APBRmetrics

[JNichols42887]

I just "created" a defensive stat called Defensive Composite Score, and you can find the link here: http://www.82games.com/nichols1.htm I know that you guys here are the best of the best in terms of basketball statistics, so I was looking for input. My methodology is explained somewhat in the article, but I can explain it further if anybody is interested. How can I make this stat better? This stat has only had a little bit of tweaking involved, and I figure you guys would have some ideas on how to make it even more realistic.

Thanks,
Jon Nichols

[kjb]

By using box score stats and DeanO's drtg, aren't you sorta double-counting the box score stats? Dean's system uses the same stats to estimate the rating. Which box score stats are you using? I'm sure blocks, steals and fouls. Defensive rebounds? On a per minute basis? Any thought of scaling DCS to 1000 instead of 100 to allow for greater differentiation between players? Do you adjust box score stats for position? In other words, guards are more likely to produce steals, centers more likely to produce blocks. Comparing the two isn't exactly an apples to apples thing.

Overall, I like the effort, and despite the questions, the results "feel" about right. Although, looking at my favorite team (the Wizards), Stevenson rates well below Jamison, which isn't right. Jamison is a truly poor defender. Stevenson isn't a shut-down defender, but he at least executes the team's defensive scheme (such as it is) and gives an effort. I'd describe him as "adequate" defensively. Jamison is "inadequate."
_________________
My blog

[JNichols42887]

Yes, it is a a bit of double counting. I was thinking of weighting either the box score stats lower or the defensive rating lower (probably that one because it seems to favor all players on good teams too much; I don't like Robert Horry being that high).

I'm only using blocks, steals, and (blocks + steals)/foul. Everything is on a per minute basis.

Scaling to 1000 would actually be a pretty good idea.

I didn't adjust for position, although I'm not exactly sure how I would do that. The problem is that all the players are ranked against every player in the league, not just at their position. I feel like since I'm doing that, I shouldn't adjust for position.

Thanks,
Jon

[davis21wylie2121]

No use of 82games' counterpart PER in your system? When I did a system like this last year, I used Counterpart PER (25%), OnCourt DRtg (25%), On/Off DRtg Differential (25%), and Dean's boxscore DRtg (25%), which gave reasonably decent results. By ignoring counterpart production and double-counting boxscore stats, you're overvaluing help defenders (guys who rack up steals and blocks by leaving their man) and undervaluing man defenders.

[JNichols42887]

Interesting. Does 82games have those numbers for each player as one number or do they only have it by position?

[Charles]

I often feel that people developing player ratings worry too much about passing the "smell test." I suppose this is done in an effort to make the numbers commercially acceptable, but too often the ratings seem to simply re-cycle the official data (often re-enforcing misconceptions.)

Therefore, I appreciate that you have included the OnCourt/OffCourt data, rather than relying solely on the league numbers. It takes considerable gumption to rank the reigning DPOY as the fourth best defensive big on his own team. But, I think you are probably on to something.

Dan Rosenbaum's superb study ranked Nene as Denver's top defender - 6th in the league over-all - while Camby didn't even make the top ten at his own position. And it doesn't appear that anything has changed since that study was done. Denver still consistently posts better defensive numbers with Nene on the court than Camby.

[davis21wylie2121]

Under each team's main page (for instance, my unfortunate favorite team), it lists "opponent production", which is just counterpart PER across all positions. And Roland, if you're lurking out there, I want to say thanks for adding that feature a few years ago -- back in the day, you'd have to compile the data position-by-position for each player, which was a real chore.

[JNichols42887]

One question about that. Correct me if I'm wrong (and I probably am), but aren't opponent PER and DRtg calculated in a somewhat similar way? Don't they just calculate how the opponents team's players at a certain position performed? In that case, wouldn't the only difference between counterpart PER and defensive rating be the difference in the way PER and offensive rating determine efficiency?

[davis21wylie2121]

Putting aside the fact that PER and points per possession "ratings" are two completely different metrics measuring two completely different things (in fact, I always say PER technically measures nothing at all -- but that's a different debate), Dean's DRtg is based entirely on boxscore stats (STL, BLK, PF, DRB, etc.) with adjustments made using team defensive stats (forced FG misses, etc. are all estimated from team stats). In other words, it's a rough estimate, to be used mainly in the absence of 82games-style advanced stats. 82games counterpart PER, on the other hand, is determined from NBA.com game logs by tracking who is theoretically guarding whom for every 5-on-5 matchup in every minute of every game. The result is a metric that tracks how each player's man performed, which is usually a very good indicator of man-to-man ability. Sometimes cross-matching fouls up the system, but it's generally a valuable stat to include in any defensive meta-rating.

[JNichols42887]

Well, it sounds like something that could be incorporated. Although, it does look like it punishes guys that play inside (Okafor has a higher counterpart PER than Walter Hermann, Raymond Felton, and Matt Carroll).

[Chronz1]

[benji]

Suggestion:
The counterpart PER includes all their rebounds, blocks, etc. though. What I do in my own defensive statistical work, is just use the counterpart data for the rest of the stuff. Most of the stuff you would need (points, FGA, FTA, eFG, TO, AST, etc.) is already on there.
_________________
a. stat. n00b.

[kjb]

I'm personally not a fan of the counterpart data because it's based on an automated matchup system. It doesn't really say how good a man defender a player actually is. It tells you what the player designated to be his counterpart produced. But it may have nothing to do with that player -- he may have been matched against someone else. Plus, there's no accounting for switches, zones, transition or help D.

For example, back in 04-05, I tracked half a season of the Wizards defense. The counterpart data said that Hughes faced 18.8 FGA per 48 minutes. My tracking put it closer to 12 FGA per 48 minutes. Counterpart data said Haywood faced 11 shots per 48 minutes -- my tracking put him at 21+. These are HUGE differences, and they're very important, in my opinion.

Now, the Wizards are/were probably more extreme than most teams in this sort of thing. I've done limited tracking of other teams, and I think counterpart data may be a little better for some of those teams. But not enough better to be used as a major part of a defensive rating system.


_________________
My blog

[Truthiness]

I enjoy the new numbers but the first thing that jumps out at me is the lack of guards being being rated higher. Its almost as if PF/C are rewarded for being big and being able to hover around the basket. The prime example for this is Tayshaun Prince being the second worst defender on the Pistons. He consistently gets the hardest matchup on the team and is probably one of the better defenders in the league.


I was also going to say that I think you are counting box score data twice but that has been pretty well covered. I think that this has potential to be a pretty good rating system but at the moment there are some early issues that I am sure you can figure out by playing with the numbers and weighting things differesntly.


Until Project Defensive Scoresheet is implimented (from Oliver's book), I think it will be really hard to have an extremely accurate rating for defense.

[JNichols42887]

Big men in general do have better ratings with DCS, but I don't really mind that. I remember reading in Basketball on Paper when Oliver pointed out that so many teams acquire big men that are good defenders and poor on offense because of the importance of having good defense inside. It would be great if it's not skewed at all, but if it had to be, I would want big men to be rewarded. And it is still possible for a big guy to have a bad DCS -- see Eddy Curry.