Which retention signals actually predict attrition
An analysis of first-term signals across 340 institutions: what predicts non-persistence, what merely correlates, and how quickly an intervention has to reach a student to matter.
Key takeaways
- Registration behaviour beats midterm grades as an early signal
- The intervention window closes around day 21 for online students
- Financial holds predict attrition more strongly than GPA below 2.5
- Advising cases with a named owner resolve 3.4× more often than alerts sent to a queue
The signals that matter
Across the institutions we studied, four signals carried most of the predictive weight in the first term: failure to register for the next term while still enrolled, an unresolved financial hold, a gap of more than ten days in LMS activity, and a missed first assignment.
Notably, midterm grades — the signal most early-alert programs are built on — arrive too late to act on for the students most likely to leave.
- Next-term registration not started by week 6
- Financial hold unresolved for more than 14 days
- LMS inactivity gap over 10 days
- First graded assignment not submitted
Speed beats sophistication
A simple flag that reaches a named advisor within a week outperforms a sophisticated model whose output lands in a monthly report. The differentiator in our data was ownership: cases assigned to a person resolved 3.4 times more often than alerts posted to a shared queue.
Measuring the intervention, not the model
Track outcome per intervention type. Most institutions discover that two of their six standard outreach templates do nearly all the work, which is far more actionable than a marginal improvement in model accuracy.