scale_epmfd: Scale polytomous item responses

Description

scale_epmfd() fits either a parametric graded response model (GRM, via mirt) or a nonparametric Mokken model (via mokken) to polytomous item-response data and filters out weak items based on user-specified thresholds.

Usage

scale_epmfd(
  object,
  method = c("auto", "mirt", "mokken"),
  a_thr = 0.5,
  H_thr = 0.3
)

Value

An object of class epmfd_scaled, a list containing:

raw: the original epmfd_raw object
method: scaling method actually used ("mirt" or "mokken")
kept: names of items retained
removed: names of items removed
model: fitted GRM model (for "mirt"), else NULL
ai: item parameter estimates (for "mirt")
a_thr: discrimination threshold used (for "mirt")
model_fit: results of mirt::M2() (if available)
Hi: vector of item scalability coefficients (for "mokken")
H_thr: scalability threshold used (for "mokken")
items: the vector containing all items names.

Arguments

object

An epmfd_raw object created by load_epmfd().

method

Scaling method. One of:

"mirt": fit a one-factor graded response model (GRM).
"mokken": perform nonparametric Mokken scale analysis.
"auto" (default): choose based on sample size (n >= 500 → GRM, otherwise Mokken).

a_thr

Numeric. Threshold for item discrimination parameter a when using GRM (default = 0.5). Items with a < a_thr are removed.

H_thr

Numeric. Threshold for item scalability coefficient H_i when using Mokken analysis (default = 0.3). Items with H_i < H_thr are removed.

Details

The function converts ordered factors to numeric before analysis.

For GRM (mirt), items are filtered by their discrimination parameter a. The overall model fit is attempted using mirt::M2(); if this fails (e.g., due to insufficient df), a warning is issued and model_fit = NULL.
For Mokken, item scalability coefficients H_i are computed and compared to H_thr.

Examples

Run this code

library(epmfd)
data<-load_epmfd(sampledata)
scale_epmfd(data)