Skip to contents

Compare Chapter 1 and Model 2a on the same data

Source: R/compare_mod_2a.R
compare_mod_2a.Rd

Fits both the Chapter 1 model (fit_chapter1_lean(), the same posterior as run_mod()) and Model 2a (run_mod_2a()) to the same data, then reports:

  1. What stays the same — the shared population means mu.par and the within-biomarker variances, side by side, with absolute differences. Because Model 2a strictly nests Chapter 1, these should agree within MCMC error; large differences would signal a problem.

  2. What Model 2a adds — the cross-biomarker covariance c_p / correlation rho_p, which Chapter 1 cannot represent (it is structurally zero there).

Use this to answer "what changed when I added cross-biomarker covariance?".

On "what improved". A point-estimate comparison shows consistency plus the new term; it does not by itself establish that Model 2a is better. A rigorous improvement claim needs a model-selection criterion (DIC/WAIC; set dic = TRUE for a best-effort DIC from each runjags fit) and, ultimately, the downstream predictive task (e.g. time-since-infection / seroincidence accuracy — MAE, RMSE, CrI coverage), which is the Chapter 2 simulation study rather than a single function.

Usage

compare_mod_2a(
  data,
  nchain = 4,
  nadapt = 1000,
  nburn = 1000,
  nmc = 1000,
  niter = 4000,
  prec_lambda = 0.25,
  dic = FALSE,
  ...
)

Arguments

data

A two-biomarker serocalculator case-data data.frame (e.g. nepal_sees).

nchain, nadapt, nburn, nmc, niter

MCMC controls applied to both fits.

prec_lambda

Factor-loading prior precision (Model 2a only).

dic

logical; attempt to extract DIC from each runjags fit (best-effort; may re-run the models and can be slow). Default FALSE.

...

Prior arguments forwarded to both prep_priors() and prep_priors_2a().

Value

A list of class "model_2a_comparison" with:

  • shared: data.frame comparing mean/var per biomarker x parameter (*_ch1, *_2a, and *_absdiff);

  • cross: Model 2a's cross-biomarker covariance/correlation summary;

  • max_mean_absdiff, max_var_absdiff: worst-case shared-parameter discrepancies (small = consistent);

  • added: the parameters whose c_p credible interval excludes zero;

  • dic_ch1, dic_2a: raw DIC objects when dic = TRUE (else NULL);

  • fits: the two underlying fit objects.