formula for RPM apparently adjusted

[dice]

lebron went from head-and-shoulders above the crowd to 3rd overnight in wins produced. zach lavine skyrockets to 14th (ahead of jimmy butler). wtf is going on here?

http://www.espn.com/nba/statistics/rpm/_/sort/WINS

[bondom34]

Read on Twitter

?s=19

I hated the new version incorporating tracking data. Well now ESPN lost even more credibility to me by just deciding to change it entirely mid season. It's way different.

[CptCrunch]

1. They are increasing τ, the precision matrix, of the gaussian prior. Box-score has more weight on DRPM.

2. They are increasing λ, the shrinkage parameter. Players need more evidence of possession level +/- to be rated as good or bad.

Here is an archived version before all-star break, around 6 games old: http://web.archive.org/web/20200212064948/http://www.espn.com/nba/statistics/rpm

The basically pwned the crap out of extreme values in DRPM, which I like since DRPM is not a reliable indicator of defense anyways. Good to not have extreme values like Giannis with 5+ RPM while Trae is like -5.

RAPM/RPM are arbitrary family of statistics anyways. Priors and shrinkage parameter choice can really be arbitrarily chosen.

[dice]

Read on Twitter

?s=19

I hated the new version incorporating tracking data. Well now ESPN lost even more credibility to me by just deciding to change it entirely mid season. It's way different.

it was already mostly box score driven until season's end (at which point it gets to 50/50 w/ on/off). from what i have read, anyway. further weighting it toward box score makes it closer to a BPM type stat. this smacks of an attempt to make the results more palatable to the average ESPN viewer

is there another respected on/off based stat floating around out there?

[dice]

1. They are increasing τ, the precision matrix, of the gaussian prior. Box-score has more weight on DRPM.

2. They are increasing λ, the shrinkage parameter. Players need more evidence of possession level +/- to be rated as good or bad.

Here is an archived version before all-star break, around 6 games old: http://web.archive.org/web/20200212064948/http://www.espn.com/nba/statistics/rpm

The basically pwned the crap out of extreme values in DRPM, which I like since DRPM is not a reliable indicator of defense anyways. Good to not have extreme values like Giannis with 5+ RPM while Trae is like -5.

RAPM/RPM are arbitrary family of statistics anyways. Priors and shrinkage parameter choice can really be arbitrarily chosen.

DRPM was the primary value of the stat. we can all pretty much gauge a player's offensive impact based on box score

and why is it ok to have extremes on offense but not on defense?

zach lavine has a significantly negative raw +/-. i.e. a team with a bad/depleted bench is doing better with it's quasi-star out of the game. he's a modest efficiency scorer. how the hell does he rate so highly?

[bondom34]

Read on Twitter

?s=19

I hated the new version incorporating tracking data. Well now ESPN lost even more credibility to me by just deciding to change it entirely mid season. It's way different.

it was already mostly box score driven until season's end (at which point it gets to 50/50 w/ on/off). from what i have read, anyway. further weighting it toward box score makes it closer to a BPM type stat. this smacks of an attempt to make the results more palatable to the average ESPN viewer

is there another respected on/off based stat floating around out there?

I usually use regular or luck adjusted RAPM from nbashotcharts.com. Some like PIPM too (b-ballindex.com I think?).

[dice]

Read on Twitter

?s=19

I hated the new version incorporating tracking data. Well now ESPN lost even more credibility to me by just deciding to change it entirely mid season. It's way different.

it was already mostly box score driven until season's end (at which point it gets to 50/50 w/ on/off). from what i have read, anyway. further weighting it toward box score makes it closer to a BPM type stat. this smacks of an attempt to make the results more palatable to the average ESPN viewer

is there another respected on/off based stat floating around out there?

I usually use regular or luck adjusted RAPM from nbashotcharts.com. Some like PIPM too (b-ballindex.com I think?).

thanks. i've heard of PIPM. so that RAPM contains no box score? no priors?

[bondom34]

it was already mostly box score driven until season's end (at which point it gets to 50/50 w/ on/off). from what i have read, anyway. further weighting it toward box score makes it closer to a BPM type stat. this smacks of an attempt to make the results more palatable to the average ESPN viewer

is there another respected on/off based stat floating around out there?

