So I have a binary dependent variable (0 and 1). Participants were randomly assigned to one of three conditions (1, 2, 3) and read one of 4 different texts (m, o, t, s). There are about 35 data points for each participant (one for each sentence they read of the text).
I did a mixed model in SPSS with ID as subjects and sentence as repeated using binary logistic regression. Condition and text were fixed factors.
According to the output, there is an interaction between condition and text. However, the estimated means don't seem right to me! I realize they will be different from the actual means, but it almost looks like the conditions were switched around.
Bargraphs of the actual means (just averaging without any modeling or anything fancy) and the estimated means are attached. Notice that in the bar graphs conditions 1 and 2 look higher for text m, but the opposite is the case for the estimated means.