The Promise of Predictive Analytics
Predictive models can process application data — grades, test scores, extracurricular profiles, essay characteristics — at scale, identifying patterns associated with academic success. Proponents argue this enables more consistent evaluation and can surface talented applicants who might be overlooked in purely human review. The efficiency gains are real: models can screen thousands of applications in seconds, allowing human reviewers to focus attention where it creates the most value.
The Fairness Challenge
The fundamental challenge is that predictive models learn from historical data — data generated by institutions operating within societies marked by systemic inequity. A model trained to predict 'academic success' as measured by historical outcomes may systematically disadvantage applicants from underrepresented backgrounds, not because they lack potential but because the training data reflects structural barriers those applicants faced. The model does not create bias; it inherits and perpetuates it.
Fitzherbert University's Approach
At Fitzherbert University, predictive analytics may assist but never replace human judgment in admissions decisions. Models serve as one input among many in a holistic review process that considers academic achievement, intellectual curiosity, character, resilience, and potential contribution to the University community. No applicant may be rejected by algorithmic assessment alone. Every rejection is reviewed by a human admissions officer with full discretionary authority.
Ongoing Auditing and Accountability
The University's four-gate validation framework applies to admissions models: technical validation, bias auditing across demographic subgroups, ethical review, and committee sign-off. Annual bias impact reports are published under the Transparency Mandate. If any model demonstrates disparate impact that cannot be justified by legitimate educational criteria, it is immediately suspended pending remediation.