Help for package GMLTM

Type:

Package

Title:

Generalized Multicomponent Latent Trait Model for Diagnosis

Version:

2.0.0

Description:

Provides Bayesian estimation of Item Response Theory models that decompose item difficulty into cognitive operations or rules. Implements the Linear Logistic Test Model (LLTM; Fischer (1973) <doi:10.1016/0001-6918(73)90003-6>), the Multicomponent Latent Trait Model for Diagnosis (MLTM-D; Embretson and Yang (2013) <doi:10.1007/s11336-012-9296-y>), and the Generalized Multicomponent Latent Trait Model for Diagnosis (GMLTM-D; Ramirez et al. (2024) <doi:10.3390/jintelligence12070067>), including a variant with correlated latent components. All models are estimated via Hamiltonian Monte Carlo using 'Stan' through the 'rstan' interface. Includes tools for prior predictive checks (Gelman et al., 2020), model validation, conditional reliability estimation, examinee mastery classification following Embretson (2019) <doi:10.1007/978-3-030-05584-4_9>, and individual diagnostic reports at the rule and component level. Supports user-defined prior distributions for all model parameters.

License:

GPL (≥ 3)

URL:

https://github.com/Eduar-Ramirez/GMLTM-D

BugReports:

https://github.com/Eduar-Ramirez/GMLTM-D/issues

Encoding:

UTF-8

LazyData:

true

Depends:

R (≥ 4.1.0)

Imports:

rstan (≥ 2.21.0), ggplot2, gridExtra, grid, utils, parallel, loo, RColorBrewer

Suggests:

testthat (≥ 3.0.0), knitr, rmarkdown

SystemRequirements:

C++17, GNU make

VignetteBuilder:

knitr

Language:

en-US

RoxygenNote:

7.3.2

NeedsCompilation:

Packaged:

2026-08-27 10:57:09 UTC; eduarramirezriveros

Author:

Eduar Ramirez [aut, cre], Marcos Jimenez [aut], Vithor R. Franco [aut], Jesus Alvarado [aut]

Maintainer:

Eduar Ramirez <edrami02@ucm.es>

Repository:

CRAN

Date/Publication:

2026-08-27 12:10:08 UTC

GMLTM: Generalized Multicomponent Latent Trait Model for Diagnosis

Description

Bayesian estimation of Item Response Theory models that decompose item difficulty into cognitive operations or rules: the Linear Logistic Test Model (LLTM), the Multicomponent Latent Trait Model for Diagnosis (MLTM-D), and the Generalized Multicomponent Latent Trait Model for Diagnosis (GMLTM-D). All models are fit via Hamiltonian Monte Carlo using rstan.

Model fitting

LLTM, MLTM, GMLTM, and GMLTM_corr fit the four supported models and share the same data/Q/priors interface. See vignette("GMLTM-intro", package = "GMLTM") for a worked example.

Reliability

reliability and enhanced_mltm_reliability estimate marginal and component-level reliability from a fitted model. conditional_reliability_tif and compare_conditional_reliability estimate reliability as a function of the ability scale. quick_reliability_check, check_reliability_data_quality, and demo_reliability_analysis support quick diagnostics.

Model checking and diagnostics

ppchecks and marginal_Pchecks perform posterior predictive checks; compute_model_validation computes WAIC/LOO-based validation statistics; extract_correlation inspects correlations among latent components; student_report builds a per-examinee mastery report.

Plotting and item structure

plot_ICC_grouped and plot_ICC_individual draw item characteristic curves; generate_Q_with_interactions expands a Q-matrix with rule-interaction columns.

Author(s)

Maintainer: Eduar Ramirez edrami02@ucm.es

Authors:

Marcos Jimenez marcosjnezhquez@gmail.com
Vithor R. Franco
Jesus Alvarado

References

Fischer, G. H. (1973). The linear logistic test model as an instrument in educational research. Acta Psychologica, 37(6), 359–374. doi:10.1016/0001-6918(73)90003-6

Embretson, S. E., & Yang, X. (2013). A multicomponent latent trait model for diagnosis. Psychometrika, 78, 14–36. doi:10.1007/s11336-012-9296-y

Embretson, S. E. (2019). Diagnostic modeling of skill hierarchies and cognitive processes with MLTM-D. In M. von Davier & Y.-S. Lee (Eds.), Handbook of Diagnostic Classification Models (pp. 185–208). Springer. doi:10.1007/978-3-030-05584-4_9

Ramirez, E. S., Jimenez, M., Franco, V. R., & Alvarado, J. M. (2024). Delving into the complexity of analogical reasoning: A detailed exploration with the Generalized Multicomponent Latent Trait Model for Diagnosis. Journal of Intelligence, 12, 67. doi:10.3390/jintelligence12070067

The Generalized Multicomponent Latent Trait Model for Diagnosis

Description

Estimate the parameters of the GMLTM-D via Bayesian Hamiltonian Monte Carlo.

Usage

GMLTM(
  data,
  Q,
  components,
  iters = 2000,
  chains = 2,
  iter_warmup = 1000,
  quantiles = c(0.025, 0.5, 0.975),
  cores = parallel::detectCores() - 1,
  priors = list(theta = list(mu = 0, sigma = 1), eta = list(mu = 0, sigma = 1), alpha =
    list(mu = 0, sigma = 1, family = "normal"), c = list(shape1 = 3, shape2 = 20)),
  ...
)

Arguments

data

An n \times p matrix or data.frame of binary responses (rows = subjects, columns = items).

Q

A p \times K matrix specifying which cognitive rules each item requires (Q-matrix).

components

A named list grouping rules into components. Each element is a numeric vector of rule indices belonging to that component. Example: list(global = c(1,2,3), local = c(4,5)).

iters

Number of post-warmup MCMC iterations per chain. Default is 2000.

chains

Number of Markov chains. Default is 2.

iter_warmup

Number of warmup iterations per chain. Default is 1000.

quantiles

Numeric vector of probabilities for posterior quantiles. Default is c(0.025, 0.50, 0.975).

cores

Number of CPU cores for parallel chains. Default is parallel::detectCores() - 1.

priors

A named list of prior hyperparameters with elements theta, eta, alpha, and c. For Normal parameters supply mu and sigma; for the guessing parameter supply shape1 and shape2 (Beta prior). priors$alpha additionally accepts family, either "normal" (default; a half-Normal, i.e. Normal truncated at 0) or "lognormal" (a Log-Normal prior on \alpha, using the same mu/sigma as the location/scale of the underlying Normal on the log scale). Unspecified elements retain defaults.

...

Additional arguments passed to rstan::sampling().

Details

GMLTM estimates the Generalized Multicomponent Latent Trait Model for Diagnosis (GMLTM-D; Ramirez et al., 2024) in its Bayesian version. This model analyses items composed of cognitive rules or operations, incorporating three IRT parameters. Rules can be grouped into distinct components.

Prior distributions: Ability (\theta) and rule difficulty (\eta) receive Normal priors. Discrimination (\alpha) receives either a half-Normal prior (priors$alpha$family = "normal", default) or a Log-Normal prior (priors$alpha$family = "lognormal"). Guessing (c) receives a Beta prior.

Value

A list of class "GMLTM" with elements:

EAP: Posterior mean estimates: theta, alpha, eta, beta, guessing.
quantiles: Posterior credible intervals for each parameter.
posterior: Full posterior samples and derived quantities.
fit: The stanfit object from rstan::sampling.
data: The original data matrix.
priors: The prior hyperparameters used.

