Author Message
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Sun Nov 05, 2006 5:24 pm Post subject: Brock2 Projections Reply with quote
Looking at some old Bill James Abstract stuff, I'm wondering if anybody's tried to do something similar to his Brock2/Brock6/etc. career-path projection system for the NBA. I have no idea where you would even begin to work up something like this for basketball (What stats would you use? Would different positions age at different rates, like in James' method?), but I thought I'd throw the idea out there to see if it had ever crossed anybody else's mind...
(Here's a link to the zipped Excel file of James' original system, in case you're not familiar with the method.)
Back to top
View user's profile Send private message Visit poster's website
ziller
Joined: 30 Jun 2005
Posts: 126
Location: Sac Metro
PostPosted: Mon Nov 06, 2006 1:57 pm Post subject: Reply with quote
I'd venture to say some are working on this type of stuff privately. As it is, it appears John Hollinger has a projection system in place that weighs both similars and aging impacts on key stats.
Positions do differ in the way and rate they "age." Ed Kupfer has done some work on this here in the forum (search "aging").
_________________
SactownRoyalty.com
tziller@gmail.com
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Mon Nov 06, 2006 5:29 pm Post subject: Reply with quote
Oh, I'm well aware of similarity -- I even used it to project PER's for my NBA Preview. I have also worked up some stuff on the effect of aging on PER:
But the beauty of James' Brockn system is that it doesn't directly involve similarity. Sure, the model was based on what similar players will do if they post x stat at y age, but when using it I'm not limited to past comparables. For instance, no 21-year-old has ever posted a PER of anything close to 28.01, so if I'm using straight comps to project LeBron James, I'll end up with Tracy McGrady at 22 (PER: 25.1), Michael Jordan at 22 (27.5), etc., and come out with a projection of 25.5. All great numbers, but LeBron is probably not going to regress to their levels this year. On the other hand, the Brock2 system would factor in age, maybe a little regression to the mean, but come out with a more reasonable age-22 projection that might build on 28.01, rather than drag LBJ's projection down because none of the comps were this good, this young.
For this reason, the most unique players would come out with much more accurate projections through Brockn than through similarity. Ben Wallace is another good example: I've got his PER dropping to 15.0 via similarity, but it hasn't been that low since 2000-01, and there's no real reason to think that it would slide all of a sudden (similarity thought it would drop last year as well, but it held steady). So I think that, in many ways, an algorithm-based projection system has more potential than a similarity-based one because it doesn't have to rely on the past for future projections. And, as an aside, I have no idea what the hell kind of system Hollinger used to get LeBron's 31.86 projection. How did that happen, using similarity?
Back to top
View user's profile Send private message Visit poster's website
THWilson
Joined: 19 Jul 2005
Posts: 164
Location: phoenix
PostPosted: Mon Nov 06, 2006 6:34 pm Post subject: Reply with quote
I never tried to confirm Hollinger's projections but I've always assumed that he:
1) finds comparable players
2) finds the ratio of (comp + 1year) / (comp)
3) finds weighted average of above ratios
4) applies ratio to subject player
Looking just at PER, McGrady stayed about steady from 21 to 22, Jordan improved by 7%, Bryant by 12% etc. Find the weighted average of these increases and apply it to 28.1.
I would think that you wouldn't want to try to project PER in isolation, but would have better results projecting points, rebounds, turnovers etc. and then calculating PER based on those projected values.
Back to top
View user's profile Send private message
ziller
Joined: 30 Jun 2005
Posts: 126
Location: Sac Metro
PostPosted: Tue Nov 07, 2006 4:30 pm Post subject: Reply with quote
Hollinger's projection system continues to boggle the mind. Not that it's bad - it's probably very good. But we may never know how he came up with the numbers he did.
And now that I've read your question more closely and looked into Brock more closely, I'm intrigued. As you say, this could help a bunch in basketball, which lacks the easier standardization of players seen in baseball and has far more unique cases in a constantly changing game.
_________________
SactownRoyalty.com
tziller@gmail.com
Back to top
View user's profile Send private message Send e-mail Visit poster's website
John Hollinger
Joined: 14 Feb 2005
Posts: 175
PostPosted: Wed Nov 08, 2006 11:52 am Post subject: Reply with quote
I explained the projection system in the last two versions of the book, but it's a bit length to stick on line. The previous poster had the basics correct -- find similar players, compare their rate of change in subsequent seasons, apply rate of change to players current stats.
The inherent weakness is that very old and very young players have few comparables (or none, in the case of Dikembe Mutombo and Alonzo Mourning), so a single fluky result can throw it way off. That's what happened with Dwight Howard and Josh Smith -- blame it on Darius Miles.
