jamovi

Hello everyone,

Im grateful for any help.

In my experiment I used a repeated measures design with a categorical outcome variable (Yes/No). This outcome variable was measured 300 times in each participant.

Participants were randomly assigned to the condition Group "ingroup" or "outgroup" and a positive and a negative wordlist (Valence). So all independent variables were categorical, too and every participant saw 300 combinations of a group stimulus combined with a word of the two valences.

Is it right to use a general mixed effect model here? I'd use Yes/No as a dummy-coded dependent variable. Valence and Group would be my fixed effect predictors. My cluster variable is participant (30 groups).

Can the cluster variable "Participant" (30 groups) account for the repeated measures design?

And is it right to let the intercept be random if i use categorical outcomes? My main question is: Is this the random efffect?

I can also provide the model I've assumed so far, still I am not sure about the random effect here.

Agreement ~ 1 + Group + Valence + (1 I Participant)

Any help is very welcome.
I hope you all have a great day!
Sharon

Hi
what you did seems quite alright. The participants are the clusters, and you need (at least) the intercept to be random. That accounts for the repeated measures. Maybe, you can also add the interaction fixed effect.

Thank you so much! Your advice was really helpful. I've got one question remaining regarding the random intercept.

Should the intercept of the Valence / the Group be random - to account for the repeated measures?
Or should the intercept of the Participants be random, as they are the cluster variable, where the repeated measures are nested actually?

I'm not sure, because either way, a random intercept would mean 0 or 1.

Kind regards,
and thanks so much already.

Sharon

Hi
the intercept of Valence (or group) does not exist in this model: intercepts are expected values for all independent variables equal to zero (the center of the sample for centered categorical independent variable). Maybe you meant whether the effect of Valence (or group) could be random. Given your design, the answer is yes. You can try and check if there's variability (variance greater than 0) for the effects of the IV