One theme that emerges in several of the conversations with experts in learning analytics is that the data being used for learning analytics is not ideal in its current format: it is largely derived from logging systems, initially designed for tracking bugs in software and monitoring server performance rather than for understanding educational processes. Dragan Gašević suggests that we need to collect data that is measuring aspects of learning, helping us, for example, to understand levels of understanding, metacognition and the affective states of students. Alyssa Wise proposes that we consider analytics right from the start when designing systems in the future:

We’ve come into the game half-way through with a lot of analytics working with the data that systems already produce – which sometimes works out well and sometimes could work out a lot better if we collected better, smarter data.

Trying to draw inferences from clickstream data, she says, is challenging. We could instead start collecting ‘data that has more information in it’, thus setting up the processes of inference in advance rather than subsequently. Alyssa also believes we could provide more insight by connecting different data sources together more effectively. She notes that, because the data is often aggregated, the temporal dynamic is lost. Knowing how long events take and how their timing relates to other events is important too, as learning is about changes over time.