Enrollments Steadily Decline After Launch
Key Observation
Nearly all courses see a drop in weekly enrollment of at least half within three years of their launch. Half of the courses reach that decline in just one year.
Background
Duke has produced 63 Coursera MOOCs since about 4.5 years ago. Over that time we have recorded the number of new public users signing up for our courses each week. The graph above shows data from all those courses for the first three years after each one launches on the platform (see below for more details on the methods used). Duke has one of the largest sets of course offerings on the platform, and also one that has a diverse array of topics.
Regardless of the subject matter or the timing of the course start date, we see a remarkably strong trend of declining enrollment. No course exhibits a ‘viral’ upswing in enrollment or benefits long from a major marketing push or course revision.
Method
Coursera provides partner institutions like Duke with access to the database of student interactions with our courses. Figure 3 shows the number of new learners enrolling in three of our courses, plotted against the real date for the start of that week. We can see here that one course (orange) started around the beginning of 2016, another in early 2017 (blue), and the last in mid-2017 (red).
When drawing on that data for weekly enrollments, I made two further adjustments to arrive at the results above. The first was to shift the time axis to represent the number of weeks after the launch of the course.
The second adjustment was to normalize the enrollment axis by averaging the first four weeks of enrollments for every course, and then divide the enrollment for every week of the course by that average. This will start every course at a y-axis value of approximately 1 (or 100% of its initial average enrollment). In this example data set, the three courses after normalization appear so.
Reproducing these two steps across all 63 courses resulted in the initial graph at the top.
Discussion
Why is it happening?
Our most plausible hypothesis so far is that Coursera’s user base is not growing as fast as their course catalog, so that the number of students is being diluted by many course choices. This might explain the overall trend, but doesn’t satisfactorily explain why interventions such as marketing or revision have minimal effect, nor why there are no courses that experience a more organic growth in enrollment.
How can we apply this knowledge?
Results like this one can help prospective and current instructors as well as course support teams better understand what to expect in the future. This result suggests that the first month of enrollment is very telling about the long term reception in the Coursera marketplace. Instructors and support staff can then make informed choices about sustainable course support options. The ineffectiveness of current marketing and revision approaches may encourage more experimentation with other options.
What else can we infer?
A small subgroup of courses performed far worse than the main trend, especially in the first year of operation. Each of these courses seems to have inadvertently caused a mismatch between student expectations (based on the course description, course title, or marketing) and their experience in the course. This may have caused poor reviews that induce a sharp decline in new enrollments. After roughly the first year, each of these courses seems to have reached a new stage where the rate of decline eases up to parallel the main trend.
For Further Info
If you would like to know more about the analysis conducted here, or would be interested in partnering in future research, please reach out to us at learninginnovation@duke.edu or share some details with us via this short form https://lile.duke.edu/contact/