Sunday, November 7, 2010

Another 'Bargnani is a Bust' Post

New readers, please go check out the basics before you leave any comments on this post.

Every time I mention that Bargnani was a terrible pick and list players from the 2006 draft that I would prefer instead of him, I still get readers would question my sanity and insult me. Of course, they never explain why I'm wrong, they just like making unsubstantiated ad hominem attacks. But I have collected even more information that should make it abundantly clear that Bargnani was a rancid pick.

How does Bargnani compare to other #1 picks?

In a previous post, I compared Bargnani's Wins Produced numbers to the last 12 years of first overall picks. This time, I've added all the advanced numbers, such as PPS, per 48 minute stats, and even PER and Win Shares (note: I do not condone use of Win Shares or PER - especially PER). I've also extended the comparison down to Olowokandi and included Darko Milicic (who was actually a second overall pick), because those are the two players Bargnani apologists inevitably claim were worse than picks than Bargnani. And remember: these numbers only include the first four seasons of each player's career. Here are the more complete numbers (spreadsheet here), this time in the form of a handy little gadget which (if it works correctly) should allow you to sort by whatever category you want:

To sort the columns, go to 'define groupings and calculations', select the column you want to sort, hit the arrow, and then click 'next' and then 'OK'. To remove a grouping, highlight the category you want to remove and click the blank button (which is underneath the one with the arrow). Unfortunately, it seems like it can only order values from smallest to largest, but it's better than giving you a plain spreadsheet. RSS users, please visit the actual website to view these numbers, as the gadget doesn't seem to work in feeds.

What do the numbers tell us? Well for one, Andrea Bargnani does not compare very favourably to the rest of these players. Ranked by WP48 he is dead last behind Olowokandi, Milicic, and Brown. If for some reason you don't "agree" with Wins Produced, ranked by PER (worth repeating: I do not condone use of PER) he is 11th (ahead of Brown, Milicic, and Olowokandi), and ranked by Win Shares (worth repeating: I do not condone use of Win Shares) he is 12th, ahead of Olowokandi and the small sample size of John Wall. If you weigh all three metrics (WP, WS, and PER) equally (which, again, I do not) and take the cumulative rank  - 'Combo Rank' - Bargnani comes out 2nd last, behind Milicic and ahead of Olowokandi. Regardless of how you slice it, Bargnani was bad for a first overall pick.

But something interesting happens when you sort these players by points per 48 minutes; all of the sudden Bargnani moves up to 7th place, ahead of notably better players like Dwight Howard, Greg Oden, and Andrew Bogut. As David Berri has mentioned over and over, the popular way of evaluating NBA players is by points scored, so it should be no surprise that so many fans (and certain GMs) have problems recognizing that Bargnani is not very productive. Casual fans are blinded by the relatively large amounts of points that Bargnani scores, failing to notice his lack of production in other areas - such as rebounding - or his shooting efficiency (Bargnani is also 4th in FG attempts).

For a good laugh, try sorting all of the players by rebounds. Bargnani places 11th, just ahead of SF Lebron James. It should be noted that Lebron entered the league when he was two years younger than Bargnani and that at the moment Lebron has actually rebounded at a better rate than Bargnani during his career.

Also, note the age of each player - Bargnani is as old as Milicic and Howard, who both entered the league out of high school. Still expecting Bargnani to 'make a leap' this season? Those who do expect a you expect a leap out of Howard and/or Milicic?

How does Bargnani compare to other members of the 2006 draft class?

In one of those previous posts up above, I compared Bargnani's Wins Produced numbers to the rest of the 2006 draft class. This time - as with the #1 picks - I've added all the advanced numbers, such as PPS, per 48 minute stats, and even PER and Win Shares (note: I do not condone use of Win Shares or PER - especially PER). Remember: these numbers only include the first four seasons of each player's career. Here are the more complete numbers (spreadsheet here), again in the form of a handy little gadget which (if it works correctly) should allow you to sort by whatever category you want:

Again, to sort the columns, go to 'define groupings and calculations', select the column you want to sort, hit the arrow, and then click 'next' and then 'OK'. To remove a grouping, highlight the category you want to remove and click the blank button (which is underneath the one with the arrow). RSS users, please visit the actual website to view these numbers, as the gadget doesn't seem to work in feeds.

