Regression coefficient interpretation for interaction (basic q)

Interfaces brms
1

Hi all,
A quick question regarding interpreting parameter values given by BRMS.

I use the following model:

Outcome ~ 1 + Timepoint + Weight + Timepoint:Weight + (1|subject)

Timepoint has 2 categorical values, t1 and t2, and weight is continuous. In the output for the main effects and interactions, I get estimates for the following parameters:

Intercept
Timepointt2
Weight
Timepointt2:Weight

My question is pretty basic but I just wanted to confirm so as not to make a silly mistake.
Is the parameter “Timepointt2:Weight” simply the estimate of the effect of Weight at timepoint t2, or is it the difference from the effect of Weight at time 1 (so that the effect of Weight at t2 is actually “Weight” + “Timepointt2:Weight”).

If anyone can clarify this it would be great!

  • Operating System: MacOS 10.12.6
  • brms Version: 2.8.0
2

This depends on the coding of your variables. Since brms works the same as lm/glm etc. in that regard, any answer you can find that on the internet about those functions will apply to brms as well.
If you didn’t specify anything else, the interaction is likely to be difference in the effect (unless you changed the coding to something else than dummy coding).

3

Thanks Paul - I’ll have a look through the documentation of lm/glm a bit further to confirm. I realised I could also create data with really clear interactions and then run the model, just to check my interpretation of what the values should mean.