Hi all, I have a modelling question with PyMC MMM package.
Here I have two OTA channels (e.g. booking.com, airbnb) - let’s say they are channels A and B. We used spend instead of the recommended impressions for the model fitting, and our target here is the revenue from channel A, channel B, and also other sources (so not just revenue from channels A and B).
By including more sources of revenue other than channels A and B, we would like to use the contribution from intercept to account for the billboard effect (i.e. people visit OTA channel A and go to another source and make the business) among other things. Also, the revenue data from OTA is not reliable because the cancellation rate is very high.
The proportion of channel spend between A and B is around 5 to 1 (i.e. if we spend $5 in channel A, we only spend $1 in channel B).
However, the fitted contribution is the opposite - approximately $5 is from channel B whereas only $1 is from channel A. I think there are some issues with the fitting process here - the fitted contribution should follow what we set for the spend here. It doesn’t make sense for channel B to get $5 of contribution with $1 spend.
May I know what can go wrong here? Things I checked include input data and prior settings.
Thanks for your help in advance.