References

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  fit$EAP$eta
  reliability(fit)

GMLTM-D with correlated latent components (general Sigma)

Description

Variant of GMLTM that replaces the independent prior on \theta (\Sigma = \sigma^2 I) with a multivariate Normal prior with a general correlation matrix \Sigma, estimated from the data via its Cholesky factor (non-centered parameterization with an LKJ prior on the correlation). This corresponds to innovation (iii) of the GMLTM-D as formulated in Ramirez et al. (2024): allowing correlations between components to reflect the natural interdependence between cognitive abilities, instead of assuming independent latent traits.

Usage

GMLTM_corr(
  data,
  Q,
  components,
  iters = 2000,
  chains = 2,
  iter_warmup = 1000,
  quantiles = c(0.025, 0.5, 0.975),
  cores = parallel::detectCores() - 1,
  priors = list(theta = list(mu = 0, sigma = 1), eta = list(mu = 0, sigma = 1), alpha =
    list(mu = 0, sigma = 1, family = "normal"), c = list(shape1 = 3, shape2 = 20)),
  lkj_eta = 1,
  ...
)

Arguments

data

An n \times p matrix or data.frame of binary responses (rows = subjects, columns = items).

Q

A p \times K matrix specifying which cognitive rules each item requires (Q-matrix).

components

A named list grouping rules into components. Each element is a numeric vector of rule indices belonging to that component. Example: list(global = c(1,2,3), local = c(4,5)).

iters

Number of post-warmup MCMC iterations per chain. Default is 2000.

chains

Number of Markov chains. Default is 2.

iter_warmup

Number of warmup iterations per chain. Default is 1000.

quantiles

Numeric vector of probabilities for posterior quantiles. Default is c(0.025, 0.50, 0.975).

cores

Number of CPU cores for parallel chains. Default is parallel::detectCores() - 1.

priors

lkj_eta

Concentration parameter of the LKJ prior on the correlation matrix \Sigma. lkj_eta = 1 (default) is uniform over the space of correlation matrices; lkj_eta > 1 concentrates the prior towards \Sigma = I (low correlations); lkj_eta < 1 towards more extreme correlations.

...

Additional arguments passed to rstan::sampling().

Details

GMLTM_corr estimates the same likelihood as GMLTM (three IRT parameters per item/component: discrimination \alpha, guessing c, and rule difficulty \eta/\beta), but replaces the prior on the subject-level ability matrix \theta with \theta_i \sim N_M(0, \Sigma), where \Sigma is a correlation matrix among the M cognitive components. \Sigma is parameterized through its Cholesky factor L_Omega (a cholesky_factor_corr), with theta = t(L_Omega %*% theta_raw) and theta_raw ~ N(0, 1) i.i.d. (non-centered parameterization, which improves posterior geometry relative to sampling \theta directly from the multivariate Normal). priors$theta is accepted for interface compatibility with GMLTM but is not used: the scale of \theta is fixed to 1 by construction (diagonal of \Sigma), and its prior is governed entirely by lkj_eta.

Value

A list of class c("GMLTM_corr", "GMLTM") with elements:

EAP: Posterior mean estimates: theta, alpha, eta, beta, guessing, and Sigma (the estimated M x M correlation matrix among components).
quantiles: Posterior credible intervals for each parameter, including Sigma.
posterior: Full posterior samples and derived quantities (including loglik, used by compute_model_validation).
fit: The stanfit object from rstan::sampling.
data: The original data matrix.
priors: The prior hyperparameters used.
lkj_eta: The LKJ concentration parameter used.

References

Embretson, S. E., & Yang, X. (2013). A multicomponent latent trait model for diagnosis. Psychometrika, 78, 14–36. doi:10.1007/s11336-012-9296-y

Lewandowski, D., Kurowicka, D., & Joe, H. (2009). Generating random correlation matrices based on vines and extended onion method. Journal of Multivariate Analysis, 100(9), 1989–2001. doi:10.1016/j.jmva.2009.04.008

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM_corr(analogy, Q, components,
                     iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  fit$EAP$eta
  fit$EAP$Sigma

The Linear Logistic Test Model

Description

Estimate the parameters of the LLTM via Bayesian Hamiltonian Monte Carlo.

Usage

LLTM(
  data,
  Q,
  iters = 2000,
  chains = 2,
  iter_warmup = 1000,
  quantiles = c(0.025, 0.5, 0.975),
  cores = parallel::detectCores() - 1,
  priors = list(theta = list(mu = 0, sigma = 1), eta = list(mu = 0, sigma = 1)),
  ...
)

Arguments

data

An n \times p matrix or data.frame of binary responses (rows = subjects, columns = items).

Q

A p \times K matrix specifying which cognitive rules each item requires (Q-matrix).

iters

Number of post-warmup MCMC iterations per chain. Default is 2000.

chains

Number of Markov chains. Default is 2.

iter_warmup

Number of warmup iterations per chain. Default is 1000.

quantiles

Numeric vector of probabilities for posterior quantiles. Default is c(0.025, 0.50, 0.975).

cores

Number of CPU cores for parallel chains. Default is parallel::detectCores() - 1.

priors

A named list of prior hyperparameters. Each element is a named list with mu and sigma. Available parameters: theta (ability, Normal prior) and eta (rule difficulty, Normal prior). Unspecified elements retain defaults. Example: priors = list(eta = list(sigma = 3)).

...

Additional arguments passed to rstan::sampling().

Details

LLTM estimates the Bayesian version of the Linear Logistic Test Model (Fischer, 1973), which extends the Rasch model by decomposing item difficulty into cognitive rules. Item difficulty is expressed as \beta_i = \mathbf{q}_i^\top \boldsymbol{\eta}, where \mathbf{q}_i is the i-th row of Q and \boldsymbol{\eta} is the vector of rule difficulty parameters.

Prior distributions: Ability (\theta) and rule difficulty (\eta) receive Normal priors. Prior sensitivity analysis is recommended.

Value

A list of class "LLTM" with elements:

EAP: Posterior mean estimates: theta, eta, beta.
quantiles: Posterior credible intervals for each parameter.
posterior: Full posterior samples and derived quantities.
fit: The stanfit object from rstan::sampling.
data: The original data matrix.
priors: The prior hyperparameters used.

References

Fischer, G. H. (1973). The linear logistic test model as an instrument in educational research. Acta Psychologica, 37(6), 359–374. doi:10.1016/0001-6918(73)90003-6

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  fit <- LLTM(analogy, Q, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  fit$EAP$eta
  reliability(fit)

The Multicomponent Latent Trait Model for Diagnosis

Description

Estimate the parameters of the MLTM-D via Bayesian Hamiltonian Monte Carlo.

Usage

MLTM(
  data,
  Q,
  components,
  iters = 2000,
  chains = 2,
  iter_warmup = 1000,
  quantiles = c(0.025, 0.5, 0.975),
  cores = parallel::detectCores() - 1,
  priors = list(theta = list(mu = 0, sigma = 1), eta = list(mu = 0, sigma = 1), alpha =
    list(mu = 0, sigma = 1)),
  ...
)

Arguments

data

An n \times p matrix or data.frame of binary responses (rows = subjects, columns = items).

