Cleveland/Denver

[bgrep14]

I am going by CTG which removes garbage time. They have Okoro's overall on/off at -1.9 and his overall on number at +1.5.
Okoro is an undersized wing who doesn't have great defensive instincts or floor awareness. He's not incredibly athletic or strong, either. Lowkey one of the worst draft picks in the 2020s.

What is CTG? Okoro doesn't play in garbage time.

Cleaning the Glass.
Whether he plays in garbage time or not, including garbage time in on/off will influence the final results since off matters as much as on.

Okoro defensive metrics are likely skewed by having to play the other teams best offensive weapon every night compared to other plays who operate off primary ball handlers. The metrics you shared just seem like a relatively lazy consumption of data to aggregate a median of plus/minus performance. Okoro is great at guarding teams point guard and shooting guards but pretty awful at defending larger wings. Okoros best position is shooting guard which the Cavs need to move one or two for SF. I think Okoro in Denver or LAL would be a slam dunk for their teams.

[BK_2020]

What is CTG? Okoro doesn't play in garbage time.

Cleaning the Glass.
Whether he plays in garbage time or not, including garbage time in on/off will influence the final results since off matters as much as on.

Okoro defensive metrics are likely skewed by having to play the other teams best offensive weapon every night compared to other plays who operate off primary ball handlers. The metrics you shared just seem like a relatively lazy consumption of data to aggregate a median of plus/minus performance. Okoro is great at guarding teams point guard and shooting guards but pretty awful at defending larger wings. Okoros best position is shooting guard which the Cavs need to move one or two for SF. I think Okoro in Denver or LAL would be a slam dunk for their teams.

How the heck is on/off a median of plus/minus performance? It's literally what it says it is. On/off is not Okoro's metrics but team metrics when Okoro plays. If Okoro is great at guarding opposing team's best players, then the Cavs should be worse on defense when Okoro is not on the floor vs when he is. Evidence suggests that that is not the case. For the most part, the Cavs are better on defense when Okoro sits, although not to a degree where you can conclude that Okoro is a bad defender. On/off for a small sample may not tell you much, but we are talking about 4 seasons' worth of on/off data where Okoro's presence makes no positive difference in the Cavs' defensive performance, in fact makes it worse by a decent amount.
And your own supposed explanation for what makes Okoro so great already explains why he is a terrible defensive piece. If you have your supposed elite defensive wing guard the opposing team's guards, then your PG or SG is guarding the opposing team's wing. Basically Okoro's best attribute (according to you) forces the defense to concede the switch that offenses normally have to work to get. Why would anyone want that?

[bgrep14]

Cleaning the Glass.
Whether he plays in garbage time or not, including garbage time in on/off will influence the final results since off matters as much as on.

Okoro defensive metrics are likely skewed by having to play the other teams best offensive weapon every night compared to other plays who operate off primary ball handlers. The metrics you shared just seem like a relatively lazy consumption of data to aggregate a median of plus/minus performance. Okoro is great at guarding teams point guard and shooting guards but pretty awful at defending larger wings. Okoros best position is shooting guard which the Cavs need to move one or two for SF. I think Okoro in Denver or LAL would be a slam dunk for their teams.

How the heck is on/off a median of plus/minus performance? It's literally what it says it is. On/off is not Okoro's metrics but team metrics when Okoro plays. If Okoro is great at guarding opposing team's best players, then the Cavs should be worse on defense when Okoro is not on the floor vs when he is. Evidence suggests that that is not the case. For the most part, the Cavs are better on defense when Okoro sits, although not to a degree where you can conclude that Okoro is a bad defender. On/off for a small sample may not tell you much, but we are talking about 4 seasons' worth of on/off data where Okoro's presence makes no positive difference in the Cavs' defensive performance, in fact makes it worse by a decent amount.
And your own supposed explanation for what makes Okoro so great already explains why he is a terrible defensive piece. If you have your supposed elite defensive wing guard the opposing team's guards, then your PG or SG is guarding the opposing team's wing. Basically Okoro's best attribute (according to you) forces the defense to concede the switch that offenses normally have to work to get. Why would anyone want that?

The metrics you're using are team based +/- metric that are not individually advanced enough to determine whether Okoro is individually a good or bad defender. If Okoro plays 40 minutes with George Niang and Dean Wade at the 4/5 instead of Mobley and Allen there's probably going to be significant defensive variances for any 1-3 position player.

[BK_2020]

Okoro defensive metrics are likely skewed by having to play the other teams best offensive weapon every night compared to other plays who operate off primary ball handlers. The metrics you shared just seem like a relatively lazy consumption of data to aggregate a median of plus/minus performance. Okoro is great at guarding teams point guard and shooting guards but pretty awful at defending larger wings. Okoros best position is shooting guard which the Cavs need to move one or two for SF. I think Okoro in Denver or LAL would be a slam dunk for their teams.

