The Higher Education Commission (HEC)’s recently released report, From Bricks to Clicks - The Potential of Data and Analytics in Higher Education, describes factors contributing to and the challenges embedded in the sector’s surge of interest in learning analytics.

Bricks to Clicks exhorts higher education institutions to consider two analytics categories: descriptive and predictive.

Descriptive Analytics

In weather terms, descriptive analytics provides information like, “It rained more in February than in March.”

Descriptive analytics considers the past and current situation with the goal of uncovering meaningful insights. While these analytical insights may ultimately inform future decision-making in the form of business intelligence, they make no prediction about future events or behaviour. They simply describe the way things are. Nottingham Trent University’s dashboard provides this level of analysis.

Predictive Analytics

Predictive analytics allows the weather reporter to say, “There is a 40 percent chance of rain tomorrow.”

Predictive analytics aims to extrapolate information from historical data and current conditions to generate an actionable idea of what might happen. HEC points out that almost no UK institution has effectively harnessed the potential of predictive analytics, with the Open University’s Analyse initiative the notable exception.

Confusion and Conflation

To this point I’ve retreaded ground mapped in From Bricks to Clicks, but I think it’s worthwhile. In my experience speaking with universities about student success and technology, I encounter quite a bit of conflation.

It’s not surprising that student success practitioners and leaders desperate for insights into complex student behaviour often ascribe predictive powers to descriptive data. It’s human nature to predict and anticipate future occurrences based on past experience. The danger in this approach is the assumption built in that what has occurred or is occurring is likely to continue in the future.

In the complicated, messy, human realm of education, that is a deeply uncertain proposition. Is a student with zero swipe-ins at the library in the past week deeply in trouble or simply occupied elsewhere? It’s so tempting to generate a predictive, forward-looking narrative from what is simply a description of what’s happening in spite of a total lack of predictive data. If you go looking for trouble, you’re likely to see it everywhere. 

It’s Raining. Now What?

I posit that what most institutions are looking for is a third flavour of analytics, which goes unmentioned in the HEC inquiry: prescriptive analytics. This category provides advice and suggestions that guide toward a solution.

In weather terms, prescriptive analytics tells us, “There is an 80 percent chance of rain tomorrow; pack your umbrella.”

Ultimately, it’s prescriptive analytics that I believe most institutions will seek out. Prescriptive analytics are the Grail Quest of learning analytics in that they are solution-focused. Descriptive and predictive analytics are important but ultimately settle in as detailed descriptors of problems without suggesting any solution or way forward for the student or students under analysis.

The HEC report tacitly acknowledges this point in its gentle endorsement of “formative” over “summative” analytics, suggesting that descriptive and predictive analytics at their best are only triggers for further conversation. Prescriptive analytics can help guide that conversation to a specific, data-driven outcome by suggesting interventions that have worked in the past.

Case Study: PAR Framework

The Predictive Analytics Reporting (PAR) Framework demonstrates how a group of colleges and universities in the United States agreed to use descriptive and predictive analytics to look for variables contributing to student loss.

Researchers were excited to see that they could identify factors that put a student at risk with more than 90% reliability. However, they quickly realized that offering a prediction of student risk without also offering recommendations for mitigating that risk was almost worse than not knowing about the risk at all. As a result of the preliminary observations, PAR’s analytical efforts quickly broadened to include:

  • Descriptive information about the student populations contained in the now more than two million student, 20 million course dataset;
  • Predictive models that include risk scores for each student in the institution;
  • Prescriptive information about interventions most likely to address the factors of risk identified in the predictive analyses at the best point and time of need.

Descriptive and predictive analytics are important for honing our understanding of what’s happened and what may yet happen. But I believe it’s prescriptive analytics, with its focus on action and data-guided advice, that represents the highest form of learning analytics. We may be three to five years away but prescriptive analytics will have its day in the sun. No umbrella needed.


Visit the Resource Center