In 22 OULAD course presentations, 44% of flags for a missing first tutor-marked assignment are for students already withdrawn

We tested a fixed withdrawal warning at the first tutor-marked assignment deadline in 22 OULAD course presentations from 2013–2014. The rule flagged a missing timely submission. The main cohort had 31,925 eligible student-course records after registration, banked-credit, and date-quality exclusions. Of 10,622 flags, 4,644 (43.72%) were for students whose recorded withdrawal date was on or before the deadline. These past events formed 69.15% of apparent true positives for final withdrawal. Excluding past withdrawals reduced precision from 63.23% to 34.66%, a change of −28.57 percentage points. The nominal 95% student-cluster bootstrap range for this change was −29.48 to −27.73 points. Among active students, later withdrawal was still more common with a flag than with a timely submission: 34.66% versus 13.99%. Changes to the cutoff day, grace period, assessment type, and treatment of undated withdrawals kept lower active-cohort precision. A separate parser checked 127,988 values, and a full repeat matched all eight analysis files. This audit measures the size of an established risk-set issue in one historical source. It does not test an intervention, a live warning system, or the models in prior papers. The package has source tables, code, provenance, figures, all output, and repeat instructions.

learning analytics · withdrawal · early warning · risk set · OULAD · assessment timing · reproducible audit