Tuesday, January 13, 2009

The WNBA and Similarity Scores



At the imitation site to Basketball-Reference.com (note the missing dash in the URL which takes you to a completely different website that B-R.com), the idea of Similarity Scores for NBA players is developed.

Similarity Scores was designed to be a fair way to compare players by their career statistical lines. The idea is "start with 1000" points and then begin to take away points from points of difference. For example, if two players had career steals totals that were 20 steals apart, we would take one point away from the Similarity Score of the two players, reducing the total score to 999. We keep doing this for other statistics and come up with a numerical value for "how similar" one player is to another.

There are three faults to this method that should be made clear:

1. It doesn't take into account the "missing years" of certain ABL players for whom I don't have ABL stats.
2. It compares currently active players to players who have either retired or might have missed years due to absence, and
3. It does not account for player position. Guards are compared to both forwards and centers, instead of other guards. I might fix this in a future version.

Given that, I created a spreadsheet to calculate these scores, and then ran the players of the Atlanta Dream through it. These are the values I obtained:

The player most similar to Betty Lennox is Deanna Nolan (976.72).
The player most similar to Erica de Souza is Jessica Davenport (991.49).
The player most similar to Katie Mattera is Kara Braxton (984.05).
The player most similar to Ivory Latta is Sidney Spencer (993.87).
The player most similar to Alison Bales is Teana Miller (991.78).
The player most similar to Iziane Castro Marques is Edna Campbell (981.57).
The player most similar to Kristin Haynie is Annie Burgess (990.42).
The player most similar to Kasha Terry is Kelly Santos (996.49).
The player most similar to Ann Strother is Navonda Moore (996.54).
The player most similar to Tamera Young is Crystal Kelly (993.59).
The player most similar to Jennifer Lacy is Michelle Marciniak (990.61).
The player most similar to Chioma Nnamaka is Wendi Willits (999.15).


The player most similar to Chamique Holdsclaw is Natalie Williams (932.25).
The player most similar to Nikki Teasley is Kedra Holland-Corn (977.81).
The player most similar to Sancho Lyttle is Sharon Manning (986.68).
The player most similar to Chantelle Anderson is Clarisse Machanguana (994.31).

(* * *)

An interesting caveat on Similarity Scores: the true definition of a Hall-of-Famer is that there is no one who is truly similar to her. Note that the really good players (Lennox, Holdsclaw, Teasley) have a smaller degree of similarity to their next closest counterpart, whereas for Chioma Nnamaka we can find someone to within 0.85 similarity. For example, the player most similar to Lisa Leslie is Tina Thompson (864.05) - Thompson was the closest person the system could find.

Thought exercise: Suppose we took the 16 current (and possibly future) Atlanta Dream players up above and matched them against a team made up of their most similar players listed above.

Who would win the match? And why?

1 comment:

Anonymous said...

Interesting list. Lacy/Spinderella is a clear miss, those two are nothing alike. I don't see Latta/Spencer as all that similar, but much of that is Sid being 8 inches taller. Young/Kelly? Really? Kedra Holland-Korn is similar to Teasley in that they're both likely to quit on the team if things are going badly. Not actually quit, Claw-style, but stop giving 100% and generally make everyone miserable. As players they're not that much alike.

OTOH, Claw/Nat is a near perfect match. Lennox/Nolan is about as good a match for Bad Azz Betty as there is.