Back to top
View user's profile Send private message AIM Address
bchaikin
Joined: 27 Jan 2005
Posts: 689
Location: cleveland, ohio
PostPosted: Wed Nov 08, 2006 2:33 pm Post subject: Reply with quote
find similar players, compare their rate of change in subsequent seasons, apply rate of change to players current stats...
you have to wonder if many nba teams even do this...
case in point - the sonics just signed 26 yr old nick collison to a $33 mil contract (5 yrs), or $6.6 mil/yr. now the key factor limiting the playing time of big men in the league is their rate of committing fouls, and collison has committed 16.4 fouls per 100 minutes played during his short career. if you look at the history of the league (for at least the past 30 years), PFs and Cs in the league who committed fouls at this high of a rate their first couple of seasons in the league (playing at least 1000 minutes in each season) go on to average only about 22-23 min/g over the length of their careers, and very few play even as much as 2000 minutes in a single season (less than 25%)...
why would you want to sign a player for this kind of money knowing this? knowing that the odds of this player ever playing significant minutes is quite low? the lakers just signed a player very similar to collison in brian cook, but paid him only $10.5 over 3 yrs ($3.5mil/yr), and he commits fouls at a lower rate...
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Mike G
Joined: 14 Jan 2005
Posts: 3608
Location: Hendersonville, NC
PostPosted: Wed Nov 08, 2006 9:19 pm Post subject: Reply with quote
Cook is about 30% better scorer. And he shoots the 3.
Collison is about 40% better rebounder. Almost twice the offensive rebounder
For overall value, I think Cook produced about 20% more wins per minute last year.
_________________
`
36% of all statistics are wrong
Back to top
View user's profile Send private message Send e-mail
Gary C
Joined: 14 Apr 2006
Posts: 69
PostPosted: Thu Nov 09, 2006 12:07 pm Post subject: Reply with quote
I've wondered about the wisdom of using "most recent season" to do similars and projections instead of "career to date." While recognizing the major inherent differences between baseball and basketball, I don't see the compelling reason for taking whole career into account with baseball projections but rejecting years 1 to y-2 completely in basketball projections.
The same point holds for lesser periods than career, like "last 3 year" or whatever.
Back to top
View user's profile Send private message
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 979
Location: Seattle
PostPosted: Thu Nov 09, 2006 12:20 pm Post subject: Reply with quote
The compelling reason is it's a lot more work. *shurg*
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 979
Location: Seattle
PostPosted: Thu Nov 09, 2006 12:23 pm Post subject: Reply with quote
Well, actually, that's not quite all of it, which relates to a larger point.
Baseball Prospectus' PECOTA is able to draw, if I recall correctly, all the way back to WWII for comparable players. We can only go back to 77-78, unless you build a system that doesn't use turnovers.
Then you add on top of that PECOTA is able to use minor-league statistics to project the development of young players, something which is far, far off with regards to basketball (whether college stats or Euroleague stats), if it ever happens at all.
The biggest problem with projecting via similarity in the NBA is that in many cases, there just aren't that many similar players. If you're talking similar over a two- or three-year stretch, the pool shrinks even further.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Fri Nov 17, 2006 4:23 pm Post subject: Reply with quote
Okay, here's a link to the Excel file of my preliminary attempt at a similar system for NBA players:
NBA Brock Spreadsheet, Version 1.0
I would encourage people to download it and please make changes, then repost the file. It is in its earliest possible evolutionary state, and I would greatly appreciate any improvements.
(Note: the column marked "P" is to signify whether or not that year is being projected, i.e. 1="yes", 2="no")
Back to top
View user's profile Send private message Visit poster's website
GCole
Joined: 08 Nov 2006
Posts: 13
PostPosted: Fri Nov 17, 2006 8:24 pm Post subject: Reply with quote
Is there supposed to be another spreadsheet with this? All the cells have "div/0".
Back to top
View user's profile Send private message
Neil Paine
Joined: 13 Oct 2005
Posts: 774
Location: Atlanta, GA
PostPosted: Fri Nov 17, 2006 8:54 pm Post subject: Reply with quote
You're supposed to enter your data into the rows, and all those "div/0!"'s will magically turn into projected numbers.
Back to top
View user's profile Send private message Visit poster's website
GCole
Joined: 08 Nov 2006
Posts: 13
PostPosted: Fri Nov 17, 2006 9:28 pm Post subject: Reply with quote
davis21wylie2121 wrote:
You're supposed to enter your data into the rows, and all those "div/0!"'s will magically turn into projected numbers.
Ok, thanks. I was confused because earlier in the thread somewhere it was mentioned about comparing like statistics from other players and I was assuming thats what you were doing.