2017 RAPM/RPM/etc. Thread 2017 S-Y NPI RAPM Added (6/26)

[og15]

What is definition of "better" in basketball?

You tell me

I've asked the question first, so it's rude from you to answer that way.

Better is determined by looking at as much as we can and making conclusions based on that. So critical watching, basic stats, efficiency, RPM, RAPM, whatever, gather it all together, critically analyze and come to a conclusion. I actually like all the different statistics, and when I have time I like to explore how they are derived if possible in order to be able to look at them critically.

In addition to basic production and advanced stats and all that, it's also important for us to examine the players role and team situation. The stats tell us about impact in a specific role on a specific team. Many times we can logically infer and see how a players type of production and role and therefore impact can be transferred to other team situations. Some players, generally the superstars will end up in a similar role and have similar impact on almost any team they go to.

Other times it's not that clear cut. There are players who in a specific role on a certain team will have very high impact which will be reflected by things like RPM, etc, and they would be suggested to have higher impact than another player who is also on a good team but maybe taking on a larger role. The problem here is that we can see situations where if they switched, the guy who certain numbers might imply is better would not be able to lead the same team to as many wins while the other guy would keep a similar level on his new team.

There's no stat I look at and just consider it something to rank players with. They can be a guide, and they can be a check against biases and overlooking things and therefore get you to re-examine your thoughts. On the other hand, to just suggest higher RPM or RAPM, or whatever just simply means better player, the end, I don't even believe the makers of these stats would stand behind that notion.

[Woodsanity]

DD is a good scorer but a weak, poor defender and not a good passer I am not surprised he doesn't rank that high.

[franktony]

I wonder how the heck Kyle Anderson appears as the best "SG" in the NBA in DRPM. I knew his defense was good in some games because his length really bothers some players, but I could never tell it was that good.

When SA drafted him, I actually expected the opposite from him. Good offensive impact and some struggles defensively because of his "slow-mo" style. Quite surprised.

[Colbinii]

When SA drafted him, I actually expected the opposite from him. Good offensive impact and some struggles defensively because of his "slow-mo" style. Quite surprised.

I expected the same thing. He has surprised me.

[Dr Positivity]

I think RPM has the Raptors starting lineup pinned pretty well. Lowry is easily the most valuable player and then you have several good but imperfect ones. Derozan and Valanciunas put up good traditional stats but don't fit into the Draymond Green-ian "spacing, D and passing" league, Patterson and Carroll have the opposite problem as they are are analytically friendly as quality defenders who space the floor but are frustrating when it comes to the still important part of putting up real boxscore stats

[homecourtloss]

I think RPM has the Raptors starting lineup pinned pretty well. Lowry is easily the most valuable player and then you have several good but imperfect ones. Derozan and Valanciunas put up good traditional stats but don't fit into the Draymond Green-ian "spacing, D and passing" league, Patterson and Carroll have the opposite problem as they are are analytically friendly as quality defenders who space the floor but are frustrating when it comes to the still important part of putting up real boxscore stats

Yes, interesting observation. I remember two years ago when JVal was leading the league in post PPP by a healthy margin and was clearly a force down low, but the Raps' offensive rating was much higher with him off the court because they were taking and making a bunch more threes,

[lorak]

Anyone knows when RAPM would be available?

[SideshowBob]

Anyone knows when RAPM would be available?

JE usually releases his 1st set of data around January (once we pass the 30 games mark).

[QRich3]

Read on Twitter

Found it a bit weird that JE considers RPM the more accurate version compared to RAPM (in this case he was talking about vanilla 1-year RAPM but still), when I thought consensus was the other way round? Does he just mean it has more predictive value? or where is he coming from?

[Dr Spaceman]

Read on Twitter

Found it a bit weird that JE considers RPM the more accurate version compared to RAPM (in this case he was talking about vanilla 1-year RAPM but still), when I thought consensus was the other way round? Does he just mean it has more predictive value? or where is he coming from?

