How do you measure outcome duration and drop-off?
Define the outcome, record when it is observed and follow it across a justified period. Distinguish evidence collected at checkpoints from evidence covering the intervals between them. If you project beyond observation, label the duration and drop-off assumptions, document their basis and test how much the conclusion depends on them.
Use this reference when a report asks whether an outcome lasted, rather than only whether it was present at one check. Build a follow-up evidence table, distinguish outcome status from missing responses and write a bounded claim. The optional calculation shows how a stated drop-off assumption affects a forecast.
A report can say that graduates were employed nine months after exit without knowing whether they stayed employed throughout those months. It can also estimate future benefits, provided the reader can tell where observation ends and estimation begins. Neither task is solved by removing the date from a favorable number.
Separate four questions that sound similar
- Outcome status: Was the defined outcome present at a particular check?
- Outcome duration: For how long did it continue, according to the evidence and definition?
- Outcome retention or reduction: How did the outcome change among a stated population over time?
- Survey attrition: Who stopped providing usable follow-up information?
Someone who misses a survey has an unknown survey result, not automatically a lost outcome. Someone employed at two dates may have had a gap between them. Someone with a lower score may still meet a defined outcome criterion. Keep these distinctions in the dataset and the report.
In SROI, drop-off is an adjustment for how an outcome's contribution reduces in later periods. It is not the same as the percentage of people who fail to answer the next survey. A forecasting assumption should not silently become a statement about measured individual experience.
Define what “lasted” means for this outcome
For employment, are you measuring any paid work at the follow-up date, continuous employment since placement, a minimum number of hours or earnings above a threshold? Each is a different outcome definition. Write the criterion before interpreting the result.
For a confidence or wellbeing measure, persistence might mean remaining above a justified threshold, retaining a specified amount of change or showing a particular pattern on a validated instrument. Do not adopt a universal one-point rule. For a service outcome, continuing access and continuing benefit may also differ.
Record the starting event. Exit is useful for a post-program question, but an outcome may begin before exit or much later. If you measure duration from first employment, preserve that date rather than treating graduation as the start of every job.
Ask affected people what persistence means in practice and what interruptions matter. This can improve the definition and follow-up questions. It does not eliminate the need to check dates, missing information and alternative explanations.
Choose evidence that fits the time claim
A current-status question tells you about a checkpoint. A carefully worded retrospective question can ask about the period since the previous check, although recall can be imperfect. Appropriate administrative records may provide dated events or regular observations. Use the source that fits the question and permitted purpose.
There is no universal rule that two post-program surveys prove duration while one can never contribute evidence. A single follow-up with a credible dated history may tell you more about an interval than two current-status questions. Conversely, two positive checkpoints do not prove uninterrupted benefit between them.
Keep planned and actual follow-up dates. Define collection windows and the rule for late responses. A nine-month report should not silently combine observations five to sixteen months after exit under one label.
Choose follow-up timing around the outcome, the decision and the burden. Three, nine and eighteen months may suit one program and miss the important changes in another. If further contact is not feasible, state the limit and consider other appropriate evidence rather than inventing a longer horizon.
Worked example: follow-up status is not continuous retention
In this fictional exercise, all 100 graduates were in paid work at exit. Everyone is eligible for the planned follow-ups. The team asks about employment at each checkpoint, not continuous employment since exit.
| Checkpoint | Status known | In paid work | Not in paid work | Status unknown |
|---|---|---|---|---|
| Exit | 100 | 100 | 0 | 0 |
| Six months | 80 | 60 | 20 | 20 |
| Twelve months | 70 | 49 | 21 | 30 |
At six months, 60 of 80 people with known status are in paid work: 75%. At twelve months, it is 49 of 70: 70%. Those are respondent-based status percentages. They do not establish a five-percentage-point decline for the same people because the observed groups may differ.
Of the full cohort, 49% are confirmed in work at twelve months, 21% are confirmed not in work and 30% have unknown status. Reporting the unknown group is more informative than silently treating it as employed or unemployed.
Now suppose 60 people have known status at both follow-ups. Their fictional transitions are:
| Six-month status | Twelve-month status | People |
|---|---|---|
| In work | In work | 35 |
| In work | Not in work | 10 |
| Not in work | In work | 7 |
| Not in work | Not in work | 8 |
Among these 60, 45 were in work at six months and 42 at twelve months. The net change is three fewer, but it includes ten moving out of work and seven moving into work. A single net number hides that movement.
Of the 45 matched people employed at six months, 35 are also employed at twelve months: 77.8%. Say “employed at both checks.” Without interval evidence, do not call this continuous employment retention. The ten observed transitions out of work are also not automatically a causal program failure or a reusable annual SROI drop-off rate.
Describe uncertainty without hiding the finding
The thirty unknown twelve-month outcomes matter. In an extreme scenario where none were employed, the full-cohort employment percentage would be 49%. If all were employed, it would be 79%. These are arithmetic bounds under the stated binary definition, not a confidence interval or a prediction.
Use the attrition review to investigate available evidence about nonresponse. Check contact problems, known exits and differences in earlier records. Do not claim that similar baseline scores prove the missing group would have the same outcomes.
For an outcome with dated onset and ending events, time-to-event methods may be appropriate. Someone still experiencing it at the last observation has an end time that has not yet been observed. More complex duration estimates require assumptions and appropriate analytical support; a final-two-waves subtraction is not a universal substitute.
When reporting, separate what the evidence confirms, what remains unknown and what is estimated. A short limitation beside the number is useful. A confident headline followed by a distant footnote is not an equivalent explanation.
