[R-sig-ME] Random effects in multinomial regression in R?

Phillip Alday ph||||p@@|d@y @end|ng |rom mp|@n|
Thu Mar 21 12:52:16 CET 2019


I honestly don't know because I don't know enough the structure of your
data.

Phillip

On 20/3/19 10:46 pm, Souheyla GHEBGHOUB wrote:
> Dear Philip,
> 
> Thank you for the clarification. I agree. 
> So does your response intend that I should just do (1|Subject) ?
> 
> Thanks again
> Souheyla
> 
> On Wed, 20 Mar 2019, 18:51 Alday, Phillip, <Phillip.Alday using mpi.nl
> <mailto:Phillip.Alday using mpi.nl>> wrote:
> 
>     Please keep the list in CC.
> 
>     I really can't provide more advice about whether to do an
>     intercept-only model or include the Pretest score in the random
>     effects without knowing more about your data. If you have multiple
>     pretest scores per subject and word, then it might make sense to
>     include them in the random effects, *depending on your data and
>     research question*. If you don't, then it definitely doesn't make
>     sense to estimate a slope (i.e a rate) from a single static
>     observation.
> 
>     Phillip
> 
>     On 20/3/19 7:33 pm, Souheyla GHEBGHOUB wrote:
>>     Hi again Phillip, 
>>
>>     My question is  :  I'd like to add random effects
>>     of */Subject/* and */Word/*, which may differ by time from pretest
>>     to posttest, but I don't have effect of */Time/* , so I can't do:
>>     |mod1 <-brm(Change~Pretest+Group+(Time|Subject)+(Time|Word)) |So should I just do (1|Subject)+(1|Word)) or
>>     (Pretest|Subject)+(Pretest|Word)) or exclude random effects?
>>     Thank you for looking into this :)
>>     Souheyla
>>     ||
>>
>>     On Wed, 20 Mar 2019 at 18:28, Phillip Alday <phillip.alday using mpi.nl
>>     <mailto:phillip.alday using mpi.nl>> wrote:
>>
>>         On 20/3/19 6:39 pm, Souheyla GHEBGHOUB wrote:
>>         > Hi Philip, 
>>         >
>>         > Thank you for the clarification. But I might have not make
>>         it clear in
>>         > my question.
>>         >
>>         > I don't have Time in my data at all because I chose to
>>         predict change
>>         > rather than having posttest and pretest responses as DV and
>>         Time as
>>         > fixed effect.
>>
>>         If Time isn't in your difference data, then it really makes no
>>         sense to
>>         have it in your model anywhere ....
>>
>>         > I chose this way because I have groups of subjects who were
>>         tested on
>>         > words, and I was not too sure whether, a simple regression with
>>         > Responses as DV and Time (Pretest/Posttest) as IV , will
>>         take into
>>         > account differences between Pretest and Posttest at the
>>         level of each
>>         > word. That is, I don't know whether it will sum the overall
>>         pretest
>>         > score of each subject then compare it to its posttest, while
>>         I want it
>>         > to compare each subject score of each word from pretest to
>>         posttest then
>>         > base its analysis on these score changes.
>>
>>         I don't want to be too harsh, but if you were unsure about
>>         that, then
>>         that's the question you should have asked first. (See also the XY
>>         problem, https://en.wikipedia.org/wiki/XY_problem)
>>
>>         >
>>         > That's why I did not want to risk it and chose /score
>>         change/ as the DV
>>         > instead. But I was faced with another problem which is
>>         absence of Time
>>         > effect by which subjects differ for my random slopes?
>>
>>         Assuming you want to compute the difference outside of the
>>         model, then
>>         you could (and I would argue should) still use the
>>         continuous/numeric
>>         difference and not a categorical thresholding of that
>>         difference as your
>>         dependent variable.
>>
>>         In that case, I would argue that there can't be a "Time" effect by
>>         subject because you are measuring the difference, which
>>         incorporates the
>>         variance at each Time in the variance of the difference. Same
>>         for word.
>>
>>         Depending on the exact structure of the test and whether there are
>>         multiple pretest scores by subject or by word, you could
>>         potentially
>>         include that as a random slope, but to make a more precise
>>         recommendation there, we need to know more about your data.
>>
>>         Best,
>>         Phillip
>>
>>
>>
>>
>>
>>         >
>>         > Best,
>>         > Souheyla 
>>         >
