I have Change from Pretest to Posttest (gain, no_gain, decline) as the DV.
Pretest and Group as covariates. This called for a multinomial regression:
mod0 <- brm(Change ~ Pretest + Group)
I’d like to add random effects of Subject and Word which might differ from pretest to posttest, but I don’t have effect of Time to do:
mod1 <- brm(Change ~ Pretest + Group + (Time|Subject) + (Time|Word))
So I thought of:
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?
When you use the change as response, there is indeed no “time” variable anymore. However, if you use the raw values of pre and posttest, you may include time both as an interaction with group and as a varying slope if you desire.
I chose to do ANCOVA (pretest as covariate) as suggested by Vickers & Altman (2001) click link. Indeed, following this model will eliminate Time variable.
Does your response intend that I do (1|subjects) , please?
If you have just one observation per person,
(1|subject) can be problematic in most (but not all) families. Which family are you using? If you just have one observation and use a normal distribution, you have to remove this term.
I have 28 words tested on every participant. So for every participant I have 28 responses in the DV, I am using categorical family.
brm(Change ~ Pretest + Group + PoS + (1|Subject) + (1| Word),
data = wf.data,
family = 'categorical',
prior = prior(normal(0, 5), class = "b"),
control = list(max_treedepth = 15))
in this case I would definitely include (1 | subject)
Thank you. I had a statistician asking me to do the hierarchy as such :
(1|Subject) + (1+Group|Word) but it does not appear accurate to me. Do you have any comments on this, please? If not that’s fine.
(1+group | word) could be very reasonable if you assume the effect of group to vary across words (which seems plausible).