Bargnani ranks 27th out of 30 - or 4th last - in terms of WP48. The players he finishes ahead of are Hilton Armstrong, Ryan Hollins, and Adam Morrison (one of the worst players of all time). If for some reason you don't "agree" with Wins Produced, ranked by PER (worth repeating: I do not condone use of PER) he is 13th (because PER rewards players who take a lot of shots, he finishes ahead of good players such as Shelden Williams and Thabo Sefolosha), and ranked by Win Shares (worth repeating: I do not condone use of Win Shares) he is 23rd. If you take the Combo Rank, Bargnani is ranked 22nd. So no matter how you slice it, Bargnani was not a good choice and not even in the top 10, top 15, or top 20 players of the 2006 draft.

But once again, something interesting happens when you sort these players by points per 48 minutes; all of the sudden Bargnani moves up to 4th place, ahead of several better players like Rajon Rondo, Paul Millsap, Leon Powe, and Ronnie Brewer. Blinded by scoring, many delusional Raptor fans will say that even if Bargnani wasn't the best player that year, at least he was in the top ten. However, we shouldn't forget about other aspects of the game when we evaluate players; in rebounding, Bargnani places 14th behind F Shawne Williams, G/F Thabo Sefolosha, and just ahead of G/F Rudy Gay and PG Rajon Rondo.

As the most comprehensive 'Bargnani is a bust' post I've written to date, hopefully it will help more people realize that Bargnani is not a very productive player. The Raptors need to trade Bargnani as soon as possible, before the rest of the league catch on to his lack of production and while his value is at its highest. Exchanging Bargnani for a more productive Centre would do wonders for the franchise and is something that should be GM Bryan Colangelo's number one priority.

 - Devin


  1. This comment has been removed by a blog administrator.

  2. .
    ..... we've overdosed on the Bargnani hate, so these kinds of "evaluations" tend to get a .... blah response. And if anything, people post comments like mine, without reading beyond the 1st line of your Post.

  3. RapthoseLeafs, I don't really care about the response I get. I have a blog and I get to post things that interest me. The Bargnani story is a classic WOW tale and I felt like these numbers were an improvement on what I had included in my previous posts.

    That being said, I rarely get any comments, and often the comments I get (like the one I deleted above) are of the useless/insulting variety. If you don't feel like reading the post, by all means, knock yourself out; but you'll be missing out on something that I found useful. Throughout the season(s), I will continue to point out how unproductive Bargnani is, until he becomes productive (not bloody likely) or gets traded (and even then, I will delight in pointing out his lack of productivity).

  4. Devin,

    You are preaching to the converted my friend. I refuse to believe that Colangelo is this dumb however, why does he hold on to Bargnani. I mean, Adam Morrison is worse, but I don't think the initial team that drafted him held on to him and then promoted him as the face of the franchise. Come on.

  5. One of the hallmarks of a well-run franchise is that they can recognize mistakes and correct them. Portland did that earlier this year when they traded away Jerryd Bayless, and they've done it before. Miami got rid of Beasley. The Bobcats got rid of Morrison. But did the Raptors get rid of Bargnani? No - they gave him a $50 million extension.

    Barring any new trade developments, yes, I do believe Colangelo is "this dumb". Bargnani was his guy and BC must think that an admission of failure somehow tarnishes his reputation. News flash Bryan: owning up to your mistakes and learning from them is much better than the George W. Bush denial method.

  6. Until BC goes, Bargnani will stay.

  7. Unfortunately, you're probably right. Which is why Toronto should be on the market for a new GM.

  8. wow I can't believe you have Oden ahead of Bargs.. the guy is a total bust supreme. And honestly that whole draft class aside from Roy and Rondo sucked huge b-balls. I was never one to like Bargs being picked #1.. I was and am still a huge Roy fan, and cannot believe the raps passed on him! But in no way is Bargs the worst as you seem to think... just as in no was was Bosh a true #1 premo franchise guy. Good, yes... but not a Mr.-clutch at all. You got too much free time to write crap like this.. I agree with the guy who posted that this article is boring.. why RapsHQ sent a link to here I will never know. I wish you luck in your hating, and hating new ppl when your Bargs-bashing wears thin. I think you should go post on the Blue Jays or Maple Leaf websites.. you'll fit in nicely there. I want my 3 minutes back!

