[R-sig-ME] random factor of type integer

[Douglas Bates]

On Jan 29, 2008 8:26 AM, Henrik Parn <henrik.parn at bio.ntnu.no> wrote:
> Dear all,

> I have a dataset where the variables that I later will use as random
> grouping factors in a lmer-model are integers. Do I first need to
> convert them to factors?

> I just tried if the example on the help page for lmer (where the
> grouping factor 'Subject' actually is a factor)...

> str(sleepstudy$Subject)
> fm1 <- lmer(Reaction ~ Days + (Days|Subject), sleepstudy)
> summary(fm1)

> ...differed from:

> Subject2 <- as.integer(as.character(sleepstudy$Subject))
> fm2 <- lmer(Reaction ~ Days + (Days|Subject2), sleepstudy)
> summary(fm2)

> ...but it did not.

The short answer is that you do not need to have grouping factors
represented as factors for lmer, but it is advisable.

Internally the variables used as grouping factors for random effects
are converted to factors before being incorporated into the flist
slot.

> Is this a general case or could the use of a random factor of integer
> type may cause problems in lmer?