SLIDE 1
Educational Predictive Analytics: Navigating Disparate Views
Aaron Springer, Victoria Chou, Francis Martin Dominguez,
- Dr. Sam Foster, Dr. Steve Whittaker
Educational Predictive Analytics: Navigating Disparate Views Aaron - - PowerPoint PPT Presentation
Educational Predictive Analytics: Navigating Disparate Views Aaron - - PowerPoint PPT Presentation
Educational Predictive Analytics: Navigating Disparate Views Aaron Springer , Victoria Chou, Francis Martin Dominguez, Dr. Sam Foster, Dr. Steve Whittaker Predictive Analytics Use Past Data To Predict Future Performance Predictive Analytics Use
SLIDE 2
SLIDE 3
SLIDE 4
Predictive Analytics Use Past Data To Predict Future Performance
SLIDE 5
Predictive Analytics Use Past Data To Predict Future Performance
SLIDE 6
SLIDE 7
SLIDE 8
SLIDE 9
Why is this worth examining?
SLIDE 10
Ensign et al., 2018
SLIDE 11
Biases in Voice Interfaces
Springer & Cramer, 2018
SLIDE 12
UCSC is examining educational predictive analytics
SLIDE 13
- Implemented tool for advising staff
○ Note-taking and sharing ○ Appointment scheduling ○ Student search
- Predictive Analytics Component
○ Not currently enabled
SLIDE 14
SLIDE 15
SLIDE 16
Research Questions
- What positive and negative effects could predictive analytics have at UCSC?
- Do advisers at UCSC want to use predictive analytics?
SLIDE 17
Envisioned Negative Effects
Self-fulfilling Prophecies Reinforcing societal biases More resource allocation to already successful students Student data safety and privacy concerns
SLIDE 18
Envisioned Negative Effects
Self-fulfilling Prophecies Reinforcing societal biases More resource allocation to already successful students Student data safety and privacy concerns
SLIDE 19
Envisioned Negative Effects
Self-fulfilling prophecies Reinforcing societal biases More resource allocation to already successful students Student data safety and privacy concerns
SLIDE 20
Envisioned Negative Effects
Self-fulfilling prophecies Reinforcing societal biases More resource allocation to already successful students Student data safety and privacy concerns
SLIDE 21
“The advising community in general is pretty good about a holistic approach and we don't just rely on this—but we're also really overworked and it's easy to fall into these traps of just relying on this data and not taking the time to get to know the student”
SLIDE 22
Envisioned Positive Effects
Identify students in need of support more quickly Easier to reach
- ut to students
who need the most help Allocate resources more equitable
SLIDE 23
Envisioned Positive Effects
Identify students in need of support more quickly Easier to reach
- ut to students
who need the most help Allocate resources more equitable
SLIDE 24
Envisioned Positive Effects
Identify students in need of support more quickly Easier to reach
- ut to students
who need the most help Allocate resources more equitable
SLIDE 25
“You could run a report and find out who are my moderate people and who are my
- reds. Who do I need to take an initiative
to just to reach out and remind, ‘hey, I'm here’ ... if you can show them ‘hey, I'm thinking about you. Just checking in.’ I think it could be really beneficial that way and bridge the gap of scary advisor and a student.”
SLIDE 26
Do Advisers Want to Use Predictive Analytics
SLIDE 27
Implications
- Training could address some negative effects
○ Strategies to fight self-fulfilling prophecies ○ Clear examples and workflows of using this system to increase equity
- System should be built in ways that encourage population oriented behaviors
SLIDE 28
Should educational predictive analytics be used at UCSC?
SLIDE 29
System Deference
- Most advisers critically evaluated system predictions
- A small subset seemed to defer to the system as ground truth
○ 2 out of 17 advisers ○ “I also expect, you know, a high rating because they're currently on subject to disqualification. [advisor looks at rating] Oh but they’re moderate, that would kind of makes sense because the student right now has a grade point balance of 69.6”
SLIDE 30
- Are advisers overly willing to defer to predictive analytics?
○ People can think that even random predictions are accurate in some systems (Springer et al., 2017)
SLIDE 31