Student Seminar Series - May 3, 2005
University of Minnesota
School of Statistics
College of Liberal Arts
A
Comparison of the Unconditional and Conditional Maximum Likelihood
Estimation Procedures for the Rasch Model as Related to Educational
Assessment
Andrew Swanlund
Tuesday, May 3, 2005
3:30 PM, 115
Ford Hall
Minneapolis, East Bank Campus
Refreshments at 3:00 PM
300 Ford Hall
Abstract
With the passing of the federal No Child Left Behind act of 2001,
there is an increased focus on accountability and standardized testing
in schools. The Rasch model of measurement is one model that is
frequently used by testing companies to calibrate and analyze the tests
which are part of many state assessment programs. This model allows for
the parameterization of student ability in conjunction with
simultaneously estimating item (or test question) difficulty. This talk
will provide a brief introduction to the field of educational
assessment, followed by a derivation of the Rasch model, its uses and
extensions, and several methods of estimating the difficulty of the
items. Specifically, the talk will compare the estimates obtained from
the traditional Unconditional Maximum Likelihood estimation procedure,
with that of the Conditional MLE approach (by fitting a Generalized
Linear Mixed Effects model).