Q

A p \times K matrix specifying which cognitive rules each item requires (Q-matrix).

components

A named list grouping rules into components. Each element is a numeric vector of rule indices belonging to that component. Example: list(global = c(1,2,3), local = c(4,5)).

iters

Number of post-warmup MCMC iterations per chain. Default is 2000.

chains

Number of Markov chains. Default is 2.

iter_warmup

Number of warmup iterations per chain. Default is 1000.

quantiles

Numeric vector of probabilities for posterior quantiles. Default is c(0.025, 0.50, 0.975).

cores

Number of CPU cores for parallel chains. Default is parallel::detectCores() - 1.

priors

A named list of prior hyperparameters with elements theta, eta, and alpha. Each is a list with mu and sigma. alpha uses a half-Normal prior (truncated at 0). Unspecified elements retain defaults.

...

Additional arguments passed to rstan::sampling().

Details

MLTM estimates the Bayesian version of the Multicomponent Latent Trait Model for Diagnosis (MLTM-D; Embretson & Yang, 2013). This noncompensatory model specifies a hierarchical relationship between components and rules.

Prior distributions: Ability (\theta) and rule difficulty (\eta) receive Normal priors. Discrimination (\alpha) receives a half-Normal prior.

Value

A list of class "MLTM" with elements:

EAP: Posterior mean estimates: theta, alpha, eta, beta.
quantiles: Posterior credible intervals for each parameter.
posterior: Full posterior samples and derived quantities.
fit: The stanfit object from rstan::sampling.
data: The original data matrix.
priors: The prior hyperparameters used.

References

Embretson, S. E., & Yang, X. (2013). A multicomponent latent trait model for diagnosis. Psychometrika, 78, 14–36. doi:10.1007/s11336-012-9296-y

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- MLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  fit$EAP$eta
  reliability(fit)

Analogy items dataset

Description

Binary item response data from a figural analogies test used to illustrate the LLTM, MLTM-D, and GMLTM-D models.

Usage

analogy

Format

A matrix with 149 rows (subjects) and 27 columns (items), where each cell contains a binary response (0 = incorrect, 1 = correct).

Source

Blum, D., Holling, H., Galibert, M. S., & Forthmann, B. (2016). Task difficulty prediction of figural analogies. Intelligence, 56, 72–81. doi:10.1016/j.intell.2016.03.001

References

Fast Bayesian Marginal Reliability

Description

Optimized computation of marginal reliability with uncertainty quantification.

Usage

bayesian_reliability_fast(theta_samples, EAP_theta)

Arguments

theta_samples

Array of posterior samples [iter, person, component].

EAP_theta

Matrix of EAP estimates [person, component].

Value

List of reliability estimates per component.

Check Data Quality for Reliability Analysis

Description

Verifies if the data is suitable for robust reliability analysis.

Usage

check_reliability_data_quality(fit)

Arguments

fit

Fitted model

Value

List with data quality diagnostics

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  dq <- check_reliability_data_quality(fit)
  print(dq)

Compare Conditional Reliability Between Components at Specific Theta Values

Description

Compare Conditional Reliability Between Components at Specific Theta Values

Usage

compare_conditional_reliability(cond_rel_obj, theta_points = c(-1, 0, 1))

Arguments

cond_rel_obj

An object of class conditional_reliability_tif returned by conditional_reliability_tif.

theta_points

Numeric vector of theta values at which to compare components. Default is c(-1, 0, 1).

Value

Invisibly returns a data frame with columns theta, reliability, and se evaluated at the requested theta_points. Called primarily for its side effect of printing a formatted summary table to the console.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  cond_rel <- conditional_reliability_tif(fit, theta_range = seq(-2, 2, 0.5))
  compare_conditional_reliability(cond_rel, theta_points = c(-1, 0, 1))

Basic Diagnostics (No External Dependencies)

Description

Computes basic MCMC diagnostics without external packages.

Usage

compute_basic_diagnostics(theta_samples)

Arguments

theta_samples

Array of posterior samples.

Value

List of diagnostic statistics.

Compute LOO and WAIC for GMLTM models

Description

This function extracts the log-likelihood from a GMLTM model and computes the Leave-One-Out Cross-Validation (LOO) and the Widely Applicable Information Criterion (WAIC). LOO is a Bayesian model comparison metric based on Pareto-smoothed importance sampling, while WAIC is a fully Bayesian criterion that estimates predictive accuracy.

Usage

compute_model_validation(fit)

Arguments

fit

A fitted GMLTM model or a list of fitted models. Models fitted with GMLTM_corr (correlated \Sigma), GMLTM (\Sigma = \sigma^2 I), MLTM, and LLTM can be freely mixed in the same list; each is labeled by model class in the output.

Value

If a single model is provided, returns a list with LOO and WAIC results. If multiple models are provided, returns a summary table with key LOO and WAIC indices, labeled by model type (e.g. "GMLTM-D (Sigma libre)" for GMLTM_corr fits, "GMLTM-D (Sigma=I)" for GMLTM fits).

References

Vehtari, A., Gelman, A., & Gabry, J. (2017). Practical Bayesian model evaluation using LOO-CV and WAIC. Statistics and Computing, 27(5), 1413–1432. doi:10.1007/s11222-016-9696-4

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit1 <- GMLTM(data = analogy, Q = Q, components = components,
                iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  compute_model_validation(fit1)

Fast Conditional Reliability (Simplified)

Description

Simplified conditional reliability using quantile-based approach.

Usage

conditional_reliability_fast(theta_samples, EAP_theta, n_bins = 5)

Arguments

theta_samples

Array of posterior samples.

EAP_theta

Matrix of EAP estimates.

n_bins

Integer. Number of ability bins (default 5 for speed).

Value

List of conditional reliability by bins.

Conditional Reliability based on Test Information Function (TIF)

Description

Calculates conditional reliability using Test Information Function for 3-parameter MLTM-D models. This approach is more precise than quantile-based partitioning methods.

Usage

conditional_reliability_tif(
  fit,
  theta_range = seq(-3, 3, 0.2),
  component = NULL,
  n_samples = 1000
)

Arguments

fit

Fitted model with \alpha, \beta, guessing parameters

theta_range

Vector of \theta values where to evaluate reliability

component

Integer or character. Specific component to analyze

n_samples

Integer. Number of posterior samples to use

Value

A list of class "conditional_reliability_tif" with elements:

theta: Numeric vector of theta values.
reliability: Numeric vector of reliability estimates at each theta value.
information: Numeric vector of test information values at each theta value.
component: Integer indicating the model component.
fit: The original fitted model object.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  cond_rel <- conditional_reliability_tif(fit, theta_range = seq(-2, 2, 0.5))
  cond_rel$global$optimal_theta

Step-by-step Usage Example

Description

Demonstration function to show how to use the optimized functions.

Usage

demo_reliability_analysis(fit)

Arguments

fit

Fitted model

Value

Invisibly returns a list with the reliability estimates computed at each step of the analysis. Called primarily for its side effect of printing a step-by-step explanation to the console.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  demo_reliability_analysis(fit)

Enhanced Reliability Analysis for GMLTM-D Models

Description

Provides comprehensive reliability analysis for General Multicomponent Latent Trait Models for Diagnosis (GMLTM-D) using Bayesian posterior distributions. Optimized for speed and minimal dependencies.

Usage

enhanced_mltm_reliability(
  fit,
  include_conditional = FALSE,
  include_hierarchical = TRUE,
  include_comparisons = TRUE,
  n_samples = NULL
)

Arguments

fit

A fitted GMLTM, MLTM, or LLTM model object containing posterior samples of theta parameters.

include_conditional

Logical. Whether to compute conditional reliability estimates across ability levels. Default is FALSE for speed.

include_hierarchical

Logical. Whether to compute hierarchical reliability for the general factor. Default is TRUE.

include_comparisons

Logical. Whether to perform Bayesian comparisons between components. Default is TRUE.

n_samples

Integer. Number of posterior samples to use (for speed control). If NULL, uses all available samples.

