2017-18 RAPM/RPM/etc. Thread

[bondom34]

This is same as saying that playoff stats don't matter because of low sample size. Yes, it's possible that was an anomaly and Curry offensive impact in the playoffs will be better in the future. But it's also possible his game doesn't translate in the playoffs and it's not as impactful. Simply dismissing it is ignorant.

No, because stats aren't entirely based on sample size.

The entire premise of RAPM is based upon needing a larger sample.

About 25% of RS games is enough to have stable RAPM results. So while 1 year playoffs RAPM isn't too informative (besides players from teams, which reached the finals), that 2014-2018 set done by Shadow is a GREAT thing.

Which is exactly what we were talking about, so...

[Trundle]

No, because stats aren't entirely based on sample size.

The entire premise of RAPM is based upon needing a larger sample.

About 25% of RS games is enough to have stable RAPM results. So while 1 year playoffs RAPM isn't too informative (besides players from teams, which reached the finals), that 2014-2018 set done by Shadow is a GREAT thing.

Which is exactly what we were talking about, so...

You were arguing against KD > Curry in 5 years PO sample and were using "small sample" as main argument. It doesn't make sense to use in this case, as that sample is FAR BIGGER than needed for RAPM.

[eminence]

About 25% of RS games is enough to have stable RAPM results. So while 1 year playoffs RAPM isn't too informative (besides players from teams, which reached the finals), that 2014-2018 set done by Shadow is a GREAT thing.

Which is exactly what we were talking about, so...

You were arguing against KD > Curry in 5 years PO sample and were using "small sample" as main argument. It doesn't make sense to use in this case, as that sample is FAR BIGGER than needed for RAPM.

Ehh, it's big enough in that enough games are played. Playoff RAPM will always have the matchup problem and the off-minutes sample being small for star players. In Curry's particular case I think he gets knocked down a couple pegs too many for missing those first round games.

[bondom34]

About 25% of RS games is enough to have stable RAPM results. So while 1 year playoffs RAPM isn't too informative (besides players from teams, which reached the finals), that 2014-2018 set done by Shadow is a GREAT thing.

Which is exactly what we were talking about, so...

You were arguing against KD > Curry in 5 years PO sample and were using "small sample" as main argument. It doesn't make sense to use in this case, as that sample is FAR BIGGER than needed for RAPM.

What eminence said.

Also I wouldn't trust 1/4 season's data for RAPM. To me that is too small a sample.

[dhsilv2]

Which is exactly what we were talking about, so...

You were arguing against KD > Curry in 5 years PO sample and were using "small sample" as main argument. It doesn't make sense to use in this case, as that sample is FAR BIGGER than needed for RAPM.

What eminence said.

Also I wouldn't trust 1/4 season's data for RAPM. To me that is too small a sample.

Full season with vanilla RAPM has issues which is why priors greatly improve the stability of the results. 1/4 season is really limited!

[Trundle]

Also I wouldn't trust 1/4 season's data for RAPM. To me that is too small a sample.

Trust is one of the fundamentals of faith, not math.

[bondom34]

Also I wouldn't trust 1/4 season's data for RAPM. To me that is too small a sample.

Trust is one of the fundamentals of faith, not math.

Wouldn't rely on as something worth using.

[eminence]

Also I wouldn't trust 1/4 season's data for RAPM. To me that is too small a sample.

Trust is one of the fundamentals of faith, not math.

Lol, this one made me laugh. And from what I've seen I agree that somewhere from 1/4 to 1/3 a regular season is how long it takes RAPM to stabilize to an acceptable degree. Now I think the other concerns I listed still apply when it comes to the playoffs.

[dhsilv2]

Also I wouldn't trust 1/4 season's data for RAPM. To me that is too small a sample.

Trust is one of the fundamentals of faith, not math.

Lol, this one made me laugh. And from what I've seen I agree that somewhere from 1/4 to 1/3 a regular season is how long it takes RAPM to stabilize to an acceptable degree. Now I think the other concerns I listed still apply when it comes to the playoffs.

NPI RAPM is starting to stabilize with 1/4 to 1/3? I have a very hard time with that assumption. Even the full year stuff often doesn't look finished when it is NPI.

[eminence]

Trust is one of the fundamentals of faith, not math.

Lol, this one made me laugh. And from what I've seen I agree that somewhere from 1/4 to 1/3 a regular season is how long it takes RAPM to stabilize to an acceptable degree. Now I think the other concerns I listed still apply when it comes to the playoffs.

