Interaction terms as media channel variables

[victoriatascau]

For two media channels where we suspect a synergistic effect (i.e., an interaction term would be suitable to capture it), can we include that interaction term as a media channel variable (besides the initial two)? To account for multicollinearity, one can use scaling. In this way, we would get a contribution percentage estimation for the interaction term, so we could point to the expected synergistic effect, but what are the implications for the modeling? Is this not advised at all?

This question is related to 1339, but it is about placing the interaction term in media channel variables, not in controls variables.

[gousesb]

Hello @victoriatascau ,

Thank you for contacting us!

Meridian is currently an additive model, and using an interaction term might break the "additive" nature of the model. As a result, the budget optimization might not work as intended.

Feel free to reach out for any further queries or feedback regarding Meridian.

Google Meridian Support Team