General Description
Policy Summary:
Predictive analytics, including operational tasks within data systems that simplify common activities (e.g., forecasting accounting entries based on historical accounting entries) and complex multivariate analyses to predict behavior and outcomes, are a valuable and integral part of modern business practices. However, with the transition to new data systems driven by Digital Transformation and the corresponding increase in the University’s capabilities, it is vital to ensure that predictive analytics are not used to perform actions that are inconsistent with Trinity’s mission and values.
Purpose:
The purpose of this Policy is to enable the University to use data in ways that support modern business practices while ensuring that our use of data is consistent with the University’s mission and values.
Scope:
This Policy applies to all University faculty, staff, and students, as well as third-parties within the University’s control (such as contractors and automated systems within university controlled software).
This Policy does not address who should have access to the data points used in predictive analytic tools, which is covered by Trinity’s Information Security Policy and the Federal Family Educational Rights and Privacy Act (FERPA).
This Policy does not address who should have access to the data points used in predictive analytic tools, which is covered by Trinity’s Information Security Policy and the Federal Family Educational Rights and Privacy Act (FERPA).
Exceptions:
Analyses conducted for class projects are exempted from this policy. If a predictive model developed for a class project is adopted for use in University operations, it must be reviewed in accordance with this Policy.
Policy Content
The Moral Risk Framework for the Ethical Use of Data is grounded in Trinity University’s values. Because these values may sometimes appear to be at odds when deciding whether to implement a predictive analytic tool, and because much of the ethical concern around the use of data relates to reinforcing or amplifying existing bias, the guiding principle for the Moral Risk Framework is Intentional Inclusion. Where bias is possible, Trinity will usepredictive analytictools to support and provide resources to ensure that no individual or group is excluded or penalized.
Before adopting a predictive analytic tool for University operations, the department in which the tool will be used must complete the Trinity University Analytics Risk Assessment Form and submit it to the vice president over their department.
The Analytics Risk Assessment Form guides end users through a series of questions to assess whether the proposed predictive analytic tool contains four factors that moderate the risk associated with using the tool.
- Potential for Damage: Potential for damage is greatest when the motivations of the end user determine how the predictive analytic tool is used. If the tool could be used to exclude or penalize individuals or groups, it is considered high risk. Additionally, regardless of the intentions of the end user, if the subjects of the tool might feel harmed by its outcome, it is considered high risk.
- Potential for Bias: Potential for bias is present when a predictive analytic tool uses data about people’s demographics, behavior, or attitudes/beliefs. This potential can be mitigated by transparency and agency, so whether a tool with potential for bias is considered medium or low risk depends on those factors.
- Lack of Transparency: Predictive analytic tools that lack transparency are often referred to as “black boxes” or “secret sauces.” If Trinity’s employees do not know what data points are being used in a tool, it is difficult for them to gauge the potential for bias. A transparent tool is one for which Trinity’s employees know what data points are being used and have the ability to decline to use data points that do not align with Trinity’s values. Another aspect of transparency is awareness on the part of the tool’s subjects that their data is being used in the tool. For example, users of LinkedIn Learning know that it makes recommendations based on their stated interests and past course completion.
- Agency: Predictive analytic tools that support agency make recommendations that individuals can choose whether or not to use and/or allow end users to access non-recommended options. Agency reduces the risk associated with use of a tool because a human is in the loop, monitoring the tool’s output and making decisions about whether or not to use it.
Based on responses to the Analytics Risk Assessment Form, predictive analytic tools are categorized into four levels of risk that require varying levels of testing and approval:
- Negligible Risk tools are generally transactional (the tool uses data about processes or things rather than about people), and they have no risk of damage or bias. They require:
- VP approval
- Pre-deployment verification that the tool produces accurate/expected results
- Low Risk tools have the potential for bias, but they enable the agency of end users/those whose data is used for the tool. They require:
- VP approval
- Pre-deployment verification that the tool produces accurate/expected results
- Annual verification that the tool produces accurate/expected results
- Annual review of the outcomes of the tool to ensure that they do not result in worse outcomes for people with protected status
- Medium and High Risk tools have the potential for damage or they have the potential for bias that is not mitigated by transparency and agency. They require:
- VP approval
- Consultation with internal elected officials (i.e., leadership of Student Government, TSEC, or Faculty Senate), as appropriate, when current students, staff, or faculty may be impacted by the tool.
- Compliance Committee approval
- Executive Planning Group (EPG) approval
- Annual verification that the tool produces accurate/expected results
- Annual review of the outcomes of the tool to ensure that they do not result in worse outcomes for people with protected status
Terms & Definitions
Terms and Definitions:
|
Term: |
Definition: |
|---|---|
| Agency | The capacity, condition, or state of acting or of exerting power. |
| Bias | Generally refers to any belief, attitude, behavior or practice that reflects, appears to be motivated by, or is perceived by the victim or victims to be motivated by an assumed superiority of one group over another. Bias is expressed through prejudice or discrimination and can be overt or covert. Bias can be directed against individuals or groups, but it can also be institutionalized into policies, practices and structures. While freedom of expression and the open exchange of ideas are a vital part of the educational discourse, bias activity dehumanizes people, erodes individual rights, debilitates morale, and interferes with the effectiveness of work and learning environments. |
| Predictive Analytics | The use of data, statistical algorithms and machine learning techniques to identify the likelihood of future outcomes based on historical data.https://www.sas.com/en_us/insights/analytics/predictive-analytics.html |
| Protected Status | Includes sex, gender, sexual orientation, gender identity, gender expression, race, color, ethnicity, religion, national origin, age, disability, genetic information, veteran status, and citizenship status. |
| Transparency | Characterized by visibility or accessibility of information especially concerning business practices. |
Attachments
Attachments:
Related Documents
Revision Management
Revision History Log:
|
Revision #: |
Date: |
Recorded By: |
|---|---|---|
| v2.0 | 7/31/2023 11:26 AM | Megan Mustain |
| v1.0 | 6/20/2023 9:11 AM | Holly Warfel |
Vice President Approval:
|
Name: |
Title: |
|---|---|
| Megan Mustain | Vice President for Academic Affairs |