R: Bootstrapping with multilevel model

Question

I am looking to calculate 95% confidence intervals around my multilevel model coefficient estimates.

I have no trouble for models with a single grouping variable, but the bootstrapping method I was following (http://www.ats.ucla.edu/stat/r/dae/melogit.htm) essentially broke when I added in an extra grouping variable.

I investigated bootMer (a newly implemented part of lme4), with the same outcome.

Here's an example of the problem:

---------------------------- EDITED to include a reproducible example ----------------------------

The helpful comments to this question and working through the example gave the answer - it is not the addition of a second grouping variable per se, but missingness in a grouping variable that causes the issue.

Here is a worked example for anyone else who runs into this.

Here is a simple function for the sake of illustration...

FUN <- function(fit) {return(fixef(fit))} 

Example data (complete)

grouper1 <- as.factor(sample(letters[1:20], 1000, replace = TRUE))
grouper2 <- sample(letters[1:2], 1000, replace = TRUE)
DV<-rnorm(1000)
IV<-rnorm(1000)
example<-data.frame(grouper1, grouper2, DV, IV)

Works fine with this data

one_grouper<-lmer(DV ~ IV + (1 | grouper1), data=example)  

> bootMer(one_grouper,FUN, nsim=1)

Call:
bootMer(x = one_grouper, FUN = FUN, nsim = 1)

Bootstrap Statistics :
       original       bias    std. error
t1* 0.005286026  0.041665542          NA
t2* 0.009642498 -0.003707219          NA
> 
> two_grouper<-lmer(DV ~ IV + (1 | grouper1) + (1 | grouper2), data=example)
> 
> bootMer(one_grouper,FUN, nsim=1)

Call:
bootMer(x = one_grouper, FUN = FUN, nsim = 1)

Bootstrap Statistics :
       original      bias    std. error
t1* 0.005286026 -0.03465914          NA
t2* 0.009642498 -0.01361108          NA

BUT, when we introduce missingness in a grouping variable...

example$missinggroups <- with(example, ifelse(randommissing=="f", NA,grouper1))

> one_grouper<-lmer(DV ~ IV + (1 | missinggroups ), data=example)  
> 
> bootMer(one_grouper,FUN, nsim=1)

Call:
bootMer(x = one_grouper, FUN = FUN, nsim = 1)


Bootstrap Statistics :
WARNING: All values of t1* are NA
WARNING: All values of t2* are NA
Warning message:
In bootMer(one_grouper, FUN, nsim = 1) : some bootstrap runs failed (1/1)

Answer 1

This was a confirmed issue and as per the comments the fix worked in the development version as of 08-Jan-2014.

That was back when the version was < 1.1-3; lme4 on CRAN has had versions > 1.1-5 since 14-Mar-2014.