Compute posterior partial correlations from a multiggm_fit object

Extracts the full array of posterior partial correlation draws from a fitted multi-GGM model. For each saved MCMC iteration and group, the precision matrix is converted to a partial correlation matrix.

Usage

posterior_pcor(fit, use = c("C_save", "Sig_save"), chain = 1L)

Arguments

fit

A multiggm_fit object (single chain) or multiggm_fit_list object (multiple chains).

use

Character; which posterior array to compute partial correlations from:

"C_save"

(default) Use the precision matrix draws directly. Faster and recommended.

"Sig_save"

Use the covariance matrix draws, inverting each to get precision first. Slower; mainly for cross-checking.

chain

Integer; which chain to use if fit is a multiggm_fit_list. Default 1.

Value

A numeric array of dimension [p, p, K, nsave]. Entry pcor[i, j, k, s] is the partial correlation between variables i and j in group k at saved iteration s. Diagonal entries are 1. Off-diagonal entries are in [-1, 1].

Details

The partial correlation between variables i and j given all other variables is defined as: $$\rho_{ij|rest} = -\Omega_{ij} / \sqrt{\Omega_{ii} \Omega_{jj}}$$ where \(\Omega\) is the precision matrix.

See also

precision_to_pcor(), posterior_ci(), fitted.multiggm_fit()

Examples

sim <- simulate_multiggm(K = 2, p = 8, n = 80, seed = 1)
fit <- multiggm_mcmc(data_list = sim$data_list, burnin = 200, nsave = 100)
pcor_draws <- posterior_pcor(fit)
# Posterior mean partial correlation for group 1
pcor_mean <- apply(pcor_draws[,,1,], c(1,2), mean)