epmfd (version 1.1.1)

misfit_epmfd: Compute person-fit statistics (polytomous data)

Description

misfit_epmfd() computes selected person-fit statistics for polytomous responses and returns an epmfd_misfit object with scores, thresholds, and logical flags per person.

Usage

misfit_epmfd(object, stats = c("auto", "lpz", "Gnp", "U3p"), cut.off = "auto")

Value

An epmfd_misfit list with:

• scaled: the input epmfd_scaled object

• method: detected method ("mirt" or "mokken")

• stats: actually computed statistics (subset of c("lpz","Gnp","U3p"))

• thresholds: named list of lists with value and tail

• scores: named list of numeric score vectors per statistic

• table: a tibble with id, one logical column per statistic, misfit_any (OR over selected stats), and misfit_final (see Details)

Arguments

object

An epmfd_scaled object (output of your scaling step).

stats

Character vector choosing which statistics to compute. Allowed values: "auto", "lpz", "Gnp", "U3p". If "auto" is present, the set is chosen based on the detected scaling method:

• for "mirt": c("lpz","Gnp","U3p")

• for "mokken": c("Gnp","U3p")

cut.off

Cut-off for Gnp and U3p. Either "auto" (default; uses PerFit’s cutoff() with its implied tail), or a single numeric value (interpreted with tail "upper" for both Gnp and U3p). lpz uses a fixed lower-tail cut-off of -1.645.

Details

Auto vs manual decision for misfit_final:

• If stats contains "auto":

  • for "mokken": misfit_final = Gnp & U3p

  • for "mirt": misfit_final = (lpz & Gnp & U3p) if any such rows exist, otherwise fallback to lpz only.

• If stats is manual (no "auto"): misfit_final is the AND over the selected statistics (if only one selected, it is used directly).

Polytomous PerFit statistics assume a common design K (number of categories) across items. This function uses object$raw$K as the global design K and maps item responses to 0..K-1 without compressing per-item gaps (unused categories are allowed and do not trigger an error).

Examples

library(epmfd)
data<-load_epmfd(sampledata)
scaling_data<-scale_epmfd(data)
misfit_result<-misfit_epmfd(scaling_data)
misfit_result
plot_misfit(misfit_result,threeD=TRUE)