Hello everybody,
I'm trying to understand if there is an effect modification by sexual orientation (Heterossexual x homossexual x bissexual ) of the association between alcohol consumption (binary exposure) and multiple partner (binary outcome).
I'm dealing with a cross-sectional study and I ran a Poisson Model to calculate the IRR, because my outcome is not rare and in these cases apparently poisson regression seems to provide better estimates than Logistic regression.
1. I tried do run stratified models to check the effect for different sexual orientation groups, but the IRR (and their respective ICs) are not comparable between the different sexual orientation strata, right?
2. Then I decided to run a non stratified model to do a formal test of a multiplicative interaction between sexual orientation and alcohol consumption, controlling for others confounders. I ran a join hypotheses test (Bonferroni) to check if the two coeficients (because I have a three-level categoric variable) of the interaction terms was statisticaly significant and it wasn't.
3. By curiousity, I decided to run the marginal effects using "margins alcohol#sexual_orientation" and for one group (homossexual) the confidence interval of who consume alcohol e who doesn't are not overlapping which suggests that there is an additive interaction between sexual orientation and alcohol. Am I right with this interpretation? Does it make sense not to have a multiplicative interaction while having an additive?
How do you suggest to formally evaluate the additive x multiplicative interaction in a Poisson model?
I'm trying to understand if there is an effect modification by sexual orientation (Heterossexual x homossexual x bissexual ) of the association between alcohol consumption (binary exposure) and multiple partner (binary outcome).
I'm dealing with a cross-sectional study and I ran a Poisson Model to calculate the IRR, because my outcome is not rare and in these cases apparently poisson regression seems to provide better estimates than Logistic regression.
1. I tried do run stratified models to check the effect for different sexual orientation groups, but the IRR (and their respective ICs) are not comparable between the different sexual orientation strata, right?
2. Then I decided to run a non stratified model to do a formal test of a multiplicative interaction between sexual orientation and alcohol consumption, controlling for others confounders. I ran a join hypotheses test (Bonferroni) to check if the two coeficients (because I have a three-level categoric variable) of the interaction terms was statisticaly significant and it wasn't.
3. By curiousity, I decided to run the marginal effects using "margins alcohol#sexual_orientation" and for one group (homossexual) the confidence interval of who consume alcohol e who doesn't are not overlapping which suggests that there is an additive interaction between sexual orientation and alcohol. Am I right with this interpretation? Does it make sense not to have a multiplicative interaction while having an additive?
How do you suggest to formally evaluate the additive x multiplicative interaction in a Poisson model?
Comment