When I run the test without an interaction term, my results look good, but when I add an interaction term I get a very long (and incorrect looking) odds ratio (Exp(B)) 217850,716. Can someone tell me what I am doing wrong please?
I do not have any missing values
My variables are
- Condition: separate = 0, combination = 1
- Gender: female = 0, male = 1
- Target presence: TP = 0, TA =1
- Choice: non-choosing = 0, choosing = 1
I made all my variables categorical (as they are all binary), and I set the reference category as 0 for gender and choice, and 1 for condition and target presence.
Hello Nastasia,
Actually, you could have left each of the independent variables as a "continuous" covariate since each of the IVs has only two levels. It shouldn't make any difference, however (though SPSS can flip categories of the IV such that the reference category is opposite from what you imagine when you identify an IV as categorical in eliciting the analysis).
What does the large result for the interaction term mean? Most likely, that when choose = 1 AND target_pres = 1, there is a substantially higher likelihood of the case being in the target category on the DV, compared to the other three combinations (e.g., choose = 1, target_pres = 0; choose = 0, target_pres = 1; or choose = 0, target_pres = 0). You could always have a look at the cell frequencies for these four combinations as one dimension of a cross-tabulation table with the levels of the DV as the other dimension, to help you better understand this.
(Of course, if SPSS has flipped the 0/1 categories for the interaction term, then the 1,1 case would have lower estimated likelihood of being in the referent DV category compared to the other 3 options: 0/1, 1/0, 0/0.)
Good luck with your work.
As we can infer from Prof Morse answer this is yet another reason to not use SPSS. Best wishes David Booth
Hello Nastasia Filipovic. SPSS is perfectly capable of estimating your model. However, you might find it more convenient to estimate your model via the GENLIN command (rather than LOGISTIC REGRESSION). As this Analysis Factor page notes, GENLIN allows you to generate tables of estimated marginal means (EMMEANS), which can help immensely in interpreting interactions.
- https://www.theanalysisfactor.com/spss-procedures-logistic-regression/
HTH.
- Badges
- Science topic