Weird Kawhi stat - why it's fallacious to accept advanced stats without context

[jpotter33]

http://www.cbssports.com/nba/news/kawhi-leonard-is-so-great-at-defense-hes-actually-hurting-the-spurs/

Interesting article. Kawhi Leonard's on/off defensive rating actually has him as a terrible defender, but the context of the situation paints an entirely different story due to offenses essentially taking him out of the defense and forcing his lesser teammates to play D. He's actually so good defensively that it has created this effect that hurts him statistically in that regard.

I think many people take advanced stats a little too seriously as the ultimate determinant of a player's worth/skill/productivity, regardless of the context or what the eye test says. But examples like this show that this isn't rocket science, and these aren't things that are 100% reflective of a player's skill set or productivity and shouldn't be taken as such.

PS - Bradley supposedly has the worst on/off defensive rating in the league. Not quite the same situation as Kawhi by any means, because he certainly has taken a step back this year. However, I'd like to know how much of that stat is contextually-based.

[tankcity!]

Bradley has been historically bad on defense if you look at his advanced metrics. This should be well known already that it's hard to determine the value of defense using stats. And yet people keep throwing the plus minus stat, and the def. rating when player is on the court stat when talking about Smart. Pick a side. Use stats, or don't.

[jpotter33]

Bradley has been historically bad on defense if you look at his advanced metrics. This should be well known already that it's hard to determine the value of defense using stats. And yet people keep throwing the plus minus stat, and the def. rating when player is on the court stat when talking about Smart. Pick a side. Use stats, or don't.

I mean, you were the one several weeks back doing just that with DRPM. You were using it as an argument that Smart isn't that great defensively. Of course, you're not using it anymore because he's now second in the league for point guards behind only Paul in that metric.

Anyways, the point I was trying to make is that you can't simply take these advanced stats at face value as ultimate truth, certainly for defensive statistics. They are certainly helpful at times, but it's clearly fallacious to not take the eye test and context into consideration when looking at these things. The fact that you think it's an either/or thing is pretty much what I'm talking about.

[tankcity!]

Bradley has been historically bad on defense if you look at his advanced metrics. This should be well known already that it's hard to determine the value of defense using stats. And yet people keep throwing the plus minus stat, and the def. rating when player is on the court stat when talking about Smart. Pick a side. Use stats, or don't.

I mean, you were the one several weeks back doing just that with DRPM. You were using it as an argument that Smart isn't that great defensively. Of course, you're not using it anymore because he's now second in the league for point guards behind only Paul in that metric.

Anyways, the point I was trying to make is that you can't simply take these advanced stats at face value as ultimate truth, certainly for defensive statistics. They are certainly helpful at times, but it's clearly fallacious to not take the eye test and context into consideration when looking at these things. The fact that you think it's an either/or thing is pretty much what I'm talking about.

I only used it because you have used it in the past. I was only using your argument against you. That was my point at the time. You aren't fair in your evaluation of players. The other night you want us to bench Bradley? That doesn't make any sense. Smart and Bradley guarded Westbrook. He played well the entire game. It's both their fault. Same with Harden.

[tarheelsxxiii]

Good points.  Thanks for sharing, Professor Potter.

[jpotter33]

Bradley has been historically bad on defense if you look at his advanced metrics. This should be well known already that it's hard to determine the value of defense using stats. And yet people keep throwing the plus minus stat, and the def. rating when player is on the court stat when talking about Smart. Pick a side. Use stats, or don't.

I mean, you were the one several weeks back doing just that with DRPM. You were using it as an argument that Smart isn't that great defensively. Of course, you're not using it anymore because he's now second in the league for point guards behind only Paul in that metric.

Anyways, the point I was trying to make is that you can't simply take these advanced stats at face value as ultimate truth, certainly for defensive statistics. They are certainly helpful at times, but it's clearly fallacious to not take the eye test and context into consideration when looking at these things. The fact that you think it's an either/or thing is pretty much what I'm talking about.

I only used it because you have used it in the past. I was only using your argument against you. That was my point at the time. You aren't fair in your evaluation of players. The other night you want us to bench Bradley? That doesn't make any sense. Smart and Bradley guarded Westbrook. He played well the entire game. It's both their fault. Same with Harden.

Once again, I have no idea what you're talking about. When is this "past" that you talk about? I've never been a big user of advanced statistics, especially defensively, because I've long held that they can't properly isolate a defender's true impact.

