[BostonCard]

This is a little bit like the bias/variance tradeoff in statistics.

It seems like in order to avoid biases in a statistical measure of a team quality, you wind up incorporating more and more variables and making the measure increasingly convoluted and less reproducible (and sometimes more proprietary).  Things like the NET and the FEI wind up giving you a summary that is not obviously confounded by things like opponent quality, but it winds up being a bit of a black box.

As always, George Box' aphorism applies: All models are wrong; some models are useful.

BC

[BostonCard]

This is a little bit like the bias/variance tradeoff in statistics.

It seems like in order to avoid biases in a statistical measure of a team quality, you wind up incorporating more and more variables and making the measure increasingly convoluted and less reproducible (and sometimes more proprietary).  Things like the NET and the FEI wind up giving you a summary that is not obviously confounded by things like opponent quality, but it winds up being a bit of a black box.

As always, George Box' aphorism applies: All models are wrong; some models are useful.

BC