How SROI drop-off differs from an observed trend
The SROI Guide, sections 3.3, 4.3 and 5.1, treats duration and later-period drop-off as explicit parts of the calculation. It describes applying a reduction to the remaining outcome in successive years and retaining the basis for the estimate. Keep those assumptions distinct from your observed follow-up results.
For a fictional calculation only, set the first-year adjusted outcome value to 100 units and assume a 20% annual drop-off for the next two years. Year two is 100 × 0.80 = 80. Year three is 80 × 0.80 = 64. The three-year total before discounting is 244 units.
This is a compounding reduction, not subtracting twenty units every year. The sequence 100, 80, 60 follows a different rule. Neither rule becomes evidence merely because the spreadsheet calculates it correctly.
If the assumed rate were 10%, the sequence would be 100, 90, 81, totaling 271. At 30%, it would be 100, 70, 49, totaling 219. This sensitivity exercise shows how the result depends on the chosen rate; it does not recommend any of these rates for your outcome.
These examples are not an SROI ratio. They omit the investment denominator and discounting, and their starting value assumes other relevant adjustments have already been addressed. Continue with the SROI calculation lesson for the broader calculation.
Keep an assumption register alongside the evidence
For each outcome, retain the definition, observed horizon, forecast horizon if any, rate or pattern assumed, supporting source, affected population, reviewer and review date. Record whether the basis is direct follow-up, another study, stakeholder evidence or a provisional scenario.
A study from another setting can inform an assumption without proving that it applies to your participants. Note relevant differences in population, delivery and measurement. If evidence is weak, show the sensitivity and avoid presenting a precise rate as certain.
Do not apply one duration to every outcome merely because it simplifies the report. A short-lived improvement in access, a skill retained for longer and a recurring annual benefit need distinct definitions. Also check that repeated values are not counting the same one-time benefit again.
When new follow-up arrives, compare it with the prior assumption. Revise transparently where justified. Preserve the old reporting snapshot so readers can distinguish new evidence from a changed calculation rule.
Write a claim that keeps its date and denominator
For the fictional cohort: “At the twelve-month follow-up, 49 of 70 graduates with known status were in paid work. Thirty of the original 100 had unknown status. Among 60 graduates observed at both six and twelve months, employment counts changed from 45 to 42. These checks do not establish uninterrupted employment.”
In a real report, add the actual collection date or range and the outcome definition. Avoid automatically updating the claim to the present tense when reusing an older figure. A March observation remains a March observation in a September presentation.
If the same report includes a forecast, place it in a separate labeled section: “Future value is modeled under these assumptions.” Do not blend projected years into an observed-results chart without a visible boundary.
Use the impact-report writing guide to organize the narrative and the report examples to explore presentation options. Keep the evidence limits attached when a chart or sentence is copied into another document.
Test the follow-up workflow before the next report
- Select one outcome. Define what counts and which date starts the relevant period.
- Reconcile the cohort. Separate eligible people, known status, unknown status and matched observations.
- Check the time claim. Identify whether the source covers a checkpoint or an interval.
- Separate forecasts. Store duration and drop-off assumptions with their rationale and sensitivity results.
- Reproduce the finding. Ask a second reviewer to recover the source, dates, denominator and rule from the report.
For a configured Sopact workflow, test how repeated surveys, supporting documents and status updates stay attached to a contact and period. Ask to see how a reviewer checks AI-prepared evidence and how the report retains its source and definition. Do not assume scheduling or reporting controls exist until the configured workflow demonstrates them.
A successful test might conclude that persistence remains unknown. That is a useful result when it identifies the next evidence needed. It is preferable to a longer-duration claim the records cannot support.
Watch: interpreting an SROI result
Watch the video · 4 minutes 13 seconds. This SROI explainer provides context for interpreting a ratio and the evidence behind it. Use the worked exercise above for the specific duration and drop-off distinction. Browse more videos in the video library.
Frequently asked questions
Can one follow-up provide evidence about duration?
It can, depending on the question and source. Current status describes a checkpoint. A credible dated history may describe an interval, with recall and other limitations made clear.
Do two positive checks prove continuous benefit?
No. The outcome may have stopped and restarted between checks. Use interval evidence if uninterrupted persistence is the claim.
Is survey attrition the same as outcome drop-off?
No. Attrition concerns missing follow-up information. Outcome reduction concerns the outcome itself; SROI drop-off is an explicit later-period adjustment. Missing responses cannot automatically supply that rate.
Should follow-up always start from program exit?
No. Choose the event appropriate to the question, such as exit, first placement or outcome onset. Preserve actual dates and explain the choice.
Can we forecast beyond the last observation?
Yes, with a justified, clearly labeled model and uncertainty or sensitivity analysis. Do not present projected years as measured outcomes.
Does a 20% annual drop-off mean subtracting twenty units each year?
Not when the rule reduces the remaining value by 20%. Starting at 100, that produces 100, 80 and 64. State the exact rule and period.
What should accompany a lasting-outcome claim?
The outcome definition, source, time horizon, collection dates, population and missingness. Explain whether the evidence covers checkpoints, an interval or an estimate.
Keep follow-up evidence within its access rules
A continuing record is only useful when access fits its purpose. Continue to the lesson on what the assistant may see, carrying the evidence, sources and review responsibilities you have defined.
Reviewed September 12, 2026. Graduate counts and forecasting values are fictional teaching examples. They are not customer results or recommended drop-off rates.