Logistic regression: dummy variables not significant, what next?

[mattdoughty]

Hi,

I'm running a logistic regression to try a predict landslides in SPSS using a categorical variable (vegetation type) with 6 possible values. In the results, the variable as a whole is significant although a couple of the dummy variables have p-values > 0.05 and thus are not significant. What should I do?

I've previously posted to ask whether collapsing classes is correct and was told that I should only do this using sound scientific theory. So, if the two classes in question could theoretically be combined (i.e. broadleaf woodland and pine woodland into one class, woodland), is it sound practice to do this in my case?

Many thanks in advance,

Matt

 

[maartenbuis]

Maybe, maybe not. That really depends on all the details of that study. So there is only one person who can answer that question: you.

 

[noetsi]

There are two ways to approach this. One is to show the dummy variables that are and are not signficant and what this tells us in terms of theory and practice. A second (which I suspect is common, but not reported) is to stress that which is signficant by reclassifying a variable. If something makes theoretical sense I can't see any reason why reclassifying in this way would be wrong. Personally I would show both models, and explain the theoretical and practical reasons for the reclassification.

 

[mattdoughty]

Thanks for the quick replies. I'm going to go ahead a group the classes and like you mentioned, I can explian my reasons in the write up.