Version 2.2 of coxme has been posted to CRAN, Windows versions and
mirrors should appear in due course. This is a major update with three
features of note:
1. A non-upwardly compatable change:
Extractor functions: beta= fixed effects, b=random effects
nlme lme4 coxme <2.2 coxme 2.2 lmekin 2.2
------------------------------------------------------
beta fixef fixef fixef fixef fixef
b ranef ranef ---- ranef ranef
var(beta) vcov vcov ---- vcov vcov
var(b) VarCorr VarCorr ranef VarCorr VarCorr
Notice that "ranef" in the prior release was out of step with
everyone
else, which became obvious when working on lmekin. Also, the constructs
"fit$coef$random" and "fit$coef$fixed" no longer work due to
some
internal rearrangments.
2. The lmekin function is now part of the package. This is the last
portion of the older "kinship" package to be updated; I will be
talking
to the current maintainer of kinship about making it depreciated. This
version of lmekin is both more stable and more capable than the old
one.
Why use lmekin? It can fit most of the models that lme or lmer can,
and uses the same formula notation as lmer. However, if lmer can fit
your needs I see no good reason to use lmekin and lots of reasons not to
use it (plots, residuals, documentation, ...). The lmekin routine can
also fit models with complex user-specified variance structures,
however, of which familial kinship correlation matrices are the
principle example.
3. More vignettes.
Terry Therneau
