Skip to contents

Simulate prior predictive replicate datasets for a probit model

Source: R/probit_prior_predictive.R
simulate_probit_prior_predictive.Rd

Draws n_states independent parameter states directly from the declared priors (no NUTS/no conditioning on outcomes) and generates a complete replicate outcome matrix from each, using the REAL fixed-effect design matrix and random-effect grouping structure of fitted_model as conditioning predictors (per Stan User's Guide-style prior predictive simulation). Reuses the identical generative mechanism as simulate_probit_posterior_predictive (Bernoulli(Phi(mu)) for identity residual, Z ~ MVN(mu, Omega); Y = I(Z>0) for correlated residual) – only the source of theta differs (prior vs. posterior draws).

Usage

simulate_probit_prior_predictive(
  fitted_model,
  n_states = 1000L,
  seed = 123L,
  prior_config_override = NULL,
  preserve_observation_mask = TRUE,
  return_replicates = FALSE
)

Arguments

fitted_model

A fitted model object as returned by fit_bayesian_multivariate_probit – supplies the real X_event design matrix, random-effect grouping structure, class panel, and prior_config_used (the priors to replicate exactly). Only design/configuration fields are read; no posterior draws are used.

n_states

Integer; number of independent prior states to draw. Default 1000.

seed

Integer seed (deterministic).

prior_config_override

Optional named list overriding any of beta_sd, tau_sd, lkj_eta from fitted_model$prior_config_used – e.g. to prior-predictive-check a CANDIDATE prior before fitting.

preserve_observation_mask, return_replicates

See simulate_probit_posterior_predictive.

Value

An "amr_prior_predictive_draws"/"amr_ppc_draws" object with the same shape as simulate_probit_posterior_predictive's return value.

Details

Returns an object of class c("amr_prior_predictive_draws", "amr_ppc_draws") – structurally compatible with compute_probit_ppc_statistics (which accepts any "amr_ppc_draws"-classed object), so the same AMR-specific discrepancy-statistic machinery can summarise the prior predictive distribution's own properties (see compute_prior_predictive_status, which additionally derives the prior-specific plausibility summaries in Part 11 of the predictive- checking specification: fraction of event-class probabilities near 0/1, fraction of degenerate all-resistant/all-susceptible profiles, and hospital-level spread).

References

Stan Development Team. "Posterior and Prior Predictive Checks." Stan User's Guide. https://mc-stan.org/docs/stan-users-guide/posterior-predictive-checks.html