Attrition Dashboard¶
Monitoring dashboard — instantiates Attrition and Dropout Monitoring
Tracks dropout as it happens — sliced by arm, site, subgroup, time, and reason — so selective loss surfaces while the study is still running, not after it ends.
A study can pass its final attrition check and still have been compromised for months without anyone noticing. Attrition Dashboard exists to shorten that blind interval. It is a live, sliceable view of dropout rates over time, disaggregated by arm, site, subgroup, calendar period, and reason code, refreshed as loss accrues rather than tallied at the end. Its defining move is to catch loss that is differential or emerging — one arm bleeding faster than another, one subgroup quietly filtered out, a rate bending upward mid-study — while there is still time to respond. Where a flow diagram is a photograph taken at the finish, the dashboard is the instrument on the dashboard: it is watching, continuously, and its entire value is the warning it gives before the data are frozen. It detects; it does not decide what to do.
Example¶
A district is running a multi-site tutoring trial across twelve schools. Overall attrition sits at a comfortable-looking 9%, and a final report would likely have called that acceptable. The dashboard tells a different story in month three. Filtered by subgroup, it shows that students without reliable home internet are dropping from the online-tutoring arm at nearly three times the rate of their peers, and the trend line is climbing. The aggregate stayed calm because those students were a minority; the dashboard's per-stratum slice made the concentrated loss visible while enrollment was still open. That early signal is what lets the team act — before the "effect of tutoring" quietly becomes "the effect of tutoring among students who could stay online."
How it works¶
Its distinguishing discipline is continuous, stratified comparison:
- Read the event register live. The dashboard is the running face of the attrition log, recomputing counts as new loss events land.
- Compute rates by stratum, not just in aggregate. Arm, site, subgroup, time window, and reason each get their own cut, because an acceptable total routinely hides a compromised slice.
- Trend, don't just tally. Rates are plotted over time so an accelerating or diverging series is caught as a shape, not a single endpoint number.
- Flag on pre-set thresholds — then hand off. Crossing a threshold raises a visible alert; the dashboard surfaces the signal but does not itself set the response.
Tuning parameters¶
- Refresh cadence — how often the view updates; faster catches fast collapse but costs monitoring load and can amplify noise.
- Stratification depth — how finely loss is sliced; deeper cuts catch subgroup filtering but multiply the comparisons and the spurious flags.
- Threshold sensitivity — how large a divergence trips an alert; tight thresholds catch more but risk alert fatigue.
- Smoothing window — how much short-run fluctuation is averaged out before a trend is called real.
- Alert routing — who is notified and how loudly; too quiet and signals are missed, too loud and they are tuned out.
When it helps, and when it misleads¶
Its strength is early, disaggregated warning: it is the mechanism most able to defeat false reassurance from low overall attrition by exposing the one arm or subgroup that a pooled rate conceals. Recognized standards for weighing differential attrition[n1] as a threat to a study's validity exist precisely because concentrated loss, not total loss, is what breaks a comparison.
It misleads when many fine-grained slices are watched at once and random fluctuation throws spurious flags — reading noise as signal — or when constant alerts breed fatigue and a real divergence is dismissed as more of the same. The guarding discipline is to pre-specify the strata and thresholds that matter before the study starts, so the dashboard tests a short list of real hypotheses rather than trawling every cut for something alarming.
How it implements the components¶
differential_attrition_monitor— its core function: comparing dropout across arms, sites, subgroups, and time so selective loss is caught, not averaged away.attrition_event_register— it consumes the register as a live feed, turning the log of individual loss events into running, sliceable rates.
It does not draw the static end-of-study denominator cascade for reporting (denominator_transition_frame, its nearest twin Participant Flow Diagram) — the dashboard is live and comparative, not a snapshot — and it detects loss but does not set the threshold-triggered response or decide continuation (attrition_response_trigger_policy, Data Monitoring Review).
Related¶
- Instantiates: Attrition and Dropout Monitoring — the dashboard is the archetype's real-time detection organ.
- Sibling mechanisms: Participant Flow Diagram · Withdrawal Reason Survey or Interview · Completer Balance Table · Missing-Data Sensitivity Analysis · Retention Outreach Protocol · Data Monitoring Review
Editorial Notes¶
Form Classification¶
Form family: Monitoring, Sensing & Alerting
Rationale: Tracks dropout as it happens — sliced by arm, site, subgroup, time, and reason — so selective loss surfaces while the study is still running, not after it ends, making its operative form repeated observation of actual state that emits measurements, status, or alerts.
Independent corroboration: The frozen evidence defines Attrition Dashboard as 'Tracks dropout as it happens — sliced by arm, site, subgroup, time, and reason — so selective loss surfaces while the study is still running, not after it ends', so its operative form is Monitoring, Sensing & Alerting.
Review outcome: Independent reviewer agreement; high confidence.
Origin Attribution¶
Primary origin: Statistics & Experimental Design
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: Experimental methodology treats overall and differential attrition as threats to internal validity requiring arm and subgroup monitoring.
Related originating lineages:
- Education & Pedagogy — Education evidence standards explicitly grade study validity by differential attrition.
- Medicine & Healthcare — Clinical-trial operations institutionalized live participant-retention monitoring.
Review resolution: Statistics and experimental design are the agreed primary lineage for monitoring loss to follow-up. Education and healthcare provide major applied traditions, but the established dashboard remains specialized to longitudinal cohort and study integrity.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The dashboard and the Participant Flow Diagram both make loss visible, but along different axes: the diagram is a static map of denominator states drawn once for the reader, while the dashboard is a live series of rates watched by the team. Confusing the two leads to the common failure of drawing a beautiful final flow diagram and never having monitored anything at all.
[n1] Differential attrition — when dropout rates or the composition of dropouts differ across comparison groups — is treated by evidence-standards bodies (such as clearinghouses that grade education and social-policy studies) as a distinct validity threat with its own tolerance thresholds, separate from overall attrition. Referenced here as the concept the dashboard is built to surface, not as a specific published standard. ↩