Orthogonal versus redundant data in education
By David Topps
Our university was recently hit by malware. It took weeks to restore most data, some lost completely. All backups were affected making a simple restore ineffectual. Why take this redundant approach in data safety? Consider flight data recorders: not all aircraft parameters are recorded or recordable. But critical analysis is afforded by reconstruction from key data such as flap angles. Automotive insurers accurately derive driving habits and risks from brake usage alone: a minimal, easily collectible dataset. Capture data that is orthogonal to, not redundant to, that currently recorded. In education, do we need to track every word written or spoken? Actions speak louder than words. Sentinel behaviours better predict anomalous performers, as learners or teachers. Activity streams, not learning objects, are now a better indicator of what a learner actually does, not what the teacher says they do. But we should beware what activities we monitor. Aside from privacy issues, simple metrics like word counts or time spent are easily spoofed. Fact regurgitation in exams does not prove comprehension. Look at resultant activities ( e.g. rich comment threads in discussions) and emergent properties from existing workflows. We have the tools, and the data. We just need to use them.