I am performing partial redundancy analysis in which I am trying to explain variation in Y with response to X after partialing out the impact of W. In this case, matrix W is a set of n-1 AEM spatial variables, where n is the number of samples. My sample size is already small, and I dont want an over determined system (number of predictors>n). Which brings me here: do I even need to count these n-1 variables since Im partialling them out anyway? In one sense they are predictors since they are used as model inputs, but not in the sense that Im trying to explain variation due to them...
Asked
Active
Viewed 13 times
0
-
The degrees of freedom (df) are the measure of the stability of a variance estimator. Suggest you test for the effective number of df. For example, see http://www.asasrms.org/Proceedings/papers/1998_119.pdf – Carl Jun 28 '21 at 18:02