I have a dichotomous dependent variable (DV), a dichotomous independent variable (IV), and a continuous covariate (CO). I want to test if the association between the DV and the IV controlling for CO exists in younger men, in younger women, in older men and in older women. I have run a logistic regression analysis four times, one for each of the four groups. A reviewer suggested running one logistic regression analysis, including gender and age group in the list of independent variables.
As far as I understand, by doing so, I will investigate if my DV and IV are associated over the effect of gender and age group, not if they are associated similarly in the four groups. Any explanations are welcome!
If you run separate regressions, how do you know if the associations are similar across groups?
If you put all the variables in one model, the interaction of the group and the main effect determines if the effect is consistent across those groups. For example if Y is the dependent variable, X is the main effect, and G is the grouping variable (let's say there's just one variable combining sex and age), then a significant X*G interaction suggests that the effect of X is not the same across all groups. With more variables, the interpretation gets more complicated, but that's the idea.
Many thanks, now I start getting it. What are the sample requirements to add interaction terms?
Do you mean, How many observations do I need for an interaction? You can think of as another term, so maybe at least 10 observations per each term and interaction.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis