View Single Post
Old 08-19-2013, 07:51 PM   #656
SebC
tromboner
 
SebC's Avatar
 
Join Date: Mar 2006
Location: where the lattes are
Exp:
Default

Quote:
Originally Posted by New Era
Yet the whole method is reliant on the classification of league strength for the measure to be considered accurate, no?
The relative league strengths are empirically determined from observable data. They're not arbitrary assumptions.

Quote:
Originally Posted by New Era View Post
Can we agree that a offensive philosophy is going to increase scoring and a defensive philosophy is going to decrease scoring?
For a team, sure. For an individual, less clear (e.g. a less offensively-talented player might receive more minutes with a defensive-minded coach, and thus score more).

Quote:
Originally Posted by New Era
Can we agree that team philosophy is going to have some effect on a player's ability to produce points?
Yes, but having the size and direction of the effect would be better.

Quote:
Originally Posted by New Era
Can we agree that a player's psychological strength is going to have an impact on his scoring?
Yes. But it would do that both at the NHL level and at lower levels. Thus, NHL-E accounts for psychological strength.

Quote:
Originally Posted by New Era
Can we also agree that a player's psychological strength may be tested by the conditions under which his team is managed?
Sure.

Quote:
Originally Posted by New Era
Now, can we agree that these factors, which are not measurable by NHL-E can have a significant affect on outcomes?
No, because things like phychological strength are captured in NHL (though perhaps not weighted properly). But I will agree that there are factors not captured by NHL-E (that are reflected as "error" in the model). The thing is, we can tell how big factors captured by NHL-E are vs. factors not captured by NHL-E by looking at the predictive value of NHL-E. So, if NHL-E explains 75% of a player's career PPG, then all the factors it doesn't capture, combined, can only explain 25%. And if those other factors can't be used predictively, then their supposed effects might not really exist at all (and that 25% is just "luck", or factors that we haven't identified or figured out how to measure yet). The NHL-E to career PPG correlation is a strong one, which means that the factors that NHL-E doesn't capture must have relatively weak effects.
SebC is offline   Reply With Quote