Re: Problem about Variance Component Estimation

linglan wrote:
Dear All:

I have Yij=Xb+Qij+Gij+Eij follow normal distribution. Xb are fix
effect while Q,G, E are random effect following normal distribution.
E(Y)=XB, E(Q)=E(G)=E(E)=0, I know the vector of Y and X. I need to

estimate Vq,Vg,Ve and b.


I use REML and fisher scoring method to get the estimation of
Variance Component first. I set initial values for three V. At r
iteration, V(r+1)=V(r)+delta(V). delta(v) is the product of fisher
information matrix and derivative of likelihood.

Shouldn't delta(V) be be the *inverse* of the information matrix times
the derivative of the log likelihood?

m00es

.