The gap, plotted
Coursera's headline rating is an average of every star a course ever received. Starisk is what the most-upvoted written reviews say. Put 798 courses of both on the same axes and the two stopped being the same measurement a long time ago.
Two distributions, one axis
Share of courses at each star rating. Coursera's ratings sit almost entirely in the top bucket; ours use the whole range.
Coursera Starisk
The same scale, all 798 of them
One dot per course, both axes 1–5. On the diagonal the two agree; below it the headline is kinder than the reviews.
Where the gap is widest
The eight largest gaps between the headline and the reviews.
- GenAI for Legal Ethics and Practicality4.4 → 1.0(3.4)
- Sequences, Time Series and Prediction4.7 → 1.5(3.2)
- Essential Tools for Data Scientists: Git, Jupyter & More4.5 → 1.4(3.1)
- Data Visualization with Python: Hands-On Projects4.5 → 1.5(3.0)
- Electronic Health Records4.0 → 1.0(3.0)
- Mastering Data Analysis in Excel4.2 → 1.3(2.9)
- Developing Front-End Apps with React4.4 → 1.6(2.8)
- Python for Data Science & AI: Hands-On Development4.6 → 1.8(2.8)
Every number here is the same one shown on a course page: the most-upvoted written reviews, weighted by the square root of their votes. Nothing was recomputed for this page.
The caveats travel with it. We read one page of reviews per course — about twenty — so a score is a reading of the most-endorsed reviews, not a survey. 166 scored courses have no Coursera rating to compare against and are absent from the A course counts as different when its score rounds to a different number of stars than Coursera's: 402 of 798 do.The full method.