Righting “Mutant” algorithmic wrongs starts with education

During the Covid-19 pandemic, data science modelling has been used to justify government actions around lockdowns, the easing of measures and the reintroduction of restrictions on a local basis. These actions continue to affect the lives and livelihoods of millions.

The importance of analytical calculation and competence has been brought home, often brutally, to households and institutions everywhere. At the same time, industries from retail through to insurance have faced huge changes. They, too, drew heavily on analytics to predict customer demand, understand changing customer behaviour, and compete effectively in an upside-down world.

The result has seen unprecedented demand for analytical skills. However, the pandemic has also shone a spotlight on the teaching methods used in data science courses. Decisions made on the back of data analytics – including the severity of lockdown restrictions and what regions should be locked down and for how long – have generated controversy and criticism.

Whether these were the ‘right’ decisions is up for debate, but it’s been made abundantly clear that the quality of algorithmic decisions has never been more important. Ensuring the fairness and success of model-based decision-making starts with the education of those building, monitoring and interpreting tomorrow’s algorithms.

Introducing analytical case studies

Simply creating more data scientists to meet demand won’t necessarily improve the quality of model-based decision-making. Universities and employers alike need to rethink which methods are most effective for teaching students the skills that they need to survive in the ‘real world’, where data is not always – or even mostly – clean and complete.

It is generally agreed by higher education institutions that case studies are one of the most effective tools for teaching analytics. These are analytical challenges based on real-world data. Case studies have long been used in business schools for more general learning and have a very good reputation. Case study ‘clearing houses’ have been set up to assure the quality of these teaching resources, and there is considerable competition to get cases accepted for publication. The best cases are studied around the world by tens of thousands of business students every year.

Businesses see case studies as ways to enhance their reputation. They often commission case studies about their organisations, using them to highlight successes or to raise brand awareness with an influential demographic.

However, until recently, the number of quantitative case studies developed for academic use has been limited by data privacy and data protection issues. It is generally hard to get hold of real-world data that is both sufficiently useful for students, and which doesn’t disclose confidential or business-sensitive information. These issues can’t be minimised or diluted, but it needs to be recognised that data science students must have case studies if they are to learn the skills they need.

The national analytical library

To address the shortfall in quantitative case studies, we need to actively seek out new analytical cases with accompanying data. Compiling these together can create a new academic case library for use by higher education institutions across the company.

The SAS Case Library provides a model for how this national analytical library could work. Experts continuously commission and curate new cases for the library. They can therefore ensure the library reflects a range of common analytical techniques applied across different industries. The cases are also tagged to show their suitability for different levels of academic attainment. Where new content is developed for existing cases, it is only added to the library after careful review.

The SAS analytical case studies are written by academics to ensure they are valuable for an academic audience. They follow a globally recognised format for academic case studies, and are grouped by industry and analytical topic, to make it easier to navigate. All the studies are data-rich and based on real-world problems. Step-by-step demos, analytical games, and ideas for datathons and hackathons are available to encourage students. Access to free software and training materials is also provided.

Drawing on this structure can form the basis of more effective teaching of data science and modelling. With real data and real use cases, data scientists in training will have a greater appreciation for how business and data interacts on a daily basis. Of course, this information shouldn’t just be limited to data science departments. All subjects can derive valuable insights from data, which plays an ever-greater role across society. The students of today will soon be the data decision makers of tomorrow. We need to ensure they are ready.

Geoffrey Taylor, Director, Academic Engagement Manager , SAS UK & Ireland