Estimate precision weights from a curveR calibration result

The curveR-ecosystem entry point. Estimates the power-law relationship between calibration-curve precision and residual variance, \(\sigma_i = \phi\, se_i^{\beta_1}\), \(w_i = 1/\sigma_i^2\), using a joint Bayesian location-scale model with a saturated cell-means location (see fit_saturated_weight()). The default scale predictor is the uncapped se_concentration from the calibration curve, so \(\phi = 1\) means "se_concentration is a calibrated residual SD".

Usage

fit_precision_weights(
  data,
  design = NULL,
  scale_predictor = c("se", "pcov"),
  plate_col = NULL,
  ...
)

Arguments

data

Either the output of as_weight_data() (class weight_data), or a calibration_result(_multiplate) / plain data.frame, in which case design is required and as_weight_data() is called internally.

design

Character vector of design-group columns (required unless data is already a weight_data).

scale_predictor

"se" (default; uncapped log10-scale SD) or "pcov" (legacy, lossy because of the cap).

plate_col

Plate grouping column for the location random intercept. Defaults to "curve_id" when present, else NULL.

...

MCMC controls and priors passed to fit_saturated_weight() (e.g. iter, warmup, chains, cores, adapt_delta, seed).

Value

A precision_weights object: a list with $estimates (phi, beta1, CIs, interpretation, diagnostics), $weights (data frame keyed by obs_id/sampleid/curve_id with se, pcov, sigma, w, w_norm), $design, $scale (conc_scale + scale_predictor), and $fit (the brms fit).

See also

as_weight_data(), predict_weights(), join_weights(), fit_saturated_weight()

Examples

# \donttest{
# mp <- curveRfreq::fit_calibration_freq_multiplate(
#         standards = std, samples = samples_with_design, ...)
# wd <- as_weight_data(mp, design = c("timeperiod", "cohort_arm"))
# pw <- fit_precision_weights(wd, iter = 1000, warmup = 500, chains = 2)
# pw$estimates$phi; pw$estimates$beta1
# }