Back in early 2015, the SAS (Stats and Act Sci) department was having a discussion about CS students in the two required STAT courses (230: Probability and 231: Statistics). While CS students are as strong as other Math students in most courses, they were systematically underperforming in STATs.

The idea that got the most traction was to pilot a special section of STAT 230, which would focus on the CS applications of probability. The hope was that by including more relevant examples, and pointing out the many important tie-ins to later CS courses, the students would be more engaged and performance would improve.

But who could design and teach such a section? I immediately and enthusiastically volunteered! I had many years’ experience teaching probability courses. And to help me develop CS-centric examples I could leverage my many close friends (including my husband) who were UW CS grads working in the tech sector. When I reached out to my network via Facebook, I received over 100 ideas.

The SAS department gave me their blessing to start working on it, so in Spring 2015, I taught the first CS section of STAT 230. To be fair to all students, examinations were identical between sections. Although I wasn’t given dedicated time to develop the new section, I was able to introduce minor changes in that first term, and make small tweaks in each subsequent term. It wasn’t necessary to overhaul the entire course in the first pilot, merely to start the process and then build upon it.

I began by incorporating interesting CS examples of STAT 230 concepts into the lectures, often substituting out toy applications (e.g. balls and urns) with more appropriate ones (e.g. binary trees, databases, sorting algorithms). I wrote CS versions of questions for quizzes and tutorials, and many of these were later added to the course notes. This had the beneficial side effect of providing better breadth of practice problems for all students and providing resources for other instructors of the CS section of the course. By ensuring materials would be available in the future I increased the probability that this differentiated section would be sustainable and my investment in the course would live past my involvement.

In the pilot course I polled students for games they liked, and used the results for my "Stats Weekly Application to Games" (SWAG). We discussed how probability applies to Poker, League of Legends, The Resistance, Hanabi, and more. Each example or application was introduced right when we had just covered the applicable theory in the course; for example, we talked about card-counting in Blackjack right after learning the Hypergeometric and Binomial distributions. In later terms I added a "Machine Learning Idea of the Week" (MLIW) segment including Bayesian classifiers, the "bag of words" algorithm, and ethical use of data.

After the first two terms of the pilot, we conducted a survey. An encouraging 59% of the students said they appreciated the tailored material and 62% said we should continue offering the special section.

After several more terms with CS sections, we then had enough data to reliably draw conclusions about the effect on student performance. While it would be easy to evaluate the change in CS students’ marks, this would be statistically invalid, since there could be other factors (easier exams perhaps, or a cohort effect) such that any change wouldn't be attributable to the CS section. (Never ask a statistician for an unqualified answer!) But what we could do was look at how the gap between CS and non-CS students had changed in terms with a CS section versus terms without.

We found that in terms without CS sections (11130 students), the average difference between CS and non-CS students was 4.2%; whereas in terms with CS sections (4126 students), the average gap closed to 0.9%.

By meeting the students where they are, using examples that are relevant to their interests and future education, and helping them see the applications of the material to their lives and studies, we have improved the performance of CS students and closed the performance gap.

Quite a few students said they were now more interested in pursuing upper year STAT courses than they were before they took STAT 230, and many said they appreciated the effort put in. But by far the most common comment received from students: do the same thing to STAT 231! :)