The lmekin function fits a linear mixed effects model, with random
effects specified in the same structure as in the coxme
function.
lmekin(formula, data, weights, subset, na.action, control,
varlist, vfixed, vinit, method = c("ML", "REML"),
x = FALSE, y = FALSE, model=FALSE,
random, fixed, variance, ...)
An object of class lmekin
.
a two-sided formula with the response as the left hand side of a
~
operator and the fixed and random effects on the right.
an optional data frame containing the variables named in the formula
.
further model specifications arguments as in lm
; see there
for details.
optional list of control options. See coxme.control
for details.
the variance family to be used for each random term. If there are
multiple terms it will be a list of variance functions.
The default is coxmeFull
.
Alternatively it can be a list of matrices, in which case the
coxmeMlist
function is used.
optional named list or vector used to fix the value of one or more of the variance terms at a constant.
optional named list or vector giving suggested starting values for the variance.
fit using either maximum likelihood or restricted maximum likelihood
if TRUE the X matrix (fixed effects) is included in the output object
if TRUE the y variable is included in the output object
if TRUE the model frame is included in the output object
In an earlier version of lmekin
the fixed and random effects were separate arguments. These arguments
are included for backwards compatability, but are depreciated.
The variance argument is a depreciated alias for vfixed.
any other arguments are passed forward to coxme.control
.
Terry Therneau
Let \(A= \sigma^2 B\) be the variance matrix of the random
effects where \(\sigma^2\) is the residual variance for the
model. Internally the routine solves for the parameters of
\(B\), computing \(A\) at the end. The vinit
and
vfixed
parmaters refer to \(B\), however.
It is possible to specify certain models in lmekin
that can not be fit with lme, in particular models with
familial genetic effects, i.e., a kinship matrix, and hence the
name of the routine. Using user-specified variance functions an even
wider range of models is possible.
For simple models the specification of the random effects follows the
same form as the lmer
function. For any model which can be fit
by both lmekin
and lmer
, the latter routine would
normally be prefered due to a much wider selection of post-fit tools
for residuals, prediction and plotting.
Much of the underlying model code for specification and manipulation
of the random effects is shared with the coxme
routine. In
fact lmekin was originally written only to provide a test routine for
those codes, and no expectation that it would find wider utility.
lmekin.object
, coxme
data(ergoStool, package="nlme") # use a data set from nlme
fit1 <- lmekin(effort ~ Type + (1|Subject), data=ergoStool)
if (FALSE) {
# gives the same result
require(nlme)
fit2 <- lme(effort ~ Type, data=ergoStool, random= ~1|Subject,
method="ML")
}
Run the code above in your browser using DataLab