Posterior predictive replicate datasets — ppc

Draws replicate outcomes the way posterior_predictive_features() does, for use in any graphical or numerical posterior predictive check: an observation-level random effect (one level per fitted row) is redrawn rather than conditioned on, mode = "new" also draws new case effects, and replicate values are NA wherever the observed outcome is missing. A latent ar() model is predicted on its fitted rows.

Usage

ppc_replicates(model, data = NULL, mode = c("same", "new"), n_draws = 100L)

Arguments

model: A fitted brmsfit.
data: Data holding the outcome, case, phase and (optional) time columns; defaults to the model's data. Rows with a missing outcome are allowed: replicate values are set to NA there.
mode: "same" replicates for the observed cases; "new" draws new case effects from the population posterior (allow_new_levels = TRUE) to assess generalization. In both modes an observation-level random effect (one level per fitted row) is redrawn rather than conditioned on, and replicate values are set to NA wherever the observed outcome is missing so both sides use the same rows.
n_draws: Number of replicate datasets.

Value

A numeric matrix with n_draws rows and one column per row of data.

Examples

if (interactive()) {
  fit <- run_scd_analysis(simulate_scd_data())$model
  yrep <- ppc_replicates(fit, n_draws = 50)
  bayesplot::ppc_dens_overlay(fit$data$outcome, yrep)
}