Some questions about on/off,rapm,epm and other advanced stats

[Grimreaper]

1. How is on/off per 48 calculated and is it better than on/off per 100 possessions?

2. Who has the highest on court rating per 48 regular season and playoffs. Highest on/off too for regular season and playoffs

3.give ur opinion on the best all in one impact metrics for measuring players impact

[Grimreaper]

From my phone screen it seems the remaining text doesn't show. it says RAPM,epm and other advanced stats

[Colbinii]

1. How is on/off per 48 calculated and is it better than on/off per 100 possessions?

You would simply figure out a players net-ON/OFF and Minutes and scale them to a minute or possession number.

I believe Basketball-Reference estimates the total possessions (PER 100) while NBA.com uses exact possessions played.

2. Who has the highest on court rating per 48 regular season and playoffs. Highest on/off too for regular season and playoffs

I would look at players in the Top 10 since 1997 (Curry, Kobe, LeBron, KG, Duncan, Dirk, CP3, Durant and also Draymond).

3.give ur opinion on the best all in one impact metrics for measuring players impact

Take a look here.

https://www.cryptbeam.com/2021/05/21/the-10-best-nba-impact-metrics/

[Doctor MJ]

1. How is on/off per 48 calculated and is it better than on/off per 100 possessions?

2. Who has the highest on court rating per 48 regular season and playoffs. Highest on/off too for regular season and playoffs

3.give ur opinion on the best all in one impact metrics for measuring players impact

Welcome Grim!

1. Per 48 refers to 48 minutes of time played. In general I prefer Per 100 because it normalizes for pace.

2. First, do you mean per season or per career? Now, unfortunately in either case I don't have a quick way to query that information, but if you're more specific I can say how I would get the necessary data.

3. So, on the idea of "all in one impact metrics" let me make a distinction:

I tend to see "all in one metrics" as metrics that incorporate everything they can in the box score, which nowadays can include +/-. I don't have strong feelings about the best of these at the moment, but Dunks & Threes' Estimated Plus Minus I'd say has seems to be the most preferred free publicly available model right now.

I'd note that by using the term "plus minus" in the model, that's a statement that they are intending the model to model impact, and so it certainly seems to qualify based on your wording.

I'll say that I prefer to do analysis with production (traditional box score) and impact (+/-) separated. Each on their own is less powerful than a model like EPM, but all-in-one can only tell you so much.

Caveat: Once player tracking gets detailed enough, I will be very, very interested in using all-in-one models for feature analysis which will give us an ability to speak to player footprint as an n-dimensional shape. The rough part here is that I think such stats may never really be available to the public - not even available with a consumer-level subscription.

[eminence]

1. They'll be on slightly different scales, but the order of /48 stats and /100 stats will be nearly the same. Technically /100 is probably a tiny bit 'better', but it's not very meaningful, and we don't have the possession data for a few of the early seasons, so need to use /48 there (or /36).

2. Not sure off the top, are you talking for career or for one season? MJ/Pippen, Duncan/Manu, KG, Dirk, LeBron, Steph/Dray, CP3, Jokic, Giannis would be some of the names to look at.

3. I'd go DARKO, but there's not that meaningful of a difference between any of the xRAPMs.

[Grimreaper]

1. How is on/off per 48 calculated and is it better than on/off per 100 possessions?

2. Who has the highest on court rating per 48 regular season and playoffs. Highest on/off too for regular season and playoffs

3.give ur opinion on the best all in one impact metrics for measuring players impact

Welcome Grim!

1. Per 48 refers to 48 minutes of time played. In general I prefer Per 100 because it normalizes for pace.

2. First, do you mean per season or per career? Now, unfortunately in either case I don't have a quick way to query that information, but if you're more specific I can say how I would get the necessary data.

3. So, on the idea of "all in one impact metrics" let me make a distinction:

I tend to see "all in one metrics" as metrics that incorporate everything they can in the box score, which nowadays can include +/-. I don't have strong feelings about the best of these at the moment, but Dunks & Threes' Estimated Plus Minus I'd say has seems to be the most preferred free publicly available model right now.

I'd note that by using the term "plus minus" in the model, that's a statement that they are intending the model to model impact, and so it certainly seems to qualify based on your wording.

I'll say that I prefer to do analysis with production (traditional box score) and impact (+/-) separated. Each on their own is less powerful than a model like EPM, but all-in-one can only tell you so much.

Caveat: Once player tracking gets detailed enough, I will be very, very interested in using all-in-one models for feature analysis which will give us an ability to speak to player footprint as an n-dimensional shape. The rough part here is that I think such stats may never really be available to the public - not even available with a consumer-level subscription.

Hi there, 1. I mean overall career playoffs and regular season,peak on court rating would be nice too.

2. Abt the on/off, isn't per 75 the usual amount of possession star players have

3. I see, is there a way I can get historical data for this advanced stats for free supposed epm and RAPM

[Doctor MJ]

3.give ur opinion on the best all in one impact metrics for measuring players impact

Take a look here.

https://www.cryptbeam.com/2021/05/21/the-10-best-nba-impact-metrics/

Love his analysis of each model but do want to point out that Englemann didn't come up with RAPM, Joe Sill did.

