Prior sensitivity analysis for an ML-UMR fit

Refit an mlumr() model across a grid of prior_beta scales (keeping the family, mean, and df fixed) and summarize how the posterior for the marginal treatment effects (delta_index, delta_comparator) moves. This is the workflow recommended by Vehtari et al.'s prior-choice wiki for judging how much of the posterior is driven by the data versus the prior.

Usage

prior_sensitivity(
  fit,
  prior_beta_scales = c(0.5, 1, 2.5, 5, 10),
  probs = c(0.025, 0.5, 0.975),
  verbose = TRUE,
  ...
)

Arguments

fit

A fitted mlumr_fit object to re-fit under alternative priors.

prior_beta_scales

Numeric vector of scales for prior_beta. Default c(0.5, 1, 2.5, 5, 10).

probs

Quantiles for summarizing each posterior (default c(0.025, 0.5, 0.975)).

verbose

Logical; if FALSE, suppresses progress messages and final printed summary table.

...

Additional arguments forwarded to mlumr() on each refit (e.g. chains, iter, refresh). Sampling defaults otherwise inherit from the original fit.

Value

A data frame (tibble-style) with one row per (scale, population, quantile) combination and columns scale, parameter, mean, sd, and the requested quantiles. Side effect: prints a summary table at the end.

Details

Only the scale of the prior_beta family is varied; its distribution (normal / student_t) and mean are preserved so comparisons are apples to apples. prior_intercept and prior_sigma are carried through unchanged from the original fit. Each value in prior_beta_scales is used as the absolute scale for every coefficient at that refit — if the original fit used per-coefficient priors, all coefficients are set to the same scale (the sweep is deliberately homogeneous so the grid reflects a single level of prior informativeness per refit, not a rescaling of existing relative differences). If the original prior_beta used an exponential family, it is swapped for a prior_normal(0, scale) at each grid point since exponential has no scale parameter to vary.

See also

prior_summary() for a one-shot description of the priors on a fit; marginal_effects() for the posterior summary quantities this sweep tracks.

Examples

if (FALSE) { # \dontrun{
sens <- prior_sensitivity(fit_spfa, prior_beta_scales = c(1, 2.5, 5))
} # }