  9. Well Devin,

    It's a Bargs post so you should expect hate.

    My question for you (I posed it to Berri but he did not responde) is the following: WP/48 is based of boxscore stats. However, somebody playing primarily on the bench would in theory have a better opportunity to accumulate box score stats and thus inflate his WP/48. Is this the case? Also, is there a "translation" of sorts that goes on when somebody gets more starters minutes. For example, WP goes down by x% when a person moves from the back up to the starting position?

  10. If you break down Bargnani's performance, it's hard to see how he can be nearly as bad as WP estimates. If you use Arturo's Gimme the Rock metric to measure his offensive productivity, he has produced 32.5 points per 30 possessions, compared to 33.6 for an average center. That's below average, obviously, but not a huge difference. That translates into about 1.5 wins below average over a full season.

    What kills him in WP, of course, is his low rebounding rate. But there's no evidence at all that this hurts the team to anything like the extent WP estimates. The team is actually above average in rebounding this season, and a bit below average over the previous 2. But the team has only been about 50 rebounds below average each season, while WP claims that Bargs is costing the team 225 boards per year. That just isn't plausible.

    This doesn't mean he was a great choice. But the notion that he's generated negative 8 wins just doesn't stand up to scrutiny.

  11. Instead of debating this mediocre - at best - player, lets just trade his ass so we never have to debate him again.

  12. @Anonymous

    Berri, might not have answered you question, but someone in the WoW network did relatively recently. I can't remember who, but I remember the result. Starters don't play a significantly larger percentage of their minutes against starters than do bench players, so the basis for qualifying WP48 doesn't actually exist.

    Chalk another one up for statistics over obvious conventional wisdom.

  13. @Guy,

    Isn't the evidence that WP works it's correlation with wins and the stability of player values as they switch teams? Your argument seems to be that the individual pieces of WP don't explain wins independently. So what? There's a reason WP was created and we don't just use Rebounds on their own to give the worth of a player. People in the WoW do often attribute undeserved statistical significance to individual pieces of the WP48 formula, but that doesn't mean that you should make the same mistake.

  14. tgt:
    I think the correlation point is one of the bigger misunderstandings about WP. A metric's high correlation to team wins tells us very little about how well it apportions credit among players. It is extremely easy to devise metrics that add up to a team's point differential, as WP does, and thus has a .95 correlation with team wins. For example, credit each player with his points scored, minus his share of team points allowed (MP/19,680). This correlates just as well as WP, but it's obviously a poor metric.

    My argument isn't that individual pieces don't explain wins. It's that they fail to predict a change in that same "piece" for the team. And that is the basis for WP. Each rebound for a player is valued at one point because one rebound for a team is worth one point, and only for that reason. So if a player rebound does not actually add a rebound for the team, than we no longer have a valid reason to credit the player with that point (1/33 of a win). Read Dr. Berri's work, and you will see he's very clear that this is how it works. If you want to say that player rebounds don't actually add team rebounds, but somehow they make teams win anyway, at that point WP just becomes a magical black box with no scientific basis behind it.

  15. Anonymous re Oden: According to string theory, you are correct. But in this universe, you'd have to be crazy to pick Bargnani over Oden (and every metric agrees with me, even silly ones). If you are a Raptor fan who doesn't like it when people criticize players/coaches/GMs (or "hating", as you called it), please go return your fan license, because you don't meet the qualifications.

    Anonymous re starters/bench: tgt had the right info. Here's the link:

    Clearly, there are some very good players who simply come off the bench, like Ginobili, Odom, Haslem, etc. Bench players play against starters at the same rate as other starters.

    Anonymous re trading Bargnani: Yes, he is clearly not a very good player. The problem is that not only is he 'not good', he's also not even mediocre, and some people still think (because of his scoring) that he's actually a good player. Some, unfortunately, we still have to be having these conversations.

