Skip to contents

Jackknife-after-Bootstrap (JAB) influence for lavaan bootstrap results

Source: R/jab_after_boot.R
jab_after_boot.Rd

Compute Jackknife-after-Bootstrap (JAB) influence values for a single model parameter and diagnose observation-level influence by comparing the full bootstrap distribution with leave-one-out (LOO) subdistributions obtained via boot.idx (stored by store_boot(keep.idx = TRUE)).

The ALL-summary (mean/SE/CI) is taken directly and consistently from standardizedSolution_boot() (standardized case) or parameterEstimates_boot() (unstandardized case). JAB influence is defined as \(I_j = n\{\bar\theta_{(-j)} - \bar\theta\}\), where \(\bar\theta\) is the bootstrap mean over all replicates, and \(\bar\theta_{(-j)}\) is the bootstrap mean over the replicates that exclude observation \(j\).

Usage

jab_after_boot(
  fit,
  param,
  standardized = TRUE,
  top_k = 5L,
  ci_level = 0.95,
  min_keep = NULL,
  plot = FALSE,
  plot_engine = c("ggplot2", "base"),
  ylab_override = NULL,
  verbose = TRUE,
  font_family = "sans"
)

Arguments

fit

A lavaan object for which store_boot(keep.idx = TRUE) has been called. Must contain fit@external$sbt_boot_ustd (with boot.idx attribute) and typically sbt_boot_std.

param

Character(1). Target parameter. Accepts "lhs op rhs", a ":=" label, or a free-parameter label.

standardized

Logical. If TRUE, use standardized bootstrap; else unstandardized.

top_k

Integer. Number of top cases (by |JAB_value|) to report.

ci_level

Numeric (0,1). Percentile CI level for LOO subdistributions.

min_keep

Integer. Minimum number of bootstrap replicates kept in each LOO subset; default max(30, floor(0.2 * B)).

plot

Logical. If TRUE, draw a JAB diagnostic plot.

plot_engine

Character. Plot engine for plot = TRUE. One of "ggplot2" or "base". Default "ggplot2" if available, otherwise "base".

ylab_override

Optional character. Override the default y-axis label in the plot.

verbose

Logical. If TRUE, print compact summaries.

font_family

Character. Graphics font family (e.g., "serif", "sans", "Times New Roman").

Value

A list with:

param

The target parameter string.

standardized

Logical flag as input.

full_summary

Data frame with ALL distribution summary: scope="ALL", param, mean, SE, CI.Lo, CI.Up.

cases_summary

Data frame (top top_k) with columns: case, JAB_value, mean, SE, CI.Lo, CI.Up.

F

The occurrence matrix \(B \times n\).

tstar

Full bootstrap vector for param.

Details

Parameter identification is robust: it accepts "lhs op rhs", ":=" user-defined name, and label. For unstandardized cases, if label is given and the column is missing in attr(unstd_boot, "boot_est_ustd"), the function will fall back to the raw matrix fit@external$sbt_boot_ustd if a column with that label exists (e.g., "b"), keeping SE/CI from the parameterEstimates_boot() row if available.

See also

store_boot, standardizedSolution_boot, parameterEstimates_boot