RMitemInfitCutoffMI: Simulation-Based Infit MSQ Cutoff Determination for Multiply...

In easyRasch2: Psychometric Analysis with Rasch Measurement Theory

Simulation-Based Infit MSQ Cutoff Determination for Multiply Imputed Data

Description

Extends RMitemInfitCutoff to work with multiply imputed datasets produced by the mice package. Runs the parametric bootstrap simulation on each imputed dataset and stacks the resulting distributions, so that the final cutoff intervals reflect both sampling variability and imputation uncertainty.

Usage

RMitemInfitCutoffMI(
  mids_object,
  iterations = 500,
  parallel = TRUE,
  n_cores = NULL,
  verbose = FALSE,
  seed = NULL,
  cutoff_method = "hdci",
  hdci_width = 0.999
)

Arguments

mids_object	A mids object (multiply imputed dataset) as returned by mice::mice(). Each completed dataset must contain only the item response columns to be analysed (i.e., no ID or grouping variables). Items must be scored starting at 0 (non-negative integers).
iterations	Integer. Total number of simulation iterations to run across all imputations. These are distributed approximately evenly across the m imputed datasets (default 500).
parallel	Logical. Use parallel processing via mirai within each imputed dataset (default TRUE). Passed to RMitemInfitCutoff.
n_cores	Integer or NULL. Number of parallel workers. Passed to RMitemInfitCutoff.
verbose	Logical. Show progress messages (default FALSE).
seed	Integer or NULL. Master random seed for reproducibility. A unique per-imputation seed is derived from this value. See easyRasch2-reproducibility for what this guarantees and how it interacts with parallel.
cutoff_method	Character string specifying how cutoff intervals are computed from the stacked distribution. Either "hdci" (default) for the Highest Density Interval via ggdist::hdci(), or "quantile" for the 2.5th/97.5th percentiles via stats::quantile().
hdci_width	Numeric. Width of the HDCI when cutoff_method = "hdci". Default is 0.999 (99.9% HDCI). Ignored when cutoff_method = "quantile".

Details

The function completes each of the m imputed datasets via mice::complete(), then calls RMitemInfitCutoff on each one. The total number of iterations is split approximately evenly across imputations (i.e., each imputed dataset receives ceiling(iterations / m) or floor(iterations / m) iterations). The per-imputation simulation results are stacked into a single distribution from which cutoff intervals are computed, naturally incorporating imputation uncertainty.

Imputed datasets that cause model convergence failures are dropped with a warning. If all imputations fail, the function stops with an error.

The mice package must be installed (it is in Suggests, not Imports).

Value

A list with the same structure as RMitemInfitCutoff, so that the result can be passed directly to RMitemInfit, RMitemInfitMI, and RMitemInfitPlot:

results

data.frame with columns iteration, imputation, Item, InfitMSQ, OutfitMSQ — the stacked simulation results from all imputed datasets.

item_cutoffs

data.frame with per-item cutoff summaries: Item, infit_low, infit_high, outfit_low, outfit_high. Computed from the stacked distribution.

actual_iterations

Total number of successful iterations across all imputations.

sample_n

Number of rows (respondents) per imputed dataset.

sample_summary

Summary statistics of estimated person parameters from the first imputed dataset.

item_names

Character vector of item names.

cutoff_method

The method used to compute cutoffs.

hdci_width

The HDCI width used.

n_imputations

Number of imputed datasets used.

iterations_per_imputation

Integer vector of requested iterations per imputed dataset.

actual_iterations_per_imputation

Integer vector of successful iterations per imputed dataset.

Interval width under multiple imputation

hdci_width defaults to 0.999 here, not to the 0.95 used by RMitemInfitCutoff. RMitemInfitMI has no corrected-p-value path yet, so under imputation the interval is still the decision rule and keeps a width chosen for that job. Combining bootstrap p-values across imputations is planned but needs its own calibration, since Johansson (2026) covers complete data only.

See Also

RMitemInfitCutoff, RMitemInfitMI, RMitemInfitPlot

Examples

if (requireNamespace("mice", quietly = TRUE) &&
    requireNamespace("iarm", quietly = TRUE) &&
    requireNamespace("ggdist", quietly = TRUE)) {
  # Create example data with ~10% MCAR missingness
  set.seed(42)
  mat <- matrix(sample(0:1, 200 * 8, replace = TRUE), nrow = 200, ncol = 8)
  mat[sample(length(mat), round(0.10 * length(mat)))] <- NA
  sim_data <- as.data.frame(mat)
  colnames(sim_data) <- paste0("Item", 1:8)

  # mice's ordinal method (`polr`) requires the items to be ordered
  # factors, so code them as such before imputing. RMitemInfitCutoffMI()
  # converts the completed factors back to numeric internally.
  sim_data[] <- lapply(sim_data, function(x) factor(x, ordered = TRUE))

  # Impute (use more imputations, e.g. m = 5+, in real analyses)
  imp <- mice::mice(sim_data, m = 2, method = "polr", seed = 123,
                    printFlag = FALSE)

  # Compute simulation-based cutoffs across imputations
  # (use more iterations, e.g. 250+, in real analyses)
  cutoff_mi <- RMitemInfitCutoffMI(imp, iterations = 50, parallel = FALSE,
                                seed = 42)
  cutoff_mi$item_cutoffs

  # Use with RMitemInfitMI()
  RMitemInfitMI(imp, cutoff = cutoff_mi)
}

easyRasch2 documentation built on Sept. 13, 2026, 1:07 a.m.