New td_sensitivity(): how comparability conclusions depend on the linking assumption. Reports the ability difference, number of flagged items and pass-rate impact under mode, purified and all-item linking; a parametric bootstrap of the mode’s stability; the items whose DIF verdict depends on the linking; and an overall robust/sensitive verdict. Motivated by the First International Mathematics Study illustration, where pervasive DIF on a short test left the linking unidentified.
td_report() gains a sensitivity argument that adds this analysis to the comparability report.
td_dif() gains c_fixed and c_se_fixed to fit the DIF model under an analyst-chosen linking.
transDIF 0.1.0
Initial release.
Per-group Rasch calibration with relative standard errors (td_calibrate()).
Robust linking by the precision-weighted mode of item differences, with a sandwich SE and a warning when two item clusters are equally plausible, and small-sample DIF detection with an empirical-Bayes spike-and-slab model and local false discovery rates (td_dif()), plus a Mantel-Haenszel baseline (td_mh()).
td_features() drops item features with no variation (with a warning) instead of failing.
Aggregate pass-rate impact with linking and DIF uncertainty (td_impact()).