I usually use regular or luck adjusted RAPM from nbashotcharts.com. Some like PIPM too (b-ballindex.com I think?).

thanks. i've heard of PIPM. so that RAPM contains no box score? no priors?

I'm not sure the priors used in this single year but I believe it has them, no box score. PIPM is a lot of box score iirc.

[dhsilv2]

No mention of adjustments to how they're using the player tracking data in the change, because that was as I understood it the change they made year over year...and no surprise made for weird stuff.

[CptCrunch]

1. They are increasing τ, the precision matrix, of the gaussian prior. Box-score has more weight on DRPM.

2. They are increasing λ, the shrinkage parameter. Players need more evidence of possession level +/- to be rated as good or bad.

Here is an archived version before all-star break, around 6 games old: http://web.archive.org/web/20200212064948/http://www.espn.com/nba/statistics/rpm

The basically pwned the crap out of extreme values in DRPM, which I like since DRPM is not a reliable indicator of defense anyways. Good to not have extreme values like Giannis with 5+ RPM while Trae is like -5.

RAPM/RPM are arbitrary family of statistics anyways. Priors and shrinkage parameter choice can really be arbitrarily chosen.

DRPM was the primary value of the stat. we can all pretty much gauge a player's offensive impact based on box score

and why is it ok to have extremes on offense but not on defense?

zach lavine has a significantly negative raw +/-. i.e. a team with a bad/depleted bench is doing better with it's quasi-star out of the game. he's a modest efficiency scorer. how the hell does he rate so highly?

Because offensive advanced stats are easy to correlate to box-scores (mainly scoring since you know efficient scoring win basketball games, also people with eyes can see that Harden and LeBron are good at scoring and passing), so we have a sense if any offensive advanced stats 'look' reasonable or not. Most tend to be reasonable.

We have no measure to quantify someone's defensive impact really so if someone were to calculate DRPM or DRAPM, there is nothing to benchmark this against other than subjective eye-test really. Therefore if we don't trust the estimate as much, we can subjectively shrink them towards the mean. The good defenders aren't as good and the bad ones aren't as bad, but of course shrinkage in this ridge context is NOT uniform. When you increase lambda, you are not subtracting/adding a constant to everyone's DRPM.

[nolang1]

1. They are increasing τ, the precision matrix, of the gaussian prior. Box-score has more weight on DRPM.

2. They are increasing λ, the shrinkage parameter. Players need more evidence of possession level +/- to be rated as good or bad.

Here is an archived version before all-star break, around 6 games old: http://web.archive.org/web/20200212064948/http://www.espn.com/nba/statistics/rpm

The basically pwned the crap out of extreme values in DRPM, which I like since DRPM is not a reliable indicator of defense anyways. Good to not have extreme values like Giannis with 5+ RPM while Trae is like -5.

RAPM/RPM are arbitrary family of statistics anyways. Priors and shrinkage parameter choice can really be arbitrarily chosen.

DRPM was the primary value of the stat. we can all pretty much gauge a player's offensive impact based on box score

and why is it ok to have extremes on offense but not on defense?

zach lavine has a significantly negative raw +/-. i.e. a team with a bad/depleted bench is doing better with it's quasi-star out of the game. he's a modest efficiency scorer. how the hell does he rate so highly?

Because offensive advanced stats are easy to correlate to box-scores (mainly scoring since you know efficient scoring win basketball games, also people with eyes can see that Harden and LeBron are good at scoring and passing), so we have a sense if any offensive advanced stats 'look' reasonable or not. Most tend to be reasonable.

We have no measure to quantify someone's defensive impact really so if someone were to calculate DRPM or DRAPM, there is nothing to benchmark this against other than subjective eye-test really. Therefore if we don't trust the estimate as much, we can subjectively shrink them towards the mean. The good defenders aren't as good and the bad ones aren't as bad, but of course shrinkage in this ridge context is NOT uniform. When you increase lambda, you are not subtracting/adding a constant to everyone's DRPM.

