How to Report Attrition Without Hiding What It Means

Forty people start your twelve-week group. Nineteen finish. Your outcomes report will say the programme “improved loneliness scores by 22% among participants.” That sentence is true and it is also close to worthless, because it describes a self-selected group of finishers and says nothing about the twenty-one people who left.

This is the most common way measurement reports mislead without technically lying. Nobody fabricates a number. They just report on the people the data made convenient to report on.

The three attrition numbers you owe a reader

Every outcomes report involving a cohort with dropout should state, near the top, not buried in a methods appendix:

Who started. The full enrolled number, not the number who completed a baseline survey.
Who finished. The number with both baseline and follow-up data.
Who you know left and why, where you know it. Not speculation — only what you actually recorded (moved away, health decline, said it wasn’t for them, unreachable).

Most reports give you number two dressed up as number one. A line like “94% of participants reported reduced loneliness” that turns out to mean 94% of the 43% who completed the programme is not a measurement error. It’s a framing choice, and funders who read enough of these reports learn to distrust the ones that don’t show their denominator.

What this means in practice: report the denominator in the same sentence as the outcome, not in a footnote. “22% improvement among the 19 of 40 participants who completed both surveys” is one sentence. There is no length excuse for splitting it.

The completer-bias problem specifically

Social prescribing evaluations are especially exposed to this because the outcomes literature they draw on is itself built almost entirely on people who stayed engaged. A qualitative meta-synthesis of social prescribing found that participants describe benefit as coming from sustained, purposeful engagement, not from contact alone — which is a finding about people who got far enough into a programme to describe that experience. It says nothing about people who tried a session and didn’t come back, because those people are rarely in the sample. Systematic reviews of social prescribing’s effect on loneliness report positive impacts across the studies included, but the reviews themselves note wide heterogeneity in design and limited attention to who was lost along the way. If your own report cites “the evidence base shows social prescribing reduces loneliness” as justification for confidence in your completer-only numbers, you are laundering someone else’s selection bias through your own.

The AARP Foundation’s 2018 survey is a useful outside check on why this matters: loneliness correlates strongly with network size and physical isolation, which are exactly the conditions that predict early dropout from a group programme. The people most likely to benefit from social connection interventions are, on the evidence, also the people most likely to be too isolated, unwell, or under-resourced to complete them. A report that only measures completers is structurally biased toward the people who needed the intervention least.

The week four problem

Programme staff who run recurring groups will tell you, informally, that the real drop happens around session three or four — after the novelty of joining wears off and before the group has built enough cohesion to be its own draw. Call this the week four problem when you write it up: if your attrition curve shows a cliff at a specific session number rather than a steady bleed, that is a design signal, not just a measurement footnote. It tells you something about the programme, not only about the participants. A steady bleed across all twelve weeks suggests something different — maybe the format never built cohesion at all. Report the shape of the curve, not just the start and end points, because the shape is diagnostic.

Phrases that hide more than they disclose

Phrase in the report	What it usually means	What to write instead
“Participants reported…”	Only completers reported	“Of the X who completed follow-up (Y% of enrollees)…”
“Retention was strong”	No stated baseline for comparison	State the actual completion rate and, if possible, a comparator from a similar programme
“Some participants did not continue for personal reasons”	Unknown reasons, dressed as known	“Reasons for leaving were not recorded for N of the M who left”
“94% satisfaction among respondents”	Survey response rate not given	Give the response rate in the same sentence

Evidence status: what attrition data can and can’t tell you

Claim	Evidence status
Completer-only outcome reporting overstates programme effect	Well established as a general measurement principle; not something the social prescribing literature has specifically quantified
Social prescribing improves self-reported loneliness and wellbeing among engaged participants	Reasonably supported across multiple systematic reviews, with the caveat below
Those same reviews adequately account for who dropped out and why	Weak — the reviews themselves flag heterogeneity and thin methodology as limitations
Isolation and low network size predict early dropout from group interventions	Plausible and consistent with survey data on who is lonely, but not directly tested as a dropout predictor in the sources here

What this does not solve

Honest attrition reporting will not make your numbers look better, and it will sometimes make a funder ask harder questions about a programme that is genuinely working for the people it retains. That is a legitimate cost, and this note does not tell you how to bear it. It also does not solve the underlying problem: most programme evaluation, however careful, describes what happened to people who showed up. It says very little about the larger, silent group who were never reached at all — who didn’t hear about the programme, couldn’t get there, or decided in the first five minutes it wasn’t for them. Attrition data is at least visible. Non-enrollment is not, and no amount of careful denominator reporting fixes that.