[R] Specifying interactions in coxme

[Patrick Phillips]

I am trying to estimate variance components from a complex random effects survival model using coxme (Package version 2.2-5). I am having problems specifying interaction effects in the random terms. I wonder if this is simply not implemented in coxme yet.

For instance, using the model

coxme(Surv(DeathAge,Dead) ~  treatment + (1 | block/tank))

Works fine (tank nested within block), but 

coxme(Surv(DeathAge,Dead) ~  treatment + (1 | block) + (1 | block:tank))

fails with "Error in getgroups(f2$group, m) : Invalid grouping factor block:tank”.

By my understanding of model specification in R, these should be equivalent statements (assuming that that the tank effects have been coded appropriately as unique within blocks). Any interaction specified using “:” appears to fail.

I am actually interested in much more complex models but can’t seem to be able to specify any interactions effects. For instance a factorial model like,

coxme(Surv(DeathAge,Dead) ~  treatment + (1 | room*block))

fails with "Error in getgroups(f2$group, m) : Invalid grouping factor room * block”.

Since I can’t directly specify the interaction effects and the factorial operator “*” does not seem to work, I appear to be SOL. It seems that only nested effects specified by the “/“ operator are allowed. Multiple nesting is also OK, which is great for a single hierarchical nesting design, but not sufficient for more complex random effects models.

Any thoughts would be most welcome.

Thanks, Patrick