Steve Blake vs. Bayless as of late...

[maxiep]

Nowhere does he claim the numbers must be wrong...I must be missing something

I come from a financial background and I'm used to financial metrics that don't really tell you anything by themselves. A lot of times they just raise a flag that something unusual might be happening when a number falls outside the norm.

There is so much static in these +/- numbers that I don't believe they say anything of substance about a player on their own...maybe the more advanced adjusted +/- do, I have no idea.

Similar to finance, when you see a weird outlier it can be useful to dig in and understand the underlying causes...but oftentimes it is unrelated to the way the original metric is pointed...FOR EXAMPLE -

Howard's +/- was +12 for the game.

If you take that number at it's face...wow Howard rules! He won the game for us...err...

But if you dig in, you realize he was on the court when Bayless made steals and started getting uber-aggresive and had a spurt of points while the Knicks were't scoring.

What I think the +/- for Howard ends up saying - wow, Bayless was really on fire for a while in the 4th quarter. ...someone could make the argument it was somehow due to Juwan's play, but eyeballing the game he got a couple rebounds but it was mostly Bayless just taking on all comers for a while...nothing to do with Juwan's play.

The concept you're trying to explain is collinearity.

 

[Blazer_Hippie]

The concept you're trying to explain is collinearity.

ack...big word!!! But from a brief wiki I don't think so...I don't think that the +/- is correlated to any singular factor (for example that a player is good or bad)...it's merely an indicator that something interesting/out of the norm might be happening while that number was created.

 

[Blazer_Hippie]

So, doesn't that mean that Bayless was absolutely brutal earlier in the game, at least in terms of point differential of his unit versus the Knicks?

Well, yes it does mean that his +/- during the earlier section of the game was significantly negative...somewhere the split between 1st and 2nd half Bayless +/- was posted...maybe in the game thread...

But, to my original point, I don't think that it really says anything on it's own about about Bayless' play in the 1st half. In this case it may correlate with his poor play...but I believe it often does not...

 

[hasoos]

Look. A very good player can be on a bad team and end up with a negative +/- for the year. All it takes is for a good player to be on a bad team for them to have a very negative +/-. It doesn't make them a bad player. It just means they are on a bad team. It is also possible for a bad player on a good team to have a very positive +/-.

In 2007-2008 year Dwayne Wade was -256 for the year, in 51 games. That means his game average was about -5. So you going to call him a shitty player? Or was the team he was out on the floor more indicative of what was actually going on?

Comon +/- supporters. Let's hear the retort.

Stats are stats. You take them with a grain of salt, but they don't tell the whole story. Sometimes they are too bold to disregard. Others are too blurry to make any sense out of.

 

[LittleAlex]

Nowhere does he claim the numbers must be wrong...I must be missing something

I come from a financial background and I'm used to financial metrics that don't really tell you anything by themselves. A lot of times they just raise a flag that something unusual might be happening when a number falls outside the norm.

There is so much static in these +/- numbers that I don't believe they say anything of substance about a player on their own...maybe the more advanced adjusted +/- do, I have no idea.

Similar to finance, when you see a weird outlier it can be useful to dig in and understand the underlying causes...but oftentimes it is unrelated to the way the original metric is pointed...FOR EXAMPLE -

Howard's +/- was +12 for the game.

If you take that number at it's face...wow Howard rules! He won the game for us...err...

But if you dig in, you realize he was on the court when Bayless made steals and started getting uber-aggresive and had a spurt of points while the Knicks were't scoring.

What I think the +/- for Howard ends up saying - wow, Bayless was really on fire for a while in the 4th quarter. ...someone could make the argument it was somehow due to Juwan's play, but eyeballing the game he got a couple rebounds but it was mostly Bayless just taking on all comers for a while...nothing to do with Juwan's play.

I just said the number was printed and calculated correctly. Minstrel pointed out that even if the number was calculated correctly it doesn't mean very much. That mirrors my opinion exactly.

 

[Denny Crane]

So you are saying that it's about the player, i.e. that +/- will help quantify the player's value on the court when you consider "chemistry." That may or may not be true, but it certainly isn't true in a single game and, statisticians say, isn't likely to be true even with a single season of data.

