Interaction terms in logistic regression

[Imicola]

Hi,

Just a quick question about interaction terms in a logistic regression model. I've been doing a logistic regression in R using backwards stepwise deletion - my final model includes the interaction between two terms (a*b), but does not contain the two terms as individual explanatory variables ( a + b ).

I was just wondering if this is ok, or do the variables in the interaction have to be kept in the model separately aswell?

Thanks,
Nicola

[Rounds]

Generally speaking it is not done. Default to main effects go in if the interaction goes in. The only time that rule seems to get violated is if there is a reason before fitting to just use the interaction. The thinking process is similar to not having an intercept in the linear model.