Jeff Leek

March 16, 2021

Design choices in taking our class online for the pandemic

Like many faculty, last summer we had to scramble to figure out how to transition what had always been an in person course to an online format. I have taught Advanced Data Science off and on for the last 8 years. While I've taught a lot of classes online, this has never been one of them! 

So last summer we set out to re-design the course to make it "pandemic ready" for students at Johns Hopkins. I had a few principles in mind when we were designing the course: 

  1. I wanted it to be mostly asynchronous because I wanted it to be flexible for students schedules. 
  2. I wanted to make sure we got in person interaction, but wanted to maximize the interaction I consider most valuable for students that you can't usually get from online courses - free form discussion with professors, TAs and other students
  3. I wanted to make sure interactive sessions were accessible to people all over the world, regardless of time zone (since many of our students couldn't in the US) 
  4. I didn't want to have sessions after working hours in the US 
  5. I wanted to make the didactic content available to anyone who was interested
  6. I didn't want assignments to be onerous to do or to grade, since I knew everyone would be under strain
  7. I wanted to find ways to be efficient with my time, since at the time we had no childcare and two working parents. 

Usually this course is taught as a lecture-based course that meets twice a week, with take home data analysis assignments of varying difficulty over the course of the term. Here is how we specifically adapted the curriculum/format for the pandemic situation. 

  1. Lectures were released Sunday night as a chapter in a bookdown book that was available online: http://jtleek.com/ads2020/
  2. We also distributed the lectures via Substack so people could subscribe to get the lectures in their inbox
  3. Homework assignments consisted of two parts - a written response submitted via Github: https://github.com/advdatasci/ and an in class discussion. Students were asked to take turns providing a summary of their response for each homework assignment (two summaries per week)
  4. Class discussions involved breakout rooms with one professor or TA assigned to each breakout room with guided questions about the homework assignment to lead the discussion and notes were distributed after each discussion. 
  5. Instead of two lectures on two different days, we scheduled two discussions one in the morning US Eastern Time and one in the afternoon US Eastern Time. These were two versions of the same discussion, so students could choose which one worked best for their time zone. 
  6. We leveraged the Open Case Studies project heavily https://www.opencasestudies.org/ for pre-built data sets and structured analysis assignments. 
  7. Most updates/notes/communication was distributed via a classroom Slack. 
  8. Homework assignments distributed via Github using the dev version of the ghclass package

I was worried about how this would work, but overall I'm about as happy with this setup as I could have been given the conditions and constraints. The student reviews from our course were also mostly positive (granted, they may have been showing some grace/empathy to their instructors as well :)). Overall the things I think worked and didn't work were:

Worked

  1. Two synchronous discussions at different times - this seemed to make it relatively straightforward for people all over the world to participate, much easier than having two lectures at the same time of day. 
  2. I really enjoyed the conversations and I think there was a lot of cross-student learning that was a pretty useful replication of the real life Hopkins experience. Interaction with students and TAs is, by far, the best part of the class in person and we got a good fraction of that going online as well. 
  3. Releasing the content publicly as a book - admittedly some of the students wished there were live lectures - but a) I find live lectures difficult to prep/deliver even in normal times, b) I find live lectures on zoom brutal and c) I like that we didn't have to require students attend a particular lecture or watch a video to get the course material. 
  4. Releasing the content via Substack - originally I was re-formatting all of the course material from the book into the platform - which was so silly in hindsight - so I just started adding a link to the relevant book chapter. I liked seeing that there was a broader audience for this material as well - with a thousand subscribers by the end of the course it felt like I was putting material out in a user friendly way. 
  5. Breakout rooms for discussion was really useful. It was much easier to have conversations among 4-8 people than it was to have a conversation among 25-40 people. 
  6. Having an easy asynchronous communication platform like Slack was really helpful for making last minute announcements, diagnosing Zoom problems, and getting quick questions from students answered. 
  7. Having shorter assignments, largely focused on reading other people's data analysis at the beginning of the term was really useful for being able to illustrate concepts without over burdening students. 
  8. ghclass was pretty useful for automatically creating repos and collecting assignments. 


Didn't Work

  1. I did not anticipate  just how much work it would be to write out all the lectures in long form. I'm glad its done now, but it made for a lot of really late Sunday nights. 
  2. The Zoom breakout rooms didn't randomize people into different groups each week. The original plan was for students to rotate among groups to get a chance to meet other students/TAs/profs and get different feedback. Overall this wasn't a total failure though, I feel like the consistent breakout rooms led to more comfortable discussion as the small groups got to know each other. 
  3. Having the class scattered across different platforms (Github/Slack/Bookdown/Substack) was a little frustrating. In retrospect I think I might try using Basecamp for everything next year, since we've been having pretty good luck managing projects on that platform. The downside, obviously being, that less people are familiar with that platform. 
  4. Some students really would have preferred at least some live lectures. I wouldn't mind doing this in the future, but given the time constraints of the pandemic I am still happy this is what hit the cutting room floor. 


I'm sure there were other things our students could report that didn't work as well :). But I have to say, the students handled the class incredibly well, were prepared for the discussions, and really contributed interesting ideas and learned from the lectures as well as from each other. So I'm going to call it a success!