All +/- for a single game says is that, for whatever reason, a player was on the floor when his team out-scored the other team by X or was outscored by the other team by X. That reason could be chemistry, good/bad luck, the quality of the players he happened to be up against during his time on the floor, small sample size, etc. The results are too noisy to tell us anything we can use. It's purely descriptive, not prescriptive.

It's a box score stat, just like rebounding or whatever.

The bold part you got right. The rest you are confusing with what +/- means beyond one game.

Within one game, you can say with certainty that he put up a lot of points in a few minutes and still had a huge negative. All his points didn't overcome the negative effect he and his teammates had playing together for whatever reason.

 

[Minstrel]

The bold part you got right. The rest you are confusing with what +/- means beyond one game.

The rest is also correct. My confusion stems from trying to parse what you think the value of a single-game +/- is.

Within one game, you can say with certainty that he put up a lot of points in a few minutes and still had a huge negative. All his points didn't overcome the negative effect he and his teammates had playing together for whatever reason.

Sure, for whatever reason. If all we have is a "for whatever reason," I'm still not sure how this is valuable. We can take it further....for whatever reason, Bayless played in a game that Portland lost. Certainly factual...how do we use this information?

 

[Denny Crane]

The rest is also correct. My confusion stems from trying to parse what you think the value of a single-game +/- is.



Sure, for whatever reason. If all we have is a "for whatever reason," I'm still not sure how this is valuable. We can take it further....for whatever reason, Bayless played in a game that Portland lost. Certainly factual...how do we use this information?

Heh.

So what does 20 rebounds for a guy in the box score tell you?

And what does it mean if he plays 82 games and grabs 21 for the whole season?

Noise!

(not)

 

[Minstrel]

So what does 20 rebounds for a guy in the box score tell you?

And what does it mean if he plays 82 games and grabs 21 for the whole season?

Noise!

(not)

I don't really understand your point. Are you saying this is the same hypothetical player (grabs 20 boards in one game and then 1 rebound in the other 81 games combined) or two hypothetical cases?

Assuming you meant the first, it means he's a terrible rebounder who had one outlier game.

The issue at hand in this post, I think, is the difference between direct measurement and indirect measurement. An individual's rebounds are a direct measure of him, because it concerns his productivity. An individual's +/- is an indirect measure of him, because it concerns his team's productivity when he's on the floor. Indirect measures require a lot more data to filter out noise, because the indirect nature is inherently noisy...it's bringing in lots and lots of confounding factors that take a huge sample size to "even out."

Even direct measures, of course, have confounding factors...a player's rebounding is affected in certain ways by his teammates. So you still need a significant sample size. It just has fewer of those confounding factors, so you don't need as large a sample size to generate meaningful conclusions about individual ability.

I still wouldn't take a single-game PER or a single-game rebound total as terribly meaningful. But it would mean more, to me, than a single-game +/-. I think direct measures, like PER or PPG or RPG, become useful more quickly than indirect measures like +/-.

 

[Denny Crane]

I don't really understand your point. Are you saying this is the same hypothetical player (grabs 20 boards in one game and then 1 rebound in the other 81 games combined) or two hypothetical cases?

Assuming you meant the first, it means he's a terrible rebounder who had one outlier game.

The issue at hand in this post, I think, is the difference between direct measurement and indirect measurement. An individual's rebounds are a direct measure of him, because it concerns his productivity. An individual's +/- is an indirect measure of him, because it concerns his team's productivity when he's on the floor. Indirect measures require a lot more data to filter out noise, because the indirect nature is inherently noisy...it's bringing in lots and lots of confounding factors that take a huge sample size to "even out."

Even direct measures, of course, have confounding factors...a player's rebounding is affected in certain ways by his teammates. So you still need a significant sample size. It just has fewer of those confounding factors, so you don't need as large a sample size to generate meaningful conclusions about individual ability.

I still wouldn't take a single-game PER or a single-game rebound total as terribly meaningful. But it would mean more, to me, than a single-game +/-. I think direct measures, like PER or PPG or RPG, become useful more quickly than indirect measures like +/-.

20 boards means his teammates didn't compete with him for those boards. It's all indirect.