  16. Guy:

    I don't really understand the confusion - rebounds are valuable. Someone who doesn't rebound hurts your team because 1) other players have to compensate and 2) often can't make up for the deficiency. Bargnani has been - and will always be - a bad rebounder. Centres need to grab rebounds. If he doesn't grab enough rebounds, that makes him less productive. If the team isn't that far below average in rebounding, it is because other players have made up the slack, and we should be singling out those who made up the slack and recognizing their contributions to the team.

  17. DDignam: Extra rebounds are clearly valuable for teams -- there is no dispute on this point among basketball analysts. The question is: what is the relationship between an individual player's rebound total and his actual contribution to team rebounds. WP assumes it's a 1:1 relationship. But this clearly isn't true. How do we know that? Because every single high-Reb48 player (+3.0 Reb48 above average) fails to improve his team's total by as much as expected. And teams with low-Reb48 players never do as poorly as predicted at the team level.

    That's NOT true for shooting efficiency: a player who shoots X% at Y FGA48 will, on average, increase or decrease team shooting efficiency exactly as predicted. A team with an efficient SG does not become less effecient at other positions. But a team that gets a lot of rebounds from center always gets below-average production from the other 4 positions (in aggregate). Teams that get a lot of rebounds from PF always get below-average rebounding from SG/PG/C/SF. And so on. That can't be a coincidence, and it proves there are massive diminishing returns when it comes to rebounds. The best research on the issue consistently shows that each additional rebound recorded by a player produces somewhere between .2 and .5 rebounds for the team (higher for ORB, less for DRB).

    Another way to think about it: if teams always manage to compensate for a low-Reb48 player by having other players get the rebounds, then how valuable can it be? That suggests the team is going to get many of these rebounds (not all) no matter who grabs it. But if this was really an important and scarce skill, then some teams with a low-Reb48 center would actually fail to get a lot of rebounds, right? That's what we see with shooting -- a low efficiency shooter hurts team efficiency. But low-Reb48 players (and high-Reb48 players) just don't have much impact on team rebounding.

    Does that clarify things for you?

  18. @Guy:
    Did I say that correlation does anything on its own? No. Did I say that correlation + stability does? Yes. Maybe you should respond to the actual argument.

    I agree that individual pieces don't explain wins, but the sum does. What's different between WP and other high correlators like net scoring is that WP is stable from year to year including when players change teams. It predicts total wins for teams accurately when adding and substracting players.

    While Berri might have built the logic of WP off of a rebound being worth a point (obviously bogus), that doesn't mean the overall metric does not work.

    The logic to build the metric: bad.
    Pieces of the metric in practice: bad.
    The full metric in practice: good.

    Unless you can fault the full metric in practice, I don't see what your point is.

    In reference to your response to DDignam:

    There are many factors in the game that aren't individually measured in WP, for example: on the ball defense, denying the other team good shots, getting back in transition, and passes to open players that miss their shots. The factors listed in WP stand in for many things they don't directly represent.

    A player that gets an extra 5 rebounds a game while the rest of the team gets 3 less IS a net of 2 rebounds, but which of the unmeasured factors and measured factors are also modified? Are other players crashing the defensive boards less and getting out in transition for easy buckets? Are the guards able to step back on offense, protect against easy buckets, and force the opponent into the half court (with a larger chance at a block, steal, or missed shot-another rebound)? Does that one player affect the other teams strategy and limit their ability to get out on the break or protect against the break?

    Despite how the metric was created, the metric works. I don't think we'll ever be able to divorce philosophy from player performance completely, but do you have anything that's more accurate?

  19. Ignore my last sentence. That was horrible on my part. I should have closed with something about the individual pieces working as surrogates for a larger scheme, and the stability of the measure attesting to its worth and accuracy in splitting player value.

  20. tgt:
    Thanks for the reply. I was just trying to keep my response concise, not ignore your point about stability. Yes, players' WP48 is fairly stable from year to year. But points scored is also quite stable, yet that doesn't make it a good way to rate players. WP48 is stable mainly because it weights rebounds very heavily, and players tend to keep getting the same number of rebounds each year. And more generally, because NBA players' boxscore stats are stable -- so any metric using those stats will be stable. But if WP is misvaluing rebounds, or undervaluing high-volume shooters, then being repeatedly wrong in a "stable" way is not much of a virtue.

