don't overwrite global (g)lmer() function

[ashenkin]

Hi there,

Recent behavior in afex seems to have changed. I need my lmer models to be class lme4:lmer (stargazer, for example, chokes on lmerTest::lmer objects). However, when i set afex_options(lmer_function = "lme4"), mixed() fails. Rewriting my code to use lme4::lmer instead of just lmer is pretty ugly. Would it be possible to allow options to be separately set for the global lmer function and the lmer afex uses internally separately? If there's a better solution, I'm all ears...

Thanks for the nice package, by the way.

Allie

[singmann]

That looks like a sensible request and bug report all in once. I will try to get back to you with a solution over the weekend as I will be at a workshop beginning of next week. Please remind me should you not hear from me within the next week.

[singmann]

Unfortunately, there is no easy way to change that afex reexports lmerTest::lmer per default. At least I am not aware of a way to alter what happens when a package is attached.

However, a simple way to circumvent the issue is to simply bind lme4::lmer to lmer in the global environment:

require(afex)

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

stargazer::stargazer(fm1)
# Error in objects[[i]]$zelig.call : 
#   $ operator not defined for this S4 class

lmer <- lme4::lmer
fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
stargazer::stargazer(fm1) # works

However, the fact that afex_options(lmer_function = "lme4") leads mixed to fail is not nice, but also unavoidable for methods 'KR' and 'S'. I think this might perhaps need a better error message.

afex_options(lmer_function = "lme4")
data(md_15.1)

t15.4a <- mixed(iq ~ timecat + (1+time|id),data=md_15.1) #default method 'KR' and 'S' fail
t15.4a <- mixed(iq ~ timecat + (1+time|id),data=md_15.1, method = "nested-KR") # still works

Please let me know if this solves the problem for you.

[ashenkin]

Thanks @singmann. I had considered overwriting lmer as you suggest, but it then breaks mixed, which is expecting lmer to point to lmerTest:lmer.

> library(afex)
> fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
> lmer <- lme4::lmer
> mixed(fm1)
Formula (the first argument) converted to formula.
Error in model.matrix.default(rh2, data = data) : 
  argument "data" is missing, with no default

The overwriting solution can be done in a number of ways (e.g. refactor all code to specify lmer4::lmer, wrap each call to mixed with two calls to afex_options, etc...). I'm wondering though: what about refactoring all calls within afex to specify lmerTest::lmer? Or rather, make the call within afex dynamic based on afex_options(lmer_function)... So, maybe one way to do it would be to, for example, define afexlmer <- lmerTest::lmer or afexlmer <- lme4::lmer depending on the option... That basically shifts the burden to you (not my intention), but I think it also kind of accomplishes what a package more or less is supposed to do: perform stuff with as little effect on the global namespace as possible.

Taking a quick look at the code, this is what you seem to do already in mixed:

if (afex_options("lmer_function") == "lmerTest") mf[[1]] <- quote(lmerTest::lmer)
else if (afex_options("lmer_function") == "lme4") mf[[1]] <- quote(lme4::lmer)

So I'm not exactly sure where the global call to lmer is getting overwritten, and I'm not sure of the need for that overwrite; perhaps it's happening in the order of package loading?

[singmann]

Your example does not work, because data is missing. If you correctly pass the data argument, everything works fine.

library(afex)
fm1 <- lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
lmer <- lme4::lmer
mixed(fm1, sleepstudy)
# Mixed Model Anova Table (Type 3 tests, KR-method)
# 
# Model: Reaction ~ Days + (Days | Subject)
# Data: sleepstudy
#   Effect    df         F p.value
# 1   Days 1, 17 45.85 ***  <.0001
# ---
#   Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘+’ 0.1 ‘ ’ 1

Please note that I do not suggest to use mixed in this way. I highly suggest to use:
mixed(Reaction ~ Days + (Days | Subject), sleepstudy)

And as you correctly assume, all calls within mixed use the value of afex_options(lmer_function). lmer is never called outside of this. And even if, this would not be affected.

More precisely, due to the NAMESPACE mechanism a function in the global environment (.GlobalEnv) to which you assign with lmer <- lme4::lmer cannot really override what is called inside a function in a package. Functions in packages are either a directly imported from a package or are called using package::function. This is true at least for all packages on CRAN. This means whatever happens in the global environment has generally no effect for code inside packages.