PER or RAPM?

[dice]

which is the better tool for ranking players?

PER:

http://insider.espn.go.com/nba/hollinger/statistics

RAPM:

https://sites.google.com/site/rapmstats/2013-rapm-non-prior-informed-updated-march-30

[islamovic]

neither

espn rating or efficency on nba.com is much better

[dice]

neither

espn rating or efficency on nba.com is much better

do either account for defense?

[islamovic]

well outside of blocks n steals there arent many d stats.

there are d rating n +/- but those are not individual stats.

[tsherkin]

NBA efficiency is a more simplistic and less effective stat similar to PER, a basic Tendex type stat. PER is a bad tool for specific comparisons and only good for a loose filter, and that only when you add context. Multi-year RAPM is a little better, but still not sufficient on its own as a rankig stat.

[lorak]

which is the better tool for ranking players?

PER:

http://insider.espn.go.com/nba/hollinger/statistics

RAPM:

https://sites.google.com/site/rapmstats/2013-rapm-non-prior-informed-updated-march-30

RAPM is by far the best publicly available single metric to rank players.

[Reservoirdawgs]

PER is a bad tool for specific comparisons and only good for a loose filter, and that only when you add context. Multi-year RAPM is a little better, but still not sufficient on its own as a rankig stat.

Can you explain the quote above? Why is PER a bad tool for specific comparisons, and why is RAPM not a sufficient ranking stat? If these two are insuficient, what are the best ranking stats?

[tsherkin]

PER is a terrible stat because it doesn't really include defense and it's weighting is arbitrary, leaning heavily towards volume scoring. It's also using per-minute stats, so you need to filter it on the basis of actual minutes played before you even THINK about using it, and then it still doesn't do a great job of differentiating players unless they're wildly different and one player is already clearly superior.

It doesn't offer any meaningful analysis.

There ARE no really good individual ranking stats, that was my point. You cannot reduce a player to a single number and expect that to be a sane way to approach things. It is entirely too minimalist and simplistic.

[Reservoirdawgs]

EDIT: Double post.

[Reservoirdawgs]

PER is a terrible stat because it doesn't really include defense and it's weighting is arbitrary, leaning heavily towards volume scoring. It's also using per-minute stats, so you need to filter it on the basis of actual minutes played before you even THINK about using it, and then it still doesn't do a great job of differentiating players unless they're wildly different and one player is already clearly superior.

It doesn't offer any meaningful analysis.

There ARE no really good individual ranking stats, that was my point. You cannot reduce a player to a single number and expect that to be a sane way to approach things. It is entirely too minimalist and simplistic.

I don't think it's that bad when comparing superstars - aka, the people who are going to be dominating the ball and playing fairly comparable minutes. I use PER along with TS%, eFG%, and ORTG and DRTG as my main measures whenever I'm comparing people. I have never found a defensive metric (other than DRTG, which has some flaws of its own) to my liking, but if you take PER and combine them with the ones I mentioned, I think it can be a meaningful tool.

[tsherkin]

I don't think it's that bad when comparing superstars - aka, the people who are going to be dominating the ball and playing fairly comparable minutes.

Again, right off of the bat, because it senselessly deflates assists and doesn't account for defense, it's a bad comparative tool. It's a good summary of box score production, nothing else.

DRTG is a largely useless tool for specific comparison because it directly includes team DRTG, something over which a single player can exert only so much info. It's useful broadly in that lower is way better, but not so much for a direct comparison of players. PER is like that: you can determine that a player is productive, but it's too limited to be of great value unless there's already a really huge gap between players. It's not so bad for looking at younger guys and determining who is being productive in limited minutes, sure, but that's also limited in its utility.

Again, PER is functionally useless. I can look at your raw averages and tell you all of the same information PER tells you... and some that it doesn't. It's not a valuable tool.

[Chicago76]

PER is a terrible stat because it doesn't really include defense and it's weighting is arbitrary, leaning heavily towards volume scoring. It's also using per-minute stats, so you need to filter it on the basis of actual minutes played before you even THINK about using it, and then it still doesn't do a great job of differentiating players unless they're wildly different and one player is already clearly superior.

