Multicollinearity between organic impressions and online media channels

[ibrahim-taher]

I'm running into an issue where an extra feature and two media channels are all collinear with each. I see high VIFs for these three features.

I'm struggling to think about how to resolve this. Removing the extra feature may give too much credit to any of media channel features and removing some of the media channel data will result in an incomplete picture of our marketing spend.

I've seen the suggestion to run multiple models where we only keep one of the collinear features and running experiments to update our priors, but I wanted to see if others have suggestions.