Comparison of Four Maximum-Likelihood Methods in Estimating the Rasch Model

• In Event: Roundtable Session 117
  In Roundtable Session: Discussions in Item Response Theory

Abstract

This study compared the joint, conditional, marginal and penalized maximum likelihood estimations (JML, CML, MML, PML) and their computer programs in estimating the Rasch model, i.e., WINSTEPS, R-packages eRm and ltm, BILOG-MG, and SAS code using PML. A simulation study showed that PML generated less biased estimates than JML but did not perform as well as CML and MML; those ML methods and computer programs are acceptable practically; R-package ltm performed best.

Author

• Tianshu Pan, Pearson