ClutchFans Game Thread: Rockets @ Mavericks 11/20/2013

The author was Arturo, and he was referring to using +- as a model over a single season. It was noted that the margin for error decreases across multiple seasons (larger sample sizes), which is why using +-, even APM has issues when used across larger sample sizes *UNLESS* the variation is well outside of 2 standard deviations.

And even then, the correlation was specific to the issue of wins, not player performance per se.

There's an excellent post up on Wages of Wins regarding the topic.

http://wagesofwins.com/2011/03/05/deconstructing-the-adjusted-plus-minus-model/

So while APM (adjusted +-) has low correlation over a single season, as you extend over multiple seasons or high minute counts (large sample sizes), you see that correlation become more relevant, though still low in terms of calculating win shares and predictive performance.

And the validity of the model goes out the window for the most part once a player switches teams (IE, APM going from Team A to Team B is non correlative, which is somewhat intuitive given changes in role and schemes).

And yes, both +- and APM are flawed models, which means that +- and APM should be taken with a rather large grain of salt on it's own. However, I would still posit that when taken in a larger context with other statistical models that it can be viewed as still useful in terms of reinforcing results that point towards specific trends.

IE, take APM + PER + EFF + WARP + VORP + WAR + TS% + AST48, etc. If they all point to similar conclusions, then it becomes far less likely that the APM is completely meaningless.

The problem with all these statistical models is on their own, especially with sample sizes smaller than multiple season is they are inherently prone to high variance, and thus the results often fall within the margin of error. It is often the tendencies of people to pull a single such stat and claim that it is indicative of performance trending upward or downward.

I prefer to view them within a statistical context (multiple models + granular data + situational information), and use them as trend predictors vs statistical projection.

In other words, I can use a series of models to indicate that something is moving in a particular direction, I just can't tell you exactly how much it's going to move or when it will stop moving.

Don't know if that made sense. :grin:

 

Well, that was frustrating.

 

great coaching guys lol

 

This is the problem with stats applied to behavioral data. Very large samples are needed because the variance is so great,and even when we have a large
sample the amount of variance accounted for when various regression techniques are used is very small. I consider basketball behavioral data.
An excellent point was made that the +/- data is useless when a player switches teams. The entire basketball scene is dynamic. Consequently, the value of any independent variable changes when the specific situation changes. Food for thought when one attempts to make a prediction.

 

Cos monta Ellis have it all?

 

In an effort to get more consistent play overall, Rockets center Dwight Howard said players might have to ask out of games more often.

James Harden is fourth and Chandler Parsons is seventh in the NBA in minutes played per game. Howard has played an average of 35.2 minutes, the 31st most in the NBA.

“We have to play four consistent quarters of basketball,” Howard said. “It can’t be one quarter or two quarters or three. It has to be four. Guys have to know their limits, know when they need a break, be humble enough to raise their hand and get somebody else in to play those minutes. A lot of times, guys cruise. I’ve cruised a couple times, wanted to stay out there longer.

“You can’t do that if you want to be a great team. You have to go out there and go hard and play 48 minutes as hard as you can, whether that’s 20 minutes, five minutes, 30 minutes (each), we have to all give it all and trust that the guy off the bench is going to give us the same effort or more. We all have to get to that level where we trust each other and trust that the guy that is going to come in for us will come in with the same intensity. Once we get there, we’ll be a tough team to beat.”