NEWS

---

irtlink 0.1.0 (2026-10-16)

Initial release

ADDED * Internal marginal maximum likelihood EM calibration

• (1PL, 2PL, 3PL): separate, concurrent (full or partial invariance, with chain, joint, and restricted joint compositions), fixed item parameter calibration (FIPC), and regularized calibration with a smooth Bayesian information criterion (SBIC).

ADDED * Polytomous items via the generalized partial credit

• model (GPCM) and the partial credit model (PCM) with separate calibration, chain and pairwise joint linking by every method, joint and pairwise Haberman linking, jackknife and approximate jackknife linking errors, EAP person scores, and a polytomous response simulator.

ADDED * Linking via link_groups() (chain, joint, restricted

• joint): mean-mean, mean-geometric mean, Haberman with robust Lq loss, Haebara, and Stocking-Lord, each joint method with a simultaneous and a pairwise variant, and a selectable reference group. The Haebara and Stocking-Lord criteria compare the full 3PL response functions when guessing parameters are present.

ADDED * Linking errors: (bias-corrected) delete-one item

• jackknife, approximate one-step jackknife, and sandwich estimators. A cluster argument deletes whole testlets instead of single items in the jackknife and the approximate jackknife.

ADDED * Item parameter covariances, including dependent_vcov

• (stacked sandwich or person bootstrap) for dependent person samples.

ADDED * DIF diagnostics: root mean square deviation (RMSD

• statistics and per-item likelihood-ratio tests, with iterative purification.

ADDED * Person scores (expected a posteriori (EAP) and weighted

• likelihood (WLE) estimates), a response data simulator (uniform and non-uniform DIF, dependent person samples), and a linking-design inspector.

ADDED * Person sampling weights (pweights, e.g. senate weights

• in every calibration strategy.

NOTE * No runtime dependencies beyond base R and Rcpp