    And let's not exaggerate WP's stability. If a high-rebound player joins a new team, the rebound rates of the other players decline accordingly. Now, he usually only takes a few rebounds from each of his individual teammates, so while their WP48 falls a bit it's not a huge amount for any one player. The year-to-year correlation for the league still looks pretty good (since most players don't change teams), but in fact there's been a significant redistribution of value.

    You say that WP "predicts total wins for teams accurately when adding and substracting players." I agree with you that this is the key test for WP and all other metrics: can it predict team performance when personnel changes? However, I don't believe there is any evidence to back up your claim. The one study I know of that examined this issue found that WP did a worse job than other metrics, and IIRC it did even worse than Points Scored alone. What evidence do you have on this point?

    You say the metric "works." But we have no idea if that is true in terms of evaluating individual players. Sure it works for teams, because the sum of players' WP48 equals the team point differential, and we know that predicts wins extremely well. But we don't need WP to tell us the Lakers' point differential -- we know that. We need it to tell us how much Kobe, Gasol, et al contributed to that differential. And there is essentially no evidence whatsoever that it does that job well.

  21. @Guy,

    It's amazing having a civilized discussion about statistics.

    Yes, points scored is stable, but it doesn't correlate with team wins. I believe the stat we agreed on correlating with team wins was net points, but that one isn't stable. You didn't create a counterargument here.

    Which study examined WP against other metrics (including points scored)? I'm rather late to the game, but I was going off Arturo's recent work on expected strength of teams this year, matched with some of my own spot checking, and what appears to be conventional wisdom around here. I didn't do a thorough check of Arturo's methodology, my spot checking isn't worth much, and conventional wisdom likely even less so I very well might be overestimating stability. The study you mentioned would be nice.

    Anyone else have any studies, or willing to help put one together?

  22. I'm not sure what you mean when you say Points doesn't correlate with wins. It correlates very highly with wins (much more than, say, rebounds). In fact, you could basically remove FGA from WP, ignore shooting efficiency entirely, and get just as good a correlation with wins (but your assessment of individual players would be MUCH less accurate). Anyway, my point is that pretty much all the main elements of a boxscore stat like WP -- rebounds, points, shooting efficiency -- are stable from season to season. So any metric based on them will be pretty stable, whether it's a good one or not.

    Arturo's comparison of metrics will be interesting to follow. It's a good year to do it, with all the team-switching. But if someone wanted to tackle this, you don't have to wait to see how the story turns out. You could use old data, and see how well 2005 WP, Statistical +/-, Win Shares, etc. "predict" team records in 2006. Even better, use a two-year lag to get more personnel changes. And actually, you should use Win Score rather than WP, just to set aside the whole "team adjustment" controversy. Then all metrics are on the same playing field, using only individual boxscore stats. Would be interesting to see....

  23. Points doesn't correlate well with wins. I don't have a graphic calculator or software on me, but it's pretty obvious from just looking:

    That team rebounds correlate less doesn't mean much to me. I'd also like to see some evidence that the modifications you suggested to WP to improve correlation with winning would actually improve correlation. Looking at its basis in wins correlated with points scored, this appears to be just as much assumption as my WP stability belief.

    I agree that waiting isn't necessary. The information's there for enough years that we should be able to verify or disprove stability.

  24. To clarify, the point of the analysis we're talking about isn't to verify "stability," it's to predict wins.

    The correlation between points and wins (or points allowed and wins) is of course reduced if you don't adjust for pace(that's true for all statistics), but it's still there. In your 2010 table, for example, the correlation with PF is .45 and with PA is .60. But if you look at points per possession, which is what we care about, then the correlation last year was .78, quite high (and about the same for PA per possession). Trust me: scoring points helps win games.

