I am not sure if this is a SAS question or a statistics question.
When you specify an interaction between a categorical variable and either another categorical or interval variable SAS generates a test of the overall interaction(in the type 3 analysis of effects) and seperately between each level of the categorical variable and every level of the variable it is interacting with.
My question is how do you interpret the results if the overall categorical variable interaction (in the type 3 analysis of effects) is signficant, but many of the interactions between levels of the variable and the other IV are not? Is interaction occuring, and how do you interpret these level by level interactions?
If some of the level by level interactions are signficant and some not, does this mean interaction is occuring at some levels, and not at others? Nothing I have read on interaction addresses this.
A separate question. What is the difference in interpretation of an interaction term when you use effect coding (the default for SAS PROC LOGISTIC) and reference coding? All discussions of interaction I have seen in the past only addressed this issue implictly in the context of reference coding.
When you specify an interaction between a categorical variable and either another categorical or interval variable SAS generates a test of the overall interaction(in the type 3 analysis of effects) and seperately between each level of the categorical variable and every level of the variable it is interacting with.
My question is how do you interpret the results if the overall categorical variable interaction (in the type 3 analysis of effects) is signficant, but many of the interactions between levels of the variable and the other IV are not? Is interaction occuring, and how do you interpret these level by level interactions?
If some of the level by level interactions are signficant and some not, does this mean interaction is occuring at some levels, and not at others? Nothing I have read on interaction addresses this.
A separate question. What is the difference in interpretation of an interaction term when you use effect coding (the default for SAS PROC LOGISTIC) and reference coding? All discussions of interaction I have seen in the past only addressed this issue implictly in the context of reference coding.