You can't get an assist if the guys you pass to miss their shots.

&c

So a one game +/- is still a box score stat, just like rebounding. By itself, it's not as useful as when you look at everything in context.

What is certain is that he put up those points and the team benefited from his time on the court to the tune of -10 points. It looks to me like the team was a sieve on defense when he was in, and he contributed to that.

 

[Minstrel]

20 boards means his teammates didn't compete with him for those boards. It's all indirect.

You can't get an assist if the guys you pass to miss their shots.

Yes, I made that point in my post (direct stats have confounding factors, too). My use of "direct" and "indirect" doesn't refer to whether there are or aren't confounding factors. It was referring to who's production is being measured...the individual's or the team's.

In a law firm with 15 partners, I'd say that evaluating a single partner on the entire firm's year revenue is less direct than evaluating a single partner on the revenue that came from his/her billable hours.

That isn't to say that one is worthless and one isn't, but they are two different things, IMO, with different merits and drawbacks. When it comes to basketball evaluations, I'd say a drawback of +/- is that it takes many more data points to tell us anything useful about an individual.

It looks to me like the team was a sieve on defense when he was in, and he contributed to that.

I think it would be more productive to evaluate his defense, for a single game, via observation than to use the entire team performance. If someone like Dante Cunningham, for example, was being beaten off the dribble constantly while Bayless was on the floor, it doesn't mean Bayless was the reason for that. As is often noted, correlation doesn't imply causation.

Bayless may well have been responsible for the team's poor defense, but I wouldn't consider the +/- from one game the way to look for that. If I had to make a determination on Bayless' defense in one game, I'd rather watch the game film a few times and keep track of his individual and team defensive responsibilities. If I could track his defense over several seasons, an adjusted +/- metric could certainly be useful.

 

[Denny Crane]

Yes, I made that point in my post (direct stats have confounding factors, too). My use of "direct" and "indirect" doesn't refer to whether there are or aren't confounding factors. It was referring to who's production is being measured...the individual's or the team's.

In a law firm with 15 partners, I'd say that evaluating a single partner on the entire firm's year revenue is less direct than evaluating a single partner on the revenue that came from his/her billable hours.

That isn't to say that one is worthless and one isn't, but they are two different things, IMO, with different merits and drawbacks. When it comes to basketball evaluations, I'd say a drawback of +/- is that it takes many more data points to tell us anything useful about an individual.



I think it would be more productive to evaluate his defense, for a single game, via observation than to use the entire team performance. If someone like Dante Cunningham, for example, was being beaten off the dribble constantly while Bayless was on the floor, it doesn't mean Bayless was the reason for that. As is often noted, correlation doesn't imply causation.

Bayless may well have been responsible for the team's poor defense, but I wouldn't consider the +/- from one game the way to look for that. If I had to make a determination on Bayless' defense in one game, I'd rather watch the game film a few times and keep track of his individual and team defensive responsibilities. If I could track his defense over several seasons, an adjusted +/- metric could certainly be useful.

I'm not talking about Bayless' defense. The team defense was a sieve, or he'd be a net positive. The stat does tell us that much and more. More you ask? It also says the rest of the team didn't score (outscore) the opposition with him on the court.

The +/- for a game is just as relevant as rebounds or assists or points. While you think the sum of all games' +/- is important or significant, I feel it's far less significant than counting the games played at +/- positive vs. negative -- that's a stat that becomes more and more meaningful with just a few games' worth of data.

 

[PapaG]

If I could track his defense over several seasons, an adjusted +/- metric could certainly be useful.

This was just being discussed in the START BAYLESS thread.

posted by bobf:

Bayless has the worst defensive rating in the entire NBA for players with 500+ minutes based on linear regressed +/- using data since 2005. It's a small sample for him, but doesn't bode well.

http://www.sportstwo.com/forums/showpost.php?p=2197477&postcount=62

 

[Minstrel]

This was just being discussed in the START BAYLESS thread.

Yeah, I saw that. Of course, Bayless doesn't have even one full season of starter minutes yet. It's definitely not a great sign. From observation, though, he has the tools and mindset to be a good defender...so hopefully that's what he'll develop into and the numbers will bear it out.