Spotting student dropouts

Sotiris Kotsiantis of the Educational Software Development Laboratory at the University of Patras, Greece, explains how student dropout rates are much higher for universities providing distance education compared with conventional education.

Ensuring the dropout rate is as low as possible is a matter of economic viability for a given course and for the reputation of the university offering distance learning. However, predicting which students will dropout is not easy.

Data on students, class distributions, and dropout rates are usually rather skew, Kotsiantis says. This skewing of the statistics makes for low error rates in predicting overall class outcomes but produces unacceptable error rates when one is considering the minority of class members or individuals. He and his colleagues have carried out a systematic study of the various analytical methods that have been tried to overcome this limitation and found them all wanting.

The most obvious problem with analysing skewed data sets is that most approaches remover the “outliers” and focus on the average. But, it's the outliers at one end of the data who are the most likely to dropout.

As such, the team has devised a new statistical approach that offers a more effective solution to the problem by looking at the data at the local level where outliers become more prominent in the data set rather than being ignored by the law of averages. They tested their approach on retrospective student data from the Hellenic Open University and were able to effectively spot the student dropouts.

The method could allow universities offering distance learning the chance to predict which students are most likely to dropout of a course and so allow them to provide appropriate counselling, advice, and support to guide such students to a more informed decision regarding their educational future.

“It is of great importance for tutors to have recognised dropout-prone students before the middle of the period in order to be able to offer additional support to those students at risk,” Kotsiantis says. The new predictive approach therefore focuses on collecting data about student successes and failures before this point in their course is reached. This data is combined with known socioeconomic factors, gender, age, previous education, marital status, and occupation if applicable.