vini_vidi_vici wrote:So theres alot going on that many have touched on.
iORTG/iDRTG are also box score metrics, so if it isnt within a boxscore, its not likely to show.
In a later chapter of Basketball on Paper, Oliver emphasized that Offensive Ratings shouldn't be viewed in a vacuum. Introducing a concept he called "Skill Curves", he acknowledged that a player's ORtg needed to be judged in conjunction with his Usage Rate, a measure of how big a role the player fills in his team's offense. The bigger the role, the more difficult it is to maintain a high ORtg; the smaller the role, the easier it is to be highly efficient. Because of this, Oliver stressed that a player's ORtg should primarily be compared to those of other players in a similar role.
Out of necessity (owing to a lack of defensive data in the basic boxscore), individual Defensive Ratings are heavily influenced by the team's defensive efficiency. They assume that all teammates are equally good (per minute) at forcing non-steal turnovers and non-block misses, as well as assuming that all teammates face the same number of total possessions per minute.
Perhaps as a byproduct, big men tend to have the best Defensive Ratings (although Oliver notes that history's best defensive teams were generally anchored by dominant defensive big men, suggesting that those types of players are the most important to a team's defensive success). A corollary to this is that excellent perimeter defenders who don't steal the ball a lot — for instance, Joe Dumars or Doug Christie — are underrated defensively by DRtg, and are prone to look only as good as their team's overall defense performs.
So as Dean notes, youre probably better to use iORTG in terms of comparing guys with similar USG rates and roles. Which is alot harder to do for obvious reasons.
That'll make sense. In reality, the original question was about the obvious difference between that metric and others used around here. Like I said, the confusion came from the poor disclosure that the nbs stats were team related stats. Now, of course, that quote clarifies any remaining questions. Thanks.
vini_vidi_vici wrote:You probably wont like hearing this, and I assume youre a novice in analytics,
If it helps I have developed models to detect money laundering that are analyzing millions of transactions as we speak, but yeah, I have never done sports analytics
vini_vidi_vici wrote:but the best way to compare players is using a myriad of statistics to get the best possible results. Its difficult to say because _____ has a similar iORTG/PER/etc.. (all encompassing statistics) to _____, they are comparable. The upside is we have so many statistics and different variations of boxscore/non boxscore metrics we can refer to. FWIW, im not a big fan of iRTGs, but thats because I think you can get better results breaking down the minutiae with different metrics and extrapolating from those instead of a one stop statistic.
You are not going to like this, but this approach caters more to people fishing for a combination of stats that back there own hypothesis. This is unfortunate but I understand it is the nature of the beast. The downside, obviously, that "proving" a formula is biased can be, depending on the formula, the basis for a phd.
I have, in fact, thought of writing an app that will take the end result you are looking for and it would search for a combination of stats proving that point. It would be useless, but priceless for arguments in here
Thanks for your time, it was a great post.
If you check the link that I posted, you'd see that the individual ratings used in basketball-reference are also entirely dependent on five-men units. They're just (poorly) modified versions of them.