Managing plagiarism in programming assignments with blended assessment and randomisation

Plagiarism is a common concern for coursework in many situations, particularly where electronic solutions can be provided e.g. computer programs, and leads to unreliability of assessment. Written exams are often used to try to deal with this, and to increase reliability, but at the expense of validity. One solution, outlined in this paper, is to randomise the work that is set for students so that it is very unlikely that any two students will be working on exactly the same problem set. This also helps to address the issue of students trying to outsource their work by paying external people to complete their assignments for them. We examine the effectiveness of this approach and others (including blended assessment) by analysing the spread of similarity scores across four different introductory programming assignments to find the natural similarity i.e. the level of similarity that could reasonably occur without plagiarism. The results of the study indicate that divergent assessment (having more than one possible solution) as opposed to convergent assessment (only one solution) is the dominant factor in natural similarity. A key area for further work is to apply the analysis to a larger sample of programming assignments to better understand the impact of different features of the assignment design on natural similarity and hence the detection of plagiarism.