>>         > On Wed, 20 Mar 2019 at 17:02, Phillip Alday
>>         <phillip.alday using mpi.nl <mailto:phillip.alday using mpi.nl>
>>         > <mailto:phillip.alday using mpi.nl <mailto:phillip.alday using mpi.nl>>>
>>         wrote:
>>         >
>>         >     Generally speaking for the parameterization of
>>         mixed-effects models in
>>         >     lme4/brms/the usual packages, it doesn't make sense to
>>         have a varying
>>         >     slope (e.g. Time|Subject) without the corresponding
>>         fixed effect. This
>>         >     is because the varying slopes are calculated as offsets
>>         from the group
>>         >     mean, i.e from the fixed effect estimate. Not doing
>>         including the fixed
>>         >     effect is equivalent to assuming the group mean is zero,
>>         which is
>>         >     usually not the assumption you want to make.
>>         >
>>         >     If you fit models with random slopes without the
>>         corresponding fixed
>>         >     effects, then there are two main problems:
>>         >
>>         >     1. The corresponding variance parameter will be
>>         mis-estimated because it
>>         >     will be the average squared distance to zero and not the
>>         average squared
>>         >     distance to the mean (and average squared distance to
>>         the mean is the
>>         >     definition of variance).
>>         >
>>         >     2. The model may not converge because the numerics are
>>         set up under the
>>         >     "zero mean" assumption. For lme4/nlme, this is the case,
>>         but I believe
>>         >     that brms may do some internal reparameterization that
>>         may avoid these
>>         >     difficulties. (And a model fit with MCMC (brms) may not
>>         have the same
>>         >     numerical issues as a model fit with MLE (lme4)).
>>         >
>>         >     In brief: just add time as a fixed effect.
>>         >
>>         >     Also: why not fit your model as a continuous model with
>>         pre vs. post as
>>         >     a contrast in the model rather reducing a continuous
>>         variable to a
>>         >     category? You can still apply a categorical distinction
>>         afterwards if
>>         >     you so desire, but in my experience, it's best to defer
>>         making things
>>         >     categorical until as late as possible (see also Frank
>>         Harrel's comments
>>         >     on prediction vs. classification:
>>         >     http://www.fharrell.com/post/classification/). Moreover,
>>         it's a lot
>>         >     easier to fit a continuous model than a multinomial one ....
>>         >
>>         >     Best,
>>         >     Phillip
>>         >
>>         >     On 18/3/19 7:11 pm, Souheyla GHEBGHOUB wrote:
>>         >     > I have *Change* from Pretest to Posttest (gain,
>>         no_gain, decline)
>>         >     as the
>>         >     > DV. Also *Pretest* and *Group* as covariates. This
>>         called for a
>>         >     multinomial
>>         >     > regression:
>>         >     >
>>         >     > mod0 <- brm(Change ~ Pretest + Group)
>>         >     >
>>         >     > *Question: *I'd like to add random effects of
>>         *Subject* and
>>         >     *Word*, which
>>         >     > may differ by time, but I don't have effect of *Time*
>>         to do:
>>         >     >
>>         >     > mod1 <- brm(Change ~ Pretest + Group + (Time|Subject)
>>         + (Time|Word))
>>         >     >
>>         >     > So I thought of this:
>>         >     >
>>         >     > mod2 <- brm(Change ~ Pretest + Group + (1|Subject) +
>>         (1|Word))
>>         >     >
>>         >     > but this also seems wrong to me. What do you think is
>>         the best way
>>         >     to treat
>>         >     > random effects in this situation, please?
>>         >     >
>>         >     > Thank you
>>         >     >
>>         >     > Souheyla Ghebghoub
>>         >     >
>>         >     >       [[alternative HTML version deleted]]
>>         >     >
>>         >     > _______________________________________________
>>         >     > R-sig-mixed-models using r-project.org
>>         <mailto:R-sig-mixed-models using r-project.org>
>>         >     <mailto:R-sig-mixed-models using r-project.org
>>         <mailto:R-sig-mixed-models using r-project.org>> mailing list
>>         >     > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>         >     >
>>         >
>>



More information about the R-sig-mixed-models mailing list