Anyone statistics here?
In a binary logistic regression, let's say you have a significant interaction between 2 continuous variables. How would you plot that interaction on a graph? Is such a thing even possible?
Can you give a concrete example of what your RVs are? Also, I thought you couldn't have your dependent RV be continuous in logistic regression?
>>7812506
Not sure about RV, but the dependent variable is a binary outcome (Not using/Using a car), and apart from covariates, the two independent variables are strength of environmental attitudes (scale treated as continuous) and time since moving home (months). Both are independent predictors, and an interaction effect is significant.
>>7812535
Alright, I'm not sure exactly how to do it, but according to http://www.ats.ucla.edu/stat/stata/seminars/interaction_sem/interaction_sem.htm, it seems like you can indeed plot interaction against the values of the covariate (plot at roughly 1 third down the page). Apparently you first need to decide exactly in which way you want to model interaction: in their example, they use difference of differences, but they mention using ratio of odds ratios in a table a bit above that. I guess unless you have some kind of insight about which one to use for your particular problem, you should try them all and picking the one that looks best compared to what you're trying to show.
I can't help much more: I'm decent at probability but I haven't done any statistics for almost a decade so I tend to just use cookbooks.
>>7812565
Ah cool, nice link, thanks man - I'll check it out
>>7812487
a few options
-create a 3d graph and show the combined main effects and interaction
-create bins (for each deciles for instance) for one of your continuous variable and interact the continuous variable with each level of the other.
you can then show the conbined effect for all levels on a 2d graph