On Mon, Aug 18, 2008 at 11:20 AM, Aaron Mackey <[EMAIL PROTECTED]> wrote: > I have a fairly large model:
>> length(Y) > [1] 3051 >> dim(covariates) > [1] 3051 211 > All of these 211 covariates need to be nested hierarchically within a > grouping "class", of which there are 8. I have an accessory vector, " > cov2class" that specifies the mapping between covariates and the 8 classes. > Now, I understand I can break all this information up into individual > vectors (cov1, cov2, ..., cov211, class1, class2, ..., class8), and do > something like this: > model <- lmer(Y ~ 1 + cov1 + cov2 + ... + cov211 + > (cov1 + cov2 + ... | class1) + > (...) + > (... + cov210 + cov211 | class8) > But I'd like keep things syntactically simpler, and use the covariates > and cov2class > variables directly. I haven't been able to find the right syntactic sugar > to get this done. Before considering the syntax you should consider the complexity of the model. Even with 3000+ observations estimating fixed effects for over 200 covariates is a risky business. To follow that up by incorporating dozens of covariates in a vector-valued random effects term with a general variance covariance matrix is not likely to be productive. When estimating variances and covariances of random effects the complexity of the estimation problem increases according to the square of the dimension of each random-effects vector. For a term of the form (1|fac) there is one variance-covariance parameter to estimate. For a term of the form (1+x|fac) there are 3 such parameters to estimate, (1+x1+x2|fac) requires 6, etc. Assuming that your 211 covariates are more-or-less equally distributed among the 8 classes you would have about 27 per class which means that you have 378 variance-covariance parameters to estimate for each of your 8 classes. Even with 3000+ observations it would be optimistic to expect to estimate that many variance-covariance parameters. You may want to move your question to the R-SIG-Mixed-Models mailing list and follow up on the model specification there. ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
