two dumb questions

[ampersand5]

I apologize if these may seem silly, I just didn't know the answers and thought someone could shed some light on the topics.

1) when running regressions on an APM (whether it be for a SPM or for draft stuff), why do we not do it by player position (or at the minimum, demarcations of guards, wings and bigs)? It seems as if what could make a guard effective is very different than what could make a centre effective. Maybe some stats are especially important coming from certain positions, or the relationship between stats differs for different positions...?

2) How do we know when a neural network would be preferable to a regression - essentially, how do we know what we are looking for is linear?

thanks

[NateTG]

1) when running regressions on an APM (whether it be for a SPM or for draft stuff), why do we not do it by player position (or at the minimum, demarcations of guards, wings and bigs)?

I don't entirely understand: are you asking why people don't include player role information as a variable when they run regressions to produce a statistical metric like APM, or are you asking why people don't incorporate position information when they regress something else with APM as the 'independent variable'?

Regardless, I think the basic reasons are going to be that it cuts down on sample size, and that good data may not be readily available.

2) How do we know when a neural network would be preferable to a regression - essentially, how do we know what we are looking for is linear?

This is part of a larger question about how to choose the most appropriate machine learning technique for a particular application:

http://en.wikipedia.org/wiki/Machine_learning

In general, it depends on the data set and the application or question you have in mind.

Because they're so simple, one approach is to start by running a linear regression and testing how well it fits the data.

I'm not an expert, but my understanding is that neural nets are good when you have a representative sample of data, don't care about how the computer is making the decision, and you want to be able to evaluate things quickly. I'm not sure there are ready examples where they're particularly appropriate for NBA analysis.

[ampersand5]

thanks - in regards to the first question, I meant - why is there not a different equation used for calculating the SPM of a guard, wing and big.

[mtamada]

Part of the answer is contained in your question. Why are the appropriate categories guard, wing, and big? What about PG, SG, SF, PF, and C? There are after all five players on the court, not three. And if we try to talk about roles, in these days of point forwards, stretch fours, point guards who look to score, a few remaining defensive specialists, etc. we've got even more than five categories of players.

Even with just three categories you get players who are in a gray area, and have to make Procrustean judgements to jam them into a pigeonhole, to mix two or three metaphors. Was Charles Barkley a wing or a big? Were Jerry West and Clyde Drexler guards or wings?

[DSMok1]

thanks - in regards to the first question, I meant - why is there not a different equation used for calculating the SPM of a guard, wing and big.

I've played with it in my BPM regressions, and it just doesn't seem to be that useful for an all-in-one stat. As Mike noted, role definition is so foggy, and not a linear continuum--and it changes, for a given player, year to year and team to team, and even game to game. Also, as Nate said, it cuts down sample size a lot.

By the way, what does Procrustean mean?

[mtamada]

Like a lot of Greek myths, the story of Procrustes is a bloody and vengeful one. But it has created a very useful metaphor.

The tl;dr version is that Procrustes was a bad guy and we don't want to imitate him by trying to make things or people fit into places where they just don't fit. Square pegs and round holes is a non-bloody modern version of this concept.

The wikipedia article is quite good at quickly summarizing the story and describing its influence over the millennia. I give it bonus points for citing Walter Murch, a sound designer and editor of Hollywood movies who is one of the most analytical, wise, and imaginative thinkers about movies that I've seen.

[Mike G]

The mrs. makes quiche without crust. Pie, too. I'm more procrustean.