NPI RAPM is starting to stabilize with 1/4 to 1/3? I have a very hard time with that assumption. Even the full year stuff often doesn't look finished when it is NPI.

Well that's my experience at least, don't usually see much (non-injury influenced) movement after that 1/3rd mark. Rookies I guess tend to move after that point.

[trex_8063]

2017 - 2018 RAPM | APBR / shadow | 1-Year / Zero Prior

Has there been any updates to include the playoffs and/or a PI model made yet?

[dice]

Trust is one of the fundamentals of faith, not math.

Lol, this one made me laugh. And from what I've seen I agree that somewhere from 1/4 to 1/3 a regular season is how long it takes RAPM to stabilize to an acceptable degree. Now I think the other concerns I listed still apply when it comes to the playoffs.

NPI RAPM is starting to stabilize with 1/4 to 1/3? I have a very hard time with that assumption. Even the full year stuff often doesn't look finished when it is NPI.

i've read that it takes 5000 minutes to stabilize (2 full seasons for a healthy superstar). and even that produces some head-scratching results. so NPI RAPM for even a full season should comprise no more than 50% of your player eval toolbox. a quarter of a season is like 10% reliable. there's some value in it, but should be taken with a mound of salt

[thekdog34]

250 day and 500 day RAPM

http://canzhiye.com/rapm/

[E-Balla]

250 day and 500 day RAPM

http://canzhiye.com/rapm/

Why such weird numbers like 250 day and 500 days? Like I would think games or seasons made more sense... I mean the NBA season isn't a square number of days it's like 180 days (6 months).

[Colbinii]

250 day and 500 day RAPM

http://canzhiye.com/rapm/

Why such weird numbers like 250 day and 500 days? Like I would think games or seasons made more sense... I mean the NBA season isn't a square number of days it's like 180 days (6 months).

Why 250 and 500 day samples? Why trailing? There is a ton of noise in RAPM for small samples. The single-season RAPM numbers that are floating out there (like ESPN’s single-season RPM) suffer from this. They might do an okay job describing how much players have contributed to winning in the given season, but those numbers are definitely not predictive. In light of that, I chose larger samples – 250 game days is about 1.5 seasons, and 500 game days is about 3 seasons. Using a trailing sample removes the arbitrary lines between seasons.

[eminence]

250 day and 500 day RAPM

http://canzhiye.com/rapm/

Any chance somebody could check the correlations between the 2 (both in absolute value and rank order)?

[E-Balla]

250 day and 500 day RAPM

http://canzhiye.com/rapm/

Why such weird numbers like 250 day and 500 days? Like I would think games or seasons made more sense... I mean the NBA season isn't a square number of days it's like 180 days (6 months).

Why 250 and 500 day samples? Why trailing? There is a ton of noise in RAPM for small samples. The single-season RAPM numbers that are floating out there (like ESPN’s single-season RPM) suffer from this. They might do an okay job describing how much players have contributed to winning in the given season, but those numbers are definitely not predictive. In light of that, I chose larger samples – 250 game days is about 1.5 seasons, and 500 game days is about 3 seasons. Using a trailing sample removes the arbitrary lines between seasons.

I get that but I'm still confused at them doing that instead of 2 and 3 seasons. Just to be 100% accurate. As it is IDK exactly where the same starts, I just have an idea.

[Clyde Frazier]

Read on Twitter

I've merely skimmed this, but it's interesting. Cool to see Rod Strickland on the 97-98 list as he seems to be written off as a talented yet lower impact player. You also have Antoine Walker and Marbury on there who were considered negative impact guys (not that I disagree). Although I wonder if it being early in their careers has anything to do with it.

[bondom34]

Read on Twitter

I've merely skimmed this, but it's interesting. Cool to see Rod Strickland on the 97-98 list as he seems to be written off as a talented yet lower impact player. You also have Antoine Walker and Marbury on there who were considered negative impact guys (not that I disagree). Although I wonder if it being early in their careers has anything to do with it.

Thought of posting it earlier, seems a tough one to track but some interesting results.

[Clyde Frazier]

Read on Twitter

I've merely skimmed this, but it's interesting. Cool to see Rod Strickland on the 97-98 list as he seems to be written off as a talented yet lower impact player. You also have Antoine Walker and Marbury on there who were considered negative impact guys (not that I disagree). Although I wonder if it being early in their careers has anything to do with it.

Thought of posting it earlier, seems a tough one to track but some interesting results.

Yup, especially with this in mind:

Read on Twitter