Paper: Improved NBA Adjusted +/- Using Regularization and Out-of-Sample Testing

[AEnigma]

I suspect Manu may be the regular season and playoff on-court leader and that Garnett may be the regular season and playoff on/off leader, although for playoff on/off it depends on how you factor Garnett’s absence in 2009, and for regular season on/off you may be able to make an incomplete career case for Jokic.

EDIT: Draymond was a good call by Eminence for playoffs on-court; forgot to think about him.

[Doctor MJ]

1. How is on/off per 48 calculated and is it better than on/off per 100 possessions?

2. Who has the highest on court rating per 48 regular season and playoffs. Highest on/off too for regular season and playoffs

3.give ur opinion on the best all in one impact metrics for measuring players impact

Welcome Grim!

1. Per 48 refers to 48 minutes of time played. In general I prefer Per 100 because it normalizes for pace.

2. First, do you mean per season or per career? Now, unfortunately in either case I don't have a quick way to query that information, but if you're more specific I can say how I would get the necessary data.

3. So, on the idea of "all in one impact metrics" let me make a distinction:

I tend to see "all in one metrics" as metrics that incorporate everything they can in the box score, which nowadays can include +/-. I don't have strong feelings about the best of these at the moment, but Dunks & Threes' Estimated Plus Minus I'd say has seems to be the most preferred free publicly available model right now.

I'd note that by using the term "plus minus" in the model, that's a statement that they are intending the model to model impact, and so it certainly seems to qualify based on your wording.

I'll say that I prefer to do analysis with production (traditional box score) and impact (+/-) separated. Each on their own is less powerful than a model like EPM, but all-in-one can only tell you so much.

Caveat: Once player tracking gets detailed enough, I will be very, very interested in using all-in-one models for feature analysis which will give us an ability to speak to player footprint as an n-dimensional shape. The rough part here is that I think such stats may never really be available to the public - not even available with a consumer-level subscription.

Hi there, 1. I mean overall career playoffs and regular season,peak on court rating would be nice too.

2. Abt the on/off, isn't per 75 the usual amount of possession star players have

3. I see, is there a way I can get historical data for this advanced stats for free supposed epm and RAPM

1. So, here's what I can quickly query for you from basketball-reference:

For combined games in multiple seasons, in the NBA/BAA, in the regular season or the postseason, sorted by descending Plus-Minus.

For combined games in one season, in the NBA/BAA, in the regular season or the postseason, sorted by descending Plus-Minus.

Note that in both cases the list goes only 200 deep, and because they're sorted by totals, players who don't play enough will miss the list.

To get what you want from the list of players in each, put it into a spreadsheet, and create a formula that divides +/- by MP and then multiply by 48.

2. Per 75 is more what NBA players play in a game than Per 100, just as Per 36 MP is more in line than Per 48. I generally prefer Per 100 because a) it's a nice round number, and b) it's the same regardless of the length of games which allows us to compare between leagues. Also, if you want something that speaks to the total of what a player did, the raw +/- already does that.

3. I don't believe EPM gives its historical data for free.

RAPM is trickier because there isn't one official RAPM metric. I'd say the most respected free stuff available right now might come fromAhmed Cheema, but if others disagree I hope they post.

[eminence]

Checked the guys I listed earlier (minus MJ/Pippen), per bbref.

Player: RS On, RS On/Off, PO On, PO On/Off

Duncan: +9.7, +8.0, +6.0, +7.5
Manu: +10.2, +6.5, +7.7, +10.2
Dirk: +5.6, +9.0, -0.2, +2.0
KG: +5.2, +11.3, +2.5, +14.5
LeBron: +6.6, +10.7, +5.9, +10.2
CP3: +6.8, +9.5, +0.9, +6.2
Steph: +7.9, +10.5, +7.6, +12.0
Dray: +9.1, +9.3, +9.2, +12.6
Giannis: +4.6, +7.0, +4.3, +8.0
Jokic: +6.7, +11.5, +1.3, +2.8

So the highest among guys I thought of to check...
RS On: Manu +10.2
RS On/Off: Jokic +11.5
PO On: Dray +9.2
PO On/Off: KG +14.5

[PooledSilver]

3.give ur opinion on the best all in one impact metrics for measuring players impact

Take a look here.

https://www.cryptbeam.com/2021/05/21/the-10-best-nba-impact-metrics/

Love his analysis of each model but do want to point out that Englemann didn't come up with RAPM, Joe Sill did.

Paper: Improved NBA Adjusted +/- Using Regularization and Out-of-Sample Testing

I’d go LEBRON over EPM, with the new padding and adjustments to it recently I’ve seen some results of it in season projections vs EPM vs Darko and it absolutely destroyed the other all in ones

A lot of all in ones since it’s about generalizing a trend to find the most accurate one in general with get a player right or wrong in a specific direction constantly if their box score overestimate or underscores their impact.

