Random-effect marginal variance factors

Source: R/random-effects-marginal-covariance.R

Extracts row-aligned design multipliers for one-column formula random-effect blocks that are intended to be marginalized by a downstream likelihood. For a block with SD parameter \(\tau\), the marginal variance contribution is \(\tau^2\) times row_multiplier. With known group covariance this factor is built from the prepared group kernel, so the dense row-space covariance factor is \(Z K Z'\).

random_effects_marginal_variance_factors(
  formula_design,
  blocks = NULL,
  require_diagonal = TRUE,
  require_one_to_one = FALSE
)

Arguments

formula_design

a BayesTools_formula_design object returned by JAGS_formula() or JAGS_formula_design().

blocks

optional character vector of random-effect block names. When NULL, only marginalized random-effect blocks are selected.

require_diagonal

whether to require the row-space covariance factor to be diagonal. This protects downstream likelihoods that can only consume row-wise variance multipliers.

require_one_to_one

whether each observation row must map to a distinct grouping level.

Value

A list of class BayesTools_random_effects_marginal_variance_factors with selected block metadata. Each block contains row_multiplier, group_map, group_levels, model_matrix, SD metadata, compile mode, and known group covariance metadata when present.

Details

Unsupported factor representations signal an error inheriting from BayesTools_random_effects_marginal_variance_unavailable. The condition also inherits from BayesTools_random_effects_error, error, and condition, and carries message and call. The shared parent includes invalid allocation support errors and does not itself imply unavailability. The unavailability condition carries stable block_name and reason fields so downstream packages can distinguish structural unavailability from invalid inputs. reason is one of "row_indexed_sd", "multiple_columns", "non_diagonal_row_covariance", or "repeated_groups".