New EPM (estimated plus minus) metric.

[OkcSinceSGA]

Read on Twitter

https://dunksandthrees.com/epm


Co creator of RPM seems pretty impressed in the comments. Looks promising. One sign of why the Clips may struggle so badly on D with Pat Beverley out. Has to be something to this list, because it's the first metric I've seen pretty much nail the top 30 NBA players, even mostly in order. There aren't any crazy anomalies that jump out to me.

[OkcSinceSGA]

Trae Young is the 11th worst defender in this. 5th best offensive player. Bradley Beal is the 2nd worst defender in the NBA in this metric.

[Reeko]

So am I supposed to believe that Pat Bev is better than Ben Simmons, Jayson Tatum and Kyle Lowry? Or that Karl Anthony Towns is a top 5 player in the league?

[THE J0KER]

EW (Estimated Wins) column is similar, and probably more relevant.

[Jabroni Lames]

Read on Twitter

https://dunksandthrees.com/epm


Co creator of RPM seems pretty impressed in the comments. Looks promising. One sign of why the Clips may struggle so badly on D with Pat Beverley out. Has to be something to this list, because it's the first metric I've seen pretty much nail the top 30 NBA players, even mostly in order. There aren't any crazy anomalies that jump out to me.

lol. Yeah, because when I think Top 30, Richaun Holmes and Daniel Theis immediately come to mind.

[dhsilv2]

So am I supposed to believe that Pat Bev is better than Ben Simmons, Jayson Tatum and Kyle Lowry? Or that Karl Anthony Towns is a top 5 player in the league?

No, when you're on a player you'll see a grey zone, more or less think of that player in that zone. People have to understand there are NO stats like this where they are ranking players. it's all "bands".

Next these never discuss "better". They discuss impactful within the system they are in.

[OkcSinceSGA]

Read on Twitter

https://dunksandthrees.com/epm


Co creator of RPM seems pretty impressed in the comments. Looks promising. One sign of why the Clips may struggle so badly on D with Pat Beverley out. Has to be something to this list, because it's the first metric I've seen pretty much nail the top 30 NBA players, even mostly in order. There aren't any crazy anomalies that jump out to me.

lol. Yeah, because when I think Top 30, Richaun Holmes and Daniel Theis immediately come to mind.

Key words, pretty much. Not to mention both check out really well in multiple metrics anyway. They are near the what 25 position? Not the same kind of anomaly as guys like Robert Covington, Tyson Chandler etc being top 3 in the league in year's past etc.

[Duke4life831]

KAT at #4 is a pretty big sore thumb in my opinion. KAT having a higher impact than guys like Jokic, Luka, Butler and Gobert makes me chuckle.

Im sure this is just like every other "PM" stat. Some good, some bad, ask one person who is a big advanced metrics guy and they will say they love it, then ask another advanced metric guy and they will tell you its useless trash. What is needed is a site that is basically just an excel sheet that is a composite of all the "PM" stats and gives us the average ranking.

[bondom34]

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

[dhsilv2]

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

I honestly think the tracking data is great, but how they're using it makes absolutely no sense. It's like they are just adamant that everything we've learned about how team defense is more valued verse man defense the last 20 years can't be right and we should skew defense to man defense. And then I'm not even sure they're using it right from there.

[bondom34]

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

I honestly think the tracking data is great, but how they're using it makes absolutely no sense. It's like they are just adamant that everything we've learned about how team defense is more valued verse man defense the last 20 years can't be right and we should skew defense to man defense. And then I'm not even sure they're using it right from there.

I don't really trust the data, and the definitions are generally poor. "Contested" shots are often not really well contested. Definitions of "open" don't fit, and tracking who's guarding who is generally not terribly fitting. Plus trusting the definition of a play type is even sketchy.

Agree on the team defense though, but from what I've seen most every metric that tries to involve player tracking data has looked worse on defense at least.

[OkcSinceSGA]

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

Interestingly enough in this guy's comments... the co creator of RPM implies he's not a fan of the new in house RPM through ESPN for whatever reason, but he goes on to praise this EPM.

[clyde21]

im all advanced stated out tbh.

[OkcSinceSGA]

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

I honestly think the tracking data is great, but how they're using it makes absolutely no sense. It's like they are just adamant that everything we've learned about how team defense is more valued verse man defense the last 20 years can't be right and we should skew defense to man defense. And then I'm not even sure they're using it right from there.

