1

I am interested in the interaction between two categorical variables that each have four states in a survival experiment.

My model: Surv(time, death) ~ size * state

My very basic question, is do I look at the individual interactions between the levels? For example, size2:state:2 or size2:state3, etc? Or do I use an anova that will combine individual effects into a single pvalue, size:state?

I realize this is very basic, so if someone can give me a simple explanation or point me to an answer I would appreciate it. I am finding with my own data that one can be significant and the other is not, so I am having a hard time interpreting.

darwin
  • 151
  • 3
  • If your categorical vars are ordered in a natural way, you can use orthogonal polynomial contrasts, and a reduced interaction, say only interaction between the linear parts (if sample size is low so you need to use the available df vicely). Without ordering that will not work? – kjetil b halvorsen Sep 14 '15 at 14:12
  • You can see at these related posts: http://stats.stackexchange.com/questions/85393/interpretation-of-interactions-between-categorical-data http://stats.stackexchange.com/questions/33709/interpreting-the-regression-output-from-a-mixed-model-when-interactions-between http://stats.stackexchange.com/questions/6294/interaction-between-ordinal-and-categorical-factor – kjetil b halvorsen Sep 14 '15 at 14:15

1 Answers1

0

I think that generally the rules from linear regression apply, just that you interpret everything in terms of effects on the hazard function.

A test for interaction can be obtained by a likelihood-ratio test between the model with main effect + interactions (size*state) and the model with only main effects size + state. To obtain the p-value of this test, see ?anova.coxph.

Theodor
  • 1,671
  • 9
  • 14