Learning analytics · open data
Which online courses lose their students, and when? A flow map of 32,593 learners
Early, and very unevenly. Across seven Open University courses, withdrawals range from 12% of students (GGG) to 45% (CCC), and half of all withdrawals happen by the week the first assignment is due. Students who hand that assignment in on time pass 64% of the time; the rest pass 13% of the time. Below, every student flows from registration to their result, course by course.
12% to 45%
of students withdraw, depending on the course: GGG keeps the most, CCC loses the most. Pass rates range from 38% (CCC) to 71% (AAA).
50%
of the 10,156 withdrawals happen by the week the first assignment is due, usually between week 1 and week 4 of a course that lasts about 35 weeks (week 8 in GGG).
64% vs 13%
passed among students who handed in the first assignment on time, against those who handed it in late or not at all. The gap holds in every course.
48% vs 26%
of clicks went to forums and live sessions for distinction students, against students who failed, in course DDD. The distinction share is the higher one in all 22 course runs.
Loading students…
- Colour shows the course; each label on the left has its swatch.
- Red channel: students withdrawing
- Diamond: first assignment due
How to read the course map
- Time runs left to right, from registration to the final result. Week 0 is the week each course run started; runs from 2013 and 2014 are pooled.
- Each river on the left is one course, AAA to GGG, as wide as the students still enrolled that week. The diamond on each river marks the week its first assignment is due.
- Red channels are students withdrawing. They drain into the grey river at the bottom, which is split by course, so you can see which course loses students and when.
- At the right, every course pours into the same four results: distinction, pass, fail, and withdrawn. The width of each course's channel into a result is how many of its students got there. Hover a course to trace its route; hover a result to see which course has the highest share.
- Every moving dot is a student (one in four in the all-courses view). Their colour shows whether they handed in the first assignment on time, late, or not at all, once it is due.
- Click a course, or pick it above the map, to open its own week-by-week map. There, rows are what students did each week: studied the materials, took quizzes, used the forums and live sessions, handed in an assignment, made no clicks, or withdrew. The small ticks above each row are the course items, placed at the week they were used most.
- Click or drag the bar strip to look at one week: its channels light up and the dots show only that week's moves. Press play to walk through the course a week at a time, or pick Whole course to see everyone flowing at once.
How the course structure was rebuilt from clicks
- Inside one course, the map uses that course's 2014J run. The dataset gives each course item a planned week, but the field is empty for almost every item (1 of 365 in DDD 2014J). So each item's week is inferred from use: the click-weighted median day it was opened.
- DDD 2014J has 361 items that anyone opened. Its map marks the 77 that at least 30% of students opened; clicks on the rest are counted on the nearest marked item of the same kind.
- The course home page is visited by everyone, every week, so it only decides a student's row when they opened nothing else that week. A student's row in a week is the kind of item they clicked most, or the hand-in row if they submitted an assessment.
- Clicks are daily totals per item, with no order inside a day. The map therefore moves week by week, not click by click.
- Items are anonymised in the dataset, so the map shows their kind and timing, not their titles.
What the data says about dropout
Courses differ more than any single signal. AAA passes 71% of its students and loses 17%; CCC passes 38% and loses 45%. GGG has the highest share of distinctions (16%) and the fewest withdrawals (12%), but also the highest fail rate (29%). The table under the map has every course.
The first tutor-marked assignment (TMA 1) is a strong early warning in every course. In DDD 2014J, 1,312 students handed it in by day 20 and 58% of them went on to pass. Of the other 491, 7% passed and most withdrew. Across all 22 course runs in the dataset, on-time students passed at least twice as often in 20 runs; the two exceptions are the AAA runs, where the gap is smaller.
Withdrawal is mostly decided early. Across all courses, half of the 10,156 withdrawals happened by the week the first assignment was due. Of the 647 students who withdrew from DDD 2014J, 384 (59%) had either unregistered by the TMA 1 deadline or made no clicks in the week before it. The bottom row of the map fills fastest in the first weeks.
How students spend their time matters as well as how much. Distinction students put 48% of their clicks into forums and live sessions, against 26% for students who failed, and the distinction share is the higher one in every run. That is a correlation, not proof that forums cause better results.
Open course EEE to compare: its first assignment is due in week 4, and in its 2014J run on-time students passed 77% of the time, against 12% for the rest.
About the data
The map uses the Open University Learning Analytics Dataset (OULAD), released under CC BY 4.0 and described by Kuzilek, Hlosta, and Zdrahal in Scientific Data (2017). It covers 22 runs of seven modules in 2013 and 2014: 32,593 students, their assessments, and 10.6 million daily click totals from the virtual learning environment. The copy used here comes from the UCI Machine Learning Repository. Student records are anonymised, and the map shows no individual information beyond what the dataset publishes.
Questions about the course map
What is OULAD?
The Open University Learning Analytics Dataset is an open dataset from the Open University in the UK. It records anonymised students on 22 runs of seven modules in 2013 and 2014: who they were, when they registered or withdrew, every assessment hand-in and score, and how many times a day they clicked each item in the online course.
Which Open University course in OULAD has the most dropouts?
Course CCC: 45% of its 4,434 students withdrew across its two runs. GGG lost the fewest (12%). AAA has the highest pass rate (71% passed or passed with distinction) and GGG the highest share of distinctions (16%).
What is the earliest sign that a student will drop out?
In this data, missing the first assignment. In module DDD 2014J, 58% of students who handed in TMA 1 by its week-2 deadline passed, against 7% of those who did not, and 59% of the students who later withdrew had already gone quiet by that deadline.
Why does the map move week by week and not click by click?
The dataset stores clicks as daily totals per student and item, with no time of day. Within a day there is no order, so the smallest honest step for a route is a day, and a week reads more clearly.
How was the course structure worked out?
The planned week for each item is missing in the dataset, so each item is placed at the click-weighted median day of its use. Items opened by at least 30% of students are drawn; the rest are merged into the nearest drawn item of the same kind.
Can I use this for my own course or platform?
The method works on any learning platform that logs who opened what and when, such as Moodle or Canvas: infer the structure from use, mark the deadlines, and watch the cohort move. The first deadline is usually the cheapest moment to reach students who are slipping.
Related tools
Most utilities process files in your browser. Any tool that uses a server or external service says so on its page.
View all tools- AI Pricing ExplorerBrowse the pricing database directly with presets, filters, pinned comparisons, source freshness, and inline history.
- Twitter Advanced SearchBuild X/Twitter queries with real-world operators beyond the built-in advanced search.
- Twitter Lists FinderSearch for Twitter lists on specific topics using advanced query operators.
- Twitter Places SearchFind Twitter place IDs for location-based searches using the _within:_ operator.
- Twitter (X) Tools HubJump into the advanced search builder, operator guide, lists finder, places lookup, and capture guides.
- Breed MatchmakerRank dog breeds against apartment life, kids, first-time ownership, low shedding, quiet homes, activity, and hot climates.
- Compare Dog BreedsCompare up to three breeds with plain-English tradeoff summaries instead of raw stat tables only.
- Dog World AtlasBrowse dog breeds by origin on a world map and jump into breed pages from each country cluster.