Logit regression interaction/moderation: what to do about it?

[Johann Kotze]

Hi all!

I have a multivariable logit regression model that has two terms showing two-way interaction (the interaction also makes biological sense.)

So what am I to do about it?

Do I just mention it in my findings?
Do I remove one of the variables? (How do I decide which one?)

I understand that interactions weaken the goodness-of-fit of a model. True?

Many thanks!

Johann

 

[hlsmith]

You should keep it in the model along with its components main effects. Are its components main effects categorical or continuous?

 

[Johann Kotze]

Thanks very much! So the point of identifying interactions in logit model is just to learn that they exist?

The study is on septic arthritis. Here are the two variables that interact:

X1 is a categorical variable denoting the period before presentation at a clinic
X2 is a dichotomous variable that denotes whether radiographic abnormalities were present

y is a dichotomous variable denoting return to function

 

[hlsmith]

Don't completely follow how X1 is defined, but it would be typical to present the odds ratio for the modifiable categorical variable in the interaction term at each of the two levels of the other interaction term. So OR for X2 @ X1 = 1 and then at X1 = 0.