proc mixed- zero covariance estimates

[archnova]

Hi,

  I have a data set where I have a dependent variable Y , a treatment variable (Trt) and set of random factors z1(10 levels), z2(2 levels) and z3(2 levels) such that z3 is nested in z2 and z2 is nested in z1 I am using proc mixed to analyze this data. Code below :

proc mixed method=REML;

class trt z1 z2 z3;

model y = trt;

random z1 z2(z1) z3(z1 z2);

run;

The problem is the covariance estimate for z2 comes out as zero. I am wondering why I am getting such an estimate. Any help/insight would be appreciated.

PS: I have also tried other methods i.e. MIVQUE0 , type 1 and type 2 and ML

Regards,

Arunava

[SteveDenham]

It most likely means that there is insufficient variability remaining after fitting z1 to get a positive estimate with the data you have in hand.  The only real solution is the most expensive one--obtain more data.

Steve Denham