Hello @paul-jdfagan,

Yes, this is common in Meridian models, we suspect that control coefficients have little impact, you can check gamma_c posterior traces using ArviZ object or see prior-posterior distributions using:

model_diagnostics.plot_prior_and_posterior_distribution(parameter='gamma_c')

Curious to see the Devs comments on this topic!

Thanks for your response @sonriks6.

My first thought was that each media channel gets only one β coefficient that applies across the entire dataset - No temporal variation in effectiveness - Unlike the baseline μ_t which varies over time through knots.

Also, is there a mechanistic explanation for how a baseline drops? is it just mirroring expected KPI so if that drops, given the fixed beta, then baseline also drops?

Hi @paul-jdfagan,

Thank you for contacting us.

I went through the graph you provided, and while the graph seems to show a constant contribution at first glance, a closer look at the tails reveals a slight divergence. This suggests the contribution might not be constant after all.

Can you confirm if the contribution is indeed meant to be constant?

Thank you,

Google Meridian Support Team