Doctor MJ has talked about this a few times; I think he and J.E. are on pretty friendly terms.

Basically, J.E. (and other statisticians) are looking for the "holy grail" of basketball statistics. Basically, a stat that could truly be used as a "player goodness" metric. Doc and others have told him (in lighter terms) this is foolish, as any good analyst will tell you the more tools there are to capture diverse viewpoints the better you can make judgements. J.E., obviously, disagrees, which is why instead of refining RAPM he's gone ahead and merged it with his box score metric. It's more a philosophical issue than a math one.

So to him, the current iteration of plus/minus stats work "better", while to people like us they have lost a lot of value in precisely the use case they were designed for.

[lorak]

Read on Twitter

Found it a bit weird that JE considers RPM the more accurate version compared to RAPM (in this case he was talking about vanilla 1-year RAPM but still), when I thought consensus was the other way round? Does he just mean it has more predictive value? or where is he coming from?

Doctor MJ has talked about this a few times; I think he and J.E. are on pretty friendly terms.

Basically, J.E. (and other statisticians) are looking for the "holy grail" of basketball statistics. Basically, a stat that could truly be used as a "player goodness" metric. Doc and others have told him (in lighter terms) this is foolish, as any good analyst will tell you the more tools there are to capture diverse viewpoints the better you can make judgements. J.E., obviously, disagrees, which is why instead of refining RAPM he's gone ahead and merged it with his box score metric. It's more a philosophical issue than a math one.

So to him, the current iteration of plus/minus stats work "better", while to people like us they have lost a lot of value in precisely the use case they were designed for.

Umm, that's not the case at all. Engelmann (and other mathematicians!) prefers RPM, because it has better predictive value (what is proven by calculations) than RAPM.

[QRich3]

Doctor MJ has talked about this a few times; I think he and J.E. are on pretty friendly terms.

Basically, J.E. (and other statisticians) are looking for the "holy grail" of basketball statistics. Basically, a stat that could truly be used as a "player goodness" metric. Doc and others have told him (in lighter terms) this is foolish, as any good analyst will tell you the more tools there are to capture diverse viewpoints the better you can make judgements. J.E., obviously, disagrees, which is why instead of refining RAPM he's gone ahead and merged it with his box score metric. It's more a philosophical issue than a math one.

So to him, the current iteration of plus/minus stats work "better", while to people like us they have lost a lot of value in precisely the use case they were designed for.

Oh cool. I'm obviously more on Doc's side philosophically (mathematically I don't have enough of a clue to be anywhere lol), but it's still odd that he's gone so far to that side. I guess it's working for his career, since there's more of a demand from the general public to get a quick one number summary of a player than a more nuanced group of different stats that require more effort to understand. A pity nonetheless.

Which brings me to my next question, is there a place where you can comfortably look up the various RAPM iterations year by year like you used to on JE's old site? or is everyone going from locally saved versions of google docs links and dropbox files like I am?

Umm, that's not the case at all. Engelmann (and other mathematicians!) prefers RPM, because it has better predictive value (what is proven by calculations) than RAPM.

Does it, in your opinion? Wasn't it in this very thread where we were talking about a few specific cases from Clippers players where it was apparent that RPM was being misguided by boxscore numbers that painted the wrong picture, but RAPM corrected them closer to (what we both believed to be) reality?

[lorak]

Does it, in your opinion? Wasn't it in this very thread where we were talking about a few specific cases from Clippers players where it was apparent that RPM was being misguided by boxscore numbers that painted the wrong picture, but RAPM corrected them closer to (what we both believed to be) reality?

Yes, but these are two different things. Mathematicians like Engelmann - unlike what Spaceman said - doesn't care about holy grail, they just want to predict things more accurately. And that job RPM does better than RAPM. However explain the past is quite different matter (but both are linked to each other) and that's why some people - me including - prefer RAPM, when they evaluate players.

