I am trying to simulate data using parameters from a glmer() model output. The model, which comes from a published paper, is as follows: DV ~ 1 + group* sex *verb type + trial number + (1 |participant) + (1 | stimulus). All three IVs (group, sex and verb type) comprise 2 levels and are deviation coded so that, for instance, female = -0.5 and male = 0.5. The DV is binary (success or failure)
I want to simulate the (log odds) probability of success for each trial for each participant, which I am calculating using the following formula (where group, sex and type are all deviation coded as in the original paper): b_intercept + (b_group * group) + (b_sex * sex) + (b_type * type) + (b_trial * trial_number) + (b_group_sex_type* group* sex* type) + (b_group_sex * group * sex) + (b_group_type * group * type) + (b_type_sex * type * sex) + random effect of participant + random effect of stimulus
But when it comes to the interaction terms in this calculation, the fact that the variables are deviation coded means I can't retrieve the specific value for each group because there are only two possible outputs and, for instance, a female (-0.5) in group 1 (-0.5) gives the same interaction term as a male (0.5) in group 2 (0.5).
In cases such as these, how do you accurately simulate data? Is there a way to incorporate the effects of interactions into your simulated IV?
