Draft Rankings

[jbrocato23]

sbs,

I'm using my own SPM variant. It's pretty similar to ASPM though.


Barncore,

Here is last year's top twenty. Only Davis, MKG, and Drummond were better than +1, and only Davis was better than +2.

Anthony Davis
Michael Kidd-Gilchrist
Andre Drummond
Jared Sullinger
Bradley Beal
Dion Waiters
Maurice Harkless
Damian Lillard
Terrence Jones
Harrison Barnes
Tyler Zeller
Meyers Leonard
Quincy Miller
Jae Crowder
Royce White
Kendall Marshall
Thomas Robinson
Jeremy Lamb
Draymond Green
John Henson

[jbrocato23]

Draft rankings now updated after the early entry deadline: http://shutupandjam.net/draft-rankings/

[bchaikin]

i don't see kansas state 6-11 C jordan henriquez (a senior) in your list of top 100 prospects. his stats his last 2 years were similar to but better (better rebounder/shot blocker) than those of say fab melo (they had the same rate of fouls committed), who was a 1st round pick last year...

where would he rank in your ratings?...

[jbrocato23]

i don't see kansas state 6-11 C jordan henriquez (a senior) in your list of top 100 prospects. his stats his last 2 years were similar to but better (better rebounder/shot blocker) than those of say fab melo (they had the same rate of fouls committed), who was a 1st round pick last year...

where would he rank in your ratings?...

bchaikin,

Obviously I can't account for every single player in the entire NCAA, so my rankings are limited to players who appeared in the top 100 at draftexpress.com and chad ford's top 100 throughout the season plus a handful of others that were super successful according to my NCAA impact metric. Anyway, I looked at Henriquez, and my model projects him at -4, which is actually good for 54th overall in my top 100, which I will update. If anyone else wants to suggest seniors that they think might be productive, let me know, and I'll gladly add them to my database.

James

[jbrocato23]

Updated the model: http://shutupandjam.net/2013/05/24/draft-model-update/


And, as always, here are the current rankings: http://shutupandjam.net/draft-rankings/

[Barncore]

Love the new draft model Jacob. Truly elite. Just wondering why you've chosen to go with college stats fromm 2002-2009? Why not go back further? One would think that the bigger the sample size, the better..

[jbrocato23]

Love the new draft model Jacob. Truly elite. Just wondering why you've chosen to go with college stats fromm 2002-2009? Why not go back further? One would think that the bigger the sample size, the better..

Barncore, this is certainly something I will consider in the future, but for now I have a pretty decent sample size and I don't think going back further will make too significant of a difference.

P.S., it's James.

[chilloutbro]

Do you still have pre-2010 draft ratings available on your site Jacob?

[Barncore]

Do you still have pre-2010 draft ratings available on your site Jacob?

I would also love to see pre-2010 draft rankings.

(and lol @ chilloutbro getting his name wrong the post right after James corrected me)

[chilloutbro]

Yeah oops. Another question though, sample size is obviously an issue in college, a guy like Otto Porter shot 42% from 3 after a 22% freshman year and is rated is a virtual lock starter by your metric. I'd like to know if it would affect the numbers greatly if you adjusted some stats up or down. If it does, it'd be nice to see some case studies. Just an idea.

[Nathan]

Out of curiosity, how do your raters deal with things like 3 point percentage? Do you just give it the raw percentage if the player has some threshold minimum amount of attempts? Do you give it a percentage and a standard deviation? Or do you give it the numbers for 3 pointers made vs. attempted and it automatically calculates the standard deviation?

Also, people talk about adding new parameters post-hoc, which I don't really understand. This means you calculate values for the majority of parameters, freeze them, then add new parameters and run a new regression? Why not just have those parameters in there in the first place? If you add new parameters, the best fit values of the old ones would likely change to some extent, right? Doesn't that mean you're not going to get the best possible fit, then, with these post-hoc adjustments?

Thanks for making these raters guys; I've really enjoyed reading them. Doolittle's "advanced" stats over on espn are just comically bad in comparison. It's nice to have people who know what they're doing.

[jbrocato23]

Chilloutbrah,

Generally speaking, adjusting any given stat is not going to make a significant difference. Adjusting stats to make a 30% three shooter a 40% shooter, for example, would result in a change of less than a tenth of a point in most cases.

Nathan,

I don't use shooting percentages in the model. Instead, I use things like 2pa and 3pa. Also, I don't use any post-hoc measurements, and I think your understanding of the implications of doing that are correct.

Barncore,

I'll probably get these numbers up. My concern is that they are in sample and thus not as useful as the out of sample results.

[Nathan]

Thanks for the info, and I think you're worrying too much about publishing results from in sample. Removing any one player, even a major outlier, from your sample shouldn't result in more than a +/- 0.1 change in any player's rating, which is (probably) small relative to the uncertainty you're working with anyway. Don't take my (highly questionable and inexperienced) word for it though; if it's not too time intensive experiment with removing an outlier from the sample, re-running the regression, and seeing how much their rating changes. If you really want an upper bound, create an imaginary super-outlier player and see how different their rating is with them in sample and with them out of sample. If you can confirm that ratings for in sample players are reasonably accurate, just publish them and add a clear disclaimer that these are not true retrojections in the way that the 2010-2012 results are. We would all be thrilled to see them

Unrelatedly, have you thought about calculating ratings for players if they had entered an earlier draft? For instance, how Porter would have rated in the 2012 draft (just using his freshman season data), or how the draft stock of Wolters changed over his four years? I'm not sure how useful this would be from a practical standpoint, it's just an avenue I haven't seen anyone go down before, so if you're also curious and it wouldn't be too time intensive I think it would be a cool way to get more interesting results out of your model.

[VJL]

If you really want an upper bound, create an imaginary super-outlier player and see how different their rating is with them in sample and with them out of sample.

Nice idea. Another approach is to the run the numbers such that it fits the parameters for each player independently using all players except ego. I did this with earlier version of my projection model, but I found that it didn't really make a difference so stopped bothering (which supports your point that 'in sample' isn't a big problem).

[Barncore]

Still very curious to see the pre-2010 rankings. I know they're in sample but it would be very interesting nonetheless.