# Interpretation of Interaction Term in Log-Linear Model

#### Use Schmitt

##### New Member
Hello All! I am currently writing my Bachelor’s Thesis and have a question about the interpretation of an interaction term in a log-linear model.
Without an interaction I’d take the coefficient of a x variable and transform it via exp(coefficient)-1 * 100. Then I am able to argue: A 1 unit increase in X is associated with a increase of … percent in Y. The coefficients are just very high (<0.3) That's why I applied the transformation.
Now I want to do that with the coefficient of an interaction term with a continuous and a binary variable. Binary variable is male/female and I want to be able to argue:
A 1 unit increase in the male sample is associated with an increase in variable Y by …percent.
A 1 unit increase in the female sample is associated with an increase in variable Y by …percent.
Is it correct to use the Stata command margins, dydx(continuous variable) at (dummy_variable=(0 1)) and then transform the output for each value of the dummy (0 1) via exp(coefficient)-1 * 100 ?
I would appreciate any help!

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
This is a cool questions, I plan to look at it tomorrow. Have you found a solutions yet. I feel like I have seen a comparable example in Andrew Gelman's and Jen Hill's Regression Stories book.

#### Use Schmitt

##### New Member
Thank you for your response hlsmith!
I have no solution yet and would be more than thankful for any hint!

#### Use Schmitt

##### New Member
My regression looks as follows: log(y1) = B0 + B1(variable2) + B2(variable3)+ B3(variable2*variable3) +controls