ForceChoice 1.0.0 (2026-07-12)

New Features

Eight Model Families

• fit.MIRT() - Multidimensional Item Response Theory (1PL–4PL) for binary dominance data, with optional Q-matrix structure.
• fit.MGPCM() - Multidimensional Generalized Partial Credit Model for polytomous ordered-category responses.
• fit.MGGUM() - Multidimensional Generalized Graded Unfolding Model for ideal-point polytomous responses with signed Q-matrix.
• fit.FCMIRT() - Forced-Choice MIRT with item-level dominance endorsement and Luce–Plackett block-level ranking (RANK, MOLE, PICK).
• fit.FCGGUM() - Forced-Choice GGUM with item-level ideal-point endorsement and block-level ranking.
• fit.TIRT() - Thurstonian IRT for forced-choice with pairwise probit comparisons and latent utility differences.
• fit.FCDCM() - Forced-Choice Diagnostic Classification Model with higher-order latent trait, DINA/DINO condensation rules, and exact attribute-profile marginalization.
• fit.FCGDINA() - Forced-Choice GDINA model with DINA, DINO, ACDM, and GDINA item-level structures for ranking, most-least, and pick responses.

Estimation Backends

• Stan (method = "stan"): Full Bayesian inference via Hamiltonian Monte Carlo (NUTS/HMC) with rstan.
• iStEM (method = "iStEM"): Fast iterative Stochastic EM with Metropolis-within-Gibbs person sampling and L-BFGS-B item optimization.
• EM (method = "EM"): Deterministic posterior-weight EM for FCGDINA.

Comprehensive Goodness-of-Fit

• Limited-information M2 statistic (Maydeu-Olivares & Joe, 2005, 2006)
• RMSEA with confidence intervals, SRMSR, RMSR
• Comparative fit: CFI, TLI, IFI
• Information criteria: AIC, AICc, BIC, CAIC, SABIC, HQIC
• Pseudo-R2 measures: McFadden, Cox–Snell, Nagelkerke, and more
• Local dependence: Yen’s Q3 (raw and adjusted)
• Posterior diagnostics: entropy, classification accuracy

Data Simulation

• sim.data.MIRT(), sim.data.MGPCM(), sim.data.MGGUM() for traditional item response data.
• sim.data.FCMIRT(), sim.data.FCGGUM(), sim.data.TIRT(), sim.data.FCDCM(), sim.data.FCGDINA() for forced-choice data.
• All simulators support controlled parameter generation, factor correlation, and rotation.

Solution Rotation

• rotate.MIRT() and rotate.matrix() for post-hoc rotation of MIRT solutions using promax or any GPArotation method.

S3 Methods

• Consistent generics across all model classes: coef(), confint(), deviance(), fitted(), logLik(), nobs(), plot(), predict(), print(), residuals(), summary(), update(), vcov().

Bug Fixes

• This is the initial CRAN release.

Notes

• The package was previously developed under the name fcirt and has been substantially rewritten and expanded to support eight model families and the iStEM estimation backend.