Skip to contents

GLS-combine surrogate columns into a single calibrated exposure

Source: R/phase1.R
gls_combine.Rd

Applies the linear back-transform W_tilde = (W - alpha_0) / alpha_1 and the GLS combiner omega = Sigma_tilde^-1 1 / (1' Sigma_tilde^-1 1) to produce a single calibrated exposure with conditional variance sigma_w_sq = (1' Sigma_tilde^-1 1)^-1.

Usage

gls_combine(W, calibration)

Arguments

W

Numeric matrix of surrogate columns, columns ordered as in calibration$W_cols.

calibration

Output of fit_me_calibration().

Value

A list with elements

  • W_bar numeric vector of length nrow(W),

  • sigma_w_sq scalar conditional variance,

  • omega numeric vector of GLS weights (length J),

  • Sigma_tilde J x J back-transformed error covariance.

Examples

sim <- generate_aft_data(n = 100, n_val = 200, seed = 1)
cal <- fit_me_calibration(sim$validation)
W   <- as.matrix(sim$survival[, cal$W_cols])
out <- gls_combine(W, cal)
head(out$W_bar)
#> [1]  9.435341 11.776481  9.556977 12.020854 11.680636  9.901770