lme4: utilities – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

utilities

Print and Summary Method Utilities for Mixed Effects

Description

The print, summary methods (including the print for the summary() result) in lme4 are modular, using about ten small utility functions. Other packages, building on lme4 can use the same utilities for ease of programming and consistency of output.

Notably see the Examples.

llikAIC() extracts the log likelihood, AIC, and related statics from a Fitted LMM.

formatVC() “format()”s the VarCorr matrix of the random effects – for print()ing and show()ing; it is also the “workhorse” of .prt.VC(), and returns a character matrix.

.prt.*() all use cat and print to produce output.

Usage

llikAIC(object, devianceFUN = devCrit, chkREML = TRUE,
        devcomp = object@devcomp)

methTitle(dims)

.prt.methTit(mtit, class)
.prt.family (famL)
.prt.resids (resids, digits, title = "Scaled residuals:", ...)
.prt.call   (call, long = TRUE)
.prt.aictab (aictab, digits = 1)
.prt.grps   (ngrps, nobs)
.prt.warn   (optinfo, summary = FALSE, ...)

.prt.VC (varcor, digits, comp, formatter = format, ...)
formatVC(varcor, digits = max(3, getOption("digits") - 2),
         comp = "Std.Dev.", formatter = format,
         useScale = attr(varcor, "useSc"), ...)

Arguments

`object`	a LMM model fit
`devianceFUN`	the function to be used for computing the deviance; should not be changed for lme4 created objects.
`chkREML`	optional logical indicating if `object` maybe a REML fit.

devcomp

for lme4 always the equivalent of object@devcomp; here a list

dims

for lme4 always the equivalent of object@devcomp$dims, a named vector or list with components "GLMM", "NLMM", "REML", and "nAGQ" of which the first two are logical scalars, and the latter two typically are FALSE or numeric.

`mtit`	the result of `methTitle(object)`
`class`	typically `class(object)`.

famL

a list with components family and link, each a character string; note that standard R family objects can be used directly, as well.

`resids`	numeric vector of model `residuals`.
`digits`	non-negative integer of (significant) digits to print minimally.
`title`	`character` string.
`...`	optional arguments passed on, e.g., to `residuals()`.

`call`	the `call` of the model fit; e.g., available via (generic) function `getCall()`.
`long`	logical indicating if the output may be long, e.g., printing the `control` part of the call if there is one.

aictab

typically the AICtab component of the result of llikAIC().

`varcor`	typically the result of `VarCorr()`.
`comp`	optional ...

formatter

a function used for formatting the numbers.

`ngrps`	integer (vector), typically the result of `ngrps(object)`.
`nobs`	integer; the number of observations, e.g., the result of `nobs`.

`optinfo`	typically `object @ optinfo`, the optimization infos, including warnings if there were.
`summary`	logical

useScale

(logical) whether the parent model estimates a scale parameter

Value

llikAIC() returns a list with components

`logLik`	which is `logLik(object)`, and
`AICtab`	a “table” of `AIC`, `BIC`, `logLik`, deviance and `df.residual()` values.

Examples

## Create a few "lme4 standard" models ------------------------------
fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
fmM <- update(fm1, REML=FALSE) # -> Maximum Likelihood

gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd),
             data = cbpp, family = binomial)
gmA <- update(gm1, nAGQ = 5)


(lA1 <- llikAIC(fm1))
(lAM <- llikAIC(fmM))
(lAg <- llikAIC(gmA))

(m1 <- methTitle(fm1 @ devcomp $ dims))
(mM <- methTitle(fmM @ devcomp $ dims))
(mG <- methTitle(gm1 @ devcomp $ dims))
(mA <- methTitle(gmA @ devcomp $ dims))

.prt.methTit(m1, class(fm1))
.prt.methTit(mA, class(gmA))

.prt.family(gaussian())
.prt.family(binomial())
.prt.family( poisson())

.prt.resids(residuals(fm1), digits = 4)
.prt.resids(residuals(fmM), digits = 2)

.prt.call(getCall(fm1))
.prt.call(getCall(gm1))

.prt.aictab ( lA1 $ AICtab ) # REML
.prt.aictab ( lAM $ AICtab ) # ML --> AIC, BIC, ...

V1 <- VarCorr(fm1)
m <- formatVC(V1)
stopifnot(is.matrix(m), is.character(m), ncol(m) == 4)
print(m, quote = FALSE) ## prints all but the first line of .prt.VC() below:
.prt.VC( V1, digits = 4, formatter = format)
## Random effects:
##  Groups   Name        Std.Dev. Corr
##  Subject  (Intercept) 24.740
##           Days         5.922   0.07
##  Residual             25.592

p1 <- capture.output(V1)
p2 <- capture.output( print(m, quote=FALSE) )
pX <- capture.output( .prt.VC(V1, digits = max(3, getOption("digits")-2)) )
stopifnot(identical(p1, p2),
          identical(p1, pX[-1])) # [-1] : dropping 1st line

.prt.grps(ngrps = ngrps(fm1),
          nobs  = nobs (fm1))
## --> Number of obs: 180, groups:  Subject, 18

.prt.warn(fm1 @ optinfo) # nothing .. had no warnings

lme4

Linear Mixed-Effects Models using 'Eigen' and S4

v1.1-26

GPL (>= 2)

Authors

Douglas Bates [aut] (<https://orcid.org/0000-0001-8316-9503>), Martin Maechler [aut] (<https://orcid.org/0000-0002-8685-9910>), Ben Bolker [aut, cre] (<https://orcid.org/0000-0002-2127-0443>), Steven Walker [aut] (<https://orcid.org/0000-0002-4394-9078>), Rune Haubo Bojesen Christensen [ctb] (<https://orcid.org/0000-0002-4494-3399>), Henrik Singmann [ctb] (<https://orcid.org/0000-0002-4842-3657>), Bin Dai [ctb], Fabian Scheipl [ctb] (<https://orcid.org/0000-0001-8172-3603>), Gabor Grothendieck [ctb], Peter Green [ctb] (<https://orcid.org/0000-0002-0238-9852>), John Fox [ctb], Alexander Bauer [ctb], Pavel N. Krivitsky [ctb, cph] (<https://orcid.org/0000-0002-9101-3362>, shared copyright on simulate.formula)

Initial release