easyRaschBayes 0.4.0

Breaking changes

• item_restscore_statistic(), infit_statistic() and q3_statistic() now condition on each person’s total score, both for the expected values and for the replicated data. They accept only Rasch-family models (bernoulli or acat, logit link, no disc). method = "theta" in infit_statistic() and q3_statistic() gives the 0.3.0 results. Results change.
  • item_restscore_statistic() adds Kreiner’s expected gamma (gamma_exp), a mid-p ppp and a max-T adjusted p_adj. gamma_diff is now observed minus expected, and gamma_obs_q* is replaced by gamma_rep_q*.
  • Conditional infit matches iarm::out_infit() for the same item parameters.
  • q3_post() adds q3_p_adj, adjusted over item pairs, and its reference line is the mean replicated Q3.
• infit_post() infit_ppp and q3_post() q3_ppp compared each draw’s observed value with the mean of the replicated values, not with the same draw’s replicated value. infit_ppp is now also P(observed > replicated) with mid-p ties, so values near 1 indicate underfit, as for q3_ppp. Results change.
• plot_targeting() and plot_targeting_hpcm() follow easyRasch2::RMtargeting(). The bottom panel shows response-category bands, with categories that are never most likely and disordered thresholds marked in red and labelled with their posterior probabilities. panel = "thresholds" gives a dot-and-whisker plot. New arguments panel, xlim, category_labels, viridis_*, row_gap and output = "list", plus person_estimate in plot_targeting(). The defaults for bins and the x-axis change, and sort_items = "location" puts the easiest item at the top. ‘scales’ is added to Imports.

New features

• New datasets phq9 (600 NHANES respondents, nine items with four categories, as in easyRasch2) and phq9_nhanes (1000 NHANES respondents, including zero responses, for the hurdle partial credit model).
• plot_icc() gains method and score_breaks for forming class intervals, as in easyRasch2::RMitemICCPlot(), and a caption stating the grouping. n_intervals is renamed class_intervals in plot_icc() and plot_icc_hpcm(), and the old name is deprecated.
• plot_icc() with dif_var separates groups by shape and line type as well as colour. plot_ipf() adds line types, and plot_tile() and plot_stackedbars() use theme_bw().
• plot_icc(), plot_icc_hpcm() and plot_residual_pca() report how many persons with missing responses they leave out.
• item_restscore_post() and infit_post() accept any item column name.
• item_parameters() n_eff is now posterior::ess_bulk(). ‘posterior’ is added to Imports.

Bug fixes

• dif_statistic() for polytomous models loaded differences in group means onto the items as DIF. Impact is now estimated separately (new impact element), and all DIF types share one sign: positive means easier for the focal group. Results change for polytomous models.
• Rating scale models gave every item the same thresholds. Results change for rating scale models. 0 + item models no longer fail in plot_residual_pca() and plot_targeting().
• person_parameters() and person_parameters_hpcm() now compute the WLE of persons with missing responses from the items they answered, with a new n_items column.
• person_parameters() and plot_icc() counted sum scores from the lowest observed category instead of category 1.
• q3_statistic(), q3_statistic_hpcm(), plot_residual_pca() and item_restscore_statistic() stop if a person and item combination occurs more than once. Previously only the last response was kept.
• plot_icc_hpcm() placed class intervals in the severity panel at their mean EAP, which made observed points look misfitting, and shifted responses down when there were no zeros. Plots change.
• plot_ipf() and plot_icc() use the model’s link for cumulative, sratio and cratio models, and stop for non-logit acat models and models that estimate disc.
• plot_residual_pca() aligns eigenvector signs across draws by the dot product with the mean direction.
• plot_icc() partial gamma used the wrong column for numeric item labels.
• posterior_epred_hurdle_acat() failed when items had different numbers of categories.
• infit_statistic_hpcm() and q3_statistic_hpcm() failed with ordered-factor responses or a single draw, and with item labels that brms renames.

Documentation

• infit_statistic() and infit_post() state that the infit ppp is conservative, also with the conditional method, and suggest item_restscore_statistic() as well.
• Corrected RMUreliability() return value, plot_residual_pca() style default and centring, and plot_icc() partial gamma sign. Added caveats to posterior_to_prior() and plot_icc_hpcm(). infit_statistic_hpcm() is no longer described as conditional infit.

easyRaschBayes 0.3.0

• New custom brms family hurdle_acat for hurdle partial credit models (H-PCM). This is written by Kristoffer Magnusson
• Several adapted (experimental) functions for H-PCM, all using the *_hpcm suffix.
  • infit_statistic_hpcm()
  • q3_statistic_hpcm()
  • item_parameters_hpcm()
  • person_parameters_hpcm()
  • RMUreliability_hpcm()
  • plot_icc_hpcm()
  • plot_targeting_hpcm()
• Bug fix - changed post processing functions to use median_hdci() instead of median_hdi() to avoid issues with multimodal posteriors.
• Bug fix - corrected the WLE person estimates in person_parameters() and person_parameters_hpcm(). The polytomous Warm bias correction had the wrong sign, biasing estimates outward; WLE now matches catR/TAM. Extreme (minimum and maximum) scores now receive proper finite WLE estimates instead of being clamped to the theta_range bounds.

easyRaschBayes 0.2.0.1

• Fix for plot_icc() theta values.

easyRaschBayes 0.2.0

• Updates:
  • infit_statistic() is now faster and defaults to not output outfit statistics, since these are of dubious value (see Müller 2020 and Johansson, 2025).
  • item_restscore_statistic() is now faster and has output similar to infit_post().
  • q3_statistic() is now faster and has output similar to infit_post().
  • plot_residual_pca() is now faster.
• Three post-processing helper functions that output a list with result table(s) and a plot:
  • infit_post() for the output of infit_statistic()
  • item_restscore_post() for the output of item_restscore_statistic()
  • q3_post() for the output of q3_statistic()
• Other new functions:
  • plot_icc() can both produce a basic conditional Item Characteristic Curves plot and optionally a DIF ICC plot, the latter also reporting a partial gamma DIF magnitude coefficient. This is inspired by iarm::ICCplot()
  • item_parameters() to retrieve item threshold locations
  • person_parameters() to retrieve person locations (latent scores) using EAP and WLE, and also a transformation table from ordinal sum score to EAP/WLE.
  • posterior_to_prior() that extracts priors from a brmsfit model that may be useful in model fit assessment. An article will be added later, exploring this further.
  • plot_tile(), plot_bars(), and plot_stackedbars() uses wide format item data to create plots to review response patterns.

easyRaschBayes 0.1.0

• Initial release.