Skip to contents

Stage 2 of the two-stage bootstrap. Takes the cached ineqx_params replicates produced by bootstrap_params, runs the variance decomposition on each at the requested (ystat, ref), and returns an ineqx_boot object with standard errors and percentile CIs.

Usage

decompose_boot_params(
  boot_params,
  ref = NULL,
  order = c("behavioral", "compositional", "pretreatment"),
  ystat = NULL
)

Arguments

boot_params

An object returned by bootstrap_params.

ref

Numeric, reference time period (longitudinal) or NULL (cross-section).

order

Character vector of length 3, decomposition ordering for the longitudinal decomposition. Default c("behavioral", "compositional", "pretreatment"). Ignored for cross-sectional fits.

ystat

Character, "Var", "CV2", or "VL". Default: the ystat stored on boot_params. "VL" requires boot_params from a log(y) fit (transform = "log"); it runs the "Var" decomposition on the log-scale params, which equals V_L.

Value

An object of class "ineqx_boot" matching bootstrap_se's return shape.

Details

Because the GAMLSS fits are already cached in the input, this call only pays the cost of B variance-decomposition evaluations — orders of magnitude cheaper than a fresh bootstrap_se() run.

ystat can override the value the params were originally extracted at: the per-replicate params$ystat is set to the requested value before decomposition. This works because the bootstrapped params object's parametric columns (mu0, sigma0, beta, lambda) are scale-agnostic; ystat only selects which decomposition formula to apply on top.