How To Identify Similar Defensive Players

[miabbett]

Hey everyone. I was just wondering if anyone could think of any defensive statistics that capture a player's defensive style of play and/or defensive ability. All I can really think of is steal percentage, block percentage, and defensive rebounding percentage, but I was wondering if there might be more stats out there that can paint a fuller picture of what a player does on the defensive end of the floor. Any ideas are greatly appreciated!

[bchaikin]

Synergy, StatsCube, 82games.com, all the video you can find...

[JimiHendrix]

Until Vantage Sports/SportVU start killing it with some publicly available stats: your eyes.

[knarsu3]

I am working on something like this using the Vantage stats (and cluster analysis).

[Crow]

Other possible stats to look at include : steal percentage per personal foul (reward/cost), block percentage / per personal foul, personal foul rate, counterpart offensive rebounding percentage, counterpart inside shot percentage and FG%, counterpart 3pt shot percentage and efG%, same for opponent team overall when on the court (and compared to off), counterpart usage rate, counterpart offensive rating (by PER, defensive rating or other metric), counterpart assist rate and then defensive real plus minus and the factor level splits if available.

[mtamada]

Also the player's height and typical position played. If two players have the same STL% and BLK%, but one of them is 6 feet tall and the other one is 7 feet tall, then at least one of them is a very unusual player. Or both of them are bad players.

A number of analysts suspect that tall players, centers in particular, have more of an effect on the defense than the other players on the court do. Defensive adjusted plus-minus (regularized or not) might measure this, but might not.

[miabbett]

Thanks for the good ideas, everyone. The reason I posed the question is because I'm trying to fine-tune my player projection model, which is based around finding similar players and extrapolating performance based on passed results. As it stands, my model first finds players who are similar in terms of height, weight, age, years of service, steal percentage, block percentage, and defensive rebound percentage. That being said, this clearly doesn't capture a perfect representation of what a player does on the defensive end of the floor. Can anyone think of any relevant metrics that would be easy to incorporate into the model?

[Mike G]

You might include PF rates. It's better to block shots, steal, and rebound without fouling much.
Last year, Tim Duncan set a modern record for Blk/PF.
http://bkref.com/tiny/EFOdN

[mtamada]

Taking PFs into account is a great idea. Ideally the data will distinguish between offensive and defensive fouls, obviously the standard statistics fail to do this but stats based on PBP data will probably let you do this.

That Tim Duncan stat is interesting. Potentially meaningful but since most people haven't been analyzing it, hard to know how meaningful.

[Mike G]

Duncan has actually never fouled anyone. The refs finally started believing him.

[miabbett]

Thanks for the ideas regarding foul data, everyone. Any other input is greatly appreciated!