For those who love stats... NCAA Player Success Rates

He is considering scoring in his similiarity metrics in the form of shooting efficiency (2pt eff, 3pt eff, ft eff). Exactly how that's tallied I don't know. Compared to the other stats he's using I'm guessing it's a similar ratio of some sort. Probably the reason he's not using PPG is because it's not normalized to give you an apples to apples comparison like a pts-per-48min does.

Durvasa is right in his last comment. The similarity scores are only to group the players and aren't an indicator of their success in the NBA. Once they were grouped he ran other calculations to rank how successful they were in the NBA and then their probability of being a success. So in the case of Nash = Tyron Lue, he's just saying they were similar when comparing their college stats he's using. The grouping has no indication of how successful they are. You would have to look at how he did his quality index / success-rate scoring on Nash & Lue to see this but one might guess by his other calculations that Nash was probably rated a 1 while Lue was rated much lower than that in his quality index etc.

The part I don't get is how he did the correlation for the 2006 draftees to estimate what their probability of success would be.

 

I didn't think the stat calculation was that bad. They don't use pts as a stat b/c of the philosophy of most stats guys, which is that efficiency is more important than raw output. This isn't a perfect assumption, since it's easier to shoot a high percentage as a role player than as a go-to scorer. Still, the article writer does consider minutes played, which should indirectly account for total production. I'm not up-to-date on the best methods for recalibrating efficiency as a function of minutes played, so I'm willing to accept his method for lack of a much better alternative.

It would be nice to see some re-weighting of the stats categories, to take into the fact that 2pt fg efficiency is probably of more interest than fouls per minute, though.

Also, re: the point on Mourning vs Harrington -- these guys were similar players, though of course Mourning is at one of the extremes of the distribution. A sharper analysis should analyze success rates based on how similar the potential draftee is to each of the NBA players, not just whether he is in the same cluster.

In any case, I like Brewer as the pick a lot. He has high overall production, and noticeably better efficiency numbers (2005-6 FG 44%, with previous seasons both higher) than Carney (44%, but the previous seasons were worse), Foye(41%), or M. Williams (41%). Roy, on the other hand, is at another level. He shot 51% last year, and in his junior year he shot 57%, as a shooting guard! Brewer (2.6 steals per game), however, owns Roy in defense (1.4 steals). 2.6 steals is insane, especially given that college games are shorter than NBA games.

 

Brewer FTW!

 

Based on this Swift should be equal to or better than pre-injury Amare and Yao should be another Manute Bol, Shawn Bradley, Chuck Nevitt, Muresan or even worse.

 

I emailed the author, and I got some feedback.

Regard the missing "points scored" category, it's actually in there. He incorporates both quantity and efficiency into the 2-pt, 3-pt, and free-throw categories. That is confusing, because he refers to it as "efficiency". But, apparently, that's not all it says. As a result, having a separate "point per minute" category would be redundant and would skew the analysis.

The final metric is:

minimum success rate, if quality index < 0.5
minimum success rate * quality index, if quality index >= 0.5

He says he did this for last years draft, and he was able to accurately predict success with 2/3 accuracy.