How the heck is on/off a median of plus/minus performance? It's literally what it says it is. On/off is not Okoro's metrics but team metrics when Okoro plays. If Okoro is great at guarding opposing team's best players, then the Cavs should be worse on defense when Okoro is not on the floor vs when he is. Evidence suggests that that is not the case. For the most part, the Cavs are better on defense when Okoro sits, although not to a degree where you can conclude that Okoro is a bad defender. On/off for a small sample may not tell you much, but we are talking about 4 seasons' worth of on/off data where Okoro's presence makes no positive difference in the Cavs' defensive performance, in fact makes it worse by a decent amount.
And your own supposed explanation for what makes Okoro so great already explains why he is a terrible defensive piece. If you have your supposed elite defensive wing guard the opposing team's guards, then your PG or SG is guarding the opposing team's wing. Basically Okoro's best attribute (according to you) forces the defense to concede the switch that offenses normally have to work to get. Why would anyone want that?

The metrics you're using are team based +/- metric that are not individually advanced enough to determine whether Okoro is individually a good or bad defender. If Okoro plays 40 minutes with George Niang and Dean Wade at the 4/5 instead of Mobley and Allen there's probably going to be significant defensive variances for any 1-3 position player.

I don't know how I can explain this to you in an easier way but if playing with Niang and Wade at 4/5 will tank the team's defense, that's not going to influence Okoro's on/off as much as you think because Okoro is being compared to someone else while Niang and Wade are playing at 4/5.
On/off tends to be fairly stable over a long period. Go look at Caruso, Lonzo, Jrue Draymond, Herb Jones, Gobert, or any number of players who are fairly clearly great defenders. You might see an outlier season here and there but you won't see a random fluctuation over the course of their careers.

[toooskies]

What is CTG? Okoro doesn't play in garbage time.

Cleaning the Glass.
Whether he plays in garbage time or not, including garbage time in on/off will influence the final results since off matters as much as on.

I don't know, but more and more I'm all the way out and those types of *adjustments* or extrapolations. If you have a sufficient sample size in terms of games and minutes, use the straight +/- and on/off numbers to determine whether a player positively impacts the game.

Otherwise you end up with a considerable delta between the adjusted data and actual data with little reason to believe the adjusted data more accurately predicts impact. If the Cavs third string pummels the other team's third string with the game was out of reach, so what? It doesn't diminish what Okoro did with the game on the line against better players.

Everyone's new favorite on this board is EPM. EPM is an estimate based on BPM - a projection itself. Assuming all the *adjustments* are properly made, and you have to assume as the veracity of the data input cannot be independently verified, at some point you have to ask whether the model itself is flawed.

In any event, if you're using adjusted numbers, they should be presented as such.

Uhh, not exactly. It first constructs a historical RAPM-trained model which estimates performance in lineup data. That model is then correlated to box score stats (although NOT strictly BPM) to produce estimates of what a player's RAPM should look like with given box scores. It then combines the RAPM of the current season (which is noisy in samples as small as full seasons) with their box score-based RAPM projections to land at the final number.

RAPM is pretty much on/off data but adjusting for the strength of your team's lineup and the opposition's. EPM is the new hotness because it makes that noisy-but-good concept of RAPM into something that can be used game-to-game in a season.

... But some of the outputs are still pretty wrong, so it's bad to take it as gospel. It's just one more stat to bucket with all the other stats. (Come up with a list of Cleveland's roster from best-to-worst at defense, for instance, then go look them up in EPM. The lists will not be close.)

[jbk1234]

Cleaning the Glass.
Whether he plays in garbage time or not, including garbage time in on/off will influence the final results since off matters as much as on.

I don't know, but more and more I'm all the way out and those types of *adjustments* or extrapolations. If you have a sufficient sample size in terms of games and minutes, use the straight +/- and on/off numbers to determine whether a player positively impacts the game.

Otherwise you end up with a considerable delta between the adjusted data and actual data with little reason to believe the adjusted data more accurately predicts impact. If the Cavs third string pummels the other team's third string with the game was out of reach, so what? It doesn't diminish what Okoro did with the game on the line against better players.

Everyone's new favorite on this board is EPM. EPM is an estimate based on BPM - a projection itself. Assuming all the *adjustments* are properly made, and you have to assume as the veracity of the data input cannot be independently verified, at some point you have to ask whether the model itself is flawed.

In any event, if you're using adjusted numbers, they should be presented as such.

