Helping School Reopening Using Operation Research (part II)
Arnaud Sahuguet & Amélie Marian | NYC Response Lab
Previously On …
Last month, we wrote an article describing a tool to help schools assign students to non-overlapping cohorts for social distancing.
A lot has happened since then:
- Schools still haven’t reopened in New York City :-(
- Most large school districts in the U.S. have decided to start the year remotely.
- We have spoken with half a dozen school principals who had expressed interest after reading the article.
- Based on these conversations, we have made some improvements to the tool.
- The code behind the tool is open-sourced and available on github.
What We Have Learned
In the rest of this article, we will share some of our learnings.
Assigning students to cohorts is only the tip of the iceberg
“Which days your children will go to school” is the most obvious and visible change for parents. But this is only one of the many problems faced by school administrators.
Remote learning has to be managed. Special computer equipment has to be provided to students and teachers. Classrooms have to be retrofitted. And when capacity is not enough, new locations have to be scouted and prepared. Not to mention transportation, lunches, bathrooms, lockers, etc.
Some teachers may choose to opt-out of in-person teaching out of medical reasons.
And on top of that, things can change drastically from one day to the next: city or state ordinance, in-school COVID-19 outbreak, the decision by parents to move their kids to remote or to in-person.
We realized that our tool only scratches the surface of the problem.
Trust is key
When interacting with school administrators and people from the NYC Department of Education (DOE), being affiliated with well-known academic institutions (Rutgers and Cornell) really helped.
Being parents with kids in the NYC school system showed that we had some skin in the game.
Having contacts at the DOE helped to get introductions to school principals: for the past year, Amélie has been part of the NYSIP team for District 2 and has worked on an integration plan to decrease disproportionality in the district’s schools. Knowing parents at various schools in the city was also key to getting introductions.
Our initial model was a good enough approximation of the problem
Our initial model was based on (a) articles we had found on the topic (mostly from the New York Times), (b) documents from the NYC DOE and (c)Amélie’s domain expertise having two kids in NYC District 2 public schools.
The examples of “constraints” we had mentioned in the article were a good fit for what we heard from school administrators. They did not come up with any new ones we could not capture.
A suggestion that emerged from the conversations was the ability to balance homerooms/group/cohorts based on a given criteria, e.g. gender. Assuming the information is available, this is something our tool can do effortlessly.
Lots of schools we spoke with already had their homeroom assignment and did not want to let the tool change that for them. We realized that assigning students to homeroom is an art rather than a science and personal knowledge about students is what makes the difference. From the tool’s point of view, this simply means adding one extra constraint per student, i.e. the homeroom that has been pre-assigned to them.
Meet your “customers” on familiar ground: use an Excel-like tool!
Our tool takes a list of students (say 500), some constraints (potentially the same order of magnitude as students, if homerooms have been pre-defined), and enriches the list with the name of cohort each student has been assigned to.
Rather than reinventing the wheel, it was very clear that the user interface of the tool had to be something like Excel (we are currently using Google Sheets).
The tool uses two worksheets: one for the list of students (above) and one for the constraints (below).
Blindspots
During our conversations, we identified two blindspots for our audience: one on the tech side, one on the use-case side.
On the tech side, few people were familiar with operation research techniques and the kind of problems that can solve. Bringing the data into a spreadsheet was not as obvious as we imagined and seems to require a lot of manual entry. FERPA (see next section) of course did not help.
On the use-case side, most school administrators we spoke with looked at our tool as a “one-of” for which the cost of entering the data by hand into the spreadsheet was on par with doing it manually. But things seem to be changing as the reopening saga unfolds. One school-principal we met last week told us they anticipate that cohorting may have to be redone every other week and for that kind of scenario, our tool could be a game-changer.
We are not dealing with Luddites. School administrators are under immense pressure, are fighting fires on a daily basis, are receiving conflicting messages from city officials and state officials, and don’t seem to be getting clear guidance from the NYC DOE. “Every school for themselves” seems to be the current situation.
FERPA is a major innovation roadblock
The Family Educational Rights and Privacy Act (FERPA) is a federal law that affords parents the right to have access to their children’s education records, the right to seek to have the records amended, and the right to have some control over the disclosure of personally identifiable information from the education records.
In our case, this means that schools cannot share with us any student data, e.g. names, or school-issued unique identifiers like OSIS (NYC DOE OSIS number is a nine-digit number that is issued to all students who attend a New York City public school).
Our tool does not care about names. It only needs to uniquely identify students, for instance when you want to make sure two students don’t end up in the same cohort. This information falls under the “directory data” exception to FERPA (but we are not lawyers). Understandably, school administrators are unwilling to take a risk unless the DOE gives them the green light. To work around these issues, we have explored various routes:
- A Google Sheets interface: Google Suite products are FERPA compliant, and public schools in NYC already use Google products for their education needs. We created a version of our tool that reads student data and constraints from a Google Sheet and writes the result of our assignment in one of the columns in the sheet ("Cohort" column in red in the screenshot above).
- A FERPA-compliant server: our tool is written in Python and can be called from a web browser or a command line. We investigated several options to have our tool hosted in a FERPA compliant server: a NYC DOE server, or a NYC city agency server, or a server from an approved third party. Unfortunately, all of these require time to go through some red-tape.
- Running the code on the school local machines: several schools have programmers, or tech teachers, who are proficient in python and could install and run the tool locally.
- Anonymizing the student data: this is the option that is currently recommended. Anonymizing data is relatively easy for school administrators, but de-anonymizing can be tedious and, because it will require some spreadsheet manipulation, is likely to introduce human errors.
What’s next?
The code of the tool is now open source. Feel free to use it and modify it as you see fit at https://github.com/NYC-Response-Lab/cohorts-public .
We have packaged the tool as a cloud service at http://cohorts.cornelltech.io . Please shoot us an email at info@cohorts.school if you want to test it.
We remain committed to helping schools. As the situation will develop, students and teachers will opt-in and out of in-person learning, social distancing guidelines will evolve, and more schools will open with a variety of staggered schedules. Our tool can help adapt to these changes and reduce both the administrative overhead and the undesirable changes in the schedule for individual students and teachers.
We are also exploring ways a similar tool could be used by small and medium businesses to re-open with staggered teams. Let us know if you want to discuss a specific use-case.
