Quick covariate screen against individual parameters and ETAs

Description

A fast, informal covariate screen built on the fit’s covariate table (fit$covtab). For every covariate it reports the association with each inter-individually-varying parameter, measured two ways: against the subject’s individual parameter estimate (the “EBE” column, from fit$individual_estimates) and against the parameter’s ETA (the “ETA” column, from fit$ebe_etas). Only parameters that have IIV (an ETA) are screened. This is a screening aid to decide what is worth a formal covariate search - it is not itself a covariate test.

Usage

ferx_cov_screen(fit, threshold = 0.2)

Arguments

  • fit: A ferx_fit object with a covtab (the model must declare a [covariates] block).
  • threshold: Minimum absolute association to report. Default 0.2.

Details

Covariates are aggregated to one value per subject first: the median for continuous covariates (so time-varying covariates collapse to a typical value) and the most frequent level for categorical covariates. The association measure depends on the covariate type (from fit$covariate_types): a signed Pearson correlation for continuous covariates, and the correlation ratio (eta, in [0, 1]) for categorical covariates. Rows are kept only when either measure is at least threshold in absolute value, and are ordered by the stronger of the two.

Seealso

Other diagnostics: [check_diagnostics](check_diagnostics.qmd)(), [ferx_cor_matrix](ferx_cor_matrix.qmd)(), [ferx_estimates](ferx_estimates.qmd)(), [ferx_eta_cov](ferx_eta_cov.qmd)(), [ferx_plot_trace](ferx_plot_trace.qmd)(), [ferx_runlog_iters](ferx_runlog_iters.qmd)(), [ferx_trace](ferx_trace.qmd)(), [ferx_warnings](ferx_warnings.qmd)(), [summary.ferx_fit](summary.ferx_fit.qmd)()

Concept

diagnostics

Value

Data frame with columns parameter, covariate, type, ebe, and eta, ordered by descending association strength. Returned invisibly; the table is also printed. Returns an empty data frame (with a message) when nothing clears the threshold, and NULL when the fit has no covariate table or no ETAs.

Examples

ex  <- ferx_example("two_cpt_oral_cov")
fit <- ferx_fit(ex$model, ex$data, method = "focei", covariance = FALSE)
ferx_cov_screen(fit)