Wednesday, March 21, 2007

Analysis: Floriani gets analytical

Ray Floriani, our man in N.J., checks in again with a stats bent:

"A little note from a statistical perspective.

In the late 90s, Arthur Linton authored the book, How To Grade And Rank Every Basketballer In History. The book deals with a formula to find the best players and distinguish various levels of expertise. The result is the EF (effectiveness factor). The abbreviated formula which allows for quick in game calculations on the bench is

(PTS + REB + Assists + Blocked Shots + Steals) – Turnovers = Total

Total/Minutes Played = EF

A quick rundown on the Georgetown players who averaged over 10 minutes per game follows. Statistics used were for Big East Conference games only. I like to use conference stats because you can get a better ‘read’ as a player (especially in the Big East) is challenged virtually every night out.

Roy Hibbert 0.804
Jeff Green 0.699
DaJuan Summers 0.517
Patrick Ewing 0.482
Jonathan Wallace 0.472
Jessie Sapp 0.470
Vernon Macklin 0.439
Jeremiah Rivers 0.221

What do these numbers mean ? Linton set up a scale in his book. A quick breakdown:

1.000+ Superstar
.900 - .999 Stars
.800 - .899 Very Good
.600 - .799 Good
Under .600 Average

You can do these evaluations at halftime and for a game. My own opinion is Linton is a bit tough on the upper rankings. In doing these calculations for about a decade I found .900 or greater a tough mark to reach. Labeling Jeff Green as just ‘good’ is a stretch. Still, the formula is not a final say. It’s another tool of analysis that will hopefully broaden our insight."

My own comments
My usual mantra when presented a stat like this is that, since it based per minute rather than per possession, you end up punishing players in slow-tempo offenses (like Georgetown's). I think this what is causing the difference between the individual player stats and scaled rankings, which, I believe, are based on NBA players. A quick glance at some stats over at indicates that NBA games averaged ~92 possessions in 2004-5. Since the Hoyas play about 60 possessions per game, you could reduce the scale to 2/3 (≈60/92) to make a comparison:

Edit - a reader (I have a reader!!) noted that I forgot to account for the difference in minutes played between college and the NBA (40 vs. 48), so I've adjusted the table accordingly

.782+ Superstar
.704 - .782 Stars
.626 - .704 Very Good
.469 - .626 Good
Under .469 Average

Now this looks reasonable:
  • Roy and Jeff are the Superstars that Ray wants them to be (with the edit, Jeff slides down just under the division into "Very Good" category)
  • Almost all of the remaining players are now coming in as "Good", with Dajuan threatening "Very Good" (now the next four are solid "Goods" - sorry Dajuan)
  • J. Rivers is once again getting hammered, as it's an offensive metric

Having said all of that, I have to admit that this is a fun stat. Of course, by working in possessions, rather than minutes, it makes it a bit harder to work out on-the-fly.

For kicks, I thought I'd run Vanderbilt's 10+ minute players, but to make the two teams comparable, I've scaled their EF's by 60/68 (the ratio of avg. possessions for G'town and Vandy, respectively). Also note that I'm using full season stats here, as I don't have access to SEC-only, so these may be a bit inflated by cupcake games.

Byars, Derrick 0.691
Foster, Shan 0.615
Neltner, Ross 0.600
Cage, Dan 0.508
Gordon, Alex 0.482
Skuchas, Ted 0.421
Drake, George 0.405
Brown, JeJuan 0.401
Beal, Jermaine 0.363

Other news
Following on the news of Joe Scott's departure from Princeton and the resulting rumors surrounding certain G'town assistant coaches, it appears that Coach Kevin Broadus has landed a head coaching job at Binghamton University. The Bearcats were previously coached by Al Walker, who resigned on March 5th.

Congratulations to Coach Broadus!


  1. Points is an imperfect metric because the formula doesn't subtract missed shots, which obviously carry an opportunity cost (though less of an opportunity cost than turnovers). You'd prefer, for instance, someone who scored 10 a night with 1.25 PPWS than someone who scored 20 on 0.8 PPWS, since on the extra 17 possessions his teammates would probably produce enough points to more than make up the difference. Similarly, you'd prefer someone who in 10 possessions shot 3-of-10 than someone who shot 3-of-3 and turned it over 7 times as long as your team's OR% was greater than 0.

    In short, tempo-freeing the stat by adjusting for possessions instead of minutes is good, but ideally you'd also adjust for offensive efficiency and quality of opposition (our boys' average opponents have Adj DEff of 88.1, Vandy's of 94.3, so you could come up with a reasonable rubric that didn't say the kid from Jackson State is the best offensive player in the country by making some adjustment that way).

    Go Hoyas.

  2. Okay, maybe I'm missing something, but if the formula itslef is a ratio in which minutes played is the denominator, why does it make any sense to reduce the scale because there are fewer minutes played in the ncaa?

  3. Re: anonymous

    I certainly agree that PPWS is a better measure of scoring efficiency, and Adj. Off. Eff. is better still for offense as a whole. I think Ray was looking for a stat that accounts for the whole offensive performance (not just scoring) but that could be easily calculated on the fly (e.g. during halftime).

    Re: hoya90

    Actually, two adjustments were made: first, I changed the scaling to account for the difference between NBA and G'town possessions (60/92), then changed the scaling again to account for the minutes difference (48/40). So, the scale was actually increased due to fewer minutes, but after it had been lowered to account for 2/3 as many possession opportunities.

    Thank you both for your comments!

  4. To the first comment: agreed. When I noted the pace/minutes adjustment, I pointed out the same thing. It's a crude, unadjusted metric, and the only one of those for which I have much affection is +/-.

    Recent/notable years in Georgetown history, not adjusted for pace and not including TOs (numbers from here:
    Ewing, 1981-82: .874
    Ewing, 1982-83: 1.046
    Ewing, 1983-84: 1.000
    Ewing, 1984-85: .979
    Mourning, 1990-91: .922
    Mutombo, 1990-91: 1.007
    Mourning, 1991-92: 1.195
    AI, 1995-96: 1.137
    Victory Page, 1996-97: 985
    Sweetney, 2000-01: .971
    Sweetney, 2001-02: 1.092
    Sweetney, 2002-03: 1.23
    Bowman, 2003-04: .841