Is Ennis the bet frosh in college basketball?

[Orangezoo]

But not R^2

Incorrect. Most frequently is it 0-1 yes. The fact is the steps before you have a model are what is problematic with these models to begin with. Its an arbitrary process as to building the model and what variables to use.

 

[Orangezoo]

Hence why I was referencing correlation. The r2 value is still associated with both pos and neg values and always was when I was in college. Just pulled an old matlab book to double check

 

[dinno28]

is he the best freshman ... no ... is he the best freshman point guard ... yes by a mile

 

[Cusefannotindc]

Hence why I was referencing correlation. The r2 value is still associated with both pos and neg values and always was when I was in college. Just pulled an old matlab book to double check

I am not sure why you brought up the R value in the first place, but R^2 can be associated with each, but since it is R squared, it can never actually be negative.

 

[Cusefannotindc]

I didnt say large differences. I said differences where you can clearly define you are comparing one set of circumstances vs another set and they are not the same. In this case the delta of vegas vs pomeroy likely contains very similar analysis and there is no way to isolate one methodology vs the other. Therefore it is too autonomous of a decision.

Disagree, training/testing. You are simply looking at how each performs out of sample, even if they are similar in data (and since there is only limited data to potentially go in, they are).

 

[Orangezoo]

I am not sure why you brought up the R value in the first place, but R^2 can be associated with each, but since it is R squared, it can never actually be negative.

I brought it up because its important. The models have to come from somewhere. How something fits on a model tells much less about it than how it was built. Defining variables be they numerical or boolean are at the root of model creation and whether you have a good model that is robust enough or not. The r2 value is associated with negatives when you are not trying to fit something and instead going back a step.

 

[Orangezoo]

Disagree, training/testing. You are simply looking at how each performs out of sample, even if they are similar in data (and since there is only limited data to potentially go in, they are).

And therefore is where not combining logic with modeling is a problem.

 

[Orangezoo]

Disagree, training/testing. You are simply looking at how each performs out of sample, even if they are similar in data (and since there is only limited data to potentially go in, they are).

You cannot compare vegas to pomeroy as apples vs oranges. Its a lie.

 

[Cusefannotindc]

You cannot compare vegas to pomeroy as apples vs oranges. Its a lie.

I don't care if they are significantly different in build if one performs significantly better than the other in out of sample. Variables should be logical, but in the end performance matters, and overfitting would reveal itself with significant out of sample testing.

 

[Orangezoo]

I don't care if they are significantly different in build if one performs significantly better than the other in out of sample. Variables should be logical, but in the end performance matters, and overfitting would reveal itself with significant out of sample testing.

And therefore it would be flawed to compare them. Thats my point. The conditions are what validates the two being separate. And in the end this example still does not validate these models as better than an individual using the data elements used to create the model to be at a similar rate of success. In sports 1 or 2 pct chance is a free throw or a rebound.

 

[Cusefannotindc]

And therefore it would be flawed to compare them. Thats my point. The conditions are what validates the two being separate. And in the end this example still does not validate these models as better than an individual using the data elements used to create the model to be at a similar rate of success. In sports 1 or 2 pct chance is a free throw or a rebound.

My point is I see significant arbitrage available. Probably time to end all of this though, we have gotten way off topic, in what is probably an area of interest to a very few.

 

[Orangezoo]

I don't care if they are significantly different in build if one performs significantly better than the other in out of sample. Variables should be logical, but in the end performance matters, and overfitting would reveal itself with significant out of sample testing.

Also the reality is that if a model requires over fitting after scrumming it out then it is something you simply cant model. But you still have to realize that pomeroy vs vegas is not a model. Its just a comparison. And one where similar methodologies are used. Prove to me someone doesnt use a simpler approach to get 80 pct.

I presume you have never participated in or conducted an internal audit of a model. The whole intent is to assess how it is fitted. You over value performance which is likely due to you experience with the type of modeling you referenced. Performance is still only as good as how well it is fitted. Anything that has a large pct of qualitative factors affecting the outcome will always end up being over fitted by those who look at it in a logical sense. As a modeler you will argue performance and still getting a good result. Combine the two and you either accept an ok model or realize there is too small of a benefit to even model it out.

 

[Orangezoo]

My point is I see significant arbitrage available. Probably time to end all of this though, we have gotten way off topic, in what is probably an area of interest to a very few.

Fair enough.

 

[Orangezoo]

And just to add. I didnt drag this out for any other reason than to say you can make a point without taking a nut shot at someones intelligence and i apologize if i seem to have gestured in that manner myself as it was just in defense of the others being called out unneccessarily. Whether you did so in the spirit of argument or not its not needed. Just dont care for those who take that stance. Otherwise it matters little.

 

[Cusefannotindc]

And just to add. I didnt drag this out for any other reason than to say you can make a point without taking a nut shot at someones intelligence and i apologize if i seem to have gestured in that manner myself as it was just in defense of the others being called out unneccessarily. Whether you did so in the spirit of argument or not its not needed. Just dont care for those who take that stance. Otherwise it matters little.

You can disagree with a model or an advanced statistic, but too many people here trash them for the sake of trashing them, especially when they don't understand them.

 

[Orangezoo]

You can disagree with a model or an advanced statistic, but too many people here trash them for the sake of trashing them, especially when they don't understand them.

Maybe but being privileged to knowledge of how statistics work does not make us einstein. And while I can appreciate passion for them the reality is statistical analysis is mathematical politics in many respects. Therefore its not always wrong to dismiss it. At the end of the day we are all cuse fans. We all have different life interests and areas where we know more than others. Sharing those things is a choice. But no need to attack those who dismiss it. Especially when its something that in fairness is not an absolute. I get your passion as I said. But same team..

 

[jordoo]

You can disagree with a model or an advanced statistic, but too many people here trash them for the sake of trashing them, especially when they don't understand them.

I understand Wins Produced better than I ever have after this thread and am even more convinced that it is a very silly stat as I first stated.

 

[4evaOrange]

This thread hurts my head...

 

[4evaOrange]

This hurt thread head my...errrrr

 

[jekelish]

You can disagree with a model or an advanced statistic, but too many people here trash them for the sake of trashing them, especially when they don't understand them.

Or you think people trash them for the sake of trashing them. Some people just might find them to be absurd because they know the game of basketball and realize that not everything boils down to some metric.

 

[Upstate]

Goodman has him #1 Frosh in the country now:

http://insider.espn.go.com/blog/jef...data=2013_insidertwitter_Top10Freshmen_social


1. Tyler Ennis, PG, Syracuse Orange
6-foot-2, 180 pounds

The Orange floor leader has moved into the top spot for a variety of reasons. Jabari Parker and Julius Randle have struggled lately, and Ennis has also quarterbacked his team to an undefeated record. Ennis runs the team, takes care of the ball, makes his teammates better and also picks his spots when to score. Ennis, who ranks third in the nation in assist-turnover ratio, averaged 11.5 points and seven assists in a pair of wins against Virginia Tech and North Carolina to help keep the Orange undefeated.