irtbem2pl 1.0.1

• Performance Optimization: Drastically improved the speed of the EM algorithm by implementing high-performance matrix algebra (%*%), reducing calibration times from minutes to seconds.
• Missing Data Support: Added native support for structurally missing data (unpresented items) by representing them as NA and mathematically adjusting the expected item frequencies (f and r) to prevent parameter estimation bias [2, 3].
• Gauss-Hermite Quadrature: Integrated Gauss-Hermite quadrature nodes and weights using the statmod package, matching ‘BILOG-MG’ standards and yielding superior precision with fewer integration points.
• Convergence Stability: Implemented parameter-change-based convergence checks to align calibration stopping rules with industrial standards [4].
• Bug Fixes: Resolved a numerical bug in the RMSEA goodness-of-fit calculation where a negative chi-square difference would produce NaN values [5]. 💡