The analysis and prediction of student progression through degree programmes : a cohort analysis of undergraduate students at the university of Cape Town

Abstract

A simplified cohort survival analysis was used to investigate the academic progression of first-time entering undergraduate students within four large bachelors' degree programmes at the University of Cape Town. The rates of graduation, academic exclusion and voluntary drop-out were quantified in relation to the matriculation authorities and prior matriculation performance of the students within each of the four cohorts. The results of the analyses served to identify specific areas of concern with regard to the internal efficiencies in student progression through each of the four degree programmes, and it is suggested that the availability of information of this type will be essential in the attainment of the institutional transformation goals set out in the 1997 White Paper on the transformation of higher education in South Africa. Significant relationships between the matriculation criteria and the final academic outcomes of students within each cohort were detected using log-near modelling. By means of multiple discriminant analysis, significant predictor variables of the final undergraduate academic outcomes within each cohort were identified. However, the relatively weak discriminatory powers of the multiple discriminant models and the poor predictive accuracy of the associated classification functions suggest the variables included in these analyses did not adequately explain the variability in the final undergraduate academic outcomes of students within the selected cohorts. The extent of the voluntary drop-out phenomenon within each of the cohorts was quantified in relation to matriculation criteria, and further analysis of the cohorts indicated that factors other than academic difficulty appeared to have prompted the greater proportion of the voluntary withdrawals. Those students who had dropped out voluntarily were therefore not included in either the log-linear models or the multiple discriminant analyses.