[R-sig-ME] Power calculation three way interaction

Clemens von Wulffen c|emen@@wu|||en @end|ng |rom gm@||@com
Sat Oct 30 14:47:23 CEST 2021

Hello Team,

I was wondering if anyone here has an answer to my question:

I am conducting a *power calculation* to find the required sample size
for *binomial
GLMER *study.
Short description of experiment: Participants rate *120 statements* as *true
or false.* Half of the statements have been repeated previously (in an
exposure task). The subjects are placed into three groups (*1 control and 2
treatments* groups called: affect reattribution and fluency reattribution).
These groups also get additional information during the second phase (the
second 60 statements to be judged from the 120 overall). We are trying to
see now *if the varying additional information* given to the two
treatment groups* changes the likelihood to rate a statement as true in the
second phase.*

I have gained parameters for my simulation from a mixture of pilot data and
previous study (since pilot data only has a results for one group).

One of the aims of the study is to get results for a *three way interaction
between status of repetition (new or repeated), judgement phase (1 or 2),
and condition (group control, af-reat, and flu-reat). *

I was wondering now how and *if I need to specify the three way interaction
in any specific way for my power calculation other than simply plugging the
variables for IV's and for the random effects? *

glmer(dv ~ fluency_reat * status.t * judge.t +
                  affect_reat * status.t * judge.t  +
                  (1 | subj) + (1 | item) , data = data,
             family = binomial(link = "logit"))

for clarification: since there are three condition groups --> the formula
gets repeated for each treatment group.

I have my code attached to this email. I personally thought that I have
specified everything and can just run it, but my supervisor said I need to
specify the three way interaction previously. I just am  not sure how to.

If anyone has had this question before and has some thoughts on it, I would
highly appreciate it.



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