I'm running a cox regression to test the effect of different variables on crime.
There's one independent variable of interest (X), for which I hypothesize a negative relationship with crime (i.e., HR<1). This is usually what I get, also when I include several control variables. However, when I insert two specific independent variables (let's call them A and B) to the model together, I get a significant HR>1 for X. When I insert each one without the other (and with all other control variables), the coefficient of X remain <1. Any idea what could cause this?
I checked for multicollinearity and all VIF's are at least 0.7. I also tried to insert an interaction term between A and B, and it is significant, but the coefficient of X is still positive (HR>1).
Thanks!