I think a lot of times when people are bent out of shape regarding someone’s defensive metrics they’re just having a different understanding of where offense ends and defense begins. In other words, the overall number that looks more at plus minus is more accurate than one that relies more on box score numbers and shrinks the gap between the best and worst ‘defenders,’ thereby compressing the overall spread.

The stuff players do on offense can have a direct effect on their team’s defense. Players like LeBron, Jokic, and Harden regularly post better defensive metrics than what someone relying on an eye test of mostly halfcourt defensive possessions would gather, and that’s likely because they’re excellent passers/decision-makers and are limiting opposing transition opportunities by consistently bending the defense and getting their teams high-percentage shots. On the other hand, players who aren’t as good at passing and have more of a gunner mentality will allow opponents to get out in transition more often.

I saw a twitter thread today where it said the Hawks have a 127 DRtg on possessions following a Trae Young missed three; to me it makes perfect sense that beyond how bad he might be at transition defense individually, his teammates will also be a step to two behind in getting back if they weren’t expecting him to pull up for a quick logo three (this largely applies to Zach Lavine as well). In that situation, I would be more comfortable saying Trae Young was chiefly responsible for the team giving up that basket than “defense is tricky so who’s to say whose fault it was?” The basket mostly resulted from the decision he made on the offensive end, so at the end of the day it doesn’t really matter to me whether it’s coming out of his offensive or defensive number as long as it’s by the same amount either way. By relying more on defensive box score numbers and shrinking the variance between the best and worst defenders, I think it’s allowing a player like Young to take most of the credit for how he improves a team’s offense while dividing up responsibility for how much easier he makes it for opponents to score on the Hawks.

[CptCrunch]

DRPM was the primary value of the stat. we can all pretty much gauge a player's offensive impact based on box score

and why is it ok to have extremes on offense but not on defense?

zach lavine has a significantly negative raw +/-. i.e. a team with a bad/depleted bench is doing better with it's quasi-star out of the game. he's a modest efficiency scorer. how the hell does he rate so highly?

Because offensive advanced stats are easy to correlate to box-scores (mainly scoring since you know efficient scoring win basketball games, also people with eyes can see that Harden and LeBron are good at scoring and passing), so we have a sense if any offensive advanced stats 'look' reasonable or not. Most tend to be reasonable.

We have no measure to quantify someone's defensive impact really so if someone were to calculate DRPM or DRAPM, there is nothing to benchmark this against other than subjective eye-test really. Therefore if we don't trust the estimate as much, we can subjectively shrink them towards the mean. The good defenders aren't as good and the bad ones aren't as bad, but of course shrinkage in this ridge context is NOT uniform. When you increase lambda, you are not subtracting/adding a constant to everyone's DRPM.

I think a lot of times when people are bent out of shape regarding someone’s defensive metrics they’re just having a different understanding of where offense ends and defense begins. In other words, the overall number that looks more at plus minus is more accurate than one that relies more on box score numbers and shrinks the gap between the best and worst ‘defenders,’ thereby compressing the overall spread.

The stuff players do on offense can have a direct effect on their team’s defense. Players like LeBron, Jokic, and Harden regularly post better defensive metrics than what someone relying on an eye test of mostly halfcourt defensive possessions would gather, and that’s likely because they’re excellent passers/decision-makers and are limiting opposing transition opportunities by consistently bending the defense and getting their teams high-percentage shots. On the other hand, players who aren’t as good at passing and have more of a gunner mentality will allow opponents to get out in transition more often.

I saw a twitter thread today where it said the Hawks have a 127 DRtg on possessions following a Trae Young missed three; to me it makes perfect sense that beyond how bad he might be at transition defense individually, his teammates will also be a step to two behind in getting back if they weren’t expecting him to pull up for a quick logo three (this largely applies to Zach Lavine as well). In that situation, I would be more comfortable saying Trae Young was chiefly responsible for the team giving up that basket than “defense is tricky so who’s to say whose fault it was?” The basket mostly resulted from the decision he made on the offensive end, so at the end of the day it doesn’t really matter to me whether it’s coming out of his offensive or defensive number as long as it’s by the same amount either way. By relying more on defensive box score numbers and shrinking the variance between the best and worst defenders, I think it’s allowing a player like Young to take most of the credit for how he improves a team’s offense while dividing up responsibility for how much easier he makes it for opponents to score on the Hawks.

