lme4 (version 1.0-6)

profile-methods: Profile method for merMod objects

Description

Methods for profile() of [ng]lmer fitted models

Usage

## S3 method for class 'merMod':
profile(fitted, which = 1:nptot, alphamax = 0.01,
	maxpts = 100, delta = cutoff/8, verbose = 0, devtol = 1e-09,
        maxmult = 10, startmethod = "prev", optimizer = "bobyqa",
	signames = TRUE, ...)
## S3 method for class 'thpr':
as.data.frame(x, ...)

Arguments

fitted

a fitted model, e.g., the result of lmer(..).

which

integer indicating which parameters to profile: default is all parameters. The parameters are ordered as follows: [object Object],[object Object],[object Object]

alphamax

maximum alpha value for likelihood ratio confidence regions; used to establish the range of values to be profiled.

maxpts

maximum number of points (in each direction, for each parameter) to evaluate in attempting to construct the profile.

delta

stepping scale for deciding on next point to profile.

verbose

level of output from internal calculations.

devtol

tolerance for fitted deviances less than baseline (supposedly minimum) deviance.

maxmult

maximum multiplier of the original step size allowed, defaults to 10.

startmethod

method for picking starting conditions for optimization (STUB).

optimizer

(character or function) optimizer to use (see lmer for details).

signames

logical indicating if abbreviated names of the form .sigNN should be used; otherwise, names are more meaningful (but longer) of the form (sd|cor)_(effects)|(group). Note that some code for profile transformations (e.

...

potential further arguments for profile methods.

x

the result of profile (or very similar structure)

Details

Methods for function profile (package stats), here for profiling (fitted) mixed effect models.

See Also

For (more expensive) alternative confidence intervals: bootMer.

Examples

fm01ML <- lmer(Yield ~ 1|Batch, Dyestuff, REML = FALSE)
system.time( tpr  <- profile(fm01ML, optimizer="Nelder_Mead") )
## ~2.6s (on a 2010 Macbook Pro)
system.time( tpr  <- profile(fm01ML))
## ~1s, + possible warning about bobyqa convergence
(confint(tpr) -> CIpr)
stopifnot(all.equal(CIpr,
  array(c(12.1985292, 38.2299848, 1486.4515,
          84.0630513, 67.6576964, 1568.54849), dim = 3:2,
        dimnames = list(c(".sig01", ".sigma", "(Intercept)"),
                        c("2.5 %", "97.5 %"))),
                    tol= 1e-07))# 1.37e-9 {64b}
xyplot(tpr)
densityplot(tpr, main="densityplot( profile(lmer(..)) )")
splom(tpr)
## Batch and residual variance only
system.time(tpr2 <- profile(fm01ML, which=1:2, optimizer="Nelder_Mead"))
## GLMM example (running time ~11 seconds on a modern machine)
gm1 <- glmer(cbind(incidence, size - incidence) ~ period + (1 | herd),
            data = cbpp, family = binomial)
system.time(pr4 <- profile(gm1))
xyplot(pr4,layout=c(5,1),as.table=TRUE)
splom(pr4)

Run the code above in your browser using DataCamp Workspace