dk7th wrote:

Bonn1997 wrote:

dk7th wrote:

mreinman wrote:

dk7th wrote:

Bonn1997 wrote:

dk7th wrote:

Bonn1997 wrote:

mreinman wrote:why are we looking at RPM when we have WS48 which is a much more respected stat?

Yeah, if you Google "problems with Real Plus Minus" you'll need a few days to read about all its problems.

The biggest problem is that the coefficients are usually statistically insignificant - or in lay terms, there's too much error in the measurement to say that a player's impact is different from zero unless the RPM is very high or very low and based on a huge sample.

what is very high or very low? and what is a large-enough sample? i look at the numbers and they dovetail pretty well with what i am seeing on the court over half a season. also, i could not find a list of win share per 48 to look at for comparison's sake. do either of you have a link for me to examine?

Well, you'd have to actually do the right statistical tests to determine the confidence intervals for each player. I doubt half a season is a large enough sample anyway from what I've read about the stat.

what about using two different statistical models and examining how well they correlate? i liked the real plus minus, mreinman said ws48 was more respected, i examined both-- briefly-- side by side, and saw similar-enough results/measures to feel confident in my conclusion.

is that not enough and if not, how not?

what are you comparing side by side?

espn's real plus minus and your bball ref ws48

Honestly, to have a really good sense of which stats to use requires grad level training in statistics and full-time dedication to learning about all of them. I don't really have that time dedication (thankfully). So I just look at all the major advanced stats: win shares, wins produced, real plus minus (it's OK to look at it), box plus minus, value over replacement, the player tracking data, and more. When I'm too busy for all that, I just look at the win shares but I realize that's not a great approach.

right, and by every measure outside of the eye test afflalo is being awarded too many minutes.

Bonn1997 wrote:

dk7th wrote:

Bonn1997 wrote:

dk7th wrote:

mreinman wrote:

dk7th wrote:

Bonn1997 wrote:

dk7th wrote:

Bonn1997 wrote:

mreinman wrote:why are we looking at RPM when we have WS48 which is a much more respected stat?

Yeah, if you Google "problems with Real Plus Minus" you'll need a few days to read about all its problems.

The biggest problem is that the coefficients are usually statistically insignificant - or in lay terms, there's too much error in the measurement to say that a player's impact is different from zero unless the RPM is very high or very low and based on a huge sample.

what is very high or very low? and what is a large-enough sample? i look at the numbers and they dovetail pretty well with what i am seeing on the court over half a season. also, i could not find a list of win share per 48 to look at for comparison's sake. do either of you have a link for me to examine?

Well, you'd have to actually do the right statistical tests to determine the confidence intervals for each player. I doubt half a season is a large enough sample anyway from what I've read about the stat.

what about using two different statistical models and examining how well they correlate? i liked the real plus minus, mreinman said ws48 was more respected, i examined both-- briefly-- side by side, and saw similar-enough results/measures to feel confident in my conclusion.

is that not enough and if not, how not?

what are you comparing side by side?

espn's real plus minus and your bball ref ws48

Honestly, to have a really good sense of which stats to use requires grad level training in statistics and full-time dedication to learning about all of them. I don't really have that time dedication (thankfully). So I just look at all the major advanced stats: win shares, wins produced, real plus minus (it's OK to look at it), box plus minus, value over replacement, the player tracking data, and more. When I'm too busy for all that, I just look at the win shares but I realize that's not a great approach.

right, and by every measure outside of the eye test afflalo is being awarded too many minutes.

When I last looked at the #s carefully, my conclusion was that all our guards were basically sixth man material and were roughly comparable.

you might look again... galloway's numbers merit increased minutes. all i have been advocating is reducing afflalo's minutes by 5 to 28mpg and increasing galloway's by 5 to 29mpg.