Uhh, not exactly. It first constructs a historical RAPM-trained model which estimates performance in lineup data. That model is then correlated to box score stats (although NOT strictly BPM) to produce estimates of what a player's RAPM should look like with given box scores. It then combines the RAPM of the current season (which is noisy in samples as small as full seasons) with their box score-based RAPM projections to land at the final number.

RAPM is pretty much on/off data but adjusting for the strength of your team's lineup and the opposition's. EPM is the new hotness because it makes that noisy-but-good concept of RAPM into something that can be used game-to-game in a season.

... But some of the outputs are still pretty wrong, so it's bad to take it as gospel. It's just one more stat to bucket with all the other stats. (Come up with a list of Cleveland's roster from best-to-worst at defense, for instance, then go look them up in EPM. The lists will not be close.)

My thing is they don't show their work. There's plenty of data with a starter, or key reserve, who plays 70 games. That data will include performances against good and bad teams. Now if they're parsing between performances against playoff-caliber teams and the bottom ten teams, and there's a significant delta, that's extremely useful information. But, they don't show their splits so you have no idea if the adjustments make sense.

If it's the extrapolations, or estimates, or adjustments that are leading to the delta between actual impact and projected performance, I'm at a loss to understand why people give more credence to the projections.

[toooskies]

I don't know, but more and more I'm all the way out and those types of *adjustments* or extrapolations. If you have a sufficient sample size in terms of games and minutes, use the straight +/- and on/off numbers to determine whether a player positively impacts the game.

Otherwise you end up with a considerable delta between the adjusted data and actual data with little reason to believe the adjusted data more accurately predicts impact. If the Cavs third string pummels the other team's third string with the game was out of reach, so what? It doesn't diminish what Okoro did with the game on the line against better players.

Everyone's new favorite on this board is EPM. EPM is an estimate based on BPM - a projection itself. Assuming all the *adjustments* are properly made, and you have to assume as the veracity of the data input cannot be independently verified, at some point you have to ask whether the model itself is flawed.

In any event, if you're using adjusted numbers, they should be presented as such.

Uhh, not exactly. It first constructs a historical RAPM-trained model which estimates performance in lineup data. That model is then correlated to box score stats (although NOT strictly BPM) to produce estimates of what a player's RAPM should look like with given box scores. It then combines the RAPM of the current season (which is noisy in samples as small as full seasons) with their box score-based RAPM projections to land at the final number.

RAPM is pretty much on/off data but adjusting for the strength of your team's lineup and the opposition's. EPM is the new hotness because it makes that noisy-but-good concept of RAPM into something that can be used game-to-game in a season.

... But some of the outputs are still pretty wrong, so it's bad to take it as gospel. It's just one more stat to bucket with all the other stats. (Come up with a list of Cleveland's roster from best-to-worst at defense, for instance, then go look them up in EPM. The lists will not be close.)

My thing is they don't show their work. There's plenty of data with a starter, or key reserve, who plays 70 games. That data will include performances against good and bad teams. Now if they're parsing between performances against playoff-caliber teams and the bottom ten teams, and there's a significant delta, that's extremely useful information. But, they don't show their splits so you have no idea if the adjustments make sense.

If it's the extrapolations, or estimates, or adjustments that are leading to the delta between actual impact and projected performance, I'm at a loss to understand why people give more credence to the projections.

I mean, in the case of EPM it's all lineup data. In the case of Cleaning the Glass, it's traceable game-to-game about what counts as garbage time. (I'm not sure exactly how because I don't have a sub.)

But even then, lineup data can be pretty much worthless with enough confounding influences. There isn't a Cavs sample size that wasn't affected by one player or another with a significant injury, and the Cavs' season schedule difficulty had a clear weak middle that corresponded with Garland and Mobley's injuries that flips everything on its head.

[jbk1234]

Uhh, not exactly. It first constructs a historical RAPM-trained model which estimates performance in lineup data. That model is then correlated to box score stats (although NOT strictly BPM) to produce estimates of what a player's RAPM should look like with given box scores. It then combines the RAPM of the current season (which is noisy in samples as small as full seasons) with their box score-based RAPM projections to land at the final number.

RAPM is pretty much on/off data but adjusting for the strength of your team's lineup and the opposition's. EPM is the new hotness because it makes that noisy-but-good concept of RAPM into something that can be used game-to-game in a season.

... But some of the outputs are still pretty wrong, so it's bad to take it as gospel. It's just one more stat to bucket with all the other stats. (Come up with a list of Cleveland's roster from best-to-worst at defense, for instance, then go look them up in EPM. The lists will not be close.)

