APBRmetrics

[Neil Paine]

OK, another college hoops-related post, but one certainly applicable to the NBA: Team Similarity Scores.

First, I dumped all of Ken Pomeroy's NCAA stats (since 2004) into a spreadsheet. Then I adjusted the 4 factors up or down based on the ratio of the team's adjusted ORtg/DRtg to its raw efficiency, and made a couple other adjustments to ensure that the national averages for offensive and defensive categories were equal.

Anyway, from there I standardized each of the 8 factors (4 offensive, 4 defensive) and pace using z-scores. Using this dataset, you can finally calculate team similarity scores (it's the same method that Ken used for individuals -- sum the differences for each team). For instance, here are the 10 most similar teams to the National Champion Kansas Jayhawks:



And the National Runner-Up Memphis Tigers:



And, finally, my favorite NCAA team ever, the 2004 National Runner-Up Georgia Tech Yellow Jackets:



Does this have a practical application? Maybe -- I'm not really sure. But I thought I'd share it, because I thought it was cool. And once the '08 NBA season is in the books, I'll do this for some of this year's pro teams as well.

[Mountain]

DW I look forward to seeing the pro version of team similarity.

it will be interesting then to compare position/player similarity and team similarity. Knowing which teams are similar at both levels will be interesting as will seeing teams with significantly different position/player construction but similar team stats. Could Eli's Count the Basket references to multi-level data analysis be brought to bear on this?

Tracking the playoff stats / achievement level and similarity of them for the groups and subgroups would be a next step. I drift forward to thinking about the possibility of a "team typology" to summarize the relationships. I'd explore the concept and see if it yielded insights / new questions.

i'd explored a little previously with regard to 4 factor similarity of leading playoff teams within a season and across seasons but going from eyeball impression of 4 factor closeness to an overall similarity summary number will be an advance.

[DavidH]

I recently undertook a similar project, creating team similarity scores from Four Factors, Adjusted Efficiency, as well as some other statistics that were easily available on kenpom.com (various shooting percentages, block/steal rate, 3FGA/FGA, A/FGM, defensive fingerprint). My series of posts, including a description of the method, begins here.

Instead of just weighting all categories equally, I assigned weights to categories, plus added extra weight to categories where the team in questions is far above or below average. I'm not wedded to this weighting system, but it's what I'm using at the moment.

I'm interested in the idea of adjusting the Four Factors data based on the efficiency adjustments. Can you go into more detail on that, or provide a link? It seems like simply multiplying them by the ratio is not the way to go, but I haven't had time to put a lot of thought into it yet.
_________________
The Audacity Of Hoops - college basketball statistical doodlings