An investigation of a discrete state Markov model for predicting college freshman grades /

Bibliographic Details
Main Author: Boudreaux, Donald Lee
Other Authors: Goetz, Ernest (degree committee member.), Matis, James H. (degree committee member.), Matthews, Tom (degree committee member.)
Format: Thesis Book
Language:English
Published: 1992.
Subjects:
Online Access:Link to OAKTrust copy
Description
Abstract:This study examined the use of a discrete state Markov model for predicting first year college grades for new freshman students entering Texas A&M University. In order to accomplish this three main tasks were performed. First, a discrete state Markov model was defined, shown to be situationally applicable, and estimated. Second, given the predominance of the regression methodology for this application, two linear regression models were developed to compare against the Markov model. The first regression involved a traditional model form based upon: high school rank SAT verbal, and SAT mathematics scores. The second regression model utilized the same basic predictor variables, but involved a more complex form--higher order powers and interaction terms. Third, a comparison was made between the Markov and regression models for an initial subsample and then cross-validated against three subsequent subsamples. The results showed that the Markov model produced higher R^2 values than the regression models. This study also discussed the use of the Markov transition probabilities for admissions purposes and documents the SAS procedure and interactive matrix language (SAS/IML) code used.
Item Description:Typescript (photocopy).
Vita.
"Major subject: Educational Psychology."
Physical Description:viii, 91 leaves : illustrations ; 29 cm
Bibliography:Includes bibliographical references.