Extract Posterior Predicted Parameters from a Hurdle or Count Model

Computes posterior predicted values for a specified distributional parameter (e.g., mean, shape, or hurdle probability) from a fitted Bayesian count or hurdle model. The function supports extracting parameters for positive counts only or for all observations.

Usage

posterior.pred(fit, dpar, count.only = TRUE)

Arguments

fit

A fitted brms model object containing data, formula, and MCMC posterior draws.

dpar

Character string specifying the distributional parameter to extract: "mu" (mean), "shape" (dispersion), or "zero" (hurdle probability).

count.only

Logical; if TRUE (default), computes predicted parameters only for positive counts (y > 0); otherwise, includes all observations.

Value

A numeric matrix of predicted parameter values for each observation (columns) and posterior draw (rows).

Details

The function performs the following steps:

1. Builds the model matrix for the chosen parameter and observation subset.

2. Extracts the corresponding posterior MCMC draws from the fitted model.

3. Computes the linear predictor via matrix multiplication of draws and model matrix.

4. Applies the link function associated with the parameter (e.g., logit, log) to obtain the predicted parameter values on their natural scale.

See also

log_pred_dist_HNB, log_pred_dist_NB, log_pred_dist_HP, posterior_linpred

Examples

if (FALSE) { # \dontrun{
# Extract posterior predicted mean (mu) for all observations
mu_pred <- posterior.pred(fit, dpar = "mu", count.only = TRUE)

# Extract hurdle probabilities
pi_pred <- posterior.pred(fit, dpar = "zero", count.only = FALSE)
} # }