Profile for merMod objects (lme4)

问题

I am not understanding how does profile work in lmer ? Sometimes it gives exactly same number of values as total number of observations, and sometimes fewer or higher than total number of observations. Also what is .zeta in the output of profile?

(2) Again

pp <- profile(fitted,"theta_",quiet=TRUE)#fitted is a fitted model

is giving values for every random and fixed effects, but

cc <- confint(pp)

is producing confidence interval for only variance-components. Why ?

In ?profile documentation, I didn't get the option quiet. How does quiet work ?

(3) Is there any advantage to run the command

confint(profile(fitted))

rather than running confint(fitted) .

Thanks in advance.

回答1:

To give a reproducible example:

library("lme4")
fm1 <- lmer(Reaction~Days+(Days|Subject),sleepstudy)

Sometimes it [profile] gives exactly same number of values as total number of observations, and sometimes fewer or higher than total number of observations.

I'm not sure what that means.

Also what is .zeta in the output of profile?

The .zeta column is the signed square root of the difference from the minimum deviance.

Specifying no value for which (profile) or parm (confint) profiles/gives CIs for all parameters; specifying "theta_" only gives results for random effects variance-covariance and residual variance (if any).

## all parameters (random and fixed)
pp0 <- profile(fm1)                 
levels(pp0$.par)
## [1] "Days"        "(Intercept)" ".sig01"      ".sig02"      ".sig03"     
## [6] ".sigma"     
# random effects only
pp1 <- profile(fm1,which="theta_")
levels(pp1$.par)
## [1] ".sig01" ".sig02" ".sig03" ".sigma"

Similarly for confint():

## all parameters
cc0 <- confint(pp0)
rownames(cc0)
## [1] ".sig01"      ".sig02"      ".sig03"      ".sigma"      "(Intercept)"
## [6] "Days"
## random-effects parameters only
cc1 <- confint(pp1)
## [1] ".sig01" ".sig02" ".sig03" ".sigma"

The quiet parameter is for confint() (not profile()). By default, applying confint() directly to a fitted model issues a message

Computing profile confidence intervals ...

to warn the user that a computationally intensive/slow process is taking place. Using quiet=TRUE suppresses this message.

The Note in ?confint.merMod says:

The default method ‘"profile"’ amounts to confint(profile(object, which=parm), signames=oldNames, ...), level, zeta) where the ‘profile’ method ‘profile.merMod’ does almost all the computations. Therefore it is typically advisable to store the profile(.) result, say in ‘pp’, and then use ‘confint(pp, level=*)’ e.g., for different levels.

In other words, if you are going to want to do anything else with the profile information (plot profiles, or compute confidence intervals for more than one alpha-level) it is more efficient to compute the profile and store it, rather than repeatedly computing the profile. If all you want are the default 95% confidence intervals then you might as well use confint(fitted_model).

来源：https://stackoverflow.com/questions/31944611/profile-for-mermod-objects-lme4

标签

lme4