Value

An object of class enhanced_mltm_reliability.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  rel <- enhanced_mltm_reliability(fit, n_samples = 100)
  rel$marginal_robust

Export Reliability Results

Description

Exports results in tabular format for publications.

Usage

export_reliability_results(reliability_obj, file_name = NULL)

Arguments

reliability_obj

Object of class enhanced_mltm_reliability

file_name

File name (optional)

Value

data.frame with tabulated results

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  rel <- enhanced_mltm_reliability(fit, n_samples = 100)
  export_reliability_results(rel)

Extract and Summarize the Latent Correlation Matrix from a GMLTM_corr Fit

Description

Extracts the posterior correlation matrix \Sigma among cognitive components from a model fitted with GMLTM_corr, together with pairwise credible intervals and a heatmap for visual inspection.

Usage

extract_correlation(fit, credible = 0.95, plot = TRUE)

Arguments

fit

An object of class "GMLTM_corr", as returned by GMLTM_corr.

credible

Width of the credible interval for each pairwise correlation, e.g. 0.95 for a 95% interval. Default is 0.95.

plot

Logical. If TRUE (default), draws a correlation heatmap of the posterior mean \Sigma on the active graphics device, using the same color scheme as the correlation diagnostics in generate_Q_with_interactions.

Details

The pairwise credible intervals are computed from the posterior draws of Sigma in fit$fit, extracted with rstan::extract() (not from fit$EAP$Sigma, which only holds the posterior mean). A pair of components is flagged as showing evidence of a real (non-zero) correlation when its credible interval excludes 0.

Value

A list of class "GMLTM_correlation" with elements:

Sigma: The M x M posterior mean correlation matrix (fit$EAP$Sigma).
pairs: A data.frame with one row per pair of components: component_1, component_2, estimate, lower, upper, and excludes_zero (logical).
credible: The credible-interval width used.
plot: A function that redraws the heatmap when called with no arguments, or NULL if plot = FALSE.

References

Embretson, S. E., & Yang, X. (2013). A multicomponent latent trait model for diagnosis. Psychometrika, 78, 14–36. doi:10.1007/s11336-012-9296-y

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM_corr(analogy, Q, components,
                     iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  corr <- extract_correlation(fit)
  corr$pairs

Generate an Extended Q-matrix with Rule Interactions and Collinearity Diagnostics

Description

This function generates interaction terms between rules within the same component, extends the Q-matrix, and evaluates the resulting matrix for collinearity issues using eigenvalues, condition indices, and variance inflation factors (VIF). If severe collinearity is detected, it attempts to iteratively remove problematic interaction terms while keeping the original rules untouched.

Usage

generate_Q_with_interactions(
  Q,
  M_list,
  max_condition_index = 30,
  min_eigenvalue = 0.1,
  plot_diagnostics = TRUE,
  verbose = TRUE,
  save_to_global = TRUE
)

Arguments

Q

A binary matrix of items by rules (original Q-matrix). Each row represents an item and each column represents a rule. Values should be 0 or 1.

M_list

A list where each element contains the indices of rules that belong to the same component/dimension. For example, list(c(1,2,3), c(4,5)) indicates that rules 1,2,3 belong to component 1 and rules 4,5 belong to component 2.

max_condition_index

Numeric. Maximum acceptable condition index. Default is 30. Values above this threshold indicate severe collinearity.

min_eigenvalue

Numeric. Minimum acceptable eigenvalue. Default is 0.1. Values below this threshold may indicate linear dependence.

plot_diagnostics

Logical. Whether to generate diagnostic plots. Default is TRUE.

verbose

Logical. Whether to print detailed diagnostic information. Default is TRUE.

save_to_global

Logical. Whether to save results to global environment. Default is TRUE.

Details

The function performs the following steps:

Validates input parameters
Generates interaction terms for rules within the same component
Performs collinearity diagnostics using multiple methods
Attempts to resolve severe collinearity by removing problematic interactions
Generates diagnostic plots and summaries

Collinearity is assessed using:

Condition indices (based on eigenvalues of correlation matrix)
Variance Inflation Factors (VIF)
Matrix rank assessment
Eigenvalue analysis

Value

A list containing:

Q_extended

The extended Q-matrix with interaction terms

M_list_extended

Updated component list including interaction terms

diagnostics

List with collinearity diagnostics

removed_interactions

Vector of removed interaction names (if any)

plots

List of diagnostic plots (if plot_diagnostics = TRUE)

References

Belsley, D. A., Kuh, E., & Welsch, R. E. (1980). Regression diagnostics: Identifying influential data and sources of collinearity. John Wiley & Sons. O'Brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality & Quantity, 41(5), 673-690. Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1995). Multivariate data analysis. Prentice Hall.

Examples


# Create a sample Q-matrix (5 items, 4 rules)
Q <- matrix(c(1,1,0,0,0,
              1,0,1,0,0,
              0,1,1,0,0,
              0,0,0,1,1), nrow=5, ncol=4, byrow=FALSE)

# Define components (rules 1-2 in component 1, rules 3-4 in component 2)
M_list <- list(c(1,2), c(3,4))

# Generate extended Q-matrix with interactions
result <- generate_Q_with_interactions(Q, M_list)

# Access results
extended_Q <- result$Q_extended
diagnostics <- result$diagnostics
plots <- result$plots

Fast Hierarchical Reliability

Description

Optimized hierarchical reliability computation.

Usage

hierarchical_reliability_fast(theta_samples)

Arguments

theta_samples

Array of posterior samples.

Value

List with component and general factor reliability.

Deprecated Alias for `student_report`

Description

informe_estudiante is a deprecated alias for student_report, kept for backward compatibility with GMLTM (< 2.0.0). It will be removed in a future release; use student_report instead.

Usage

informe_estudiante(...)

Arguments

...

Arguments passed on to student_report.

Value

See student_report.

Integration with Enhanced Reliability Analysis

Description

Integrates conditional TIF analysis with existing reliability functions.

Usage

integrate_with_enhanced_reliability(
  fit,
  include_conditional_tif = TRUE,
  theta_range = seq(-3, 3, 0.1),
  ...
)

Arguments

fit

A fitted GMLTM, MLTM, or LLTM model object.

include_conditional_tif

Logical. Whether to compute conditional TIF-based reliability. Default is TRUE.

theta_range

Numeric vector of theta values for conditional reliability evaluation. Default is seq(-3, 3, 0.1).

...

Additional arguments passed to enhanced_mltm_reliability.

Value

A list combining enhanced reliability results and, optionally, conditional TIF-based reliability.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  result <- integrate_with_enhanced_reliability(fit,
               theta_range = seq(-2, 2, 0.5))
  result$conditional_tif

Marginal Proportions Predictive Checks

Description

Computes and visualizes marginal success proportions, including predicted values, confidence intervals, RMSR, SRMR, and bias estimation.

Usage

marginal_Pchecks(fit, interval = 0.95)

Arguments

fit

MLTM object containing model results.

interval

Probability associated with the credibility intervals (default = 0.95).

Details

marginal_Pchecks calculates marginal prediction intervals and observed success proportions. It prints a table with observed vs. predicted values, generates a forest plot for visualization, and computes key fit indices: RMSR, SRMR, and bias.

