Were Students on a Repeat Attempt as Likely to Be Active 12 Weeks In? A Cohort Retention Reading
Updated October 2026
Group people by when they first appeared, count each group at its own age, and read a gap with its range.
Students on a repeat attempt of a module were less likely than students on a first attempt to be active twelve weeks after they first appeared: 29.7% against 42.1%, a gap of 12.4 percentage points with a 95% range of 5.5 to 19.3 points below zero. At four weeks the gap is also below zero, but that reading depends on who is included in the group, and we hold it loosely. Here is the data, the two readings, and how to read a gap when one group is small.
The data and the question
The question: are students on a repeat attempt of a module as likely to be active as students on their first attempt, 4 weeks and 12 weeks after they first appear? "Active" means at least one click on the course site in that week only. It does not mean continuous activity, and a student can be active in week 12 without having been active in every week before it.
Data: one module presentation, CCC 2014J, from the Open University Learning Analytics dataset [1] [2]. We used two columns of information per student: when they were first active on the course site, and whether they were on a first attempt or a repeat attempt.
A cohort here is the week of a student's first click. We took the students first active in presentation weeks 0 to 3, which gives four cohorts of 1,801, 256, 117 and 16 students, pooled into one group of 2,190. Of these, 1,988 were on a first attempt and 202 on a repeat attempt. Students whose first activity came after week 3 are not in the group.
The retention measure is one age at a time. A student counts at age N if they had at least one click in the Nth week after their own first active week [3] [4]. Age 4 and age 12 were the only two ages compared, and we fixed them, the groups and the test before running this analysis. The test is a two-sided two-proportion test with a continuity correction, with 95% ranges, and a Holm correction over the two comparisons.
The dates in the file are day offsets from the start of the presentation, placed on a Monday so that weeks line up. They are a label, not the actual calendar, so this page speaks only in weeks since first active.
What the table shows at 12 weeks
| Group | Students | Active at age 12 | Percent |
|---|---|---|---|
| First attempt | 1,988 | 837 | 42.1% |
| Repeat attempt | 202 | 60 | 29.7% |
Students on a repeat attempt minus students on a first attempt: -12.4 points, 95% range -19.3 to -5.5 points, Holm-adjusted p = 0.0017 (unadjusted p = 0.00084).
The whole range sits below zero, so the data are consistent with repeat-attempt students being less likely to be active twelve weeks in. The range is wide, because the repeat-attempt group is small.
What it shows at 4 weeks, and why to hold it loosely
| Group | Students | Active at age 4 | Percent |
|---|---|---|---|
| First attempt | 1,988 | 1,560 | 78.5% |
| Repeat attempt | 202 | 141 | 69.8% |
Students on a repeat attempt minus students on a first attempt: -8.7 points, 95% range -15.5 to -1.8 points, p = 0.0063 (Holm-adjusted also 0.0063).
This one is also below zero in this pooled group, and it is the weaker of the two readings. Among only the students who first appeared in week 0, the gap at four weeks could not be told apart from no gap. Add the students who first appeared in weeks 1 to 3, and it can. So the age-4 reading depends on who is included, and we do not present it as a stable finding. If you replicate this, report it with the population stated beside it.
How to read a gap with a small group
The repeat-attempt group has 202 students, against 1,988 on a first attempt. A small group gives wide ranges, as the two 95% ranges above show. Our reading rule: take the size of the gap as roughly the range, not as the single number. A gap of -12.4 with a range from -19.3 to -5.5 says "somewhere from about 5.5 to about 19.3 points below zero", not "exactly 12.4".
Two phrases are easy to mix up when a range includes zero. "Cannot tell" means the data are too thin to separate the groups. Our addition: "the same" would need a different test and a stated margin, and a range that includes zero supports the first reading, not the second.
What this does not tell you
- It is not causal. The groups are defined by prior attempts. Students on a repeat attempt can differ in other ways, for example prior exposure to the material, and the table does not show why the two groups differ.
- It is one module presentation. CCC 2014J is one course in one term, and the pooled cohorts share it.
- It is two ages. No other age was tested, so the table says nothing about the weeks in between.
Run it on your own file
You need a person ID and an activity date for each activity, and optionally a group column. Set the period to week, because the free tool defaults to a month. The tool groups people by the period of their first activity and reports retention one age at a time. Run the free cohort retention tool on your own file.
Questions people ask
These questions are inferred from the sources, not taken from observed search queries.
What is cohort retention?
The share of a group that started together who come back, period by period. The start event and the return event define the group and the count [3].
What is the difference between "exactly in period N" and "in period N or any later one"?
The first counts a person only if they return in exactly the stated period. The second counts them if they return in that period or any later one [3] [4]. The measure on this page is the first kind: active in week N only.
Why not average all the students?
Cohorts differ, and totals alone can hide differences between groups [5]. Reading by age, group by group, shows where they part.
Does this show that repeat attempts cause lower activity?
No. It describes who was active, not why. Students on a repeat attempt can differ in other ways, for example prior exposure to the material.
Why is the age-4 result held loosely?
Because it depends on who is included. Among only the students first active in week 0, the four-week gap could not be told apart from no gap; adding those first active in weeks 1 to 3 makes it distinguishable.
Data: Kuzilek, Hlosta and Zdrahal, Open University Learning Analytics dataset, UCI Machine Learning Repository, DOI 10.24432/C5KK69 (CC BY 4.0). Changes: grouped by week of first activity and compared by MCP Analytics; the dates in the extract are day offsets, not the actual calendar.
References
- Kuzilek, J., Hlosta, M. and Zdrahal, Z. (2017). Open University Learning Analytics dataset. Scientific Data 4:170171. doi:10.1038/sdata.2017.171
- Kuzilek, J., Hlosta, M. and Zdrahal, Z. Open University Learning Analytics dataset. UCI Machine Learning Repository. doi:10.24432/C5KK69 (CC BY 4.0)
- Adobe. Analytics, retention rates. Read October 2026. Page
- Hackle. Retention analysis. Read October 2026. Page
- Wikipedia. Cohort analysis. Page
Three checks on this reading
What we checked before we trusted the numbers.
One clear meaning of active
Active means at least one click on the course site in that week only, not continuous activity, and the same rule is used for every student.
A range, not a guess
At twelve weeks the gap is -12.4 points with a 95% range of -19.3 to -5.5, Holm-adjusted p = 0.0017. The repeat-attempt group is small (202 students), so the range is wide.
Who is included
Students first active in weeks 0 to 3 are included, pooled from four cohorts. Among only the students who first appeared in week 0, the four-week gap could not be told apart from no gap; adding the later starters makes it clear, so hold the four-week reading loosely.