And regarding Bradley, if I ever wanted to bench Bradley in OKC it was because of his stupid live turnovers and his forced attempts. He did not "play well the entire game," and that was clear to see. He was an off-guard that had 4 turnovers, and he had about 5 forced possessions at the rim that also cost us, including one down the stretch that was crucial. He was a -12 for a reason.

Bradley is a fine defender, but he's not very good on bigger guards like Westbrook, which is what all of us on the thread were saying. Westbrook will get his, but at least Smart has the size and strength to challenge him and make it a difficult, which he did for most of the game. When they put the smaller defenders of Bradley and Rozier on him, he either shot right over them or bullied them down to a favorable position on the block.

Once again, it's these kinds of contextual factors and the actual eye test that you don't take into consideration in your analysis.

[kozlodoev]

http://www.cbssports.com/nba/news/kawhi-leonard-is-so-great-at-defense-hes-actually-hurting-the-spurs/

Interesting article. Kawhi Leonard's on/off defensive rating actually has him as a terrible defender, but the context of the situation paints an entirely different story due to offenses essentially taking him out of the defense and forcing his lesser teammates to play D. He's actually so good defensively that it has created this effect that hurts him statistically in that regard.

I think many people take advanced stats a little too seriously as the ultimate determinant of a player's worth/skill/productivity, regardless of the context or what the eye test says. But examples like this show that this isn't rocket science, and these aren't things that are 100% reflective of a player's skill set or productivity and shouldn't be taken as such.

PS - Bradley supposedly has the worst on/off defensive rating in the league. Not quite the same situation as Kawhi by any means, because he certainly has taken a step back this year. However, I'd like to know how much of that stat is contextually-based.

I confess I haven't read the article, but if offenses take him out of the defense, every advanced stat will show that he's barely being scored against (think opp. PER and the likes). Sure, on/off and the likes will be a horrible indicator but that's not supposed to be news, or is it?

[jambr380]

Did you just say fellatious?



   But, on the subject at hand, that is really interesting about Leonard - tp. One would just assume he would lead all metrics. I guess this is promising news for Bradley.

[Moranis]

http://www.cbssports.com/nba/news/kawhi-leonard-is-so-great-at-defense-hes-actually-hurting-the-spurs/

Interesting article. Kawhi Leonard's on/off defensive rating actually has him as a terrible defender, but the context of the situation paints an entirely different story due to offenses essentially taking him out of the defense and forcing his lesser teammates to play D. He's actually so good defensively that it has created this effect that hurts him statistically in that regard.

I think many people take advanced stats a little too seriously as the ultimate determinant of a player's worth/skill/productivity, regardless of the context or what the eye test says. But examples like this show that this isn't rocket science, and these aren't things that are 100% reflective of a player's skill set or productivity and shouldn't be taken as such.

PS - Bradley supposedly has the worst on/off defensive rating in the league. Not quite the same situation as Kawhi by any means, because he certainly has taken a step back this year. However, I'd like to know how much of that stat is contextually-based.

I confess I haven't read the article, but if offenses take him out of the defense, every advanced stat will show that he's barely being scored against (think opp. PER and the likes). Sure, on/off and the likes will be a horrible indicator but that's not supposed to be news, or is it?

that is basically the gist of the article.  they used the Chicago game as an example.  Basically Butler and Wade when guarded by Kawhi just stood way out of the play and let the rest of the team play 4 on 4 so the rest of the team tore up the crappier Spurs defenders while Butler or Wade did almost nothing when guarded by Kawhi.

[tankcity!]

Bradley has been historically bad on defense if you look at his advanced metrics. This should be well known already that it's hard to determine the value of defense using stats. And yet people keep throwing the plus minus stat, and the def. rating when player is on the court stat when talking about Smart. Pick a side. Use stats, or don't.

I mean, you were the one several weeks back doing just that with DRPM. You were using it as an argument that Smart isn't that great defensively. Of course, you're not using it anymore because he's now second in the league for point guards behind only Paul in that metric.

Anyways, the point I was trying to make is that you can't simply take these advanced stats at face value as ultimate truth, certainly for defensive statistics. They are certainly helpful at times, but it's clearly fallacious to not take the eye test and context into consideration when looking at these things. The fact that you think it's an either/or thing is pretty much what I'm talking about.

I only used it because you have used it in the past. I was only using your argument against you. That was my point at the time. You aren't fair in your evaluation of players. The other night you want us to bench Bradley? That doesn't make any sense. Smart and Bradley guarded Westbrook. He played well the entire game. It's both their fault. Same with Harden.

Once again, I have no idea what you're talking about. When is this "past" that you talk about? I've never been a big user of advanced statistics, especially defensively, because I've long held that they can't properly isolate a defender's true impact.

And regarding Bradley, if I ever wanted to bench Bradley in OKC it was because of his stupid live turnovers and his forced attempts. He did not "play well the entire game," and that was clear to see. He was an off-guard that had 4 turnovers, and he had about 5 forced possessions at the rim that also cost us, including one down the stretch that was crucial. He was a -12 for a reason.

Bradley is a fine defender, but he's not very good on bigger guards like Westbrook, which is what all of us on the thread were saying. Westbrook will get his, but at least Smart has the size and strength to challenge him and make it a difficult, which he did for most of the game. When they put the smaller defenders of Bradley and Rozier on him, he either shot right over them or bullied them down to a favorable position on the block.

Once again, it's these kinds of contextual factors and the actual eye test that you don't take into consideration in your analysis.

You sound like a politician

[jpotter33]

http://www.cbssports.com/nba/news/kawhi-leonard-is-so-great-at-defense-hes-actually-hurting-the-spurs/

Interesting article. Kawhi Leonard's on/off defensive rating actually has him as a terrible defender, but the context of the situation paints an entirely different story due to offenses essentially taking him out of the defense and forcing his lesser teammates to play D. He's actually so good defensively that it has created this effect that hurts him statistically in that regard.

I think many people take advanced stats a little too seriously as the ultimate determinant of a player's worth/skill/productivity, regardless of the context or what the eye test says. But examples like this show that this isn't rocket science, and these aren't things that are 100% reflective of a player's skill set or productivity and shouldn't be taken as such.

PS - Bradley supposedly has the worst on/off defensive rating in the league. Not quite the same situation as Kawhi by any means, because he certainly has taken a step back this year. However, I'd like to know how much of that stat is contextually-based.

I confess I haven't read the article, but if offenses take him out of the defense, every advanced stat will show that he's barely being scored against (think opp. PER and the likes). Sure, on/off and the likes will be a horrible indicator but that's not supposed to be news, or is it?

that is basically the gist of the article.  they used the Chicago game as an example.  Basically Butler and Wade when guarded by Kawhi just stood way out of the play and let the rest of the team play 4 on 4 so the rest of the team tore up the crappier Spurs defenders while Butler or Wade did almost nothing when guarded by Kawhi.

What's funny is that this is basically what happened in The Waterboy to Bobby Boucher in the Bourbon Bowl game lol The other team stopped playing offense so that his defense couldn't hurt them. (Only a semi-serious analogy, by the way.)

[mmmmm]

The reason Bradley's 'advanced stats' (really, just anything based on plus-minus data) is so bad this year is the same reason why Isaiah's isn't massively positive for the first time in his Celtics career.

Basically it is a classic small-sample size issue that is skewed by some outliers.

Both Avery and Isaiah were on the floor for moments of dramatic runs in the GSW, DEN and WAS games that were played without Al Horford or Jae Crowder.  The DEN game also was played without Kelly Olynyk.

Those three games account for a total plus/minus impact of -59 for Bradley and -58 for Isaiah.

So just those three games still account for a -3.5 impact on Bradley's Net Rating (points/100 possessions) for this season so far (and -4.1 on Isaiah's).   That's a heavy cost considering those three games amount to a sample size of just 211 possessions for Avery and just 197 for Isaiah.

In Avery's larger sample of 1471 possessions played in all the other games, his plus/minus net rating is +1.8.   

In Isaiah's larger sample of 1210 possessions played in all other games, his plus/minus net rating is +5.3.

Now, I don't want to reverse-cherry pick here and just toss the data from the three mentioned games in the toilet and just use the rest of the season.

Instead, the right thing to do is recognize that a 24 game sample of plus-minus data is basically worth very, very little. 

Well, I'd say more exactly what it is worth, but we keep the language clean here.

We need, at a minimum, a full season to fully mitigate these outliers as well as positive outliers that may happen during the season (such as if we were to blow out some other team playing shorthanded).   Whether even a full season is really enough statistics to support drawing conclusions based on plus/minus data (or it's derivatives) is, of course, a larger debate in itself that I won't go into here.