Create table with Rasch PCM model item fit values for each item.

Defaults to using conditional estimates for MSQ values (Müller, 2020) estimated using the iarm package. Use method = "unconditional" for the "old" unconditional MSQ values (using eRm).

Usage

RIitemfitPCM(
  dfin,
  samplesize,
  nsamples,
  zstd_min = -1.96,
  zstd_max = 1.96,
  msq_min = 0.7,
  msq_max = 1.3,
  fontsize = 15,
  fontfamily = "Lato",
  output = "table",
  tbl_width = 65,
  method = "conditional",
  simcut = FALSE,
  gf
)

Arguments

dfin

Dataframe with item data only

samplesize

Desired sample size in multisampling (recommended range 200-500)

nsamples

Desired number of samples (recommended range 10-50)

zstd_min

Lower cutoff level for ZSTD

zstd_max

Upper cutoff level for ZSTD

msq_min

Lower cutoff level for MSQ

msq_max

Upper cutoff level for MSQ

fontsize

Set fontsize for table

fontfamily

Set font family for table

output

Defaults to output a table. Optional "dataframe" or "quarto"

tbl_width

Set table width in percent

method

Defaults to "conditional". Optional "unconditional"

simcut

Set to TRUE if you want to use simulation based cutoff values

gf

The output object from RIgetfit() is needed when simcut = TRUE

Details

Since version 0.2.0 (2024-08-15), it is highly recommended to replace rule-of-thumb cutoff values with simulation based cutoffs. See details in ?RIgetfit() for an easy way to get and set appropriate cutoff values.

ZSTD is inflated with large samples (N > 500). There is an optional function to use a reduced sample size and run analysis using multiple random samples to get the average ZSTD for each item over all runs.

If you are using Quarto, the YAML execute setting "cache: yes" will be a useful chunk option to speed things up if you render often. 30-50 samples seems to produce stable output, but 4-8 is probably sufficient for a quick look at the approximate ZSTD statistics. It is recommended to use sample size 200-500, based on Hagell & Westergren (2016) & Müller (2020).