Validation and Probabilities of Misclassification of Derived Discriminant Models for the Academic Performance of Pre-major Engineering Students

Victor V. Hafalla Jr., MAAS, REE


The study sought to validate recently derived models which discriminate between high, moderate and low achieving pre-engineering students of the University of Baguio using data of students for SY 2005-2006. Specifically, the study sought to provide profiles of the different student groups formed, compare classification matrices, and determine hit ratios, misclassification probabilities, and profiles of misclassified cases on the analysis and holdout samples. Results of the study confirmed the feasibility of the derived models for discriminating students during their pre-major years into high, moderate and low-achievers. The discriminant functions formed may be used to profile incoming freshmen of the College of Engineering. This measure, together with the Freshmen Engineering Students’ Academic Intervention Program (FESAIP), results to selective or prescriptive entry of the first year students to the College, leading to a more academically capable student body. It can also be used to assess the proportion of pre-major engineering students belonging to each of the aforementioned groups. The resulting bias of the derived discriminant model of 3.1% is modest as compared to the model’s hit ratio of 92.0%. Furthermore, results of the study confirm findings from previous studies that the discriminant functions are most accurate in classifying high and low achieving students. The classification accuracy of the model derived also shows that it is substantially better than a chance model (92%>CPRO=44.56, Press’ Q=70>6.63). The profiles of the analysis and holdout samples on the different factor constructs and variables indicate that Factor 3 (Parental Influence and SES) and its variables do not show significant mean differences between the groups. Profiles of the misclassified cases show that most of the misclassification occurs in classifying individuals originally from moderate achiever group into low achiever group. It may be possible to re-estimate the discriminant functions combining moderate and low achiever groups or make use of extreme-poles comparison. It is also possible to derive discriminant models for incoming third year engineering students taking into account their second year GPA, instead of their high school GPA. Further validation of the classificatory power of the derived models may also be gained by re-estimating and testing (computing hit ratios from classification matrices) using the U-method or jackknife approach.

Source: UB Research Journal, Vol. XXXII, No. 1, January – June 2008