[R-sig-ME] New re-written version of lmerTest will soon be released - please test, use and comment
rune.haubo at gmail.com
Fri Mar 23 12:53:18 CET 2018
The lmerTest package has been re-written and the new version will soon
be released as package version ≥ 3.0.
The lmerTest package provides p-values in type I, II or III anova and
summary tables for linear mixed models (lmer model fits cf. lme4) via
Satterthwaite’s degrees of freedom method; a Kenward-Roger method is
also available via the pbkrtest package. Model selection and
assessment methods include step, drop1, anova-like tables for random
effects (ranova), least-square means (LS-means; lsmeans) and tests of
linear contrasts of fixed effects (contest).
The most important changes for end-users are:
1. More robust and less error-prone implementation with much better
2. Faster evaluation of summary and anova tables - much faster for
time-consuming model fits
3. New functions include drop1(), contest(), as_lmerModLmerTest(),
show_test(), and ranova() – see details below.
Given that the codebase has been rewritten completely from scratch it
is not unlikely that a few ‘childhood diseases’ are lurking in the
details. Please help us cure out such maladies by installing and
testing the new package and report bugs if you see them at
https://github.com/runehaubo/lmerTestR/issues. Comments and
suggestions are most welcome.
The new lmerTest package can be installed with
# install.packages("devtools") # run if you don't have devtools installed.
The new user interface is almost 100% backward compatible with
previous versions (see details below).
Details of "New features" and "Changes to the user interface" are
given below -- full details are available at
That document also contains technical details relevant for packages
that depend on lmerTest. We will be reaching out to the maintainers of
packages that depend on lmerTest in a separate email since a couple of
packages needs to be tweaked.
1. ranova() - ANOVA-like table of random effects via likelihood ratio
tests with methods for both lmerMod and lmerModLmerTest objects.
ranova() is similar to the old rand() and essentially produces a
drop1() table for random-effect terms. ranova() can either test
reduction of random-effect terms to simpler structures or it can test
removal of entire random-effect terms; the rules for how complex
random-effect terms are reduced is described in help(ranova).
2. drop1() - F-tests of fixed-effect terms using Satterthwaite or
Kenward-Roger methods for denominator degrees of freedom. These
‘single term deletion’ tables are useful for model selection and tests
of marginal terms. Compared to the likelihood ratio tests of
lme4::drop1 the F-tests and p-values of lmerTest::drop1 are more
accurate and considerably faster since no additional model fitting is
3. as_lmerModLmerTest() - an explicit coerce function from class
'lmerMod' to 'lmerModLmerTest'.
4. contest() - tests of contrasts, i.e. tests of linear functions of
the fixed-effect coefficients. A user-friendly interface for tests of
contrasts with outputs either as a summary-like table of t-tests or an
anova-like table of F-tests (or a list of either). Contrasts can
optionally be tested for estimability. Contrasts are allowed to be
rank-deficient as the rank is automatically detected and appropriate
adjustments made. Methods for 'lmerModLmerTest' as well as 'lmerMod'
objects – the latter avoids the Satterthwaite specific computations
when the Kenward-Roger method is used.
5. A show_test() function which operates on anova tables and LS-means
tables (produced by ls_means) makes it possible to see exactly which
functions of the coefficients are being tested. This is very helpful
when differences between type I, II and III anova tables are being
considered and discussed.
6. An ls_means() functions is provided as an alias for lsmeansLT. As
the name implies the function computes the so-called least-squares
means (classical Yates contrasts) as well as pairwise differences of
7. lmerTest::lmer returns an object of class 'lmerModLmerTest'
(previously 'merModLmerTest') to clarify that 'lmerModLmerTest'
extends 'lmerMod' – not 'merMod'. The merMod class includes
generalized and nonlinear mixed models and lmerTest is only designed
for linear mixed models.
8. Test coverage has been greatly improved providing confidence that
lmerTest functionality works as expected even in boundary situations
(e.g. such that anova and summary tables have the expected format even
if there are no fixed effects - and even if the intercept has been
suppressed as well.)
9. The computational approach is to let lmerTest::lmer compute the
required Hessian and derivatives needed for evaluation of degrees of
freedom and t- and F -tests. Previously these quantities were computed
following calls to anova and summary methods which meant that the
computationally intensive parts had to be evaluated anew with each
summary or anova table; the model even had to be refitted as well.
With the new implementation refitting the model is avoided and the
required Hessian and derivaties have to be computed only once per
Changes to the user interface:
The user interface has been updated with new functionality and extra
features as described above. We have tried our best to keep the new
version backward compatible, but a few things from the old API have
been changed out of necessity:
1. In step() the argument type is now being ignored as drop1() is
always used for reduction of the fixed-effect structure. type used to
indicate the type of anova table to use for tests of fixed-effect
terms, but since it only makes sense to remove marginal terms and
drop1() provides the test of these terms it does not make sense to use
anything but drop1() to test the fixed-effect terms. Furthermore type
II and III anova tables provide identical tests of marginal terms.
Additionally, the arguments fixed.calc, lsmeans.calc,
difflsmeans.calc, and test.effs are deprecated and attempts to set
them leads to a warning; also keep.effs has been reduced to keep.
2. The documented behavior of rand/ranova has not changed, but being a
new implementation it may behave differently in ‘corner’ cases.
3. anova tables have a column 'F value' (previously 'F.value') being
consistent with anova.lm and anova.merMod.
4. The headers for summary and anova tables have been modernized.
1. lmerTest::lmer produces an object of (S4) class 'lmerModLmerTest'
(previously 'merModLmerTest') which extends the 'lmerMod' class -
objects of class 'lmerMod' are produced by lme4::lmer.
2. anova and summary methods for objects of class 'lmerModLmerTest'
are S3 and should only be called using method dispatch, i.e. they
should be called with objects of class 'lmerModLmerTest' as the first
argument. If you have an object of class 'lmerMod' and want to compute
p-values use anova(as_lmerModLmerTest(object)) or
summary(as_lmerModLmerTest(object)). (The previous lmerTest package
defined S4 anova and summary methods.)
Rune Haubo B. Christensen
for the lmerTest authors:
Alexandra Kuznetsova, Per Bruun Brockhoff, and Rune Haubo Bojesen Christensen
More information about the R-sig-mixed-models