Skip to contents

Predict Method for rhierMnlRwMixture Objects

predict.rhierMnlRwMixture.Rd

Predict Method for rhierMnlRwMixture Objects

Usage

# S3 method for class 'rhierMnlRwMixture'
predict(
  object,
  newdata = NULL,
  type = "DeltaZ+mu",
  burn = 0,
  nsim = 10,
  mode = "coefficients",
  r_verbose = TRUE,
  force_tree_eval = FALSE,
  ...
)

Arguments

object

A fitted rhierMnlRwMixture object.

newdata

Optional list containing data for prediction. Structure depends on mode and type:

  • For mode = "coefficients" with type %in% c("DeltaZ", "DeltaZ+mu", "SigmaZ"): requires newdata$Z, a matrix with npred rows for prediction units (if model was fit with Z).

  • For mode = "posterior" with type = "choice_probs": requires newdata$nlgtdata, a list of length nlgt (original number of units). Each element \\[[i]] must contain $X, the design matrix (T_i*p) x nvar for unit i. Also requires newdata$p, the number of alternatives.

  • For mode = "prior" with type = "choice_probs": requires newdata$Z (if model fit with Z, determining npred), newdata$p, and newdata$X (a list of length npred, each element \\[[i]] having design matrix (T_i*p) x nvar).

type

Type of prediction within the selected mode:

  • "DeltaZ": Expected part-worths of the representative consumer, \(\Delta(Z)\).

  • "DeltaZ+mu": Expected part-worths plus the mean of the unobserved heterogeneity component, \(\Delta(Z) + \mu_1\). The package supports only ncomp = 1.

  • "choice_probs": Predictive choice probabilities (use with mode = "posterior" or mode = "prior").

  • "SigmaZ": Draws of the heteroscedastic covariance matrix \(\Sigma(Z)\). Available only for models fit with Prior$vartree (class marker "bayesm.HART.HeterCov").

burn

Integer, number of initial MCMC draws to discard.

nsim

Integer, number of draws from the heterogeneity distribution per posterior draw for mode = "prior" and type = "choice_probs".

mode

Prediction mode:

  • "coefficients": coefficient-level outputs (DeltaZ, DeltaZ+mu, SigmaZ).

  • "posterior": posterior predictive output (type = "choice_probs").

  • "prior": prior predictive output (type = "choice_probs").

r_verbose

Logical, print progress updates?

...

Additional arguments passed to underlying prediction functions (e.g., mc.cores, verbose for BART DeltaZ predictions via pwbart).

Value

Depends on type:

  • For type %in% c("DeltaZ", "DeltaZ+mu"): 3D array [npred, nvar, ndraws_out] of predicted expected part-worths.

  • For type = "SigmaZ": 4D array [npred, nvar, nvar, ndraws_out] of covariance draws at each prediction unit.

  • For mode = "posterior", type = "choice_probs": List of length nlgt. Each element \\[[i]] is a 3D array [T_i, p, ndraws_out] of posterior predictive choice probabilities for unit i.

  • For mode = "prior", type = "choice_probs": List of length npred. Each element \\[[i]] is a 3D array [T_i, p, ndraws_out] of prior predictive choice probabilities for prediction unit i.