Thursday, July 17, 2008

2009 Player Projections: UVa, NCSU, and GT

I've uploaded the player projection reports for Georgia Tech (Mo Miller is poised for a breakout), N.C. State (the real Brandon Costner was the one we saw in 2007; Ben McCauley's great sophomore season looks like it was a fluke), and Virginia (a healthy Solomon Tat could be the Wahoos' best player). Boston College, Florida State, Miami, North Carolina, and Wake Forest have all posted their '08-'09 rosters on their official sites, so I'll have them up relatively soon.

There's a new section in the reports this year, and it's the key to the player comps. Each player-season in the database has been assigned these eight numbers, and each number represents a facet of the player's game:
  • Size is height, with an adjustment for bulk using Body Mass Index (BMI). Skinny players tend to play shorter than their height, and bulky ones play taller. Each standard deviation of BMI adds or subtracts an inch from height.
  • Physicality is Free Throw Attempts plus Personal Fouls, divided by 1 plus the ratio of 3-Point Attempts to Field Goal Attempts.
  • Activity shows how often a player does something that shows up in the box score.
  • Shooting is the average of: FT%; 3G% times 3GM, divided by 3; and 2G% times 2GM, divided by 8.
  • Scoring is simply Points scored.
  • Passing is Assists minus Turnovers.
  • Rebounding is the number of Rebounds recorded.
  • Defense is height-normalized Blocks plus Steals.
All the numbers used are per-40. Each category is then normalized to a 0-to-100 scale. The similarity scores are based on a combination of the correlation and the closeness of the two players' profiles. The projections based on the new system are about 25% tighter on average than the ones from a year ago.


Steven said...

Nice work as usual, Vince. Love this stuff.

One question: what's up with Courtney Fells's minutes projection?

Vince said...

I think Fells might have slept with the computer's wife or something. Courtney's comps lost about 10% of their playing time on average from junior to senior year, and each of his top 9 comps in the regression saw at least a 15% reduction in playing time from their junior to senior season, so it must just be that his regression hits all the indicators for loss of playing time.

Floor time is ridiculously unpredictable. It doesn't really correlate all that strongly with anything in particular, probably because there are so many external factors--injuries, coaching preferences, positional logjams, etc.--that there's just no good way of predicting it. Even the players with the tightest minutes predictions have a standard deviation of about 6.5 minutes per game.

That said, there is a tendency for coaches to reduce--sometimes drastically--the playing time of decent-but-not-great seniors in favor of underclassmen who might not be as good or as consistent in hopes that the playing time will develop the younger guy. Tracy Smith, Johnny Thomas, and CJ Williams will probably all squeeze Fells' minutes a little bit.