0
$\begingroup$

I have two logistic models assessing the correlates of a binary outcome in males and females, respectively. The predictors are the same in both groups.

I want to see if the effects of the predictors (the odds ratios) are different for males and females, i.e. compare the same dependent and independent variables across the category of gender. How can I accomplish this?

$\endgroup$
1
  • 1
    $\begingroup$ If you have two models, one for males whereas the other one for females, with exactly the same outcome variable, why not run a single model with gender as a predictor interacted with the other predictors to examine whether the effects of predictors vary depending on gender? $\endgroup$
    – user139190
    Jun 19, 2019 at 14:39

1 Answer 1

1
$\begingroup$

What you're asking about is effect modification, which you test by including interactions. If you fit a logistic model with gender interacting with every predictor, then it will produce the exact same fit as the fully stratified models (this is not exactly true for linear regression because the interaction model would constrain the error variance to be equal whereas the stratified model would not).

In the interaction model, you can tell whether a predictor has a different effect based on gender by seeing whether the interaction is significant, and look at the value of the interaction coefficient relative to the main effects to see how the effects are different. You may have collinearity between interactions if you put them all in the same model, so you may want think about testing interactions one at a time, or making more specific hypotheses about which variables have gender-specific effects.

$\endgroup$
1
  • $\begingroup$ Thank you so much .Yes that is effect modification. I have 8 predictors and most of them are categorical (both ordinal and binary 0,1) . If I create interaction term for all the dummies in a multivariate model, how I can interpret the result if some categories are not significant? Can I just put them in the model as continuous variables? $\endgroup$
    – Neda
    Jun 20, 2019 at 7:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.