Wednesday, 1 May 2013

Supressor variables

Hejsan - as we say in Sweden.

Nothing for the shoots contingent, but a bit of a science query.

I have been noticing with real and simulated data that controlling for certain variables (say Z) improves basic correlations (between X and Y). It turns out this is due to Z acting as a supressor variable. This is essentially why:

It has to due with Z accounting for variance in Y that would be viewed as error in the simple correlation between X and Y. This seems to be overlooked a lot in ecology and evolution - it barely comes up in the literature - probably because it is a rare event.

Did you guys ever learn about this?