Model Specification
This section explains how PanelMMM is defined: the model structure, media
transforms, priors, panel dimensions, optional time variation, and calibration
hooks.
Pages
- Model Overview - The actual
PanelMMMmean structure, scaled-space formulation, and optional components. - Adstock and Saturation - Supported media
transforms, their priors, and the
adstock_firstcomposition order. - Priors and Configuration - Default prior keys,
model_config, transform-prior overrides, and directional control priors. - Time-Varying Parameters - How
time_varying_interceptandtime_varying_mediauseSoftPlusHSGP. - Seasonality and Trends - Built-in yearly seasonality plus custom Fourier, trend, and event effects.
- Panel Dimensions - How
dimschange the shape of the data and parameters. - Choose an Estimator - Compare the released time-series, FE, and CRE contracts and choose the identifying variation that matches the modelling question.
- Fixed-effects Estimator - The released FE contract with one unit dimension, its within-design checks, and its limits.
- Correlated-random-effects Estimator
- The released CRE contract with one unit dimension, its transformed summary basis, estimability checks, and interpretation limits.
- Calibration - Likelihood-based evidence updates, the existing saturation-lift equations, cost-per-target penalties and Python/YAML access.
- Experimental evidence - Import a GeoLift study, compare its full intervention with a time-series model, and calibrate a new fit using explicit uncertainty and independence assumptions.