I have three independent variables (all binary) and two interaction variables that I would like to test for their effect on the dependent variable "Awareness_level". This dependent variable is also binary.
I assigned participants randomly to one of 8 different conditions. In the assigned condition they were exposed to 1 of 8 advertisements (advertisements were varied based on different combinations of the 3 independent variables). All participants were asked the same questions to assess awareness: one question to assess recall and one question to assess recognition.
I restructured the data in SPSS so that for each subject there are two rows of data. The first row (index variable awareness_type is 1) shows recall and the second row (index variable awareness_type is 2) shows recognition. I created the variable awareness_level as a target variable. See picture (might be more clear).
To add an intercept to the model that is different for every participant, I choose to fit a binomial logistic mixed model by running a Generalized linear mixed model with a binomial regression in SPSS.
I think I should add all the independent variables (used to create the different conditions) + the interaction terms as fixed factors, while adding a random effect for subject ID.
Should I also add the index variable "awareness_type" as a independent variable?
I used this manual from SPSS as a basis (although it is for multinomial logistic mixed model) http://pic.dhe.ibm.com/infocenter/spssstat/v20r0m0/index.jsp?topic=%2Fcom.ibm.spss.statistics.cs%2Fglmm_cablesurvey_howto.htm
They interpret the fit of the model by looking at the accuracy / classification table. But if I'm correct you should compare the accuracy rate to the proportional by chance accuracy rate right? How do I calculate this one?
And wouldn't it be better to assess the model fit based on the information criterion instead of accuracy rate? And which one is better than, Bayesian or Akaike corrected?
Would be great if you can help me!!