    The low correlation with rebounds should actually concern you. Dr. Berri's claim is that securing possessions is a very important part of winning in the NBA, perhaps the most important. But if that were true, it should be the case that winning teams are good rebounding teams (in general), and losing teams are not. That's pretty much the definition of being an important factor, right? But rebounding is NOT what distinguishes good and bad teams:
    10 Best Teams: .264 ORB%, .742 DRB%
    10 Worst teams: .258 ORB%, .726 DRB%
    That's a difference of less than 1 rebound per game between the best and worst teams. Instead, it's scoring efficiency and opponent scoring efficiency -- these are MUCH more important for winning in the NBA. Overall, possessions explain about 15% of the difference between good and bad teams, while shooting efficiency (O and D0 explains about 80%.

    So another good test for WP, PER, WS, etc. at the player level is how well they correlate with shooting efficiency vs. possessions. WP correlates very highly with rebounds/possessions, much more than it seems like it should if it is capturing what really produces wins in the NBA.

  25. the importance of rebounding has to do with possessions. if teams score on one point for every three of their possessions (i'm guessing this would be very low), then a rebound equates to 2/3 of a point because (1/3 of a point that your team scores plus the 1/3 of a point that the other team did not score). basketball is not so much about scoring as it is about scoring more than the other team. rebounding helps by not only giving you chances to score, but also taking away the other team's chance to score. using this logic, it is very easy to understand why wins produced values rebounds so much.

  26. Anon: no one disputes that rebounds are valuable. The problem is that NBA teams all rebound at a roughly similar level, or more precisely, within a very narrow range. If there isn't a big difference, then by definition it can't explain wins and losses.

    Suppose every NBA team had a FG% of exactly 45%. Then we would have to say that shooting efficiency explains/predicts nothing about who win games in the NBA. And that would be true whether the number was 35% or 45% or 55% (although total scoring would change dramatically). Things only matter when some teams are much better than others. And NBA teams simply aren't that different in rebounding ability.

    In baseball, people used to think that stolen bases were very important. Then Bill James discovered that good teams and bad teams stole the same number of bases, and people realized it wasn't as important as they had thought. (In contrast, good teams hit a LOT more HRs, which are important.)

    This doesn't mean it's not good to get rebounds -- it is. And on rare occasions, a team really stands out there. But in the big picture, it's a pretty small part of success in the NBA. The evidence is really unambiguous.

  27. i know about bill james (i've also read moneyball). basketball is not baseball. there are a lot more moving parts and variables. outside of point differential (in a very vague and basic sense), you can't just look at one stat and determine which teams are the best. also, point differential itself doesn't explain WHY teams are better. rebounding may be fairly similar from one team to another, but if one team rebounded much better, then they would have a huge advantage compared to a weak rebounding team. however, this is only a single variable. wins produced is not solely focused on rebounds. it focuses on multiple variables and converts proficiency at all relevant variables into a single understandable number. normally, i am against aggregate stats, but there is a high correlation from wins produced to actual wins.

  28. Thank you Anonymous, that is what I was getting ready to respond with on the Team Stats post.

    FG%, 3pa/fg, pace, fta/fga, TOs, fg defence - all affect the availability of rebounds. Good teams that are weak in rebounding may make up for it in other ways.

  29. Anon: I'm not sure what you mean when you say "rebounding may be fairly similar from one team to another, but if one team rebounded much better, then they would have a huge advantage compared to a weak rebounding team." If rebounding is similar, then there aren't any teams with a huge advantage. That's the point.

    Devin: Of course you are right that a weak rebounding team could have other strengths and thus still be good. But if rebounding is important for winning in the NBA, then good teams should -- on average -- have a big rebounding advantage. And the best rebounding teams should - again, on average -- be very successful teams. This is really definitional, not a matter of personal opinion.

    But none of those things are true. Winners don't rely on rebounds, and rebounding doesn't lead to winning. In contrast, the relationship to shooting efficiency is very strong: winning teams are MUCH more efficient. Why can we see the importance of shooting, while rebounding remains invisible?

    And if you don't believe that a statistical association with winning matters, then I'm not sure why you believe in WP, since it's based entirely on the same principle of statistical association.

  30. obviously he is not a 5 but an bench warmer long 4 with no reb and no def. skills at all.
    He's soft and belongs on the Eurolegue Developement league (if that even exists).
    Lets trade is Pasta harsss while we can for a 1st round pick nxt yr and rebuild a franchise around DD and other rookies jus like OK.