Nearly two years ago, I caught the argument about which of the three rookies -- Grant Hill, Jason Kidd, and Glenn Robinson -- should win the Rookie of the Year award. At that point, I looked at my numbers and the winner was quite clear. Below are segments of my comment and a response that criticized the use of statistics to evaluate players. At the end, I answer a number of the criticisms.
My comments on the Rookie of the Year> > >Brad Umland (S1082650@cedarville.edu) wrote: > > >: It'll be *very* close. I hope that people take Kidd's sub-40% shooting > >: percentage into consideration as well. There is no way that Kidd has been > >: as consistent as Hill or Robinson this year. > > >I hope that the voters take the fact that Kidd has been thrown into the > >fire at POINT GUARD from game 1. Point guard is a much more difficult > >position to learn, especially for a guy with two years of college and who > >starts the first game. His shooting has not been that good, but that's > >secondary for a point guard, isn't it? > Ah, the problem of how to balance a point guard's shooting with assists. One way to balance these things is with Tendex, but I prefer to look at how they contribute to the team's points per possession rating. Below are stats for Hill, Kidd, and Robinson that I will explain,
Scoring possessions are roughly the number of team scoring possessions (possessions on which at least one point was scored) the player contributed to via a made shot, an assist, or offensive rebound. These are weighted according to difficulty. Poss. is the number of total possessions used by the player either as a scoring possession or in a wasted possession. Floor% is the player's scoring poss. divided by total possessions. Rtg (rating) is how many points produced by a player per 100 total possessions used. Then points produced is the points produced by a player thru assists, shots made, offensive rebounds. It is usually about twice scoring possessions, depending on whether the player makes a lot of three pointers and their free throw ability. These numbers can be measured, but I estimate them through a rather complex procedure. Anyway, what they say is that Hill is having the most efficient rookie season, while Robinson is producing the most points. Kidd falls in the middle efficiency-wise. All players have improved through the season. Hill's efficiency is pretty good for a rookie; certainly not Jordanesque. Jordan had a floor % of about 57% his rookie year and produced more points than these guys. Note that these numbers do not include defensive contributions. It is important, but I only have an approximate technique for estimating the player's defensive rating, or how many points per 100 possessions he gives up. For Hill, it is 108.6. For Kidd, it is 106.4. For Robinson, it is 108.8. (Strictly, I should not have all these significant figures, especially for the defense.) Putting these together with the offense, it should be pretty clear that Hill is best. You can just take the difference between offensive and defensive ratings or you can use the formula This shows Hill with a 0.437 win%, Kidd with a .302 win%, and Robinson with a 0.140 win%. Given the records of their teams, it makes sense that they don't have great percentages. I will point out that last year's ROY was a much closer affair statistically. Neither Hardaway nor Webber were as good offensively as Hill is this year, but they played in a league where hand checking was OK and the three point line was farther away. Both players were about 0.500 players last year. Hardaway is now an 0.823 player and Webber is a 0.406 player. | ||||||||||||||||||||||||
A Response From Michigan> Sc. Poss. Poss. Floor% Rtg Points Prod. >Hill 586 1082 0.541 106.9 1157 >Kidd 451 926 0.487 101.2 937 >Robins. 672 1420 0.473 97.4 1383 More statistical calculations. How does this help? Does it take into account who a player is passing to? For instance, it doesn't take a lot of great passing to get an assist when you hit Allan Houston for a jumper. It takes a hell of a lot of effort to get an assist passing to Mark West. How does that change your rating system? Does it change your rating if you keep hitting shooters with great passes and they keep missing? Does that make you less of a player? And what about rebounds? What about blocks? Are those disappearing into your unexplained defense calculation? And who cares how many p/g a player gives up. If they are all covering the same man, then that's great, but they aren't. If one of these players is a defensive weakness, then won't that person give up artificially low p/g against, as the coach keeps him off the other team's best scorer? And won't the best defender have an artificially high p/g against, as he is given the most difficult assignment game in and game out? I put big money that if you use this statistic on every player in the NBA, you will come out with some laughably funny results as to who is better than who. Dikembe will drop like a rock, but we saw last year what his presence can do. Stockton, on the other hand, would love it if awards and salaries were doled out according to this statistic. But stats don't win basketball games - so until we can get a hotline to the great basketball divinity somewhere, we will have to continue to use our eyes, minds and instincts to decide who is best, and we won't always agree. So stop trying to prove it to me with inane statistics, thank you. |
I will answer the criticism point-by-point because the Michigander makes some good ones and I think the answers deserve to be heard.
> Sc. Poss. Poss. Floor% Rtg Points Prod. >Hill 586 1082 0.541 106.9 1157 >Kidd 451 926 0.487 101.2 937 >Robins. 672 1420 0.473 97.4 1383 > >More statistical calculations. >How does this help? Does it take into account who a player is passing to?
