The existing Sensitivity Analysis and Marginal Effects notebook covers response curve sensitivity but not the workflow of assessing how much channel contributions are prior-driven vs data-driven by refitting with varied priors.
A documentation page demonstrating this prior perturbation workflow would fill a gap — particularly relevant for models with limited data or correlated channels, where the posterior may be dominated by priors. This came up in an EAP advisory discussion as a recommended diagnostic that currently has no documented example.
[@drbenvincent suggested this]
The existing Sensitivity Analysis and Marginal Effects notebook covers response curve sensitivity but not the workflow of assessing how much channel contributions are prior-driven vs data-driven by refitting with varied priors.
A documentation page demonstrating this prior perturbation workflow would fill a gap — particularly relevant for models with limited data or correlated channels, where the posterior may be dominated by priors. This came up in an EAP advisory discussion as a recommended diagnostic that currently has no documented example.
[@drbenvincent suggested this]