Research

Towards an Inclusive Analytics for Australian higher education

Bret Stephenson1, Andrew Harvey2 and Qing Huang1

Executive summary

Artificial intelligence (AI) and machine learning (ML) applications now quietly power countless automated decision-making, and predictive processes, across university business areas and throughout the student lifecycle. The recent challenge of the COVID-19 crisis, and the emergency shift to online learning, has also notably increased institutional interest in the adoption of AI/ML “business solutions.” While advanced data analytics techniques can be responsibly deployed to advance student equity interests, if adopted uncritically they can also amplify social inequalities and historical injustice, often by stealth. Moreover, it is increasingly difficult for non-specialist university leaders and decision-makers to anticipate how the AI/ML applications their institutions adopt may be working to undermine their own strongly held commitments to student equity and diversity. The proprietary nature of commercial analytics-powered products and services can also serve to frustrate a university’s attempts to audit the impact of these processes on equity students and equity interests more broadly.

In this report we identify the potential benefits of advanced analytics for student equity, and the institutional and cultural changes required for such potential to be fulfilled. We also argue, however, that the growing use of analytics involves risks and threats to student equity, further underlining the importance of institutional change, including educative and regulatory reform. We begin this report by providing a brief overview of the uses of advanced analytics within the higher education context. Analytic techniques now inform vast areas of the university and traverse the whole student lifecycle: from the recruitment and admissions of prospective students, through to the building of employability profiles of graduates. Part 1 also reviews many of the important conceptual and practical challenges involved in the quantification or datafication of equity, and equity cohorts, within the Australian context.

In Part 2 of this report, we outline the potential of analytics to protect and advance student equity. We highlight at least three related ways in which improved outcomes might be delivered for marginalised students, and for the broader cause of equity overall. First, analytics can help us to discover discrimination, including within historical processes such as admissions and course guidance. Second, analytics can help us to identify individual disadvantage, and move beyond our reliance on group membership assumptions. In the Australian context, this opportunity potentially enables institutions to move beyond the six conventional equity groups to consider dynamic behavioural indicators at the individual level. Proponents of analytics have long touted this capacity for a more sophisticated, individuated understanding of risk and disadvantage. We note, however, that this is a complex and contested area, and frequently requires consideration of protected characteristics, e.g., equity group membership, stereotype risks, and the perils of what has been called a “colour blind” approach. Third, we argue that analytics provides an opportunity to assess emergent and contingent forms of disadvantage, such as the impacts of COVID-19.

While the effective use of advanced analytics can demonstrably improve student equity, its uncritical adoption, and a failure to maintain effective oversight, can result in a dramatic undermining of equity goals. The report outlines these potential perils, across the four stages of the machine learning lifecycle: 1) data collection and preparation; 2) model building/learning or algorithmic training; 3) model evaluation and verification; and 4) model deployment. We consider various forms of “data bias” including historical bias, representation bias, measurement bias and aggregation bias, which can have the effect of reifying stereotypes, misrepresenting individuals, and exacerbating inequity. Further risks include “algorithmic bias” and interventions based on predictive analytics, such as tailored communications to “at risk” student. These analytics-driven interventions may be either ineffective or even counter-productive, in some cases leading to self-fulfilling prophecies of failure. Threats to privacy are also rising and remain priority areas for ethical and equitable implementation.

Collectively, our analysis reveals that advanced analytics is widely used but not closely monitored for its equity impacts throughout the university and its many business areas. Where equity protections do exist, these are typically limited to the teaching and learning functions of the university.

In the final section, we address the Fairness, Accountability and Transparency in Machine Learning (FATML) movement and provide a brief account of its constituent elements and limitations. While FATML typically focusses on technical aspects of machine learning, such as mathematical definitions of algorithmic fairness, we argue that this focus is necessary but insufficient to support student equity within the university. Defining, conceptualising, and prioritising equity requires an understanding of existing structural inequity and the broader frameworks in which advanced analytics operate. Such understanding itself requires engagement with philosophical, political and policy questions, and the expertise of staff outside the technology and analytics domains.

Ultimately, we highlight a pressing need for institutions to embed greater data literacy and equity consciousness across their organisations. Harnessing the potential of analytics to improve student equity requires a comprehensive institutional approach, and a range of sophisticated strategies and practices. In our discussion we briefly address the need for broader education and professional learning among both academic and professional staff. While not every staff member needs to understand the technical processes of advanced analytics, this knowledge needs to be distributed across the university, including within areas responsible for many of the analytics-informed interventions, e.g., student support staff, academic progression staff, and equity practitioners. Further initiatives could include greater monitoring and evaluation of the deployment of advanced analytics and of data-informed interventions. Such oversight could be partly provided by regulatory committees that embody specialist knowledge and that might operate similarly to existing ethics committees. Greater oversight and regulation should not, however, be an excuse for stifling institutional innovation. A program of inclusive analytics should instead provide the appropriate safeguards within which innovation can be leveraged to benefit all students in a spirit of inclusivity. Advanced analytics provides both an opportunity and a threat to student equity. Active engagement, education, and oversight are required to ensure that the emancipatory promise of technology can be fulfilled in the present age of advanced data analytics.

Recommendations

  1. That universities develop data analytics policies and procedures that protect equity interests throughout the full student lifecycle and across all business areas.
  2. That universities broaden distribution of analytic expertise, particularly within the DVC (Academic) divisions.
  3. That universities broaden distribution of equity and ethics expertise, particularly including within data analytics (institutional research and performance), Information Services, and ICT divisions of the university.
  4. That universities increase professional education of staff, including academics, engaged with analytics projects at each stage of the development and deployment process.
  5. That universities establish in-house regulatory structures and professional expertise to ensure equity and fairness are protected through the deployment of advanced data analytics, e.g., standing committees to oversee analytics, similar to ethics committees.
  6. That universities ensure that analytics-informed interventions are tailored, based on behavioural factors, and designed to reduce self-fulfilling prophecies based on immutable characteristics.
  7. That universities regularly monitor and evaluate the analytics project lifecycle for impact on equity and “fairness” interests.
  8. That universities work towards benchmarking/collective agendas, potentially involving Universities Australia (UA) leadership.
  9. That universities conduct and facilitate further interdisciplinary research into the intersection of equity in higher education and advanced data analytics as an urgent priority.

Read the full report: Towards an Inclusive Analytics for Australian Higher Education


This research was conducted under the NCSEHE Research Grants Program, funded by the Australian Government Department of Education, Skills and Employment.


1La Trobe University
2Griffith University, formerly La Trobe University

Posted 17 March 2022 By ncsehe