THE BOOK--Playing The Percentages In Baseball

The other big sites for adjusted +/- are basketballvalue.com, 82games.com, and basketballprospectus.com.

Didn’t through the entire thing due to some time constraints, but I do like it. The plus/minus work done by the good basketball analysis is extremely interesting.

I like 82games.com a lot.  They combine the linear weights aspect of analysis… combining points, shooting %, rebounds, steals, blocks, assists, etc… and combine with a +/- on/off court statistic, which I think helps take into account defense, matchups, etc.

My use of the stats is to look at both components and the total.  If the linear weights component is higher than the +/-, that tells me the player is better offensively than defensively.  If the +/- is higher than the linear weights number, that tells me the player is good defensively (or “intangibly”, since we don’t fully grasp everything that contributes to someone’s net effect on the court).

This is just my interpretation though.  I’ve never played basketball and don’t watch too many games a year.  But for the most part, the 3-year values on that site seem like they’re spot-on.  The few names that seem out of place (Thaddeus Young?!) may just be very underrated defensively.

This methodology has, btw, been rejected in basketball.  The big problem with it is that the sport is very matchup-dependent.  If you had one player with a good plus/minus, it’s not as though you could swap him in for a player with bad plus/minus.  You’d need to know who they were on the floor with, who they were guarding.

It also ended up being pretty meaningless for guys who played a lot of minutes, because they’re only in the game when it’s close.  When they come out, leverage is low, and the end of the bench is in the game, and weird things happen to scoring.  And then it’s not very helpful for the guys at the end of the bench because they only play in low-leverage situations.

I think WOWY is great when you have virtually no substitutions (fielding on a per game basis) or when you have so many substitutions that you don’t always get your matchup (hockey.) Basketball is in the middle…

Clearly, you need to include WE and LI as parameters…

Btw, I would reject the notion that this has been rejected.

This is no different than a strength of schedule adjustment.  There are matchups in play, and that simply means the regression equation is going to be more complicated than a simple x1 + x2 +… + x10 = +3 or whathaveyou. 

It definitely has the basis.

Let me rephrase that: Roland has rejected the “Roland rating”: (aka adjusted +/-)

http://www.slamonline.com/online/nba/2009/03/changing-the-stat-quo/

@Hawerchuk:

The Roland rating is not adjusted--it is pure plus/minus, which is rather worthless.  Adjusted plus/minus accounts for who was on the court for both teams at the time--and that makes the evaluation actually mean something for the individual player.  I agree, the Roland rating is worthy of rejection as a player metric.

Whichever form of it you choose - raw +/-, On-Off +/- (the “Roland” Rating) or some form of leverage- or competition-weighted +/-, Roland has rejected them all.  He counts events, watches D-matchups.  It’s quantitative scouting.

The basic problem with basketball plus/minus is that no distinction is made between

1. items that the player has a lot of control over, such as his own shooting % or rebounding,

2. items which he has some control over, such as his teammaters fg% and

3. items which he has little or no control over, such as his teammates or opponents ft%

Additionally, these match-ups look at how 5 man units perform. It’s amazing how little time any 5 man unit performs together, so there is a large sample error as a result

Thirdly, if you have 2 players, a starter and a back-up, playing virtually all of the minutes at a spot, you are impacted by how good/bad your backup is.

as a result of this and other problems, the stats have a lot of noise, and little correlation year-to-year. I think they need to determine how much to regress these different areas.

I agree that you need to look at the adjusted plus/minus in conjunction with player-events.  If, for example, a player happens to have a high plus/minus, but there is no indication in any of his individual stats that he could have been directly responsible, then you have to “regress” that plus/minus heavily.

As an example, if you have Wayne Gretzky and Dave Semenko each with a +50, then unless we can show that Semenko had a defensive presence of some sort that could have led to that +50, then we have to presume that alot of that +50 is noise.

So, we can say that the +50 for each player is actually:
+50 +/- 20

And when we looked at player-events, that it becomes:
+60 +/- 10 Gretzky
+10 +/- 20 Semenko

Something like that.

Ok, I read the interview.  I would guess that Roland sees it similar to how I’m describing it here.