I was wondering if you could help me understand a problem I have with a regression. Basically, SPSS excludes one of my variables because it claims that the variable is constant, but I just don't believe it!

I've got four variables that I'm concerned with, all are binary:

isSpeaking: 0 = No, 1 = Yes

isNodding: 0 = No, 1 = Yes

TaskRole: 0 = Instructor, 1 = Learner

RecipientRole: 0 = Secondary, 1 = Primary.

Each data point is a time point per person. isSpeaking and RecipientRole are exclusive, by that I mean if isSpeaking is 1, then there will be no data for RecipientRole (as a person can't be a speaker and a recipient at the same time here). If isSpeaking is 0, then there should be a value for RecipientRole. So a person may be speaking, nodding, a learner and no entry for recipient role.

When I perform a correlation analysis there is an interaction between isNodding and isSpeaking (and lots of other interactions):

isSpeaking is constant w.r.t. RecipientRole as expected.

However, when I try to predict isNodding (in a binary logistic regression), the isSpeaking variable is excluded all together (see point b):

However, it's just not constant for the selected cases! It's constant w.r.t to RecipientRole in that whenever there is a value in the RecipientRole variable there is a 0 in isSpeaking.. but surely this just means there's no interaction between them?

Am I missing something with how a regression works here?

Thanks,

Regards

Stuart