Value

A list containing:

items: A table of fitted values and prediction intervals for each item.
rmsr: The Root Mean Square Residual (RMSR).
srmr: The Standardized Root Mean Square Residual (SRMR).
bias: The difference between the total observed and predicted proportions.

The function also generates:

A forest plot visualizing prediction intervals and observed success probabilities.

References

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  fit <- LLTM(analogy, Q, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  marginal_Pchecks(fit)

Plot Method for conditional_reliability_tif Objects

Description

Plot Method for conditional_reliability_tif Objects

Usage

## S3 method for class 'conditional_reliability_tif'
plot(x, ...)

Arguments

x

An object of class conditional_reliability_tif.

...

Additional arguments passed to plot_conditional_reliability.

Value

Invisibly returns NULL. Called for its side effect of producing a reliability and/or information plot via plot_conditional_reliability().

Simple Plot Method (Minimal Dependencies)

Description

Creates basic plots using base R (no ggplot2 dependency).

Usage

## S3 method for class 'enhanced_mltm_reliability'
plot(x, type = "marginal", component = NULL, ...)

Arguments

x

Object of class enhanced_mltm_reliability.

type

Character. Type of plot: "marginal", "conditional", "comparison".

component

Integer or character. Specific component for conditional plots.

...

Additional plotting parameters.

Value

Invisibly returns NULL. Called for its side effects.

Grouped Item Characteristic Curves (ICC) Plot

Description

Generates Item Characteristic Curves (ICCs) for a group of items displayed in a grid layout with a shared legend.

Usage

plot_ICC_grouped(
  fit,
  Q,
  components,
  page = 1,
  n_items_per_page = 9,
  ncol = 3,
  nrow = 3
)

Arguments

fit

A fitted GMLTM model object containing EAP parameter estimates.

Q

The Q-matrix indicating the association between items and rules.

components

A list where each element is a vector of rule indices per component.

page

Integer specifying which page of items to display.

n_items_per_page

Number of items to include per page. Default is 9.

ncol

Number of columns in the layout grid. Default is 3.

nrow

Number of rows in the layout grid. Default is 3.

Details

Displays one legend shared across all plots and ensures consistency across theta and probability axes. Ideal for publications or appendices.

Value

A composed ICC grid with one shared legend, plotted to the active device.

References

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  plot_ICC_grouped(fit, Q, components, page = 1)

Individual Item Characteristic Curves (ICC)

Description

Returns a list of individual ICC ggplot2 plots (one per item).

Usage

plot_ICC_individual(fit, Q, components)

Arguments

fit

A fitted GMLTM model object.

Q

The Q-matrix for rule-item associations.

components

A list indicating rule groupings per component.

Value

A list of individual ggplot objects, one per item.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  plots <- plot_ICC_individual(fit, Q, components)
  print(plots[[1]])

Plot All Components Separately

Description

Creates individual plots for each component in the analysis.

Usage

plot_all_components(
  results,
  plot_type = "both",
  include_ci = TRUE,
  color_scheme = "blue",
  save_plots = FALSE,
  output_dir = NULL,
  ...
)

Arguments

results

An object of class conditional_reliability_tif.

plot_type

Character. Type of plot: "reliability", "information", or "both". Default is "both".

include_ci

Logical. Whether to include credible interval bands. Default TRUE.

color_scheme

Character. Color scheme for plots. Default "blue".

save_plots

Logical. Whether to save plots to disk. Default FALSE.

output_dir

Character. Directory for saved plots. Default NULL (no files written). Only creates directory and saves when both save_plots = TRUE and output_dir is non-NULL.

...

Additional arguments passed to plot_conditional_reliability.

Value

Invisibly returns NULL. Called for its side effect of producing reliability plots for all model components.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  cond_rel <- conditional_reliability_tif(fit, theta_range = seq(-2, 2, 0.5))
  plot_all_components(cond_rel, save_plots = TRUE, output_dir = tempdir())

Plot Comparisons (Base R)

Description

Plot Comparisons (Base R)

Usage

plot_comparison_base(comparison_data, ...)

Plot Comparison of Conditional Reliability Between Components

Description

Plot Comparison of Conditional Reliability Between Components

Usage

plot_comparison_conditional(x, include_ci = TRUE, color_scheme = "blue", ...)

Create Comparison Plot of All Components

Description

Creates a single plot comparing conditional reliability across all components.

Usage

plot_components_comparison(
  results,
  include_ci = TRUE,
  show_optimal_points = TRUE,
  add_reference_lines = TRUE,
  color_palette = "Set2",
  save_plot = FALSE,
  filename = NULL,
  ...
)

Arguments

results

An object of class conditional_reliability_tif.

include_ci

Logical. Whether to include credible interval bands. Default TRUE.

show_optimal_points

Logical. Whether to mark optimal theta points. Default TRUE.

add_reference_lines

Logical. Whether to add horizontal reference lines at 0.7, 0.8, 0.9. Default TRUE.

color_palette

Character. RColorBrewer palette name for component colors. Default "Set2".

save_plot

Logical. Whether to save the plot. Default FALSE.

filename

Character. Output filename if save_plot = TRUE. Default NULL (no file written). Only writes when both save_plot = TRUE and filename is non-NULL.

...

Additional graphical arguments.

Value

Invisibly returns NULL. Called for its side effect of producing a comparative reliability plot across model components.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  cond_rel <- conditional_reliability_tif(fit, theta_range = seq(-2, 2, 0.5))
  plot_components_comparison(cond_rel)

Plot Conditional Reliability (Base R)

Description

Plot Conditional Reliability (Base R)

Usage

plot_conditional_base(conditional_data, component = NULL, ...)

Plot Conditional Reliability Results

Description

Creates plots for conditional reliability analysis with multiple visualization options

Usage

plot_conditional_reliability(
  results,
  component = NULL,
  plot_type = "both",
  include_ci = TRUE,
  ci_level = 0.95,
  color_scheme = "blue",
  add_reference_lines = TRUE,
  save_plot = FALSE,
  filename = NULL,
  ...
)

Arguments

results

Object of class conditional_reliability_tif

component

Integer or character. Component to plot (NULL for first component)

plot_type

Character. Type of plot: "reliability", "information", "both", "comparison"

include_ci

Logical. Include confidence intervals

ci_level

Numeric. Confidence level (0.90 or 0.95)

color_scheme

Character. Color scheme: "blue", "viridis", "custom"

add_reference_lines

Logical. Add reference lines for reliability levels

save_plot

Logical. Save plot to file

filename

Character. Filename if saving plot. Default NULL (no file written). Only writes when both save_plot = TRUE and filename is non-NULL.

...

Additional plotting parameters

Value

Invisibly returns NULL. Called for its side effect of producing one or two plots (reliability curve, test information function, or both) depending on plot_type.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  cond_rel <- conditional_reliability_tif(fit, theta_range = seq(-2, 2, 0.5))
  plot_conditional_reliability(cond_rel, component = 1)

Plot Marginal Reliability (Base R)

Description

Plot Marginal Reliability (Base R)

Usage

plot_marginal_base(marginal_data, ...)

Plot a Prior Predictive Check for the GMLTM-D

Description

Draws a histogram of the simulated proportion-correct distribution from prior_predictive_check, with configurable reference lines marking a substantively plausible range, in the style of Fig. 3 of Gelman et al. (2020): a prior that concentrates simulated data outside that range is a sign the prior is too informative, too diffuse, or otherwise misspecified.

