NotesFAQContact Us
Collection
Advanced
Search Tips
ERIC Number: ED517917
Record Type: Non-Journal
Publication Date: 2010
Pages: 214
Abstractor: As Provided
Reference Count: N/A
ISBN: ISBN-978-1-1241-4315-6
ISSN: N/A
A Semi-Parametric Bayesian Mixture Modeling Approach for the Analysis of Judge Mediated Data
Muckle, Timothy Joseph
ProQuest LLC, Ph.D. Dissertation, University of Illinois at Chicago
Existing methods for the analysis of ordinal-level data arising from judge ratings, such as the Multi-Facet Rasch model (MFRM, or the so-called Facets model) have been widely used in assessment in order to render fair examinee ability estimates in situations where the judges vary in their behavior or severity. However, this model makes certain assumptions about the distribution of examinee abilities, namely, that this distribution follows an asymptotic normal model. In cases where the examinee distribution does not follow a normal distribution, the resulting examinee ability estimates may be substantially inaccurate. This dissertation introduced a generalized semiparametric version of the MFRM, which relaxes the assumption of normality by specifying a nonparametric prior (Dirichlet Process) on the distribution of examinee abilities. The goal of such a generalization was to specify a better-fitting model with improved predictive accuracy. The semiparametric model was illustrated on three data sets consisting of judge-rendered ordinal ratings. Posterior distributions for all model parameters, including examinees, items, and judges, were summarized and presented for all three data sets. Also, the model-data fit of the semiparametric model was compared to that of the traditional Facets model, as well as that of various random effect-extensions of the semiparametric model. Generally speaking, when evaluating the Deviance Information Criterion (DIC) among all models, there was accumulated evidence in favor of the semiparametric model with random vectors, incorporating random effects, either items or judges, in addition to examinees. A final comparative analysis evaluated model-data fit using two different formulations of Gamma priors for the [alpha] parameter in the Dirichlet Process, one concentrating prior support on small values of [alpha], and one weakly informative prior, which spread prior support over a wider range of [alpha] values. The DIC values resulting from fitting these models to the three datasets suggested that the relaxed a prior provided improved fit in a majority of the cases. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://www.proquest.com/en-US/products/dissertations/individuals.shtml.]
ProQuest LLC. 789 East Eisenhower Parkway, P.O. Box 1346, Ann Arbor, MI 48106. Tel: 800-521-0600; Web site: http://www.proquest.com/en-US/products/dissertations/individuals.shtml
Publication Type: Dissertations/Theses - Doctoral Dissertations
Education Level: N/A
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: N/A