Package {HighDimenCDM}

Type: Package
Title: Stochastic EM Algorithm for High-Dimensional Cognitive Diagnosis Models
Version: 0.1.0
Author: Yuxuan Mei [aut, cre], Wenchao Ma [aut], Kevin Wang [aut], Gongjun Xu [aut]
Maintainer: Yuxuan Mei <mei00060@umn.edu>
Description: Provides stochastic expectation-maximization (stEM) algorithms for estimating high-dimensional cognitive diagnosis models. The package implements stochastic EM algorithms for cognitive diagnosis models with a large number of attributes. It includes estimation functions and example datasets for model fitting and analysis. The methods are described in Ma, W., Wang, K., and Xu, G. (Accepted). "Parameter estimation of cognitive diagnosis models with stochastic EM algorithm." Behaviometrika.
License: GPL-3
LazyData: TRUE
Depends: R (≥ 3.5)
Imports: coda, GDINA, Rcpp (≥ 0.12.1)
LinkingTo: Rcpp, RcppArmadillo
Encoding: UTF-8
NeedsCompilation: yes
Config/roxygen2/version: 8.0.0
RoxygenNote: 8.0.0
Packaged: 2026-07-17 02:26:55 UTC; meiyuxuan
Repository: CRAN
Date/Publication: 2026-07-24 10:10:05 UTC

International Personality Item Pool (IPIP) data

Description

The IPIP-NEO-120 data are based on the public-domain personality inventory developed from the International Personality Item Pool (IPIP; Johnson, 2014).

Usage

IPIP

Format

A list of responses and Q-matrix with components:

dat

Responses of 4271 participants to 120 items.

Q

The 120 \times 30 Q-matrix.

Details

The data consist of responses from 4271 participants to 120 dichotomized items and the corresponding Q-matrix relating the 120 items to 30 personality facets.

Author(s)

Yuxuan Mei, The University of Minnesota, mei00060@umn.edu

References

Johnson, J. A. (2014). Measuring thirty facets of the five factor model with a 120-item public domain inventory: Development of the ipip-neo-120. Journal of Research in Personality, 51, 78–89.

Mental health CDM data (Tan2023)

Description

In this study, we analyze the responses of 719 college students (34.6% men, 83.9% White, and 16.3% first-year students) to 40 items measuring mental health problems such as alcohol-related problems, anxiety, hostility, and depression.

Usage

Tan2023

Format

A list of responses and Q-matrix with components:

dat

Responses of 719 participants to 40 items.

Q

The 40 \times 4 Q-matrix.

Details

The data were previously analyzed in Tan et al. (2022). The items measure four attributes: alcohol-related problems, anxiety, hostility, and depression.

Author(s)

Yuxuan Mei, The University of Minnesota, mei00060@umn.edu

References

Tan, Z., de la Torre, J., Ma, W., Huh, D., Larimer, M. E., & Mun, E.-Y. (2022). A tutorial on cognitive diagnosis modeling for characterizing mental health symptom profiles using existing item responses. Prevention Science.

Examples


data(Tan2023)
str(Tan2023)
fit <- stEM(
  dat = Tan2023$dat,
  Q = Tan2023$Q
  )

Batch variance calculation

Description

Computes variance of parameter estimates across batches in the stochastic EM algorithm.

Usage

batch.var(plist, n)

Arguments

plist

A list of parameter estimates collected across batches.

n

Number of batches

Value

A vector of batch variance estimates.

Author(s)

Wenchao Ma, The University of Minnesota, wma@umn.edu

Gibbs Sampling Kernel for stEM Algorithm

Description

Performs Gibbs sampling updates of attribute profiles and item parameters within the stochastic EM algorithm.

Usage

kernel(dat, Q, J, reduced.profiles, alpha, B, ip.only = TRUE)

Arguments

dat

A required N \times J data matrix of N examinees to J items. Missing values are currently not allowed.

Q

A required J \times K item-attribute association matrix, where J is the number of items and K is the number of attributes.

J

The number of items.

reduced.profiles

A list of reduced attribute profiles for each item.

alpha

The initial latent attribute profiles.

B

The number of Gibbs sampling iterations.

ip.only

If TRUE, only item parameter estimates are returned.

Value

a list with elements

ip

estimated item parameters from Gibbs sampling iterations

alpha

latent attribute profiles from the final Gibbs sampling iteration

a

saved latent attribute profiles across Gibbs sampling iterations

item.parm

final item parameter estimates

Author(s)

Wenchao Ma, The University of Minnesota, wma@umn.edu

Simulated CDM data with 10 attributes

Description

Simulated responses for 2000 examinees generated with the script in 'data-raw/simN2000K10.R' using 'GDINA::simGDINA()'.

Usage

simN2000K10

Format

A list with components:

data

A 2000 \times 30 response matrix.

Q

The 30 \times 10 Q-matrix used for simulation.

true.params

The true item category response probabilities used to generate the data.

Examples

data(simN2000K10)
str(simN2000K10)

Simulated CDM data with 15 attributes

Description

Simulated responses for 2000 examinees generated with the script in 'data-raw/simN2000K15.R' using 'GDINA::simGDINA()'.

Usage

simN2000K15

Format

A list with components:

data

A 2000 \times 45 response matrix.

Q

The 45 \times 15 Q-matrix used for simulation.

true.params

The true item category response probabilities used to generate the data.

Examples

data(simN2000K15)
str(simN2000K15)

Simulated CDM data with 20 attributes

Description

Simulated responses for 2000 examinees generated with the script in 'data-raw/simN2000K20.R' using 'GDINA::simGDINA()'.

Usage

simN2000K20

Format

A list with components:

data

A 2000 \times 60 response matrix.

Q

The 60 \times 20 Q-matrix used for simulation.

true.params

The true item category response probabilities used to generate the data.

Examples

data(simN2000K20)
str(simN2000K20)

Stochastic EM Algorithm for High-Dimensional Cognitive Diagnosis Models

Description

Estimates model parameters for cognitive diagnosis models using the stochastic expectation-maximization (stEM) algorithm with sequential Gibbs sampling.

Usage

stEM(
  dat,
  Q,
  item.parm = NULL,
  maxitr = 100,
  eps1 = 2,
  eps2 = 0.4,
  frac1 = 0.1,
  frac2 = 0.5,
  verbose = FALSE
)

Arguments

dat

A required N \times J data matrix of N examinees to J items. Missing values are not currently supported.

Q

A required J \times K item-attribute association matrix, where J is the number of items and K is the number of attributes.

item.parm

Optional initial item parameter estimates.

maxitr

Maximum number of iterations.

eps1

A positive convergence criterion used during the burn-in stage of the stochastic EM algorithm.

eps2

A positive convergence criterion used after the burn-in.

frac1

Proportion of the first part of the Markov chain used in the Geweke diagnostic.

frac2

Proportion of the last part of the Markov chain used in the Geweke diagnostic.

verbose

Logical; if TRUE, prints progress messages during the estimation process.

Value

a list with elements

catprob.parm

estimated item category response probabilities

ip

estimated final item parameters obtained by averaging across retained batches

alpha

estimated attribute mastery profiles

alpha.list

attribute mastery profile estimates from each retained batch

total.number.of.batch

total number of retained batches used for estimation

final.chain

final length of the Markov chain

burn.in.size

the number of burn-in iterations discarded

plist

estimated item parameters from each retained batch

Examples


dat <- simN2000K20$data
Q <- simN2000K20$Q
fit <- stEM(dat = dat, Q = Q)
fit