topic Re: Unequal variances on NLMIXED in Statistical Procedures

Unequal variances on NLMIXED

marcioadg — Thu, 14 Aug 2014 21:10:42 GMT

This is the part of the code I used for equal variances and work fine.

model y ~ normal(L + U*(z1) + lvar, s2e);

random lvar ~ normal(0,blockvar) subject=block out=blups;

For unequal variances, I tried the one below and it outputted errors in the log.

model y ~ normal(L + U*(z1) + lvar, s2a s2b);

random lvar ~ normal(0,blockvar) subject=block out=blups;

Could you please help me identify what I am missing,

Thank you,

Marcio

Re: Unequal variances on NLMIXED

SteveDenham — Fri, 15 Aug 2014 11:49:29 GMT

The normal argument to the MODEL statement takes only a mean and variance. I think you will have to code up the log likelihood function with the two variances, and use the general argument.

Steve Denham

Re: Unequal variances on NLMIXED

marcioadg — Fri, 15 Aug 2014 19:03:13 GMT

Thank you Steve,

But I am not sure how to do that. I looked into some papers (http://www.education.umd.edu/EDMS/fac/Harring/MIsc/WUSS-Paper.pdf) and websites (SAS/STAT(R) 9.22 User's Guide) but I am not being able to do it.

Would you recommend any resource on how to do that?

Thank you,

Marcio

Re: Unequal variances on NLMIXED

SteveDenham — Mon, 18 Aug 2014 11:19:07 GMT

I have never done it, so I can't offer much beyond searching this site and the SAS-L archives. I'm sorry.

Steve Denham

Re: Unequal variances on NLMIXED

lvm — Tue, 19 Aug 2014 20:22:21 GMT

It is not clear what you are trying to do. I suspect you have two groups and want a different variance for each group?