Prior movement is an empirical identification diagnostic. Contrasts that
move substantially when a weakly identified prior is changed should not be
interpreted as data-driven. This helper reuses the principle in
documentation/validation/estimability_gamma.R.
Usage
prior_sensitivity(
object,
newdata,
prior = c("gamma", "beta", "all"),
tighter = 0.5,
looser = 2,
reference = NULL,
...
)Arguments
- object
A
cmlnmr()fit.- newdata
One target population, as for
relative_effects().- prior
Which scales to vary: the interaction prior, component-effect prior, or all configurable scale priors.
- tighter, looser
Positive multipliers for the fitted prior scales.
- reference
Reference treatment. Defaults to the fit reference.
- ...
Named arguments overriding the stored refit call, such as fewer sampling iterations for a screening run.