Programming the statistical procedures from SAS

proc glimmix order of parameters for spatial correlation?

Reply
Occasional Contributor
Posts: 9

proc glimmix order of parameters for spatial correlation?

Hi,

I'm fitting a binomial model using Proc Glimmix in sas 9.2. My input for the random effect looks something like this:

random _residual_ / subject=rcgroup type = SP(exp)(X Y);

In the Parms statement you can pass two start (or final with noiter option) values for parameters estimated in sp(exp). The problem is that you simply give the parameter values in order (the order sets what they mean) but no documentation tells you which one comes first and fully describes what they are. Does anyone have this information?

Once distance (dij) between observations is known, SAS gives this as the exponential (exp) formula for sp(exp).

cov=var*exp(-d/A)

where var is variance d is distance and A is a shape parameter. Usually, in fitting a variogram using the exponential function, you need the nugget, sill, and range. I am assuming that SAS only estimates var and A here. But is this true? From other experience, it seems that variance does not need to be estimated. Further, when I arbitrarily set the parameters for sp(exp) and give the noiter command. Only one parameter estimate returns as I had specified, does this mean that one has been modified or that neither have and only one has been reported and the other non-matching value is something else?

For example

random _residual_ / subject=rcgroup type = SP(exp)(X Y);
parms (0.096) (0.9983)/noiter;

returns this in output


Covariance Parameter Estimates

Standard
Cov Parm Subject Estimate Error

SP(EXP) rcgroup 0.09600 0.001113
Residual 1.0649 0.004435



Thanks,
Seth
Ask a Question
Discussion stats
  • 0 replies
  • 206 views
  • 0 likes
  • 1 in conversation