In susanne-frick/MFCblockInfo: compute block information for multidimensional forced-choice questionnaires

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)
knitr::opts_knit$set(
  root.dir = "../data_raw"
)

Dataset

This example uses a dataset from Wetzel & Frick (2020) which is available on OSF: \url{https://osf.io/z9w6s/}

library(MFCblockInfo)

Questionnaire design

Specify the loading matrix with items in rows and traits in columns, 1=positively keyed item, -1=negatively keyed item

design.load.MT <- read.table("design_load_bft-matched.txt", sep="\t", header=TRUE)
head(design.load.MT)

The TIRT design matrix of pairwise comparisons (termed $\Lambda$ in \textcite{brown2011}) can be created with create.design.mat, given number of blocks (K) and block size (nb).

K <- 20
nb <- 3
design.mat <- create.design.mat(K=K, nb=nb)
design.mat[1:6,1:6]

Read and format Mplus parameter estimates

mt <- MplusAutomation::readModels("tirt_bft-matched_intercepts.out", what=c("parameters"))
mt.pars <- mt$parameters$unstandardized

Make sure to extract unstandardized parameter estimates.

#theta
scores.mt <- read.table("scores_bft-matched_intercepts.dat", quote="\"", comment.char="")[,61:66] # last column is ID
colnames(scores.mt) <- c("N","E","O","A","C","id_all")

#pair thresholds
thresh.mt <- mt.pars[mt.pars$paramHeader=="Thresholds",]$est

Use load2mat and uni2mat to convert estimates for loadings and uniquenesses to matrices, respectively. (uni2mat assumes a diagonal matrix of error covariances, i.e. errors do not covary between items.)

#loadings
load.mt <- load2mat(mt.pars, design.load.MT, design.mat)
head(load.mt)

#uniqenesses
uni.mt <- uni2mat(mt.pars, "^MT")
uni.mt[1:6,1:6]

Combine these objects to a list, for convenience. Thresholds should be inverted ($\times(-1)$) to intercepts.

#combine to data
# ! intercepts = -thresholds
data.mt <- list(traits=scores.mt, int=-thresh.mt, loads=load.mt, uni=uni.mt, K=K, nb=nb)

The two dataset is available via data("data.mt").

susanne-frick/MFCblockInfo documentation built on Oct. 20, 2024, 8:26 p.m.