[R-sig-ME] Plotting post-fitting inference effects in a glmmTMB model

[Fox, John]

Dear Ben, Daniel, and Julian,

A reproducible example would have been nice, but it's easy to create one:

-------- snip ---------

> library("glmmTMB")
> example("glmmTMB")

. . .

> library("effects")
Loading required package: carData
lattice theme set by effectsTheme()
See ?effectsTheme for details.
> ef <- Effect("sd", m0) # sd is character
> plot(ef) # error
Error in UseMethod("droplevels") : 
  no applicable method for 'droplevels' applied to an object of class "character"
> 
> dat$sd <- as.factor(dat$sd)
> mf0 <- update(m0)
> ef0 <- Effect("sd", mf0)
> plot(ef0) # works

And as Ben suggests, the problem isn't unique to glmmTMB models.

The temporary fix, as in the example above, is to change the character predictor to a factor. That shouldn't be necessary: Until the current version, Effect() didn't accommodate character predictors. Now that it does, plot.eff() should work with them too and it doesn't. I'll investigate the bug and fix it.

Best,
 John


  -------------------------------------------------
  John Fox, Professor Emeritus
  McMaster University
  Hamilton, Ontario, Canada
  Web: http::/socserv.mcmaster.ca/jfox

> On Aug 27, 2019, at 9:31 AM, Ben Bolker <bbolker using gmail.com> wrote:
> 
> 
>  Yes, quite weird as effects:::plot.eff (the relevant method) appears
> to check that variables are factors every time it tries to applies
> droplevels().  A reproducible example would be great (although
> technically maybe not this list's problem, as it's not clear it's
> directly mixed-model-related?)
> 
> On 2019-08-27 8:06 a.m., Daniel Lüdecke wrote:
>> Dear Julian,
>> 
>> without a reproducible example, it's a bit difficult to guess how to solve
>> your issue. Is one of your predictors of class "character" and should
>> probably be coerced to factor before you fit your model?
>> 
>> Else, you could also try the "ggeffects" package
>> (https://strengejacke.github.io/ggeffects/), where you have three functions:
>> ggpredict(), ggeffect() and ggemmeans(), each calling the related functions
>> "predict()", "effects::effect()" or "emmeans::emmeans()" internally to
>> compute marginal effects. There is a plot() method as well, based on ggplot2
>> (see examples here
>> https://strengejacke.github.io/ggeffects/articles/introduction_plotcustomize
>> .html or here
>> https://strengejacke.github.io/ggeffects/articles/introduction_plotmethod.ht
>> ml). 
>> 
>> Best
>> Daniel
>> 
>> -----Ursprüngliche Nachricht-----
>> Von: R-sig-mixed-models [mailto:r-sig-mixed-models-bounces using r-project.org] Im
>> Auftrag von Julian Gaviria Lopez
>> Gesendet: Dienstag, 27. August 2019 12:01
>> An: r-sig-mixed-models using r-project.org
>> Betreff: [R-sig-ME] Plotting post-fitting inference effects in a glmmTMB
>> model
>> 
>> Hello,
>> 
>> 
>> I have the next model:
>> 
>>> zipoisson2  <- glmmTMB(Observations ~ CAP * Condition + (1|ID), 
>>> contrasts=list(CAP="contr.sum",Condition="contr.sum"), data=mDATA, 
>>> ziformula=~ 1 , family=poisson)
>> 
>> 
>> Plotting the effects ("effects" package), I obtain:
>> 
>> (ae <- allEffects(zipoisson2))
>> 
>> model: Observations ~ CAP * Condition
>> CAP*Condition effect
>>        Condition
>> CAP        aapaff      aff      neu     pneu
>>  apreC3 3.222636 2.731724 2.777021 2.573694
>>  insC5  1.406698 3.173030 2.104875 1.420336
>>  preC1  2.883115 3.641252 2.541221 2.635991
>>  preC5  2.184075 2.310564 3.137734 1.995430
>> 
>> plot(ae)
>> Error in UseMethod("droplevels") :
>>  no applicable method for 'droplevels' applied to an object of class
>> "character"
>> 
>> Question 1: Has anyone encountered the same issue? Any hint to solve it?
>> 
>> Question 2: Maybe is there any other package works better with glmmTMB model
>> 
>> Thanks in advance for any hint about it.
>> 
>> Julian,
>> 
>> 	[[alternative HTML version deleted]]
>> 
>> _______________________________________________
>> R-sig-mixed-models using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> 
>> --
>> 
>> _____________________________________________________________________
>> 
>> Universitätsklinikum Hamburg-Eppendorf; Körperschaft des öffentlichen Rechts; Gerichtsstand: Hamburg | www.uke.de
>> Vorstandsmitglieder: Prof. Dr. Burkhard Göke (Vorsitzender), Prof. Dr. Dr. Uwe Koch-Gromus, Joachim Prölß, Marya Verdel
>> _____________________________________________________________________
>> 
>> SAVE PAPER - THINK BEFORE PRINTING
>> 
>> _______________________________________________
>> R-sig-mixed-models using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> 
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models