Skip to contents

Near-degeneracy diagnostic for the posterior Omega correlation matrix

Source: R/plot_omega_diagnostics.R
plot_omega_degeneracy_diagnostic.Rd

A multivariate-probit correlated-residual fit can push Omega toward a near-singular boundary (very high pairwise correlations across the board) without that being visible from any single Omega\[i,j\] entry. For each posterior draw this computes the smallest eigenvalue of the full D x D Omega matrix – a value approaching 0 means that specific draw's correlation matrix is nearly singular – and reports the distribution across draws, plus the condition number (largest / smallest eigenvalue) as a supplementary summary.

Usage

plot_omega_degeneracy_diagnostic(
  fit,
  class_cols,
  title_base = "",
  degenerate_threshold = 0.05
)

Arguments

fit

List returned by fit_bayesian_multivariate_probit().

class_cols

Character vector of all class names, in the fit's canonical order.

title_base

Character. Prefixed to the plot title.

degenerate_threshold

Numeric. Draws with smallest eigenvalue below this are counted as "near-degenerate" in the subtitle. Default 0.05.

Value

A ggplot object, or NULL if this is not a correlated-residual fit, or Omega draws are unavailable.