It doesn't offer any meaningful analysis.

There ARE no really good individual ranking stats, that was my point. You cannot reduce a player to a single number and expect that to be a sane way to approach things. It is entirely too minimalist and simplistic.

I'm not as harsh on PER as you are, but I agree. As someone spends more and more time with advanced stats, this is a conclusion that I think most people reach. Where I do think it is somewhat useful is that it at least utilizes weighting that are intuitive with respect to the concept of production per possession. For someone who is fairly new in the use of advanced stats, it does provide a framework to understand the rough impact TOVs, TS%, rebounding rates, assist rates, etc have on efficiency and pts scored. When ued as a learning tool to promote someone's ability to mentally aggregate and place importance on more discrete forms of production, it is useful. But again, once someone's ability to understand the subcomponents of production improves, the limitations become fairly apparent.

Kind of a side issue, but somewhat relevant to the topic. The underlying advanced stats that "feed" PER are limited in their own ways too. TOV rates are a decent example. It's not really enough to say that player X has a TOV rate of 15, while player Y has a TOV rate of 11, and to conclude that player X is more TOV prone. There is a lot of variance in TOV rates across the league due to role. I've found it useful to regress other things to arrive at an expected TOV rate to compare to actual TOV rate to assess players within their roles. In order of significance:

-AST: players who are responsible for distribution have higher TOV rates.
-ORB: players who play closer to the basket on O tend to get more ORBs. They also tend to play in tighter space, so they end up turning the ball over more frequently.
-FTA: players who get to the line more often tend to have more TOVs
-two least important factors: 2FGA and 3FGA.

I find it more useful to say something like, "For a guy who gets to the line and distributes as much as he does, his TOV rate is actually pretty low" rather than, "He turns the ball 10% more than league avg."

[mysticbb]

PER is a terrible stat because it doesn't really include defense and it's weighting is arbitrary, leaning heavily towards volume scoring.

The weights are not arbitrary. Hollinger explained why he picked those weights in his books, and it makes a lot of sense. You may want to look into it further, because once you try to understand what Hollinger is doing, you may actually agree that this "leaning towards volume scoring" isn't such a bad idea. For that, you may want to check how good players in average are in late shotclock situation when they have to rush a shot, compare that to the expected breakeven points in Hollinger's PER and you will be surprised how close that is. ;)

[tsherkin]

I read what he wrote, mystic, and I still disagree with his reasoning because it's not contextually accurate. Meantime, I also don't agree with him overvaluing volume scoring, since it's one of the least understood and least useful things in the league for the most part.

[RealRapsFan]

PER is both the least predictive and least explanitory of the commonly available stats.

Unfortunately 'picking weightings' is not good science no matter how rational it may seem when you do it.

[LobCityClips]

Is RAPM saying Blake was the 4th best impact player in the league?

[bondom34]

Is RAPM saying Blake was the 4th best impact player in the league?

Huh? Don't know where that question came from, but no, it has him 18th:
http://stats-for-the-nba.appspot.com/ratings/2013.html

[Dr Positivity]

I don't think it's that bad when comparing superstars - aka, the people who are going to be dominating the ball and playing fairly comparable minutes.

Again, right off of the bat, because it senselessly deflates assists and doesn't account for defense, it's a bad comparative tool. It's a good summary of box score production, nothing else.

DRTG is a largely useless tool for specific comparison because it directly includes team DRTG, something over which a single player can exert only so much info. It's useful broadly in that lower is way better, but not so much for a direct comparison of players. PER is like that: you can determine that a player is productive, but it's too limited to be of great value unless there's already a really huge gap between players. It's not so bad for looking at younger guys and determining who is being productive in limited minutes, sure, but that's also limited in its utility.

Again, PER is functionally useless. I can look at your raw averages and tell you all of the same information PER tells you... and some that it doesn't. It's not a valuable tool.

Is this not true of an individual's defensive impact anyways? I am not a fan of Oliver's DRTG because it includes the boxscore weighting stuff, I actually think the best version of a WS-y stat, would be to compare Oliver's ORTG with the player's on-court DRTG.