My thing is they don't show their work. There's plenty of data with a starter, or key reserve, who plays 70 games. That data will include performances against good and bad teams. Now if they're parsing between performances against playoff-caliber teams and the bottom ten teams, and there's a significant delta, that's extremely useful information. But, they don't show their splits so you have no idea if the adjustments make sense.

If it's the extrapolations, or estimates, or adjustments that are leading to the delta between actual impact and projected performance, I'm at a loss to understand why people give more credence to the projections.

I mean, in the case of EPM it's all lineup data. In the case of Cleaning the Glass, it's traceable game-to-game about what counts as garbage time. (I'm not sure exactly how because I don't have a sub.)

But even then, lineup data can be pretty much worthless with enough confounding influences. There isn't a Cavs sample size that wasn't affected by one player or another with a significant injury, and the Cavs' season schedule difficulty had a clear weak middle that corresponded with Garland and Mobley's injuries that flips everything on its head.

That's what I'm saying though. They say they're adjusting for strength of opponent and lineups, but their end results don't arrive at where youd expect them to arrive. Garland misses the easiest part of our schedule and comes back for the hardest part with all of Mitchell, Strus, Wade and Mobley out. You'd expect his adjusted, extrapolated, or projected numbers to better than his actual on/off or +/-. Instead just the opposite occurred. If you're spitting out results like that, and Garland is hardly alone, show your work so that people can make sense of it.

As far as CTG, they should at least provide the objective criteria for what constitutes *garbage time.* That way people can verify they're doing what they say they're doing.

[toooskies]

My thing is they don't show their work. There's plenty of data with a starter, or key reserve, who plays 70 games. That data will include performances against good and bad teams. Now if they're parsing between performances against playoff-caliber teams and the bottom ten teams, and there's a significant delta, that's extremely useful information. But, they don't show their splits so you have no idea if the adjustments make sense.

If it's the extrapolations, or estimates, or adjustments that are leading to the delta between actual impact and projected performance, I'm at a loss to understand why people give more credence to the projections.

I mean, in the case of EPM it's all lineup data. In the case of Cleaning the Glass, it's traceable game-to-game about what counts as garbage time. (I'm not sure exactly how because I don't have a sub.)

But even then, lineup data can be pretty much worthless with enough confounding influences. There isn't a Cavs sample size that wasn't affected by one player or another with a significant injury, and the Cavs' season schedule difficulty had a clear weak middle that corresponded with Garland and Mobley's injuries that flips everything on its head.

That's what I'm saying though. They say they're adjusting for strength of opponent and lineups, but their end results don't arrive at where youd expect them to arrive. Garland misses the easiest part of our schedule and comes back for the hardest part with all of Mitchell, Strus, Wade and Mobley out. You'd expect his adjusted, extrapolated, or projected numbers to better than his actual on/off or +/-. Instead just the opposite occurred. If you're spitting out results like that, and Garland is hardly alone, show your work so that people can make sense of it.

As far as CTG, they should at least provide the objective criteria for what constitutes *garbage time.* That way people can verify they're doing what they say they're doing.

Well, maybe your assumptions are wrong. For instance: you assume Garland missed time mostly against easy opponents. In truth, of the 25 games he missed, 12 were against playoff teams.

[jbk1234]

I mean, in the case of EPM it's all lineup data. In the case of Cleaning the Glass, it's traceable game-to-game about what counts as garbage time. (I'm not sure exactly how because I don't have a sub.)

But even then, lineup data can be pretty much worthless with enough confounding influences. There isn't a Cavs sample size that wasn't affected by one player or another with a significant injury, and the Cavs' season schedule difficulty had a clear weak middle that corresponded with Garland and Mobley's injuries that flips everything on its head.

That's what I'm saying though. They say they're adjusting for strength of opponent and lineups, but their end results don't arrive at where youd expect them to arrive. Garland misses the easiest part of our schedule and comes back for the hardest part with all of Mitchell, Strus, Wade and Mobley out. You'd expect his adjusted, extrapolated, or projected numbers to better than his actual on/off or +/-. Instead just the opposite occurred. If you're spitting out results like that, and Garland is hardly alone, show your work so that people can make sense of it.

As far as CTG, they should at least provide the objective criteria for what constitutes *garbage time.* That way people can verify they're doing what they say they're doing.

Well, maybe your assumptions are wrong. For instance: you assume Garland missed time mostly against easy opponents. In truth, of the 25 games he missed, 12 were against playoff teams.

And the Cavs lost half those games. Also, 13 out of 25 is most of the games he missed. Also, nothing I said about the Cavs being down three starters and Wade in March is inaccurate and we played 12 playoff teams in March versus 6 lottery teams. If that's what you're supposed to be adjusting for, it's not adding up.