extract specific values from lmer summary - R-SIG-mixed-models

Sun, May 4, 2014 5:11 PM #

* yes, in general these sorts of question should go directly to
r-sig-mixed-models at r-project.org

 * it's mildly surprising but not shocking that this format doesn't work
any more.

 * without testing, here are a few idioms that should work better:


for(i in 1:nlevels(vital)) {
   lmer.sheep <- glmer(y ~ (1|year),
      family = binomial,
      data = subset(sheep.dat,vital==levels(vital)[i])
}

or

for(i in 1:nlevels(vital)) {
   lmer.sheep <- glmer(y ~ (1|year),
      family = binomial,
      data = sheep.dat,
      subset = vital==levels(vital)[i])
}

 or

model.list <- vector(nlevels(vital),mode="list")
model.list[[1]] <- glmer(y ~ (1|year),
      family = binomial,
      data = sheep.dat,
      subset = vital==levels(vital)[1])
for (i in 2:nlevels(vital)) {
  model.list[[i]] <- update(model.list[[1]],
       subset = vital==levels(vital)[i]
}

  etc.





-------- Original Message --------
Subject: Re: extract specific values from lmer summary
Date: Sun, 04 May 2014 15:25:42 -0800
From: Laura Prugh <lprugh at alaska.edu>
To: Ben Bolker <bbolker at gmail.com>

Hi Ben,
Sorry to bother you again (please let me know if you prefer I post these
questions to a list?)--I updated my version of R to the newest and reran
the code, which had run fine on the older version. It's now giving me an
error referring to how I'm trying to get it to loop through my
dataset--my code is:
for(i in 1:nlevels(vital)) {
lmer.sheep <- glmer(y[vital==levels(vital)[i]] ~
(1|year[vital==levels(vital)[i]]), family = binomial, data = sheep.dat)
}
and it gives the error:
Error in FUN(X[[1L]], ...) :
  Invalid grouping factor specification, year[vital == levels(vital)[i]]

When I put "year[vital == levels(vital)[i]]" directly into R, it appears
to work properly, and this syntax worked yesterday on the older R
version. Is a new syntax required or could this perhaps be a bug?

thanks,
Laura

On 5/2/2014 1:18 PM, Ben Bolker wrote:

On 14-05-02 03:53 PM, Laura Prugh wrote:

Hi Ben,
I am trying to save specific outputs from the lmer summary, and I was
hoping you might be able to tell me the code to do this. I'm running a
random effects model with only the intercept as a fixed effect. I want
to extract the fixed effect intercept estimate and the variance (or
standard deviation) of the random effect. I seem to have figured out how
to get the fixed effect (model at fixedef) but have had no luck getting the
random effect. I guessed a variety of things and model at ranef returned
something, but it didn't match anything in the summary output. Here is
the output of my model, and I would like to extract and save the st dev
of the random effect (0.21287). Any help would be greatly appreciated.
If you generally know of a resource for figuring out the syntax to
extract specific values from summary outputs, that would be great
because this is a recurring problem for me! By the way, I plan to have
as many of my students at possible attend the workshop you're offering
at Chena HS near Fairbanks in August.
cheers,
Laura

    This is really more of an r-sig-mixed-models at r-project.org question:
I'm forwarding the answer there.

   methods(class="merMod") is a good way to check the available accessor
methods for (g)lmer fits, as is ?getME (an lme4-specific function for
getting at lower-level components of a (g)lmer fit in a safe/stable way).

lmer.sheep <- lmer(y ~ (1|year),family=binomial, data = coy,

+ control=list(msVerbose=F,maxIter=1000,maxFN=3000))

lmer.sheep

Generalized linear mixed model fit by the Laplace approximation
Formula: y ~ (1 | year)
Data: coy
Fixed effects:
Estimate Std. Error z value Pr(>|z|)
(Intercept) -0.8083 0.2168 -3.729 0.000192 ***
---
Signif. codes: 0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1

mn <- lmer.sheep at fixef
mn

(Intercept)
-0.8083453

lmer.sheep at ranef

[1] 0.09490052 0.09490052 0.03168227 -0.01783168 -0.14832171 -0.04832885

    A reproducible example would be slightly better, but here I believe
you're looking for

   fixef(lmer.sheep)[1]  ## intercept
   sqrt(unlist(VarCorr(lmer.sheep))  ## std dev of year effect

   It looks like you're still using an older version of lme4: I believe
the incantations I gave above are backward compatible, but there are a
fair number of convenience methods in the newer (>1.0) version that
might be helpful (e.g. as.data.frame(VarCorr(lmer.sheep)))

   Ben Bolker