In theory it does. As I said in the original post, assists are weighted by how easy the assisted shot is. I explain the theory behind this in The Fundamentals for Analyzing Basketball and in Jordan vs. Olajuwon: Who's Better?. But to state it briefly here, an assist is more valuable if it sets up a player with an easy shot than if it sets up a player with a hard shot. This is why inside players are guarded more tightly than someone above the three-point line. This is why we typically consider "outstanding passes" to be those that create an easy shot, rather than those that yield jump shots from 25 feet. This is why we notice more frequently who made the pass to an inside player than who made the pass to a jump shooter. In theory, this is what my methods do. They do not yet completely do this, but they approximate it.
>For instance, it doesn't take a lot of great passing to get an assist >when you hit Allan Houston for a jumper. It takes a hell of a lot of >effort to get an assist passing to Mark West. How does that change your >rating system?
It doesn't. The assist to West is more valuable than the one to Houston. This, as explained above, is because Mark West is much more likely to make his shot from 2 ft than the errant Houston is from 20 ft.
>Does it change your rating if you keep hitting shooters >with great passes and they keep missing? Does that make you less of a >player?
Theoretically, yes, it makes you less of a player if you "keep hitting shooters with great passes and they keep missing". If you are making great passes to players who are consistently missing, then you are making the wrong choice in giving them the ball -- it doesn't make sense to pass to a 0% shooter, even if they are wide open under the basket. The great players know that it doesn't make a lot of sense to thread a needle to someone who can't shoot from that position. Yes, sometimes players miss shots that they normally would make, but in the long term, "sometimes" balances out.
>And what about rebounds? What about blocks? Are those disappearing into >your unexplained defense calculation?
These are in the calculation. Yes, they do "disappear". One of the things I have tried to do with my work is to prevent too much from "disappearing". It is important to get the big picture with statistics like floor percentage and defensive ratings, but it is also important to keep track of the details by remembering that steals are different from blocks. Summary statistics like defensive ratings or TENDEX do hide these details for the purpose of getting the big picture. Although TENDEX does it fairly arbitrarily, my defensive ratings have a solid theory behind them, as laid out in either The Fundamentals for Analyzing Basketball and in Jordan vs. Olajuwon: Who's Better?.
>And who cares how many p/g a player >gives up. If they are all covering the same man, then that's great, but >they aren't. If one of these players is a defensive weakness, then won't >that person give up artificially low p/g against, as the coach keeps him >off the other team's best scorer? And won't the best defender have an >artificially high p/g against, as he is given the most difficult >assignment game in and game out?
All good points. Doug Steele has a method for getting away from this bias. I have been working on a method myself, one that is based more in theory than Doug's.
>I put big money that if you use this statistic on every player in the >NBA, you will come out with some laughably funny results as to who is >better than who. Dikembe will drop like a rock, but we saw last year what >his presence can do. Stockton, on the other hand, would love it if awards >and salaries were doled out according to this statistic.
Dikembe Mutombo's offense looks horrible, but his defense looks pretty darn good, making him an overall valuable player by my numbers, not as good as David Robinson or Hakeem Olajuwon, but important to any defense. You can see that with his trade to Atlanta: the Hawks team defense has improved a lot (exactly average last year and 4.8 points per 100 possessions better than average this year, through 12/29), but their offense has gotten worse (1.3 better than average last year and 0.2 worse than average this year). You can also see this in Denver's statistics: the Nuggets offense has surprisingly improved somewhat this year (2.9 worse than average last year and 2.2 better than average this year), but their defense has become the worst in the league (exactly average last year and 7.8 worse than average this year
Stockton does look good with my method, which the Hall of Fame will confirm. I have never come up with a season where he should win the MVP or even one where he's been any better than Karl Malone, but he is always very very good.
>But stats don't >win basketball games - so until we can get a hotline to the great >basketball divinity somewhere, we will have to continue to use our eyes, >minds and instincts to decide who is best, and we won't always agree.
As a former coach and scout, I agree that we have to be qualitative. The tools I have developed are aids to making distinctions among players, not answers in themselves. Sometimes, these tools make clear distinctions, as they did for the '95 ROY. Two years later, I'm pretty happy that they picked the best of the three, with Kidd recently getting traded, Robinson still not great but rapidly improving, and Hill turning into a valid superstar. When these tools don't make as clear a distinction, then we have to rely on more qualitative descriptions.
> So stop trying to prove it to me with inane statistics, thank you.
My so-called "inane statistics" correlate at about 95% with team winning percentage. So unless you think winning percentage is inane, these statistics have a pretty good track record.