Reach and Frequency in National Level Model

[annie-gulati]

For R&F (Reach and Frequency), we are trying to include in National level model but we are getting extreme large value (in millions) of R-hat.

Attached is the snippet that we implemented in our code.

[dattatreyam23]

Hi @annie-gulati ,

Thank you for reaching out to the Google Meridian support team.

I went through the code snippet you have provided, and it appears to be correct. Large R-hat values indicate that the model is having difficulty achieving convergence. If you haven't already, please review our documentation on achieving MCMC convergence, which can be found at Getting MCMC convergence.

Additionally, knowing if all parameters, or just one or two, have extremely large R-hat values would be helpful for diagnosis. For example, if the tau_t parameter has a large R-hat value, you should consider modifying the knots parameter. Also, does the model converge when using impressions instead of reach and frequency?

If the issue still persists, please provide a few more details on your dataset: (1) number of channels included; (2) whether the data is weekly or daily, etc. This will help us figure out the exact point of error.

Feel free to reach out if you have any further questions.

Thanks,
Google Meridian Support Team