Usage

plot_prior_predictive_check(
  x,
  type = c("global", "by_item", "by_component"),
  plausible_range = c(0.2, 0.9),
  bins = 30
)

Arguments

x

An object of class "GMLTM_prior_predictive_check", as returned by prior_predictive_check.

type

Which simulated distribution to plot: "global" (default, one value per replicate), "by_item" (item-level values pooled across items and replicates), or "by_component" (component-level values pooled across components and replicates; requires x to have been computed with by_component = TRUE).

plausible_range

Numeric vector of length 2 giving the substantively plausible range for the proportion correct, drawn as dashed reference lines. Default is c(0.2, 0.9).

bins

Number of histogram bins. Default is 30.

Value

A ggplot2 object.

Examples

Q <- structure(
  c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
    1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
    1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
    0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
    1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
  dim = c(27L, 5L),
  dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                          "subt_seg","mov_point")))
components <- list(global = c(1, 2, 3), local = c(4, 5))
ppc <- prior_predictive_check(Q, components, N = 150, S = 200)
plot_prior_predictive_check(ppc)

Posterior Predictive Checks (PPC) for Model Fit Evaluation

Description

This function generates posterior predictive checks by plotting the observed and simulated total scores distribution, comparing empirical data against model predictions. It also computes fitted values and prediction intervals.

Usage

ppchecks(fit, nsim = 100, interval = 0.95, ...)

Arguments

fit

A fitted MLTM object containing model results.

nsim

Number of simulated posterior samples (default = 100).

interval

Probability associated with the credibility intervals (default = 0.95).

...

Additional graphical parameters to customize the plot.

Details

The function simulates multiple datasets from the posterior distribution and compares the empirical distribution of total scores with the predicted distribution. It overlays the observed and predicted distributions using a histogram with transparency.

The fitted values, along with their credibility intervals, are computed and returned.

Value

A list containing:

items: A table of fitted values and prediction intervals for each item.
subjects: Fitted and observed mean scores per subject.
ysim: Simulated response matrices.

The function also generates:

A histogram comparing observed vs. predicted total scores.

References

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  fit <- LLTM(analogy, Q, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  ppchecks(fit)

Print method for GMLTM objects

Description

Print method for GMLTM objects

Usage

## S3 method for class 'GMLTM'
print(x, digits = 3, ...)

Arguments

x

An object of class "GMLTM", as returned by GMLTM.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print Method for GMLTM_batch_report Objects

Description

Print Method for GMLTM_batch_report Objects

Usage

## S3 method for class 'GMLTM_batch_report'
print(x, digits = 3, ...)

Arguments

x

An object of class "GMLTM_batch_report", as returned by student_report_batch.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print method for GMLTM_corr objects

Description

Print method for GMLTM_corr objects

Usage

## S3 method for class 'GMLTM_corr'
print(x, digits = 3, ...)

Arguments

x

An object of class c("GMLTM_corr", "GMLTM"), as returned by GMLTM_corr.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print method for GMLTM_correlation objects

Description

Print method for GMLTM_correlation objects

Usage

## S3 method for class 'GMLTM_correlation'
print(x, digits = 3, ...)

Arguments

x

An object of class "GMLTM_correlation", as returned by extract_correlation.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print method for GMLTM_prior_predictive_check objects

Description

Print method for GMLTM_prior_predictive_check objects

Usage

## S3 method for class 'GMLTM_prior_predictive_check'
print(x, digits = 3, ...)

Arguments

x

An object of class "GMLTM_prior_predictive_check", as returned by prior_predictive_check.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print Method for GMLTM_student_report Objects

Description

Print Method for GMLTM_student_report Objects

Usage

## S3 method for class 'GMLTM_student_report'
print(x, digits = 3, ...)

Arguments

x

An object of class "GMLTM_student_report", as returned by student_report.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print method for LLTM objects

Description

Print method for LLTM objects

Usage

## S3 method for class 'LLTM'
print(x, digits = 3, ...)

Arguments

x

An object of class "LLTM", as returned by LLTM.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print method for MLTM objects

Description

Print method for MLTM objects

Usage

## S3 method for class 'MLTM'
print(x, digits = 3, ...)

Arguments

x

An object of class "MLTM", as returned by MLTM.

digits

Number of decimal digits to display. Default is 3.

...

Currently ignored.

Value

Invisibly returns x.

Print Method for Enhanced MLTM Reliability

Description

Print Method for Enhanced MLTM Reliability

Usage

## S3 method for class 'enhanced_mltm_reliability'
print(x, digits = 3, ...)

Arguments

x

An object of class "enhanced_mltm_reliability".

digits

Integer. Number of decimal places to display. Default is 3.

...

Currently unused.

Value

Invisibly returns x. Called for its side effect of printing the enhanced reliability analysis results to the console.

Print Method for Data Quality Diagnostics

Description

Print Method for Data Quality Diagnostics

Usage

## S3 method for class 'reliability_data_quality'
print(x, ...)

Arguments

x

An object of class "reliability_data_quality".

...

Currently unused.

Value

Invisibly returns x. Called for its side effect of printing the data quality diagnostics to the console.

Print Method for Reliability Profile

Description

Print Method for Reliability Profile

Usage

## S3 method for class 'reliability_profile'
print(x, ...)

Arguments

x

An object of class "reliability_profile".

...

Currently unused.

Value

Invisibly returns x. Called for its side effect of printing a formatted reliability profile summary to the console.

Prior Predictive Check for the GMLTM-D

Description

Simulates parameters directly from the priors declared for GMLTM/GMLTM_corr, simulates response data from those parameters under the GMLTM-D's conjunctive structure, and returns the resulting distribution of simulated proportion-correct – globally, per item, and (optionally) per component – across replicates. This lets the analyst check whether a candidate prior implies substantively plausible data before spending compute on fitting the model to real data (Gelman et al., 2020, Sect. 2.4).

Usage

prior_predictive_check(
  Q,
  components,
  N,
  S = 1000,
  priors = list(theta = list(mu = 0, sigma = 1), eta = list(mu = 0, sigma = 1), alpha =
    list(mu = 0, sigma = 1, family = "normal"), c = list(shape1 = 3, shape2 = 20)),
  by_component = TRUE
)

Arguments

Q

A p \times K Q-matrix specifying which cognitive rules each item requires, identical in role to the one passed to GMLTM.

components

A named list grouping rules into components, identical in role to the one passed to GMLTM.

N

Number of simulated examinees per replicate.

S

Number of replicates. Default is 1000, following the number of prior predictive draws used in Fig. 3 of Gelman et al. (2020); values from 500 to 1000 are typical.

priors

A named list of prior hyperparameters with the same structure accepted by GMLTM's priors argument (elements theta, eta, alpha, c), including priors$alpha$family ("normal" or "lognormal"). Parameters are simulated under exactly this prior, so the check reflects what will actually be fit – not an approximation of it.

by_component

Logical. If TRUE (default), also returns the simulated proportion-correct restricted to the items of each component.

Details

In each of the S replicates: rule difficulties \eta are drawn as \eta_k \sim \text{Normal}(\mu_\eta, \sigma_\eta) for each rule of each component, and item difficulty is \beta = Q \eta; abilities \theta are drawn i.i.d. \text{Normal}(\mu_\theta, \sigma_\theta) for N simulated examinees; guessing c is drawn \text{Beta}(\text{shape1}, \text{shape2}) per item; and discrimination \alpha is drawn from priors$alpha$family – half-Normal (\text{Normal}(\mu_\alpha, \sigma_\alpha) truncated to (0, \infty)) or Log-Normal(\mu_\alpha, \sigma_\alpha) – with one draw per unique combination of rules that a component's items require (the same grouping GMLTM itself uses, so alpha has as many free values here as it would have as a Stan parameter). Response probabilities follow the GMLTM-D's noncompensatory structure: for item i of examinee n, p_{ni} = c_i + (1 - c_i) \prod_{m} \text{plogis}(\alpha_{im}(\theta_{nm} - \beta_{im}))^{C_{im}}, and y_{ni} \sim \text{Bernoulli}(p_{ni}).

