Operationalizing predictive factors of success for entry level students of computer science (South Carolina)

This study undertakes to implement a predictive model of student success in the introductory course of computer science (CPSC 101) at a major southern university in the United States. The central issue is moving a predictive model from an "explanatory" state to an "operational" state within a student advising framework. The study's premise is that what works for analytical purposes may diverge greatly when the model is implemented within a "real world" institutional framework and in the process encounters questions of data accuracy, and availability.; The study analyzes the achievement of 1,014 students who took CPSC 101 between the fall term of 1996 through the spring term of 2004. The primary independent variable under scrutiny is the Clemson Math Placement Test (CMPT) which is used to place students in their first calculus course by the mathematical sciences department, a co-requisite for taking the initial computer science course. The relationship between the student's score on the CMPT and a student's performance in computer science has been historically assumed by the computer science department, but never tested until this study.; The analytical design uses multiple and logistic regression processes, the former to define a predictive model for student achievement in an introductory computer science course and the latter to test the efficacy of the model. The model developed and tested shows weakness overall with an explanatory R2 values of .168 and a decided inability to deal with the case of predicting the unsuccessful student, incorrectly classifying the student outcome 60% of the time. Further, the underlying data elements supporting the prediction are extremely limited and in some cases, questionable in their validity and utility. Much of the predictive model's failure can be traced to the differing environments separating a purely analytical or "explanatory" model, and the compromises that must be made to bring that model to bear "operationally" for predictions in real world situations.