APBRmetrics

DSMok1 · Last edited by DSMok1 on Tue Oct 27, 2009 9:14 am; edited 1 time in total

NCAA Statistical Plus/Minus 2008-2009

I realized this morning that Statistical Plus/Minus was an ideal tool for evaluating NCAA players. It does not have to deal with the issues related to competition like other box-score stats do--for I can force the team's totals to equal their efficiency margin, as calculated by Ken Pomeroy. I already have on hand every NCAA player's box scores, heights, and approximate ages from ESPN (I calculated Wins Produced for NCAA players, but realized there is no coherent way to adjust for opposition), so it wasn't hard to do.

I used the update for statistical plus/minus calculated by Neil Paine over on the Basketball-Reference Blog. Mostly, the calculation was straight-forward. The difference was that adjusting to force each team's total to match their efficiency differential was much more significant. I tweaked a couple of things, also--I scaled the NCAA minutes per game by 48/40, and scaled the ages upward also (so the mean and stdev would match NBA players). I did not use the latest iteration of statistical plus/minus (SPM), because I didn't need an age-squared term (age is positive, in general, in college...). Other than that, I just used the formula directly.

The standard error term was a puzzle (as Ilardi has noted). I ended up fitting an exponential curve to the data Rosenbaum provided to estimate the standard error term using minutes alone. (The result I had was stderr = 152.342min^(-0.76055)+1.954212.)

The stats are per 66.5 possessions (or 40 minutes of average-pace play).

Here is the Google Spreadsheet. I have discovered that Google Spreadsheets are a pain to work with! Just give me my Excel....

There are 3 basic ways to rate players with statistical plus/minus. First of all, the standard +/-. Of course, players with only a few minutes totally mess those numbers up. Secondly, there is the total points added by that player... probably the best overall rating. And finally, there is the confidence over average--the confidence (using the stderr of the measurement) that a given player is above average. That is how the sheet is arranged by default. Technically, a regressed SPM could also be compiled, but I haven't bothered.

Comments, everyone?

Here are the top 10 players by confidence over average:
1. Stephen Curry
2. DeJuan Blair
3. Lester Hudson
4. Ty Lawson
5. Blake Griffin
6. James Harden
7. Terrence Williams
8. Eric Maynor
9. Kenneth Faried
10. Tony Gaffney

and by total points contributed:
1. Stephen Curry
2. Lester Hudson
3. Blake Griffin
4. James Harden
5. Terrence Williams
6. DeJuan Blair
7. Ty Lawson
8. Talor Battle
9. Eric Maynor
10. Toney Douglas

EDIT: New Spreadsheet Update

I have updated the spreadsheet--the team adjustment is now applied to the individual players' SPM based on their minutes played. So if the adjustment is upward (like for a good team) the most boost will be given to the player with the most minutes. I used a recursive algorithm to force the adjusted SPM's to total the team's efficiency differential.

Spreadsheet HERE

jmethven · Joined: 16 May 2005 Posts: 51

Great stuff. I'd looked into calculating SPM for college players before for draft projections and was pleased with the results. Even though it's an NBA-derived regression, they certainly pass the laugh test.

Now the real trick would be developing a SPM projection system. Or for that matter, getting college APM data so there is something to compare to!

DSMok1 · Posted: Thu Oct 22, 2009 4:28 pm Post subject:

In looking at the results, it occurs to me that the shot creation squared term (TSA^2) is a bit over-powered in college because of the outliers it produces from disparities in competition quality. Thus very good players in lower levels (Stephen Curry, Lester Martin) have somewhat over-inflated numbers because of the inordinate number of shots they took. All other terms are linear, so they are not effected in the same way by outliers like this. It seems that this is the (somewhat minor) flaw with using the NBA weights for SPM for the NCAA.

tawtaw · Joined: 25 Jun 2008 Posts: 28 Location: Oregon

Yeah, that's pretty good. Hudson and Curry are great college players, though perhaps a bit overvalued by this metric. The kid from Morehead St. has some good all-around numbers, but I'm skeptical about a kid that scores 13pts per game against relatively weak competition being among the very best players .

Overall, really strong work.

Crow · Joined: 20 Jan 2009 Posts: 811

Any interest in appending strength of schedule estimates (existing or your own) or adjusting the SPM's by SOS?

DSMok1 · Posted: Fri Oct 23, 2009 8:11 am Post subject:

DSMok1 · Posted: Fri Oct 23, 2009 8:13 am Post subject:

DLew · Joined: 13 Nov 2006 Posts: 224

I think you're right in saying that the squared true shot attempts term is causing some funny things to happen because everything else, including the efficiency adjustment is linear. I think the solution to this would be to use a SPM formula that is entirely linear. I'm sure it would not be difficult for you to derive one, or for Neil to run his regression with the squared term removed.

Kevin Pelton · Posted: Fri Oct 23, 2009 11:14 am Post subject:

DSMok1 · Posted: Fri Oct 23, 2009 12:02 pm Post subject:

Qscience · Joined: 22 Jun 2009 Posts: 70 Location: Phoenix, Arizona

Scouting Lester Hudson I can never forget the game he put on against the Memphis Tigers. He had 3 NBA players double and triple teaming him and he still made them look silly at times.
He is going to be the next Eddie House once he gets his NBA shooting range down.

Crow · Joined: 20 Jan 2009 Posts: 811

tawtaw · Joined: 25 Jun 2008 Posts: 28 Location: Oregon

bchaikin · Joined: 27 Jan 2005 Posts: 687 Location: cleveland, ohio

...I just think that somewhere along the line the methodology is giving too much credit to small conference guys...

kenneth faried is the best rebounder in the ohio valley conference in some time, but a few of the best rebounders from the OVC have had a cup of coffee in the nba, including popeye jones and james singleton (both of murray state), and bob mccann (morehead state)...

actually faried's (6-8, 225 lb) and singleton's (6-8, 230 lb) college stats are fairly similar, faried the better rebounder with a better rate of steals, singleton a bit better shot blocker and the better overall shooter. faried was a bit better scorer per minute...

tawtaw · Joined: 25 Jun 2008 Posts: 28 Location: Oregon

Yeah, I don't doubt Faried can play in the NBA. When you talking about one of the top dozen players in the country though, I'd think we'd be talking about more than just a cup of coffee in the NBA. It'll be interesting to follow his progress this year for sure. I don't doubt that this metric has uncovered him as a bit of a hidden gem.