Validate Model 2a parameter recovery — validate_recovery

Simulation-based check that run_mod_2a() recovers a known cross-biomarker correlation. It simulates two-biomarker longitudinal data with a chosen c_vec (sim_case_data_2a()), fits Model 2a, and returns a table comparing the true cross-biomarker correlation with the posterior estimate per kinetic parameter.

Recovery is expected to be accurate for well-identified parameters (peak, decay) and to attenuate gracefully for weakly-identified ones (baseline); true nulls should yield credible intervals covering zero.

Usage

validate_recovery_2a(
  n = 120,
  mu_g = c(0, 3, 2.3, -4, -1),
  mu_a = c(0.2, 3.1, 2.2, -3.8, -1.1),
  sigma_g = diag(c(0.09, 0.16, 0.09, 0.16, 0.09)),
  sigma_a = diag(c(0.09, 0.16, 0.09, 0.16, 0.09)),
  c_vec = c(0.054, 0.08, 0, 0.064, 0),
  noise_sd = 0.15,
  seed = 1,
  ...
)

Arguments

n: integer number of subjects. Default 120.
mu_g, mu_a, sigma_g, sigma_a, c_vec: Model 2a truth for sim_case_data_2a(). Sensible defaults are provided.
noise_sd: Residual SD on the log scale. Default 0.15.
seed: RNG seed for the simulation. Default 1.
...: MCMC controls forwarded to run_mod_2a() (e.g. nchain, niter, nmc, nburn, nadapt).

Value

A data.frame with columns param, true_rho, cor_med, cor_lo, cor_hi, and verdict ("recovered", "null ok", or "review").