Compute hierarchical saturation probabilities and link parameters

Computes posterior draws of the link-scale parameters or coverage probabilities ($\phi$) across cohorts and locations from an [imugap_fit] object.

Usage

compute_phi(object, link_scale = FALSE, posterior_size = NULL, ...)

Arguments

object: an object of class imugap_fit returned by [sampling()].
link_scale: logical scalar; return link-scale parameters ($\eta$) instead of probabilities ($\phi$)? (default: FALSE).
posterior_size: integer scalar; optional number of posterior draws to retain from the tail of each chain (default: NULL, keeping all draws).
...: additional arguments (currently ignored).

Value

a 3D numeric array with dimensions c(draw, cohort, location) containing the reconstructed posterior draws.

Details

In multi-layer models, sampling() retains orthonormalized spatial offsets (off_layer) and parent aggregation shifts (mu_offset) while omitting the full unrolled parameter matrix (raw_phi_mat) by default to conserve memory. compute_phi() reconstructs the complete link-scale and probability profiles across the location hierarchy: $$\eta_{l, t} = \eta_{\text{parent}(l), t} + \mu_{\text{parent}(l), t} + \delta_l$$ $$\phi_{l, t} = g^{-1}(\eta_{l, t})$$ where $g^{-1}$ is the configured inverse-link function (e.g. logit or probit).

Examples

if (FALSE) { # interactive()
# \donttest{
data("fit_sim", package = "imuGAP")
phi_draws <- compute_phi(fit_sim, posterior_size = 10)
dim(phi_draws)
# }
}