Hi,

I am fitting a model with a continuous outcome and an interaction between two binary predictors on the righthand side. In addition to measuring the interaction as a departure from additivity, which is what the coefficient of the interaction term implies, I would also like to calculate the interaction effect on the multiplicative scale. VanderWeele and Knol (2014, p. 60) point out that this can be done by fitting a log-linear regression. I am wondering if, instead, such a measure can be calculated using margins and nlcom?

After margins, I ran:

nlcom ln((_b[4._at]/_b[3._at])/(_b[2._at]/_b[1._at])).

But I am not sure if this is producing the estimate I am interested in. Any advice would be much appreciated.

I should mention that the reason I do not want to run a log-linear regression (in this context, I use this term to mean taking the log of the continuous outcome) is because my outcome is a standardized scale and has negative values.

Thank you!
Robbie Dembo