubt that their efficiency would fall? To believe that is to believe, among other things, that team defense matters not at all — even double- or triple-teaming a shooter won’t impact his efficiency. So the real question is: how big is the tradeoff within the range of usages we actually see?

    I wouldn’t claim to know how to measure it with any precision. But I think it must be significant. For example, it must be the case that efficiency starts to drop off quickly when a player starts to exceed about a 30-35% usage rate. If not, stars would have usage rates above that, but we don’t. I think it’s also pretty clear that most low-usage players would see a drop if compelled to take 30% of their team’s shots.

    I haven’t checked recently, but my impression is that the correlation between usage and efficiency is close to zero. If that is true, and there is no U-E tradeoff, then you have to believe NBA and college coaches — every single one of them — is a complete freakin’ idiot. I find that totally implausible.

    • Alex says:

      Evan is the one who made the recent comment on the APBR board, and is working on incorporating usage into his model of productivity. But perhaps it will end up being a small tweak. I occasionally see people explain why thy think the tradeoff happens, but I never read anyone say that it’s tough to find or really has to be accounted for on a case-by-case basis. I know you’re more familiar with some of the older boards posts and the like; maybe the discussion used to be more measured.

      I agree there has to be a game theory-related trade-off at some point. That would help explain why end-of-game scoring is so terrible; when everyone knows that Kobe is getting the ball and isn’t going to pass it, defense becomes easy. But it doesn’t mean that the trade-off has to exist, or be a strong force, at a lower range. Many NBA role and bench players were very good at the college level; obviously the step up in talent has an effect (I assume practically everyone is worse in the NBA than they were in college), but I don’t find it indisputable that they couldn’t stay efficient in a larger role if they had to.

      The correlation depends on what productivity measure you like, but in my data set it’s positive and generally about .2 (it’s much higher for PER). I don’t know if I think every coach is a complete idiot, but I do believe that virtually all of them are sub-optimal. Some of that is forced on them; you can’t let one guy shoot all the time because the other players get unhappy and don’t play as well when they do have an opportunity, or in other facets of the game, for example. You have to cater to the fans, or the owner, which creates some issues in some cases. But I also believe that there are holes in the strategy of pretty much every coach in any sport.

      • Guy says:

        I don’t know that we disagree a lot. But I think you are underestimating how radical a position it would be to think there could be little/no tradeoff in the 10-30% usage range. It really would imply that coaches are idiots. The most obvious strategy in the world is “let the best guy shoot.” It’s what we all played on the playground at one time (unfortunately for me). And yet we see literally scores of players in the NBA with <20% usage rates and above-average efficiency. If there's little/no tradeoff, you are saying coaches are dumber than kids on the playground.

        And the second-most-obvious strategy would be "everyone shoots 20%" — that is "fair" to all your players, and makes the defense's job as hard as possible. Yet we never see that: almost all teams have a few high usage guys, some average, and some low. If players are all trying to maximize their P/G, then the only possible outcome would be universal 20% usage — any player violating that protocol would quickly find they never received a pass, and would quickly be brought to heel. So either players care about more than maximizing their own point total, and/or coaches tell them how often to shoot (I think both).

        I think every NBA coach and GM would tell you that for any given efficiency rate, the number of players who could maintain that efficiency at 20% usage is smaller than the number who could do it at 10% usage, and the number who could do it at 30% is smaller still. (If they don't believe that, then neither usage patterns nor NBA salaries make a lick of sense.) Do you really think they are all wrong? If so, do this thought exercise: imagine the NBA changes its rules such that a team may choose anyone to take foul shots, not only the fouled player. What would happen? In a world in which coaches don't understand the importance of efficiency, and let players decide how often to shoot based on their egos and financial self-interest (I call this "BerriWorld"), we might find no change in who takes FTs. Or perhaps all players would take an equal number of FTs. But that is not what would happen. Instead, every single team would do what you or I would do: quickly identify our best % foul shooter, and give him that job every time. The idea that teams that employ PhD statisticians wouldn't figure this out in 30 seconds is just silly. And so why should be believe (arrogantly) that current usage patterns are the result of stupidity, rather than the outcome of complex game theory considerations that analysts have not (yet) been able to fully model?

        At a minimum, I would say anyone who rejects the U-T tradeoff has the burden of offering an alternative theory for the usage patterns we observe. And "players are selfish" doesn't cut it, because it can't explain what we observe….

        • EvanZ says:

          Guy, I agree. This is where domain expertise comes in (and you don’t need much to figure this one out). The other way to think about it is that you have guys like Tyson Chandler who are insanely efficient yet have very low USG (career 14%). WoW tells you that he helps his team by only taking efficient shots. But that reasoning implicitly tells us that there are many “inefficient” shots that must be taken. If every player had a 14% USG, the team USG would be 70%. Who would take those other 30% of the shots? Obviously, the point is that Chandler *can’t* increase his USG significantly without seriously diminishing his efficiency. That is so obvious on the face of it, that it’s hard to believe anyone would argue the point.

          • Guy says:

            Evan: perhaps you could use distance data to help demonstrate this point? I assume that players like Chandler — high-efficiency/low-usage (HELU) — invariably take a lot of their shots from close in. That would explain their failure to take more shots — teams presumably always want to take close shots, but cannot always succeed in getting the ball to a close-in player in a good position to shoot within the constraint of the shot clock. So, the only way these players could take more shots would be to shoot from further out, where their efficiency would certainly fall of.

