I am studying LMM and am confused about the degrees of freedom associated with testing covariance parameters on the boundaries of the parameter space using restricted maximum likelihood, REML.
Suppose I have a model built from a 2-level clustered dataset and have 3 random effects, but am testing to see if only 1 of them is zero. What would be the degrees of freedom for this test?
Also, am looking for some background information on this, but not anything too lengthy.
I tried looking up information online but couldn't find anything direct or specific. Am not trying to read a full book on this.