Use plot_prior_predictive_check to visualize the resulting distribution against a substantively plausible range.

Value

A list of class "GMLTM_prior_predictive_check" with elements:

global: Numeric vector of length S: the overall simulated proportion correct in each replicate.
by_item: An S x p matrix of simulated proportion correct per item.
by_component: An S x M matrix of simulated proportion correct restricted to each component's items, or NULL if by_component = FALSE.
N, S: The arguments used.
Q, components, priors: The arguments used (priors after applying defaults).

References

Gelman, A., Vehtari, A., Simpson, D., Margossian, C. C., Carpenter, B., Yao, Y., Kennedy, L., Gabry, J., Burkner, P.-C., & Modrak, M. (2020). Bayesian workflow. arXiv. doi:10.48550/arXiv.2011.01808

Examples

Q <- structure(
  c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
    1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
    1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
    0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
    1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
  dim = c(27L, 5L),
  dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                          "subt_seg","mov_point")))
components <- list(global = c(1, 2, 3), local = c(4, 5))

ppc_normal <- prior_predictive_check(Q, components, N = 150, S = 200)
summary(ppc_normal$global)

ppc_lognormal <- prior_predictive_check(
  Q, components, N = 150, S = 200,
  priors = list(alpha = list(mu = 0, sigma = 1, family = "lognormal"))
)
summary(ppc_lognormal$global)

Quick Reliability Check

Description

Ultra-fast reliability check for initial assessment.

Usage

quick_reliability_check(fit, n_samples_quick = 500)

Arguments

fit

Fitted model object.

n_samples_quick

Integer. Number of samples for quick analysis (default 500).

Value

Named vector of reliability estimates.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  fit <- LLTM(analogy, Q, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  quick_rel <- quick_reliability_check(fit)
  print(quick_rel)

Marginal reliability

Description

Estimate the the marginal reliability of the GMLTM.

Usage

reliability(fit)

Arguments

fit

A fitted model object of class "GMLTM", "MLTM", or "LLTM", as returned by GMLTM, MLTM, or LLTM.

Details

reliability computes the classical marginal reliability coefficient for each cognitive component's ability estimates, from the posterior mean and posterior variance of theta across subjects: r_{xx} = s^2 / (s^2 + e), where s^2 is the variance of the EAP theta estimates across subjects (true-score variance) and e is the average posterior variance of theta across subjects (error variance).

Value

A named numeric vector with one marginal reliability coefficient per cognitive component (length 1 for LLTM fits, which have a single ability dimension).

References

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  reliability(fit)

Fast Reliability Comparisons

Description

Optimized pairwise comparisons of reliability between components.

Usage

reliability_comparison_fast(marginal_results)

Arguments

marginal_results

List of marginal reliability results.

Value

List of comparison results.

Quick Reliability Profile Analysis

Description

Provides a quick overview of reliability characteristics for each component.

Usage

reliability_profile(cond_rel_obj)

Arguments

cond_rel_obj

An object of class conditional_reliability_tif returned by conditional_reliability_tif.

Value

A list of class "reliability_profile" with elements:

summary: Data frame with columns theta, reliability, lower, and upper.
component: Integer indicating which model component was analysed.

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components,
               iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  cond_rel <- conditional_reliability_tif(fit, theta_range = seq(-2, 2, 0.5))
  reliability_profile(cond_rel)

Usage Instructions

Description

Prints detailed usage instructions.

Usage

reliability_usage_instructions()

Value

Invisibly returns NULL. Called for its side effect of printing step-by-step usage instructions to the console.

Examples

reliability_usage_instructions()

Individual Student Diagnostic Report for GMLTM-D Models

Description

Builds a per-component and per-rule diagnostic report for a single examinee from a fitted GMLTM/GMLTM_corr model: posterior EAP and credible interval of \theta, a mastery cutline \gamma_m per component and \tau_{km} per rule, derived from a mastery probability y, binary mastery vectors, a decision-confidence index at both levels, and the difficulty of the rules/items involved in each component. Follows the mastery-diagnosis procedure for the MLTM-D described in Embretson (2019, Chap. 9, Sects. 9.2.2 and 9.4.2).

Usage

student_report(
  fit,
  Q,
  components,
  student_id,
  y = 0.5,
  theta_cut = NULL,
  credible = 0.95,
  digits = 3,
  include_plots = TRUE
)

Arguments

fit

An object of class "GMLTM" (as returned by GMLTM or GMLTM_corr).

Q

A p \times K Q-matrix, identical to the one used to fit fit, specifying which rules each item requires.

components

A named list grouping rules into components, identical to the one used to fit fit.

student_id

Identifier of the examinee: either a row name of fit$data (character/numeric matching rownames(fit$data)) or an integer row index in 1:nrow(fit$data).

y

Mastery probability used to derive the cutlines \gamma_m and \tau_{km} (Embretson, 2019, Sect. 9.2.2). Default is 0.50. Ignored for components overridden via theta_cut (\tau_{km} is always derived from y; see Details).

theta_cut

Optional fixed cutline(s) on the \theta scale, used instead of deriving \gamma_m from y. Either a single number (applied to every component) or a numeric vector of length length(components) (optionally named with the component names). Default is NULL (derive \gamma_m from y). Does not affect the rule-level cutlines \tau_{km}.

credible

Width of the credible interval for \theta, e.g. 0.95 for a 95% interval. Default is 0.95.

digits

Number of decimal digits used in the printed summary and in the plot subtitles. Default is 3.

include_plots

Logical. If TRUE (default), builds the ggplot2 objects described under plots below. If FALSE, plots is NULL and that (comparatively expensive) step is skipped, leaving only the numeric summary; this is used by student_report_batch to compute a lightweight summary for every requested student while only rendering plots for the current page.

Details

For each component m, the mastery cutline \gamma_m is the value of \theta_m at which the mean predicted probability of solving the items that load on component m, \bar{P}_m(\theta_m) = \text{mean}_{k}\left[c_k + (1-c_k)\, \text{plogis}(\alpha_{km}(\theta_m - \beta_{km}))\right], equals y (Embretson, 2019, Eq. in Sect. 9.2.2). It is found numerically with uniroot using the item EAP estimates fit$EAP$alpha, fit$EAP$beta, and fit$EAP$guessing. If theta_cut is supplied, it is used directly as \gamma_m and y is not used to derive it.

In addition, a rule-level cutline \tau_{km} is computed for each rule k belonging to component m, using the same root-finding procedure as \gamma_m but restricted to the items that both load on component m and require rule k specifically (Embretson, 2019, Sect. 9.2.2). If no such item exists – the rule never appears on its own among the component's items – \tau_{km} is undefined: a warning is issued and that rule is omitted from domain_rules and confidence_rules for that component. Rule-level cutlines are always derived from y and are not affected by theta_cut.

