0
$\begingroup$

Suppose I have a model with an interaction between two dummy variables

$$ \begin{equation} wage = \beta_0 + \beta_1 male+ \beta_2 white + \beta_3 educ + \beta_4 (male* white) + \epsilon \end{equation} $$

and that the coefficient for $ \beta_4 $ is significant.

Am I able to say that the impact of being white is significantly conditioned by gender (i.e. that the coefficient for white is different across men and women)? In the case of a continuous moderator, I would plot the conditional effect of the independent variable for different values of the moderator and see whether this effect is statistically different from zero. Should I do something similar also in this case? Thanks a lot.

Improve this question
$\endgroup$
1
  • $\begingroup$ When variable interactions like this start being important, consider a Structural Equation Modeling (SEM) approach. Interaction terms will get the job done, but with less clarity/interpretability. $\endgroup$
    – Mox
    Commented Feb 7, 2023 at 18:35

1 Answer 1

Reset to default
0
$\begingroup$

You can say both the impact of being white is significantly conditioned by gender and equally the impact of gender is significantly conditioned by being white. You can plot the distribution of $wage$ in each of the four subgroups, for instance using box-and-whisker plots, violin plot, and even scatter plots. See for example Figure R9 here: https://wellcomeopenresearch.org/articles/4-63

Improve this answer
$\endgroup$

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.