584 – Innovative Statistical Methods Applied to the Social Sciences
Newton-Type Algorithms for the Estimation of Item Response Theory Model
Xinming An
SAS Institute
Yiu-Fai Yung
SAS Institute
In the �eld of item response theory, G-H quadrature based EM algorithm proposed by Bock and Aitkin (1981) has been widely recognized as the gold standard for model estimation because of its several appealing properties. However, these advantages are overshadowed by a number of important issues that has not been resolved successfully. Furthermore, recent developments in item factor analysis, for example, con�rmatory analysis, also impair EM’s advantages. On the other hand, Newton algorithms do not suffer these problems, but are computationally more expensive than EM. During the last twenty years, statistical researches have been impacted dramatically by the advances in computational sciences. Thus it is worthwhile to apply these computational advances to Newton and EM type algorithms and re-evaluate their relative advantages. To this end, the focus of this research is to (1) introduce some Newton type algorithms for item factor analysis; and (2)investigate the computational properties of these Newton type algorithms as compared with the EM algorithm.