The student is classified as mastering component m (domain = 1) if \theta_{m}^{EAP} \geq \gamma_m, and as non-mastering (domain = 0) otherwise; the same comparison of \theta_m^{EAP} against \tau_{km} classifies mastery of rule k (domain_rules). Decision confidence is the proportion of posterior draws of \theta_m for that student that agree with the classification: the proportion \geq \gamma_m (or \tau_{km}) when classified as mastery, or the proportion < \gamma_m (or \tau_{km}) when classified as non-mastery (Embretson, 2019, Sect. 9.4.2).

The difficulty of the rules assigned to each component is read from fit$EAP$eta (rule-level difficulty, on the \theta scale), and the difficulty of the items that load on each component is read from fit$EAP$beta (item-level difficulty). Q and components are required because fit does not store the item/component structure matrix C; it is recomputed internally the same way as in plot_ICC_individual.

Value

A list of class "GMLTM_student_report" with elements:

student: The resolved identifier of the examinee (as it appears in rownames(fit$data)).
theta: A list with EAP, lower, upper (named by component) and credible.
cutline: A list with y, gamma (named by component), tau (named list by component of named numeric vectors of \tau_{km} by rule, excluding any rule omitted for lack of items) and fixed (logical, TRUE if theta_cut was supplied).
domain: Named binary vector (1 = mastery, 0 = non-mastery) per component.
domain_rules: Named list by component of named binary vectors (1 = mastery, 0 = non-mastery) per rule, excluding any rule omitted for lack of items.
confidence: Named decision-confidence index per component.
confidence_rules: Named list by component of named decision-confidence indices per rule, same exclusions as domain_rules.
difficulty: A list with rules (named list of rule-level \eta difficulties per component) and items (named list of item-level \beta difficulties per component).
plots: Named list of ggplot2 objects: one per component, showing the rule difficulty continuum, the mastery cutline, and the student's \theta estimate with its credible interval (colored green for mastery, red for non-mastery), inspired by Figs. 4 and 5 of Embretson & Yang (2013); plus badges, a single scannable summary of every component and rule with a green/red mastery badge, complementing the continuum plots. NULL if include_plots = FALSE.

References

Embretson, S. E., & Yang, X. (2013). A multicomponent latent trait model for diagnosis. Psychometrika, 78, 14–36. doi:10.1007/s11336-012-9296-y

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)
  report <- student_report(fit, Q, components, student_id = 11)
  report$domain
  report$domain_rules
  report$plots$global
  report$plots$badges

Paginated Batch Diagnostic Reports for GMLTM-D Models

Description

Runs student_report over many examinees at once. A lightweight numeric summary (posterior EAP of \theta, mastery and decision confidence per component and per rule) is always computed for every requested student, since it only reads quantities already stored in fit$EAP/fit$posterior$theta. The comparatively expensive step – building the ggplot2 objects returned by student_report – is only performed for the students in the current page, so large batches stay cheap to page through.

Usage

student_report_batch(
  fit,
  Q,
  components,
  student_ids = NULL,
  offset = 0,
  limit = 10,
  y = 0.5,
  theta_cut = NULL,
  credible = 0.95,
  digits = 3
)

Arguments

fit

An object of class "GMLTM" (as returned by GMLTM or GMLTM_corr).

Q

A p \times K Q-matrix, identical to the one used to fit fit, specifying which rules each item requires.

components

A named list grouping rules into components, identical to the one used to fit fit.

student_ids

Vector of examinee identifiers to include in the batch, in either form accepted by student_report's student_id (row names of fit$data, or integer row indices). Default is NULL, meaning every student in fit$data (seq_len(nrow(fit$data))).

offset

Integer. Number of students in student_ids to skip before the current page. Default is 0.

limit

Integer. Maximum number of students in the current page (i.e. with full reports/plots generated). Default is 10.

y, theta_cut, credible, digits

Passed through to student_report; see its documentation.

Details

The current page covers student_ids[(offset + 1):(offset + limit)]. If offset is at or beyond length(student_ids), the page is empty (reports has length 0) rather than an error; summary is unaffected, since it always covers every requested student regardless of paging.

Value

A list of class "GMLTM_batch_report" with elements:

summary: A data.frame with one row per requested student (student, theta_<component>, domain_<component>, confidence_<component> for each component, and domain_rule_<rule>, confidence_rule_<rule> for each rule), built without generating any plots.
reports: Named list (by resolved student label) of full student_report objects, including their plots, for the students in the current page only.
page: A list with offset, limit, and total (the number of requested students in student_ids).

Examples


  if (!requireNamespace("rstan", quietly = TRUE)) return()
  data(analogy)
  Q <- structure(
    c(0,0,1,0,1,0,1,0,1,1,0,1,1,1,0,1,1,1,0,1,0,1,0,0,1,0,1,
      1,0,0,0,0,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1,1,1,1,0,1,1,1,0,
      1,0,0,0,0,0,1,0,0,1,0,0,0,0,0,0,0,1,0,1,0,1,1,0,1,0,0,0,
      0,0,0,0,0,0,1,0,1,0,0,1,0,0,1,0,0,1,0,0,1,0,0,0,0,0,0,0,
      1,0,0,0,1,1,0,1,1,1,1,1,1,1,0,1,1,0,1,1,1,1,0,1),
    dim = c(27L, 5L),
    dimnames = list(NULL, c("rot_fig","rot_trap","reflection",
                            "subt_seg","mov_point")))
  components <- list(global = c(1, 2, 3), local = c(4, 5))
  fit <- GMLTM(analogy, Q, components, iters = 200, iter_warmup = 100, chains = 1, cores = 1)

  # First page: students 1-10 (summary covers all 149 examinees either way)
  page1 <- student_report_batch(fit, Q, components, limit = 10)
  page1
  page1$reports[["1"]]$plots$badges

  # Next page: students 11-20, individual reports for students 1-10 are
  # still fully available in page1$reports
  page2 <- student_report_batch(fit, Q, components, offset = 10, limit = 10)
  page2$reports[["11"]]

Summary Method for Conditional Reliability Analysis

Description

Summary Method for Conditional Reliability Analysis

Usage

## S3 method for class 'conditional_reliability_tif'
summary(object, digits = 3, ...)

Arguments

object

An object of class conditional_reliability_tif.

digits

Integer. Number of decimal places to display. Default is 3.

...

Currently unused.

Value

Invisibly returns object. Called for its side effect of printing a formatted summary of conditional reliability statistics to the console.

Package {GMLTM}

GMLTM: Generalized Multicomponent Latent Trait Model for Diagnosis

Description

Model fitting

Reliability

Model checking and diagnostics

Plotting and item structure

Author(s)

References

See Also

The Generalized Multicomponent Latent Trait Model for Diagnosis

Description

Usage

Arguments

Details

Value

References

See Also

Examples

GMLTM-D with correlated latent components (general Sigma)

Description

Usage

Arguments

Details

Value

References

See Also

Examples

The Linear Logistic Test Model

Description

Usage

Arguments

Details

Value

References

See Also

Examples

The Multicomponent Latent Trait Model for Diagnosis

Description

Usage

Arguments

Details

Value

References

See Also

Examples

Analogy items dataset

Description

Usage

Format

Source

References

Fast Bayesian Marginal Reliability

Description

Usage

Arguments

Value

Check Data Quality for Reliability Analysis

Description

Usage

Arguments

Value

See Also

Examples

Compare Conditional Reliability Between Components at Specific Theta Values

Description

Usage

Arguments

Value

See Also

Examples

Basic Diagnostics (No External Dependencies)

Description

Usage

Arguments

Value

Compute LOO and WAIC for GMLTM models

Description

Usage

Arguments

Deprecated Alias for `student_report`