Author Message
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 979
Location: Seattle
PostPosted: Sat Jan 08, 2005 2:23 am Post subject: Team similarity Reply with quote
I finally got to one of the many projects on my to-do list: A team similarity system based on Dean Oliver's Four Factors going back to 1979-80. I equally weighted each of the four factors on offense and defense, as well as pace, squaring the differences.
To take it for a test run, I decided to look at two of the "surprise" teams of 2004-05, Phoenix and Seattle.
Code:
Team Year Sim
---------------------
Atlanta 1990 97.2
Portland 1984 96.8
Seattle 1989 96.3
Seattle 1992 96.3
Milwaukee 2000 96.2
Atlanta 1986 95.8
Houston 1980 95.2
Atlanta 1988 95.1
Miami 1994 95.1
Seattle 1993 94.9
---------------------
AVerage 46.3
Gol State 1993 96.1*
Well, as a Sonics employee, it's disappointing to see that the comparable teams only averaged 46.3 wins. The projection for +.500 teams is likely to underestimate, but that's still disappointing.
It's interesting to see so many Sonics teams on the list. Milwaukee is one of the teams I've mentioned as a comparable team in terms of perimeter orientation, and Golden State circa TMC (though by 1993 Mitch Richmond was gone) was a team Dean mentioned when we talked about similar teams. They shoot up when you factor out pace (I have one score with, and one without.)
Now, the Suns:
Code:
Team Year Sim
---------------------
LA Lakers 1981 92.1
Sacramento 2004 91.9
LA Lakers 1996 90.6
S Antonio 1984 90.1
Charlotte 1993 89.3
LA Lakers 1982 89.2
Sacramento 1999 89.2
Orlando 1996 88.9
Orlando 1995 88.7
Dallas 2000 88.6
---------------------
Average 50.1
Phoenix's comparable teams, alas, faired somewhat better. You'll notice a pair of Showtime Lakers teams, as well as a pair of Shaq-Penny Orlando squads and two recent Sacramento teams. Still, the 2004-05 Phoenix combination appears to be pretty rare; I haven't looked at a lot of team similarity scores, but 92.1 seems to be a pretty low top score. Things don't get much better when you factor out pace.
Dan Rosenbaum has talked about the oddity of having such a good team that's so weak on the glass, and apparently that does show up numerically.
Back to top
View user's profile Send private message Send e-mail Visit poster's website
Ed Küpfer
Joined: 30 Dec 2004
Posts: 786
Location: Toronto
PostPosted: Sat Jan 08, 2005 3:43 am Post subject: Re: Team similarity Reply with quote
admin wrote:
I finally got to one of the many projects on my to-do list: A team similarity system based on Dean Oliver's Four Factors going back to 1979-80. I equally weighted each of the four factors on offense and defense, as well as pace, squaring the differences.
I have found that similarity depends crucially on the weights. This is why I am not a big fan of similarity scores, as the weights chosen are essentially arbitrary.
Not necessarilry so with the four factors. Recently, I regressed WIN% against the four (times 2) factors (standardized), and came up with the following weights:
Code:
O_EFG% 1.0
O_FOUL% 0.3
O_OR% 0.4
O_TO% -0.6
D_EFG% -0.9
D_FOUL% -0.4
D_OR% -0.3
D_TO% 0.6
where FOUL% = (0.44 * FTA) / POSS, and TO% = TO/POSS.
The team similarity idea is a good one. I'll see if I can look into it on the weekend.
_________________
ed
Back to top
View user's profile Send private message Send e-mail
kjb
Joined: 03 Jan 2005
Posts: 865
Location: Washington, DC
PostPosted: Sat Jan 08, 2005 7:08 am Post subject: Reply with quote
What's the similarities list for this year's Wizards using the method you applied to the Sonics and Suns?
Back to top
View user's profile Send private message AIM Address Yahoo Messenger
Kevin Pelton
Site Admin
Joined: 30 Dec 2004
Posts: 979
Location: Seattle
PostPosted: Sat Jan 08, 2005 2:12 pm Post subject: Re: Team similarity Reply with quote
Ed Kupfer wrote:
Not necessarilry so with the four factors. Recently, I regressed WIN% against the four (times 2) factors (standardized), and came up with the following weights
The thing is ... those weights are great for finding teams of similar overall ability, but I'm not that concerned about doing that. If teams with similar skills had wildly different overall performance, that's interesting to me. I want to see if certain types of teams fare better the following season, or aren't as successful as they should be in the playoffs (a claim many have made about both Phoenix or Seattle), etc.
Code:
Team Year Sim
---------------------
Denver 2004 97.7
New Jersey 1999 97.1
Gol State 1986 96.9
Portland 1989 96.9
Charlotte 1992 96.7
Indiana 1980 96.6
Denver 1985 96.3
Gol State 1996 96.3
Cleveland 1980 96.1
Washington 1995 95.9
---------------------
Average 35.2
Atlanta 1996 98.9*
NewOrleans 2004 98.0*
Indiana 1981 97.8*
The similar teams to the Wizards aren't very good, and that does illustrate some of the weakness of using my method; the reason for that is their eFG% ratings are relatively low compared to their other ratings. Intriguingly, taking out pace changes the scores dramatically (and improves the quality of the comparable teams).
Denver 2004 is an interesting comp, in that it was another team that improved after being terrible the year before. This is something experts know to be a "hidden indicator", but I have no idea what it means.
Code:
Team Year Sim
---------------------
SanAntonio 2004 98.5
Detroit 1990 95.6
New York 1992 95.6
Miami 1997 94.7
New York 1994 94.7
Portland 1999 94.6
SanAntonio 1992 94.6
Phoenix 1983 94.6
Portland 1991 94.1
Indiana 2004 93.7
---------------------
Average 56.6
I decided to throw in San Antonio's more conventional style to see if they showed up winning more games, and the answer is a resounding yes. If one name comes to mind looking at that list, that name is "Pat Riley".
Back to top
View user's profile Send private message Send e-mail Visit poster's website
KnickerBlogger
Joined: 30 Dec 2004
Posts: 180
PostPosted: Sat Jan 08, 2005 6:56 pm Post subject: Re: Team similarity Reply with quote
[quote="Ed Kupfer"]
admin wrote:
where FOUL% = (0.44 * FTA) / POSS
Just curious, why use FTA *.44 instead of FTM?
Back to top
View user's profile Send private message Visit poster's website AIM Address Yahoo Messenger
Ed Küpfer
Joined: 30 Dec 2004
Posts: 786
Location: Toronto
PostPosted: Sat Jan 08, 2005 7:08 pm Post subject: Re: Team similarity Reply with quote
KnickerBlogger wrote:
Ed Kupfer wrote:
where FOUL% = (0.44 * FTA) / POSS
Just curious, why use FTA *.44 instead of FTM?
I see getting to the line as the skill of interest.
From a defensive standpoint, you have no control over whether your opponent makes a FT shot (although you may have some control over who you foul). It doesn't make much sense to me to look at defensive FTMs. I use FTAs on offense to be directly comparable to defense.
I'm pretty sure using FTM or FTA will give similar results, especially for unstandardised numbers. There simply isn't much veriation between teams to make any difference.
_________________
ed