Author: Jeff Coon
Department of Cognitive Sciences
Editor-in-Chief: Alex Bower
February 20, 2022
Statistics courses are fundamental to virtually every STEM major, yet they are often viewed as a necessary evil. Students generally choose their major because they enjoy learning about a subject that interests them, and they are typically far less excited about the math involved in establishing those facts. Beyond students’ enjoyment, statistics courses pose an equity issue because they can act as intimidating gatekeepers (Onwuegbuzie, 1999). For students who come from underserved backgrounds – and who might struggle to envision themselves as scientists – the sudden dose of math can be overwhelming.
Flipped classrooms can be valuable in this context, as such a course design can provide social support and guidance when students are doing one of the most difficult and stressful parts of the course, namely applying what they have learned to complete graded assignments. I thus planned to implement such a course design when asked to teach the final course in the Cognitive Sciences statistics sequence.
However, I was specifically asked to teach the course remotely. With the course involving considerable coding in R, this arrangement posed a significant obstacle for implementing a flipped classroom. When students work over Zoom, we can all share screens with each other, but the viewer cannot directly tinker with the code. This makes it difficult for me to help them with errors and for them to learn, both from me and from each other.
Indeed, such a setup resembles classic “yoked learning” experiments in psychology. In these experiments, two participants are trying to learn about the same environment, but only one participant actively interacts with the environment. The other “yoked” participant can only watch what they do (e.g., Walls et al., 1975). Results generally show that, even though both participants see the same information, the participant who can actively interact with the environment learns more about it. Active participants can focus on gaps in their knowledge and test any hypotheses that might occur to them (Gureckis & Markant, 2012). I began wondering whether a flipped classroom would be worthwhile if students were not able to actively work on code collaboratively.
Fortunately, I had help from DTEI’s online instructional experts as a Summer Graduate Fellow. They encouraged me to explore Ed Discussion, a digital learning platform to which UCI already had a subscription. Thanks to Alex Fang at Ed Discussion, I was able to use a feature called Workspaces. The Workspaces are like editable Google Docs, but for runnable code. I could assign groups to their own Workspace, where they could work on the same code and run it to see if it worked. As the administrator, I could pop in and out of Workspaces to check on their progress.
There were certainly hiccups in using the Workspaces. In particular, the students found the console to be somewhat confusing. To run their entire script, they needed to type the file name into the main console. But to type code directly into the console, they first needed to enter an R-specific mode, which they would have to exit if they subsequently wanted to run the entire script.
Another issue was that students ultimately needed to submit a PDF, produced by knitting an R Markdown file. This format let me include open-ended conceptual questions as well as coding questions. However, it also meant that students needed to copy whatever work they did in the Workspaces over to the appropriate code chunks in the R Markdown file. Understandably, some students decided it was more efficient to work directly in the R Markdown file. They were correct, so long as they never encountered coding issues that they couldn’t solve. But if they did run into trouble, they had to transfer their code to the Workspace for me to work with it. They often tried to cut corners, only transferring the part where they ran into trouble, which was hardly any better than them sending me a screenshot of what they believed to be the problematic section of code. Fortunately, most students who regularly attended class began the quarter by diligently using the Workspaces and only cut corners as they became more confident. Thus the Workspaces acted as a self-differentiating resource, providing support for students who were nervous about coding.
In the future, I might tweak my approach. Ed Discussion offers a more structured feature called “Lessons,” where instructors can mix conceptual questions with coding challenges in a pre-set sequence. The coding challenges are all within a specific coding environment (e.g., R), avoiding the confusion some students encountered with using the console in Workspaces. The Lessons feature also allows students to submit their work directly, avoiding the issue of students making their code more difficult to share by working directly in R. While I do think there is value in having students implement all aspects of an R script, including loading and manipulating data, there are probably ways of constructing Lessons to achieve a similar outcome.
That being said, the Workspaces worked quite well overall. Many students pointed to the group assignments as one of the most effective aspects of the course, providing social accountability as well as practical and moral support. I was especially gratified to see that, of the 18 students who had indicated on a pre-course survey that they were very worried about their coding abilities detracting from their performance in the course, 16 of them earned at least an A-.
My experience speaks to the value of creating flipped classrooms when teaching a remote statistics course, as well as the challenges involved. The most fundamental challenge of creating such a classroom is that I was asking students to learn a new technology, and I was already asking them to work with new concepts in a relatively new coding language. Yet the effort ultimately proved worthwhile, as the flipped classroom provided students with crucial support when it came time to apply the concepts they had learned.
References
Gureckis, T. M., & Markant, D. B. (2012). Self-Directed Learning: A Cognitive and Computational Perspective. Perspectives on Psychological Science, 7(5), 464–481. https://doi.org/10.1177/1745691612454304
Onwuegbuzie, A. J. (1999). Statistics Anxiety among African American Graduate Students: An Affective Filter? Journal of Black Psychology, 25(2), 189–209. https://doi.org/10.1177/0095798499025002004
Walls, R. T., Rude, S. H., & Gulkus, S. P. (1975). Model and Observer Learning of Low, Medium, and High Level Concepts. Psychological Reports, 37(2), 671–675. https://doi.org/10.2466/pr0.1975.37.2.671