You can pretty much come up with counter examples to your whole Trae Young defense theory.

Wesley Matthews and Donte DiVincenzo have godly DRPM for wing players. Are they actually stud defenders or are they benefitting from the Bucks' historic defensive efficiency?

The whole point of RAPM family statistics to decouple team effect form individual effects. Although the decoupling is not perfect, it is better than nothing.

You can't uniformly blame Trae for ruining his team defense while crediting Bucks players for their stout team defense.

[nolang1]

Because offensive advanced stats are easy to correlate to box-scores (mainly scoring since you know efficient scoring win basketball games, also people with eyes can see that Harden and LeBron are good at scoring and passing), so we have a sense if any offensive advanced stats 'look' reasonable or not. Most tend to be reasonable.

We have no measure to quantify someone's defensive impact really so if someone were to calculate DRPM or DRAPM, there is nothing to benchmark this against other than subjective eye-test really. Therefore if we don't trust the estimate as much, we can subjectively shrink them towards the mean. The good defenders aren't as good and the bad ones aren't as bad, but of course shrinkage in this ridge context is NOT uniform. When you increase lambda, you are not subtracting/adding a constant to everyone's DRPM.

I think a lot of times when people are bent out of shape regarding someone’s defensive metrics they’re just having a different understanding of where offense ends and defense begins. In other words, the overall number that looks more at plus minus is more accurate than one that relies more on box score numbers and shrinks the gap between the best and worst ‘defenders,’ thereby compressing the overall spread.

The stuff players do on offense can have a direct effect on their team’s defense. Players like LeBron, Jokic, and Harden regularly post better defensive metrics than what someone relying on an eye test of mostly halfcourt defensive possessions would gather, and that’s likely because they’re excellent passers/decision-makers and are limiting opposing transition opportunities by consistently bending the defense and getting their teams high-percentage shots. On the other hand, players who aren’t as good at passing and have more of a gunner mentality will allow opponents to get out in transition more often.

I saw a twitter thread today where it said the Hawks have a 127 DRtg on possessions following a Trae Young missed three; to me it makes perfect sense that beyond how bad he might be at transition defense individually, his teammates will also be a step to two behind in getting back if they weren’t expecting him to pull up for a quick logo three (this largely applies to Zach Lavine as well). In that situation, I would be more comfortable saying Trae Young was chiefly responsible for the team giving up that basket than “defense is tricky so who’s to say whose fault it was?” The basket mostly resulted from the decision he made on the offensive end, so at the end of the day it doesn’t really matter to me whether it’s coming out of his offensive or defensive number as long as it’s by the same amount either way. By relying more on defensive box score numbers and shrinking the variance between the best and worst defenders, I think it’s allowing a player like Young to take most of the credit for how he improves a team’s offense while dividing up responsibility for how much easier he makes it for opponents to score on the Hawks.

You can pretty much come up with counter examples to your whole Trae Young defense theory.

Wesley Matthews and Donte DiVincenzo have godly DRPM for wing players. Are they actually stud defenders or are they benefitting from the Bucks' historic defensive efficiency?

The whole point of RAPM family statistics to decouple team effect form individual effects. Although the decoupling is not perfect, it is better than nothing.

You can't uniformly blame Trae for ruining his team defense while crediting Bucks players for their stout team defense.