[Dr Spaceman]

Oh cool. I'm obviously more on Doc's side philosophically (mathematically I don't have enough of a clue to be anywhere lol), but it's still odd that he's gone so far to that side. I guess it's working for his career, since there's more of a demand from the general public to get a quick one number summary of a player than a more nuanced group of different stats that require more effort to understand. A pity nonetheless.

Yup. Although his work with RAPM itself got him a job with the Phoenix front office, IIRC, so who really knows.

I'm not really privy to the math either, but as I understand it it shouldn't be too difficult to release RAPM every once in a while either since he's already don't the hard work of collecting PBP data. But sounds like he's trending away from that as well, although I'm obviously ignorant on the matter.

Which brings me to my next question, is there a place where you can comfortably look up the various RAPM iterations year by year like you used to on JE's old site? or is everyone going from locally saved versions of google docs links and dropbox files like I am?

Mostly we rely on spreadsheets in someone's google drive for the last 5 years or so (or just RPM). Doc collected all of J.E.s data in a single spreadsheet that spans 99-12 excepting 01, but stopped when J.E. started mislabeling and obfuscating things on his website. We hat GotBuckets, but they didn't renew their URL lease. Your best bet is to look for the links in this thread; I believe SSB catalogues what he sees as well.

[QRich3]

Yes, but these are two different things. Mathematicians like Engelmann - unlike what Spaceman said - doesn't care about holy grail, they just want to predict things more accurately. And that job RPM does better than RAPM. However explain the past is quite different matter (but both are linked to each other) and that's why some people - me including - prefer RAPM, when they evaluate players.

So how is it that they reached the conclusion that something that's more inaccurate in explaining the past because of being misguided by numbers that don't actually say what they're used for (steals/blocks = good defense, etc.), will predict the future better?

Mostly we rely on spreadsheets in someone's google drive for the last 5 years or so (or just RPM). Doc collected all of J.E.s data in a single spreadsheet that spans 99-12 excepting 01, but stopped when J.E. started mislabeling and obfuscating things on his website. We hat GotBuckets, but they didn't renew their URL lease. Your best bet is to look for the links in this thread; I believe SSB catalogues what he sees as well.

Cool, yeah, I think I have most of those compiled in a folder in my home computer, thought there would be a place where it's all neatly organized (someone do a website, you'll definitely get a bunch of clicks!)

Thanks to both for the answers by the way

[Dr Spaceman]

Does it, in your opinion? Wasn't it in this very thread where we were talking about a few specific cases from Clippers players where it was apparent that RPM was being misguided by boxscore numbers that painted the wrong picture, but RAPM corrected them closer to (what we both believed to be) reality?

Yes, but these are two different things. Mathematicians like Engelmann - unlike what Spaceman said - doesn't care about holy grail, they just want to predict things more accurately. And that job RPM does better than RAPM. However explain the past is quite different matter (but both are linked to each other) and that's why some people - me including - prefer RAPM, when they evaluate players.

Then what exactly are PER and WS designed to do?

Look I don't have the exact quote in front of me but it doesn't seem like something that would just be made up.

[lorak]

So how is it that they reached the conclusion that something that's more inaccurate in predicting the past because of being misguided by numbers that don't actually say what they're used for (steals/blocks = good defense, etc.), will predict the future better?

You can't "predict the past" ;]

And that conclusion about predictive power was reached by running tests: based on players' RAPM/RPM and minutes distribution you can predict points differential. You do it for every game in X last seasons and margin of error would be lower if you would use RPM than RAPM.

BTW, using blended metrics (boxscore/PM with player tracking data) gives even better results.

[QRich3]

haha you managed to quote me before my ninja edit

Without having any background on working statistics, I would think at first that the fact it managed to predict things better in the past should not mean it's inherently a better predictor, specially in this case where it seems evident that the boxscore data only makes it more probable to be plainly wrong.

Anyway, it will be a pity if RAPM ends up fading out totally and being replaced by RPM altogether.

[Clyde Frazier]

Read on Twitter

[Clyde Frazier]

Read on Twitter

Read on Twitter