The relationship between selection test results and performance of students at the University of Stellenbosch Business School

Andrag, H. W. (2005-03)

Thesis (MBA)--Stellenbosch University, 2005.

Thesis

ENGLISH ABSTRACT: Business schools, businesses and prospective students will benefit from an indication of prospective MBA students’ likelihood of success in their studies. This study examines the relationship between GMAT and SHL selection tests and performance of students at the University of Stellenbosch Business School. The aim is to establish the feasibility of using models derived from the selection tests in order to identify students who are unlikely to succeed. It was found that variables analyzed in GMAT and SHL numeric and verbal tests as well as the SHL OPQ 32-test correlate significantly with weighted average marks on the MBA programme. Significant correlations were also found between GMAT and SHL numeric and verbal tests and the marks obtained in subjects with high failure rates. Different variables correlate significantly with weighted average marks depending on the mode of study. Said correlations were however found to be too weak to build a model to predict, with accuracy, the performance of a student based solely on the results of selection tests. Adding the subject Computer Literacy strengthens the models to the extent that discriminant analysis can identify many of the students whom would be expected to fail. Prediction efficiency of discriminant models is however not high enough to allow its sole use as basis for exclusion of prospective students. Linear models could not predict any of the students who failed to achieve a weighted average mark of 50% or above. Linear regression models could however explain 27.8% to 52.6% of variability in weighted averaged depending on the method of study and selection test taken. Linear regression and discriminant models can thus be used as part of a judgement based selection process or as a basis for the provision of guidance to individuals, it is however not suitable for use as sole measure in admissions decisions.

AFRIKAANSE OPSOMMING: Besigheidskole, besighede en voornemende studente sal baat vind indien hul ‘n indikasie kan kry van die waarskynlikheid van ‘n voornemende student se sukses. Die studie ondersoek die verhouding tussen GMAT en SHL toetse en prestasie van studente aan die Universiteit van Stellenbosch Bestuurskool. Die doel was om vas te stel of dit wesenlik is om modelle, wat van die toelatingstoetse afgelei is, te gebruik om studente wat waarskynlik nie sal slaag nie, te identifiseer. Daar is gevind dat veranderlikes in die GMAT en SHL numeriese en verbale toelatingstoetse sowel as die SHL OPQ32 toets wesenlik korreleer met die geweegde gemiddelde punt vir die MBA program. Wesenlike korrelasies is ook tussen GMAT en SHL numeriese en verbale toelatingstoetse en die punte behaal in vakke met hoeë druipsyfers gevind. Verskillende veranderlikes korreleer op ‘n wesenlike vlak met geweegde gemiddelde punte afhangende van die metode van onderrig. Bogenoemde korrelasies is egter nie sterk genoeg om ‘n model te bou, suiwer gebaseer op toelatingstoetse, wat met akkuraatheid die prestasie van ‘n student kan voorspel nie. Deur die vak Rekenaargeletterdheid by te voeg kan die model sodanig versterk word dat diskriminante analise baie van die studente wat sou druip, kon identifiseer. Die voorspellings effektiwiteit is egter nie hoog genoeg om diskriminante modelle as enigste basis vir die weiering van studente te gebruik nie. Lineêre regressie modelle kon nie enige van die studente wat gedruip het identifiseer nie. 27.8% tot 52.6% van ‘n variansie in geweegde gemiddelde punt kan egter deur lineêre regressie modelle voorspel word, afhangende van die metode van onderrig en toelatingstoets wat geskryf is. Lineêre en diskriminante modelle kan gebruik word as deel van ‘n oordeel gebaseerde keuringsproses of as basis vir die voorsiening van raad aan individue. Dit is egter nie geskik vir gebruik as enigste keuringsmaatstaf nie.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/80763
This item appears in the following collections: