How exactly do they calculate RAPM on the defensive end? Does it account for bad rotations on his teammates parts, etc. Because I just don't think Harden has been that bad of a defender this season
RAPM doesn't attempt to assign credit/blame for individual possession outcomes. It looks globally at the offensive and defensive efficiency with all combinations of players on the floor and uses some fancy math to estimate how good each player is on the offensive and defensive end. The estimates don't always match our subjective assessment of the players, but its another (objective) perspective to consider.
I see. So if you put Harden on a team like the Pacers, his defensive RAPM would be very high? And likewise for Offensive RAPM on a great offensive team like the Spurs or something?
Not necessarily. Its not simply how well the team performs with that player on the floor. It accounts for the other guys on the floor as well. Its possible that a good team defense can mask Harden's defensive weaknesses, or maybe it doesn't at all and (for example) he turns a great defensive team into merely a very good defensive team when he is on the floor. The basic concept here with adjusted +/- is it estimates ratings for all players in such a way that the predicted outcomes for every stint of every NBA game (a period of time with no substitutions) is as close as possible to the actual outcomes. The idea is that the best estimates for every player will result in a minimized standard error. There are different types of regression techniques one can do for this, RAPM being a little more sophisticated than the basic OLS regression and resulting usually in more accurate estimates. Accuracy in this context refers to smaller standard errors.
Thanks, wise I could rep again. So is RAPM basically considered the best stat right now for evaluating a players impact right now?
Some would say yes, but there's probably no single "best stat" out there. You generally can divide player rating stats into three categories: "top-down" stats which rate players according to team performance (RAPM is a prime example of this), "bottom-up" stats which rate players based on individual actions they perform (PER, ASPM, WS/48, WP48, lots of examples ...), and hybrid stats which make use of some combination of these (e.g. xRAPM, Simple Rating from 82games.com). I think its a good idea to consider both top-down and bottom-up evaluations of players and try to evaluate the player from many different angles.
Casspi is a good example of using the multiple angles. XRAPM showed him to be okay defensively , bad offensively. In a system where he isn't asked as much to be a creator, his RAPM shows he's okay in both places for the right team. He is shooting slightly less from midrange, but does look like he should be used more in the corners. I think the player tracking and synergy stats will add another useful element in player evaluation when these stats get incorporated with RAPM.
Sometimes it's better to use your own eyes. Even a cross-eyed test will tell you that number on Howard is crap. Howard is the best player on this team and easily a top ten player in the league and any number that says otherwise is crap. Use your eyes.
But not everyone's eyes are in agreement, particularly when it comes to Dwight, so the "eye test" isn't necessarily any more accurate.
RAPM is not terribly accurate. It ranks right about where PER ranks. xRAPM (a completely opaque stat that is not replicable by the public) is actually very accurate. I cannot get excited about this posts information.
The formula for xRAPM is public, though you have to read through some threads on the APBRmetric board to piece it all together. It is a composite of a statistical plus-minus formula (which is box-score based) plus RAPM. The incorporation of box-score stats helps to reduce the noise levels and makes it more accurate particularly for offensively-tilted players.
There have been some talk this year about Anthony Davis' true impact on the team. For example, people say he's a great defender. Yet the Pelicans give slightly LESS points when he's off the court. And the Pelicans are only 22-27. For all the stuff about Harden's ball hogging and lack of defense, the Rockets were 45-37 last year. Anthony Davis may be awesome, but he doesn't seem to be changing the teams' fortunes the way everyone make him out to be this year. In a nutshell, that's what RAPM is, how a player's presence change the team's ability to win. It is true that this allows some role players to be "overrated" due to some incredible fits. Or make some stars look lesser because of bad situations. But in the end, isn't that pretty important? If Anthony Davis is not helping his team win, then the Pelicans should find a coach that can get the most out of Davis.
I personally cannot find the formula. Would you kindly point me towards the source? I have never seen anyone on APBR claiming that they could replicate xRAPM or ASPM.
I found this post from a few years back which provides some links that can help answer your question on xRAPM: As far ASPM, details can be found here: http://godismyjudgeok.com/DStats/aspm-and-vorp/ The authors of xRAPM and ASPM stats both post on APBRMetrics, so you could always ask detailed questions about it over there.
They are significantly different stats. Win Shares is not adjusted for lineups. For example, last year Harden had a 3.0 DWS. Asik had a 3.5 DWS. With xRAPM(which is slightly different from the current RAPM but same philosophy), had Harden at -0.9 and Asik at 5.4. Which stat you use is of course your preference. I cannot say which is definitively better. But they are definitely NOT the same thing.
I remember Morey once saying that Rockets have really smart fans, I think this thread is a fine example of what he's talking about. I had no idea we had so many RAPM savvy guys guys here, very informative stuff here guys, I love it.