I think there are other issues like lineup synergy and individual performance meaningfully increasing or decreasing suddenly on both ends as a season goes on

[PooledSilver]

3.give ur opinion on the best all in one impact metrics for measuring players impact

Take a look here.

https://www.cryptbeam.com/2021/05/21/the-10-best-nba-impact-metrics/

Love his analysis of each model but do want to point out that Englemann didn't come up with RAPM, Joe Sill did.

Paper: Improved NBA Adjusted +/- Using Regularization and Out-of-Sample Testing

I’d go LEBRON over EPM, with the new padding and adjustments to it recently I’ve seen some results of it in season projections vs EPM vs Darko and it absolutely destroyed the other all in ones

A lot of all in ones since it’s about almost finding a trend to find the most accurate one in a general sense when applied to everyone, but they can get a player right or wrong in a specific direction constantly if their box score overestimate or underscores their impact. LEBRON for example pretty clearly undershoots lebron consistently when you cross reference his RAPM and impact, esp defensively

Luck adjusted Rapm too I’m 50/50 on, the 3 point adjustments specifically. If something is radically different from the regular rapm at a consistent level year to year then I question if it’s luck versus reality, and I’ve heard multiyear luck data is blah. Free throws I think are fair though.

I think there are other issues like lineup synergy and individual performance meaningfully increasing or decreasing suddenly on both ends as a season goes on

[Grimreaper]

Checked the guys I listed earlier (minus MJ/Pippen), per bbref.

Player: RS On, RS On/Off, PO On, PO On/Off

Duncan: +9.7, +8.0, +6.0, +7.5
Manu: +10.2, +6.5, +7.7, +10.2
Dirk: +5.6, +9.0, -0.2, +2.0
KG: +5.2, +11.3, +2.5, +14.5
LeBron: +6.6, +10.7, +5.9, +10.2
CP3: +6.8, +9.5, +0.9, +6.2
Steph: +7.9, +10.5, +7.6, +12.0
Dray: +9.1, +9.3, +9.2, +12.6
Giannis: +4.6, +7.0, +4.3, +8.0
Jokic: +6.7, +11.5, +1.3, +2.8

So the highest among guys I thought of to check...
RS On: Manu +10.2
RS On/Off: Jokic +11.5
PO On: Dray +9.2
PO On/Off: KG +14.5

Thanks for the info

[Dutchball97]

Relying on just one advanced metric is always going to give a skewed picture because every stat has their biases in terms of what player profiles consistently get higher or lower scores. The more inputs the better generally alhough there definitely are stats where you can wonder if including them actually adds anything or actually detracts from it. EPM and LEBRON are the stats that influence my view the most but I like to take BPM and raw +- into account as well. Another thing I do is look at team ORTG and DRTG, especially when contrasting it with +- to get a better picture if someone is providing a lot of lift because the floor for the team is so low or if they're pushing a team to greater heights.

[Djoker]

This thread has additional plus minus data going back to 1994 from Harvey Pollack's Statistical Yearbooks. colts18 ranked the best regular seasons starting from 1994 and then SpreeS updated until 2023.

viewtopic.php?f=344&t=1343246&start=200

And then here is a thread that contains a lot of MJ plus-minus data. It has his entire playoff career and large pieces of the regular season (~60% of career) with large samples from 1991 onwards.

viewtopic.php?f=64&t=2314587

[Grimreaper]

This thread has additional plus minus data going back to 1994 from Harvey Pollack's Statistical Yearbooks. colts18 ranked the best regular seasons starting from 1994 and then SpreeS updated until 2023.

viewtopic.php?f=344&t=1343246&start=200

And then here is a thread that contains a lot of MJ plus-minus data. It has his entire playoff career and large pieces of the regular season (~60% of career) with large samples from 1991 onwards.

viewtopic.php?f=64&t=2314587

Where did u get the bull games to track from and the other posters u linked ??

[Djoker]

This thread has additional plus minus data going back to 1994 from Harvey Pollack's Statistical Yearbooks. colts18 ranked the best regular seasons starting from 1994 and then SpreeS updated until 2023.

viewtopic.php?f=344&t=1343246&start=200

And then here is a thread that contains a lot of MJ plus-minus data. It has his entire playoff career and large pieces of the regular season (~60% of career) with large samples from 1991 onwards.

viewtopic.php?f=64&t=2314587

Where did u get the bull games to track from and the other posters u linked ??

MJ's Bulls playoff games are generally easy to find. The majority are free on YouTube or Facebook. Some I bought from collectors like for instance most of the 1989 Bulls-Knicks series.

I'm currently tracking Orlando Shaq and in the process of getting more 80's Lakers games to track Magic/Kareem. Will probably do other players like Hakeem and Bird at some point too.

I only track playoff games. Regular season games are harder to find but some collectors I know have thousands of games and will part with them for a small fee.