            On the other hand, if there are HELU players who are mainly shooting from 20 feet, they should almost certainly increase their usage. And the existence of many such players would I think be evidence in favor of the idea of widespread sub-optimal coaching/shooting. I’m just guessing there aren’t many such creatures…..

  4. EvanZ says:

    Late to the party. I have started to incorporate USG in ezPM, using Eli’s regression. Right now, that means a player is credited (or debited) 0.25*(USG-20%). For example, a player who has 36% USG would gain 4 points in ezPM100. But the coefficient (0.25) is subject to change. At some point, I will run a regression against ORAPM (assuming the USG term can be aggregated with the offensive part of ezPM), and “tune” the coefficient to maximize correlation.

    I do agree with Guy’s original comment that you can’t look strictly at whether TS% goes up or down with USG. It’s also important to take into account shot selection. Bargnani is shooting out of his mind right now with respect to 10-22 ft jump shots. I for one don’t expect him to keep up that efficiency, but anything is possible.

  5. Johannes says:

    Hi, you referred to an earlier article from you were you estimated the usage effect. Unfortunately I couldn’t find it and to be honest I am not sure if it is even possible to regress it easily.

    If you assume players only increase usage for rational reasons (rational in terms of wins) you would not expect a big usage effect. I can think of 3 rational reasons to increase my usage:
    1 – I improved my skills (learned a new shot, got better in drawing fouls…. whatever)
    2 – The team tactic changes which makes it easier for me to score
    3 – I play in a worse team, so shooting with lower efficiency is still better than letting them shoot

    Only in the last instance I would expect a small drop in the efficiency. So given that rational behavior would predict only small usage changes, you can neither reject the rationality assumption nor measure the usage effect.

    The only thing I can imagine how you could do that, is by looking at teams how lose their high usage player due to injury. I did that actually for the Mavericks last year when Nowitzki was injured and his teammates FG% dropped almost 3% in the 11 Games without him. (either Nowitzki-effect or usage effect or random or adaption problems…)

    But you would need to do that for all injured high usage players and also correct for his substitutes. (so basically only look at the players who are usually with him on the floor)

    • EvanZ says:

      “Only in the last instance I would expect a small drop in the efficiency. So given that rational behavior would predict only small usage changes, you can neither reject the rationality assumption nor measure the usage effect.”

      That the size of the effect is “small” is only your assumption, right? For example, we might look at #2 options the year after a #1 option leaves a team. Bargnani’s USG increased from 22->28% last season in the absence of Bosh. His TS% went down by 20 points from the year before, which is arguably not a “small” change. The year Iverson left Philly, Iguodala became the #1 option (more or less). His USG went from 14.7->22.6%. His TS% dropped from 59.8% to 56.2%. That’s certainly not a small change. His TS% never again topped 56%. These are a couple of instances. We could look for others. I’m not saying that this will always be the case. What I am saying, however, is that there are (presumably) many such instances that we could find like this and put together some statistically meaningful sample size.

      • Johannes says:

        It is of course my assumption, that changes are in general small because in most of the times you don’t have these extreme changes.
        Since I don’t know how he regressed for usage, I thought he might have looked at all changes of usage between seasons. And there I certainly would expect only a very small effect. Which however does not disprove that there is a strong usage effect.

        If you take all these cases you might find a strong effect. That is what I would expect as well. But even than it will be smaller than it actually is, because only the players who can increase their usage will actually do it (if somewhat rational). Chandler will not start throwing threes, but Iguodala is probably an extremely talented scorer and is actually able to increase his usage with only a reasonable drop in efficiency (better than 20% if Chandler would start jump shooting).

        • EvanZ says:

          Johannes, I think we are in agreement. When I first read your comment, I think I misunderstood and thought you were arguing against the usage-efficiency tradeoff. Now I see you were arguing that the measurement technique simply may not have picked it up.

          We agree 100% that Chandler and other low usage players cannot simply increase usage arbitrarily without significantly impacting negatively their efficiencies.

    • Alex says:

      Johannes – just to clarify (I’m not sure who you were referring to at the start of your comment), I have never tried to estimate the usage effect. The post I linked to is by Eli Witus. He measured the effect by using line-up level data to predict offensive rating for a line-up based on the usage rates of the players in that line-up. Five guys who typically had lowish usage rates, and thus had to increase theirs when on the floor together, tended to be less efficient than you would guess whereas five guys who had highish rates, and thus got to lower them, tended to be more efficient. But the effect is very noisy, which has always been my major complaint with the finding. I think his method is a reasonable alternative to what you suggest doing with injuries, but with many more opportunities to see the effect (since any given player just has to be substituted out, not be injured).

      • Johannes says:

        Sorry, I thought you wrote that at some point… That would also explain why I could not find it in your blog. 🙂

        “I think his method is a reasonable alternative to what you suggest doing with injuries, but with many more opportunities to see the effect (since any given player just has to be substituted out, not be injured).”

        The problem with that method is however the big noise… An injury works best because it forces the remaining players to step up by keeping many other things constant… However, I can thank of many problems with that approach either and like you said there are only limited meaningful samples.

  6. Pingback: Usage Should Be Pretty Easy to Change | Sport Skeptic

Leave a Reply

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

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

Google+ photo

You are commenting using your Google+ 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 )


Connecting to %s