Interaction

Nov 2005
127
4
Glasgow, Scotland
Hello, I have a theoretical question...

I was analyzing a data set using logistic regression. I had over 12 independent variables, so I used a method for variable selection ( Collett's algorithm ), which resulted in 1 categorial variable and 1 continous. since I had over 12 variables, I ignored interactions ( since it's not practical to look at them all ). My question is, now that I got a final model with 2 significant IV, should I include their interaction, or ignore it like I did from the start ?
when I include it, the categorial variable stops being sifnificant, only the continous one and the interaction are sifnificant.

what sould I do in this case ?
 
Jan 2009
591
134
Hello, I have a theoretical question...

I was analyzing a data set using logistic regression. I had over 12 independent variables, so I used a method for variable selection ( Collett's algorithm ), which resulted in 1 categorial variable and 1 continous. since I had over 12 variables, I ignored interactions ( since it's not practical to look at them all ). My question is, now that I got a final model with 2 significant IV, should I include their interaction, or ignore it like I did from the start ?
when I include it, the categorial variable stops being sifnificant, only the continous one and the interaction are sifnificant.

what sould I do in this case ?

I think you've answered your question. If the interaction diminishes the a variable, then the interaction becomes more significant.

Just wondering, but with only 12 variables, why is it not practical to look at them all?

If you were doing all the calulations (by hand) without the aid of a computer, then it's understandable -- however, you have internet access, and that implies a personal computer.
 
Nov 2005
127
4
Glasgow, Scotland
Let's assume I have only 12, and not more, the numer of combinations of 2 out of 12, is 66, if I add the combinations of 3,4, ...., I get a crazy number ( 66 is also too much, isn't it ? ).

so I understand that you suggest to include the interaction, even if I ignored interactions the whole way. should I keep the variable that now is not significant ?
 
Nov 2005
127
4
Glasgow, Scotland
I think I need some help interprating the results...

the picture attached is the result of the model containing 1 continous independent variable (A) and one categorial (B) devided to dummy variables. the interaction included.
how do I interprate the results ? especially the interactions ?
I'll appreciate it if you can give me an exmaple line of a conclution...
cheers !
(Speechless)
 

Attachments