Page 1 of 1


Posted: Wed Feb 20, 2019 10:04 pm
by Italian Stallion
How reliable is NBA DRPM at measuring individual defense?

I don't want to put too much weight on eye test, but I find too many players I think are clearly good or bad as individual defenders that happen to be on a team that is the opposite of what they are and the team quality seems to get passed to their DRPM to at least some extent.

For example:

Frank Ntilikina is woeful on offense, but he's clearly one of the best defenders on the Knicks, who happen to be one of the worst defensive teams in the NBA. Even when Frank is doing a great job, the TEAM defense often fails. DRPM shows Ntilikina as being a poor defender and much worse than last year. I'm not buying it. He may not be an elite defender, but imo there's no way he's a huge negative.

DRPM says that Kyrie Irving is a plus defender this year. I'm not buying that either. He plays on an excellent defensive team that covers for him well, but imo his own contribution is negative.

DRPM says that Jokic is a solidly plus defender. I'm not buying that either. He just happens to play on an excellent defensive team that covers for him well.

I don't know the formula or underlying math well enough, but I'd be willing to bet anything that if Ntilikina went to Boston and Kyrie came to NY their DRPM's would shift significantly in opposite directions. Somehow the formula is not teasing out individual defense from team defense well enough.


Posted: Wed Feb 20, 2019 10:37 pm
by Mike G
Theory #1 -- Just as an elite defender is not going to be very effective if nobody else on his team is playing defense, a mediocre defender can do just enough to be effective, when the other 4 guys are mostly good on D.

Theory #2 -- In boxscore stats, there's no real measure of defensive effectiveness, so it's often just some form of: D = overall -minus - offense. RPM is something like 2/3 Boxscore plus-minus and 1/3 RAPM?

Denver's top 6 players in DRPM are either C or PF. One wing player, Gary Harris, is barely positive. Den is NBA avg in DRtg.
Celtics top 5 are 4 bigs plus Marcus Smart. Irving is right at team avg.

Ntilikina is about half a point below avg for the Knicks on DRPM, but he's best among the PG they have.
Knicks are defensively 3.0 pts worse than avg with him on the floor and 4.4 below avg when he's off.

Maybe there's no such thing as "individual defense" in the NBA. You're basically part of a system -- on offense, too, but esp. on Defense.


Posted: Thu Feb 21, 2019 1:05 am
by Crow
I dunno the proportions of the Boxscore component to RAPM (not sure if JE has disclosed). But I do remember that the proportion shifts during the season toward more RAPM. I am not sure what share is RAPM at start of season but I assume by end of season it is at least 2/3rds RAPM.

Any clarifications Jerry?

There is individual defense but it is not clear what share it is. Years ago I did a video review of a game and found (for that game) that about 1/3rd of possessions ended with a defender other than the original matchup. If you were a team with the player tracking data you could find the closest defender and sort the data into a counterpart defender at end and switch and then run separate RAPM for own man possessions and switches. Own man at end doesn't mean sole responsibility and switch may not mean a free pass. But having the RAPM split might help understand abilities and issues better than not having it.

Another RAPM I've identified and wonder if ever has been run or ever will.


Posted: Thu Feb 21, 2019 1:42 am
by xkonk
This article doesn't give you the DRPM number specifically, but from a quick read-through I think you can piece together a) superstar type players don't change much when they change teams, presumably on offense or defense; b) defense is a bit more reliable than offense but still not a great correlation from year to year and especially not if you change teams ... nge-teams/


Posted: Thu Feb 21, 2019 5:37 am
by Crow
Find DRAPM factors (rd11490 on twitter, occasionally Jerry Engelmann, or others), compare with boxscore based four factors, individual, team, on / off, matchup. Analyze everything you can find. The answer is not simple or fixed.

If you want the answer for gambling purposes, split data by opponent player and team types and quality, home / road, rest/not, etc. may help or mislead. Do it yourself or join a syndicate with higher analytic talent / effort? Up to you.

To try to know just for interest / yakking may or may not be lighter. Depends how much you want to try and know


Posted: Thu Feb 21, 2019 9:50 pm
by Italian Stallion
I'm not looking for information for gambling purposes (though I would not eliminate that possibility if I thought I had a profitable insight). I'm just trying to understand and measure player value better. There are too many things missing from the boxscore and some of the things included are hard to weight properly because it really sort of "depends" sometimes. But when I start looking at +/-, lineup data, and various adjusted +/- ratings I seem to run into a different set of problems. I may be looking for a solution that does not exist, but it would help if I at least understood some of the flaws in the models I am looking at.


Posted: Fri Feb 22, 2019 7:04 pm
by tarrazu
Might find some of my links here: viewtopic.php?f=2&t=9574#p33949 and link to Shea's site about his defensive ratings applicable to this discussion.