PsyMetricTools 1.2.4

Examples of the factor-analysis functions (EFA_modern(), EFA_plot(), Standardized_solutions(), Standardized_solutions_cfa(), multi_cfa(), fit_index_Table(), plot_multi_sem(), combine_likert_sem(), safe_cfa(), cor_afe(), extract_fit(), factor_summary() and the boot_cfa*() / boot_efa*() families) now simulate Likert items from correlated latent factors. They used independent sample(1:5) draws, so the models had no common factor and could fail to converge on some platforms: the EFA_plot() example stopped with “fit measures not available if model did not converge” in the CRAN donttest check.
Version 1.2.3 was submitted to CRAN but withdrawn before review; its changes are included here.

PsyMetricTools 1.2.3

New run_sempowerlab(): launches the SemPowerLab Shiny application, which plans the sample size of a predictive design with latent variables (two correlated latent predictors and a latent criterion) by wrapping semPower::semPower.powerRegression(). It draws the model, plots the power curve, crosses the effect with the average factor loading, and writes the reproducible script and a draft of the Participants section. The application moves here from the InterconectaR package. shiny, bslib and semPower added to Suggests.
The application passes the criterion as the first factor, as semPower expects, and every output (cards, curve, script and paragraph) follows the specification that produced the result rather than inputs edited afterwards.

boot_cfa() now returns the scaled (robust) fit indices when the model is fitted with a robust test statistic. Since lavaan 0.6-14 the test option is a vector (c("standard", "scaled.shifted") for WLSMV), and the internal helpers is_robust_estimator_lavaan() and lavaan_estimator() inspected only its first element. That element is always "standard", so every robust fit was classified as non-robust and fit_measures1 carried cfi, tli and rmsea instead of cfi.scaled, tli.scaled and rmsea.scaled. With ordinal data the gap is not cosmetic: in a 23-item five-factor model the reported CFI was .988 against a true robust value of .947, which made the bootstrap figure contradict the fit table. Both helpers now test every declared statistic with any(test_types %in% ...), the same way boot_efa() already did. As a side effect the estimator column now reports WLSMV instead of DWLS_variant.
The column names keep their current form (CFI, TLI, RMSEA, …) so the plotting functions boot_cfa_plot(), boot_cfa_plot_enhanced(), boot_cfa_density(), boot_cfa_raincloud() and calculate_per_fit() are unaffected; only the values change.

The PDF manual now builds without LaTeX errors: the Unicode symbols >=, <=, -> and R^2 are written in portable form in the documentation of boot_cfa_plot_enhanced(), plot_mediation_chord(), plot_mediation_donuts() and plot_path_mediation(). This was the error that caused the 1.2.0 submission to be archived by CRAN.
DESCRIPTION now declares R (>= 4.1.0), required by the use of the native pipe |> / lambda \(...) syntax in invertir_items() and split_data_stratified_clustered().

plot_path_cfa() rewritten: latent variables as circles, observed variables as rectangles, fan-out arrows, black-and-white default.

plot_multi_sem() now supports multi-row panel layouts via the new nrow and ncol arguments. By default it keeps the previous behaviour (a single row with one column per model); when a grid is requested, leftover cells are left blank. Saved-figure dimensions scale with the grid (width_per per column, height per row).

filtrar_aberrantes() - Filter aberrant response patterns using Mahalanobis distance
invertir_items() - Reverse score items
split_data_two() / split_data_three() - Split data for cross-validation
smote_multiclass() - SMOTE oversampling for imbalanced multiclass data

easy_invariance() - Measurement invariance analysis with ordinal data following Wu and Estabrook (2016)