Dear All, any ideas would be most appreciated. In a logistic regression predicting outcome (self-harm at 24 months yes or no), we are interested in whether treatment model (A or B) and self-mutilation at intake (present or absent) interact. Is it possible to compute an interaction term based on these two independent dichotomous variables (as one could do in the case of 2 continuous IDs) , as the samples are not very large? If this does not make sense, would the way to go about this be to split the file into say those who self-harmed at intake and those who did not, and look at treatment model as a the only ID? many thanks Ioanna