Leaving the UW

10 May 2018

It has been six months since I left my job as a data scientist at the University of Washington. I joined the UW in 2012 with the personal goal of helping students graduate with less debt. I (naively) thought this would be possible by making a university more efficient. If students can graduate faster, if the right courses have more space, if UW funds are spent more intelligently, then everyone would benefit.

I spent 5.5 years doing this, over 3 of them as the UW data scientist studying student behavior. During that I built dozens of tools to help the UW run more efficiently, such as:

Tools to Help Students

Identify which students are trying to get into competitive degree programs and aren’t likely to be admitted. Identify and encourage alternate majors as ‘backups’, and the best set of courses to prepare them for any backups they’re interested in.
Find the fastest and cheapest ways to graduate from every single degree. For example, in Computer Science, there was over 2 quarters’ worth of difference between different course ‘sequences’. Make that information available to students.
Show a student their personal fastest and/or cheapest way to graduate.
Identify the AP/IB tests and community college classes that help prospective students graduate faster. Let them tailor the information for the career goals. Make this information publicly available.
Identify which classes make or break a student’s interest in a degree. Again, make that information public.

Tools to Help Faculty/Staff

Identify which students will leave the UW in the next 30-90 days, and their risk factors. Make this available to the appropriate advisors/faculty so they can proactively communicate and help retain struggling students.
Identify which courses fill up quickly, many of which have far more students interested than available space. Expose this information to academic departments so they can help their students. Expose it to students so they can plan for course-registration headaches.
Identify buildings and classrooms that are not used much. Build a course allocation system so a department can find ‘nearby’ rooms/labs for their classes, and other departments can ‘rent out’ their spare classroom space.. In the long run, this could save hundreds of millions of dollars in construction costs.
Shows instructors the basic composition of the students in each class: the most common previous courses taken, typical GPA ranges, which other courses are being taken concurrently, the current/likely degree programs for their students.
Show academic advisors which classes their students need to take to graduate, along with an estimated time-to-graduate for every person they counsel.

None of these ever saw the light of day. To my chagrin and horror, I realized there were no incentives for most staff/faculty to help students graduate with less debt, to help their department run efficiently, or to make bold/risky decisions.

Looking back, my work was doomed to fail, and I was blinded by hope, and didn’t see the clues:

I never read or heard the phrase ‘student debt’ by anyone employed full time at the UW. Not even once.
Doing machine learning and stats on student data, the terms I heard most were “enterprise” and “politics around funding”.
No one was willing to try anything new without scientific proof that it would work. Usually not even then.
I told a well-connected Vice Provost how to save hundreds of millions of dollars by not building buildings. I was quickly told to drop the subject.
The most common questions I heard to any suggestion were “what will this do to our budget?” and “what if I lose my job over this?”.

Eventually I realized that my efforts were not helping students, and they never would. If I wanted to change the UW to help its students, its faculty, and its research, I would need to make organizational & political changes, not technical ones. That’s a job for a different person.

So, I left for a different job, with the same motivation as in 2012: to make the world more equitable and just.