Sure you can in that you’re only as strong as your weakest link defensively. I’m sure Matthews and Divincenzo are benefitting from some good three-point variance when they’re playing without Giannis, but it also goes to my larger point that when you focus too much on the box score you run into the issue that individual field goal attempts, points, assists, turnovers, etc. can vary widely in how much they help or hurt the team. In the Trae Young example, the larger point was that certain shot attempts can pretty much amount to live-ball turnovers if they don’t go in. Another common example of this would be when a player drives to the basket and hits the deck while putting up an off-balance shot attempt or trying to draw contact; maybe such a shot is more likely to go in than a mid-range pull-up, but if it misses it’s much more likely to result in a fast break for the other team. The difference between giving the other team a fast break and not doing so is going to be big enough to overcome a number of good half-court defensive possessions.

edit: also Matthews and DiVincenzo are 5th and 8th among shooting guards (oops, actually 3rd and 4th now that it’s adjusted to the ‘better’ version with more weight on box score stats). Either way I wouldn’t exactly call that ‘godly’ given that the Bucks have the #1 defense in the league and, last I checked, don’t experience much of a drop-off (if any) in defensive rating when Giannis isn’t in the game. Obviously they have multiple players playing good defense for them to be where they are, and as I already mentioned they could be deriving some ‘defensive’ value from taking good shots within the flow of the offense and rarely turning the call over (Matthews is at the highest 3-point attempt rate of him is career and is averaging just 1 turnover per 36 minutes).

[grindtime22]

Its just another basketball reference stat now. Leaning on box scores for defense is silly.

[CptCrunch]

I think a lot of times when people are bent out of shape regarding someone’s defensive metrics they’re just having a different understanding of where offense ends and defense begins. In other words, the overall number that looks more at plus minus is more accurate than one that relies more on box score numbers and shrinks the gap between the best and worst ‘defenders,’ thereby compressing the overall spread.

The stuff players do on offense can have a direct effect on their team’s defense. Players like LeBron, Jokic, and Harden regularly post better defensive metrics than what someone relying on an eye test of mostly halfcourt defensive possessions would gather, and that’s likely because they’re excellent passers/decision-makers and are limiting opposing transition opportunities by consistently bending the defense and getting their teams high-percentage shots. On the other hand, players who aren’t as good at passing and have more of a gunner mentality will allow opponents to get out in transition more often.

I saw a twitter thread today where it said the Hawks have a 127 DRtg on possessions following a Trae Young missed three; to me it makes perfect sense that beyond how bad he might be at transition defense individually, his teammates will also be a step to two behind in getting back if they weren’t expecting him to pull up for a quick logo three (this largely applies to Zach Lavine as well). In that situation, I would be more comfortable saying Trae Young was chiefly responsible for the team giving up that basket than “defense is tricky so who’s to say whose fault it was?” The basket mostly resulted from the decision he made on the offensive end, so at the end of the day it doesn’t really matter to me whether it’s coming out of his offensive or defensive number as long as it’s by the same amount either way. By relying more on defensive box score numbers and shrinking the variance between the best and worst defenders, I think it’s allowing a player like Young to take most of the credit for how he improves a team’s offense while dividing up responsibility for how much easier he makes it for opponents to score on the Hawks.

You can pretty much come up with counter examples to your whole Trae Young defense theory.

Wesley Matthews and Donte DiVincenzo have godly DRPM for wing players. Are they actually stud defenders or are they benefitting from the Bucks' historic defensive efficiency?

The whole point of RAPM family statistics to decouple team effect form individual effects. Although the decoupling is not perfect, it is better than nothing.

You can't uniformly blame Trae for ruining his team defense while crediting Bucks players for their stout team defense.

Sure you can in that you’re only as strong as your weakest link defensively. I’m sure Matthews and Divincenzo are benefitting from some good three-point variance when they’re playing without Giannis, but it also goes to my larger point that when you focus too much on the box score you run into the issue that individual field goal attempts, points, assists, turnovers, etc. can vary widely in how much they help or hurt the team. In the Trae Young example, the larger point was that certain shot attempts can pretty much amount to live-ball turnovers if they don’t go in. Another common example of this would be when a player drives to the basket and hits the deck while putting up an off-balance shot attempt or trying to draw contact; maybe such a shot is more likely to go in than a mid-range pull-up, but if it misses it’s much more likely to result in a fast break for the other team. The difference between giving the other team a fast break and not doing so is going to be big enough to overcome a number of good half-court defensive possessions.

edit: also Matthews and DiVincenzo are 5th and 8th among shooting guards (oops, actually 3rd and 4th now that it’s adjusted to the ‘better’ version with more weight on box score stats). Either way I wouldn’t exactly call that ‘godly’ given that the Bucks have the #1 defense in the league and, last I checked, don’t experience much of a drop-off (if any) in defensive rating when Giannis isn’t in the game. Obviously they have multiple players playing good defense for them to be where they are, and as I already mentioned they could be deriving some ‘defensive’ value from taking good shots within the flow of the offense and rarely turning the call over (Matthews is at the highest 3-point attempt rate of him is career and is averaging just 1 turnover per 36 minutes).

