ERIC Number: ED395995
Record Type: RIE
Publication Date: 1996-Mar-8
Reference Count: N/A
An Estimation Procedure for the Structural Parameters of the Unified Cognitive/IRT Model.
Jiang, Hai; And Others
L. V. DiBello, W. F. Stout, and L. A. Roussos (1993) have developed a new item response model, the Unified Model, which brings together the discrete, deterministic aspects of cognition favored by cognitive scientists, and the continuous, stochastic aspects of test response behavior that underlie item response theory (IRT). The Unified Model blends psychometric and cognitive science viewpoints and promises to allow the practitioner to recover cognitive information from simple, well-designed tests. This paper proposes an estimation procedure for the structural model parameters of the Unified Model that uses the marginal maximum likelihood estimation approach of Bock and Aitkin (1981) and the EM algorithm of A. P. Dempster, N. M. Laird, and D. B. Rubin (1977). In the maximization (M) step of the EM algorithm, because of the difficulties in computing the second derivative (Hessian) matrix and the possibility of multiple local maxima, using an alternative maximization procedure is proposed. This procedure, called Evolution Programming (Z. Michalewicz, 1994), has good properties in finding a global extremum. A simulation study is then given to show the effectiveness of the estimation procedure. (Contains 2 figures, 7 tables, and 16 references.) (Author/SLD)
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: N/A
Authoring Institution: N/A
Identifiers: EM Algorithm
Note: Paper presented at the Annual Meeting of the National Council on Measurement in Education (New York, NY, April 9-11, 1996).