lz and lz* Person-Fit Statistics to Detect Aberrant Response Pattern

• In Event: Novel Approaches for Model Fit

Abstract

This paper investigates two nonparametric person-fit statistics, lz and lz*. Even though Tendeiro, Meijer, & Niessen (2016) developed Perfit R package by which these two statistics can be calculated, research comparing the performance of these two statistics are sparse in literature. In this study, we compared lz and lz* statistics calculated in PerFit as well as in SAS program developed for our operational analysis. The performance of lz and lz* is then compared by specifying different estimation methods and different item response theory models. The results show that depending on different estimation methods and/or different item response theory models, the flagged aberrant item response patterns differ a lot between lz and lz*.

Authors

• Yi Lu, The Federation of State Boards of Physical Therapy

• Yu Zhang, The Federation of State Boards of Physical Therapy

• Lorin Mueller, Federation of State Boards of Physical Therapy