Your points are good, but there is no way to systematically adjust for what you are describing. My point is that you cannot discount Trae's RPM stats just because you have some interesting possibly factually driven observations. If you want to adjust, you need to do it in a way that all players' stats are adjusted for in certain ways. We shouldn't be using narrative to single out specific player and their stats based on largely on 'feels' with no data-driven evidence or model.

3/4th are pretty godly considering that there are ~450 NBA players, ~300 rotations players in the league.

[nolang1]

You can pretty much come up with counter examples to your whole Trae Young defense theory.

Wesley Matthews and Donte DiVincenzo have godly DRPM for wing players. Are they actually stud defenders or are they benefitting from the Bucks' historic defensive efficiency?

The whole point of RAPM family statistics to decouple team effect form individual effects. Although the decoupling is not perfect, it is better than nothing.

You can't uniformly blame Trae for ruining his team defense while crediting Bucks players for their stout team defense.

Sure you can in that you’re only as strong as your weakest link defensively. I’m sure Matthews and Divincenzo are benefitting from some good three-point variance when they’re playing without Giannis, but it also goes to my larger point that when you focus too much on the box score you run into the issue that individual field goal attempts, points, assists, turnovers, etc. can vary widely in how much they help or hurt the team. In the Trae Young example, the larger point was that certain shot attempts can pretty much amount to live-ball turnovers if they don’t go in. Another common example of this would be when a player drives to the basket and hits the deck while putting up an off-balance shot attempt or trying to draw contact; maybe such a shot is more likely to go in than a mid-range pull-up, but if it misses it’s much more likely to result in a fast break for the other team. The difference between giving the other team a fast break and not doing so is going to be big enough to overcome a number of good half-court defensive possessions.

edit: also Matthews and DiVincenzo are 5th and 8th among shooting guards (oops, actually 3rd and 4th now that it’s adjusted to the ‘better’ version with more weight on box score stats). Either way I wouldn’t exactly call that ‘godly’ given that the Bucks have the #1 defense in the league and, last I checked, don’t experience much of a drop-off (if any) in defensive rating when Giannis isn’t in the game. Obviously they have multiple players playing good defense for them to be where they are, and as I already mentioned they could be deriving some ‘defensive’ value from taking good shots within the flow of the offense and rarely turning the call over (Matthews is at the highest 3-point attempt rate of him is career and is averaging just 1 turnover per 36 minutes).

Your points are good, but there is no way to systematically adjust for what you are describing. My point is that you cannot discount Trae's RPM stats just because you have some interesting possibly factually driven observations. If you want to adjust, you need to do it in a way that all players' stats are adjusted for in certain ways. We shouldn't be using narrative to single out specific player and their stats based on largely on 'feels' with no data-driven evidence or model.

3/4th are pretty godly considering that there are ~450 NBA players, ~300 rotations players in the league.

My point is that you can’t cleanly separate offense from defense, so I’d rather look at a player’s overall RPM/RAPM than try to nitpick the defensive component too much based on eye test. And when doing that, I think there is value in a metric looking more at the score of the game than trying to be the umpteenth different linear weighting of box score numbers when evaluating defense based on box score numbers is always going to be garbage in, garbage out to some extent. There is no box score number that says anything about transition defense, and I don’t have to single out any particular players to say that shot selection or the propensity to make certain types of turnovers is going to have an impact on their team’s defense before the defensive possession technically starts.

[GeorgeMarcus]

Read on Twitter

?s=19

I hated the new version incorporating tracking data. Well now ESPN lost even more credibility to me by just deciding to change it entirely mid season. It's way different.

Explains a lot. Ugh...