The OU centre for STEM pedagogy
Predicting the passing probabilities of an online module has always been an interest in higher education. It helps in planning, decision making, providing feedback to students and also in helping students work around any difficulties or milestones that are predicted to affect their results. Different models have been built to predict the passing probability as a binary variable (pass/fail) based on a set of explanatory variables.
In distance learning, and specifically at The OU educational system, Associate lecturers (ALs) are the main contact point with their allocated students. Across successive presentations, ALs are expected to build their expertise about the modules they are teaching and also on the anticipated characteristics and abilities of their students. We believe that incorporating the ALs’ opinion and expertise together with the data can be highly beneficial in building predictive models and will markedly enhance the predicted probabilities.
Moreover, for more accurate planning and to tailor the feedback to the required level of each student, a better model should allow the prediction of probabilities for all possible pass grades (pass1, pass 2 and so on) instead of just (pass/fail).
The proposed project aims at building a Bayesian multinomial logistic model with more than two outcomes that allows for the experts’ knowledge (ALs in this case) to be quantified as a subjective prior distribution for the Bayesian analysis. To the best of our knowledge, ALs knowledge and expertise on students performance have never been implemented with the available data during the modelling stage of analysis. It is anticipated that including the ALs input in this way at the modelling stage will help predicting more accurate probabilities of students’ results.
We aim for a two-year project where the methods and tools are used in the first year to build the models, elicit the ALs’ opinion and predict the probabilities of different outcomes. This will provide a full prediction system that can be used for future presentations. In the second year of the project we aim to test and evaluate our proposed system. The predicted probabilities computed with ALs' input will be compared to those based only on the data and both will be compared to the actual module results. This should give a clear idea on the impact of incorporating ALs’ knowledge into modelling the analysis.
The proposed system will provide a tool that can be used in successive presentations of different modules, where ALs will be able to input their own knowledge and opinion on students performance to model the probabilities of passing the module. Based on these accurately predicted probabilities of different outcomes, both ALs and students will efficiently work together on the actual students needs according to their predicted level of competency in the module work. For example, student support can be efficiently tailored to individual students and hence improve students’ satisfaction and/or retention. For example, students with potential risk of failure can be identified to get more support.