I don't really trust the data, and the definitions are generally poor. "Contested" shots are often not really well contested. Definitions of "open" don't fit, and tracking who's guarding who is generally not terribly fitting. Plus trusting the definition of a play type is even sketchy.

Agree on the team defense though, but from what I've seen most every metric that tries to involve player tracking data has looked worse on defense at least.

This is something I've noticed too. You'll have NBA.com say 17 of 25 shots were contested.. then Synergy will say 9 of 25 etc. It's hard to keep track when they have different criteria. I personally trust Synergy above all though, because I watch the games very closely, and it seems to align closest with what I PERSONALLY also define as an open shot etc (aka eye test). For me a contested shot means a guy is very close to you with a hand up pretty much. If he's trapped on a screen and 5 feet away... or running towards you, but still a good ways away with hands down.. that's not a contested shot to me.

[bondom34]

I honestly think the tracking data is great, but how they're using it makes absolutely no sense. It's like they are just adamant that everything we've learned about how team defense is more valued verse man defense the last 20 years can't be right and we should skew defense to man defense. And then I'm not even sure they're using it right from there.

I don't really trust the data, and the definitions are generally poor. "Contested" shots are often not really well contested. Definitions of "open" don't fit, and tracking who's guarding who is generally not terribly fitting. Plus trusting the definition of a play type is even sketchy.

Agree on the team defense though, but from what I've seen most every metric that tries to involve player tracking data has looked worse on defense at least.

This is something I've noticed too. You'll have NBA.com say 17 of 25 shots were contested.. then Synergy will say 9 of 25 etc. It's hard to keep track when they have different criteria. I personally trust Synergy above all though, because I watch the games very closely, and it seems to align closest with what I PERSONALLY also define as an open shot etc (aka eye test). For me a contested shot means a guy is very close to you with a hand up pretty much. If he's trapped on a screen and 5 feet away... or running towards you, but still a good ways away with hands down.. that's not a contested shot to me.

From what I recall, contested shot definition is pretty unreliable in general for both (remember reading this years ago though so maybe something changed). Bigger issue is that "open" to an NBA player isn't really open to a normal person and you can't tell from a camera angle how much room a guy has.

I kinda wish they'd throw it all out personally for defense.

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

Interestingly enough in this guy's comments... the co creator of RPM implies he's not a fan of the new in house RPM through ESPN for whatever reason, but he goes on to praise this EPM.

FYI though, that's not the original RPM creator. Pretty sure Engelmann was the guy there (Ilyardi has him tagged I believe). Seems to be saying its kind of a home brewed RPM. Personally not a fan, but will see how it tests out before making sure.

I was skeptical of RAPTOR to start for some of the same reasons, and thus far haven't been a fan there. I'd still trust LARAPM and box scores personally.

[CptCrunch]

Code: Select all

Estimated plus-minus (EPM) is an all-in-one player metric that is both retrodictive and predictive. It's retrodictive in that a weighted sum will approximate team ratings, and EPM values weighted by playing time in the following season correlates nicely with win %. It also lends information about future wins when you have an idea of how much a player will play.

EPM is a one-season Regularized Adjusted Plus-Minus (RAPM) calculation with a Bayesian prior. The prior is an estimation of long-term RAPM using stats and some player-tracking-derived metrics. Both the RAPM calculations and modeling are performed by dunksandthrees.com.

Raw plus-minus and player "Net Ratings" are very noisy and, in the opinion of dunksandthrees.com, can oftentimes be misleading even in large samples. The advantage of RAPM based metrics is that there is a lot done to control for factors that cause much of the noise in raw plus-minus, such as who a player is playing with and against.

Estimated Plus-Minus is called such because any all-in-one metric is simply trying to estimate a player's true value which is unknown. Although RAPM-based metrics are much better than raw plus-minus/net rating, it is still only an estimate, and a thorough analysis of a player should accompany the use of EPM, such as with other stats and video. The role and context of a player should be considered as well. A more detailed explanation of EPM may come later.

It should be noted that a different prior for defense is used before the player-tracking era (pre 2014 season). Also, EPM at the game level (i.e. for the 5-game moving average trend) only uses a prior (no short-term RAPM calculation).

Nothing offensively wrong in the description unlike the write up for PIPM.

This is just this dude's version of ESPN's RPM. If ESPN/Engelmann is allowed to name their stat RPM, I don't see how this guy can't devise his own EPM.

I do like his prior description more than ESPN's box-score based one.

This guy's web dev skills are quite good, much better than the average analyst capable of understanding regularized regressions. I wouldn't be surprised if this site turns dark soon after he is hired.

[zimpy27]

The website interface is really cool.


It sounds like he's using long-term RAPM to check his model against which uses single-season player tracking stats.

Not a bad idea but I'd like to see some in-season testing to see if the variations are consistent not just for whole seasons but half seasons as well.

It's actually the way I was thinking of doing it myself as it's the way we make models of biology but I might be reading into it a bit much, I can only see his twitter feed. Any more information?

[Roy The Natural]

I honestly think the tracking data is great, but how they're using it makes absolutely no sense. It's like they are just adamant that everything we've learned about how team defense is more valued verse man defense the last 20 years can't be right and we should skew defense to man defense. And then I'm not even sure they're using it right from there.

I don't really trust the data, and the definitions are generally poor. "Contested" shots are often not really well contested. Definitions of "open" don't fit, and tracking who's guarding who is generally not terribly fitting. Plus trusting the definition of a play type is even sketchy.

Agree on the team defense though, but from what I've seen most every metric that tries to involve player tracking data has looked worse on defense at least.

This is something I've noticed too. You'll have NBA.com say 17 of 25 shots were contested.. then Synergy will say 9 of 25 etc. It's hard to keep track when they have different criteria. I personally trust Synergy above all though, because I watch the games very closely, and it seems to align closest with what I PERSONALLY also define as an open shot etc (aka eye test). For me a contested shot means a guy is very close to you with a hand up pretty much. If he's trapped on a screen and 5 feet away... or running towards you, but still a good ways away with hands down.. that's not a contested shot to me.

We'll yea... Quite simply all we can really look at fairly accurately is how team defense looks with certain players on the floor. Individual defensive ratings are quite frankly not going to be accurate until there is actual exhaustive game tape review involved. Missed rotations need to be involved. The difference in a Giannis contest and a Lillard contest needs to be properly valued. You have post defense vs rim protection vs help defense vs perimeter defense. Most players aren't good at all of them. Quite simply amalgamation stats for individial defense aren't going to do much for your any time soon. We can however state with a large enough sample size that a team seems to give up less points when a certain group of players is on the floor, but without actual game time review we can't pinpoint exactly why, we can just guess and supplement the findings with the eye test.

This metric does seem to do a good job with offense though imo.

[zimpy27]

I'd like to see some idea of what goes into it first, looking at quick glance there are some oddball data points, and I've been very skeptical of the new RPM. Anything with tracking data really.

I honestly think the tracking data is great, but how they're using it makes absolutely no sense. It's like they are just adamant that everything we've learned about how team defense is more valued verse man defense the last 20 years can't be right and we should skew defense to man defense. And then I'm not even sure they're using it right from there.

I don't really trust the data, and the definitions are generally poor. "Contested" shots are often not really well contested. Definitions of "open" don't fit, and tracking who's guarding who is generally not terribly fitting. Plus trusting the definition of a play type is even sketchy.

Agree on the team defense though, but from what I've seen most every metric that tries to involve player tracking data has looked worse on defense at least.

The raw data collected should be pretty good. The form the public get it in is obviously compressed in to definitions. I have never tried to use their data before but does anyone know if you can get raw data from second spectrum?

[Duke4life831]

im all advanced stated out tbh.

This is where Im at as well. I used to be happy to see a new advanced stat come out, my thought process used to be "cool the more stats the better" to now my thought process is "cool, add it to the other 10 advanced stats people cherry pick for when it suits their player".

I feel like talking about the game has lost some of its nuance. Again dont get me wrong I like the data and I use it. But lots of times when talking about players the argument goes something like this.

Person 1: Player x is great, he is 4th in BPM
Person 2: BPM is trash, RPM is actually the best one
Person 1: RPM is trash, it has LeBron as the best defender and AD as a horrible defender
Person 3: Actually RAPM is the best stat, BPM is way to simple and RPM is ESPN garbage
Random Person 4: I like PER
Everyone else: Person 4 is an idiot

We have PER, BPM, VORP, WS, RPM, RAPM, PIPM, RAPTOR, CARMELO, EPM. Not to mention other stats like On/Off Offensive and defensive rating (which is different on different sites...) NET rating and so much more.

We rarely see the comparing of players foot work, and other more nuanced topics. The fact that we are around 10 or so advanced stats, my mind has checked out haha.